Multiwfn 3.7

Multiwfn
− A Multifunctional Wavefunction Analyzer −
Software Manual
with tutorials and abundant examples in Chapter 4
Version 3.7
2020-Aug-14
Official site: http://sobereva.com/multiwfn
Tian Lu
[email protected]
Beijing Kein Research Center for Natural Sciences (www.keinsci.com)

!!!!!!!!!! ALL USERS MUST READ !!!!!!!!!!
1. Please feel free to ask questions about using Multiwfn by posting topic on
Multiwfn English forum (http://sobereva.com/wfnbbs) or Multiwfn Chinese forum
(http://bbs.keinsci.com/wfn)! I am always willing to provide as much help as I can for
any Multiwfn user!!! \()/
2. To very quickly understand how to use Multiwfn to carry out very common
tasks, please check "Multiwfn quick start.pdf" in the Multiwfn binary package.
3. The BEST way to get started is reading through Chapter 1, Sections 2.1~2.5 of
Chapter 2, and then follow the tutorials in Chapter 4. After that if you want to learn
more about theoretical backgrounds and details of Multiwfn, then read Section 2.6~2.8
of Chapter 2 and Chapter 3. Note that the tutorials and examples given in Chapter 4
only cover most important and frequently used functions rather than all functions of
Multiwfn.
4. Different functions of Multiwfn require different type of input file, please
carefully read Section 2.5 for explanation.
5. Usually Multiwfn runs in interactive mode, however Multiwfn can also be run
in command-line mode, and it is very convenient to write shell script to make Multiwfn
deal with a large number of input files. See Sections 5.2 and 5.3 for detail.
6. If you do not know how to copy the output of Multiwfn from command-line
window to a plain text file, consult Section 5.4. If you do not know how to enlarge
screen buffer size of command-line window of Windows system, consult Section 5.5.
7. If error “No executable for file l1.exe” appears in screen when Multiwfn is
invoking Gaussian, you should set up Gaussian environment variable first. For
Windows version, you can refer Appendix 1. (Note: Most functions in Multiwfn DO
NOT require Gaussian installed on your local machine)
8. The so-called ”current folder” in this manual and in prompts of Multiwfn refers
to the path where you are invoking Multiwfn. If you boot up Multiwfn by clicking the
icon of executable file in Windows platform, the “current folder” is the folder
containing Multiwfn executable file. In the case of command-line mode, if you are in
“D:\study\” directory when invoking Multiwfn, then “D:\study\” is “current directory”.
9. Please frequently check Multiwfn website and update the program to the latest
version. Multiwfn is always in active development, useful new functions are frequently
added, bugs are continuously fixed and efficiency is continuously improved.
i
Linux and Mac OS USERS MUST READ
1. If you are new to Multiwfn, using Windows version is highly recommended,
since this version of Multiwfn is more stable than others, the GUI effect is the best and
you do not need to set up your system.
2. See Section 2.1.2 and 2.1.3 on how to install Linux and Mac OS versions of
Multiwfn, respectively.
3. When graphical user interface (GUI) appears, the graph cannot be shown
automatically, you have to do something to active the graph, e.g. clicking "up" button
or dragging a scale bar in the GUI.
4. Transparent style does not work when showing isosurfaces in GUI window. But
if you choose to save the graph as image file, the transparent isosurfaces can be rendered
normally in the resulting graphical file.
5. Using Mac OS version is deprecated, some functions may do not work normally.
ii
About the manual
This content in the manual is organized in the following sequence:
Chapter 1 Overview: Briefly describes what is Multiwfn and what Multiwfn can
do.
Chapter 2 General information: Introduces all aspects beyond specific functions
of Multiwfn, such as how to install, how to use, supported file types, etc.
Chapter 3 Functions: Describes all functions and related theories of Multiwfn in
detail.
Chapter 4 Tutorials and examples: Plenty of practical examples are provided for
helping users to use Multiwfn and understand value of various analysis methods. The
Section 4.A contains some advanced tutorials, important overviews and special topics.
Chapter 5 Skills: Some useful skills in using Multiwfn.
Appendix
In general, italics font is used for inputted commands, journal names and filenames
throughout this manual. Very important contents are highlighted by red color or bolded.
You can directly jump to specific section by choosing corresponding entry in
bookmark window of your pdf reader.
The purpose of this manual is not only helping users to use Multiwfn, but also
introducing related theories. I hope any quantum chemist can benefit from this manual.
Albeit I have tried to write this manual as readable as possible for beginners, some
topics request the readers have basic knowledges of quantum chemistry. If the readers
have carefully read the book "Quantum Chemistry 7ed" written by Ira. Levine, then
they will never encounter any difficulty during reading through the whole manual.
iii
Contents
!!!!!!!!!! ALL USERS MUST READ !!!!!!!!!! ..........................................i
Linux and Mac OS USERS MUST READ ............................................ ii
1 Overview ...............................................................................................1
2 General information ............................................................................9

2.1 Install....................................................................................................................................... 9
2.1.1 Windows version .............................................................................................................. 9
2.1.2 Linux version ................................................................................................................... 9
2.1.3 Mac OS version .............................................................................................................. 10
2.2 Using Multiwfn ..................................................................................................................... 10
2.3 Files of Multiwfn .................................................................................................................. 11
2.4 Parallel implementation ........................................................................................................ 12
2.5 Input files and wavefunction types........................................................................................ 12
2.6 Real space functions.............................................................................................................. 18
2.7 User-defined real space function ........................................................................................... 30
2.8 Graphic formats and image size ............................................................................................ 43
3 Functions ............................................................................................44
3.2 Showing molecular structure and viewing orbitals / isosurfaces (0) ..................................... 44
3.3 Outputting all properties at a point (1) .................................................................................. 47
3.4 Outputting and plotting specific property in a line (3) .......................................................... 47
3.5 Outputting and plotting specific property in a plane (4) ....................................................... 48
3.5.1 Graph types .................................................................................................................... 49
3.5.2 Setting up grid, plane and plotting region ...................................................................... 53
3.5.3 Options in post-processing interface .............................................................................. 55
3.5.4 Setting up contour lines .................................................................................................. 56
3.5.5 Plot critical points, paths and interbasin paths on plane graph....................................... 57
3.6 Outputting and plotting specific property within a spatial region (5) ................................... 58
3.7 Custom operation, promolecular and deformation properties (subfunction 0, -1, -2 in main
function 3, 4, 5) ........................................................................................................................... 62
3.7.1 Custom operation for multiple wavefunctions (0) ......................................................... 62
3.7.2 Promolecular and deformation properties (-1, -2) .......................................................... 62
3.7.3 Generation of atomic wavefunctions.............................................................................. 63
3.7.4 Sphericalization of atom wavefunction .......................................................................... 64
3.8 Checking & Modifying wavefunction (6) ............................................................................. 64
3.9 Population analysis and atomic charges (7) .......................................................................... 67
3.9.1 Hirshfeld atomic charge (1)............................................................................................ 67
3.9.2 Voronoi deformation density (VDD) atom population (2) ............................................. 68
iv
3.9.3 Mulliken atom & basis function population analysis (5) ............................................... 69
3.9.4 Löwdin atom & basis function population (6) ............................................................... 70
3.9.5 Modified Mulliken atom population defined by Ros & Schuit (SCPA) (7) ................... 71
3.9.6 Modified Mulliken atom population defined by Stout & Politzer (8) ............................ 72
3.9.7 Modified Mulliken atom population defined by Bickelhaupt (9)................................... 72
3.9.8 Becke atomic charge with atomic dipole moment correction (10) ................................. 72
3.9.9 Atomic dipole moment corrected Hirshfeld atomic charges (ADCH, 11) ..................... 73
3.9.10 CHELPG (Charges from electrostatic potentials using a grid based method) ESP fitting
atomic charge (12)................................................................................................................... 75
3.9.11 Merz-Kollmann (MK) ESP fitting atomic charge (13) ................................................. 77
3.9.12 AIM atomic charge (14) ............................................................................................... 78
3.9.13 Hirshfeld-I atomic charge (15) ..................................................................................... 78
3.9.14 CM5 atomic charge (16) .............................................................................................. 82
3.9.15 Electronegativity Equalization Method (EEM) atomic charge (17) ............................. 83
3.9.16 Restrained ElectroStatic Potential (RESP) atomic charge (18) .................................... 85
3.9.16.1 Theory ................................................................................................................... 85
3.9.16.2 Usage and some details ......................................................................................... 90
3.9.17 PEOE (Partial equalization of orbital electronegativity) or Gasteiger charge (19) ...... 93
3.10 Orbital composition analysis (8) ......................................................................................... 95
3.10.1 Output basis function, shell and atom composition in a specific orbital by Mulliken,
Stout-Politzer and SCPA approaches (1, 2, 3) ......................................................................... 95
3.10.2 Define fragment 1 and 2 (-1, -2) .................................................................................. 97
3.10.3 Output composition of fragment 1 and inter-fragment composition by Mulliken, Stout-
Politzer and SCPA approaches (4, 5, 6)................................................................................... 97
3.10.4 Orbital composition analysis by natural atomic orbital approach (7) .......................... 98
3.10.5 Calculate atom and fragment contributions by Hirshfeld or Hirshfeld-I method (8,10)
................................................................................................................................................. 99
3.10.6 Calculate atom and fragment contributions by Becke method (9) ............................. 100
3.10.7 Calculate atom and fragment contributions by AIM method (11) .............................. 100
3.10.100 Evaluate oxidation state by LOBA method (100) .................................................. 101
3.11 Bond order analysis (9) ..................................................................................................... 101
3.11.1 Mayer bond order analysis (1).................................................................................... 101
3.11.2 Multi-center bond order analysis (2, -2, -3) ............................................................... 103
3.11.3 Wiberg bond order analysis in Löwdin orthogonalized basis (3) ............................... 106
3.11.4 Mulliken bond order analysis (4) and decomposition (5)........................................... 106
3.11.5 Orbital occupancy-perturbed Mayer bond order (6)................................................... 107
3.11.6 Fuzzy bond order (7) .................................................................................................. 108
3.11.7 Laplacian bond order (8) ............................................................................................ 108
3.11.8 Decompose Wiberg bond order in NAO basis as atomic orbital pair contributions (9)
............................................................................................................................................... 109
3.11.9 Intrinsic bond strength index (IBSI) (10) ................................................................... 110
3.11.10 AV1245 index (approximate multi-center bond order for large rings) and AVmin .. 111
3.12 Plotting total density-of-states (DOS), partial DOS, overlap population DOS, local DOS and
photoelectron spectrum (10) ..................................................................................................... 113
v
3.12.1 Theory ........................................................................................................................ 113
3.12.2 Input file ..................................................................................................................... 115
3.12.3 Options for plotting DOS and basic usage ................................................................. 116
3.12.4 Local DOS.................................................................................................................. 118
3.12.4 Photoelectron spectrum .............................................................................................. 118
3.13 Plotting IR, Raman, UV-Vis, ECD, VCD, ROA and NMR spectra (11) ........................... 120
3.13.1 Theory ........................................................................................................................ 120
3.13.2 Input file ..................................................................................................................... 123
3.13.3 Usage and options ...................................................................................................... 126
3.13.4 Plotting multiple systems together and weighted spectrum ....................................... 128
3.13.5 Plotting NMR spectrum ............................................................................................. 129
3.14 Topology analysis (2) ........................................................................................................ 131
3.14.1 Theory ........................................................................................................................ 131
3.14.2 Search critical points .................................................................................................. 133
3.14.3 Generate topology paths ............................................................................................. 135
3.14.4 Generate interbasin surfaces....................................................................................... 136
3.14.5 Visualize, analyze, modify and export results ............................................................ 137
3.14.6 Calculate the aromaticity indices based on topology properties of electron density.. 138
3.15 Quantitative analysis of molecular surface (12) ................................................................ 139
3.15.1 Theory ........................................................................................................................ 139
3.15.2 Numerical algorithm .................................................................................................. 143
3.15.2.1 Analysis on the whole molecular surface ............................................................ 143
3.15.2.2 Analysis on local molecular surface .................................................................... 144
3.15.3 Parameters and options .............................................................................................. 146
3.15.4 Options at post-processing stage ................................................................................ 148
3.15.5 Special topic: Hirshfeld and Becke surface analyses ................................................. 152
3.16 Processing grid data (13) ................................................................................................... 156
3.16.0 Visualize isosurface of present grid data (-2) ............................................................. 156
3.16.1 Output present grid data to Gaussian cube file (0) ..................................................... 156
3.16.2 Output all data points with value and coordinate (1) ................................................. 156
3.16.3 Output data points in a XY/YZ/XZ plane (2, 3, 4)..................................................... 156
3.16.4 Output average data of XY/YZ/XZ planes in a range of Z/X/Y (5,6,7) ..................... 156
3.16.5 Output data points in a plane defined three atom indices or three points (8,9) .......... 156
3.16.6 Output data points in specified value range (10) ........................................................ 157
3.16.7 Grid data calculation (11) ........................................................................................... 157
3.16.8 Map values of a cube file to specified isosurface of present grid data (12) ............... 158
3.16.9 Set value of the grid points that far away from / close to some atoms (13) ............... 158
3.16.10 Set value of the grid points outside overlap region of two fragments (14) .............. 159
3.16.11 If data value is within certain range, set it to a specified value (15) ........................ 159
3.16.12 Scale data range (16) ................................................................................................ 159
3.16.13 Show statistic data of the points in specific spatial and value range (17) ................ 159
3.16.14 Calculate and plot integral curve in X/Y/Z direction (18) ....................................... 160
3.17 Adaptive natural density partitioning (AdNDP) analysis (14) .......................................... 161
3.17.1 Theory ........................................................................................................................ 161
vi
3.17.2 Input file ..................................................................................................................... 163
3.17.3 Options ....................................................................................................................... 164
3.18 Fuzzy atomic space analysis (15) ...................................................................................... 166
3.18.0 Basic concepts ............................................................................................................ 166
3.18.1 Integration of a real space function in fuzzy atomic spaces (1) ................................. 169
3.18.2 Integration of a real space function in overlap spaces (8) .......................................... 170
3.18.3 Calculate atomic and molecular multipole moments (2) ............................................ 170
3.18.4 Calculate atomic overlap matrix (3) ........................................................................... 172
3.18.5 Calculate localization and delocalization index (4).................................................... 173
3.18.6 Calculate PDI (5)........................................................................................................ 175
3.18.7 Calculate FLU and FLU-π (6,7) ................................................................................. 176
3.18.8 Calculate condensed linear response kernel (9) ......................................................... 177
3.18.9 Calculate para linear response index (10) .................................................................. 178
3.18.10 Calculate multi-center delocalization index (11)...................................................... 179
3.18.11 Calculate information-theoretic aromaticity index (12) ........................................... 179
3.19 Charge decomposition analysis and plotting orbital interaction diagram (16) .................. 180
3.19.1 Theory ........................................................................................................................ 180
3.19.2 Input file ..................................................................................................................... 183
3.19.3 Usage.......................................................................................................................... 184
3.20 Basin analysis (17) ............................................................................................................ 186
3.20.1 Theory ........................................................................................................................ 186
3.20.2 Numerical aspects ...................................................................................................... 187
3.20.3 Usage.......................................................................................................................... 191
3.21 Electron excitation analysis (18) ....................................................................................... 195
3.21.A Basic information about electron excitation analysis module ................................... 195
3.21.0 Check, modify and export configuration coefficients of an excitation (-1) ............... 198
3.21.1 Analyze and visualize hole-electron distribution, transition density, and transition
electric/magnetic dipole moment density (1) ........................................................................ 199
3.21.1.1 Theory ................................................................................................................. 200
3.21.1.2 Usage and Functions ........................................................................................... 207
3.21.2 Plot atom/fragment transition matrix of various kinds as heat map (2) ..................... 210
3.21.3 Analyze charge-transfer based on density difference grid data (3) ............................ 214
3.21.4 Calculate ∆r index to measure charge-transfer length (4) .......................................... 216
3.21.5 Calculate transition electric dipole moments between all states and electric dipole
moment of each state (5) ....................................................................................................... 217
3.21.6 Generate natural transition orbitals (NTOs) (6) ......................................................... 219
3.21.7 Calculate ghost-hunter index (7) ................................................................................ 221
3.21.8 Calculate interfragment charge transfer in electron excitation via IFCT method (8) . 222
3.21.9 Generate and export transition density matrix (9) ...................................................... 224
3.21.10 Decompose transition dipole moment as molecular orbital pair contributions (10) 227
3.21.11 Decompose transition dipole moment as basis function and atom contributions (11)
............................................................................................................................................... 228
3.21.12 Calculate Mulliken atomic transition charges (12) .................................................. 229
3.21.13 Generate natural orbitals of specific excited states (13)........................................... 230
vii
3.21.14 Calculate Λ index to characterize electron excitation (14)....................................... 230
3.21.15 Print major MO transitions in all excited states ....................................................... 232
3.22 Orbital localization analysis (19) ...................................................................................... 232
3.23 Visual study of weak interaction (20)................................................................................ 237
3.23.1 Noncovalent interaction (NCI) anaylsis (1) ............................................................... 237
3.23.2 NCI analysis based on promolecular density (2)........................................................ 246
3.23.3 Averaged NCI analysis (NCI analysis for multiple frames. 3) ................................... 247
3.23.4 Density Overlap Regions Indicator (DORI) analysis (5) ........................................... 248
3.23.5 Independent Gradient Model (IGM) analysis based on promolecular density (10) ... 249
3.23.6 IGM analysis based on Hirshfeld partition of molecular density (IGMH) (11) ......... 254
3.23.7 Visualization of van der Waals potential (6) .............................................................. 255
3.23.8 Interaction region indicator (IRI) analysis (4) ............................................................ 256
3.24 Energy decomposition analysis (21) ................................................................................. 257
3.24.1 Energy decomposition analysis based on molecular forcefield (EDA-FF) ................ 257
3.24.2 Shubin Liu's energy decomposition ........................................................................... 260
3.24.5 Simple energy decomposition analysis ...................................................................... 262
3.100 Other functions, part 1 (100) ........................................................................................... 262
3.100.1 Draw scatter graph between two functions and generate their cube files ................ 262
3.100.2 Export various files or generate input file of quantum chemistry programs ............ 263
3.100.3 Calculate molecular van der Waals volume ............................................................. 264
3.100.4 Integrate a function over the whole space ................................................................ 265
3.100.5 Show overlap integral between alpha and beta orbitals ........................................... 266
3.100.6 Monitor SCF convergence process of Gaussian....................................................... 267
3.100.8 Generate Gaussian input file with initial guess combined from fragment wavefunctions
............................................................................................................................................... 268
3.100.9 Evaluate interatomic connectivity and atomic coordination number ....................... 270
3.100.11 Calculate overlap and centroid distance between two orbitals ............................... 271
3.100.12 Perform biorthogonalization between alpha and beta orbitals ............................... 272
3.100.13 Calculate HOMA and Bird aromaticity index ........................................................ 274
3.100.14 Calculate LOLIPOP (LOL Integrated Pi Over Plane) ............................................ 276
3.100.15 Calculate intermolecular orbital overlap ................................................................ 276
3.100.16 Calculate various quantities in conceptual density functional theory (CDFT)....... 277
1 Theory ............................................................................................................................ 277
2 Usage.............................................................................................................................. 279
3 Special topic: Orbital-weighted Fukui function and dual descriptor.............................. 280
3.100.18 Yoshizawa's electron transport route analysis ........................................................ 281
3.100.19 Generate promolecular .wfn file from fragment wavefunctions ............................ 283
3.100.20 Calculate Hellmann-Feynman forces ..................................................................... 284
3.100.21 Calculate properties based on geometry information for specific atoms ............... 284
3.100.22 Detect π orbitals, set occupation numbers and calculate π composition ................ 286
3.100.23 Fit function distribution to atomic value ................................................................ 288
3.100.24 Obtain NICSZZ value for non-planar or tilted system ............................................ 289
3.200.1 Calculate core-valence bifurcation (CVB) index and related quantities .................. 291
viii
3.200.2 Calculate atomic and bond dipole moments in Hilbert space .................................. 296
3.200.3 Generate cube file for multiple orbital wavefunctions ............................................. 298
3.200.4 Generate iso-chemical shielding surfaces (ICSS) and related quantities ................. 299
3.200.5 Plot radial distribution function for a real space function ........................................ 300
3.200.6 Analyze correspondence between orbitals in two wavefunctions ............................ 301
3.200.7 Parse output of (hyper)polarizability task of Gaussian and evaluate relevant quantities
............................................................................................................................................... 302
3.200.8 Study (hyper)polarizability by sum-over-states (SOS) method and two- or three-level
model analyses ...................................................................................................................... 308
3.200.8.1 Calculation of (hyper)polarizability .................................................................. 308
3.200.8.2 Two-level and three-level model analyses for hyperpolarizability.................... 311
3.200.9 Calculate average bond length and average coordinate number .............................. 313
3.200.10 Output various kinds of integral between orbitals .................................................. 314
3.200.11 Calculate center, the first and second moments of a real space function ............... 315
3.200.12 Calculate energy index (EI) or bond polarity index (BPI) ..................................... 316
3.200.13 Evaluate orbital contributions to density difference or other grid data .................. 317
3.200.14 Domain analysis (obtaining properties within isosurfaces of a function) .............. 319
3.200.15 Calculate electron correlation index ....................................................................... 321
3.200.16 Generate natural orbitals, natural spin orbitals and spin natural orbitals based on the
density matrix in .fch/.fchk file ............................................................................................. 321
3.200.17 Calculate Coulomb and exchange integral between two orbitals ........................... 323
3.200.18 Calculate bond length/order alternation (BLA/BOA) and angle/dihedral alternation
............................................................................................................................................... 324
3.200.20 Bond order density (BOD) and natural adaptive orbital (NAdO) analyses ............ 325
3.300.1 Viewing free regions and calculating free volume in a box ..................................... 329
3.300.2 Fitting atomic radial density as linear combination of multiple STOs or GTFs ...... 330
3.300.2.1 Algorithm and technical details ......................................................................... 331
3.300.2.2 Usage................................................................................................................. 333
3.300.3 Visualize (hyper)polarizability via unit sphere and vector representations .............. 336
3.300.4 Simulating scanning tunneling microscope (STM) image ....................................... 338
3.300.5 Calculate electric dipole, quadrupole and octopole moments of present system
analytically ............................................................................................................................ 341
4 Tutorials and Examples ..................................................................343

Prologue and generation of input files ...................................................................................... 343
4.0 View orbitals and structure.................................................................................................. 346
4.0.1 Viewing molecular orbitals of cycloheptatriene ........................................................... 346
4.0.2 Viewing natural bond orbitals (NBO) of ethanol ......................................................... 348
4.0.3 Using Multiwfn+VMD to rapidly plot high-quality orbital isosurface map ................ 350
4.1 Calculate properties at a point ............................................................................................. 352
4.1.1 Show all properties of triplet water at a given point .................................................... 352
4.1.2 Calculate ESP at nuclear positions to evaluate interaction strength of H2O∙∙∙HF ........ 354
4.2 Topology analysis................................................................................................................ 356
ix
4.2.1 Atoms in molecules (AIM) and aromaticity analysis for 2-pyridoxine 2-aminopyridine
............................................................................................................................................... 356
4.2.2 LOL topology analysis for acetic acid.......................................................................... 361
4.2.3 Plot real space function along bond path ..................................................................... 363
4.2.4 Decompose properties at a critical point as orbital contributions ................................ 364
4.2.5 Easily plot high quality AIM topology map in VMD visualization program based on
Multiwfn outputs ................................................................................................................... 367
4.2.6 Topology analysis in special regions: G-C...G-C base pair as an example .................. 370
4.2.7 Topology analysis based on attractors located by basin analysis: spin density of biradical
as an example ........................................................................................................................ 373
4.3 Output and plot various properties in a line ........................................................................ 375
4.3.1 Plot the spin density curve of triplet formamide along carbon and oxygen atoms....... 375
4.3.2 Study Fermi hole and Coulomb hole of H2 .................................................................. 376
4.3.3 Study interatomic interaction via PAEM-MO method ................................................. 379
4.4 Output and plot various properties in a plane ..................................................................... 382
4.4.1 Illustration of plotting color-filled map and contour line map ..................................... 382
4.4.1.1 Plotting electron density of hydrogen cyanide ...................................................... 382
4.4.1.2 Plotting electron localization function (ELF) for FOX-7 ...................................... 385
4.4.2 Shaded surface map with projection effect of electron localization function (ELF) of
monofluoroethane ................................................................................................................. 388
4.4.3 Plotting plane map without contributions from some atoms ........................................ 389
4.4.4 Contour map of electrostatic potential of chlorine trifluoride ...................................... 391
4.4.5 Contour map of two orbital wavefunctions .................................................................. 393
4.4.6 Gradient+contour map with topology paths of electron density of hydrogen peroxide
............................................................................................................................................... 395
4.4.7 Deformation map of electron density of acetyl chloride .............................................. 399
4.4.8 Draw difference map of electron density and ELF for water tetramer with respect to its
constituent monomers ........................................................................................................... 400
4.4.9 Draw LOL-π map for porphyrin to reveal favorable electron delocalization path ....... 403
4.4.10 Plotting gradient line and vector field map of electrostatic potential to reveal electric
field of LiF ............................................................................................................................ 405
4.5 Generate grid data and view isosurface map ....................................................................... 407
4.5.1 Electron localization function of chlorine trifluoride ................................................... 408
4.5.2 Laplacian of electron density of 1,3-butadiene ............................................................ 409
4.5.3 Calculate ELF-α and ELF-π to study aromaticity of benzene ...................................... 410
4.5.4 Use Fukui function and dual descriptor to study favorable site of electrophilic attack for
phenol.................................................................................................................................... 413
4.5.4.1 Fukui function ....................................................................................................... 413
4.5.4.2 Dual descriptor ...................................................................................................... 415
4.5.5 Plot difference map of electron density to study electron transfer of imidazole coordinated
magnesium porphyrin............................................................................................................ 418
4.5.6 Study electron delocalization range function EDR(r;d) of anionic water dimer .......... 422
4.5.7 Study orbital overlap distance function D(r) of thioformic acid .................................. 422
4.6 Modify and check wavefunction ......................................................................................... 423
x
4.6.1 Delete certain Gaussian functions ................................................................................ 423
4.6.2 Remove contributions from certain orbitals to real space functions ............................ 424
4.6.3 Translate and duplicate graphene primitive cell wavefunction to periodic system ...... 426
4.7 Population analysis and atomic charge calculation ............................................................. 427
4.7.0 Mulliken population analysis on triplet ethanol ........................................................... 427
4.7.1 Calculate Hirshfeld and CHELPG atomic charges as well as fragment charge for chlorine
trifluoride .............................................................................................................................. 429
4.7.2 Calculate and compare ADCH atomic charges with Hirshfeld atomic charges for
acetamide .............................................................................................................................. 431
4.7.3 Calculate condensed Fukui function and condensed dual descriptor ........................... 432
4.7.4 Illustration of computing Hirshfeld-I atomic charges .................................................. 434
4.7.5 Calculating EEM atomic charges for ethanol-water cluster ......................................... 435
4.7.6 Determining correspondence between basis functions and atomic orbitals via population
analysis .................................................................................................................................. 438
4.7.7 Illustration of deriving RESP charges and normal ESP fitting charges with extra
constraints ............................................................................................................................. 439
4.7.7.1 Example 1: Deriving RESP charges for dopamine in ethanol environment .......... 440
4.7.7.2 Example 2: Taking multiple conformations into account during RESP charge
calculation of dopamine ............................................................................................................ 442
4.7.7.3 Example 3: Imposing equivalence constraint in ESP fitting of Dimethyl phosphate
.................................................................................................................................................. 444
4.7.7.4 Example 4: Evaluation of atomic charges of aspartic acid residue with equivalence
and charge constraints ............................................................................................................... 446
4.7.7.5 Example 5: Example of setting equivalence constraint according to local or global
point group symmetry ............................................................................................................... 448
4.7.7.6 Example 6: RESP charge calculation with additional fitting centers .................... 452
4.7.7.7 Skill 1: Using two times of one-stage fitting to equivalently realize standard RESP
two-stage fitting ........................................................................................................................ 455
4.7.7.8 Skill 2: Quickly obtaining RESP charges from molecular structure file by only one
command ................................................................................................................................... 456
4.7.7.9: Special topic: Calculation of RESP2 charges....................................................... 457
4.7.8 Examine electrostatic potential reproducibility of atomic charges............................... 459
4.7.9 Calculate PEOE (Gasteiger) charge ............................................................................. 462
4.8 Molecular orbital composition analysis .............................................................................. 463
4.8.1 Analyze acetamide by Mulliken method ...................................................................... 463
4.8.2 Analyze water by natural atomic orbital method.......................................................... 466
4.8.3 Analyze acetamide by Hirshfeld and Becke method .................................................... 469
4.8.4 Calculate oxidation state by LOBA method ................................................................. 471
4.8.5 Quantifying extent of spatial delocalization of orbitals via orbital delocalization index
(ODI) ..................................................................................................................................... 473
4.8.6 Calculate orbital composition contributed by AIM basins and other type of basins .... 476
4.9 Bond order analysis ............................................................................................................. 478
4.9.1 Mayer bond order and fuzzy bond order analysis on acetamide .................................. 479
4.9.2 Multi-center bond order analysis on Li6 cluster and phenanthrene .............................. 482
xi
4.9.3 Calculate Laplacian bond order (LBO) ........................................................................ 485
4.9.4 Decomposition analysis of Wiberg bond order in NAO basis for formaldehyde ......... 486
4.9.5 Study orbital contributions to Mulliken bond order for C-C bond of CH3CONH2 ...... 487
4.9.6 Using intrinsic bond strength index (IBSI) to measure strength of chemical bonds .... 489
4.9.11 Example of using AV1245 and AVmin indices to study aromaticity .......................... 491
4.9.11.1 Using AV1245 and AVmin to study local aromaticity of phenanthrene .............. 491
4.9.11.2 Using AV1245 and AVmin to measure global aromaticity of porphyrin ............. 492
4.10 Plot density-of-states (DOS) maps .................................................................................... 495
4.10.1 Plot total, partial and overlap DOS for N-phenylpyrrole ........................................... 495
4.10.2 Plot local DOS for 1,3-butadiene ............................................................................... 503
4.10.3 Plot DOS map for unrestricted open-shell system: Na3O@Si12C12 ........................... 505
4.10.4 Plot photoelectron spectrum (PES) for Cr3Si12− cluster ............................................. 507
4.10.5 Plot MO-PDOS map to reveal PDOS contributed by different MOs for cyclo[18]carbon
............................................................................................................................................... 508
4.11 Plot various kinds of spectra ............................................................................................. 511
4.11.1 Plot infrared (IR) spectrum for NH3BF3 ..................................................................... 511
4.11.2 Plot UV-Vis spectrum and contributions from individual transitions for acetic acid . 512
4.11.3 Plot electronic circular dichroism (ECD) spectrum for asparagine ............................ 514
4.11.4 Plot conformational weighted UV-Vis and ECD spectra for plumericin .................... 517
4.11.5 Plot Raman spectrum for 2-methyloxirane based on Raman intensity....................... 521
4.11.6 Simultanously plot multiple systems .......................................................................... 522
4.11.7 Plot Raman optical activity (ROA) spectrum for chiral molecule S-methyloxirane .. 524
4.11.8 Skill: Plot spectrum for a batch of files via shell script .............................................. 525
4.11.9 Skill: Use spikes to indicate position of transition levels ........................................... 526
4.11.10 Examples of plotting NMR spectrum ....................................................................... 528
4.11.10.1 Plotting 1H and 13C NMR spectra for acetaldehyde .......................................... 528
4.11.10.2 Plotting NMR spectra for pyridine based on scaling method ............................ 530
4.11.10.3 Plotting conformation weighted NMR spectrum for valine .............................. 531
4.11.10.4 Plotting multiple systems simultaneously ......................................................... 533
4.12 Quantitative analysis of molecular surface ....................................................................... 534
4.12.1 Electrostatic potential analysis on phenol molecular surface ..................................... 534
4.12.2 Average local ionization energy analysis (ALIE) on phenol molecular surface......... 541
4.12.3 Atomic local molecular surface analysis for acrolein................................................. 543
4.12.4 Fukui function distribution on local molecular surface of phenol.............................. 545
4.12.5 Becke surface analysis on guanine-cytosine base pair ............................................... 546
4.12.6 Hirshfeld surface analysis and fingerprint plot analysis on urea crystal .................... 548
4.12.7 Predict density of molecular crystal of FOX-7 .......................................................... 553
4.12.8 Quantitative analysis of orbital overlap distance function D(r) on thioformic acid
molecular surface .................................................................................................................. 554
4.12.9 Evaluate vdW surface area of the whole system as well as individual fragment ....... 555
4.12.10 Quantification of area of sigma-hole and pi-hole ..................................................... 557
4.12.11 Basin-like analysis of molecular surface for electrostatic potential ......................... 561
4.12.12 Estimate kinetic diameter for small molcules .......................................................... 563
4.12.13 Using local electron affinity to reveal electrophilic regions..................................... 565
xii
4.13 Process grid data ............................................................................................................... 566
4.13.1 Extract data points in a plane ..................................................................................... 566
4.13.2 Perform mathematical operation on grid data ............................................................ 567
4.13.3 Scaling numerical range of grid data.......................................................................... 568
4.13.4 Screen isosurfaces in local regions............................................................................. 568
4.13.4.1 Screen isosurfaces inside or outside a region ...................................................... 568
4.13.4.2 Screen isosurfaces outside overlap region of two fragments .............................. 569
4.13.5 Acquire barycenter of a molecular orbital .................................................................. 571
4.13.6 Plot charge displacement curve .................................................................................. 572
4.13.7 Evaluation of electron density overlap ....................................................................... 574
4.14 Adaptive natural density partitioning (AdNDP) analysis .................................................. 576
4.14.1 Analyze Li5+ cluster.................................................................................................... 576
4.14.2 Analyze B11− cluster ................................................................................................... 578
4.14.3 Analyze phenanthrene ................................................................................................ 582
4.14.4 Analyze Au20 cluster................................................................................................... 585
4.15 Fuzzy atomic space analysis ............................................................................................. 586
4.15.1 Study delocalization index of benzene ....................................................................... 586
4.15.2 Study aromaticity of phenanthrene by PDI, FLU, FLU-π and PLR ........................... 587
4.15.3 Calculate fragment dipole moment to exhibit local polarity ...................................... 589
4.16 Charge decomposition analysis and plotting orbital interaction diagram ......................... 591
4.16.1 Closed-shell interaction case: COBH3 ....................................................................... 592
4.16.2 Open-shell interaction case: CH3NH2 ........................................................................ 596
4.16.3 More than two fragments case: Pt(NH3)2Cl2 .............................................................. 598
4.17 Basin analysis .................................................................................................................... 599
4.17.1 AIM basin analysis for HCN and Li6 ......................................................................... 599
4.17.2 ELF basin analysis for acetylene ................................................................................ 606
4.17.3 Basin analysis of electrostatic potential for H2O........................................................ 609
4.17.4 Basin analysis of electron density difference for H2O ............................................... 613
4.17.5 Study source function in AIM basins ......................................................................... 615
4.17.6 Local region basin analysis for polyyne ..................................................................... 617
4.17.7 Evaluate atomic contribution to population of ELF basins ........................................ 618
4.17.8 Calculating high ELF localization domain population and volume (HELP, HELV).. 619
4.18 Electron excitation analysis............................................................................................... 621
4.18.1 Using hole-electron analysis to fully characterize electron excitations ..................... 621
4.18.1.1 Example 1: NH2-biphenyl-NO2 ........................................................................... 622
4.18.1.2 Example 2: Ru(bpy3)2+ cation in water ............................................................... 632
4.18.2 Illustration of transition density (matrix) and transition dipole moment density (matrix)
analysis .................................................................................................................................. 635
4.18.2.1 Analyzing transition density and transition dipole moment density in real
space.......................................................................................................................................... 636
4.18.2.2 Plotting and analyzing transition density matrix (TDM) .................................... 639
4.18.2.3 Plotting and analyzing transition dipole moment matrix .................................... 645
4.18.2.4 Investigating transition density and transition density matrix between excited
states .......................................................................................................................................... 647
xiii
4.18.3 Analyze charge-transfer during electron excitation based on electron density difference
............................................................................................................................................... 650
4.18.4 Calculate ∆r and Λ indices to characterize various electron excitations for N-
phenylpyrrole ........................................................................................................................ 652
4.18.6 Generate and analyze natural transition orbitals (NTOs) for uracil ........................... 655
4.18.8 Using IFCT method and heat map of charge transfer matrix to study interfragment
charge transfer during electron excitation ............................................................................. 657
4.18.8.1 IFCT analysis for 4-nitroaniline .......................................................................... 658
4.18.8.2 Plotting heat map of charge transfer matrix to intuitively understand nature of
electron excitation ..................................................................................................................... 660
4.18.9 Generate transition density matrix and transform it to orbital representation ............ 662
4.18.10 Gain molecular orbital pair contributions to transition dipole moment ................... 664
4.18.11 Plot transition dipole moment vector contributed by molecular fragments as arrows
............................................................................................................................................... 665
4.18.13 Study electronic structure of a single excited state and difference between two excited
states ...................................................................................................................................... 668
4.19 Orbital localization analysis .............................................................................................. 671
4.19.1 Localizing molecular orbital of 1,3-butadiene by Pipek-Mezey method ................... 671
4.19.2 Analyze variation of localized molecular orbitals for SN2 reaction .......................... 674
4.19.3 Characterize Re-Re bond of [Re2Cl8]2- anion............................................................. 675
4.19.4 Study bond dipole moment based on two-center LMOs for CH3NH2 ....................... 677
4.20 Visual study of weak interactions...................................................................................... 679
4.20.1 Studying weak interaction in 2-pyridoxine 2-aminopyridine by NCI method ........... 679
4.20.2 Studying weak interaction in DNA by NCI method based on promolecular density . 682
4.20.3 Visually studying weak interaction for water in bulk environment by aNCI method 684
4.20.4 Simultaneously revealing covalent and noncovalent interactions in phenol dimer by IRI
analysis .................................................................................................................................. 689
4.20.5 Simultaneously revealing covalent and noncovalent interactions in phenol dimer by
DORI analysis ....................................................................................................................... 691
4.20.6 Visualizing and analyzing van der Waals potential .................................................... 691
4.20.6.1 Example 1: Helicene ........................................................................................... 692
4.20.6.2 Example 2: Cyclo[18]carbon .............................................................................. 693
4.20.10 Visualize and quantify weak interactions by Independent Gradient Model (IGM) .. 696
4.20.10.1 Example 1: Guanine-cytosine (GC) base pair ................................................... 696
4.20.10.2 Example 2: C60-coronene dimer ........................................................................ 704
4.20.10.3 Example 3: Oxazolidinone trimer ..................................................................... 705
4.20.11 Using IGM based on Hirshfeld partition of molecular density (IGMH) to study weak
interaction.............................................................................................................................. 708
4.21 Energy decomposition analysis ......................................................................................... 710
4.21.1 Examples of energy decomposition analysis based on forcefield (EDA-FF) ............. 710
4.21.1.1 Example 1: Water dimer ...................................................................................... 710
4.21.1.2 Example 2: Circumcoronene-Cytosine-Guanine trimer ...................................... 714
4.21.2 Shubin Liu's energy decomposition analysis for ethane rotation barrier.................... 720
4.100 Other functions (Part 1) ................................................................................................... 721
xiv
4.100.4 Calculate kinetic energy and nuclear attraction potential energy of phosgene by
numerical integration ............................................................................................................ 721
4.100.8 Perform simple energy decomposition by using combined fragment wavefunctions
............................................................................................................................................... 722
4.100.12 Perform biorthogonalization analysis for orbitals of unrestricted open-shell
wavefunction ......................................................................................................................... 723
4.100.13 Calculate HOMA and Bird aromaticity index for phenanthrene ............................ 727
4.100.14 Calculate LOLIPOP for phenanthrene ................................................................... 728
4.100.15 Calculate intermolecular orbital overlap integral of DB-TTF ................................ 729
4.100.16 Example of easily calculating various quantities defined in the conceptual density
functional theory (CDFT) ..................................................................................................... 730
4.100.16.1 Automatically calculate conceptual density functional theory quantities for
phenol........................................................................................................................................ 730
4.100.16.2 Illustration of studying orbital-weighted Fukui function and orbital-weighted
dual descriptor ........................................................................................................................... 733
4.100.18 Yoshizawa's electron transmission route analysis for phenanthrene ...................... 739
4.100.19 ELF analysis on the promolecular wavefunction combined from fragment
wavefunctions ....................................................................................................................... 740
4.100.21 Calculate molecular diameter and length/width/height .......................................... 742
4.100.22 Analyze π electron character of non-planar system: cycloheptatriene ................... 744
4.200.4 Study iso-chemical shielding surface (ICSS) and magnetic shielding distribution for
benzene ................................................................................................................................. 749
4.200.5 Plot radial distribution function of electron density ................................................. 753
4.200.6 Studying correspondence between orbitals in different wavefunction files ............. 757
4.200.6.1 Revealing relationship between HF and MP2 orbitals of CH3NH2 ................... 757
4.200.6.2 Study contribution of lone pair of nitrogen to MOs of dopamine ..................... 758
4.200.7 Parse output file of “polar” task of Gaussian to obtain (hyper)polarizability and
calculate related quantities .................................................................................................... 760
4.200.8 Studying polarizability and hyperpolarizability based on sum-over-states (SOS)
method................................................................................................................................... 764
4.200.8.1 Calculate polarizability and hyperpolarizability for NH3 .................................. 764
4.200.8.2 Perform two- and three-level model analysis for first hyperpolarizability of
NH2-biphenyl-NO2 .................................................................................................................... 768
4.200.12 Calculate energy index (EI) and bond polarity index (BPI) ................................... 771
4.200.13 Study orbital contributions to density difference ................................................... 772
4.200.13.1 Contribution of MOs to Fukui function f − of phenol ...................................... 772
4.200.13.2 Contribution of NBO orbitals to Fukui function f − of 1,3-butadiene ............. 774
4.200.13.3 Contribution of NBO orbitals to density difference between S0 and S1 states
of H2CO .................................................................................................................................... 776
4.200.14 Application of domain analysis .............................................................................. 778
4.200.14.1 Integrate real space functions within reduced density gradient (RDG)
isosurface to study weak interaction quantitatively .................................................................. 778
4.200.14.2 Visualize molecular cavity and calculate its volume by domain analysis
xv
module....................................................................................................................................... 781
4.200.18 Studying bond length/order alternation (BLA/BOA) as well as alteration of bond
angle and dihedral for specific paths ..................................................................................... 785
4.200.18.1 BLA and BOA of thiophene oligomer............................................................. 785
4.200.18.2 Study variation of bond lengths, bond angles and dihedrals in the ring of
cyclo[18]carbon ........................................................................................................................ 787
4.200.20 Using bond order density and natural adaptive orbital to study chemical bonds ... 789
4.200.20.1 Plot bond order density for N2 molecule ......................................................... 789
4.200.20.2 Study BOD and NAdO orbitals for C-C bonds in butadiene .......................... 790
4.300.1 Visualizing free regions and calculating free volume for a porous system .............. 793
4.300.2 Example of fitting radial atomic density as STOs or GTFs ..................................... 795
4.300.2.1 Crudely fitting radial density of silicon as several STOs .................................. 795
4.300.2.2 Accurately fitting radial density of bromine as many GTFs ............................. 798
4.300.3 Example of using unit sphere representation to visually study (hyper)polarizability
............................................................................................................................................... 799
4.300.3.1 First-order hyperpolarizability of CH3NHCHO ................................................ 800
4.300.3.2 Polarizability and second-order hyperpolarizability of cyclo[18]carbon .......... 802
4.300.4 Example of simulating scanning tunneling microscope (STM) image .................... 804
4.300.4.1 Simulating constant height STM image for phenanthrene ................................ 805
4.300.4.2 Simulating constant current STM image for phenanthrene ............................... 806
4.300.5 Calculate electric dipole moment and multipole moments ...................................... 808
4.A Special topics and advanced tutorials ................................................................................. 809
4.A.1 Study variation of electronic structure along IRC path ............................................... 809
4.A.2 Calculation of spin population .................................................................................... 814
4.A.3 Overview of methods for studying aromaticity ........................................................... 815
4.A.4 Overview of methods for predicting reactive sites ...................................................... 819
4.A.5 Overview of methods for studying weak interactions ................................................. 821
4.A.6 Calculate odd electron density and local electron correlation function ....................... 824
4.A.7 Study (hyper)polarizability density ............................................................................. 827
4.A.8 Analyze wavefunction higher than CCSD level .......................................................... 835
4.A.9 Calculate TrEsp (transition charge from electrostatic potential) charges and analyze
excitonic coupling ................................................................................................................. 836
4.A.10 Intuitively exhibiting atomic properties by coloring atoms....................................... 841
4.A.11 Overview of methods for studying chemical bonds .................................................. 844
4.A.12 Overview of methods for analyzing electron excitation ............................................ 850
4.A.13 Plot electrostatic potential colored van der Waals surface map and penetration graph of
van der Waals surfaces .......................................................................................................... 856
4.A.14 Very easily rendering cube files as state-of-the-art isosurface map via VMD script . 863
4.A.15 Calculating information-theoretic quantities and some relevant quantities............... 866
5 Skills ..................................................................................................867
5.1 Make Multiwfn support more quantum chemistry programs .............................................. 867
5.2 Running Multiwfn in silent mode ....................................................................................... 867
xvi
5.3 Running Multiwfn in batch mode ....................................................................................... 869
5.4 Copy outputs from command-line window to clipboard..................................................... 871
5.5 Make command-line window capable to record more outputs ........................................... 872
5.6 Rapidly load a file into Multiwfn ........................................................................................ 872
5.7 Make use of cubegen utility in Gaussian package to reduce computational time of electrostatic
potential analyses ...................................................................................................................... 873
5.8 Some tips on achieving ideal drawing quality .................................................................... 875
Appendix ................................................................................................876
1 Setting up running environment for Gaussian in Windows.................................................... 876
2 The routines for evaluating real space functions .................................................................... 876
3 Detail of built-in atomic densities .......................................................................................... 878
4 Details about supplying inner-core electron density for the wavefunctions involving pseudo-
potential..................................................................................................................................... 879
5 Check sanity of wavefunction ................................................................................................ 880
6 Special functions .................................................................................................................... 881
xvii
1 Overview
1 Overview
Multiwfn is an extremely powerful wavefunction analysis program, it supports almost all of
the most important wavefunction analysis methods. Multiwfn is free, open-source, high-efficient,
very user-friendly and flexible. Multiwfn can be downloaded at Multiwfn official website
http://sobereva.com/multiwfn. This code is developed by Tian Lu at Beijing Kein Research Center
for Natural Sciences (http://www.keinsci.com).
Input files supported by Multiwfn

Multiwfn accepts many kinds of files for loading wavefunction information: .mwfn (Multiwfn
wavefunction file), .wfn/.wfx (Conventional / Extended PROAIM wavefunction file), .fch
(Gaussian formatted check file), .molden (Molden input file), .31~.40 (NBO plot files) and .gms
(GAMESS-US and Firefly output file). Other types such as Gaussian input and output
files, .cub, .grd, .pdb, .xyz and .mol files can be used for specific functions.
Briefly speaking, Multiwfn can perform wavefunction analyses based on outputted file of
almost all well-known quantum chemistry programs, such as Gaussian, ORCA, GAMESS-US,
Molpro, NWChem, Dalton, xtb, PSI4, Molcas, Q-Chem, MRCC, deMon2k, Firefly, CFour,
Turbomole ...
Special points of Multiwfn

(1) Very comprehensive functions. Almost all of the most important wavefunction analysis
methods have been well supported by Multiwfn.
(2) Extremely user-friendly. Multiwfn is designed as an interactive program (but can also run
silently and be embedded into shell script), prompts shown on screen in each step clearly tell users
what should input next. Multiwfn also never prints obscure messages, therefore there is no any
barrier even for beginners. In addition, all wavefunction analysis theories are very detailedly
documented, and there are more than one hundred of well-written examples in the manual;
furthermore, there is a "quick start" document that guides new users to master common analyses
immediately. Moreover, the developer always very timely and patiently replies all users' questions
in Multiwfn official forum.
(3) High flexiblity. The design of the overall framework, functions and user interface of
Multiwfn is rather flexible, but this does not sacrifice ease of use. Different modules of Multiwfn
are organically integrated together to make numerous analyses that single module cannot realize
feasible
(4) High efficiency. The code of Multiwfn is substantially optimized. Most parts are
parallelized by OpenMP technology. For computationally intensive tasks, the efficiency of Multiwfn
exceeds analogous programs significantly.
(5) Results can be visualized directly. A high-level graphical library DISLIN is invoked
internally and automatically by Multiwfn for visualizing results, most plotting parameters are
controllable in interactive interface. This remarkably simplified wavefunction analysis, especially
for studying distribution of real space functions.
1
1 Overview
Major functions of Multiwfn

PS: Despite that there are so many functions in Multiwfn and the manual is quite thick, you
can easily learn how to realize the functions listed below by checking the "Multiwfn quick start"
document in Multiwfn binary package.
 Showing molecular structure and viewing orbitals (MO, NBO, natural orbital, NTO, LMO,
etc.). Speed of generating orbital graph is extremely fast, and the use is very convenient.
 Outputting all supported real space functions as well as gradient and Hessian at a point.
Value can be decomposed to orbital contributions.
 Calculating real space function along a line and plot curve map.
 Calculating real space function in a plane and plot plane map. Supported graph types
include filled-color map, contour map, relief map (with/without projection), gradient map
and vector field map.
 Calculating real space function in a spatial scope, data can be exported to Gaussian-type
cube grid file (.cub) and can be visualized as isosurface.
 For the calculation of real space functions in one-, two- and three-dimensions, user can
define the operations between the data generated from multiple wavefunction files.
Therefore one can calculate and plot such as Fukui function, dual descriptor and density
difference very easily. Meanwhile promolecule and deformation properties for all real
space functions can be calculated directly.
 Topology analysis for any real space function, such as electron density (AIM analysis),
Laplacian, ELF, LOL, electrostatic potential and so on. Critical points (CPs) can be
located, topology paths and interbasin surfaces can be generated, and then they can be
directly visualized in a 3D GUI window or be plotted in plane map. Value of various real
space functions can be calculated at critical points or along topology paths. CP properties
can be decomposed as orbital contributions.
 Checking and modifying wavefunction. For example, print orbital and basis function
information, manually set orbital occupation number and type, translate and duplicate
system, discard wavefunction information from specified atoms.
 Population analysis. Hirshfeld, Hirshfeld-I, VDD, Mulliken, Löwdin, Modified Mulliken
(including three methods: SCPA, Stout & Politzer, Bickelhaupt), Becke, ADCH (Atomic
dipole moment corrected Hirshfeld), CM5, CHELPG, Merz-Kollmann, RESP (Restrained
ElectroStatic Potential), AIM (Atoms-In-Molecules) and EEM (Electronegativity
Equalization Method) are supported. Electrostatic interaction energy of two given
fragments can be calculated based on atomic charges.
 Orbital composition analysis. Mulliken, Stout & Politzer, SCPA, Hirshfeld, Hirshfeld-I,
Becke, natural atomic orbital (NAO) and AIM methods are supported to obtain orbital
composition. Orbital delocalization index (ODI) can be outputted to quantify extent of
spatial delocalization of orbitals.
 Bond order/strength analysis. Mayer bond order; multi-center bond order in AO or natural
atomic orbital (NAO) basis (up to 12-centers); Wiberg bond order in Löwdin
orthogonalized basis; Mulliken bond order; AV1245 index; intrinsic bond strength index
(IBSI). Mayer and Mulliken bond order can be decomposed to orbital contributions.
Wiberg bond order can be decomposed to contribution from various NAO interactions.
 Plotting total, partial, overlap population density-of-states (TDOS, PDOS, OPDOS) and
2
1 Overview
MO-PDOS. Up to 10 fragments can be very flexibly and conveniently defined. Local

DOS (LDOS) can also be plotted for a point as curve map or for a line as color-filled map.
Furthermore, plotting photoelectron spectrum (PES) based on (generalized) Koopmans'
theorem is fully supported.
 Plotting IR (infrared), normal/pre-resonance Raman, UV-Vis, ECD (electronic circular
dichroism), VCD (vibrational circular dichroism), Raman optical activity (ROA) and
NMR spectra. For vibrational spectra, not only harmonic spectrum, but also anharmonic
fundamental, overtone and combination bands can be plotted. Spin-orbit coupling effect
could be incorporated for electronic spectra. Abundant parameters (broadening function,
FWHM, scale factor, etc.) can be customized by users. Maxima and minima of the
spectrum can be printed and directly labelled on the map. Total spectrum can be
decomposed to individual contribution of each transition. Spectrum of multiple systems
can be conveniently plotted together. Plotting conformational weighted spectrum can be
very easily realized. Spikes can be added to bottom of the graph to indicate position and
degeneracy of transition level.
 Quantitative analysis of molecular surface. Surface properties such as
(total/positive/negative/polar/nonpolar) surface area, enclosed volume, average value and
std. of mapped functions can be computed for overall molecular surface or local surfaces;
Various kinds of GIPF descriptors and molecular polarity index can be evaluated; minima
and maxima of mapped functions on the surface can be located; area of characteristic
region corresponding to e.g. /-hole and lone pair can be calculated based on ESP;
Basin-like analysis on molecular surface for arbitrary mapped function can be realized.
 Processing grid data (can be loaded from .cub/.grd or generated by Multiwfn). User can
perform mathematical operations on grid data, set value in certain range, extract data in
specified plane, plot integral curve, etc.
 Adaptive natural density partitioning (AdNDP) analysis. The interface is interactive and
the AdNDP orbitals can be visualized directly. Energy and orbital composition of AdNDP
orbitals can be obtained.
 Analyzing real space functions in fuzzy atomic spaces (defined by Becke, Hirshfeld or
Hirshfeld-I partitions). These quantities can be computed: Integral of real space functions
in atomic spaces or in overlap region between atomic spaces, dipole and multipole
moments of atom/fragment/molecule, atomic multipole moments, atomic overlap matrix
(AOM), localization and delocalization indices (LI, DI), condensed linear response kernel,
multi-center DI, as well as five aromaticity indices, namely FLU, FLU-, PDI, PLR and
information-theoretic index.
 Charge decomposition analysis (CDA) and extended CDA analysis. Orbital interaction
diagram can be plotted. Infinite number of fragments can be defined.
 Basin analysis. Attractors can be located for any real space function, corresponding basins
can be generated and visualized at the same time. All real space functions can be
integrated in the generated basins. Electric multipole moments, basin/atomic overlap
matrix (BOM/AOM), localization index (LI) and delocalization index (DI) can be
calculated for the basins. Atomic contribution to basin population can be obtained. High
ELF localization domain population and volume (HELP and HELV) and be evaluated.
 Electron excitation analyses: Visualizing and analyzing hole-electron distribution,
3
1 Overview
transition density, transition electric/magnetic dipole moment and charge density

difference; calculating Coulomb attractive energy between hole and electron (exciton
binding energy); calculating Mulliken atomic transition charges and TrEsp (transition
charge from electrostatic potential); decomposing transition dipole moment to MO pair
contribution or basis function/atom contribution; analyzing charge-transfer by the method
proposed in JCTC, 7, 2498; plotting atom/fragment transition density matrix, transition
dipole moment matrix and charge transfer matrix as heat maps; calculating r index
(JCTC, 9, 3118) and  index (JCP, 128, 044118) to reveal electron excitation character;
calculating transition dipole moments between excited states; generating natural transition
orbitals (NTOs); calculating ghost-hunter index (JCC, 38, 2151); calculating amount of
interfragment charge transfer via IFCT method; generating natural orbitals for a batch of
excited states; quickly examining major MO transitions in all excited states.
 Orbital localization analysis: Pipek-Mezey (based on Mulliken or Löwdin population) and
Foster-Boys localization methods are supported. Composition, energy and dipole moment
of the resulting LMOs can be derived, shape and center of the LMOs can be easily
visualized. Furthermore, based on the LMOs, oxidation states can be evaluated via LOBA
method (PCCP, 11, 11297).
 Visual study of weak interaction: RDG/NCI method (JACS, 132, 6498); aNCI method
(noncovalent interaction analysis in fluctuating environments, JCTC, 9, 2226); DORI
method (JCTC, 10, 3745); independent gradient model (IGM) method (PCCP, 19, 17928);
IGM based on Hirshfeld partition of molecular density (IGMH); Interaction region
indicator (IRI). Individual regions enclosed by isosurface of these real space functions can
be integrated for quantitative analysis. Becke and Hirshfeld surface analyses, as well as
fingerprint analysis are also supported. Van der Waals potential (DOI:
10.26434/chemrxiv.12148572) can be visualized.
 Energy decomposition analysis (EDA): EDA based on UFF/AMBER/GAFF molecular
force fields (EDA-FF); Simple energy decomposition (relies on Gaussian); Shubin Liu's
energy decomposition (EDA-SBL)
 Other functions (incomplete list): Automatic calculation of all quantities involved in
conceptual density functional theory; integrating a real space function over the whole
space by Becke's multi-center method; evaluating overlap integral between alpha and beta
orbitals; evaluating overlap and centroid distance between two orbitals; generating
Gaussian input file with initial guess combined from multiple fragment wavefunctions;
calculating van der Waals volume; calculating HOMA and Bird aromaticity indices;
calculating LOLIPOP index; calculating intermolecular orbital overlap; Yoshizawa's
electron transport route analysis; calculating atomic and bond dipole moment in Hilbert
space; plotting radial distribution function for real space functions; plotting iso-chemical
shielding surface (ICSS); calculating overlap integral between orbitals in two different
wavefunctions; parsing output of (hyper)polarizability task of Gaussian and calculate
numerous related quantites; studying polarizability and 1st/2nd/3rd hyperpolarizability by
sum-over-states (SOS) method; outputting various kinds of integrals between orbitals;
evaluating the first and second moments and radius of gyration for a real space function;
exporting loaded structure/wavefunction to many popular formats such
as .wfn, .wfx, .molden, .fch, NBO .47, .pdb, .xyz and yield input file for a lot of known
4
1 Overview
quantum chemistry codes; calculating bond polarity index (BPI); domain analysis
(obtaining properties within isosurfaces defined by a real space function); calculating
electron correlation indices; detecting  orbitals and evaluating orbital  composition;
evaluating molecular diameter and length/width/height; perform biorthogonalization
between alpha and beta orbitals to maximally pair them; evaluating interatomic
connectivity and atom coordination numbers based on geometry; evaluating core-valence
bifurcation (CVB) index; evaluating orbital contributions to density difference (e.g. Fukui
function) or other kind of grid data; calculating bond length/order alternation (BLA/BOA)
as well as bond angle and dihedral alterations; bond order density (BOD) and natural
adaptive orbital (NAdO) analyses; viewing free regions and calculating free volume in a
box; fitting atomic radial density as STOs or GTFs; unit sphere and vector representation
of (hyper)polarizability tensor; simulating scanning tunneling microscope (STM) image;
evaluating electric dipole/quadrupole/octopole moments, etc.
Real space functions supported by Multiwfn

Real space function analysis is one of the most powerful features of Multiwfn, more than 100
real space functions are supported and listed below, detailed descriptions can be found in Section
2.6 and 2.7 of the manual:
• Electron density
• Gradient norm of electron density
• Laplacian of electron density
• Value of orbital wavefunction
• Electron spin density
• Hamiltonian kinetic energy density K(r)
• Lagrangian kinetic energy density G(r)
• Electrostatic potential from nuclear / atomic charges
• Electron localization function (ELF) defined by Becke and the one defined by Tsirelson
• Localized orbital locator (LOL) defined by Becke and the one defined by Tsirelson
• Local information entropy
• Electrostatic potential (ESP)
• Reduced density gradient (RDG) with/without promolecular approximation
• sign(2) (product of the sign of the second largest eigenvalue of electron density Hessian
matrix and electron density) with/without promolecular approximation
• Exchange-correlation density, correlation hole and correlation factor
• Average local ionization energy
• Source function
• Interaction region indicator (IRI)
• Electron delocalization range function EDR(r;d) and orbital overlap distance function D(r)
• The g function defined in Independent Gradient Model (IGM)
Others (incomplete list): potential energy density, electron energy density, strong covalent
interaction index (SCI), shape function, local temperature, bond metallicity, linear response kernel,
local electron affinity/electronegativity/hardness, ellipticity of electron density, eta index, on-top
pair density, numerous forms of DFT exchange-correlation potential, numerous forms of DFT
kinetic energy density, Weizsäcker potential, Fisher information entropy, Ghosh/Shannon entropy
5
1 Overview
density, integrand of Rényi entropy, steric energy/potential/charge, Pauli potential/force/charge,

quantum potential/force/charge, PAEM, density overlap regions indicator (DORI), region of slow
electrons (RoSE), PS-FID, single exponential decay detector (SEDD), electron linear momentum
density, electric/magnetic dipole moment density, local electron correlation function, magnitude of
electric field, orbital-weighted Fukui function and dual descriptor, van der Waals potential.
Implementing a new real space function into Multiwfn is extremely easy, as illustrated in
Section 2.7 of the manual.
Things that Multiwfn can do

The analyses that Multiwfn support for different topics are briefly listed below, you can easily
find related manual sections by searching "Multiwfn quick start.pdf" document. Do not forget to
ask question in Multiwfn official forum when you are confused!
• Visualizing various kinds of orbitals generated by various programs in various forms
• Characterizing chemical bonds: Various form of AIM analyses; studying real space functions
(ELF, LOL, 2, kinetic/potential energy density, valence , fragment density difference,
deformation density, source function, bond ellipticity, bond degree, eta index, V(r)/G(r), SCI, DORI,
PAEM, IGM...); various kinds of bond orders analysis (Mayer, Laplacian, Mulliken, Wiberg, Fuzzy
and multi-center bond orders, as well as decomposition analysis for Mayer, Mulliken and Wiberg
bond orders); intrinsic bond strength index (IBSI); localization/delocalization index; orbital
localization analysis; bond order density (BOD) and natural adaptive orbital (NAdO) analyses;
various methods of measuring bond polarity and bond dipole moment; charge decomposition
analysis (CDA); overlap population density-of-states (OPDOS); energy decomposition analysis and
so on. See Section 4.A.11 of manual for an overview. Variation of various properties of chemical
bonds during scan and IRC processes can also be easily studied via shell scripts, see Section 4.A.1
of manual.
• Characterizing electron distribution and variation: Atomic charges (AIM, Mulliken, SCPA,
Hirshfeld, Hirshfeld-I, Voronoi, Löwdin, ADCH, CM5, EEM, CHELPG, MK, RESP, RESP2... );
total and spin population analyses for basis functions/shells/atoms/fragments; atomic dipole and
multipole moment analysis; plotting / basin analysis / domain analysis for density difference; charge
displacement curve
• Aromaticity and electron delocalization analyses: ICSS; AdNDP; ELF-/; LOL-/;
HOMA; Bird; multi-center bond order; AV1245 and AVmin; NICS; Shannon aromaticity; FLU and
FLU-; PDI; ATI; PLR; ∆DI; density curvature perpendicular to ring plane and so on. See Section
4.A.3 of manual for an overview
• Characterizing intramolecular and intermolecular weak interactions: AIM analysis (bond path
visualization and analysis of various properies at bond critical point); visual analyses of weak
interactions (NCI, IGM, IGMH, IRI, DORI); atom and atom pair g indices based on IGM or IGMH;
quantitative molecular surface analysis for electrostatic potential (ESP); plotting ESP in various
form; plotting van der Waals potential; energy decomposition analysis based on forcefield (EDA-
FF); Hirshfeld/Becke surface analysis; LOLIPOP; mutual penetration distance and penetration
volume analysis; atomic charge and multipole moment analysis; charge transfer analysis (density
difference map, CDA, variation of population ...); ELF and core-valence bifurcation (CVB) index
and so on. See Section 4.A.5 of manual for an overview
• Electron excitation analysis: Analysis of hole and electron (distribution,
6
1 Overview
atom/fragment/orbital contribution, centroid position, displacement and overlap, exciton binding

energy); charge transfer analysis (IFCT, density difference...); NTO; overlap and centroid distance
between crucial MOs; plotting atom/fragment transition density matrix and charge transfer matrix;
∆r index; decomposition of transition dipole moment to basis function/atom/fragment/MO pair
contributions; transition dipole moment between various excited states; transition atomic charge;
ghost-hunter index; revealing variation of electronic structure (bonding and population) during
excitation; printing major MO transitions in all excited states and so on. See Section 4.A.12 of
manual for an overview
• Prediction of reactive sites and reactivity analysis: ESP and ALIE analyses on molecular
surface; atomic charges; orbital composition analysis for frontier molecular orbitals; population of
 electron; orbital overlap distance function analysis; automatically calculating all important
quantities defined in the framework of conceptual density functional theory (Fukui function and
dual descriptor as well as their condensed form and orbital-weighted variants, Mulliken
electronegativity, hardness, electrophilicity and nucleophilicity index, softness, condensed local
softness, relative electrophilicity and nucleophilicity, etc.); evaluating contribution of orbitals (MO,
NBO, NAO, etc.) to Fukui function. See Section 4.A.4 of manual for an overview
• Prediction properties of molecular condensed phase: Using ESP distribution on vdW surface
to empirically predict heat of vaporization, heat of sublimation, density of molecular crystal, boiling
point, heat of fusion, surface tension, pKb and so on. Molecular polarity can be quantified. See
Section 3.15.1 of manual
• Plotting spectra: IR, Raman, UV-Vis, ECD, VCD, ROA, NMR and photoelectron spectra
• Characterizing molecular structure: Evaluating molecular volume, surface area,
length/height/weight, diameter, interatomic connectivity and atomic coordination number, average
bond length of atomic cluster, cavity volume, bond length alternation (BLA) as well as bond angle
and dihedral alternations, kinetic diameter and so on. Free regions (pores) in a box can be visualized
and volume can be evaluated
• (Hyper)polarizability study: Parsing Gaussian output file of "polar" task and calculating many
data related to (hyper)polarizability; Calculating quantities related to Hyper-Rayleigh scattering
(HRS); plotting (hyper)polarizability density; obtaining atomic contribution to (hyper)polarizability;
calculating (hyper)polarizability by means of sum-over-states (SOS) method; two-level and three-
level model analyses; unit sphere and vector representation of (hyper)polarizability tensor
• Electric conduction analysis: TDOS and PDOS; orbital overlap analysis between
neighbouring monomers; Yoshizawa's transport route analysis; bond length/order alternation
(BLA/BOA)
• Many others: Teaching structure chemistry; simulating scanning tunneling microscope (STM)
image; converting file formats containing geometry or wavefunction information; studying electron
correlation effect; realizing ELF-tuning and LOL-tuning for DFT functionals; evaluating oxidation
state by LOBA method; studying distribution of real space functions (in terms of radial distribution
function, centroid, first and second moments, integral over whole space and local region...);
evaluating  or  component in molecular orbitals and so on
Citing Multiwfn
If Multiwfn is used in your research, at least this paper must be cited:
7
1 Overview
Tian Lu, Feiwu Chen, Multiwfn: A Multifunctional Wavefunction Analyzer, J. Comput.

Chem. 33, 580-592 (2012)
Other my papers should be cited according to the method and function employed in your work.
Please carefully check How to cite Multiwfn.pdf document in the Multiwfn binary package
Discussion zones
There are two Multiwfn official forums, with different languages. You can discuss anything
about Multiwfn and wavefunction analysis in either one. If you encountered problems in using
Multiwfn, please do not hesitate to post topic on these forums!
Multiwfn English forum: http://sobereva.com/wfnbbs
Multiwfn Chinese forum: http://bbs.keinsci.com/wfn
BTW: Multiwfn Youtube channel contains many valuable illustration videos of Multiwfn, it is
highly recommended to look at them and subscribe this channel
8
2 General information
2.1 Install
2.1.1 Windows version
What you need to do is just uncompressing the program package, then you can start to use by
double-clicking the icon.
A few functions in Multiwfn rely on Gaussian, if you need to carry out these analyses, you
need to setup environment variables for Gaussian manually, see Appendix 1.
It is strongly suggested to set "nthreads" in settings.ini to actual number of CPU physical cores
of your machine, so that all computing power of your CPU could be utilized during calculation. See
Section 2.4 for more detail.
If you want to make Multiwfn able to directly open .chk file produced by Gaussian, set
"formchkpath" in settings.ini to actual path of formchk executable file in Gaussian package.
2.1.2 Linux version
• Uncompress the Multiwfn binary package

• Make sure that you have installed motif package, which provides libXm.so.4, full version of
Multiwfn cannot boot up without this file. The motif is freely available at
http://motif.ics.com/motif/downloads. If you are a CentOS or Red Hat Linux user and have not
installed motif, you can directly run yum install motif to install it; alternatively, you can download
corresponding rpm package (e.g. motif-2.3.4-1.x86_64.rpm) and manually install it; If you are an
Ubuntu user, you can run sudo apt-get install libmotif4 to install it, or download deb package (e.g.
libmotif4_2.3.4-1_amd64.deb) and manually install it.
• Add below lines to ~/.bashrc file (using e.g. vi ~/.bashrc command)
export KMP_STACKSIZE=200M
ulimit -s unlimited
These lines mean removing limitation on stacksize memory, and defining stacksize of 200MB
for each thread during parallel calculations, see Section 2.4 for detail.
• Run /sbin/sysctl -a|grep shmmax to check if the size of SysV shared memory segments is
large enough (unit is in bytes); if the value is too small, Multiwfn may crash when analyzing big
wavefunction. To enlarge the size, for example you can add kernel.shmmax = 512000000 to
/etc/sysctl.conf and reboot system, then the upper limit will be enlarged to about 512MB.
• Assume that you are using Bash shell, and you have decompressed the Multiwfn package as
“/sob/Multiwfn_3.6_bin_Linux” folder, you should add below lines into ~/.bashrc file:
export Multiwfnpath=/sob/Multiwfn_3.6_bin_Linux
export PATH=$PATH:/sob/Multiwfn_3.6_bin_Linux
• Run below command to add executable permission to Multiwfn executable file:

chmod +x /sob/Multiwfn_3.6_bin_Linux/Multiwfn
9
• Configure the settings.ini file in Multiwfn folder in the same way as described in last Section
After re-entering the terminal, you can boot up Multiwfn anywhere by simply running Multiwfn.
Linux version of Multiwfn works well on Red Hat Enterprise Linux 6 & 7, CentOS 6 & 7 and Ubuntu 12/14/16.
I can not guarantee that the program is completely compatible with all other Linux distributions. If system prompts
you that some dynamical link libraries (.so files) are missing when booting up Multiwfn, try to find and install the
packages which contain corresponding .so files.
If you encounter difficulty when running/compiling Multiwfn due to missing or incompatibility of some
graphics related library files, and meantime you do not need any visualization function of Multiwfn, you can
run/compile Multiwfn without GUI supported, all functions irrelevant to GUI and map plotting will still work
normally. Please check document of compilation method in source code package on how to compile this version, the
pre-compiled binary of this version can also be downloaded from Multiwfn website (termed as "noGUI" version).
2.1.3 Mac OS version
If you are using relatively old Mac OS version, e.g. OS X 10.8, simply follow the instruction
given in this section. If you intend to install Multiwfn in relatively new Mac OS systems, such as
OS X 10.11 EI Capitan and MacOS 10.12 Sierra, please follow steps in this page:
https://wiki.ch.ic.ac.uk/wiki/index.php?title=Mod:multiwfn. Since I am not a Mac user, I am sorry
that I am unable to provide much help if you encounter difficulty in Mac platform.
• Uncompress the program package to e.g. /Users/sob/Multiwfn. Notice that the path
(including file name) should less than 80 characters.
• Download .dmg file of Mac OS version of motif from http://www.ist-
inc.com/downloads/motif_download.html and then install it. The motif package I installed is
openmotif-compat-2.1.32_IST.macosx10.5.dmg.
If your system does not natively support X11 (i.e. OS X Mountain Lion), you should download
XQuartz from http://xquartz.macosforge.org/landing and install it.
• Add below sentense to your .profile file (e.g. /Users/sob/.profile) to make them take effect
automatically, then reboot your terminal. If the .profile is unexisting, you should create it manually.
export KMP_STACKSIZE=64000000
KMP_STACKSIZE defines stacksize (in bytes) for each thread in parallel implementation, see
Section 2.4 for detail.
• Run sysctl -a|grep shmmax to check if the size of SysV shared memory segments is large
enough (unit is in bytes), if the value is too small, Multiwfn may crashes when analyzing big
wavefunction. In order to enlarge the size, you should edit or create the file /etc/sysctl.conf, and add
kern.sysv.shmmax = 512000000 to it and reboot system, then the upper limit will be enlarged to
about 512MB.
• Move the libdislin_d.11.dylib in the Multiwfn package to /usr/local/lib folder.
• Set Multiwfnpath environment variable if needed, see point 5 of Section 2.1.2.
• Configure the settings.ini file in the same way as described in Section 2.1.1.
2.2 Using Multiwfn
Using Multiwfn is very easy, by simply reading the prompts shown on screen, you will know
what should input next. If you get stuck, please read corresponding section carefully in Chapter 3
or corresponding tutorials in Chapter 4.
In Windows, usually Multiwfn is booted up by directly double-clicking the icon of the
10
executable file, then you should input the path of the file to be loaded. You can also boot up Multiwfn
via command line, and the same time the path of input file may be also given, for example you can
run Multiwfn /sob/test.wfn.
If the input file is in current directory, you can input file name without the path of directory. If
the input file is just the one that last time used, you can simply input the letter o after entering
Multiwfn (the path of the input file successfully read at last time was recorded in settings.ini). If the
input file is in the same folder as the one last time used, for convenience the path can be replaced
by symbol ?. For example, last time you loaded C:\sob\wives\K-ON\Mio.wfn, this time you can
simply input ?Azusa.fch to load C:\sob\wives\K-ON\Azusa.fch. If you preferred to choose input file
in GUI window, you can directly press ENTER button after entering Multiwfn, then a GUI window
will be shown for selecting input file.
You can press CTRL+C or click “×” button at right-top of Multiwfn window any time to exit
Multiwfn, but a more graceful way of exiting Multiwfn is inputting q in main menu. When graphical
window is showing on screen, you can click “RETURN” button to close the window, if there is no
such button, clicking right mouse button on the graph to close it.
If you want to load another file into Multiwfn, you can reboot Multiwfn or start a new Multiwfn
instance. Alternatively, in main menu you can input r to initialize Multiwfn and load a new file, at
the meantime the settings.ini will also be reloaded. However, please notice that the safest way of
loading a new file is rebooting Multiwfn.
Multiwfn can also run via silent mode instead of interactive mode, by which users do not need
to press any keyboard button during running. This is useful for batch processing, please consult
Sections 5.2 and 5.3.
Supported arguments
For convenience, there are a few arguments may be added when running Multiwfn via
command line:
• -nt: Number of threads for parallel calculation
• -uf: Index of user-defined function
• -silent: Run Multiwfn in silent mode
• -set: Path of settings.ini
For example
Multiwfn COCl2.fch -nt 36 -set /sob/tmp/settings.ini -silent
The priorities of these arguments are higher than those in settings.ini.
2.3 Files of Multiwfn
You will find following files after uncompressing Multiwfn package, only the bolded files are
indispensable for running Multiwfn:
• Multiwfn.exe (Windows) or Multiwfn (Linux/Mac OS) : The executable file of Multiwfn.
• libiomp5md.dll (Windows) : Intel OpenMP Runtime library.
• settings.ini : All detail parameters for running Multiwfn are recorded here, most of them do
not need to be frequently modified. When booting up, Multiwfn will try to find and use this file in
current folder, if it is not presented in current folder, the file in the path defined by "Multiwfnpath"
11
environment variable will be used; if the file still cannot be found, default settings will be used
instead. If you run Multiwfn via command line, you can also directly specify position of this file via
"-set" argument, for example: Multiwfn test.wfn -set /sob/3.7/settings.ini.
The meanings of all parameters in settings.ini are not documented in this manual systematically,
since they have already been commented in detail, only those important will be mentioned in this
manual. I suggest you read through the settings.ini and find out the ones useful for you.
• “examples” folder : Some useful files, scripts and the files involved in the examples of
Chapter 4.
• LICENSE.txt : The terms that all users must follow.
• Multiwfn quick start.pdf : A short document lets new user immediately understand how to
use Multiwfn to carry out very common tasks.
• How to cite Multiwfn.pdf : Please properly cite Multiwfn according to this document.
2.4 Parallel implementation
Most time-consuming codes of Multiwfn have been parallelized by OpenMP. If your computer
is SMP architecture, you can greatly benefit from parallelization (For grid calculation, the speed up
ratio versus the number of CPU cores is nearly linear). To enable parallel mode, just modify
“nthreads” parameter in settings.ini to your situation. For example, your computer have a physical
12-cores CPU installed, then change “nthreads” to 12.
If Multiwfn crashes during parallel calculation for very large system, try to enlarge
“ompstacksize” in settings.ini (for Windows version) or enlarge the value of the environment
variable KMP_STACKSIZE (for Linux or Mac OS version).
2.5 Input files and wavefunction types
Wavefunction types supported by Multiwfn include restricted/unrestricted single-determinant

wavefunction, restricted open-shell wavefunction and post-HF wavefunction (in natural orbital
formalism).
For basis function, Cartesian or spherical harmonic Gaussian functions with angular moment
up to h are supported.
There is no upper limit of the number of atoms / basis functions / GTFs / orbitals in Multiwfn,
the actual upper limit is only decided by available memory on your computer.
Multiwfn determines the input file type by file extension. Notice that different function need
different types of information, you should choose proper type of input file, see the table below. For
example, the wavefunction represented by GTFs is enough for Hirshfeld population, so you can
use .mwfn/.fch/.molden/.gms/.31~.40/.wfn/.wfx file as input, but .pdb, .xyz, .mol, etc. do not carry
any wavefunction information hence cannot be used. While generating grid data of RDG function
with promolecular approximation only requires atom coordinates, so all supported file formats can
be used (except for plain text file). The requirement on information types by each function is
commonly described at the end of corresponding section in Chapter 3 by red text.
12
Contained information types

File Format
Basis Atom Grid Atomic
GTFs
functions coordinates data charges
.fch/.fchk/.chk     
.mwfn .molden .gbw     
.gms     
NBO .31~.40     
.wfn and .wfx     
.pdb .xyz .mol .mol2 .gjf .gro     
.chg and .pqr     
.cub/.cube     
.vti, DMol3 .grd, .dx     
Plain text file     
Multiwfn wavefunction file (.mwfn) : This format is defined and supported since Multiwfn
3.7. This is the most ideal format for wavefunction storage and exchange purposes. This file records
all information for wavefunction analysis in a strict, concise, compact and extensible format. The
introduction and definition of this format is given in my paper: ChemRxiv (2020) DOI:
10.26434/chemrxiv.11872524.
AIM wavefunction file (.wfn) : This format was first introduced by Bader’s AIMPAC program,
and currently supported by a lot of mainstream quantum chemistry softwares, such as Gaussian,
ORCA, GAMESS-US/UK, Firefly, Q-Chem and NWChem. The information in .wfn file include
atomic coordinates, elements, orbital energies, occupation numbers, expansion coefficients of
Cartesian Gaussian type functions (GTF). Supported angular momentum of GTF is up to f. The wfn
file does not contain any virtual orbital. The generation method of .wfn file is documented at the
beginning of Chapter 4.
Note: Although GTFs with angular moment of g and h are not formally supported by original .wfn format, if g
and h-type GTFs are recorded in following manner, then Multiwfn is able to recognize them: 21~35 in "TYPE
ASSIGNMENT" correspond to YZZZ, XYYY, XXYY, XYZZ, YZZZ, XYYZ, XXXX, XXXY, XZZZ, XXYZ,
XXXZ, XXZZ, YYYY, YYYZ, ZZZZ, respectively. 36~56 correspond to ZZZZZ, YZZZZ, YYZZZ, YYYZZ,
YYYYZ, YYYYY, XZZZZ, XYZZZ, XYYZZ, XYYYZ, XYYYY, XXZZZ, XXYZZ, XXYYZ, XXYYY, XXXZZ,
XXXYZ, XXXYY, XXXXZ, XXXXY, XXXXX, respectively. The sequence shown here in fact is also the sequence
used in the .wfn outputted by Molden2AIM and Gaussian09 since B.01.
AIM extended wavefunction files (.wfx) : This is a format introduced as an extension of .wfn,
and it was supported by Gaussian 09 since B.01 revision. Relative to .wfn format, .wfx supports
higher data record precision and infinitely high GTF angular moment. The most special point of this
format is the newly added electron density function (EDF) field, that is using multiple GTFs to
represent inner core electron density of the wavefunction in which effective core potential (ECP) is
used. Due to this, the result of electron density analysis for the wavefunction using ECP is nearly
identical to that for full electron wavefunction. Currently the real space functions supported EDFs
in Multiwfn include: electron density, its gradient and Laplacian, local information entropy, reduced
density graident as well as Sign(λ2(r)). Meanwhile topology analyses of electron density and its
13
Laplacian also take into account EDFs. Notice that EDF informations have neither effect on ESP
nor the real space functions that relied on wavefunction (e.g. kinetic energy density, ELF). If you
you want to analyze these properties for heavy elements, you should use full-electrons basis sets, at
lease small-core ECP. Currently the only supported GTF type in EDF field is S-type (actually S-
type is enough for fitting inner density, since which is nearly spherical symmetry). Like .wfn,
Multiwfn does not allow virtual orbitals presented in .wfx file.
Multiwfn has a powerful built-in EDF library, taken from Molden2aim program developed by
Wenli Zou. As long as the input file contains GTF information (e.g. .fch, .wfn, .molden, .gms...),
Multiwfn always automatically loads EDF information from this library for the atoms using
pseudopotential. Only when you use .wfx file as input and the .wfx itself already contains EDF field,
the EDF information will be loaded from the .wfx file rather than from EDF library. See Appendix
4 for more details.
Note that although some programs other than Gaussian can also generate .wfx file (e.g. ORCA),
these .wfx files are unable to provide EDF field.
Notice: For certain version of Gaussian (e.g. G09 B.01), I found that the EDF field recorded in .wfx for rare
cases is problematic, namely the number of electrons represented by EDF field is unequal to the actual number of
core electrons exhibited by ECP. In order to verify if the EDF field is correct, you can use subfunction 4 in main
function 100 to obtain the integral of total electron density over the whole space, if the result is approximately equal
to the total number of electrons (core+valence electrons), that means the EDF field is correct.
Gaussian formatted checkpoint file (.fch/.fchk): Checkpoint file of Gaussian program (.chk)
can be converted to formatted checkpoint file (.fch/.fchk) via formchk utility in Gaussian package.
There is no any difference between .fch and .fchk. "fch" ("fchk") is the default extension generated
by Windows (Linux) version of formchk.
If you want to make Multiwfn able to directly load .chk file, you must set "formchkpath" in
settings.ini to actual path of formchk executable file in Gaussian package. In this case Multiwfn will
automatically invoke formchk to convert .chk file to .fch/fchk file, and if conversion is successful,
the .fch/fchk will be loaded and then be automatically deleted once loading is finished.
.fch/.fchk contains richer information than .wfn/.wfx files, virtual orbital wavefunctions are
also recorded, and meanwhile it provides basis function information for Multiwfn. If you want to
use .fch/.fchk file as carrier for post-HF wavefunction, read the beginning of Chapter 4 carefully!
Notice that for single-determinant wavefunctions, before some calculations involving real
space function (e.g. main function 2, 4, 5), virtual orbitals higher than LUMO+10 will be deleted
automatically to speed up calculations, therefore you cannot analyze those orbitals after the
calculation, unless you reboot Multiwfn. If you want to disable this treatment, set "idelvirorb" in
settings.ini to 0.
The .fchk file generated by Q-Chem and PSI4 can also be used as input file of Multiwfn. (If
the .fchk file was generated by relatively old version of Q-Chem, you must set “ifchprog” in
settings.ini to 2. You do not need to do this if your Q-Chem version is equal or newer than 5.0).
Molden input file (.molden or .molden.input or molden.inp) : Currently, a wide variety of

quantum chemistry packages, such as Molpro, Molcas, ORCA, Q-Chem, CFour, Turbomole, PSI4,
MRCC and NWChem are able to produce input file of Molden visualization program. This type of
file records atomic coordinates, basis set definition, information of all occupied and virtual orbitals
(including expansion coefficient of basis functions, occupation number, spin, energy and symmetry),
meanwhile there is no information only specific for Molden. So in fact, Molden input file can be
14
regarded as a standard and general file format for exchanging wavefunction information. For
Multiwfn, this type of file can provide atomic coordinate, basis function information and GTF
information.
Beware that the Molden input files produced by a lot of program are quite non-standard!
Currently Multiwfn only formally supports the Molden input file generated by Molpro, ORCA, xtb,
Dalton, NWChem (only for spherical harmonic functions and meantime symmetry is disabled),
MRCC (only for spherical harmonic functions), deMon2k, BDF. If the Molden input file you used
is generated by other programs, the analysis result may or may not be correct, you should first use
the methods described in Appendix 5 to check if the wavefunction has been correctly loaded.
Hint: Multiwfn fully supports the Molden input file standardized by molden2aim utility (see Section 5.1 for
detail), which is able to properly recognize Molden input files generated by many other quantum chemistry codes,
such as CFOUR and Molcas.
.molden file only formally supports basis functions up to g angular moment. However,
subfunction 2 of main function 100 can generate .molden file containing h functions, and Multiwfn
can then normally load it. Multiwfn can also normally load .molden file generated by ORCA and
Dalton even if h functions are presented.
Although Molden input file also supports Slater type orbital (STO), Multiwfn can only utilize
the Molden input file recording Gaussian type basis functions.
One drawback of Molden input file is that it does not explicitly record nuclear charges as other
formats such as wfn and fch, therefore the results relying on nuclear charges (e.g. electrostatic
potential) will be problematic when pseudo-potential is used. To address this problem, Multiwfn
loads atomic indices in the file (i.e. the third column in [Atoms] field) as nuclear charges, thus if
you manually change the atomic index to the number of atomic valence electrons that explicitly
represented in the quantum chemistry calculation, then the result will be correct. If you are confused
about this point, please check this post: http://sobereva.com/wfnbbs/viewtopic.php?pid=721.
A disadvantage of using Molden input file as wavefunction carrier is that its format is not as
compact as .fch. Due to this reason, for the same wavefunction, loading speed of .molden file is
much slower than .fch. Therefore, if you need to frequently analyze a .molden file, I suggest you
use subfunction 2 of main function 100 to convert it to .fch format.
In analogy to .fch file, Multiwfn may delete virtual orbitals higher than LUMO+10. To avoid
this, you should set "idelvirorb" in settings.ini to 0.
The way of generating Molden input file by some quantum chemistry programs are described
at the beginning of Chapter 4. If you are an ORCA user and you do not want to manually convert
the .gbw file to Molden input file via the orca_2mkl utility in ORCA, you can set "orca_2mklpath"
in settings.ini to actual path of orca_2mkl executable file in ORCA folder, then Multiwfn will be
able to directly load .gbw file.
PS: Detailed description about .molden format can be found on Molden official site:
http://www.cmbi.ru.nl/molden/molden_format.html.
GAMESS-US or Firefly output file (.gms): If you want to use GAMESS-US or Firefly
(originally known as PC-GAMESS) output file as input file, you can change its suffix as .gms, then
Multiwfn will properly recognize it. Currently, I can only guarantee that output file of
HF/DFT/TDDFT calculation with default NPRINT option can be normally loaded by Multiwfn. If
the point group is not C1, Multiwfn will be unable to deal with the output file.
The role of .gms is similar as .molden and .fch file, i.e. all of them provide atomic coordinates,
15
GTF and basis function informations.

Since I am not a experienced Firefly user, I cannot guarantee that the compatibility with Firefly output files is
as good as GAMESS-US output files. For the former I only tested DFT single point task and TDDFT task.
Plot files of NBO program (.31~.40): The main purpose of supporting these file types is for
visualizing PNAO/NAO/PNHO/NHO/PNBO/NBO/PNLMO/NLMO/MO (their orbital coefficients
are recorded in .32~.40 respectively), .31 recorded basis function information. After boot up
Multiwfn, you should input the path of .31 file first, and then input the path of one of .32~.40 files
(for simplicity, you can only input the suffix when the filenames are identical).
Notice that among all types of the orbitals generated by NBO program, only using NBO or
NLMO to calculate real space functions is meaningful!
Protein data bank format (.pdb), .xyz, MDL Molfile (.mol), .mol2: These are the most
widely used formats for recording atom coordinates. They do not carry any wavefunction
information, but for the functions which only require atom coordinates, using these kind of files as
input is adequate. An advantage of .mol and .mol2 with respect to .pdb and .xyz is that they contain
atomic connectivity table, which is need by a few functions of Multiwfn, e.g. Calculation of EEM
atomic charges. If .xyz file contains multiple frames, only the first frame will be loaded.
Notice that the .mol file supported by Multiwfn is V2000 version, both the maximum number
of atoms and bonds that can be recorded are 999. More description about .mol format can be found
in https://en.wikipedia.org/wiki/Chemical_table_file.
Gaussian job file (.gjf): This means input file of Gaussian program, it can provide atom
coordinate information as well as the number of  and  electrons information to Multiwfn. Note
that the atoms must be recorded as Cartesian coordinate.
Charge files (.chg): This type of plain text file can be generated by some functions of Multiwfn
(e.g. population analysis functions), it contains element names (less than or equal to two characters),
atom coordinates (first three columns, in Å) and charges (the fourth column), users can modify them
manually. This file is free-formatted, all fields must be delimited by white-space. This file can
provide atomic charge information, the main use of which is to visualize electrostatic potential and
analyze it on molecular surface based on atomic charges, the electrostatic interaction energy based
on atomic charges can also be evaluated by subfunction -2 of main function 7 using .chg as input
file. When .chg file is loaded, the sum of all atomic charges as well as electric dipole moment
calculated using the atomic charges will be shown on screen.
An example of .chg file of water molecule is given below:
O 0.000000 0.000000 0.119308 -0.301956
H 0.000000 0.758953 -0.477232 0.150977
H 0.000000 -0.758953 -0.477232 0.150977
.pqr file: This format is very similar to .pdb format, but with different content. Behind the
columns corresponding to atomic X/Y/Z coordinate, there are two columns recording atomic
charges and atomic radii, respectively (the number of decimals of the two columns is not important,
the fields must be delimited by white-space). This kind of file can provide atom information as well
as atomic charge information to Multiwfn. Below is an example .pqr file of water. The REMARK
16
field could exist to record comments, they will be skipped during loading the file.
REMARK From file m1charges.out
REMARK ESP charges
ATOM 1 O O 1 1 0.000 0.123 0.000 -0.680698 2.9000
ATOM 2 H O 1 1 0.757 -0.490 0.000 0.340338 2.6000
ATOM 3 H O 1 1 -0.757 -0.490 0.000 0.340361 2.6000
Gaussian-type cube file (.cub or .cube): This is the most popular volumetric data format, it
can be generated by vast computational chemistry softwares and can be recognized by the majority
of molecular graphics programs. Atom coordinates, a set of grid data of real space function or
multiple sets of grid data of molecular orbitals could be recorded in this file. After a cube file is
loaded into Multiwfn, one can choose main function 0 to visualize isosurfaces, or use main function
13 to process the grid data.
Multiwfn only supports cube file of rectangle grid, that means the three translation vectors are
parallelized with X, Y and Z Cartesian axes, respectively. However, if the grid of loaded cube file
is not rectangle, you can still normally use the grid data calculation function (subfunction 11 of main
function 13).
.gro file: GROMOS format. This kind of file is most frequently employed in GROMACS
molecular dynamics program. .gro file can only provide atomic information for Multiwfn. Note that
since this file records atom name rather than element, Multiwfn automatically guesses the actual
element based on atom names and residue names during loading. However, sometime the guessed
element may be incorrect, therefore it is recommended to examine the printed molecular formula
after loading the file.
.vti, .dx and DMol3 grid file (.grd):

.vti is "ParaView VTK Image Data" format, which can record scalar field and vector field. This
kind of file can be generated by e.g. GIMIC 2.0 and ParaView programs. Only .vti file containing
scalar data of ASCII type is supported. Briefly speaking, this file is very similar to .cub file, but no
atom information is presented.
.dx is volumetric data format that can be exported by e.g. Volmap plugin of VMD program.
.grd file is the volumetric data format mainly used by DMol3 program. Only .grd file
corresponding to rectangle grid could be loaded. Also, no atom information is recorded in .grd.
Plain text file: This file type is only used for special functions, such as plotting DOS graph,
plotting spectrum, generating Gaussian input file with initial guess. See explanations in
corresponding sections.
Gaussian output file also belongs to this type. However, if you set "iloadGaugeom" in
settings.ini to 1, then Multiwfn will load geometry (the final one) and number of electrons from this
file.
17
2.6 Real space functions
The "Real space functions" in this manual referred to as the functions whose variables are
coordinate of the three-dimension space of present system. Real space function analysis is one of
the most important functions of Multiwfn, the supported real space functions are listed below. All
wavefunctions are assumed to be real type, all units are in atomic unit (a.u.).
Notice that for speeding up calculation, especially for big system, when evaluating a
exponential function (except for some real space functions, such as 12, 14 and 16), if the exponent
is more negative than -40, then this evaluation will be skipped. The default cutoff value is safe
enough and cannot cause detectable loss of precision even in quantitive analysis, you can also
disable this treatment or adjust cutoff, see “expcutoff” in settings.ini.
1 Electron density
2
 (r ) = i  i (r ) = i  Cl .i  l (r )
2
i i l
where i is occupation number of orbital i,  is orbital wavefunction,  is basis function. C is

coefficient matrix, the element of the ith row jth column corresponds to the expansion coefficient
of orbital j respect to basis function i. Atomic unit for electron density can be explicitly written as
1/Bohr-3 (which corresponds to 6.74833/Å-3, since 1 Bohr = 0.529177Å).
When input file does not contain GTF information, this function will be calculated as
promolecular density, which is approximate molecular electron density simply constructed by
superposing built-in spherically averaged free-state atomic density of all atoms in the system. See
Appendix 3 on how the built-in atomic density were derived.
It is worth to note that distribution character of valence electron density is much more
informative than electron density, this point was thoroughly discussed in my work Acta Phys. -Chim.
Sin., 34, 503 (2018) DOI: 10.3866/pku.Whxb201709252.
2 Gradient norm of electron density

2
  (r)    (r)    (r ) 
2 2
 (r) =   +   +  
 x   y   z 
3 Laplacian of electron density
 2  ( r )  2  (r )  2  (r )
 2  (r ) = + +
x 2 y 2 z 2
The positive and negative value of this function correspond to electron density is locally
depleted and locally concentrated respectively. The relationships between ∇2 𝜌 and valence shell
electron pair repulsion (VSEPR) model, chemical bond type, electron localization and chemical
reactivity have been built by Bader and many other researchers.
If “laplfac” in settings.ini is set to other value rather than the default one 1.0, ∇2 𝜌 will be
18
multiplied with this value. Setting it to negative value is convenient for analysis of electron density
concentration.
The ∇2 𝜌 outputted by Multiwfn is in atomic unit, which can be explicitly written as 1/Bohr-5
(corresponding to 24.09874/Å-5, since 1 Bohr = 0.529177Å).
4 Value of orbital wavefunction

 i (r) =  Cl ,i  l (r)
l
When you select this function, you will be prompted to input the index of orbital i.
5 Electron spin density

Spin density is defined as the difference between alpha and beta density
 s (r) =   (r) −   (r)

If “ipolarpara” in settings.ini is set to 1, then spin polarization parameter function will be
returned instead of spin density
  (r ) −   (r)
 (r) = 
 (r ) +   (r )
The absolute value of  going from zero to unity corresponds to the local region going from
unpolarized case to completely polarized case.
6 Hamiltonian kinetic energy density K(r)

The kinetic energy density is not uniquely defined, since the expected value of kinetic energy
operator   | −(1 / 2) |   can be recovered by integrating kinetic energy density from
2
alternative definitions. One of commonly used definition is
1
K (r) = − ii* (r) 2i (r)
2 i
7 Lagrangian kinetic energy density G(r)

Relative to K(r), the local kinetic energy definition given below guarantee positiveness
everywhere, hence the physical meaning is clearer and more commonly used. G(r) is also known as
positive definite kinetic energy density.
1 1 
  i (r )    i (r )    i (r )  
2 2 2

G(r ) = i  i (r ) = i   +   + 
2
 
2 i 2 i  x   y   z   
1  i   i
Since  i = (i i ) = 2i i  i , there is an equivalent form of G(r):
2
  .
8 i i
K(r) and G(r) are directly related by Laplacian of electron density
19
2  (r) / 4 = G(r) − K (r)
8 Electrostatic potential from nuclear / atomic charges
ZA
Vnuc (r) = 
A r − RA
where RA and ZA denote position vector and nuclear charge of atom A, respectively. If pseudo-
potential is used, then Z is the number of explicitly expressed electrons. When .chg file is used as
input, Z will stand for the atomic charges recorded in the file (the fourth column), at this time Vnu is
useful for analyzing the difference between exact electrostatic potential and the electrostatic
potential reproduced by atomic charges.
Notice that at nuclear positions, this function will be infinite and may cause some numerical
problems in program, hence at these cases this function always returns 1000 instead of infinity.
9 Electron localization function (ELF)

The larger the electron localization is in a region, the more likely the electron motion is
confined within it. If electrons are completely localized, then they can be distinguished from the
ones outside. Bader found that the regions which have large electron localization must have large
magnitudes of Fermi hole integration. However, the Fermi hole is a six-dimension function and thus
difficult to be studied visually. Becke and Edgecombe noted that spherically averaged like-spin
conditional pair probability has direct correlation with the Fermi hole and then suggested electron
localization function (ELF) in the paper J. Chem. Phys., 92, 5397 (1990). The ELF used in Multiwfn
is generalized for spin-polarized system, see Acta Phys. -Chim. Sin., 27, 2786 (2011) for
introduction. For a review, see Chapter 5 of Theoretical Aspects of Chemical Reactivity (2007).
1
ELF(r ) =
1 + [ D(r ) / D0 (r )]2
where
1   (r )   (r ) 
2 2
1
D(r ) = i  i (r ) − + 
2
2 i 8    (r )   (r ) 
 
3
D0 (r ) = (6 2 ) 2 / 3 [  (r )5 / 3 +   (r )5 / 3 ]
10
For closed-shell system, since  =   = (1 / 2) , D and D0 terms can be simplified as
1  (r )
2
1
D(r ) = i  i (r ) −
2
2 i 8  (r )
D0 (r ) = (3 / 10)(3 2 ) 2 / 3  (r )5 / 3
Savin et al. have reinterpreted ELF in the view of kinetic energy, see Angew. Chem. Int. Ed.
Engl., 31, 187 (1992), which makes ELF also meaningful for Kohn-Sham DFT wavefunction and
even multi-configuration wavefunction. The first term of D(r) can be seen as the exact kinetic
20
energy density of the noninteracting electron system defined by KS-DFT theory, namely
 S (r ) = (1 / 2)i i (r ) , while the second term is equivalent to Weizsäcker kinetic energy
2
density  W (r ) = (1 / 8)  /  , therefore the D(r ) =  S (r ) −  W (r ) reveals the excess

2
kinetic energy density caused by Pauli repulsion and it is known as Pauli kinetic energy density. The
D0(r) can be interpreted as Thomas-Fermi kinetic energy density TF, which is the exact kinetic
energy density of noninteracting, uniform electron gas. Since D0(r) is introduced into ELF as
reference, what the ELF reveals is actually degree of relative localization.
ELF is within the range of [0,1]. A large ELF value means that electrons are greatly localized,
indicating that there is a covalent bond, a lone pair or inner shells of the atom involved. ELF has
been widely used for a wide variety of systems, such as organic and inorganic small molecules,
atomic crystals, coordination compounds, clusters, and for different problems, such as the revealing
atomic shell structure, classification of chemical bonding, verification of charge-shift bond,
studying aromaticity.
Notice that there is a deficiency of ELF, sometimes with r going beyond from molecular
boundary, D(r) decreases faster than D0(r) and then ELF reaches 1 (completely localized). To
overcome the problem, Multiwfn automatically adds a minimal value 10-5 to D(r), this treatment
almost does not affect the ELF value in interesting regions. You can also disable this treatment by
modifying “ELF_addminimal” in settings.ini to 0.
Tsirelson and Stash put forward an approximate version of ELF in Chem. Phys. Lett., 351, 142
(2002), in which the actual kinetic energy term in D(r) is replaced by Kirzhnits type second-order
gradient expansion, that is
(1 / 2)i i (r )   TF (r ) + (1 / 72)  (r ) /  (r ) + (1 / 6)2  (r )
2 2
so that ELF is totally independent from wavefunction, and then can be used to analyze electron
density from X-ray diffraction data. Of course Tsirelson’s ELF can also be used to analyze electron
density from quantum chemistry calculation, but is not as good as the ELF defined by Becke owing
to the approximation introduced in kinetic energy term, however, qualitative conclusions can still
be recovered in general. If you want to use Tsirelson’s definition of ELF, change “ELFLOL_type”
in settings.ini from 0 to 1.
If “ELFLOL_type” is set to 2, another formalism will be used:
1
1 + D(r ) / D0 (r )
A real space function closely related to ELF is SCI index, it is very useful for identifying strong
covalent bonds, see introduction of user-defined function 37 in Section 2.7.
10 Localized orbital locator (LOL)

This is another function for locating high localization regions likewise ELF, defined by
Schmider and Becke in the paper J. Mol. Struct. (THEOCHEM), 527, 51 (2000).
21
 (r )
LOL(r ) =
1 +  (r )
where
D0 (r )
 (r ) =
(1 / 2) i  i (r )
2
D0(r) for spin-polarized system and closed-shell system are defined in the same way as in ELF.
LOL has similar expression compared to ELF. Actually, the chemically significant regions that
highlighted by LOL and ELF are generally qualitative comparable, while Jacobsen pointed out that
LOL conveys more decisive and clearer picture than ELF, see Can. J. Chem., 86, 695 (2008).
Obviously LOL can be interpreted in kinetic energy way as for ELF, however LOL can also be
interpreted in view of localized orbital. Small (large) LOL value usually appears in boundary (inner)
region of localized orbitals because the gradient of orbital wavefunction is large (small) in this area.
The value range of LOL is identical to ELF, namely [0,1].
Multiwfn also supports the approximate version of LOL defined by Tsirelson and Stash (Acta.
Cryst., B58, 780 (2002)), namely the actual kinetic energy term in LOL is replaced by second-order
gradient expansion, as what they do for ELF. This Tsirelson’s version of LOL can be activated by
setting “ELFLOL_type” to 1.
For special reason, if “ELFLOL_type” in settings.ini is changed from 0 to 2, another formalism
1
will be used: LOL(r ) = .
1 + [1 /  (r )]2
11 Local information entropy

Information entropy is a quantification of information, this theory was proposed by Shannon
in his study of information transmission in noise channel, nowadays its application has been largely
widened to other areas, including theoretical chemistry. For example, Aslangul and coworkers
attempted to decompose diatomic and triatomic molecules into mutually exclusive space by
minimizing information entropy (Adv. Quantum Chem., 6, 93 (1972)), Parr et al. discussed the
relationship between information entropy and atom partition as well as molecular similarity (J. Phys.
Chem. A, 109, 3957 (2005)), Noorizadeh and Shakerzadeh suggested using information entropy to
study aromaticity (Phys. Chem. Chem. Phys., 12, 4742 (2010)). The formula of Shannon’s
information entropy for normalized and continuous probability function is
S ( x) = −  P( x) ln P( x) d x
For chemical system, if P(x) is replaced by (r)/N, then the integrand may be called local
information entropy of electrons
 (r)  (r)
S (r) = − ln
N N
where N is the total number of electrons in current system. Integrating this function over whole
space yields information entropy.
22
12 Total electrostatic potential (ESP)
ZA  (r ')
VESP (r ) = Vnuc (r ) + Vele (r ) =  − dr'
A r − RA r −r'
where Z is nuclear charge, the index A cycles over all atoms. If pseudo-potential is used, then ZA is
the number of explicitly represented electrons. The atomic unit (a.u.) of ESP is Hartree/e, where e
is elementary charge. Other commonly used unit are eV/e and kcal/(mole), however for simplicity,
in Multiwfn they are printed as eV and kcal/mol, respectively. The value in eV/e unit is equal to the
value in SI unit (J/C).
This function measures the electrostatic interaction between a unit point charge placed at r and
the system of interest. A positive (negative) value implies that current position is dominated by
nuclear (electronic) charges. Molecular electrostatic potential (ESP) has been widely used for
prediction of nucleophilic and electrophilic sites for a long time. It is also valuable in studying
hydrogen bonds, halogen bonds, molecular recognitions and the intermolecular interaction of
aromatics. Moreover, based on statistical analysis, Murray and coworkers found a set of functions
called GIPF, see J. Mol. Struct. (THEOCHEM), 307, 55, which connects ESP in molecular surface
and macroscopic properties. There are a lot of reviews on ESP, interested readers are suggested to
consult WIREs Comput. Mol. Sci., 1, 153 (2011), Theor. Chem. Acc., 108, 134 (2002), Chapter 17
of the book Chemical Reactivity Theory-A Density Functional View, the entry "Electrostatic
Potentials: Chemical Applications" (page 912) in the book Encyclopedia of Computational
Chemistry and Chapter 7 of the book Reviews in Computational Chemistry vol.2.
By the way, if you only want to obtain the electrostatic potential contributed by electrons,
namely Vele(r), you can use the 14th user-defined function. If you want to omit contribution of
specific nucleus during evaluating VESP(r), use 39th user-defined function. See corresponding
entries in Section 2.7 for details.
Currently there are two sets of internal code in Multiwfn for evaluating ESP, they can be
selected by "iESPcode" in settings.ini. iESPcode=1 corresponds to the old and slow code written by
Tian Lu, iESPcode=2 (default) corresponds to the newer and fast code provided by Jun Zhang based
on his LIBRETA electron repulsion integral library. If the latter code is employed, citing J. Chem.
Theory Comput., 14, 572 (2018) is recommended. In addition, in many kinds of analyses, if you are
using .fch/fchk as input file and Gaussian has been installed on your machine, you can allow
Multiwfn to invoke the cubegen utility in Gaussian package to calculate ESP, the speed is usually
even faster than "iESPcode=2", see Section 5.7 for detail.
Note: Some functions in Multiwfn modify orbital occupation. For example, user can use subfunction 26 of main
function 6 to manually modify occupation number of specific orbitals. Only the ESP calculated by internal code with
"iESPcode=1" directly reflect the effect of the modification on occupation. In the case of using "iESPcode=2", in
order to represent the effect of modification in the ESP calculation, you should first export the modified wavefunction
to .wfn/wfx/molden/mwfn (via subfunction2 of main function 100), then reboot Multiwfn and load the file.
13 Reduced density gradient (RDG)

RDG and sign(2) are a pair of very important functions for revealing weak interaction
regions, they are collectively employed in NCI method, see J. Am. Chem. Soc., 132, 6498 (2010)
for detail, and application examples in Section 3.23.1. RDG is defined as
1  (r )
RDG(r ) =
2(3 ) 2 1/ 3
 (r ) 4 / 3
23
Notice that there is a parameter “RDG_maxrho” in settings.ini, if the value is set to x, then
RDG function will be set to an arbitrary big value (100.0) where the electron density is larger than
x. This mechanism allows uninteresting regions to be shielded when viewing isosurfaces of weak
interaction regions. By default x is 0.05, you can nullify this treatment by setting the parameter to
zero.
14 Reduced density gradient (RDG) with promolecular approximation

Weak interaction has significant influence on conformation of macromolecules, binding mode
of proteins and ligands; however reproduction of electron density by ab initio and grid data
calculation of RDG for such huge systems are always too time-consuming. Fortunately, it is found
that weak interaction analysis under promolecular density is still reasonable. Promolecular density
is simply constructed by superposing electron densities of free-state atoms and hence can be
evaluated extremely rapidly
 pro (r ) =   Afree,fit (r − R A )
A
where  Afree, fit (r ) is pre-fitted spherically averaged electron density of atom A. The atomic
densities for H~Lr are built-in data of Multiwfn, among which the data for H~Ar are taken from
supplemental material of J. Am. Chem. Soc., 132, 6498 (2010), while those for other elements are
evaluated according to the description in Appendix 3. For elements heavier than Lr the promolecular
approximation is not currently available.
For efficiency consideration, if contribution from H, C, N or O atom to the function value at a specific point is
less than 0.00001, then the contribution will not be calculated, for huge system this treatment improves efficiency
several times and the result is almost unperturbed. You can also disable this treatment by setting “atomdenscut” in
settings.ini to 0.
The parameter “RDGprodens_maxrho” in settings.ini is the counterpart of “RDG_maxrho” in
the case of promolecular approximation.
15 sign(2)
(r) = sign[2 (r)] (r)
where sign[λ2(r)] means the sign of the second largest eigenvalue of electron density Hessian matrix
at position r.
16 sign(2) with promolecular approximation

The actual electron density used to evaluate sign[λ2(r)] is approximated by promolecular
electron density, see the description in real space function 14.
17 Exchange-correlation density, correlation hole and correlation factor

These functions involve advanced topics, in order to clarify their physical meanings and to
avoid confusion of the symbols used in a wide variety of literatures, I think it is worth to use much
more texts to introduce theoretical background. For more detail discussions, please consult Chapter
2 of A Chemist's Guide to Density Functional Theory 2ed and Chapter 5 of Methods of molecular
quantum mechanics (2ed, McWeeny).
24
======= Theoretical basis ======

Pair density is defined as
 (r1, r2 ) = N ( N − 1)    (x , x ...x ) d 1 d  2 d x3 d x 4 ...d x N

2
1 2 N
where Ψ is system wavefunction, r is space coordinate, σ is spin coordinate, x is space-spin

coordinate. Pair density denotes the probability that finding an electron at r1 and another electron at
r2, regardless of the spin type. If we perform double-integration for pair density over the whole
space, we will get N(N-1), reflecting the nature that there are N(N-1) electron pairs in present system.
Obviously, pair density can be decomposed to contributions from different spin types of electron
pairs
 (r1 , r2 ) =   (r1 , r2 ) +   (r1 , r2 ) +   (r1 , r2 ) +   (r1 , r2 )

If the electron motions are completely independent with each other, then the probability density
of finding two electrons with spin σ1 at r1 and with σ2 at r2 respectively should simply be
  (r1 )  (r2 ) . Of course, in real world electrons always interacting with each other, so their
1 2
motions are correlated. The pair density thereby should be corrected by exchange-correlation
density Г
   (r1, r2 ) =   (r1 )   (r2 ) + XC (r1, r2 )

1 2 1 2 1 2
If we have already known that an electron with spin σ1 presents at r1, then the probability of
finding another electron with spin σ2 at r2 is known as conditional probability (This function is also
known as Lennard-Jones function)
   (r1 , r2 )
1 2
XC (r1 , r2 ) 1 2
1 2 (r1 , r2 ) = =  
( r ) + 2
  (r1 ) 1
2
  (r1 ) 1
Correlation hole reveals the decrease of probability of finding another electron with spin σ2 at
r2 when an electron with spin σ1 presents at r1 owing to electron correlation effect
1 2 XC1 2 (r1 , r2 ) 1 2 2  1 2 (r1 , r2 )

hXC (r1 , r2 ) = 1
=  (r , r ) −  (r ) = 1
−   2 (r2 )
 (r1 ) 1 2 2
 (r1 )
Correlation factor is a function closely related to correlation hole
1 2
1 2 hXC (r1 , r2 ) XC1 2 (r1 , r2 )
f XC (r1 , r2 ) = =
  2 (r2 )  1 (r1 )   2 (r2 )
Collectively, one can write out
   (r1, r2 ) =   (r1 )   (r2 ) +   (r1 )hXC

1 2 1 2 
(r1, r2 ) =   (r1 )   (r2 ) [1 + f XC
1 1 2 
(r1, r2 )]
1 2 1 2
ГXC can be decomposed to the sum of exchange correlation (also called as Fermi correlation)
part ГX and Coulomb correlation part ГC, therefore hXC can be straightforwardly decomposed to
exchange hole hX (also called as Fermi hole) and Coulomb hole hC as follows. Likewise, fXC can be
decomposed to fX and fC
25
XC1 2 (r1 , r2 ) = X1 2 (r1 , r2 ) + C1 2 (r1 , r2 )

1 2
hXC (r1 , r2 ) = hX1 2 (r1 , r2 ) + hC1 2 (r1 , r2 )
1 2
f XC (r1 , r2 ) = f X1 2 (r1 , r2 ) + f C1 2 (r1 , r2 )
Fermi correlation only presents between like-spin electrons; while Coulomb correlation occurs
between any two electrons. Fermi correlation is much more important than Coulomb correlation,
even at Hartree-Fock level, Fermi correlation is always well represented due to the anti-symmetry
requirement of Slater determinant, while Coulomb is completely ommited. Only post-HF
wavefunction is capable to simultaneously exhibit Fermi and Coulomb correlation effects.
Commonly we only focus on Fermi hole while neglecting Coulomb hole. One can easily show that
integration of hX over whole space is exactly equal to -1, hence Fermi correlation perfectly avoided
self-pairing problem, which may cause significant rise in system energy; while the integration for
hC is zero, this is mainly why Coulomb correlation has less influence on system energy.
It is also rather straightforward to obtain the pair density and conditional probability when only
exchange correlation or Coulomb correlation is taken into account.
The total π, Ω, ГXC (or ГX, ГC) and hXC (or hX, hC) for an electron with spin σ, regardless the
spin of another electron can be defined as
  ,tot (r1 , r2 ) =   (r1 , r2 ) +   (r1 , r2 )

 ,tot (r1 , r2 ) =  (r1 , r2 ) +  (r1 , r2 )
XC,tot (r1 , r2 ) =  
XC (r1 , r2 ) +  XC (r1 , r2 )
 ,tot  
hXC (r1 , r2 ) = hXC (r1 , r2 ) + hXC (r1 , r2 )
======= Technical aspects ======

For single-determinant wavefunctions, exchange-correlation density for an  electron can be
explicitly written as
occ occ
XC (r1 , r2 ) = − i (r1 ) j (r2 ) j (r1 )i (r2 )

XC,tot (r1 , r2 ) =  * *
i j
 
Note that XC (r1, r2 ) = XC (r1, r2 ) = 0 for this type of wavefunction. To obtain the
expression for β electron, just replace  by β, similarly hereafterin.
For post-HF wavefunction, exact evaluation of pair density requires two-particle density matrix
(2PDM). Unfortunately 2PDM is very difficulty to be obtained, mainstream quantum chemistry
packages including Gaussian cannot directly output it. In Multiwfn, exchange-correlation density
for post-HF wavefunction is approximately evaluated by natural orbital formalism. Note that the
approximate method is not unique, see J. Chem. Theory Comput., 6, 2736 (2010) for discussion.
The most popular form among them, which is firstly derived by Müller, is currently implemented
in Multiwfn as below. See Phys. Lett. A, 105, 446 (1984), also see Mol. Phys., 100, 401 (2002) for
extensive discussion (especially equation 32).
26
XC,approx
,tot
(r1 , r2 ) = − 
i j i (r1 ) j (r2 ) j (r1 )i (r2 )
* *
i j
Obviously, if occupation numbers of natural spin orbitals are integer (0 or 1), then XC,approx
,tot
reduces to single-determinant form. so XC,approx

,tot
can be regarded as a general form to evaluate
exchange-correlation density. Note that post-HF wavefunction has taken Coulomb correlation
between unlike-spin electrons into account, however there is no way to separate 
XC,approx and
  ,tot
XC,approx from  XC,approx .
The exchange-only part of Г for post-HF wavefunction can be approximately evaluated as

below (of course, there is no exchange correlation between β electron pair)
X,approx X,approx (r1 , r2 ) = − 
(r1 , r2 ) =  i ji (r1 ) j (r2 ) j (r1 )i (r2 )
,tot * *
i j
So Coulomb-only part of Г can be evaluated as (including both  and β pair contributions)
C,approx
,tot
(r1 , r2 ) = XC,approx
,tot
(r1 , r2 ) − 
X,approx (r1 , r2 )
=  [(
i j − 
i j )i (r1 ) j (r2 ) j (r1 )i (r2 )]
* *
i j
Since we already have explicit expression to calculate ГXC term, other quantities introduced
eariler can be easily computed according to the relationships between them and ГXC. Recall that
  (r ) = i  i (r ) .
2
i
Postscript: One can show that XC,approx

,tot
(r1, r2 ) also exactly holds the requirement that

integration of r2 over the whole space is equal to −  (r1 ) . However, in common, integrating r2
over the whole space for X,approx

,tot
(r1, r2 ) and C,approx
,tot
(r1, r2 ) deviate from −   (r1 ) and
 ,tot  ,tot
zero, respectively, which are basic properties of exact form of  X and C .
======= Usage ======

In Multiwfn, r1 is seen as reference point and r2 is seen as variable, to define the coordinate of
reference point, just modifying “refxyz” in settings.ini before booting up.
“paircorrtype” parameter in settings.ini controls which type of correlation effect will be taken
into consideration in calculation of Г. Since correlation hole and correlation factor are calculated
based on Г, this setting also affects them. For single-determinant wavefunction, =1 and =3 are
equivalent and =2 is meaningless, because Coulomb correlation is completely ommited.
=1: Only consider exchange correlation
=2: Only consider Coulomb correlation
=3: Consider both exchange and Coulomb correlation
“pairfunctype” parameter in settings.ini controls which function and which spin will be
27
calculated by real space function 17, see below, those enclosed by parentheses are for single-
determinant wavefunction cases. Of course, for closed-shell system, the results for  spin are exactly
identical to those for β spin. Note that correlation factor for post-HF wavefunction case is undefined.
 ,tot   ,tot 
=1: h (h ) =2: h (h )
 
=4: Undefined ( f ) =5: Undefined ( f )
 ,tot   ,tot 
=7:  ( ) =8:  ( )
=10:   when paircorrtype =1 (   ) =11:   when paircorrtype =1 (   )
=12:  any,any (  any,any )

For example, if paircorrtype=1 and pairfunctype=2, for post-HF wavefunction, what will be
 ,tot
calculated is hX (r1, r2 ) , which is equivalent to hX (r1 , r2 ) since β electron pairs have no
exchange correlation. This quantity can be interpreted as Fermi hole at r2 caused by a  electron
present at r1.
18 Average local ionization energy

Average local ionization energy is written as (Can. J. Chem., 68, 1440 (1990))
  (r ) | 
i i |
I (r ) = i
 (r )
where i(r) and i are the electron density function and orbital energy of the ith molecular orbital,
respectively. Hartree-Fock and typical DFT functionals such as B3LYP are both suitable for
computing 𝐼 .̅ Lower value of 𝐼 ̅ indicates that the electrons at this point are more weakly bounded.
𝐼 ̅ has widespread uses, for example revealing atomic shell structures, measuring electronegativity,
predicting pKa, quantifying local polarizability and hardness, but the most important one may be
predicting reactive site of electrophilic or radical attack. It is proved that the minima of ALIE on
vdW surface are good indicator to reveal which atoms are more likely to be the preferential site of
electrophilic or radical attack.. There are also many potential uses of 𝐼 ̅ waiting for further
investigation. Excellent reviews of 𝐼 ̅ have been given by Politzer et al, see J. Mol. Model., 16, 1731
(2010) and Chapter 8 of the book Theoretical Aspects of Chemical Reactivity (2007).
Since 𝐼 ̅ is dependent upon orbital energies, while orbital energy for post-HF wavefunction is
undefined, therefore when post-HF wavefunction is used, 𝐼 ̅ will be simply outputted as zero
everywhere.
If the parameter "iALIEdecomp" in settings.ini is set to 1, in main function 1, not only 𝐼 ̅ will
be outputted, the contribution from each occupied MOs will also be outputted, the contribution due
to MO i is defined as
 i (r ) |  i |
I i (r ) =
 (r )
28
19 Source function
Source function was proposed by Bader and Gatti, see Chem. Phys. Lett., 287, 233 (1998).
2  (r')
SF (r, r') = −
4 r − r'
It can be shown that
 (r ) =  SF (r, r') d r'

where r' ranges entire space. This equation suggests that SF(r,r') represents the effect of electronic
Laplacian at r' on electron density at r. If at r' the electron is concentrated (namely Laplacian is
negative, also suggesting potential energy dominates kinetic energy), then r' will be a source for the
electron density at r; conversely, if at r' the electron is depleted, then r' diminish the electron density
at r. If the range of integration in above formula is restricted to a local region  and we get a value
S(r,), then S(r,)/ρ(r)100% can be regarded as the contribution from region  to the electron
density at r. Source function has many uses, when it is used to discuss bonding problems, usually
bond critical points are taken as r. A very comprehensive review of theoretical background and
applications of source function is given by Gatti in Struct. & Bond., 147, 193 (2010).
In Multiwfn, source function has two modes: (1) If "srcfuncmode" in settings.ini is set to 1,
then r' is regarded as variable, while r is regarded as fixed reference point, whose coordinate is
determined by “refxyz” in settings.ini. This is default mode, useful to study effect of electronic
Laplacian at everywhere on specific point (2) If "srcfuncmode" is set to 2, then r becomes variable
and r' becomes reference point, this is useful to study effect of electronic Laplacian at specific point
on everywhere.
When r=r', this function will return −2(r')/0.001 to avoid numerical problem.
20,21 Electron delocalization range function EDR(r;d) and orbital overlap distance
function D(r)
Content of this section and all analysis code of EDR(r;d) and D(r) was kindly contributed by Arshad Mehmood
and then slightly adapted by Tian Lu.
The electron delocalization range function EDR(r;d) (J. Chem. Phys., 141, 144104 (2014); J.
Chem. Theory Comput., 12, 3185 (2016); Angew. Chem. Int. Ed., 56, 6878 (2017)) quantifies the
extent to which electrons at point r in a wave function occupy orbital lobes of size d. EDR(r;d) is
built from the nonlocal one-particle reduced density matrix (1-RDM)  (r, r' ) = i (r ) (r' )
i
as
EDR(r; d ) =  g d (r, r ') (r, r ')dr '
 | r − r ' |2 
3/4
 2 
gd (r, r ') =  2   −1/2 (r ) exp  − 
d   d2 
Here (r) is the electron density at point r. The prefactor ensures that the EDR is between -1 and
+1. The Multiwfn implementation evaluates the EDR on grids, for a single global input value of
distance d. Section 4.5.6 illustrates an example.
At each point, the orbital overlap distance function D(r)=argmaxdEDR(r;d) corresponds to the
distance d that maximizes EDR(r;d). Compact, chemically "hard" regions of small D(r) are
29
distinguished from diffuse, chemically "soft" regions of large D(r). Atomic averages of valence-
electron D(r) complement the information obtailed from atomic partial charges. Plots of D(r) on
density isosurfaces, and quantitative analysis of such surfaces, complements molecular electrostatic
potentials. The Multiwfn implementation evaluates EDR(r;di) on a grid of distances di, then uses a
three-point numerical fit to find the maximum. Sections 4.5.7 and 4.12.8 illustrate example
calculations.
22, 23 g function defined in Independent Gradient Model (IGM) method

The g function is defined as below in the original paper of Independent Gradient Model (IGM)
method (Phys. Chem. Chem. Phys., 19, 17928 (2017)):
 g (r) = g IGM (r) − g (r) =  abs[

A
A (r)] −  
A
A (r )
where A stands for atomic density of atom A, the abs() operator makes each of the three gradient
components to be its absolute value. Specifically, Multiwfn supports two ways to calculate the g:
• Real space function 22: Corresponding to the g calculated under promolecular
approximation, namely the A corresponds to spherically density of atom A in its isolated state, in
this case only geometry information is needed to be supplied by input file, since the atomic densities
are built-in data in Multiwfn (see Appendix 3 for detail).
• Real space function 23: Corresponding to the g calculated based on Hirshfeld partition of
actual molecular electron density, thus wavefunction must be supplied by input file. This definition
was proposed by me. In this case the A is defined as wA, where  is molecular electron density
calculated in usual way, wA is Hirshfeld weighting function of atom A (see Section 3.9.1 for its
definition).
The g function at bond critical point in weak interaction region is shown to be closely related
to interaction strength. This function can also be plotted as plane map or isosurface map to reveal
all bonding regions. Very detailed introduction of IGM method is given in Section 3.23.5, and if you
want to study weak interaction due to specific two or more fragments, you should use the function
described in this section.
2.7 User-defined real space function
In real space function selection menu you can find a term named "User-defined real space
function". In order to avoid lengthy list of real space functions, numerous uncommonly used real
space functions are not explicitly presented in the list. However, if you want to use them, you can
set "iuserfunc" parameter in settings.ini to one of the indices (see below), then the user-defined
function will be pinned to corresponding function. For example, before running Multiwfn, if you
set "iuserfunc" to 2, then the user-defined real space function will be equivalent to density of beta
electrons. An alternative way of setting user-defined function is inputting iu in the main manu, you
can input the index of the user-defined function.
In fact, the user-defined function corresponds to “userfunc” function in source file function.f90.
By filling proper code yourself, the functions supported by Multiwfn can be easily extended. For
30
examples, after filling the code "userfunc=fgrad(x,y,z,'t')**2/8/fdens(x,y,z)" into proper place of

“function userfunc” in function.f90 and recompile Multiwfn, the integrand of Weizsäcker kinetic
energy functional, namely  W [  ] =   (r) / [8 (r)]d r , will be ready for use. When you
2
write your own code you can refer to existing codes, a list of built-in functions is given in Appendix
2 of this manual.
Prebuilt user-defined functions

-2 Promolecular density calculated based on built-in sphericalized atomic densities. You can check
Appendix 3 on how these atomic densities are produced. Options -3 and -4 of main function 6 could
be used to exclude contribution of some atoms to promolecular density.
Please note that this promolecular density is slightly different to the promolecular density used in real space
functions 14 and 16, because in which the atomic densities from H to Ar are directly taken from the original paper
of NCI method rather than calculated according to the method described in Appendix 3.
-1 Value evaluated by trilinear interpolation from grid data. The grid data can be generated by main
function 5, or loaded from .cub/.grd file when Multiwfn boot up. This function is quite useful. For
example, by making use of this function you can plot the grid data as curve map and plane map via
main functions 3 and 4, respectively; you can also gain atomic contribution to the total value via
subfunction 1 of fuzzy atomic space analysis module (main function 15)
-3 The same as -1, but use 3D cubic B-spline interpolation instead. The cost is higher than trilinear
interpolation but smoother and usually more accurate especially when grid spacing is relatively large.
However, the function interpolated via cubic B-spline shows unwanted feature where the variation of actual
function cannot be well approximated by polynomial and in this case the result is even worse than trilinear
interpolation. For example, cusp character of electron density at nucleus cannot be faithfully represented. In addition,
the function shows fluctuant feature when the actual function shows a steep slope close to a region of small slope,
or when the grid points inadequately represents the actual function.
0 This function corresponds to a constant value of 1.0
1 Alpha density:   (r) = [ (r) +  s (r)] / 2
2 Beta density:   (r) = [ (r) −  s (r)] / 2
3 Integrand of electronic spatial extent <R**2>:  (r)  ( x 2 + y 2 + z 2 )
1  (r )
2
1 2  (r )
4 Weizsäcker potential (closed-shell form): VW (r ) = −
8  2 (r ) 4  (r )
5 Integrand of Weizsäcker functional (closed-shell form):  W (r ) =  (r ) /[8 (r )] , which is

2
the exact kinetic energy density of any one-orbital system (one or two electrons, or any number of
bosons). If you need spin polarized form, use user-defined function 1200.
6 Radial distribution function of electron density: 4   (r )  ( x + y + z ) . Clearly,

2 2 2
spherical symmetry of electron density is assumed.
7 Local temperature, in kB-1 (PNAS, 81, 8028): T (r) = [2G(r)] / [3 (r)]
8 Average local electrostatic potential (J. Chem. Phys., 72, 3027 (1980)): VESP (r ) /  (r )
31
9 Shape function: (r)/N, where N is the total number of electrons
10 Potential energy density (Virial field): V (r) = − K (r) − G(r) = (1 / 4)

2
 (r) − 2G(r)
11 Electron energy density: E (r) = G(r) + V (r) = − K (r) (Sometimes E is written as H)
12 Local nuclear attraction potential energy: −  (r )  Vnuc (r )
13 Kinetic energy density per electron: G(r ) /  (r ) This quantity at bond critical point is useful
to discriminate covalent bonding and closed-shell interaction
 (r ')
14 Electrostatic potential from electrons: Vele (r ) = VESP (r ) − Vnuc (r ) = −  r − r' dr' . The
negative of this function is also known as Hartree potential.
15 Bond metallicity: J (r) =  (r) / 2  (r) At bond critical point,  J >1 indicates metallic
interaction, see J. Phys.: Condens. Matter, 14, 10251 (2002).
36(3 π 2 ) 2 / 3  (r ) 5 / 3
16 Dimensionless bond metallicity:  m (r ) = At bond critical point, larger
5  2  (r )
value corresponds to stronger metallicity of the bond, see Chem. Phys. Lett., 471, 174 (2009).
17 Energy density per electron: E(r)/(r), this value at BCP is called as bond degree parameter (BD),
see J. Chem. Phys., 117, 5529 (2002) for detail, in which the authors advocate that for covalent
bonds (viz. EBCP<0), the BD renders covalence degree, and the stronger the interaction the greater
the BD magnitude; while for closed-shell interactions (viz. EBCP>0), the BD can be viewed as
softening degree, and the weaker the interaction the larger the BD magnitude.
18 Region of Slow Electrons (RoSE), which is defined in Chem. Phys. Lett., 582, 144 (2013) and
D0 (r ) − G(r )
has a pattern very similar to ELF, with value space of [-1,1]:  =
D0 (r ) + G(r )
19 Single exponential decay detector (SEDD), which is highly analogous to ELF. Its updated
definition in J. Chem. Theory Comput., 10, 3745 (2014) is implemented:

  ( (r ) /  (r )) 2  
2

SEDD( r ) = ln 1 +   

   (r )  

20 Density overlap regions indicator (DORI), defined in J. Chem. Theory Comput., 10, 3745 (2014):
DORI(r) =  (r) /[1 +  (r)] , where  (r) = [( (r) /  (r)) 2 ]2 /[ (r) /  (r)]6 . DORI is
mainly used to reveal interatomic interaction regions, see Section 3.23.3 for more information
21 Integrand of X component of electric dipole moment: −x(r)
22 Integrand of Y component of electric dipole moment: −y(r)
23 Integrand of Z component of electric dipole moment: −z(r)
24 Approximate form of DFT linear response kernel for closed-shell (Phys. Chem. Chem. Phys., 14,
32
 i* (r1 ) j (r1 ) *j (r2 ) i (r2 )

3960 (2012)):  (r1 , r2 )  4  
iocc jvir i −  j
25 Magnitude of fluctuation of the electronic momentum: P(r ) =  (r) / [2 (r)] , which is
useful to discuss bonding and very similar to reduced density gradient, see Theor. Chem. Acc., 127,
393 (2010)
26 Integrand of Thomas-Fermi kinetic energy functional (closed-shell form): TF(r)=CTF(r)5/3,
where CTF=(3/10)(32)2/3=2.871234. This is the exact kinetic energy density of noninteracting,
uniform electron gas. If you need spin polarized form, use user-defined function 1200.
−   i (r )  i
2
27 Local electron affinity: EAL (r ) = ivir

, which is very similar to average local
  i (r )
2
ivir
ionization energy, but i cycles all unoccupied orbitals. See J. Mol. Model., 9, 342 (2003). Illustration
of applying this function for a practical molecule is given in Section 4.12.13.
28 Local Mulliken electronegativity:  L (r) = [ I (r) + EAL (r)] / 2 , see J. Mol. Model., 9, 342
(2003)
29 Local hardness: L (r) = [ I (r) − EAL (r)] / 2 , see J. Mol. Model., 9, 342 (2003)
Note: In order to use EAL, L and L, the wavefunction must contain both occupied and unoccupied
orbitals, therefore such as .mwfn, .fch, .molden and .gms rather than .wfn/.wfx file must be used.
Meanwhile, "idelvirorb" in settings.ini must be set to 0. Besides, these quantities are meaningful
only when minimal basis set is used.
30 Ellipticity of electron density:  (r) = 1 (r) / 2 (r) − 1 , where 1 and 2 are the lowest and
the second lowest eigenvalues of Hessian matrix of , respectively. At bond critical point (BCP), 1
and 2 are both negative and exhibit the curvatures of electron density in the two orthogonal
directions that perpendicular to the bond. The  at BCP is often viewed as an indicator of asymmetric
distribution of electron density around the bond, the higher deviation to axisymmetric distribution,
the larger the  value at BCP.
31 eta index:  (r) = 1 (r) / 3 (r) , where 1 and 3 are the lowest and the highest eigenvalues
of Hessian matrix of , respectively. It was argued that the value of  at bond critical point is less
than unity for closed shell interactions and increases with increasing covalent character, see Angew.
Chem. Int. Ed., 53, 2766 (2014), as well as J. Phys. Chem. A, 114, 552 (2010) for discussions.
32 Modified eta index by Tian Lu:  ' (r) = 1 (r) / 3 (r) − 1 . Similar to , but negative value of
' corresponds to closed shell interactions.
33 Potential acting on one electron in a molecule (PAEM), see J. Comput. Chem., 35, 965 (2014):
ZA 1  (r, r' )
VPAEM (r ) = −
| r − R A |  (r )  | r − r' |
+ d r' , π is pair density. It can be further written as
A
33
 ,tot  ,tot
1 XC (r, r' ) + XC (r, r' )
VPAEM (r ) = −VESP (r ) + VXC (r ) = −VESP (r ) +
 (r )  | r − r' |
d r' , where
VESP is the total molecular electrostatic potential and VXC is exchange-correction potential. In
Multiwfn, the exchange-correlation density Γ is evaluated in terms of Müller approximation, see
part 17 of Section 2.6 for detail. It was shown in the original paper that PAEM may be useful to
distinguish covalent and noncovalent interactions. An example of application of PAEM can be found
in Section 4.3.3.
34 The same as 33, but now VXC directly corresponds to DFT exchange-correlation potential. Its
specific form can be chosen via "iDFTxcsel" parameter, see the end of this section for detail. This
form of PAEM is several times more computationally economical than 33 but only supports closed-
shell wavefunction.
35 | V (r ) | / G(r ) . In J. Chem. Phys., 117, 5529 (2002) it was proposed that this quantity at BCP
can be used to discriminate interaction types. <1 corresponds to closed-shell interaction; >2
corresponds to covalent interaction; while >1 and < 2 corresponds to intermediate interaction.
36 On-top pair density, namely the two positions of the pair density are identical: (r,r). See such
as Int. J. Quantum Chem., 61, 197 (1995) for discussion. "paircorrtype" parameter in settings.ini
mentioned earlier affects the result.
37 The strong covalent interaction index (SCI) defined in J. Phys. Chem. A, 122, 3087 (2018) and
further examined in J. Mol. Model., 24, 213 (2018). This function was shown to be very useful for
identifying very strong covalent bonds. SCI is expressed as SCI(r) = 1 /  (r) , where
 (r) = [ S (r) −  W (r)] / TF . The meaning of S, W and TF are described in Section 2.6 where
ELF is introduced. It is easy to find SCI index is closely related to ELF, which can be written as
ELF(r) = 1 /[1 +  2 (r)] .

38 The angle between the second eigenvector of Hessian of electron density and the vector
perpendicular to a given plane, which can be defined by option 4 of main function 1000; the unit
vector normal to the plane will be shown on screen, assume that you use three points A, B, C to
define the plane and you get vector u, but what you really want is -u, you can then input the points
again but in reverse sequence, i.e. C, B, A. In J. Phys. Chem. A, 115, 12512 (2011) this quantity
along bond paths was used to reveal π interaction.
39 Electrostatic potential without contribution of a specific nucleus:
ZA  (r ' )
Vn (r ) =  r−R
A K
−
r − r'
dr' , where the nucleus K can be set by option 3 of main function
A
1000 (which is hidden in main interface but can be chosen). Vn is a useful quantity, for example if
K is chosen as index of a hydrogen, then the value correlates with its pKa, because in this case Vn
approximately reflects the binding energy of a proton at the position of K and rest of the system; In
addition, J. Phys. Chem. A, 118, 1697 (2014) showed that Vn can be used to quantitatively predict
interaction energy of the weak interactions dominated by electrostatic effect (viz. H-bonds, halogen-
bonds, dihydrogen bonds), see Section 4.1.2 for introduction and example.
It is worth to mention that in main function 1, when you request Multiwfn to print properties at nuclear position
34
of an atom, the electrostatic potential without contribution of nuclear charge of this atom is automatically printed.
40 Steric energy density:  (r ) /[8 (r )] , which is equivalent to integrand of Weizsäcker

2
functional.
1 |  (r ) |2 1  2  (r )
41 Steric potential:  s (r ) = − . The negative of this quantity is also
8 [  (r ) +  ]2 4  ( r ) + 
known as one-electron potential (OEP). Notice that the  is a very small term artifically introduced
to avoid the denominator converges to zero faster than nominator. The value of  can be determined
by "steric_addminimal" in settings.ini. If  is set to 0, then the original expression of steric potential
is recovered.
42 Steric charge: qs (r ) =  S (r ) /( −4 ) .

2
43 The magnitude of steric force: FS (r ) =| −S (r ) | .
Notice that the  term, which is mentioned above, also affects steric force and steric charge.
Discussions of steric energy/potential/force/charge can be found in J. Chem. Phys., 126, 244103
(2007).
44,45,46 Damped Steric potential, steric force based on damped steric potential, directly damped
Steric force: Documented privately
50 Shannon entropy density: sS (r ) = − (r ) ln  (r ) .
51 Fisher information density: iF (r ) =|  (r ) | /  (r ) . Note that this quantity is only different

2
from Weizsäcker functional by a constant factor of 1/8.
52 Second Fisher information density: iF (r ) = −

2
 (r) ln  (r) . The relationship between
Shannon entropy and Fisher information can be found in J. Chem. Phys., 126, 191107 (2007). The
integration of this function over the whole space is exactly identical to that of Fisher information
density.
53 Ghosh entropy density or Ghosh-Berkowitz-Parr (GBP) entropy density, in kB (PNAS, 81, 8028
(1984)): s(r) = (3 / 2) (r){ + ln[t (r) / tTF (r)]} , where  = 5 / 3 + ln (4ck / 3) , ck is
Thomas-Fermi constant (3/10)(32)2/3=2.871234, Lagrangian kinetic energy density G(r) is chosen
to be the kinetic energy density term t(r).
54 The same as 53, but G(r)−2(r)/8, which is the kinetic energy density exactly corresponding to
Eq. 22 of PNAS, 81, 8028 (1984), is employed as the t(r). In rare cases this definition of kinetic
energy density leads to a very small negative value, since at these points tTF(r) is also very close to
zero, in order to normally get result, ln[t(r)/tTF(r)] in this case is simply set to zero.
55 Integrand of quadratic form of Rényi entropy: 2(r)
56 Integrand of cubic form of Rényi entropy: 3(r)
 (r )
57 g1 (r ) = 
2
 (r ) ln
 0 (r )
35
  2  (r )  2  0 ( r ) 
58 g2 (r ) =  (r )  −
  (r ) 0 (r) 
2
  (r ) 
59 g3 (r ) =  (r )  ln
  0 (r ) 
Note that user-defined functions 57, 58, 59 can only be studied by plotting as map in main
functions 3, 4, 5, and meantime deformation and promolecular maps are not available.
60 Pauli potential: V = VESP − VXC − VW, where VESP is total electrostatic potential (as shown in
Section 2.6), VXC is exchange-correlation potential (its specific form is determined by "iDFTxcsel"
in settings.ini, see below), VW is Weizsäcker potential. This quantity corresponds to Eq. 17 of
Comput. Theor. Chem., 1006, 92 (2013) and is only applicable to closed-shell cases.
61 The magnitude of Pauli force: F (r ) =| −Vθ (r ) | .
62 Pauli charge: qθ (r ) =  Vθ (r ) /( −4 ) .

2
63 Quantum potential: Vq = Vθ + VXC  VESP − VW
64 The magnitude of quantum force: Fq (r ) =| −Vq (r ) |
65 Quantum charge: qq (r) = qθ (r) + qXC (r)
66 The magnitude of electrostatic force: FESP (r) =| −VESP (r) |
As discussed in Phys. Chem. Chem. Phys., 19, 1496 (2017), there is a very close relationship
between steric force, quantum force and electrostatic force, which are user-defined function 43, 64
and 66, respectively.
67 Electrostatic charge: qESP (r ) =  VESP (r ) /( −4 )

2
68 Energy density of electronic part of electrostatic term of Shubin Liu's energy decomposition:
  (r) ZA 
 e (r ) =  ( r )   d r −  , which is clearly equivalent to ESP multiplied by
 | r − r | A | r − R A | 
electron density. To learn more about Shubin Liu's energy decomposition, see Section 3.24.2.
69, -69 Energy density of quantum part of Shubin Liu's energy decomposition: s(r) − W(r) + XC(r),
where s is Hamiltonian kinetic energy density (if userfunc = 69) or Lagrangian kinetic energy
density (if userfunc = -69). W is integrand of Weizsäcker functional (same as userfunc=5), XC is
exchange-correlation energy density (same as userfunc=1000), whose form can be chosen by
"iDFTxcsel" parameter in settings.ini, see below for detail.
3 (r ) 9  2 (r )
70 Phase-space-defined Fisher information density (PS-FID): if (r ) = = ,
k BT (r ) 2 G(r )
where T(r) is local temperature as shown above. This function has very similar characters to ELF
and LOL, the spatial localization of electron pairs can be clearly revealed. See Chem. Phys., 435,
36
49 (2014) for introduction and illustrative applications.
71, 72, 73, 74 Electron linear momentum density (EMD) in 3D representation. The electron linear
momentum operator is − i , for an orbital the expectation value of linear momentum in X is
− i    , so X component of EMD for a wavefunction can be defined as (the imaginary sign
is ignored) px (r ) = −   (r) (r) / x , similar for Y and Z components. The 71, 72, 73th
i
i
*
i i
user-defined functions correspond to X, Y, Z component, respectively. The magnitude (74) of EMD
is defined as ptot (r ) = p x2 (r ) + p 2y (r ) + p z2 (r ) .
75, 76, 77, 78 Magnetic dipole moment density (MDMD). The operator for magnetic dipole moment
is the angular momentum operator (see Theoret. Chim. Acta, 6, 341 (1966))
          
− i (r  ) = −i î  y − z  + ˆj  z − x  + kˆ  x − y 
  z y   x z   y x 
where i, j, k are unity vectors in X, Y and Z directions, respectively. Therefore, X, Y and Z
component of MDMD may be defined as follows (the imaginary sign is ignored)
  (r )  (r ) 
m x (r ) = −i i* (r )  y i −z i 
i  z y 
  (r )  (r ) 
m y (r ) = −i i* (r )  z i −x i 
i  x z 
  (r )  (r ) 
mz (r ) = −i i* (r )  x i −y i 
i  y x 
The 75, 76, 77th user-defined functions correspond to X, Y, Z component, respectively. The
magnitude (78) of MDMD is mtot (r ) = mx2 (r ) + m 2y (r ) + mz2 (r ) .
79 Gradient norm of electron energy density: |E(r)|

80 Laplacian of electron energy density: 2E(r)
81, 82, 83 X, Y, Z component of Hamiltonian kinetic energy density, respectively
84, 85, 86 X, Y, Z component of Lagrangian kinetic energy density, respectively
87 Local total electron correlation function: I T (r ) = 1

4 [ (1 −  )]
i
i i
1/ 2
| i (r ) |2 . The i denotes
index of natural spin orbital, the  is corresponding occupation number. This and below two
functions are very useful for revealing electron correlation in different regions, see J. Chem. Theory
Comput., 13, 2705 (2017) for introduction, illustrative example can also be found in Section 4.A.6.
Note that in some cases,  may be marginally larger than 1.0 or negative, Multiwfn automatically
set it to 1.0 and 0.0 respectively to make the calculation feasible.
88 Local dynamic electron correlation function:
I D (r ) = 1
4 {[ (1 −  )]
i
i i
1/ 2
− 2i (1 − i )} | i (r ) |2
37
89 Local nondynamic electron correlation function: I ND (r ) = 1

2  (1 −  ) |  (r) |
i
i i i
2
, it is clear
that IT(r) = ID(r) + IND(r).
92, 93, 94 Van der Waals potential, repulsion potential and dispersion potential based on UFF
forcefield parameters, see Section 3.23.7 for detailed introduction. The unit is in kcal/mol, the probe
atom can be set by "ivdwprobe" in settings.ini.
95, 96, 97, 98 Orbital-weighted f +, f −, f 0 Fukui functions as well as orbital-weighted dual descriptor,
respectively. They are originally proposed in J. Comput. Chem., 38, 481 (2017) and J. Phys. Chem.
A, 123, 10556 (2019). They are useful in studying local reactivity for systems whose frontier
molecular orbitals are (quasi-)degenerate, see Section 3.100.16.3 for brief introduction and Section
4.100.16.2 for illustrative examples. The  parameter in these functions can be set by option 6 of
main function 1000 (a hidden option). Since these functions involve virtual orbitals, you should use
mwfn/fch/molden/gms as input file. Only closed-shell single-determinant wavefunction is
supported.
99 Interaction region indicator (IRI): IRI = [  (r)]a / (|  | +b) , where a and b correspond to
"uservar" and "uservar2" in settings.ini, respectively. If "uservar" is 0, a will be recommended value

1.1, while if "uservar2" is 0, then b will be recommended value 0.0005. IRI is able to reveal both
chemical bond regions and weak interaction regions like DORI, but computational cost is much
lower and graphical effect is evidently better. See Section 3.23.8 for more introduction and 4.20.4
for example of applying this function in practical system.
100 Disequilibrium (also known as semi-similarity): Dr (r ) =  (r ) . See illustrative applications

2
in Int. J. Quantum Chem., 113, 2589 (2013).

+ +
101 Positive part of ESP: 𝑉ESP . In the region where VESP is positive, 𝑉ESP = 𝑉ESP ; where ESP is
+
negative, 𝑉ESP = 0.
− +
102 Negative part of ESP: 𝑉ESP , defined similarly as 𝑉ESP .
103 Magnitude of electric field |F|. Since electric field vector is simply negative gradient vector of
ESP, therefore this quantity corresponds to norm of gradient of ESP.
900, 901, 902 X, Y and Z coordinate variables, respectively: x, y, z.
1000 Integrand of DFT exchange-correlation functionals, which is also known as exchange-

correlation energy density.
1100 DFT exchange-correlation potential, i.e. VXC (r) =  EXC [  (r)] /  (r)
For user-defined function 1000 and 1100, "iDFTxcsel" in settings.ini is used to select XC
functional. 0~29 are X part only, 30~69 are C part only, 70~99 are whole XC. For example, if
iDFTxcsel=32, then iuserfunc=1000 corresponds to integrand of LYP, and iuserfunc=1100
corresponds to LYP potential.
Notice that both 1000 and 1100 are applicable for closed-shell cases. But currently only 1000
can be applied to open-shell cases.
[See below for details]
38
Available options of "iDFTxcsel" parameter

Currently only LSDA and some GGA functionals are supported
0 LSDA exchange: − (3 / 2)3 /( 4 ) [  (r )4 / 3 +   (r)4 / 3 ] . For closed-shell cases the

1/ 3
equivalent form is − (3 / 4)(3 /  )  (r) 4 / 3 .

1/ 3
1 Becke 88 (B88) exchange, see Phys. Rev. A, 38, 3098 (1988).

2 Perdew-Burke-Ernzerhof (PBE) exchange, see Phys. Rev. Lett., 77, 3865 (1996).
3 Perdew-Wang 91 (PW91) exchange, see Electronic Structure of Solids '91; Ziesche, P., Eschig,
H., Eds.; Akademie Verlag: Berlin, 1991; p. 11.
30 Vosko-Wilk-Nusair V (VWN5) correlation, see Can. J. Phys., 58, 1200 (1980).
31 Perdew 86 (P86) correlation, see Phys. Rev. B, 33, 8822 (1986).
32 Lee-Yang-Parr (LYP) correlation, see Phys. Rev. B, 37, 785 (1988).
33 Perdew-Wang 91 (PW91) correlation, see the reference of its exchange counterpart.
34 Perdew-Burke-Ernzerhof (PBE) correlation, see the reference of its exchange counterpart.
70 Becke 97 (B97) exchange-correlation, see J. Chem. Phys., 107, 8554 (1997).
71 Hamprecht-Cohen-Tozer-Handy with 407 training molecules (HCTH407) exchange-
correlation, see J. Chem. Phys., 114, 5497 (2001).
80 SVWN5 exchange-correlation.
81 BP86 exchange-correlation.
82 BLYP exchange-correlation.
83 BPW91 exchange-correlation.
84 PBEPBE exchange-correlation.
85 PW91PW91 exchange-correlation.
About iuserfunc=1200
The kinetic energy density (KED) denotes integrand of kinetic energy functional. User-defined
function 1200 is a collection of all KEDs supported by Multiwfn, which were systematically
examined in J. Chem. Phys., 150, 204106 (2019). The "iKEDsel" in settings.ini is used to choose
the form of KED.
Note that if "iKEDsel" is not at its default value (0), then the S term (Lagrangian kinetic energy
density) of ELF and strong covalent interaction index (SCI) will be replaced with corresponding
KED.
There is an subfunction 92 in the main function 1000 (hidden in the main menu), it can calculate
integral of all KEDs over the whole space, namely evaluating kinetic energy based on the kinetic
energy functionals. The GEA4 KED is calculated only when "iKEDsel" is set to 24 since it needs
Laplacian of electron density, while other KEDs only need electron density and its gradient. Integral
of Hamiltonian KED is not calculated by this function since the result is identical to that of
Lagrangian KED.
• iKEDsel=1: Hamiltonian KED, identical to real space function 6
• iKEDsel=2: Lagrangian KED, identical to real space function 7
39
• iKEDsel=3: Thomas-Fermi KED:  TF (r ) =  

  
= ,
TF = C
  
= ,
TF [  (r )]5 / 3 ,
where CTF = 10
3
(6 ) 2 2/3
= 4.557799872 is Thomas-Fermi constant in spin polarized case.
 (r )
2
• iKEDsel=4: Weizsäcker KED:  W (r ) = 

 = ,  8   (r )
Most of below mentioned KEDs can be represented using a general form

  
 (r ) = 
(r ) F [ s (r )] , where s (r ) =  (r ) /  (r )
  4/3
TF
, F is called as enhancement
 = , 
factor, see J. Chem. Phys., 127, 144109 (2007) for detail. This paper systematically introduced and
compared a variety of existing KEDs. Note that a lot of expressions given in this paper are wrong,
while the formulae given below are absolutely correct, and all of them are explicitly written as spin
polarized form, if they will be involved in your work please cite my paper J. Chem. Phys., 150,
204106 (2019); citation of each KED is also given in this paper. More information and comparison
about KEDs can be found in Phys. Rev. A, 46, 6920 (1992) and J. Chem. Phys., 100, 4446 (1994).
  1 
• iKEDsel=5: Second order gradient expansion approximation, FGEA 2 = 1 + s  (r ) 2 
 72CTF 
  1 
• iKEDsel=6: Thomas-Fermi + 1/5 Weizsäcker KED, FTF 5 W = 1 + s  (r ) 2 
 40CTF 
  1  2
• iKEDsel=7: Thomas-Fermi + Weizsäcker KED, FTFvW = 1 + s (r ) 
 8CTF 
  1.067  2 
• iKEDsel=8: Thomas-Fermi + b/9 Weizsäcker KED, FTF 9 W = 1 + s (r ) 
 72CTF 
 0.313 0.187
• iKEDsel=9: N-dependent Thomas-Fermi KED, FTF −N = 1 + − 2 / 3 , where N is the
N 1/ 3 N
number of total electrons in the system
 2
  1 1  (r )
• iKEDsel=10: Pearson KED,  Pear =  TF + , where  is fixed to
1 + [ sr (r ) /  ]6 72   (r )
1, sr (r) = s (r) /[2(6 2 )1 / 3 ] , similarly hereinafter
 9b3 x 4 + a3 x 3 + a2 x 2 + a1 x + 1
• iKEDsel=11: DePristo-Kress Pade KED, FDK Pade = , where
b3 x 3 + b2 x 2 + b1 x + 1
x = [s (r)]2 /(72CTF ) , a1=0.95, a2=14.28111, a3=-19.57962, b1=-0.05, b2=9.99802 and

b3=2.96085
 0.0044188[ s (r )]2
• iKEDsel=12: Lee-Lee-Parr KED, FLLP = 1 +
1 + 0.0253s (r ) arcsinh[ s (r )]
40
 1
• iKEDsel=13: Ou-Yang–Levy 1 KED, FOL 1 = 1 + [ s (r )]2 + 0.00187s (r )
72CTF
 1 0.0245s (r )
• iKEDsel=14: Ou-Yang–Levy 2 KED, FOL 2 = 1 + [ s (r )]2 +
72CTF 1 + 2 5 / 3 s  (r )
 0.0055[ s (r )]2 0.072s (r )

• iKEDsel=15: Thakkar KED, FThak =1+ −
1 + 0.0253s (r ) arcsinh[ s (r )] 1 + 25 / 3 s (r )
 [ s (r )]2
• iKEDsel=16: Becke 86A KED, FB 86 A = 1 + 0.0039
1 + 0.004[ s (r )]2
 [ s (r )]2
• iKEDsel=17: Becke 86B KED, FB 86 B = 1 + 0.00403
{1 + 0.007[ s (r )]2 }4 / 5
• iKEDsel=18: DePristo-Kress 87 KED,
 1 + 0.861504s (r )

FDK 87 = 1 + 0.00132327[ s (r )] 2
1 + 0.044286[ s (r )]2
• iKEDsel=19: Perdew-Wang 86 KED,
   
86 = {1 + 1.296[ sr (r )] + 14[ sr (r )] + 0.2[ sr (r )] }
2 4 6 1 / 15
FPW
• iKEDsel=20: Perdew-Wang 91 KED,

1 + a1sr (r ) arcsinh[ b  sr (r )] + {a2 − a3e −100[ s r ( r )] }[sr (r )]2
2

FPW 91 = , where a1=0.19645,
1 + a1sr (r ) arcsinh[ b  sr (r )] + a4 [ sr (r )]4
a2=0.2743, a3=0.1508, a4=0.004 and b=7.7956
• iKEDsel=21: Lacks-Gordon 94 KED
 {1 + a2 [ sr (r )]2 + a4 [ sr (r )]4 + a6 [ sr (r )]6 + a8 [ sr (r )]8 + a10[ sr (r )]10 + a12[ sr (r )]12}b
FLG 94 =
1 + 10−8 [ sr (r )]2
, where a2=(10-8+0.1234)/0.024974, a4=29.790, a6=22.417, a8=12.119, a10=1570.1, a12=55.944 and
b=0.024974
 1 1.412
• iKEDsel=22: Acharya-Bartolotti-Sears-Parr KED, FABSP = [ s (r )]2 + 1 − 1 / 3
8CTF N
 1  2  1.303 0.029 
• iKEDsel=23: Gázquez-Robles KED, FGR = [ s (r )]2 + 1 − 1 − 1 / 3 + 2 / 3 
8CTF  N  N N 
• iKEDsel=24: Fourth order gradient expansion approximation,
(6 2 ) − 2 / 3   2  2
9 2   (r )   (r )
2
1   (r ) 
4
  1 / 3    (r )  
 GEA 4 =  GEA 2 + [  (r )]    −  
+  
540 

 (r )  8  ( r )  (r ) 3  (r ) 

Note that if the "uservar" in settings.ini is not equal to zero, all above mentioned KEDs will be
41
added by "2 /uservar" term. For example, when iuserfunc=1200, iKEDsel=5 and uservar=6, the
user-defined function will correspond to GEA2+2 /6 (which corresponds to the KED employed
by Tsirelson type of ELF and LOL, as shown in Section 2.6).
1201 Difference between KED selected by "iKEDsel" and Weizsäcker KED.

1202 Difference between KED selected by "iKEDsel" and Lagrangian KED.
1203 Absolute difference between KED selected by "iKEDsel" and Lagrangian KED.
42
2.8 Graphic formats and image size
Multiwfn supports a lot of mainstream graphic formats, including:

1 Postscript (ps)
2 Encapsulated postscript (eps)
3 Portable document format (pdf)
4 Windows metafile format (wmf)
5 Graphics interchange format (gif)
6 TIFF (tiff)
7 Portable network graphics (png)
8 Windows bitmap format (bmp)
9 Scalable vector graphics (svg)
The graphic format of the picture exported by Multiwfn is controlled by “graphformat”
parameter in settings.ini, you can set this parameter to the texts in the parentheses listed above, the
default format is “png”.
For curve maps, the height and weight of the image file are controlled by “graph1Dsize”
parameter in settings.ini. “graph2Dsize” is responsible for two-dimension data plotting (color-filled
map, contour map, relief map, etc.). “graph3Dsize“ is responsible for three-dimension data plotting
(isosurface graph, molecular structure graph, etc.).
Tip 1: If the graph is mainly composed of lines, e.g. contour line map and curve map, the best
formats are pdf and svg. However, if you need to embed the resulting graph to Office, commonly
wmf format should be used.
Tip 2: If you want to make background of exported image file transparent, please look this
video illustration: https://youtu.be/E7lAGac3aDM.
43
3 Functions
3 Functions
This chapter introduces all functions of Multiwfn in detail, the numbers in the parentheses of
secondary titles are indices of corresponding functions in main menu, the numbers in the parentheses
of tertiary titles are indices of corresponding options in corresponding submenus. Different
functions of Multiwfn require different types of input file, the information needed by each function
are shown in the final line of corresponding section. Please choose proper type of input file
according to the table in Section 2.5.
[There is no Section 3.1 due to history reason ;-D]
3.2 Showing molecular structure and viewing orbitals /

isosurfaces (0)
When main function 0 is selected, information of all atoms as well as basic information of
featured orbitals (e.g. index of HOMO and LUMO, HOMO-LUMO gap) will be printed on text
window, meanwhile molecular structure will be shown in a GUI window. If the input file contains
orbital information, the orbitals can be viewed by selecting corresponding orbital index in the list at
right-bottom corner of the window, or by directly inputting orbital index in the text box, see below
screenshot. If the input file is .cub or .grd format, then one can view isosurface of the grid data in
this interface.
44
3 Functions
Since most widgets in this GUI are self explanatory, I only mention some worthnoting points
here.
Viewing structure
"Ratio of atomic size" is the ratio of one-fourth of atom radius shown on screen to its van der
Waals (vdW) radius, so if the slide bar is dragged to 4.0, then what to be shown is vdW surface.
There are three predefined drawing styles of system structure, you can activate one of them via
“Use CPK style”, “Use vdW style” and “Use line style” in “Other settings” in the menu.
The perspective can be adjusted by "Left", "Right", "Up", "Down", "Zoom in" and "Zoom out"
buttons. To exactly control the viewpoint, you can also directly input value of rotation angle by
selecting "Set rotation angle" option in "Set perspective" in the menu. In addition, you can rotate
your mouse wheel in the drawing region to zoom in and zoom out.
Multiwfn determines if two atoms are bonded by empirical distance criteria, if the distance
between two atoms is short than 1.15 times of the sum of their CSD covalent radii, they will be
considered as bonded. You can adjust this criteria by dragging "Bonding threshold" slide bar.
Note: If your input file contains connectivity information, such as .mol and .mol2, then the bonds will be
displayed directly according to the provided connectivity.
The color of bonds, atomic labels and atom spheres can be set by parameters "bondRGB",
"atmlabRGB" and "atmcolorfile" in settings.ini, respectively. See corresponding comments for
detail. After adjusting these parameters, you need to reboot Multiwfn to make them take effect. You
can also directly change atomic label color in the GUI via “Other settings”-“Set atomic label color”.
Viewing orbitals
By clicking "Orbital info." and then selecting corresponding options, basic orbital
information can be printed on text window.
All orbital indices are listed in the right-bottom box, default selection is “none” (no orbital is
shown). If you click an orbital index, Multiwfn will calculate grid data of wavefunction value for
corresponding orbital, and then the orbital isosurface appears immediately. Green and blue parts
correspond to positive and negative regions, respectively. The “Isovalue” slide bar controls the
isovalue of the isosurface.
For efficiency consideration, the default quality of grid data is relatively coarse, the number of
grid points can be set by "Isosur. quality", the larger number leads to the smoother isosurface (Note
that after you adjust the number, present isosurface will be deleted). You can also set default number
of grid points through changing “nprevorbgrid” in settings.ini.
For unrestricted wavefunction, the alpha and beta orbitals are recorded separately. Assume that
there are Na alpha orbitals and Nb beta orbitals, then the first Na and latter Nb orbitals in the orbital
selection list correspond to alpha and beta orbitals, respectively. Whereas in the orbital selection
box, negative index corresponds to beta orbitals. For example, if you input -9, then the 9th beta
orbital will be shown, and the orbital selection list will be automatically switched to the Na+9 term.
Sometimes it is useful to display two orbitals simultaneously; for example, analysis of phase
overlapping of two NBOs. This can be realized by "Show+Sel. Isosur#2" (Sel.=Select) check box.
This check box is inactive by default, once you have chosen an orbital (corresponding isosurface
will be referred to as Isosurface #1), this check box will be actived. After you clicked the check box,
the isosurface of the orbital you newly selected (which will be referred to as Isosurface #2) will be
shown together with the orbital you previously selected, yellow-green and purple parts correspond
to positive and negative regions respectively. If the check box is deselected, Isosurface #2 will
45
3 Functions
disappear, then if you reselect the check box the same Isosurface #2 will be redrawn.
The representation styles of Isosurface #1 and #2 can be adjusted by suboptions in "Isosur.
#1" and "Isosur. #2" individually, available styles include solid face (default), mesh, points, solid
face+mesh and transparent face. (Notice that Multiwfn cannot plot only one isosurface as
transparent face). The colors for face and mesh/points can be set by corresponding suboptions too,
users will be prompted to input Red, Green, Blue components for positive and negative parts in turn,
the component should between 0.0 and 1.0. For instance, 1.0,0.0,0.0 corresponds to pure red while
1.0,1.0,0.0 corresponds to Yellow. The default values for positive and negative parts can be set by
"isoRGB_same" and "isoRGB_oppo" in settings.ini, respectively. Opacity for transparent face can
also be customized, valid range is from 0.0 (completely transparent) to 1.0 (completely opaque).
Before showing orbital isosurfaces, Multiwfn first sets up a box internally, then calculates
orbital wavefunction value at the points evenly placed in the box. The default extension distance for
setting up the box is controlled by "Aug3D" parameter in settings.ini. The default value is suitable
and efficient for general cases; however, in rare cases (e.g. visualizing Rydberg orbitals) you may
need to manually enlarge the extension distance, you can either change the Aug3D to modify the
default setting, or select "Other settings"-"Set extension distance" and then input a value to modify
extension distance for present instance.
When orbital isosurface is portrayed as solid face, in rare cases the plotting effect is not quite
good, for example positive and negative regions are difficult to be distinguished, in these cases you
can try to adjust lighting setting by choosing "Other settings"-"Set lighting" or properly rotate the
system to bypass this issue.
Tools submenu
In the "Tools" of the menu bar, there are several options:
• Write settings to GUIsettings.ini: Write visualization state (e.g. color, view angle, molecular
representation, text size, etc.) to GUIsettings.ini in current folder. If "Multiwfnpath" environment
variable has been defined, the GUIsettings.ini will be written to the folder defined by this
environment variable instead.
• Load settings from GUIsettings.ini: Load visualization state from GUIsettings.ini in current
folder, so that you can quickly retrieve previous visualization state. If "Multiwfnpath" environment
variable has been defined, Multiwfn will first try to load this file from the folder defined by this
environment variable.
• Measure geometry: By inputting two, three and four atom indices in the boxes and press
ENTER button, the bond length, angle and dihedral will be returned.
• Batch plotting orbitals: Via this tool you can very conveniently save a lot of selected orbitals
to respective image file in current folder, see https://youtu.be/SHwrQhqBHZ0 for video illustration.
• Select fragment: After selecting it and input an atom index, the whole fragment where the
atom attributes to will be highlighted, and the indices of all atoms in the fragment will be returned.
This is useful when you perform some analyses, in which you need to input atom indices in the
fragment of interest.
More functions and illustration of the use of the main function 0 are described in Section 4.0.
Information needed: GTFs (only for viewing orbitals), atom coordinates, grid data (only for
viewing isosurface of grid data)
46
3 Functions
3.3 Outputting all properties at a point (1)
Input coordinate of a point or index of an atom, then the values of all real space functions
supported by Multiwfn at the point or corresponding nuclear position will be printed on screen, as
well as each component of gradient and Hessian matrix of specified function (default is electron
density). The function can be specified by f?, for example f9 select ELF, you can input allf to check
all available functions. The orbital whose wavefunction value will be outputted can be selected by
command o?, for example o4 choose the 4th orbital. If the input file merely contains atom
coordinates (such as pdb file), then only limited functions based on electron density will be outputted,
the electron density used is promolecular density constructed from fitted free atom density, see the
introduction of real space function 14 in Section 2.6.
If input d, value of a real space function at a given point can be decomposed into orbital
contributions. The contribution from orbital i is evaluated as follows: First set occupation number
of all orbitals to zero except for orbital i, and then calculate real space function as usual. Note that
sum of contribution of all orbitals may be different to the result when all orbitals are simultaneously
into account, since many real space functions contain non-linear operators, such as ELF.
Electrostatic potential is the most expensive one among all of the real space functions supported
by Multiwfn. If you are not interested in it, you can set "ishowptESP" parameter in settings.ini to 0
to skip calculation of electrostatic potential.
Information needed: GTFs (depending on the choice of real space function), atom coordinates
3.4 Outputting and plotting specific property in a line (3)
In this function, what you should do is just selecting a real space function and then define a
line. There are two ways to define the line:
(1) By inputting indices of two atoms, the line will be automatically extended by a small
distance in each side, the extended distance can be adjusted by "aug1D" in settings.ini or by the
option “0 Set extension distance for mode 1”, default value is 1.5 Bohr.
(2) By inputting the coordinates of the two endpoints.
Generally, the calculation only takes a few seconds, then curve map pops up, like this:
47
3 Functions
The gray dashed line indicates the position of Y=0. If the line is defined by the second way,
two red circles with Y=0 will appear in the graph, they indicate the position of the two nuclei. Click
right button on the graph and then you can select what to do next, you can redefine the scale of Y-
axis, export the data to line.txt in current directory, save the graph to a file, locate minimal and
maximal positions and so on. Note that the process for searching stationary points and the position
where Y equals to specified value is based on the data you have calculated, that means the finer the
points, the more accurate X position you will get.
The data points are evenly distributed in the line, the number of points is 3000 by default,
which is fine enough for most cases. The number can also be adjusted by “num1Dpoints” parameter
in settings.ini. Of course, the more points the more time is needed for calculating data. Notice that
for ESP calculation, the number of points is decreased to one-sixth automatically, because it is much
more time-consuming than other task.
Information needed: GTFs (except for ESP from nuclear/atomic charges and promolecular
approximation version of RDG and sign(2)), atom coordinates
3.5 Outputting and plotting specific property in a plane (4)
The basic steps of using this function are listed below. Users can finish all operations by simply
following program prompts, there are only a few key points are needed to be described.
1. Select a real space function
2. Select a graph type
3. Set the number of grid points in both dimensions
4. Define a plane
5. View the graph
6. Post-processes (adjust plotting parameters, save graph, export data to plain text file, etc.)
48
3 Functions
Worthnotingly, after you enter this function, you will see real space function selection menu; if you select option
111 (a hidden option), then the real space function to be plotted will be Becke weight of an atom or Becke overlap
weight between two atoms. If you select option 112 (another hidden option), Hirshfeld weight of a given atom or
fragment will be plotted.
If the parameter "iplaneextdata" in settings.ini is set to 1, then the data in the plane will not be calculated by
Multiwfn internally but directly loaded from an external plain text file. The coordinate of the points in the plane will
be automatically outputted to current folder, you can use third-party tools (e.g. cubegen utility of Gaussian) to
calculate function value at these points.
3.5.1 Graph types
Currently Multiwfn supports seven graph types for exhibiting data in a plane.
1 Color-filled map. This type of map uses different colors to represent real space function
value in different regions. By default, "Rainbow" coloring method is employed to map the function
value to color, and if the function value exceeds lower (upper) limit of color scale, then the region
will be filled by black (white). By selecting "Set color transition" in post-processing menu, you can
choose other color transition methods.
By selecting “Enable showing contour lines“ in post-processing menu, contour lines can be
plotted together on the graph.
2 Contour line map. This map uses solid lines to represent positive regions, and dashed lines
to exhibit negative regions.
49
3 Functions
The number of isovalue lines can be adjusted by user (see Section 3.5.4), you can also mark
the isovalues on the contour lines by using the option “2 Enable showing isovalue on contour lines”
in post-processing step.
Only for contour line map, if the real space function you selected to be plotted is orbital
wavefunction, you can not only plot one orbital by inputting one orbital index, but also plot two
orbitals simultaneously by inputting two orbital indices (e.g. 3,5), see Section 4.4.5 for example.
Option 4 in post-processing menu is used to toggle if showing atoms labels or reference point
(involved in some real space functions such as correlation hole and source function) in the graph.
The size and form (i.e. if showing atom index or element name) of the atom labels can be set by
"pleatmlabsize" and "iatmlabtype" parameters in settings.ini, respectively. The reference point is
represented as a blue asterisk on the graph. By default only when the distance between the atoms
and the plane is smaller than "disshowlabel" in settings.ini the corresponding atom labels will be
shown. To change this distance threshold, you can either adjust this value in setting.ini or choose
option 17 in post-processing menu. If "iatom_on_plane_far" parameter in settings.ini is set to 1,
then even if the distance is larger than the threshold, the label will still be shown, but in thin text
rather than bold text.
Option 8 in post-processing menu is used to toggle if showing bonds. The bond is shown for
an atom pair if these two conditions are satisfied: (1) The distance between the two atoms is smaller
than bonding threshold, which can be adjusted in the GUI window of main function 0 using
corresponding scale bar. (2) Both atoms are close enough to the plotting plane (smaller than the
"disshowlabel" parameter mentioned above).
Option 15 in post-processing is used to plot a contour line corresponding to vdW surface
(electron density=0.001 a.u., which is defined by R. F. W Bader in J. Am. Chem. Soc., 109, 7968
(1987)). This is useful to analyze distribution of electrostatic potential on vdW surface. Such a
contour line can be plotted in gradient line and vector field map too by the same option. Color, label
size and line style of the contour line can be adjusted by Option 16.
Notice that the discussion in above two paragraphs also applies to color-filled map, gradient
50
3 Functions
lines map and vector field map (see below).
3 Relief map. Use height to represent value at every point. If values are too large they will be
truncated in the graph, you can choose to scale the data with a factor to avoid truncation. The graph
is shown on interactive interface, you can rotate, zoom in/out the graph.
4 Shaded surface map and 5 Shaded surface map with projection. The relief map is shaded
in these two types. The latter also plots color-filled map as projection. The meshs on the surface can
disabled at post-processing stage.
51
3 Functions
6 Gradient line map with/without contour lines. This graph type represents gradient
direction of real space function, you can determine if the contour lines is also shown on the graph.
Note that since gradients of real space function are needed to be evaluated, and graphical library
needs to take some time to generate gradient lines, the calculation and plotting costs are evidently
higher than other graph types.
In the option "11 Set detailed parameters of plotting gradient line" at the post-processing menu,
you can set various plotting parametes:
• Suboption 1: Integration step for gradient lines, the smaller the value the finer the graph
• Suboption 2: Interstice between gradient lines, the smaller the value the denser the lines
• Suboption 3: Criteria for plotting new gradient line, try to play with it and you will know how this
parameter affects the graph
• Suboption 4: Integration method. The RK4 is the most robust but most expensive. The default
RK2 is good balance between accuracy and speed
• Suboptions 5 and 6: Color and width of gradient lines
7 Vector field map with/without contour lines. This graph type is very similar to last graph
type, however the gradient lines are replaced by arrows, which distribute on grids evenly and
represent gradient vectors at corresponding point. You can set color of arrows, or map different
colors on arrows according to magnitude of function value, you can also invert the direction of
arrows. The option 10 is worth mentioning, if you set upper limit for scaling to x by this option, then
if the norm of a gradient vector exceeds this value, the vector will be scaled so that its norm equals
to x.
52
3 Functions
3.5.2 Setting up grid, plane and plotting region
When program asking you to input the number of grid points in both dimensions, you can input
such as 100,150, which means in dimensions 1 and 2 the number of grid points are 100 and 150,
respectively, so total number is 100*150=15000, they are evenly distributed in the plotting region.
For “Relief map”, “Shaded surface map” and “Shaded surface map with projection”, commonly I
recommend 100,100; if this value is exceeded, the lines in the graph will look too crowd. For other
graph types I recommend 200,200. Of course the picture will become more pretty and smoother if
you set the value larger, but you have to wait more time for calculation. Bear in mind that total ESP
calculation is very time-consuming, you’d better use less grid points, for previewing purpose I
recommend 80,80 or less.
Multiwfn provides 7 modes to define the plotting plane:

1. XY plane: User inputs Z value to define a XY plane uniquely.
2. XZ plane: User inputs Y value to define a XZ plane uniquely.
3. YZ plane: User inputs X value to define a YZ plane uniquely.
4. Define by three atoms: Input indices of three atoms to define a plane by their nuclear
coordinates.
5. Define by three points: Input coordinates of three points to define a plane.
6. Input origin and transitional vector: This way is only suitable for expert, the two inputted
translation vectors must be orthogonal.
53
3 Functions
7. Define a plane parallel to a bond and meantime normal to a plane defined by three atoms
For modes 1~5, the actual plotting region is a subregion of the plane you defined. Multiwfn
automatically sets the plotting region to tightly enclose the whole molecule (for modes 1, 2 and 3)
or cover the three nuclei / points you inputted (for modes 4 and 5), finally the plotting region is
extended by a small distance to avoid truncating the interesting region. The extension distance is
4.5 Bohr by default, if you find the region you interested in is still be truncated, simply enlarging
the value by option “0 Set extension distance for plane type 1~5”, you can also directly modify the
default value, which is controlled by “Aug2D” parameter in settings.ini.
Below diagram illustrates how the actual plotting region is determined when you select mode
4 or 5 to define the plotting plane. The X and Y axes shown in the graph correspond to the actual X
and Y axes you finally see. Evidently, the input sequence of the three points or atoms directly affects
the graph.
For the modes 4 and 5, if you find the content is skewed in the final graph, or the interesting
part is not located at the center of the graph, you can choose "-1: Set translation and rotation of the
map for plane types 4 and 5" before selecting mode 4 or 5. For example, if you find the content in
the graph your previously plotted should be translated by (-3,1.5) Bohr and then rotated by 35, then
in this option you should first input -3,1.5 and then input 35, the resulting graph will meet your
expectation. (A practical instance of using this option was posted on http://bbs.keinsci.com/thread-
11037-1-1.html)
For mode 6, the plotting region is determined as follows, in which each black arrow denotes
translational vector 1, each brown arrow denotes translational vector 2, blue point denotes origin
point. The number of arrows is the number of grids set by users. Evidently, this mode enables users
to fully control the plotting plane setting.
54
3 Functions
Mode 7 is very useful when you want to define a plotting plane cutting a bond but there is no
proper reference atoms to use mode 4, below graph is such an instance, the purple rectangle is the
plane you want to plot:
To define this plane, you should select mode 7, and then input 3,5 to use these two atoms to define
the axis that the plotting will be parallel to, and then input 2,3,10 (or 2,5,10 etc.) to use them to
define a plane that the plotting plane will be normal to. After that you need to input the length of X
and Y axes, e.g. 10 Bohr and 7 Bohr, respectively.
3.5.3 Options in post-processing interface
After plotting the graph, you will see a menu, in which there are a lot of options used to adjust
or improve the quality of the graph. Since many of them have already been introduced above and
some of them will be mentioned in next sections, and lots of them are self explanatory, only a few
will be mentioned here.
-9 Only plot the data around certain atoms: Sometimes in the graph only a few regions are
interesting; if you want to screen other regions, you may find this option useful. After selecting this
option, assume that you input 2,4,8-10, then only the real space function around atoms 2,4,8,9,10
will be plotted (the data to be plotted then in fact is the original plane data multiplied by the Hirshfeld
weight of the fragment you inputted). Next time you select this option, the original data will be
recovered.
-7 Multiply data by a factor: This is mainly used to scale the range of the plane data.
-6 Export the current plane data to plane.txt in current folder: After using this option to
export the plane data, you can very conveniently using third-part plotting softwares such as
Sigmaplot to redraw the data.
55
3 Functions
-2 Set stepsize in X, Y (,Z) axes: This option determines the spacing between the labels in the
coordinate axes. If axis labels are not shown, that means you need to decrease the stepsizes.
-1 Show the graph again: After adjusting plotting parameters, choose this option to replot the
graph to check the effect.
0 Save the graph to a file: Export the graph to a graphic file in current folder. See Section 2.8
on how to determine the graphic format and size.
3.5.4 Setting up contour lines
For graph types 1, 2, 6 and 7, the contour lines can be plotted (if not shown, select "2 Enable
showing contour lines" in the post-processing menu). There is also an option “Change setting of
contour lines” in the post-processing menu. In this interface, values of current contour lines are first
listed and you can modify them by using below suboptions:
Option 1: Save current setting and return to upper menu. Then if you select “Show the graph
again”, the graph with new isovalue setting will appears.
Option 2: Input a new value to replace old value of a contour line.
Option 3: Add a new contour line and input the isovalue for it.
Option 4: Delete some contour lines.
Option 6: Export current isovalue setting to a plain text file, you can use this function to save
multiple sets of your favourite isovalue settings for different systems and real space functions.
Option 7: Load isovalue setting from external file, the format should be identical to the file
outputted by subfunction 6.
Option 8: Generate isovalues according to arithmetic sequence, user need to input initial value,
step size and total number. For plotting ELF/LOL, I suggest user input 0,0.05,21 to generate
isovalues in the range 0.0~1.0 with step size 0.05. You can choose if cleaning existing contour lines,
if you select n, then new contour lines will be appended to old ones.
Option 9: Like function 8, but according to geometric series.
Option 10: Some contour lines can be bolded with this function, by default no line is bolded.
To bold some lines, select this function and input how many lines you want to bold, then input
indices of them in turn. If there are some lines already been bolded, selecting this function will
unbold all of them.
Option 11: Set color for positive contour lines, you need to input a color index.
Option 12: Set line style for positive contour lines, you need to input two integer number, the
first one denotes the length of line segment, the second one denotes the length of interstice between
line segment. For example 10,15 means positive contour lines are composed of line segments with
length of 10 and spaces with length of 15 alternatively.
Option 13, 14: Like option 11 and 12, but for negative part.
Option 15: If this option is selected, the positive and negative lines will be set to "crimson,
width=6, solid line" and "blue, width=6, long dashed line", respectively. Then the saved picture will
be very suitable for publication; as you can see in the resulting graphical file, the two kinds of lines
are clear and can be distinguished easily.
56
3 Functions
3.5.5 Plot critical points, paths and interbasin paths on plane graph
CPs and paths can be plotted on color-filled map, contour line map, gradient line map and
vector field map. Below is a typical electron density gradient line map containing critical points,
topology paths and interbasin paths.
In order to append the CPs and paths on plane map, before drawing the plane graph using main
function 4, you need to enter topology analysis module (main function 2), then search CPs and
generate paths. After that, return to main menu and draw plane graph as usual, you will find that the
CPs and paths have appeared on the graph. In the graph, brown, blue, orange, green dots denote (3,-
3), (3,-1), (3,+1), (3,+3) critical points, respectively. Bold dark brown lines depict bond paths.
In the option “4 Set details of plotting critical points and paths” at post-processing menu, you
can choose which types of CPs are allowed to be shown, and you can set size of markers, thickness,
distance threshold and color of path lines. By default, if the vertical distance between a CP or a point
in a path and the plotting plane exceeds 0.5 Bohr, the CP or path point will not be shown, the
thresholds can be altered by “8 Set distance threshold for showing CPs” and “9 Set distance
threshold for showing paths”.
Interbasin paths (deep blue lines in above graph) are derived from (3,-1) CPs, these paths
dissect the whole space into individual atomic basins. You can also append the interbasin paths on
contour line map, gradient line map and vector field map. In order to draw interbasin paths, you
should first confirm that at least one (3,-1) CP has been found in topology analysis module and it is
close enough to current plotting plane (smaller than "disshowlabel" in settings.ini), then you can
find a option "Generate and show interbasin paths" in post-processing stage, choose it, wait until
57
3 Functions
the generation of interbasin paths is completed, then replot the plane graph again, you will find these
interbasin paths have already presented.
If you hope the interbasin paths become shorter or longer, choose option "Set stepsize and
maximal iteration for interbasin path generation" in post-processing stage before generating
interbasin paths, you will be prompted to input stepsize and the number of iterations, the maximum
length of yielded interbasin paths equals to product of the two values. Note that if distance between
a point in interbasin path and the plotting plane exceeds the "disshowlabel" in settings.ini,
corrsponding points in the interbasin path will not be shown on the plane graph.
Information needed: GTFs (depending on the choice of real space function), atom coordinates
3.6 Outputting and plotting specific property within a

spatial region (5)
The main purpose of this function is calculating grid data or generating Gaussian-type cube
file (.cub) for specific real space function. The .cub file is supported by a lot of chemistry
visualization softwares, such as VMD, GaussView, ChemCraft and Molekel. The isosurface of
generated grid data can be viewed directly in Multiwfn too. This function can also be used to
calculate function values for a set of points recorded in an plain text file.
The basic procedure of using this function is:
(1) Select a real space function
(2) Set up grid
(3) Use post-processing options to visualize isosurfaces, modify and export data
In the real space function selection menu, if you select option 111 (a hidden option), then the real space function
to be calculated will be Becke weight of an atom or Becke overlap weight between two atoms. If you select option
112 (another hidden option), Hirshfeld weight of a given atom or fragment will be calculated.
Setting up grid
Multiwfn provides many modes for setting up grid point:
Mode 1: Low quality grid, about 125000 points in total (corresponding to 50*50*50 grid,
assuming the spatial region is cubic), this mode is recommended for previewing purpose.
Mode 2: Medium quality grid, about 512000 points in total (corresponding to 80*80*80 grid,
assuming the spatial region is cubic). For small molecular, this quality is enough for most analysis.
Mode 3: High quality grid, about 1728000 points in total (corresponding to 120*120*120 grid,
assuming the spatial region is cubic).
For modes 1, 2 and 3, the actual number of points in each direction is automatically determined
by Multiwfn so that grid spacing in each direction are nearly equal. The method of determining
spatial scope of grid data is illustrated below in two-dimension case.
58
3 Functions
First, Multiwfn sets up a box (red dashed line) to just enclose the entire molecule, and then the
box is suitably extended in each dimension to avoid truncating boundary part of isosurfaces, the
green rectangle in below graph is the actual spatial scope of grid data. If the extension distance is
inappropriate, you can customly set the value by using option “-10 Set extension distance of grid
range", or directly set default value by modifying “Aug3D” in settings.ini.
The so-called “Low quality”, “Medium quality” and “High quality” are only relative to small
system. If spatial range of the grid data is wide, then density of grid points of mode 3 will not be
high and should be called “Medium quality”, and mode 2 should be called “Low quality” at this
time. While for very small system, the density of mode 1 is already adequate for general purposes.
Mode 4: Specify the number of grid points or grid spacing in X, Y and Z directions by yourself,
the spatial scope is determined automatically as shown above.
Mode 5: Specify all details of grid setting by user, including the number of points in X, Y and
Z directions, initial point, translation vectors. This mode is useful for experienced users.
Mode 6: Specify the center coordinate, number of points and extension distance in X, Y and Z
directions. For example, the center coordinate you inputted is 2.3,1.0,5.5, the extension distance in
X, Y and Z is k,k,m, then the coordinates of two most distant endpoints are (2.3-k,1.0-k,5.5-m) and
(2.3+k,1.0+k,5.5+m). This mode is useful for analyzing local properties.
Mode 7: Like function 6, but inputting indices of two atoms instead of inputting center
coordinate, the midpoint between the two nuclei will be set as center. This mode is very useful for
weak interaction analysis by RDG function. For examples, we want to study the weak interaction
region between the dimer shown below, and we found C1 and C14 may enclose this region, so we
input 1,14. see Section 3.23.1 for example. If the two atom indices are identical, then the nuclear
coordinate will be set as center.
59
3 Functions
Mode 8 Use grid setting (origin, number of points and grid spacing) of an existing cube file.
You will be prompted to input the file name. This mode is useful to generate multiple cube files with
exactly identical grid setting.
Mode 10 Set up the grid data in a GUI window. This mode is quite convenient, the box is
clearly visible in the graphic window, and you can directly change box length, box center and grid
sapcing by dragging corresponding scale bars (the X,Y,Z directions share the same grid spacing).
The total number of grids under present grid setting can be directly seen at the right bottom side of
the window.
60
3 Functions
Notice that the more grid points you set, the finer the isosurface graph you will get, however
the more time is needed for calculating grid data and generating isosurface graph. The increase of
calculation time with the number of points is nearly linear.
After setting grid points, the program starts to calculate grid data, once the calculation is
finished, the minimum/maximum and corresponding coordinate, the sum of all/positive/negative
data multiplied by differential element are printed on screen immediately. If what you calculated is
electron density, the molecular dipole moment evaluated based on grid data is also printed out. Then
you will see post-processing menu.
Post-processing options
In the post-processing menu if the option “-1 Show isosurface graph” is selected, a GUI
window will pop up, which shows the isosurface of grid data, all widgets are self explanatory, you
can change isosurface value by both dragging slide bar (upper and lower limit are -5 and +5
respectively) or inputting precise value in text box (then press “Enter” button in you keyboard); if
“Show data range” is selected, the spatial scope will be marked by a blue frame as the one in above
picture (if spatial scope exceeds the range of coordinate axis, the frame will no be displayed). The
isosurfaces with the same and reverse sign of current isovalue are in green and blue respectively.
Option 1 can export isosurface graph to graphic file in current directory. Option 2 can export
grid data to .cub file in current directory. By selecting option 3 the grid data will be exported to plain
text file output.txt in current directory. Using option 4 you can set isovalue without entering GUI
window and dragging slide bar, it is useful for batch process and in command-line environment.
With options 5~8, you can perform addition, subtraction, multiplication and division operations on
the grid data, respectively.
Special note: Calculate data for a set of arbitrarily distributed points

If you want to use Multiwfn to calculate real space function value for a set of points (may be
arbitrarily distributed), in the interface for setting up grid, you should choose "100 Load a set of
points from external file" and then input the path of the plain text file recording the points. In the
file the first line should be the number of points recorded, and followed by X/Y/Z coordinates of all
points. For example
1902
-3.3790050 -2.0484700 0.0274117
-3.3844930 -1.9472468 -0.2500274
-3.4064601 -1.9221635 -0.1287148
-3.4118258 -1.9232203 -0.0003350
-3.4059106 -1.9218280 0.1351179
....
The coordinates must be given in Bohr. The format is in free format, and can have more than four
columns of data, but the columns after the fourth one will be simply ignored.
Multiwfn will load coordinates of the points from this file, and then calculate function value
for them. Finally, the X/Y/Z coordinate and function value of all points will be outputted to a plain
text file, whose path is specified by users.
Information needed: Atom coordinates, GTFs (depends on the choice of real space function)
61
3 Functions
3.7 Custom operation, promolecular and deformation

properties (subfunction 0, -1, -2 in main function 3, 4, 5)
3.7.1 Custom operation for multiple wavefunctions (0)
In main function 3, 4 and 5, there is a subfunction allow you to set custom operation for
multiple wavefunctions. Supported operators include + (add), − (minus), * (multiply), / (divide),
there is no upper limit of the number of wavefunctions involved in custom operation. For example,
if the first loaded wavefunction after booting up Multiwfn is a.wfn, then in the setting step of custom
operation you inputted 2 (viz. there are two wavefunctions will be put into “custom operation list”
and thus will be operated with a.wfn in turn), then you inputted -,b.wfn and *,c.wfn, the property
finally you get will be [(property of a.wfn) - (property of b.wfn)] * (property of c.wfn). If you are
confused, you can consult the example in Sections 4.5.4 and 4.5.5.
Sometimes the molecular structure in the first loaded file and that in the subsequently loaded
files are not identical, the grid points you set will be for the first loaded file, all of the other files will
share the same grid setting.
Avoid using custom operation in conjunction with main functions -4, -3 and 6, otherwise you
may get absurd result.
3.7.2 Promolecular and deformation properties (-1, -2)
If you selected subfunction -1 in main functions 3, 4 or 5 before choosing a property (viz. a

real space function), what you finally get will be promolecular property. Promolecular property is
the superposition property of atoms in their free-states
P pro (r ) =  PAfree (r − R A )
A
If the property you chose is electron density, then the promolecular property is generally
referred to as promolecular density
 pro (r ) =   Afree (r − R A )
A
This is an artificial density that corresponds to the state when molecule has formed but the density
has not relaxed.
Deformation property is the difference between actual property and promolecular property of
a molecule under the same geometry
P def (r) = P mol (r) − P pro (r)

If the property is chosen as electron density, then the deformation property is generally called
deformation density or known as electronic bonding charge distributions (BCD), which is very
useful for analyzing charge transfer and bonding nature. Application example of deformation
density can be found from my paper Acta Phys. -Chim. Sin., 34, 503 (2018).
62
3 Functions
3.7.3 Generation of atomic wavefunctions
Evaluation of promolecular and deformation properties and some functions in Multiwfn

request atomic wavefunction files, such as calculating Hirshfeld, VDD and ADCH charges, fuzzy
atomic space analysis and orbital composition analysis based on Hirshfeld partition, etc. The process
of generating atomic .wfn files are exactly the same.
After you select subfunctions -1 or -2 to study promolecular and deformation properties,
Multiwfn checks whether .wfn files of all elements involved in present system have been presented
in “atomwfn” subdirectory of current directory, if not, Multiwfn automatically invokes Gaussian to
generate the missing element .wfn files and sphericalizes their densities. If the path of Gaussian
executable file (“gaupath” parameter in settings.ini) is incorrect or has not been defined, Multiwfn
will ask you to input the path of Gaussian executable file. The basis set for generating the element
wavefunctions can be arbitrarily set by user, however it is suggested to use the same basis set as the
molecular wavefunction.
The newly generated element wavefunction files or those taken from “atomwfn” directory are
stored in “wfntmp” subdirectory in current directory. They will be translated to actual position of
the atoms in present system, meanwhile atomic indices will be added to the .wfn filename (e.g. "Cr
30.wfn"). These are the files will directly be used to calculate promolecular and deformation
properties.
Details and skills

If you want to avoid the step of generating element wavefunctions every time, you can move
the.wfn files that without number suffix (such as C .wfn) from “wfntmp” directory to “atomwfn”
directory (if “atomwfn” directory is non-existing, build it by yourself), next time if Multiwfn
detected that all needed element .wfn files have already been presented in“atomwfn” folder, then
Multiwfn will directly use them. Multiwfn only invokes Gaussian to calculate the missing
element .wfn files.
“atomwfn” subdirectory in “examples” directory contains element wavefunction files of all
first four rows, they were generated under 6-31G*, and they have been sphericalized. If you want
to use them, simply copy the "atomwfn" directory to current folder.
There is a quick way to generate all first four-rows element wavefunction files: the file
“examples\genatmwfn.pdb” contains all first four-rows atoms, load it into Multiwfn and generate promolecular
property, after the calculation of element wavefunctions is finished, copy the .wfn files that without suffix in “wfntmp”
directory to “atomwfn” directory.
If your system involves elements heavier than Kr, Multiwfn is unable to generate atom
wavefunction files by invoking Gaussian and sphericalize their density automatically; in that cases
you have to calculate and sphericalize atom wavefunctions manually, and then put them into
“atomwfn” directory, Multiwfn will directly use them.
The default theoretical method for generating wavefunction for main group elements in first
four-rows (index from 1 to 20 and 31 to 36) is ROHF, for the transition metals in the fourth row,
UB3LYP is used. In general, the promolecular and deformation properties are insensitive to
theoretical method. If you want to specify theoretical method by yourself, you can input theoretical
method and basis set at the same time with slash as separator, for example BLYP/6-311G*. Do not
add “RO” or “U” prefix since they will be added automatically. If error occurs during generation of
atom wavefunctions, please check Gaussian input and output files in “wfntmp” directory carefully.
63
3 Functions
Notice that the maximum character length of the path of .wfn file permitted by Gaussian is
only 60! The path will be truncated and cause error if the length exceeded this threshold. So do not
put Multiwfn in the directory with too long path!
3.7.4 Sphericalization of atom wavefunction
The main purpose of Multiwfn supporting promolecular and deformation property is for
generating promolecular and deformation density, however, electron density of most elements in
free and ground state is not in spherical symmetry, hence will lead to orientation dependence
problem. To tackle it, atomic electron density must be sphericalized. However, there is no unique
way to do this. In Multiwfn, atom electron density is sphericalized by modifying atom wavefunction
artificially, here I describe the detail. If you want to skip the sphericalization step, simply set the
“ispheratm” in settings.ini to 0.
For elements in IV A group, Multiwfn uses sp3 configuration to replace s2p2 ground state by
default. This treatment is reasonable, since in most molecules these atoms are in sp3 hybridization.
For VI A, VII A and Fe, Co, Ni, Multiwfn equalizes the occupation number of orbitals within the
same shell; for example, oxygen has two singly occupied 2p orbitals and one doubly occupied 2p
orbitals in ground state, the number of electrons in this shell is 4, so Multiwfn sets the occupation
number of all the three orbitals to 4/3. This method works because the shape of occupied orbitals
are always nearly identical, regardless of the original occupation number is one or two. However,
the difference between virtual orbital and occupied orbital is remarkable, thus Multiwfn uses another
method to sphericalize electron density of elements in III A group, as well as Sc, Ti and V. In this
method, the singly occupied orbitals are duplicated and rotated. Taking boron as example, assume
that the singly occupied 2p orbital is directing along Z-axis, Multiwfn replicates this orbital twice
and turn them toward X-axis and Y-axis respectively, finally the occupation numbers of the three
orbitals are set to 1/3. (Users needn’t to check the orientation of singly occupied orbital by
themselves). If you hope that Multiwfn sphericalizes atoms in IV A group in such manner instead
of using sp3 configuration, set “SpherIVgroup” in settings.ini to 1.
Notice that the sphericalization methods used in Multiwfn are closely related to wavefunction
type, the methods fail if unrestricted wavefunction is used for main group elements or restricted
open-shell wavefunction is used for transition metals. These methods also fail if Hartree-Fock
method is used for transition metals, because orbital order produced by HF is different from most
DFT cases (the HF’s order is wrong, 4s is higher than 3d).
3.8 Checking & Modifying wavefunction (6)
This function provides a lot of subfunctions for checking and modifying loaded wavefunction,
all subfunctions take effect immediately, so after modification you can print related information to
check if your operations are correct. Once you finished all modifications of wavefunction, you can
save current wavefunction to “new.wfn” file or return to main menu, all following tasks which make
use of wavefunction information will be affected. The title in this interface shows the number of
GTFs, orbitals, atoms and alpha/beta electrons of current wavefunction.
64
3 Functions
Below is explanation of each subfunction in this catagory.

Subfunction -1: Return to main menu.
Subfunction -3 and -4: Defining molecular fragment. In -3 you directly define fragment by
inputting atomic indices; while in -4 you should input the indices of the atoms to be excluded, the
remaining atoms will constitute the fragment. The GTFs whose centers do not belong to this
fragment will be discarded (can be imagined as corresponding GTF coefficients in all orbitals are
set to zero). This function affects all the following tasks which make use of GTF information, in
particular, the calculations involving real space functions.
When use these two functions? For large molecules, what you are interested in may be only a
small region, by discarding atoms in other region, the calculation speed of time-consuming task
such as generating grid data can be increased evidently. Besides, sometimes you may only want to
obtain real space function contributed by certain atoms, these two functions can fulfill your purpose.
This function can be safely used together with main function 1~6. For other tasks, DO NOT
use this function if you do not understand what you are exactly doing, otherwise you may obtain
wrong or meaningless results. Notice that this function is irreversible, the discarded GTFs cannot
be recovered, unless you reboot the program and reload the wavefunction file.
Subfunction 0: Save current modified wavefunction to new.wfn file in current directory.
Multiwfn can be used as an .mwfn/.fch/.molden...→wfn format converter through this function.
Notice that the orbitals with zero occupation number and the atoms do not have GTFs will be
automatically discarded during saving.
Subfunction 1: Print information of all GTFs, including the centers they are belonging to, GTF
types and exponents.
Subfunction 2: Print information of all basis functions, including the shells/centers they
attributed to, types, and corresponding GTF index ranges.
Subfunction 3: Print basic information of all orbitals, including energies, occupation numbers,
orbital types (Alpha, beta or alpha+beta).
Subfunction 4: Print detail information of an orbital, including the expansion coefficients with
respect to GTFs (along with GTF information), note that the coefficients include GTF normalization
constants.
Subfunction 5: Print coefficient matrix in basis function (not the coefficients with respect to
GTFs), only available when the input file contains basis function information.
Subfunction 6: Print one-particle density matrix in basis functions, only available when the
input file contains basis function information.
Subfunction 7: Print various kinds of integral matrix between basis functions, including
overlap integrals, electric/magnetic dipole moment integrals, kinetic energy integrals, velocity
integrals, electric quadrupole and octopole integrals. For overlap matrix, eigenvalues are printed
together, which are useful for checking linear dependency. Only available when the input file
contains basis function information.
Subfunction 11: Swap centers or types or exponents or orbital expansion coefficients of two
GTFs, or swap all information of two GTFs at once (identical to swap record order of two GTFs,
thus does not affect any analysis result)
Subfunction 21 to 24: Set the center, type, exponent and expansion coefficient of a specific
GTF in an orbital respectively.
Subfunction 25: Set the expansion coefficients of some GTFs that satisfied certain conditions
65
3 Functions
in some orbitals. The conditions you can set for GTFs include: Index range of GTFs, index range of
atoms, GTF types (you can input such as YZ, XXZ). The selected GTFs are intersection of these
conditions. When program is asking you for inputting range, you can input such as 3,8 to select
those from 3 to 8, input 6,6 to select only 6, especially, input 0,0 to select all.
If the input file also contains basis function information, by this option you can also set
expansion coefficients of basis functions.
Subfunction 26: Set occupation number of some orbitals. You can directly set them to a
specific value, or add, minus, multiply, divide them by a given value. This function is very useful
for shielding the contributions from certain orbitals to real space functions, namely setting their
occupation numbers to zero before calculating real space functions.
Subfunction 27: Set orbital type of some orbitals, then wavefunction type will be
automatically updated, please check the prompts shown on the screen.
Subfunction 28: Set energy of some orbitals. You can directly set them to a specific value, or
add, minus, multiply, divide them by a given value. This function is useful when you want to rectify
the orbital energies using a given relationship (e.g. J. Am. Chem. Soc., 121, 3414 (1999)) before
plotting density-of-states (DOS) map.
Subfunction 31: Translate the whole wavefunction and all atom coordinates of current system
by inputting translation vectors and their units.
Subfunction 32: Translate and duplicate the whole wavefunction and all atom coordinates of
current system, users need to input translation vectors, their units and how many times the system
will be translated and duplicated according to the translation vectors. This function is useful for
extending the primitive cell wavefunction outputted by Gaussian PBC function to supercell
wavefunction, of course you can calculate supercell wavefunction directly by Gaussian, but much
more computational time will be consumed. Notice that constructing wavefunction of supercell by
this way is only an approximation, because the mix between orbitals of neighbouring primitive cell
to new supercell orbitals is completely ignored.
Subfunction 33: Rotate wavefunction, namely X→Y, Y→Z, Z→X
Subfunction 34: Set occupation number of all inner molecular orbitals in present system to
zero, namely only contribution from valence orbitals will be reserved for following studies. The
number of inner MOs is automatically determined, howver, the atoms using pseudo-potential are
not taken into account.
Subfunction 35: This function is very useful if you want to discard contribution of orbitals
with certain irreducible representation (IRREP) in all kinds of analyses. After you enter this
subfunction, IRREP of all occupied orbitals will be shown, the "N_orb" denotes the number of
occupied orbitals belonging to the corresponding IRREP. If you want to discard contribution of
some IRREPs in the succeeding analyses, you can select option 1 and input the index of the IRREPs,
then the occupation number of corresponding orbitals will be set to zero and thus their contributions
are eliminated, and you will also see their status are changed from "Normal" to "discarded". Then
you can choose option 0 to save wavefunction and quit. You can also choose option 2 to recover the
original occupation number of all orbitals, or choose 3 to reverse the status of every IRREP between
"Normal" and "Discarded". Note that this subfunction only works for .mwfn, .molden and .gms files,
because only these files record orbital IRREPs (beware that the IRREPs in the .molden file produced
by many quantum chemistry programs are missing or incorrect). Besides, only restricted and
unrestricted SCF wavefunctions are supported.
66
3 Functions
Subfunction 36: Invert phase of some orbitals (i.e. replacing expansion coefficients of basis
functions and GTFs of these orbitals with their negative values).
Information needed: Basis function (only for subfunction 2, 5, 6, 7), GTFs, atom coordinates
3.9 Population analysis and atomic charges (7)
This module is used to calculate population number of basis functions, shells, atoms, fragments,
or atomic charges. There are some points should be noted:
Regardless of which subfunction you choose, if you want to evaluate fragment
charge/population (the sum of charge/population of the atoms in a fragment), you should select
option "-1 Define fragment" first and input atom indices to define the fragment. Then once the
calculation of atomic charges is finished, fragment charge will be printed along with atomic charges.
After atomic charge calculation, you can choose to output atomic coordinates with calculated
charges to [name of loaded file].chg file in current directory, see Section 2.5 for detail about .chg
format.
If the file loaded when Multiwfn boots up is .chg format, you can use option "-2 Calculate
interaction energy between fragments based on atomic charges" in present function to evaluate
interfragment electrostatic interaction energy between two given fragments by classic Coulomb
formula (see below) using the atomic charges in this .chg file.
q A qB
ECoul−int =  
Afrag 1 Bfrag 2 | R A − R B |
This feature is useful for studying electrostatic interaction component in total molecular interaction
energy, as well as evaluating intermolecular exciton coupling energy based on atomic transition
charges (see Section 4.A.9 on how to derive them).
3.9.1 Hirshfeld atomic charge (1)
Hirshfeld is a very popular atomic population method based on deformation density partition,
Hirshfeld charge is defined as (Theor. Chim. Acta (Berl.), 44, 129 (1977))
q A = −  wAHirsh (r )  def (r )dr
where
 def (r ) =  (r ) −  pro (r )
 pro (r ) =   Afree (r − R A )
A
 Afree (r − R A )
w AHirsh (r ) =
 pro (r )
The advantages of Hirshfeld population are:
1. Result is qualitatively consistent with general chemical concepts such as electronegativity
67
3 Functions
rule.
2. The weighting function w for space partition has clear physical meaning.
3. Unlike the methods based on integrating electron density such as AIM charge, what the
Hirshfeld charge reflects is the amount of transfered electron density during molecule formation,
the density not transfered is not involved.
4. Insensitive to the quality of wavefunction.
5. Although calculating Hirshfeld charge needs integration in real space, due to the smooth
integrand, sophisticated density functional theory (DFT) grid-based integration schemes can be
directly used, so Hirshfeld population is high-efficient.
6. The wide application field. Deformation density data can also be obtained by X-ray
crystallography experiments. Moreover, the applicability of Hirshfeld population is not constrained
by the type of wavefunction, the method can be directly applied to solid system, where the
wavefunction generally be described by plane-wave functions.
The disadvantages of Hirshfeld population are the charge is always too small and the poor
reproducibility of observable quantities, such as molecular dipole moment and ESP, the reason is
Hirshfeld population completely ignores atomic dipole moments.
After calculation, if the printed sum of all Hirshfeld charges is very close to integer, that means
the quadrature is accurate; if not, that means the outputted Hirshfeld charges are unreliable, you
need to increase the density of integration points by setting “radpot” and “sphpot” to larger value
and calculate again. For balancing computational time and accuracy, the default value of “radpot”
and “sphpot” are 75 and 434, respectively, you can set them to 100 and 590 respectively to obtain
more accurate results.
An example is given in Section 4.7.1.
Information needed: GTFs, atom coordinates
3.9.2 Voronoi deformation density (VDD) atom population (2)
The only difference between VDD and Hirshfeld population is the weighting function w. In
VDD population, the Voronoi cell-like partition is used, each cell corresponds to an atom, see J.
Comput. Chem., 25, 189 (2004) for details.
The results of VDD population are similar to Hirshfeld population in common, because the
magnitude of deformation density is always small, so there is no significant change in charges when
different weighting functions are used. The outputted terms are identical to Hirshfeld population.
Personally, I suggest you use Hirshfeld population instead of VDD.
Note that if the sum of VDD atomic charge deviates from molecular net charge evidently, that
means the numerical integration accuracy is not satisfactory, and hence you should enlarge "sphpot"
parameter in settings.ini to improve the result. "radpot" parameter also influences the result, but not
so significantly as "sphpot".
68
3 Functions
3.9.3 Mulliken atom & basis function population analysis (5)
Theory
Mulliken analysis is the oldest population method based on orbital wavefunction, supported by
almost all quantum chemistry packages.
Orthonormality condition of spin orbital wavefunction entails 1 =   (r)

2
d r , if we assume
that the orbital is real type and insert the linear combination equation  i (r ) = C
a
a ,i  a (r ) into
it, we get
2
 
1 =    Ca ,i  a (r )  d r = Ca2,i +  Ca ,i Cb,i S a ,b =  Ca2,i + 2 Ca ,i Cb,i S a ,b
 a  a a ba a a b a
where S a ,b =  a (r )  b (r ) d r , the normality of basis functions are used in derivation. The first
term is “local term”, denotes the net population of each basis function in orbital i, the second term
is “cross term”, denotes the shared electrons between basis function pairs in orbital i. Certainly the
local terms should be completely attributed to corresponding basis functions, however for cross
terms the partition method is not unique. Mulliken defined the population of basis function a in spin
orbital i as
 i ,a = C a2,i +  C a ,i Cb,i S a ,b
ba
That is each cross term 2C a ,i Cb,i S a ,b is equally partitioned to corresponding two basis functions.
The population number of atom A is simply the sum of population numbers of all basis
functions attributed to atom A in all orbitals. Mulliken atomic charge is defined as
q A = Z A −  ni , A = Z A −  i   i ,a
i i aA
where  is orbital occupation number, ni,A is the contribution from orbital i.

Mulliken analysis is not ideal for practical application due to these shortcomings:
1. Poor reproducibility of observable properties such as molecular dipole moment
2. The “equal partition” of cross term has no strict physical meaning
3. Very high basis set dependence. In particular, diffuse function must not be presented,
otherwise the result will be misleading.
4. Occasionally, meaningless result occurs (population number is negative).
Usage
Since amount of outputted information is huge for large systems, before selecting below
subfunctions, you can use option -1 to change the default output destination from screen to a specific
plain text file.
· Subfunction 1 (Output Mulliken population and atomic charges): Population of basis
69
3 Functions
functions, population of basis function shells, population of each angular moment of atomic orbitals,
as well as atomic charges are printed.
·Subfunction 2 (Output gross atomic population matrix and decompose it): Gross atom
population matrix is printed, from which you can get local terms of each atom (diagonal element)
and cross terms between each atom pair (non-diagonal element multiplies 2). The matrix is defined
as
 A,B =  iA,B = i  Ca ,i Cb,i Sa ,b
i i aA bB
Note that the last row of outputted matrix is the sum of corresponding column elements, that
is the total population number of corresponding atom. You can also choose to decompose the matrix
to contribution of each occupied orbital, the Ωi matrices will be outputted to groatmdcp.txt in

current directory.
By the way, the quantity 2iA, B is just the Mulliken bond order between atom A and B
contributed from orbital i, as we will see in Section 3.11.4.
·Subfunction 3 (Output gross basis function population matrix and decompose it): Gross
basis function population matrix can be outputted by this option for further analyzing detail of
electron distribution, the matrix element is defined as
a ,b =  ai ,b =  i C a ,i Cb,i S a ,b
i i
Likewise in subfunction 2, the last row of the outputted matrix is total population number of
corresponding basis function. You can also select to output contribution of each occupied orbital,
namely a,i b , to grobasdcp.txt in current directory (notice that this file is extremely large even for
medium-sized system).
·Subfunction 4 (Output orbital contributions to atomic populations to atmpopdcp.txt): This
option is used to output contributions of occupied orbitals to atomic populations, namely all
  
i
i
a A
i,a terms, to atmpopdcp.txt in current folder.
An example of Mulliken population analysis is given in Section 4.7.0.
Information needed: Basis functions
3.9.4 Löwdin atom & basis function population (6)
The only difference between Löwdin and Mulliken population analysis is whether Löwdin
orthogonalization is performed first. In Löwdin orthogonalization, the transformation matrix is
X = Us 0.5U H
where U is eigenvector matrix of overlap matrix, matrix s is diag(1, 2...), where {} is eigenvalue
set of overlap matrix. After Löwdin orthogonalization, the overlap matrix becomes identity matrix,
70
3 Functions
and new coefficient matrix is X −1C .

Ostensibly, Löwdin population avoids the partition for cross terms since they have become
zero, actually, the partition step is no more than hidden in the orthogonalization. Since Löwdin
orthogonalization still has no strong physical meaning, it cannot make conclusion that Löwdin
population is better than Mulliken. In view of practical results, one also found Löwdin charges have
no any evident advantages relative to Mulliken charges, though some people argued that Löwdin
charges have better basis set stability and reproducibility of molecular dipole moment. Besides,
Mayer et. al. found Löwdin population has rotation dependence to some extent when cartesian type
Gaussian basis functions are used, however the dependency can be safely ignored in generally, for
detail please see Chem. Phys. Lett., 393, 209 (1968) and Int. J. Quantum Chem., 106, 2065 (2006).
3.9.5 Modified Mulliken atom population defined by Ros & Schuit

(SCPA) (7)
Some people had proposed several different partition methods of cross term to improve
Mulliken analysis, they are generally called as modified Mulliken population analysis (MMPA). In
the method proposed by Ros and Schuit (Theo. Chim. Acta, 4, 1 (1966)), the composition of basis
function a in orbital i is defined as
C a2,i
 i ,a =
C b
2
b ,i
only the square of coefficients are presented in the formula, so this method is also called C-squared
Population Analysis (SCPA). By inserting the identity
1 =  C a2,i +  C a ,i Cb,i S a ,b
a a ba
into the right most of above formula,  can be rewritten as
C a2,i
 i ,a = C 2
+  C C b , i S a ,b
C
a ,i 2 a ,i
b ,i a ba
b
It is clear that when calculating composition of basis function a in orbital i, what is partitioned is
not the cross terms between atom a and other atoms, but the total cross term of all atom pairs. The
SCPA atomic charges can be calculated in the same manner as Mulliken charges by using the newly
defined . Relative to Mulliken population, the advantage of SCPA is that negative value of
population number never occurs.
71
3 Functions
3.9.6 Modified Mulliken atom population defined by Stout & Politzer

(8)
Stout and Politzer defined the  as (Theoret. Chim. Acta, 12, 379 (1968))
C a2,i
 i ,a = C 2
+ 2C a ,i Cb,i S a ,b
C a2,i + Cb2,i
a ,i
ba
That is cross terms are partitioned according to the ratio of the squares of corresponding coefficients.
Ostensibly, this definition has more consideration on the unbalanced nature of cross term, however
in practical applications the results are even worse than Mulliken, therefore this method is rarely
used now. Besides, Grabenstetter and Whitehead had pointed out that this MMPA definition has
unitary transformation dependence, so if the molecule is rotated the results changed (the dependence
is remarkable and cannot be ignored).
3.9.7 Modified Mulliken atom population defined by Bickelhaupt (9)
The population number of a basis function defined in Organometallics, 15, 2923 (1996) is
 a =  i C a2,i +  wa,b  2 i C a ,i Cb,i S a ,b
i b a i
where the weight of basis function a for partitioning the total cross term between a and b in all
orbitals is
 C k
2
a ,k
wa ,b = k
 C +  C
i
i
2
a ,i
j
j
2
b, j
Essentially, it is equivalent to define  as

 i ,a = C a2,i +  wa ,b 2C a ,i Cb,i S a ,b
ba
This method is similar to the MMPA defined by Stout and Politzer, the difference is in the latter
the weight wa,b is only related to local terms of basis functions a and b in current orbital, while in
present method the weight is related to total local terms of basis functions a and b.
3.9.8 Becke atomic charge with atomic dipole moment correction (10)
In the paper J. Chem. Phys., 88, 2547 (1988), Becke proposed a weighting function for
converting whole space integral to multiple single-center spherical integrals, although the weighting
function is not intend for population analysis, Multiwfn still makes an attempt to use this weighting
function as atomic space to obtain atomic charges. The Becke charge can be defined as
72
3 Functions
qA = Z A −  wA (r )  (r) d r
The radii used for evaluating the Becke weighting function (or say Becke atomic space) can be
controlled by uses, see corresponding options shown on screen. For details about Becke weighting
function please see J. Chem. Phys., 88, 2547 or Section 3.18.0. According to my experiences, by
using the default "modified CSD" radii, Becke charge is reasonable for typical organic systems, but
not very appropriate for ionic systems. For detail about "modified CSD" radii, see the end of Section
3.18.0.
After the Becke charges are calculated, atomic dipole moment correction will be performed
automatically. The correction process is identical to the one used for correcting Hirshfeld charge
(see next section). After the correction the charges will have better electrostatic potential
reproducibility and can exactly reproduce molecular dipole moment.
3.9.9 Atomic dipole moment corrected Hirshfeld atomic charges

(ADCH, 11)
Basic characteristic and usage

The main reason why Hirshfeld charges are too small and have poor reproducibility of
observable quantity is that atomic dipole moments are completely neglected. In the ADCH method
proposed by me (J. Theor. Comput. Chem., 11, 163 (2012)), atomic dipole moment of each atom is
expanded to correction charges placed at neighbour atoms, then ADCH charge is just the sum of
original Hirshfeld charge and correction charge. ADCH atomic charges are very reasonable in
chemical sense, molecular dipole moment is exactly reproduced, the reproducibility of ESP is close
to the atomic charges obtained from fitting ESP. Compared to another method that try to improve
Hirshfeld charges, namely Hirshfeld-I (see Section 3.9.13), the computational cost of ADCH charge
is negligible. Owing to its many advantages, ADCH is a highly recommended atomic charge model.
For an extensive comparison of atomic charge models, see Acta Phys. -Chim. Sin, 28, 1 (2012).
Before doing ADCH correction, Hirshfeld charge will be calculated first. In the summary field,
“corrected” and “before” correspond to ADCH charge and Hirshfeld charge respectively. At final
stage, the “Error” means the the difference between molecular dipole moment produced by ADCH
charges and the one produced by actual electron density, “Error” is always equals to or very close
to zero, because ADCH charges in principle exactly reproduce molecular dipole moment.
If you would like to obtain the detail of charge transfer between atoms in the atomic dipole
moment correction process, you can set “ishowchgtrans” in settings.ini to 1.
Algorithm detail about expanding atomic dipole moment

Full details of ADCH charge can be found from its original paper, here I only mention how the
atomic dipole moment correction (ADC) is realized.
Assume that we want to expand atomic dipole moment of atom A as correction charges, clearly,
the sum of all correction charges should be zero, and the dipole moment evaluated based on the
correction charges should be exactly equal to atomic dipole moment, namely below conditions
should be satisfied
73
3 Functions
 q
B
A→ B =0
μ A =  q A→ B R B
B
where qA→B is transferred charge from atom A to B, in other words, it is the correction charge on
atom B due to A. The index B cycles all atoms in the system. RB is relative coordinate (column
vector) of atom B with respect to atom A
The correction charges are expected to be distributed only around atoms neighouring to A, this
could be realized by minimizing function F:
( qA→ B )2
F = +   qA→ B +  (μ A −  qA→ B R B )
B  AB B B
where  and  are Lagrangian multipliers used to satisfy the two constraint conditions, the AB is a
function decreases rapidly as increase of distance between A and B, and its detailed form is
dependent of atomic radii of A and B, see original paper of ADCH for more information. Obviously,
this design of F suppresses occurrence of large correction charge on the atoms far away from atom
A.
It is shown that after some manipulations, the working equation of evaluating qA→B is
 AB
q A→ B = [(R B − R B )T  Λ −1  μ A ]
 AB

B
where superscript "T" is sign of transpose, and

Λ = R B  (R B )T − R B R B
T
 R AB B  AB R B  (R B )T
RB = B
, R B  (R B ) T
= B
 B
AB  B
AB
If the atom A is in local planar region, the  matrix will be exactly or almost singular matrix,
in this case inversed matrix of  obviously cannot be obtained. Notice that the solution to this
problem in current implementation is slightly different to the one introduced in the ADCH original
paper. In current Multiwfn, the  matrix is first diagonalized, the eigenvalues with absolute value
less than 10-5 will be simply set to zero (the corresponding eigenvector typically perpendicular to
the local plane of atom A), and remaining eigenvalues are inversed, now this matrix has
corresponded to the -1 in the new local coordinate. Then R B − R B and μ A are transformed
to the new local coordinate by doing left multiplication of transpose of eigenvector matrix of  on
them. Finally, the qA→B is obtained using above formula.
An example of calculating ADCH charges is given in Section 4.7.2.
74
3 Functions
3.9.10 CHELPG (Charges from electrostatic potentials using a grid

based method) ESP fitting atomic charge (12)
Theory
CHELPG (J. Comput. Chem., 11, 361 (1990)) is one of most widely used electrostatic
potentials (ESP) fitting charge models. Compared to CHELP and Merz-Kollman methods, CHELPG
charges have better rotational invariance, mostly due to the fitting points are distributed in cubic
grid manner.
In CHELPG model, a box is defined first to enclose the whole molecule, extension distance in
each side is 2.8 Å, see the red box in the following picture
Fitting points are evenly distributed in the box, the default spacing is 0.3 Å. For any fitting
point, if the distance between the point and any nucleus is smaller than vdW radius of the atom, or
the distances between the point and all nuclei are larger than 2.8 Å, then the fitting point will be
discarded. The purple dots shown above are finally used fitting points.
Like other ESP fitting methods, in CHELPG, the deviation function shown below is minimized
to make the ESP calculated by atomic charges (Vq) close to the ESP calculated based on
wavefunction (V) as good as possible.
F ( q1 , q2 ...qN ) = [Vq (ri ) − V (ri )]2
i
where ri is coordinate of fitting point i. {q} are fitted point charges, their positions are referred to as
fitting centers. Notice that q does not necessarily correspond to atomic charge, the fitting centers
can be defined at arbitrary positions.
It can be shown that the minimization of F could be formulated to below matrix equation
Aq = B
which can be further explicitly written as
 A11 A12  A1N 1  q1   B1 
A A22  A2 N 1  q2   B2 
 21    
     1    =   
    
 AN 1 AN 2  ANN 1   q N   BN 
 1 1 1 1 0     qtot 
75
3 Functions
with
1 V (ri )
A A, B =  BA =  qtot =  qA
i riAriB i riA A
where A and B are indices of fitting centers and N is total number of fitting centers. The column
vector q is what we need, its first N elements correspond to charges of the fitting centers. The q
could be easily evaluated as q=A-1B.
Usage
In the interface there are many options, which are introduced below.
• Option 1: If selecting this option, Multiwfn will start to calculate ESP value at each fitting
point, then fitted atomic charges of all fitting centers will be outputted on screen. RMSE and
RRMSE are also outputted automatically, they measure quality of fitting, the smaller value suggests
that the fitted charges have better ESP reproducibility. RMSE and RRMSE are defined as (where N
is the number of fitting points)
[V (r ) − V (r )]
q i i
2
RM SE = i
N
[V (r ) − V (r )]
q i i
2
RRM SE = i
V (r ) i
i
2
• Options 2~4: These options are used to set parameters for distributing fitting points. The
default values are reasonable and should not be changed without special reasons.
• Option 4: If you hope fitting points only distribute over certain fragment, you can select this
option and input atom indices, then a fitting point will be taken into account only if the atom closest
to it belongs to the given atom list.
• Option 5: During the ESP fitting, three kinds of ESP can be calculated and used, namely
"Nuclear + Electronic" (default), "Electronic" and "Transition electronic". Commonly you should
not change this option. The "Transition electronic" should be chosen if you want to evaluate the so-
called TrEsp (transition charge from electrostatic potential), please check Section 4.A.9 for details.
• Option 6: If you have chosen this option once to switch its status to "Yes", then after
calculation, coordinates with ESP values of all fitting points can be exported to ESPfitpt.txt or
ESPfitpt.pqr in current folder, the former is more readable, while the latter can be directly loaded
into VMD program to visualize the fitting points (and can be colored according to the "charge"
column, which records ESP values). In addition, fitting points with absolute difference between
exact ESP and the ESP evaluated based on atomic charges can be exported to ESPerr.pqr so that
ESP reproduction error in various molecular regions can be visualized in VMD. See Section 4.7.8
for illustration.
• Option 10: Choose the atomic radii used in fitting. The atomic radii employed in fitting affects
the distribution of fitting points, and thus influence the resulting charges. There are three modes can
be chosen to set the radii:
(1) This is default mode, the radii defined in original paper of CHELPG are used. However,
only the elements in the first three rows are defined, if your system contains other elements,
Multiwfn will ask you to input their vdW radii in turn. If you do not have proper radii in hand, you
76
3 Functions
can directly press ENTER button, then vdW radius of UFF forcefield of corresponding element
multiplied by 1/1.2 will be used, which is commonly a reasonable choice.
(2) Employing the UFF radii scaled by 1/1.2 for all elements. The UFF radii can be find in
Table 1 of UFF original paper J. Am. Chem. Soc., 114, 10024 (1992).
(3) Load radii of all elements involved in present system from external file. Multiwfn will ask
you to input the file path during fitting, the format of the file should like follows, the unit of the radii
should be in Å, all elements in the system must be defined:
H 1.2
O 1.8
Cu 2.2
• Option -1: For flexibility consideration, by using this option, coordinates of fitting points are
allowed to be read from external file to replace the CHELPG fitting points. The format of the file
should be
numdata
X Y Z [ESPval]  For fitting point 1
X Y Z [ESPval]  For fitting point 2
...
where the ESPval is an optional term, which denotes precalculated ESP value at corresponding point.
If numdata is a negative value, then the ESP values used in charge fitting will be read from the
fourth column rather than calculated by Multiwfn.
• Option -2: If you have special reasons (e.g. you want to place additional fitting centers at lone
pair or -hole region to enhance description of ESP around corresponding region), then coordinates
of additional fitting centers can be read from external file by using this option. The format of the
file should be
numdata
X Y Z  For additional fitting center 1
X Y Z  For additional fitting center 2
...
where numdata denotes how many entries are in this file. X, Y, Z are coordinates (in Å).
• Option -3: If you simply want to examine reproducibility for ESP at the fitting points of given
atomic charges, then you can use this option to load atomic charges from specific .chg file. Then
when you select option 1 to start the ESP fitting process, no ESP fitting charges will be yielded, only
the RMSE and RRMSE of the given atomic charges will be outputted (note that if you have chosen
option -2 to load additional fitting centers, then the number of charges in the loaded .chg file should
be identical to the total number of fitting centers). If you only want to study reproducibility of ESP
for the fitting points around specific fragment, you can choose option 4 and input the atom indices.
3.9.11 Merz-Kollmann (MK) ESP fitting atomic charge (13)
Merz-Kollmann (MK) charge is another well known charge model derived from ESP fitting,
77
3 Functions
see J. Comput. Chem., 11, 431 (1990). The only difference between MK and CHELPG is grid setting.
In MK, the fitting points are evenly distributed on the layers of 1.4, 1.6, 1.8 and 2.0 times the vdW
radii of each atoms, if distance between a fitting point and any atom is smaller than 1.4 times of its
vdW radius, then this fitting points will be discarded.
In the MK module of Multiwfn, the density of points per Å2 on the MK layers can be set by
option 2, the number of layers and the scale factor of atomic vdW radii used to define the layers can
be set by option 3, the atoms used to construct the MK fittings points can be set by option 4. For
description of other options and outputs, see last section.
Like the CHELPG module, in the MK module you can also choose the mode for determining
atomic radii used in the fitting by option 10. Notice that the original paper of MK method does not
explicitly present atomic radii. In Multiwfn, the default MK radii for the first and second rows and
P, S, Cl are in line with those in MK code of Gaussian, while the default radii for Na~Si are defined
as 1.57, 1.65, 1.65, 1.80 Å, respectively.
In the example in Section 4.7.8, I exemplified how to use the MK module to investigate ESP
reproducibility of given atom charges on all MK fitting points and on the fitting points
corresponding to specfic atoms. In addition, the example showed how to visualize ESP reproduction
error at various molecular regions.
3.9.12 AIM atomic charge (14)
AIM (Atoms-in-molecules) population denotes the number of electrons in AIM basin, and
accordingly, nuclear charge minusing AIM population yields AIM charge, which is also known as
Bader charge. AIM charges can be calculated in basin analysis module, please check the example
given in 4.17.1 on how to do this. Related theories and algorithms of basin analysis module are
introduced in Section 3.20.
3.9.13 Hirshfeld-I atomic charge (15)
Hirshfeld-I (HI) method was proposed in J. Chem. Phys., 126, 144111 (2007), it is an important
extension of Hirshfeld method. It is believed that the atomic space defined by Hirshfeld is not quite
ideal, because it does not respond actural molecular environment. In HI, atomic spaces are gradually
refined via an iterative scheme. After convergence, the final HI atomic spaces are evidently more
physically meaningful than the Hirshfeld ones.
There are some variants of HI method, including the Hirshfeld-E proposed in J. Chem. Theory
Comput., 9, 2221 (2013), the Hirshfeld-I proposed in J. Comput. Chem., 32, 1561 (2011) and the
fractional occupation Hirshfeld-I (FOHI) proposed in J. Chem. Theory Comput., 7, 1328 (2011).
Also, there is an iterative atomic space method name iterated stockholder atoms (ISA), which is
closely related to HI, see Chem. Commun., 2008, 5909. However, since in my personal viewpoint
78
3 Functions
these variants and ISA are not quite important or have special physical meaning, they are not
implemented in Multiwfn.
Basic theory of HI method

In HI method, the weighting function of atom A at iteration n is defined as
 A( n −1) (r − R A )  A( n −1) (r − R A )
w(An ) (r − R A ) = 
  B( n−1) (r − R B )
B
 pro
( n −1)
(r )
where r and R denote spatial coordinate and position of atomic nucleus, respectively. The atomic
density involved in HI is obtained by linear interpolation between spherically averaged free-state
atomic density of adjacent charge states:
 A( n ) (r ) = (qhigh − q (An−1) ) Afree,low (r ) + (q(An−1) − qlow ) Afree,high(r )

where qhigh and qlow are ceiling and floor integer of atomic charge of atom A at n-1 iteration, while
 Afree,high and  Afree,low are spherically averaged atomic density of atom A at these two charge states,
respectively. For example, if atom A carries 0.2 charge at n-1 step, then its radial atomic density at
step n will be computed as
 A( n ) (r ) = 0.8   Afree,q=0 (r ) + 0.2   Afree,q=+1 (r )

Before the HI iteration, radial density of all atoms are initialized to their neutral state.
Based on HI atomic weighting functions, HI atomic charges can be straightforwardly obtained.
The HI iteration continues until all atomic charges converged to a given criterion.
Note: It is clear that the HI atomic charges yielded at the first iteration are in principle identical to Hirshfeld
charges. However, this is not exactly true in Multiwfn, because currently the rule of generating spherically averaged
free-state atomic density involved in HI is not completely identical to that used in Hirshfeld, but the difference is
very small and can be ignored. It is never incorrect to directly take the atomic charges printed at the first iteration of
HI as Hirshfeld charges.
Usage
After entered the HI analysis module, you will find many options, you can adjust convergence
criterion and maximum number of cycles, also you can switch the algorithm for realizing HI
iterations (see below), and you can decide if printing atomic charges every iteration.
If you select option 0, Multiwfn will start the HI calculation. Before this, Multiwfn first checks
atomic radial density files (.rad) in "atmrad" folder, if they are available for all elements in the
current system, HI calculation will directly start; if they are not available, Multiwfn will try to
calculate atomic .wfn files by invoking Gaussian and convert them to .rad files, and then start the
HI iterations.
After convergence of HI iterations, HI atomic charges and fragment charge/population (if you
have defined fragment) are printed. Note that the printed HI atomic charges have been properly
normalized to eliminate noise of numerical integration (i.e. making sum of HI population of all
atoms equal to total number of electrons). Finally, you can select if outputting resulting atomic
charges as .chg file.
There is an option "-3 Switch if speeding up calculation using distance cutoff". By default this treatment is
enabled to significantly reduce cost for large system. If you find the HI charges are problematic or the calculation
cannot be normally finished, you can select this option once to disable this treatment and retry. Alternatively, you
can select it twice, then you will be prompted to input a cutoff value for this treatment, the larger the value, the more
accurate the result and more robust the calculation will be, however, at the expense of increasing computational cost.
The default cutoff is 2.0, which should be very safe for almost all cases.
79
3 Functions
About preparing atomic radial density files (.rad files)

Before igniting HI iteration, atomic radial density files (.rad files) must be available for all
elements in the systems at different charged states. There are three ways to provide them:
• Using built-in .rad files
The easiest way of providing the .rad files is simply copying the "atmrad" folder from
"examples" directory to current folder, then Multiwfn will directly utilize the .rad files in it during
HI calculation. This "atmrad" folder contains .rad file of all elements (except for lanthanides and
actinides) of all possible charged states, therefore in this case HI charges can be directly computed
without any additional step.
The .rad files in this folder were carefully generated based on highly accurate densities. Specifically, elements
of the first three rows were calculated at PBE0/def2-QZVPP level, all other main group elements were calculated by
PBE0/ANO-RCC with DKH2 relativisitic Hamiltonian, while all transition metal atoms were calculated by
PBE0/Sapporo-DK3-QZP-2012 with DKH2. For each element of each charged state, ground state spin multiplicity
was employed, and meantime wavefunction stability test and wavefunction optimization were carried out to make
sure that the wavefunction is stable and thus the resulting radial density is realistic.
• Let Multiwfn automatically invoke Gaussian to generate .rad files
If needed .rad files are not found in "atmrad" folder, Multiwfn will invoke Gaussian to calculate
atomic wavefunctions and generate .rad files, there are below two steps. Note that .rad file cannot
be generated in this way for lanthanides and actinides.
(1) Generating atomic .wfn files in "atmrad" folder: Multiwfn generates Gaussian input files
(.gjf) in "atmrad" subdirectory in current folder and invokes Gaussian to run them to generate
atomic .wfn files for all elements involved in present system. Charge states ranging from -2 to +2
are taken into account, while calculation of meaningless charge states are skipped. For example, its
meaningless to calculate -2 and +2 charge states for alkali elements, since in practical systems it is
impossible that an alkali atom has atomic charge between -1 to -2, or between +1 to +2. The
multiplicity of each charge state of each element is set to ground state (this cannot be well guaranteed
for all cases, for example, ground state at different calculation levels may even be different. However,
this is never an important problem, so please do not concern it too much). The path of Gaussian
executable file should be set by "gaupath" in settings.ini file, if it is not properly set, you will be
prompted to input it in the Multiwfn interface. You also need to input the keywords of Gaussian
used to calculate the atomic .wfn files. After Gaussian calculation has finished, the atomic .wfn files
are generated in "atmrad" folder, the file name directly corresponds to element name and charge
state. For example, the file corresponding to -1, 0, +1 and +2 charged states of Be will be generated
in "atmrad" folder as Be-1.wfn, Be_0.wfn, Be+1.wfn and Be+2.wfn, respectively. If corresponding
file has already been found in the "atmrad" folder, then the file will be directly used and not be
recalculated. If you find Gaussian calculation is failed (mostly due to SCF unconvergence problem),
you should carefully check Gaussian output file and properly adjust the keywords used.
(2) Converting atomic .wfn files to .rad files: As mentioned earlier, HI calculation requires
spherically averaged atomic densities. However, the electron density corresponding to the
atomic .wfn files generated at last step often does not meet this requirement. For example, carbon
at its neutral ground state has s2p2 configuration, and thus C_0.wfn corresponds to elliptical density
distribution. To get spherically averaged representation of atomic densities, Multiwfn automatically
loads each atomic .wfn files in "atmrad" folder, calculates spherically averaged radial density, and
write the data as .rad file, whose name is identical to its parental .wfn file. For example, C_0.wfn
will be converted to C_0.rad. The .rad file is a plain text file, the first line is the number of data
points, the first and second columns respectively correspond to radial distance with respect to
nucleus (in Bohr) and corresponding electron density. Note that if needed .rad file has already
80
3 Functions
presented in "atmrad" folder, then the conversion of .wfn→.rad will be skipped.

Note: If you are using Windows version of Multiwfn and Gaussian cannot be invoked properly, please read
Appendix 1 to set environment variable for Gaussian.
It is best (but never compulsory) that the atomic .wfn files are generated at the same calculation level as the
molecule under study, so that the result has strict physical meaning. If you want to let Multiwfn regenerate all needed
atomic .wfn/.rad files at a specific level prior to HI analysis, evidently you should clean up the "atmrad" folder before
calculation.
• Manually generate .rad files
It is also possible to manually calculate .rad files and put them in the "atmrad" folder, so that
they will be used in HI calculation. At least .rad files corresponding to below charged states must
be provided for the elements in the present system, otherwise Multiwfn will try to employ Gaussian
to calculate missing ones.
IA, VIIIA: -1,0,1
IIA: -1,0,1,2
IIIA, IVA, VA, VIA: -2,-1,0,1,2
VIIA: -1,0,1,2
All transition metals: -1,0,1,2
For lanthanides and actinides, the consideration of charged states is up to you. It is suggested to
consider -1, 0, 1 and 2.
As an example, if you want to calculate HI charges for water, you should provide O-2.rad, O-
1.rad, O_0.rad, O+1.rad, O+2.rad, H-1.rad, H_0.rad and H+1.rad in "atmrad" folder of current
directory. Generating .rad file is easy. Taking generation of O-1.rad as example, you should use
Gaussian or other code to calculate an oxygen atom with charge of -1 and spin multiplicity of 2.
Then load the resulting wavefunction file (e.g. wfn/molden/fch...) into Multiwfn, enter main
function 1000 and select subfunction 14, you will immediately obtain a .rad file with same name as
the input file.
Note that if your system contains lanthanides or actinides, this is the only way of preparing
corresponding .rad files and thus calculating HI charges.
Appendix: Two numerical algorithms of HI

Multiwfn provides below two algorithms to realize HI methods, the results are exactly the same,
they only differ by efficiency and memory requirement.
(1) Fast & large memory requirement (default): This algorithm computes as much as possible
data before starting iteration. Therefore, once initialization stage is done, the HI iteration can be
finished rapidly. The drawback of this algorithm is that large amount of memory is needed, the
memory consumed is at least 7natmnatmnradnang, where natm is the number of total atoms, nrad and
nang corresponds to the number of radial and angular integration grid per atom. Evidently, this
algorithm is unable to be applied for very large systems unless you have huge physical memory.
(2) Slow & low memory requirement: This algorithm utilizes very low amount of memory,
however, the data needed to be calculated in each iteration is much more than algorithm (1), thus
the total computational cost is much higher.
For both the two algorithms, the higher number the integration grid, the more accurate the
result. By default nrad and nang are properly set by Multiwfn. If the system only consists of first two
rows elements, then nradnsph will be 30*170=5100. If you want to manually set nrad and nsph, you
can set "iautointgrid" in settings.ini to 0 and change "radpot" and "sphpot" parameters.
An example of computing HI charges is given in Section 4.7.4.
81
3 Functions
3.9.14 CM5 atomic charge (16)
CM5 charge was proposed by Truhlar et al. in J. Chem. Theory Comput., 8, 527 (2012). This
charge is somewhat akin to the ADCH charge proposed by me, both of them are defined as applying
additional correction to Hirshfeld charges. Unlike ADCH, which is free of empirical parameters,
CM5 method contains global parameters as well as parameters for individual elements. The CM5
parameters are optimized for best reproduction of highly accurate experimental or theoretical
molecular dipole moment.
Worthnotingly, there is a remarkable difference between the basic feature of ADCH and CM5:
ADCH charges can exactly reproduce molecular dipole moment corresponding to present
calculation level, that means if the level used is very high (e.g. CCSD/aug-cc-pVTZ), then the dipole
moment calculated by the resulting ADCH charge must be very close to the molecular dipole
moment in real world; while if the used level is poor (e.g. HF/6-31G), then the ADCH charges will
be almost useless. In contrast, the CM5 charges do not attempt to reproduce molecular dipole
moment at present calculation level, but to reproduce real molecular dipole moment, therefore even
if low level such as B3LYP/6-31G* is used, which is certainly unable to give good dipole moment
result, the CM5 charges derived at this level commonly are still able to yield molecular dipole
moment at acceptable accuracy.
The expression of CM5 charge is
qiCM 5 = qiHirsh +  Tij Bij

j i
Bij = exp[ − ( rij − Ri − R j )]
where rij is distance between atom i and j, Bij may be regarded as their Pauling bond order, Ri and
Rj are their atomic covalent radii, which are defined as follows: For Z=1~96, the average between
CSD radii and Pyykkö radii are used, while for Z=97~118, the Pyykkö radii are employed. The
global parameter  equals to 2.474 Å-1. The Tij is defined as Dij if both i and j are attributed to H, C,
O, N, note that Dij=0 when i and j belong to the same element and Dij=-Dji. All the involved six Dij
parameters (H-C, H-N, H-O, C-N, C-O, N-O) are tabulated in the original paper. For other cases, Tij
is defined as Di - Dj, the optimized D parameters for all elements through out the whole periodic
table are provided in the supplemental material of CM5 original paper.
Like usual Hirshfeld and ADCH calculations, after you enter this function, I suggest you select
option 1 to use the build-in sphericalized atomic densities in free-states, since it is the most
convenient. Then Multiwfn starts calculation of Hirshfeld charges, and then print CM5 charges. If
you want to gain detailed information about the CM5 correction process during the calculation, you
can set "ishowchgtrans" parameter in settings.ini to 1.
82
3 Functions
3.9.15 Electronegativity Equalization Method (EEM) atomic charge (17)
Theory
The EEM method is also known as Full Equalization of Orbital Electronegativity (FEOE), it
was firstly proposed in J. Am. Chem. Soc., 107, 829 (1985). The idea of EEM is very clear. The
electronegativity of atom i can be written as
qj
 i = (  i0 +  i ) + 2(i0 + i )qi + 
j i ri. j
where  i0 and i0 are Sanderson electronegativity and Parr-Pearson hardness of corresponding
element, respectively, while  i and  j are fitted parameters. qi denotes atomic charge of
atom i, and ri,j is distance between atoms i and j. For convenience, above formula is commonly
expressed as
qj
 i = Ai + Bi qi +  
j i ri , j
where  is global parameter, A and B are element parameters. It can be seen that, atomic
electronegativity in a chemical system is function of atomic charges.
EEM method assumes that in a chemical system, all atoms have equal electronegativity, hence
the EEM charges can be determined by solving linear equations according to below conditions
1 =  2 = ... =  N =  mol
q
i
i =Q
where Q stands for net charge of the whole system, mol is molecular electronegativity
It is easy to show that the working matrix equation for solving EEM charges can be written as
follows
 B1 
r1, 2  
r1, N − 1  q1   − A1 
 B2  
− 1  q2   − A2 
 r2 ,1 r2 , N    
        =   
    
 rN ,1

rN , 2  BN − 1  qN   − AN 
 1 1  1 0    mol   Q 
Since EEM matrix element is very simple and the above matrix equation can be solved easily, EEM
charges could be evaluated rather rapidly even for very large systems.
Parameters
The result of EEM charges are directly dependent of EEM parameters , {A} and {B}. There
is no unique way to determine the parameters, the most common way to obtain them is fitting, so
that the resulting EEM charges are close to quantum chemistry atomic charges (e.g. NPA, CHELPG,
Mulliken) as much as possible. Many papers presented EEM parameters fitted to various kinds of
atomic charges calculated at different levels. In Multiwfn, below EEM parameters can be directly
chosen:
83
3 Functions
 Parameters fitted to Mulliken charges at HF/STO-3G, see Int. J. Mol. Sci., 8, 572-582 (2007).
Available for H, C, N, O, S
 Parameters fitted to CHELPG charges at B3LYP/6-31G* and at HF/6-31G*, see J. Comput.
Chem., 30, 1174 (2009). Available for H, C, N, O, F, S, Cl, Br, Zn
 Parameters fitted to NPA charges at B3LYP/6-311G*, they were extracted from SI of J.
Cheminform., 8, 57 (2016). Available for H, C, N, O, F, P, S, Cl, Br
Note that for some elements, their A and B parameters also depend on multiplicity, namely the
maximal formal bond order between this atom and its neighboring atoms, so that influence of
chemical environment can be taken into account.
With above mentioned parameters, EEM charges have good reproducibility of target atomic
charges for typical organic systems, but do not expect EEM method can work well for systems with
complicated electronic structure, since common training set of EEM parameters only include
organic systems with typical bonding.
Usage
Since formal bond order is involved in the EEM calculation, you must use MDL molfile (.mol)
or .mol2 as input file, because in all file types that supported by Multiwfn, only this file contains
connectivity information between atoms. Do not forget that there is a severe limitation of .mol
format, namely the number of atoms cannot exceeds 999, therefore .mol2 must be used for very
large system. The .mol/.mol2 file can be outputted by many programs, such as GaussView and
OpenBabel. OpenBabel is recommended to use, since the format of .mol file outputted by
OpenBabel is very standard. Do not use GaussView to generate .mol/.mol2 file if dashed bond is
presented (e.g. for benzene, each C-C bond is described by default as a single bond with a dashed
bond, the .mol file is unable to record such non-Lewis representation of bonding).
After you load a .mol/.mol2 file and then enter present function, you can directly choose option
0 to calculate EEM charges, the molecular electronegativity will also be outputted together. The
default EEM parameters are those fitted for reproducing B3LYP/6-31G* CHELPG charges.
Before calculation, you can choose option "1 Choose EEM parameters", present EEM
parameters will be shown on screen, and then you can select a built-in parameter set that you want
to use. Alternatively, you can load parameters from external file using suboption 0, the format of
parameter file should mimic to this:
0.302000
H 1 2.38500 0.73700
C 1 2.48200 0.46400
C 2 2.46400 0.39200
N 1 2.59500 0.46800
N 2 2.55600 0.37700
O 1 2.82500 0.84400
O 2 2.78900 0.83400
The first line is , after that, defining parameter for each element of each multiplicity. The second,
third and fourth columns are multiplicity, A and B, respectively. Free format is used. Using
corresponding option, the present parameters can also be exported to EEMparm.txt in current folder.
If the system under study is an ionic system, do not forget to use option "2 Set net charge" to
set net charge to actual status before calculation!
Notice for GaussView users: The .mol/.mol2 file generated by GaussView, at least for version 6.0.16, is not
completely correct when there are conjugated rings. In the connectivity field of .mol file, the aromatic ring should
84
3 Functions
be represented as alternate variation of single and double bonds, however in the .mol (.mol2) generated by
GaussView, the multiplicity of the bonds in aromatic rings is recorded as 4 (ar), clearly it is unreasonable. This issue
could be fixed if you use OpenBabel to convert the .mol/.mol2 to a new .mol/.mol2 file.
An example of calculating EEM charges is given in Section 4.7.5.
Information needed: Atom coordinates with connectivity (.mol or .mol2 file)
3.9.16 Restrained ElectroStatic Potential (RESP) atomic charge (18)
This module is quite powerful and flexible, it can calculate the standard RESP charge proposed
by Kollman and coworkers in J. Phys. Chem., 97, 10269 (1993), and can also calculate ESP fitting
charges under various equivalence and charge constraints. In Section 3.9.16.1 I first describe basic
ideas of RESP charge and related treatments, then in Section 3.19.16.2 the usage of this module is
introduced. If you are not interested in relevant theories, you can skip Section 3.9.16.1. However, if
you are not familiar with ESP fitting method, you should at least read Section 3.9.10 first to gain
minimal knowledges.
Some more discussions about this topic can be found in my blog article "Principle of RESP charge and its
calculation in Multiwfn" (in Chinese, http://sobereva.com/441).
3.9.16.1 Theory
Theory Part 1: Conformation dependency, equivalence constraint and penalty function

The ESP fitting charges generated by MK and CHELPG methods introduced in Section 3.19.10
and 3.19.11 can be perfectly used for molecular modeling purpose for rigid molecules, however,
they are not quite suitable for modeling flexible molecules, due to below reasons:
(1) The results are highly dependent on conformation. Flexible molecules have many different
conformations, the conformation often changes during molecular dynamics simulation, while the
ESP fitting charges are highly sensitive to the conformation. If only one conformation is used to
calculate the ESP fitting charges and the simulation is conducted based on these charges, the
dynamic behavior of the molecule may be wrong and the relative energies between different
conformations may be seriously incorrect, since single set of fixed atomic charges is unable to
equitably well describe all relevant conformations.
(2) The atomic charges fitted at single conformation do not faithfully reflect chemical
equivalency of atoms. For example, three hydrogens on the methyl group of methanol are
chemically equivalent. During molecular dynamic simulation in room temperature, the methyl
group can rotate frequently, therefore, three hydrogens should share the same charge. However the
calculated charges are not identical regardless of choice of conformation used in the ESP fitting
procedure (since this system does not have triple rotational symmetry along the methyl bond axis),
clearly this problem also brings some irrationality to the simulation phenomenon.
(3) The quality of fitted charge of buried atoms is poor. The ESP fitting points are distributed
in the vicinity of and outside the van der Waals surface. For atoms connected to multiple atoms
(such as sp3 hybridized carbon), especially the atoms inside a macromolecule, because of their far
distances to fitting points, their atomic charges have low fitting quality and large numerical
uncertainty. Moreover, as the conformation changes, the charge fluctuation of these buried atoms
tends to be significant, therefore the presence of these atoms further aggravate the conformational
85
3 Functions
dependence of the ESP fitting charges.

ESP fitting charges cannot be well used for modeling flexible molecules without solving above
problems.
For the above problem (1), a good solution is to simultaneously consider multiple
conformations during the ESP fitting process. One first determines the weight of each conformation,
then constructs A matrix and B vector using fitting points of various conformers with consideration
of conformational weights, then the solved ESP fitting charges can at least be able to well reproduce
the ESP of those conformations with relatively large weights. This idea has been examined in J. Am.
Chem. Soc., 114, 9075 (1992). Of course, this way of considering multiple conformations can be
very expensive for flexible molecules with many rotatable bonds, because the number
conformations increases exponentially with the increase in rotatable bonds.
For the above problem (2), equivalence constraints can be imposed on chemically equivalent
atoms in the fitting process so that their atomic charges are the same (Another way is to calculate
the ESP fitting charges as usual, and then average the charges of chemical equivalent atoms.
However, the charge obtained in this way is not as ideal as employing equivalence constraints).
For the above problem (3), the solution proposed in Kollman's RESP paper is to add a
hyperbolic penalty function  rstr
2
= a [( qA2 + b2 )1 / 2 − b] to the function of measuring
A
reproducibility of the ESP calculated based on wavefunction, where index A corresponds to atomic
index of non-hydrogen atoms. The penalty function involves a tightness parameter b and a restraint
strength parameter a. The former is generally set to 0.1, while the latter can be adjusted in the actual
calculation. The larger the a, the stronger the tendency of the atomic charge to be pulled down, and
meantime the worse the ESP reproducibility becomes. Obviously, the parameter a should be
properly selected, generally a value less than or equal to 0.001 is employed. It has been found that
introduction of this form of penalty function significantly lowers charges of buried atoms, while
other atoms, in particular polar atoms, are not evidently affected. Kollman believes that this
treatment also significantly reduces the conformational dependence of the ESP fitting charge. After
introducing the hyperbolic penalty function, the ESP fitting procedure can no longer be solved in
one step, iteration is needed until changes of all atomic charges are small enough.
Below I give detailed derivation of the working equation used to calculate ESP fitting charges
under above special considerations. When equivalence constraint is employed, the function to be
minimzed using least square method in the ESP fitting procedure will be
2
  1   
F =  Vi −  q A   +    n Aq A − qtot 
 i  A a A ria   A 
where i cycles fitting points, Vi is the ESP calculated based on wavefunction at point i, {q} is the set
of uniquely derived atomic charges, ria denotes distance between point i and atom a, which belongs
to equivalence constraints A. nA is the number of atoms constrained to be equivalent in batch A. If
nA=1, that means A just corresponds to an atom without equvalence constraint.
Minimization of F with respect to variables yields
86
3 Functions
F
= 0 =  n Aq A − qtot
 A
F  1  1
= 0 = −2   Vi −  q A   + nB 
qB  bB rib  i  A a A ria 
The second equation can be further reorganized as follows, with considering the fact that the
value of Lagrangian multiplier  is arbitrary
F  1  1
= 0 =    Vi −  q A   + 
qB  bB rib  i  A a A ria 
1 Vi
 q  r r
A
A
a A bB i
+  = 
bB i rib
ia ib
The set of linear equations can be formulated as a matrix equation
 A11 A12  A1N n1   q1   B1 

A A22  A2 N n2   q2   B2 
 21    
         =     Aq = B
    
 AN 1 AN 2  ANN nN  qN   BN 
 n1 n2  nN 0      qtot 
with
1 Vi
AA,B =  BA =  qtot =  q A
aA bB i ria rib aA i ria A
where A and B are atomic indices, and there are totally N atoms. Once construction of the A and B
is completed, the charge vector can be easily obtained as q=A-1B.
Now we consider the case that penalty function is added to the function F. Given that
 rstr
2
/ qA = aqA ( qA2 + b2 )−1 / 2
by incorporating it into the above expression of F / qB = 0 , we finally get
1 Vi
 q  r r
A
A
a A bB i
+ aqA ( qA2 + b2 ) −1 / 2 +  = 
bB i rib
ia ib
1
The diagonal elements of A should thus be AA, A =  r r
a , b A i
+ a ( q A2 + b2 ) −1 / 2 , while the non-
ia ib
diagonal terms of A should keep unchanged. In practical calculation, the {q} in initial A is set to
zero, then updated {q'} is obtained by solving the matrix equation, after that {q'} is used to construct
the A of the second iteration. The iteration is repeated until charge variation of all atoms is smaller
than a given threshold.
The easiest way of taking multiple conformations into the ESP fitting procedure is replacing
the A matrix and B vector with their weighted averaged counterparts, as suggested in J. Am. Chem.
Soc., 114, 9075 (1992):
87
3 Functions
A =  w A  B =  w B 
 
where  is index of conformer and w is corresponding weight. The weights could be determined in
different ways, the common way is calculating relative Gibbs free energies and then evaluating
weights according to Boltzmann distribution.
Theory Part 2: The standard RESP charge

The Restrained ElectroStatic Potential (RESP) charge proposed by Kollman et al. in J. Phys.
Chem., 97, 10269 (1993) may be the most suitable atomic charge model for molecular simulation
of flexible molecules. It takes advantage of many aforementioned ideas and largely solves the above
mentioned problems in MK/CHELPG charges. The fitting process of the RESP charges is divided
into the following two steps.
• Step 1: During the charge fitting, a hyperbolic penalty function with a=0.0005 is used to
impose a weak charge restraint on non-hydrogen atoms. Charges of all atoms are fitted, and no
equivalence constraint is employed. This step allows atomic charges to change with the greatest
degree of freedom to make polar atoms fit the ESP as well as possible.
• Step 2: Using the hyperbolic penalty function with a=0.001 to impose a strong restraint on
non-hydrogen atoms. This step only allows charge of sp3 hybridized carbons, methylene carbons
and hydrogens attached on them to be fitted, while charges of all the other atoms keep fixed at the
value obtained at step 1. Equivalence constraint is applied to hydrogens on each −CH3, =CH2,
−CH2− group.
The reason why the RESP charge fitting is divided into two steps is because the authors found
that only by doing so, the problems of normal ESP fitting charges in modeling flexible molecules
could be largely solved without causing too much damage on the reproducibility of ESP. Since RESP
charge is fairly ideal for molecular dynamic modeling purpose, it has been employed by many
famous forcefields, such as AMBER, GAFF and GLYCAM.
Notice that although conformation dependency has been diminished to large extent in the subtly
designed RESP fitting process, if you want to obtain a set of atomic charges that can equally well
describe all important conformations, you still need to explicitly take multiple conformations in the
RESP fitting procedure.
The MK type of fitting points are employed in original paper of RESP, however, changing to
CHELPG type of fitting points is also completely reasonable.
Theory Part 3: ESP fitting with charge constraints

When calculating the ESP fitting charges, various constraints can be added via the Lagrangian
multiplier method. The most significant one should be constraint on net charge of specific fragments.
The charge constraints can achieve many special purposes:
(1) Biomacromolecules, polymers and other systems are all polymerized one by one. Each
component of such a macromolecule is called residue. The atomic charges of the whole system are
made up of that of individual residues. It is obvious that the net charge of each residue should be an
integer. If we want to derive atomic charges of a given residue, we can cap the two terminals of the
residue with appropriate group or fragment, and then impose charge constraint on the residue
segment during ESP fitting process so that its net charge exactly corresponds to a desired integer.
(2) Some force fields, such as GROMOS, use the charge-group concept to reduce the error of
the electrostatic interaction evaluated via cut-off method. Each charge group contains several atoms,
and all the atomic charges are summed to an integer. For example, the total charge of a carboxyl
88
3 Functions
group should be 0, and its charge should become -1 after dissociation of the proton. In order to
obtain a set of ESP fitting charges that compatible with the charge group concept, charge constraints
can be utilized to maintain the charge of each segment as a specified integer value.
(3) Sometimes one wants to calculate ESP fitting charges based on wavefunction of dimer or
multimer, and hopes that the charge of each monomer is exactly integer, this purpose could be
realized by employing charge constraints.
Technically, adapting charge constraint into ESP fitting in terms of Lagrangian multiplier is
straightfoward, we only need to properly modify the form of matrix equation. For example, we want
to add below constraints:
q1 + qN = 0.5
q2 = −0.2
Then below terms should be added to the function F, which is to be minimized:
 ' (q1 + qN − 0.5) +  ' ' (q2 + 0.2)

correspondingly, two new equations appears
F F
= 0 = q1 + qN − 0.5 = 0 = q2 − 0.2
 '  ' '
and
F F F F F F
= 0 = 0 + ' = 0 = 0 + '' = 0 = 0 + '
q1 q1 q2 q2 qN qN
where F0 is the function F without charge constraint. Clearly, the current ESP fitting problem in
matrix equation form can be given as
 A11 A12  A1N n1 1 0  q1   B1 

A A22  A2 N n2 0 1  q2   B2 
 21    
      0 0      
    
 AN 1 AN 2  ANN nN 1 0   q N  =  BN 
 n1 n2  nN 0 0 0     qtot 
    
 1 0 0 1 0 0 0   '   0.5 
 0 1 0 0 0 0 0  ' '  − 0.2
In practical programming implementation, when multiple conformations, equivalence
constraint, charge constraint and penalty function are simultaneously taken into account, the ESP
fitting calculation is carried out in following process: Because of introduction of the hyperbolic form
of penalty function, the A and q should be updated alternately until convergence criterion is reached.
In each iteration, only the first NatomNatom block of A matrix and first Natom elements of B vector
are constructed with consideration of conformation weights and penalty function, then remainder
parts of A and B are filled according to charge constraint. Finally, according to equivalence
constraint, the corresponding rows of A are combined together (e.g. if atoms 3, 6, 7 are constrained
to be equivalent, then these three rows should be summed up) to form a temporary matrix, whose
columns are further properly combined according to equivalence constraint to form Aeqv matrix.
Similarly, the rows of B vector are tranformed to Beqv according to equivalence constraint. After that,
89
3 Functions
solving the equation qeqv=Aeqv-1Beqv and correspondingly updating atomic charges according to the
given equivalency relationship. In the next cycle, the diagonal terms of A matrix are updated using
the atomic charges obtained in last cycle, while non-diagonal terms of A and all elements of B vector
are not needed to be changed. Multiwfn simply uses zero as initial charges for the atoms to be fitted.
3.9.16.2 Usage and some details
Options in the RESP module

Any input file carrying GTF information could be used for this modules. After loading input
file and entering this RESP module (subfunction 18 of main function 7), you will find many options,
as described below
·Option 1: If you just want to calculate the standard RESP charges defined by Kollman et al.,
you should simply select this option, then the RESP charges will be calculated and printed. Since
this calculation contains two steps, it will be referred to as "two-stage RESP fitting".
·Option 2: If you simply need to calculate normal ESP fitting charges with/without specific
constraints, you should select this option. This process only includes one step, therefore it will be
referred to as "one-stage ESP fitting".
·Option -1: In the calculation of standard RESP charges and normal ESP fitting charges
with/without additional constraints, multiple conformations could be taken into the fitting process.
By selecting this option, you will be asked to input path of a plain text file containing conformer
list, each line of this file consists of file path and weight for each conformer. For example:
D:\a\conf1.fch 0.2
D:\a\conf2.fch 0.75
D:\b\conf.fch 0.05
Evidently, the sum of all weights must be exactly or very close to 1.0. After that, in the charge
calculation, all files involved in this file will be loaded and calculated in turn (if you use this feature,
the input file loaded when Multiwfn boots up will be unimportant, it can even only contain structure
information of present system, therefore you can also use e.g. .pdb and .xyz as input file).
·Option 3: By default, MK type of fitting points is employed, if you want to change to
CHELPG type of fitting points, or you want to modify detailed setting of distribution of fitting
pionts (such as point density), you can use this option. Note that the density of fitting points under
default setting is already high enough, thus it does not need to be further enhanced without special
reason.
·Option 4: This option is used to set parameters of hyperbolic penalty function for non-
hydrogen atoms. The a used in "one-stage fitting" (0.0005 is employed by default), the respective a
parameters used in the first and second stages of the standard RESP fitting, as well as the b parameter
can be customized. Also, this option is able to manually define maximum number of RESP iterations
and convergence threshold of charge variation.
·Option 5: This option is use to set equivalence constraint. You can customize the constraint
by providing a plain text file containing entries of equivalence constraints. For example, if the file
content is
4,6,9-11
5,7
Then there will be two equivalence constraints, the first one requires that atoms 4, 6, 9, 10, 11 share
the same charge, the second one requires that atoms 5 and 7 share the same charge. The equivalence
90
3 Functions
constraint defined in this way take effect for both "one-stage fitting" and the first step of "two-stage
RESP fitting".
Note that for "one-stage fitting", by default hydrogens in each CH3 and CH2 group are
constrained to be equivalent. You can modify or simply remove this equivalence constraint setting
via this option.
Suboption 10 can generate plain text file named eqvcons_H.txt containing equivalence
constraint setting of "hydrogens in each CH3 and CH2 group are the same". You can then manually
modify this file to meet your special requirement.
Subfunction 11 can generate plain text file containing equivalence constraint of symmetrically
equivalent atoms in local region or the entire system. The point group of the selected atoms will be
detected and each class of equivalent atoms will be written to eqvcons_PG.txt in current folder. This
feature is quite useful in certain cases.
·Option 6: This option is use to set charge constraint in "one-stage fitting" or the first step of
"two-stage RESP fitting". By default no charge constraint is employed for them, however in this
option you can provide a plain text file to customize the rule of charge constraint. For example, if
the file content is
4,6,9-11 0.8
5,7 -0.3
Then sum of charges of atoms 4, 6, 9, 10, 11 will be constrained to 0.8 during the fitting, while
sum of charges of atoms 5 and 7 will be constrained to -0.3.
·Option 7 and details about determination of connectivity: To calculate the standard RESP
charge, or to calculate the normal ESP fitting charges but requiring the charges of the hydrogens in
each CH3 and CH2 group to be equivalent, interatomic connectivity is needed for automatically
determining which atomic charges should be fitted and which hydrogens should be constrained to
be equivalent. By default, if distance between two atoms is less than 1.15 times the sum of their
CSD covalent radii, then they will be regarded as bonded. If you feel that the current connection
relationship does not match your expectation, you can select option 7 to read the connectivity from
a specific .mol file, the .mol format contains a field recording connectivity information and can be
generated by many visualization programs such as GaussView. (Alternatively, you can modify the
threshold for judging bonding in main function 0, in which you can gradually change the bonding
threshold until the bonding relationship shown in the graphical window completely in line with your
expectation, the threshold will be retained and applied to the calculation in the RESP module).
·Option 8: This option enables Multiwfn directly load fitting points and corresponding ESP
values from Gaussian output file. If you have selected this option once, then during the ESP fitting
charge calculation process, Multiwfn will no longer attempts to determine position of fitting points
and calculate ESP values, but ask you to input path of a Gaussian output file of pop=MK or
pop=CHELPG task in combination with IOp(6/33=2) keyword. In addition, Gaussian also has a
keyword IOp(6/42=x), where x is the number of fitting points per Å2 for pop=MK task. x is
recommended to set to 6, which corresponds to Multiwfn default setting. Notice that since the code
in Multiwfn and in Gaussian for generating MK fitting points is different, the result calculated with
and without loading Gaussian pop=MK output file must have slight deviation.
In general, this option is not needed, but if you prefer to perform time-consuming calculations
on server and use Multiwfn to realize analyses on a poorly configured PC, then this function will be
useful. In addition, if you may calculate ESP fitting charges for a system many times (due to some
91
3 Functions
reasons, such as each time of calculation uses a different setting), then if Gaussian output file
containing fitting points information is generated, calculation of ESP values in each time of ESP
fitting can be avoided and thus a lot of time could be saved.
• Option 9: This option is used to set additional fitting centers for the RESP fitting, which may
be used to enhance representation of ESP due to e.g. lone pairs and -holes. The coordinates are
read from a given text file, whose format should look like
numdata
X Y Z Additional fitting center 1 of conformer 1
[blank line]
[blank line]
where numdata denotes the number of additional fitting centers for each conformer (the number is
the same for all confomers). X, Y, Z denote the coordinates in Å. You can set arbitrary number of
additional fitting centers for arbitrary number of conformers. The additional center definition
between different conformers should be separated by a blank line, as illustrated above. Note that
these fitting centers are regarded as polar non-hydrogen atoms in the RESP charge fitting procedure,
however, their radii are set to zero (i.e. they do not affect number and distribution of ESP fitting
points). An example of utilizing this feature is given as "Example 6" of Section 4.7.7.
·Option 10: This option is used to set the atomic radii employed in the fitting, please check
corresponding description in the CHELPG section (Section 3.9.10) for detail. Notice that the default
mode is "automatic", in this case, if the fitting points of MK type are employed, then MK radii will
be adapted (see Section 3.9.11 on how the MK radii are defined in Multiwfn); if fitting points of
CHELPG type are employed, then CHELPG radii will be adapted.
·Option 11: This option is used to choose the type of ESP that you want to fit. Commonly,
this option should not be changed, the default type of ESP is the ESP defined in usual way. However,
if you intend to use this RESP module to derive atomic transition charges, you should choose this
option and change the ESP type to “3 Transition electronic”. Please check Section 4.A.9 for more
information and example.
More information about equivalence constraint and charge constraint

In the user-provided equivalence constraint file involved in option 5, no atom can be shared by
multiple entries. For example, if the first entry is 2~7, while the second entry is 5, 8~10, the result
will be completely meaningless, because both of them involve atom 5.
Intersection between different sets of charge constraints is allowed, for example, you can
require charge of atom 5 is 0.35 while sum of charges of atoms 3~8 is 1.0.
Charge constraint could also be used in combination with equivalence constraint; however,
there should not be intersection between any charge constraint entry and equivalence constraint
entry. For example, charges of atoms 2, 5, 9 are required to be identical, and meantime you constrain
total charge of 5, 10~15, 17~19 to be 0.15, such a combination never works since both of them
involves atom 5.
Speeding up calculation
92
3 Functions
Since calculation of ESP on fitting points is a computationally demanding step, while

calculation speed of ESP of internal code of Multiwfn is usually slower than the cubegen utility in
Gaussian package, if Gaussian is available on your machine and the input file is .fch/fchk, it is
suggested to allow Multiwfn to invoke cubegen to evaluate ESP to reduce cost of deriving ESP
fitting charges. You simply need to set "cubegenpath" parameter in settings.ini to actual path of
cubegen executable file in your machine. See Section 5.7 for detail.
FAQ: Why sometimes spatially equivalent atoms have different charges?

You may frequently find a phenomenon that spatially equivalent atoms often have marginally
different charges. The reason is that the distribution of fitting points does not always happen to be
coincident with molecular point group. There are two ways to relieve this problem:
(1) Write an equivalence constraint file and use option 5 to load it to make the spatially
equivalent atoms share exactly the same charge during fitting. Writing this file is quite easy even if
the system is large, because by subfunction 11 of option 5 you can make Multiwfn automatically
recognize point group of local fragment or the entire system and write corresponding equivalence
constraint setting to eqvcons_PG.txt. This point is fully exemplified in "Example 6" in Section 4.7.7.
(1) Choose option "3 Set method and parameters for distributing fitting points", select
CHELPG, and then select "1 Set grid spacing", input a value much smaller than the default one. The
smaller the grid spacing, the better the atomic charges satisfy the point group symmetry. This
treatment does not completely solve the problem but only relieve it, and it increases computational
cost because the number of points to be calculated is increased. Therefore way (1) is prefered over
this way.
Many examples of evaluating standard RESP charges and normal ESP fitting charges with
various constraints, as well as some special skills can be found in Section 4.7.7.
The RESP2 method is an extension of RESP method. The RESP2 charge is more suitable than
RESP charge for molecular dynamics simulation in condensed phase since it better takes solvent
effect into account. See Section 4.7.7.9 for instance on how to calculate RESP2 charge.
3.9.17 PEOE (Partial equalization of orbital electronegativity) or

Gasteiger charge (19)
Theory
The PEOE (Partial equalization of orbital electronegativity) charge is also known as Gasteiger
charge or Gasteiger-Marsili charge, the idea was firstly proposed by Gasteiger and Marsili in
Tetrahedron Lett., 34, 3181 (1978), and then matured in Tetrahedron, 36, 3219 (1980). PEOE charge
is determined according to interatomic connectivity, chemical environment and partially based on
electronegativity equalization principle. The major advantage of PEOE method is that it is able to
estimate atomic charges for huge system with negligible computational cost. However, there are
several drawbacks in this method: (1) Only limited elements are supported (2) Actual electronic
structure is not taken into account, the charges only reflect bonding types and connectivities (3) Poor
reproducibility to many observable quantities, such as dipole moment and electrostatic potential. (4)
93
3 Functions
Cannot be used for charged systems. Despite that there are many shortcomings, PEOE is still very
popular for crude estimation of atomic charges purpose, and have been widely employed in
molecular docking and drug design fields.
The principle of PEOE method is simple. It defines atomic electronegativity as follows
 A = a + bqA + cqA2
where qA is atomic charge of atom A, while a, b and c are parameters derived by Gasteiger et al.
based on ionization potential and electron affinity of corresponding element at neutral and charge
states. The parameters are dependent of atomic hybridization state, which can be determined
according to the number and type of bonded atoms.
PEOE charges are calculated via iterative process. In every iteration, certain amount of electron
is transferred between each pair of bonded atoms. In iteration n, the variation of atomic charge of A
due to its bonded atom B is calculated as:
q(An ) = f n [  B( n −1) −  A( n −1) ] /  Aq=1 when  B( n −1)   A( n −1)

q(An ) = f n [  B( n −1) −  A( n −1) ] /  Bq=1 when  A( n −1)   B( n −1)
The f is damping factor, which is usually set to 0.5, 𝐴𝑞=1 is the electronegativity of atom A at q=1
𝑞=1 𝑞=1
state, clearly 𝐴 = 𝑎 + 𝑏 + 𝑐. However, note that, only for hydrogen, the 𝐴 is always equal
to 20.02 eV. Since the f n term decreases swiftly with increase of iteration number n, the flow of
electrons between atoms is increasingly suppressed. In constrast to the EEM method described in
Section 3.19.5, the PEOE does not finally meet electronegativity equalization condition. The main
reason that PEOE method violates this condition is that the formula for evaluating  employed by
PEOE method is not quite rigorous, therefore damping factor must be introduced to compensate this.
Clearly, the total amount of charge variation of atom A in iteration n can be written as
  ( n −1) −  ( n −1)  ( n −1) −  A( n −1) 

q(An ) = f n   K q=1 A +  L 
 K A L  Lq=1 
where looped atoms K and L satisfy  K( n−1)   A( n−1) and  A( n−1)   L( n−1) , respectively.
The atomic charges at iteration n are updated as
q(An ) = q(An−1) + q(An ) ,

then the atomic electronegativities {(n)} are recalculated based on {q(n)} via the aforementioned
equation.
The iteration stops when maximum of charge variation is smaller than a given threshold. In
Multiwfn the threshold is set to 0.0001, in this case the iteration can usually converge after a dozen
of cycles. In Multiwfn the f parameter is fixed to 0.5.
The initial charges, namely {q(0)}, are set to zero. However, for very few kinds of atoms the q(0)
are set to specific value; for example, the =O atom in sulfonyl group has q(0) of -1.
Usage
Calculating PEOE charges are quite easy. You only need to load a file containing geometry
information into Multiwfn, then enter subfunction 19 of main function 7, the determined PEOE
parameters in the currenet system as well as calculated charges will be printed immediately.
If you want to monitor variation in atomic charges and atomic electronegativities in each cycle,
94
3 Functions
you can set "outmedinfo" in settings.ini to 1 before booting up Multiwfn.

Currently the PEOE module of Multiwfn supports H, C, N, P, O, S, F, Cl, Br, I, most parameters
are taken from Tetrahedron, 36, 3219 (1980), while some of them are supplemented from parameter
file of the Antechamber utility in AmberTools.
The interatomic connectivity involved in PEOE calculation is automatically guessed according
to bond length and atomic covalent radii. If the input file is in .mol or .mol2 format, then the
connectivity will be directly load from the input file.
It is strongly encouraged to optimize the geometry at least by lowest acceptable level that can
qualitatively represent the system, so that the actual connectivity can be correctly guessed according
to bond lengths. In addition, the hybridization state of tri-coordinated nitrogen atoms is determined
by the relative position to the three atoms bonded to it, thus it is even more important to provide an
optimized geometry when the system contains nitrogens.
Charged system is not supported by PEOE method.
An example of evaluating PEOE charges is given in Section 4.7.9.
Information needed: Atom coordinates
3.10 Orbital composition analysis (8)
Notice that the word “orbital” here is not restricted to molecular orbital, for example, if the
input file carries natural bond orbitals (NBO), then what will be analyzed is NBOs. There is an
excellent paper compared various orbital composition analysis approaches, see Acta Chim. Sinica,
69, 2393 (2011) (in Chinese, http://sioc-journal.cn/Jwk_hxxb/CN/abstract/abstract340458.shtml).
No matter what analysis method you choose, if you request Multiwfn to print composition of
various atoms in an orbital, in the output you can find a value "Orbital delocalization index" (ODI).
The lower the value, the stronger the orbital delocalization. When you intend to quantitatively
compare extent of spatial delocalization of various orbitals, you will find this index quite useful.
This ODI is detailedly described and illustrated in Section 4.8.5.
3.10.1 Output basis function, shell and atom composition in a specific

orbital by Mulliken, Stout-Politzer and SCPA approaches (1, 2, 3)
Mulliken, SCPA and Stout-Politzer methods support decomposing orbital to basis function,
shell and atom compositions. Actually I have introduced the theories in Sections 3.9.5, 3.9.6 and
3.9.7, i,a100% is just the composition of basis function a in orbital i, if we sum up all the
compositions of basis functions that within a shell we will get shell composition, and if we sum up
all the compositions of shells that attributed to the same atom we will get atom composition.
These approaches rely on basis expansion, in current Multiwfn version you must
use .mwfn, .fch, .molden or .gms as input file.
When you entered “Orbital composition analysis” submenu from main menu, select which
method you want to use for decomposition, and then input the index of orbital, the result will be
printed on screen immediately, you can also input -1 to print basic information of all orbitals to find
95
3 Functions
which one you are interested in. By default, only those terms with composition larger than 0.5%
will be printed, this threshold can be adjusted by “compthres” in settings.ini.
If the basis functions stored in .mwfn/.fch/.molden file are spherical harmonic type, then the
label of basis functions printed will look like D+1, F-3 rather than XX, XYY. The labels of spherical
harmonic basis functions used in Multiwfn are completely identical to Gaussian program, the
conversion relationship is:
D 0=-0.5*XX-0.5*YY+ZZ
D+1=XZ
D-1=YZ
D+2=√3/2*(XX-YY)
D-2=XY
F 0=-3/2/√5*(XXZ+YYZ)+ZZZ
F+1=-√(3/8)*XXX-√(3/40)*XYY+√(6/5)*XZZ
F-1=-√(3/40)*XXY-√(3/8)*YYY+√(6/5)*YZZ
F+2=√3/2*(XXZ-YYZ)
F-2=XYZ
F+3=√(5/8)*XXX-3/√8*XYY
F-3=3/√8*XXY-√(5/8)*YYY
G 0=ZZZZ+3/8*(XXXX+YYYY)-3*√(3/35)*(XXZZ+YYZZ-1/4*XXYY)
G+1=2*√(5/14)*XZZZ-3/2*√(5/14)*XXXZ-3/2/√14*XYYZ
G-1=2*√(5/14)*YZZZ-3/2*√(5/14)*YYYZ-3/2/√14*XXYZ
G+2=3*√(3/28)*(XXZZ-YYZZ)-√5/4*(XXXX-YYYY)
G-2=3/√7*XYZZ-√(5/28)*(XXXY+XYYY)
G+3=√(5/8)*XXXZ-3/√8*XYYZ
G-3=-√(5/8)*YYYZ+3/√8*XXYZ
G+4=√35/8*(XXXX+YYYY)-3/4*√3*XXYY
G-4=√5/2*(XXXY-XYYY)
H 0=ZZZZZ-5/√21*(XXZZZ+YYZZZ)+5/8*(XXXXZ+YYYYZ)+√(15/7)/4*XXYYZ
H+1= √ (5/3)*XZZZZ-3* √ (5/28)*XXXZZ-3/ √ 28*XYYZZ+ √ 15/8*XXXXX+ √ (5/3)/8*XYYYY+ √

(5/7)/4*XXXYY
H-1= √ (5/3)*YZZZZ-3* √ (5/28)*YYYZZ-3/ √ 28*XXYZZ+ √ 15/8*YYYYY+ √ (5/3)/8*XXXXY+ √

(5/7)/4*XXYYY
H+2=√5/2*(XXZZZ-YYZZZ)-√(35/3)/4*(XXXXZ-YYYYZ)
H-2=√(5/3)*XYZZZ-√(5/12)*(XXXYZ+XYYYZ)
H+3=√(5/6)*XXXZZ-√(3/2)*XYYZZ-√(35/2)/8*(XXXXX-XYYYY)+√(5/6)/4*XXXYY
H-3=-√(5/6)*YYYZZ+√(3/2)*XXYZZ-√(35/2)/8*(XXXXY-YYYYY)-√(5/6)/4*XXYYY
H+4=√35/8*(XXXXZ+YYYYZ)-3/4*√3*XXYYZ
H-4=√5/2*(XXXYZ-XYYYZ)
H+5=3/8*√(7/2)*XXXXX+5/8*√(7/2)*XYYYY-5/4*√(3/2)*XXXYY
H-5=3/8*√(7/2)*YYYYY+5/8*√(7/2)*XXXXY-5/4*√(3/2)*XXYYY
96
3 Functions
3.10.2 Define fragment 1 and 2 (-1, -2)
Before doing composition analysis for fragments by Mulliken, Stout-Politzer and SCPA
approachs, you have to define fragment in advance. If what you are interested in is only composition
of one fragment rather than the composition between two fragments (cross term composition), you
only need to define fragment 1. The content of fragment can be chosen to basis functions, shells,
atoms or mixture of them, whatever you choose, only the indices of corresponding basis functions
are recorded eventually. Notice that the "fragment" I referred here has no any relationship with the
"fragment" involved in Section 3.1, the fragment defined here does not disturb wavefunction at all.
All supported commands in the interface of defining fragment are self explanatory, so I will
not reiterate them but only give an examples, that is define fragment as all P-shells of atom 3: First,
type command all, information of all basis functions are listed, find out the shells that attributed to
center 3 and contain X, Y and Z type of basis functions (viz. PX, PY and PZ). Assume that the
indices of such shells are 3, 6 and 7, then input s 3,6,7 to add them into fragment. If you want to
verify your operation, input all again and check if asterisks have appeared in the leftmost of
corresponding rows, the marked basis functions are those that have been included in the fragment.
Finally, input the letter q to save current fragment and return to last menu, the indices of basis
functions in the fragment will be printed at the same time.
By default, fragments do not have any content. Each time you enter the fragment definition
interface, the status of fragment is identical to that when you leave the interface last time. So, if you
have defined the fragment earlier and you want to completely redefine it, do not forget to use “clean”
command to empty the fragment first.
3.10.3 Output composition of fragment 1 and inter-fragment

composition by Mulliken, Stout-Politzer and SCPA approaches (4, 5, 6)
After you defined fragment 1, the fragment composition analysis based on Mulliken, Stout-
Politzer and SCPA approaches is available. The fragment composition is the sum of all basis function
compositions within the fragment, in this function the fragment compositions of all orbitals are
printed on screen at the same time. If the analysis method you chose is Mulliken (subfunction 4) or
Stout-Politzer (subfunction 5), below component terms are outputted together with total
composition:
c^2 term: The sum of square of coefficients of basis functions within fragment 1, namely

afrag1
Ca2,i  100% .
Int.cross: The sum of internal cross terms in fragment 1, namely
 
afrag1 bfrag 2
Ca ,iCb,i Sa ,b  100% .
Ext.cross: Fragment 1 part of the total cross term between fragment 1 and all other atoms,
97
3 Functions
namely  
afrag1 bfrag1
wa ,b 2Ca ,iCb,i Sa ,b  100% .
It is clear that total composition of fragment 1 equals to c^2 term + Int.cross + Ext.cross.
If the fragment 2 is also defined (you must have already defined fragment 1), in subfunction 5
(Mulliken) or subfunction 5 (Stout-Politzer) the cross term between fragment 1 and fragment 2 in
each orbital, namely  
afrag1 bfrag 2
2Ca ,iCb,i Sa ,b  100% will be outputted too. “Frag1 part” and
“Frag2 part” correspond to the components of cross term attributed to fragment 1 and fragment 2
respectively, for Mulliken analysis the two terms are of course exactly equal due to the “equal
partition”.
3.10.4 Orbital composition analysis by natural atomic orbital approach

(7)
This function is used to calculate orbital composition based on natural atomic orbitals (NAOs).
This idea was proposed in my paper Acta Chim. Sinica, 69, 2393 (2011) http://sioc-
journal.cn/Jwk_hxxb/CN/abstract/abstract340458.shtml.
Theory
The first step of the famous natural bond orbital (NBO) analysis is converting original basis
functions to NAOs based on density matrix. Resulting NAOs can be classified into three categories:
• Core-type NAOs, describing inner core densities, their occupation numbers are almost equal
to integer
• Valence-type NAOs, describing valence densities, generally they have high occupation
numbers
• Rydberg-type NAOs, mainly displaying characteristics of polarization and delocalization of
electrons, the occupation numbers of them are very low
Core and valence NAOs are collectively called as minimal set, they have strong physical
meaning and have one-to-one correspondence with "actual" atomic orbitals, so they are what we
should be most concerned. Occupied MOs are almost exclusively contributed by minimal set NAOs.
Rydberg NAOs do not have clear physical interpretation, their contributions can be ignored in
occupied MOs, however they often have great contribution to virtual orbitals.
Since NAOs is an orthonormal set, if we have MO coefficient matrix in NAO basis, we can get
contribution from a NAO to specific MO by simply squaring corresponding expansion coefficient
and then multiplying it by 100%. Composition of an atom can be calculated as sum of composition
of minimal set NAOs in this center.
This orbital composition calculation method based on NAOs has great basis set stability as
Hirshfeld approach, it is especially suitable for analyzing composition of occupied orbitals.
However for virtual orbitals, since contribution from Rydberg NAOs is often large, this method no
longer works well.
Input file
The MO coefficient matrix in NAO basis cannot be generated by Multiwfn itself, you need to
provide an output file of NBO program containing this matrix as Multiwfn input file. By default
98
3 Functions
NBO program does not output this matrix, so you need to manually add NAOMO keyword between
$NBO ... $END field in NBO input file. The NBO program we referred here may be stand-alone
NBO program (also known as GENNBO), or NBO module embedded in quantum chemistry
software, such as L607 in Gaussian.
Options
After loading proper input file and enter present function, you will find following options in
the interface:
-1 Define fragment: This option is used to define fragment, which is needed by fragment
contribution analysis (option 1). All commands are self explanatory.
0 Show composition of an orbital: Print contribution from NAOs, shells and atoms to a
specific MO. At the meantime, contributions from core, valence and Rydberg type of NAOs are
reported respectively.
1 Show fragment contribution to a batch of orbitals: Print contribution from the fragment
defined by option -1 to specific orbitals.
2 Select output mode: This option controls which set of terms will be printed by option 0,
there are four modes:
(0) Show all terms
(1) Show non-Rydberg terms
(2) Show the terms whose contributions are larger than specific criterion
(3) Show non-Rydberg terms whose contributions are larger than specific criterion (default)
3 Switch spin type: You can find this option if the current system is open shell. You can select
the spin of the MOs to be analyzed.
Information needed: MO coefficients in NAO basis
3.10.5 Calculate atom and fragment contributions by Hirshfeld or

Hirshfeld-I method (8,10)
Hirshfeld and Hirshfeld-I weighting function (see Sections 3.9.1 and 3.9.13, respectively) can
also be used for decomposing orbital to atom and fragment compositions, the composition of atom
 (r ) wA (r ) d r  100% . The composition of a fragment is simply the sum of

2
A in orbital i is i
the compositions of the atoms that belongs to the fragment. These methods have great basis set
stability and are always more reliable and reasonable than Mulliken and MMPA. In fact the
Hirshfeld partition is already good enough, the more sophisticated and computationally demanding
Hirshfeld-I partition is not necessary.
If you choose to use Hirshfeld partition, you will be prompted to select the way to generate
atomic densities for constructing Hirshfeld weighting function, I strongly suggest using the built-in
atomic densities rather than using atomic .wfn files, since the former is much more convenient. If
you choose to use Hirshfeld-I partition, regular HI iterations will be performed first to yield
converged atomic weighting functions (if you are confused by the operations, please consult the
99
3 Functions
example of computing HI charges in Section 4.7.4 and the implementation details of Hirshfeld-I
introduced in Section 3.9.13).
Before calculating orbital composition, data initialization is automatically carried out. Once it
is finished, you can input the orbital index that you are interested in. Because numerical quadrature
always introduces some errors, so the sum of all atom compositions is not exactly equals to 100%,
the deviation might be relatively significant in rare cases, so Multiwfn normalizes results
automatically and prints them under the title “After normalization”.
If you want to view composition of an atom in specific range of orbitals at the same time,
choose option -2, then input the atom index and the index range of orbitals.
If you wish to study contribution of a fragment to orbitals, use -9 to define a fragment first,
then when you input an orbital index, the contribution of the fragment will be outputted along with
the contributions of all atoms. Also, you can choose -3 to calculate the contribution from the
fragment you defined to a range of orbitals.
If selecting option -4, program will calculate composition of every atom in every orbitals and
then export all of them to orbcomp.txt in current folder.
Information needed: Atom coordinates and GTFs
3.10.6 Calculate atom and fragment contributions by Becke method (9)
This function is very similar to the function introduced in Section 3.10.5, the only difference
is that Becke partition is used instead of Hirshfeld partition. For most cases, their results are in
qualitative agreement with each other. Using Becke partition instead of Hirshfeld partition has a
prominent advantage, namely the atomic wavefunction files are not needed, since the Becke atomic
space can be simply constructed based on atomic radius. For more detail about Becke partition, see
Section 3.18.0. An example is given in Section 4.8.3.
3.10.7 Calculate atom and fragment contributions by AIM method (11)
Multiwfn is also able to compute orbital composition based on atoms-in-molecules (AIM)

partition of molecular space. In this partition method, each atomic basin corresponds to space of an
atom, see Section 3.20 on detail about the concept of basin and AIM partition. To calculate orbital
composition under AIM partition, you should use subfunction 11 of basin analysis module (main
function 17), see Section 4.8.6 for example.
Usually I do not recommend to calculate orbital composition in this way, because the cost is
significantly higher than other ways while the result is not better.
100
3 Functions
3.10.100 Evaluate oxidation state by LOBA method (100)
This function is an implementation the LOBA method proposed in Phys. Chem. Chem. Phys.,
11, 11297 (2009). LOBA (localized orbital bonding analysis) is a method used to evaluate atomic
oxidation state based on orbital composition of localized MOs (LMOs). The idea is very simple: if
an atom has N nuclear charges, and its composition in M occupied LMOs are larger than a threshold
(e.g. 50%. In this case the electrons in these LMOs can be approximately viewed as completely
attributed to the atom. If a LMO is doubly occupied, it should be counted twice), then the oxidation
state of the atom is N-M.
In my opinion, this idea can also be extended to define fragmental oxidation state, namely if
the sum of nuclear charge in a fragment is N, and the fragment contribution to M LMOs are larger
than a certain threshold, then the fragment oxidation state will be N-M.
To use this function, you should provide .mwfn, .fch or .molden file containing LMOs (or
NBOs). For example, if you are a Gaussian user, you can use the .fch file resulting from
pop=saveNBO or pop=saveNLMO task as input. Alternatively, you can directly use Multiwfn to
carry out orbital localization to generate .fch containing localized orbitals. An example is given in
Section 4.8.4.
The result of LOBA method somewhat depends on the choice of orbital composition analysis
method. Multiwfn employs Hirshfeld method for LOBA analysis, it is very robust.
3.11 Bond order analysis (9)
In the bond order analysis module, you can directly select an option to analyze bond order by
corresponding method.
If you want to obtain total bond order between atoms in two molecular fragments, you can use
option -1 to define fragments 1 and 2 prior to bond order analysis. Then if you choose an option to
calculate bond order, the total bond order IRS between the two fragments will be calculated as follows
by summing up interatomic bond orders, and meantime be outputted along with two-center bond
orders
I RS =  I AB
AR BS
Evidently, interfragment bond order calculation is not available for multi-center bond order analysis, orbital
occupancy-perturbed Mayer bond order and Wiberg bond order decomposition analysis.
3.11.1 Mayer bond order analysis (1)
The Mayer bond order between atom A and B is defined as (Chem. Phys. Lett, 97, 270 (1983))
I AB = I AB + I AB

= 2 [( P S ) ba ( P S ) ab + ( P  S ) ba ( P  S ) ab ]
aA bB
where P and P are alpha and beta density matrix respectively, S is overlap matrix. Above formula
 
can be equivalently rewritten using total density matrix P = P + P and spin density matrix
101
3 Functions
P S = P − P 
I AB =  [( PS ) ba ( PS ) ab + ( P S S ) ba ( P S S ) ab ]
aA bB
For restricted closed-shell circumstance, since spin density matrix is zero, the formula can be
simplified to
I AB =  ( PS ) ab ( PS ) ba
aA bB
Generally, the value of Mayer bond order is in agreement with empirical bond order, for single,
double and triple bond the value is close to 1.0, 2.0 and 3.0 respectively. For unrestricted or restricted
open-shell wavefunction, alpha, beta and total Mayer bond orders will be outputted separately. By
default, only the bonds whose bond order exceed 0.05 will be printed on screen, the threshold can
be adjusted by “bndordthres” parameter in settings.ini, you can also select to export full bond order
matrix.
Moreover, Multiwfn outputs total and free valences, the former is defined as
V A = 2 (PS )aa − (PS )ab (PS )ba
aA aA bA
The latter is defined as

FA = VA −  I AB =  ( P s S ) ab ( P s S ) ba
B A aA bA
where Ps is spin density matrix, namely P minusing P.

For restricted closed-shell wavefunctions free valences are zero since Ps=0, thus total valence
of an atom is simply the sum of the related bond orders
VA = I
B A
AB
Total valence (also known as atomic valence) measures atomic bonding capacity, while free valence
characterizes the remaining ability of forming new bonds by sharing electron pairs.
For unrestricted or restricted open-shell system, there is another way to calculate total bond
order rather than summing up alpha and beta bond orders, that is summing up alpha and beta density
matrices to form total density matrix first and then calculate Mayer bond order by using restricted
closed-shell formula, this treatment is sometimes called “generalized Wiberg bond order“, these
total bond orders are printed following the title “Mayer bond order from mixed alpha&beta density
matrix”.
Similar to Mulliken population, Mayer bond order and the multi-center bond order described
below are sensitive to basis set, so do not use the basis sets having diffuse functions, otherwise the
bond order result will be unreliable.
Although Mayer bond order was originally defined for single-determinant wavefunctions, for
post-HF wavefunctions, Multiwfn calculates Mayer bond orders via exactly the same formulae as
shown above based on corresponding post-HF density matrix. The reasonableness of this treatment
has been validated in Chem. Phys. Lett., 544, 83 (2012).
Some applications of Mayer bond order can be seen in J. Chem. Soc., Dalton Trans., 2001,
2095.
102
3 Functions
3.11.2 Multi-center bond order analysis (2, -2, -3)
In main function 9 there are three options (2, -2, -3) used to calculate multi-center bond order,
they are very similar and will be introduced below in turn. Finally, a notable point about the input
order of atomic indices is mentioned.
Option 2: Standard multi-center bond order

The multi-center bond order (MCBO), which is also known as multi-center index (MCI), was
originally proposed in Struct. Chem., 1, 423 (1990), in some sense it may be viewed as an extension
of Mayer bond order to multi-center cases. Multiwfn supports it up to 12 centers.
Three/four/five/six-center bond orders are defined respectively as
I ABC =  ( PS ) ab ( PS ) bc ( PS ) ca
aA bB cC
I ABCD =   ( PS ) ab ( PS ) bc ( PS ) cd ( PS ) da
aA bB cC dD
I ABCDE =  ( PS ) ab ( PS ) bc ( PS ) cd ( PS ) de ( PS ) ea
aA bB cC dD eE
I ABCDEF =   ( PS ) ab ( PS ) bc ( PS ) cd ( PS ) de ( PS ) ef ( PS ) fa
aA bB cC dD eE f F
Similarly, infinite-center bond order can be written as

I ABCDEF K =   ( PS ) ab ( PS ) bc ( PS ) cd ( PS ) ka
aA bB cC kK
For open-shell cases, there are two definitions of the MCBO, the first one is the sum of alpha
part and beta parts:
I ABCDEF K = I ABCDEF K + I ABCDEF


K
 
= 2 n −1   ( P  S ) ab ( P  S ) bc ( P  S ) cd  ( P  S ) ka 
aA bB cC kK 
 
+ 2 n −1   ( P  S ) ab ( P  S ) bc ( P  S ) cd  ( P  S ) ka 
aA bB cC kK 
Another definition is using the mixed density matrix, this is not rigorous as above:
I ABCDEF K =   ( P mixed S ) ab ( P mixed S ) bc ( P mixed S ) cd  ( P mixed S ) ka
 
P mixed
=P +P
For unrestricted or restricted open-shell wavefunction, the output of MCBO analysis consists
of four terms, which have been explained above: (1) The result from alpha density matrix (2) The
result from beta density matrix (3) The sum of the result of alpha and beta parts (4) The result from
mixed alpha&beta density matrix. Commonly, if you are only interested in total MCBO, you should
use (3).
The computational cost for MCBO increases exponentially with the number of centers; for ring
size larger than 12 centers the cost will be prohibitively high, so Multiwfn does not support higher
number of centers. If you need to calculate MCBO for more than 12 centers, you have to calculate
103
3 Functions
AV1245 index instead, see Section 3.11.10 for introduction.

Notice that the MCBO for different number of centers are not directly comparable, since the
result is not in the same magnitude. However, in Phys. Chem. Chem. Phys., 18, 11839 (2016), it was
shown that the normalized MCBO is comparable for different ring size and can be simply calculated
as MCBO1/n, where n is the number of centers. For example, at B3LYP/6-31G* level, the MCBO
for H3+, benzene (6 centers) and naphthalene (12 centers) are 0.2963, 0.0863 and 0.0080,
respectively, while the normalized results are 0.667, 0.665 and 0.617, respectively. When MCBO is
negative, the normalized value will be calculated as -|MCBO|1/n. Commonly, if you need to compare
MCBO between different number of centers, you should take the normalized MCBO from the
information printed by Multiwfn, else using raw MCBO value is suggested.
Multiwfn is able to automatically search multi-center bonds. If you input -3 when Multiwfn
asks you to input atom combination, all three-center bond orders will be calculated, only those larger
than the threshold you inputted will be printed. Similarly, four-, five- and six-center bonds can be
searched by inputting -4, -5 and -6 respectively. Due to efficiency consideration, the search may be
not exhaustive. Also note that the search is based on mixed alpha&beta density matrix for open-
shell cases.
There is a hidden option -3 in main function 9, it is used to calculate MCBO under Löwdin
orthogonalized basis. The only difference between this option and the option 2 described above is
that this option performs Löwdin orthogonalization for basis functions before calculating the MCBO.
Since this method does not have obvious advantage over the standard MCBO definition, this option
is rarely used and thus invisible in the interface. However, if you have interesting, you can have a
try.
Option -2: Multi-center bond order in natural atomic orbital (NAO) basis
The most severe drawback of the MCBO is its high basis set dependency. In particular, if
diffuse functions are involved, then MCBO may be completely meaningless. In order to tackle this
problem, I proposed an alternative way (to be published) to calculate the MCBO, and the idea is
implemented as option -2.
Option -2 is very similar to option 2 (as introduced above), the only difference is that the
MCBO is calculated based on natural atomic orbital (NAO) basis rather than based on the basis
functions originally defined by the basis set. Since NAO is an orthonormal set and thus overlap
matrix S is an identity matrix, the formula can be simplified as (using closed-shell form for example)
I ABCDEF K =   Pab Pbc Pcd  Pka
The MCBO calculated in this manner has very good stability with respect to change in basis
set. Even if diffuse functions are presented the result is still fully reliable. According to my
experience, if no basis function shows diffuse character, the results given by option 2 and -2 will be
very similar, though not exactly identical.
In order to use option -2, the output file of NBO module embedded in Gaussian or standalone
NBO program (namely GENNBO) should be used as input file, and DMNAO keyword must be used
to make NBO print density matrix in NAO basis. If you are a Gaussian user, for example, you can
use output file of below instance as input file of Multiwfn (DO NOT use .fch file for this analysis!).
#p PBE1PBE/6-311G** pop=nboread
opted
104
3 Functions
0 1
C 0.00000000 1.38886900 0.00000000
... [ignored]
H -2.14060700 1.23588000 0.00000000
$NBO DMNAO $END
In this function, if you only input indices of two atoms, then the result is just Wiberg bond
order under NAO basis, which is completely identical to that printed by bndidx keyword of NBO
program.
Influence of input order of atomic indices on the result

Both the direction (e.g. A,B,C,D vs. D,C,B,A) and permutation (e.g. A,B,C,D vs. B,D,C,A ...)
of the inputted atomic indices can influence the calculated MCBO, below I describe this point in
detail.
• Input direction
Due to the mathematical form of the original MCBO (i.e. the one calculated by option 2), the
result of MCBO may relies on input direction. For example, the result yielded by inputting A,B,C,D
can be different from that by inputting D,C,B,A. The reason is clear: The term corresponding to
A,B,C,D is (PS)ab(PS)bc(PS)cd(PS)da, while if we invert the input order, the term will become
(PS)dc(PS)cb(PS)ba(PS)ad. Although both P and S are symmetry matrices, their product PS is not
necessarily symmetry, so the two terms are not equivalent. In my own viewpoint, in order to obtain
more reasonable result, if in a ring the atom connectivity is A-B-C-D-E-F (A also connects to F),
one should calculate A,B,C,D,E,F and F,E,D,C,B,A respectively and then take their average. If you
want Multiwfn to directly prints the averaged value, you can set "iMCBOtype" in settings.ini to 1,
in this case you do not need to manually perform the calculation twice, however, of course, the
computational cost is doubled compared to normal case.
An advantage of using option -2 to calculate MCBO in NAO basis and using option -3 to
calculate it in Löwdin orthogonalized basis is that the result is irrelevant to the input direction, this
is because in these cases the overlap matrix S is not explicitly involved and the density matrix P is
a symmetry matrix.
• Index permutation
Permutation of inputted atomic indices in the calculation of MCBO can significantly alter the
result. For example, the result of inputting 1,2,3,4,5,6 may be very different to that of inputting
2,4,3,5,6,1, regardless of which form of MCBO is used. If your aim is to study aromaticity and
characterize cyclic delocalization of electrons over a ring, you should input the atomic indices in
clockwise or anti-clockwise order, or take their average as mentioned above.
Some people advocated that it is needed to take all possible permutations into account to get a
definitive result, see J. Phys. Org. Chem., 18, 706 (2005); that means for a region consisting of six
atoms, the bond order of (B,C,A,D,E,F), (C,A,B,D,E,F), (D,B,C,A,F,E) and so on (6!=720 in total)
are all required to be taken into account. An explicit definition is given below, see Eq. 9 of Phys.
Chem. Chem. Phys., 18, 11839 (2016):
1
I perm =  I A,B,C...
2n Pˆ ( A,B ,C ...)
105
3 Functions
where n is the number of atoms involved in the calculation, 𝑃̂ is permutation operator that
generates all possible permutation sequences. The MCBO in this form is significantly more
expensive than the one in original definition, and it is not suitable for measuring aromaticity or
cyclic delocalization. However, it may be useful in measuring "global" electron delocalization
among atoms in a cluster-like region.
If you want to make Multiwfn directly print Iperm, you can set "iMCBOtype" in settings.ini to
2, then if you calculate MCBO as usual (via any of options 2, -2 and -3), the printed result will
correspond to Iperm.
Information needed: Basis functions (options 2, -3), NBO output file with DMNAO keyword
(option -2)
3.11.3 Wiberg bond order analysis in Löwdin orthogonalized basis (3)
Wiberg bond order is defined as follows, see footnote in Tetrahedron, 24, 1083 (1968)
I AB =  Pab2
aA bB
The original definition of Wiberg bond order is only suitable for the wavefunction represented
by orthogonal basis functions such as most of semiempirical wavefunctions, and only defined for
restricted closed-shell system. Actually, Mayer bond order can be seen as a generalization of Wiberg
bond order, for restricted closed-shell system and orthonormal basis functions (namely S matrix is
identity matrix) cases their results are completely identical.
In this function, Multiwfn first orthogonalizes basis functions by Löwdin method and then
performs usual Mayer bond order analysis. The threshold for printing is controlled by “bndordthres”
in settings.ini too.
As shown in J. Mol. Struct. (THEOCHEM), 870, 1 (2008), the Wiberg bond order calculated
in this manner, say WL, has much less sensitivity to basis set than Mayer bond order (whereas for
small basis sets, their results are closed to each other). One should be aware that WL tends to
overestimates bond order for polar bonds in comparison with Mayer bond order.
Commonly, if there is not special reason, using Mayer bond order is more preferred.
Notice that numerous literatures used NBO program to calculate Wiberg bond order, the result
must be somewhat different to that produced by present function, because in NBO program the
Wiberg bond orders are calculated in the basis of natural atomic orbitals (NAO), which are generated
by OWSO orthogonalization method. Multiwfn is also possible to calculate Wiberg bond order
under NAO basis, and furthermore, the result can be decomposed as atomic orbital pair contributions,
see Section 3.11.8 for details.
3.11.4 Mulliken bond order analysis (4) and decomposition (5)
Mulliken bond order is the oldest bond order definition, it is defined as
106
3 Functions
I AB = i  2Ca ,i Cb,i Sa ,b = 2 Pa ,b S a ,b

i aA bB aA bB
Mulliken bond order has low agreement with empirical bond order, it is deprecated for
quantifying bonding strength, for which Mayer bond order always performs better. However,
Mulliken bond order is a good qualitative indicator for bonding (positive value) and antibonding
(negative value). The threshold for printing results is controlled by “bndordthres” parameter in
settings.ini.
Mulliken bond order is easy to be decomposed to orbital contributions, the contribution from
orbital i to bond order AB is
i
I AB = i  2Ca ,i Cb,i S a ,b
aA bB
From the decomposition, we can know which orbitals are favourite and unfavourite for specific
bonding.
3.11.5 Orbital occupancy-perturbed Mayer bond order (6)
Orbital occupancy-perturbed Mayer bond order was firstly proposed in J. Chem. Theory
Comput., 8, 908 (2012). Put simply, by using this method one can obtain how large is the
contribution from specific orbital to Mayer bond order.
Orbital occupancy-perturbed Mayer bond order can be written as
,
I A ,B = I AB ,
+ I AB = 2 [( PX S ) ba ( PX S ) ab + ( PX S ) ba ( PX S ) ab ]
aA bB
The only difference between this definition and Mayer bond order shown in Section 3.11.1 is that
P  and P  have been replaced by PX and PX respectively. PX stands for the density
matrix generated when occupation number of a specific orbital is set to zero. The difference between
I A ,B and Mayer bond order can be regarded as a measure of contribution from the orbital to Mayer
bond order. Bear in mind, because Mayer bond order is not a linear function of density matrix, the
sum of I A ,B for all orbitals is not equal to Mayer bond order generally.
In Multiwfn, you only need to input indices of two atoms, then I A ,B for all occupied orbitals
and the difference between I A ,B and Mayer bond order will be outputted. The more negative
(positive) the difference, the more beneficial (harmful) to the bonding due to the existence of the
orbital.
You can also use another way to calculate I A ,B , that is using wavefunction modification
module (main function 6) to manually set occupation number of a specific orbital to zero, and then
calculate Mayer bond order as usual, but this manner may be tedious if you want to calculate I A ,B
107
3 Functions
for many orbitals.
This kind of analysis is illustrated in Section 4.9.1 and Section 4.19.3.
3.11.6 Fuzzy bond order (7)
Fuzzy bond order (FBO) was first proposed by Mayer in Chem. Phys. Lett., 383, 368 (2004):
BAB = BAB + BAB


= 2[( P S A )   (P S B ) + (P  S A )   (P  S B ) ]
 
S   =  wA (r )   (r )  (r )dr
A 
where S is overlap matrix of basis functions in fuzzy atomic spaces. In Multiwfn, Becke's fuzzy
atomic space with sharpness parameter k=3 in conjunction with modified CSD radii is used for
calculating FBO. (See Section 3.18.0 for introduction of fuzzy atomic space).
Commonly the magnitude of FBO is close to Mayer bond order, especially for low-polar bonds,
but much more stable with respect to the change in basis set. According to the comparison between
FBO and delocalization index (DI) given in J. Phys. Chem. A, 109, 9904 (2005), FBO is essentially
the DI calculated in fuzzy atomic space. See Section 3.18.5 for detail about DI.
Calculation of FBO requires performing Becke's DFT numerical integration, due to which the
computational cost is larger than evaluation of Mayer bond order. By default, 40 radial points and
230 angular points are used for numerical integration. This setting is able to yield accurate enough
results in general. If you want to further refine the result, you can set the number of radial and
angular points by "radpot" and "sphpot" in settings.ini manually, and ensure that "iautointgrid" has
been set to 0.
The threshold for printing results is controlled by “bndordthres” parameter in settings.ini.
3.11.7 Laplacian bond order (8)
In J. Phys. Chem. A, 117, 3100 (2013) (http://pubs.acs.org/doi/abs/10.1021/jp4010345), I
proposed a novel definition of covalent bond order based on the Laplacian of electron density 
2

in fuzzy overlap space, called Laplacian bond order (LBO). The LBO between atom A and B can be
simply written as
LA,B = −10   w (r)w (r)  (r) d r

2
A B
2  0
where w is a smoothly varying weighting function proposed by Becke and represents fuzzy atomic
space, hence wAwB corresponds to fuzzy overlap space between A and B. Note that the integration
is only restricted to negative part of 

2
 . The physical basis of LBO is that the larger magnitude
108
3 Functions
the integral of negative 

2
 in the fuzzy overlap space, the more intensive the electron density
is concentrated in the bonding region, and therefore, the stronger the covalent bonding.
The reasonableness and usefulness of LBO were demonstrated by applying it to a wide variety
of molecules and by comparing it with many existing bond order definitions. It is shown that LBO
has a direct correlation with the bond polarity, the bond dissociation energy and the bond vibrational
frequency. The computational cost of LBO is low, also LBO is insensitive to the computational level
used to generate electron density. In addition, since LBO is inherently independent of wavefunction,
one can in principle obtain LBO by making use of accurate electron densities derived from X-ray
diffraction data.
In Multiwfn, Becke's fuzzy atomic space with sharpness parameter k=3 in conjunction with
modified CSD radii is used for calculating LBO. (See Section 3.18.0 for detail about fuzzy atomic
space). The threshold for printing results is controlled by “bndordthres” parameter in settings.ini.
Note that in current implemention, LBO is particularly suitable for organic system, but not for
ionic bonds since in these cases a better definition of atomic space should be used to faithfully
exhibit actual atomic space. LBO is also not very appropriate for studying the bond between two
very heavy atoms (heavier than Ar), because these bonds are often accompanied by insignificant
charge concentration in the fuzzy overlap space, even though the bonding is doubtless covalent.
3.11.8 Decompose Wiberg bond order in NAO basis as atomic orbital

pair contributions (9)
Theory
As mentioned in Section 3.11.3, Wiberg bond order is expressed as
I AB =  Pab2
aA bB
The data is calculated for atom pair. Since the expression is simply a linear combination of square
of density matrix element, it is straightfoward to decompose Wiberg bond order as basis function
pair contribution (this idea is to be published). For example, Pab2 is simply the contribution from
interaction between basis functions a and b. Since one-to-one correspondence between basis
function and atomic orbital is lacking when extended basis set is used, in order to make the
decomposition method full of physical meaning, the decomposition is best to be carried out under
natural atomic orbitals (NAOs). Each non-Rydberg type of NAO uniquely corresponds to an atomic
orbital, thus, by above decomposition method, Wiberg bond order at the atomic orbital scale can be
obtained.
In addition, contribution from interaction between atomic orbital shells i and j can be obtained
as  P
ashell i bshell j
2
ab .
By the way, it is also possible to decompose multi-center bond order as atomic orbital
contribution. For example, three-center bond order is expressed as
109
3 Functions
I ABC =  PabPbc Pca

aA bB cC
Clearly, PabPbc Pca can be regarded as contribution from interacting between NAOs a, b and c.
However, decomposition of multi-center bond order under NAO basis has not been implemented.
Usage
After entering this function, simply input indices of two atoms, the nonnegligible contribution
from NAO pairs and NAO shell pairs together with total Wiberg bond will be printed.
The input file of this function is completely identical to the option -2 described in Section
3.11.2, namely the output file of NBO program containing density matrix information (i.e.
"DMNAO" keyword is required).
An illustrative example of this decomposition analysis is given in Section 4.9.4.
3.11.9 Intrinsic bond strength index (IBSI) (10)
Theory
The intrinsic bond strength index (IBSI) was proposed in J. Phys. Chem. A, 124, 1850 (2020)
to quantify strength of chemical bonds, it may also be used to compare strength of weak interactions.
The IBSI is defined in the framework of independent gradient model (IGM), which is very detailedly
described in Section 3.23.5. The IBSI is expressed as
(1 / d 2 )   g pair d r
IBSI =
(1 / d H2 2 )   g H2 d r
where d is the distance between the two atoms for which the interaction is to be studied. The integral
in the numerator is equivalent to the atom pair g index defined by me between the two atoms, see
Section 3.23.6 for detail. The denominator is the data for reference system, the 𝑑H2 and the integral
are the bond length and atom pair g index of H2 in its equilibrium structure, respectively. In the
original paper of IBSI, it was shown that the IBSI value is modestly positively correlated with
strength of covalent bond. Furthermore, it was found that magnitude of IBSI of transition metal
coordinate bond is markedly smaller than that of covalent bond, and magnitude of IBSI of weak
interactions is even much lower, this feature of IBSI may be used to distinguish type of interaction.
Implementation
It is important to note that in the IBSI paper the authors calculated the IBSI using the IGM
based on gradient-based partition (IGMGBP), however this form of IGM is not supported by
Multiwfn. Currently Multiwfn supports the original form of IGM, namely IGM based on
promolecular approximation (IGMpro), and also supports IGM based on Hirshfeld partition of
molecular density (IGMH), which I believe is better than IGMGBP. Different forms of IGM
correspond to different ways of evaluating gradient of atomic density, and thus the value of the
integral in the IBSI expression is correspondingly different. In Multiwfn, the IBSI can be computed
based on both IGMpro and IGMH, their results are very different, I found the result based IGMpro is
obviously closer to the original paper of IBSI.
110
3 Functions
Usage
To calculate IBSI, you simply need to enter main function 9 and select subfunction 10, then
select option 0 to start calculation.
Before calculation, Multiwfn asks you to choose quality of integration grid, evidently the better
the grid, the higher the cost, while the more accurate the result. According to my experience, for
IGMpro, "medium quality" is already able to give quantitatively accurate result; while for IGMH, at
least "high quality" should be used if you have requirement on accuracy.
You can choose which form of IGM will be used in the IBSI calculation by option 2. If the
input file contains wavefunction information, by default IGMH is used, while if the input file only
contains geometry information (e.g. .pdb, .mol, .xyz...), only IGMpro can be used. Note that IGMH
is much more expensive than IGMpro, since its formule to evaluate gradient of atomic density is
much more complicated.
If you are only interested in the interactions in a local region, you can choose option 3 to define
the region to be studied, only the IBSI between the atoms in the defined fragment will be evaluated
and outputted, the cost is correspondingly lower than the IBSI calculation for the whole system,
especially when the system is huge.
The reference value can be set by option 4, it corresponds to the denominator of the IBSI
formule. Of course, this value must be different for IGMpro and IGMH. The default reference values
of both IGMpro and IGMH cases were calculated for H2 with experimental structure (0.74144 Å), in
the latter case B3LYP/6-311G** wavefunction was employed. Commonly, the reference value is
not needed to be changed. However, for calculating IBSI in terms of IGMH, if you want to obtain
the reference value at your current calculation level to pursue stricter result, you can load
wavefunction file of H2, then enter the present function, set reference value to 1.0, then use option
0 to start to calculate IBSI, the result can be employed as reference value for studying practical
molecules.
In order to avoid excessive output, by default only the data for the atom pairs with separation
smaller than 3.5 Å are printed, because IBSI should be negligible if the separation is larger. If you
need to adjust the distance printing threshold, use option 5.
Example of calculating IBSI is given in Section 4.9.6.
Information needed: Atom coordinates (for IBSI based on IGM), GTF information (for IBSI
based on IGMH)
3.11.10 AV1245 index (approximate multi-center bond order for large

rings) and AVmin
Theory
The multi-center bond order (MCBO), as introduced in Section 3.11.2, is a very rigorous and
popular way of characterizing aromaticity. Unfortunately, the computational cost of MCBO
increases sharply with increase of the number of atoms involved in the ring, therefore commonly
MCBO can only be applied to rings composed of up to 12 atoms.
In order to quantify aromaticity for larger rings, in Phys. Chem. Chem. Phys., 18, 11839 (2016)
111
3 Functions
the author proposed the AV1245 index. This index can be regarded as an approximation of MCBO,
the key advantage of AV1245 over MCBO is that the computational cost of AV1245 only increases
linearly with number of atoms, therefore AV1245 can be very easily used to characterize aromaticity
for rings composed of dozens of atoms and may even be used for rings containing several hundreds
of atoms.
The definition of AV1245 in the original paper is "average all the 4c-ESI values along the ring
that keeps a positional relationship of 1, 2, 4, 5". Here I clarify its definition. As an instance, for
below ring,
its AV1245 is calculated as
AV1245=[ESI(1,2,4,5)+ESI(2,3,5,6)+ESI(3,4,6,1)+ESI(4,5,1,2)+ESI(5,6,2,3)+ESI(6,1,3,4)] / 6
where the nc-ESI (n-center electron sharing index) can be directly obtained via the multi-center
bond order with all possible permutations (namely I perm, see Section 3.11.2 for detail) by below
relationship
2
ESI = I perm
(n − 1)!
where n is the number of atoms. Evidently, 4c-ESI = (4c-Iperm) / 3.
The central idea of AV1245 is based on the fact that in an aromatic ring, the resonance between
1-2 bond and 4-5 bond is strong, as illustrated below. This feature can be captured by ESI(1,2,4,5).
Larger value of AV1245 of a cyclic path implies stronger delocalization and thus larger
aromaticity of the ring. Since magnitude of AV1245 is small, it is often multiplied by 1000 when
presenting the data.
It is worth to note that in the original paper of AV1245, the MCBO was calculated in terms of
atomic overlap matrix under AIM partition, this way of calculation is not only expensive but
complicated. In Multiwfn, the MCBO involved in the AV1245 is calculated in usual way, namely
based on density matrix and overlap matrix. Since in this case the calculation of 4-center I perm is
fairly cheap, the AV1245 can be quickly obtained even for large systems and macro-rings. However,
due to the difference in the calculation of MCBO, the result of AV1245 produced by Multiwfn is
smaller than that in the original paper. In addition, when directly calculating AV1256 by Multiwfn
(in other words, calculating AV1256 in original basis functions), the employed basis set should not
contain diffuse functions, otherwise the AV1256 will be meaningless.
Multiwfn also supports calculating AV1245 in natural atomic orbital (NAO) basis, in this case
reasonable result can be obtained even if diffuse functions are presented. If there is no diffuse
function, the result calculated in original basis functions and that in NAO basis is nearly the same.
112
3 Functions
The AVmin index was proposed in J. Phys. Chem. C, 121, 27118 (2017) and further discussed
in Phys. Chem. Chem. Phys., 20, 2787 (2018). It corresponds to minimal absolute value of all 4c-
ESIs involved in the calculation of AV1245. Unlike AV1245, AVmin quantifies lowest degree of
conjugation in the whole pathway, therefore it has unique value in distinguishing aromaticities of
different delocalization pathways, since according to common intuition, aromaticity of a path should
be predominated by the local region mostly disconnecting the delocalization over the whole path.
In other words, AVmin is able to determine the bottleneck of aromaticity of a given path.
Usage
• Calculating AV1245 and AVmin in original basis functions
You should enter subfunction 19 of main function 200, then input indices of the atoms in the
ring in the order of connectivity (clockwise or counterclockwise along the ring). After that, the
AV1245 together with its constituent 4c-ESI values as well as AVmin will be outputted
Multiwfn provides a convenience for inputting the atomic indices for large rings. If you input d first and press
ENTER button, then you can input the atomic indices in arbitrary order, because in this case the actual order will be
automatically guessed according to interatomic connectivities. However, this input mode cannot be used when any
atom in the ring connects to more than two other atoms in the ring.
• Calculating AV1245 and AVmin in NAO basis

The operation process is the same as "Calculating AV1245 and AVmin in original basis
functions", however, you should use output file of standalone NBO program or the NBO module
embedded in quantum chemistry codes as input file of Multiwfn, and meantime the "DMNAO"
keyword must be employed in the NBO analysis. Note that in this case the aforementioned "d" mode
of index inputting is not available.
Examples of employing AV1245 and AVmin to study aromaticity of small rings and large rings
are given in Section 4.9.11.
Information needed: Atom coordinates, basis functions
3.12 Plotting total density-of-states (DOS), partial DOS,

overlap population DOS, local DOS and photoelectron
spectrum (10)
3.12.1 Theory
Density-of-states (DOS) is an important concept of solid physics, which represents the number
of states in unit energy interval, since energy levels are contiguous, so DOS can be plotted as curve
map. In isolated system (such as molecule), the energy levels are discrete, the concept of DOS is
questionable and some people argued that DOS is completely valueless in this situation. However,
if the discrete energy levels are broadened to curve artifically, DOS graph can be used as a valuable
tool for analyzing the nature of electron structure.
113
3 Functions
The original total DOS (TDOS) of isolated system can be written as

TDOS(E ) =   ( E −  i )
i
where {} is eigenvalue set of single-particle Hamilton,  is Dirac delta function. If  is replaced by
broadening function F(x), such as Gaussian, Lorentzian and pseudo-Voigt function, we get
broadened TDOS.
The normalized Gaussian function is defined as
x2
1 − FWHM
G ( x) = e 2c 2
where c =
c 2 2 2 ln 2
FWHM is acronym of “full width at half maximum”, it is an adjustable parameter in Multiwfn,
the larger FWHM the TDOS graph looks more smooth and analysis is easier to perform, but more
fine-structure is masked.
The normalized Lorentzian function is defined as
FWHM 1
L( x ) =
2 x + 0.25  FWHM2
2
Pseudo-Voigt function is weighted linear combination of Gaussian function and Lorentzian

function:
P( x) = wGaussG( x) + (1 − wGauss ) L( x)
Obviously, if G(x) and L(x) are normalized, normalization condition for P(x) always holds
regardless of the select of wGauss.
The curve map of broadened partial DOS (PDOS) and overlap DOS (OPDOS) are very
valuable for visual study of orbital composition. PDOS of fragment A is defined as
PDOS A ( E ) =  i , A F ( E −  i )
i
where i,A is the composition of fragment A in orbital i. Note that the word "projected DOS" used
in some literatures is essentially equivalent to the partial DOS. The OPDOS between fragment A
and B is defined as
OPDOSA,B ( E ) =   iA,B F ( E −  i )
i
where  iA, B is the composition of total cross term between fragment A and B in orbital i. I have
discussed how to calculate  and  in Section 3.10.3.
The local DOS is a special kind of DOS, which will be introduced in Section 3.12.4. Since the
TDOS is closely related to photoelectron spectra (PES), present module is also able to PES, this
point will be introduced in Section 3.12.5.
Example of DOS: Ferrocene

Below is DOS map of a typical molecule ferrocene, Lanl2DZ basis set in combination with
Lanl pseudopotential is used for iron, 6-31G* is employed for other elements. In this system Z-axis
is perpendicular to cyclopentadienyl groups.
114
3 Functions
The graph clearly exhibits orbital characteristics in different energy ranges, each discrete lines
corresponds to a MO. The curves are yielded by applying broadening function to the discrete lines.
The left and right Y-axes correspond to the curves and discrete lines, respectively. Notice that only
relative height rather than absolute height of curves is meaningful. It is obvious that the major
contribution from s, px and py atomic orbitals of carbon (magneta curve) is due to low-lying MOs
rather than frontier MOs. The major composition of MOs around -0.25 a.u. comes from pz orbital
of carbon (blue curve) and iron atom (red curve). Inspection of the green OPDOS curve, which
corresponds to the bonding between carbon pz and iron atom, suggests that carbon pz orbitals are
very important for stablization of ferrocene, since OPDOS has large positive value in these ranges.
HOMO is almost purely contributed by iron orbitals, however its slight overlap with carbon pz
orbitals is still beneficial to bonding. For all virtual MOs, OPDOS curve is in negative region and
shows antibonding characteristic, this is due to the unfavorable overlapping in orbital phase, as can
been seen from LUMO isosurface.
3.12.2 Input file
.mwfn/.fch/.molden/.gms files can be used as input. You can also use output file of single point
task of Gaussian program as input (pop=full keyword must be specified). For generality, Multiwfn
supports plain text file as input too, the format is free, there is no upper limit of the number of
orbitals. The format of the file should be
nmo inp
energy occ [strength] [FWHM]  For orbital 1
...
energy occ [strength] [FWHM]  For orbital nmo
where energy and occ denote orbital energy and occupation number, respectively. nmo is the number
of orbitals recorded in this file. inp is input type, there four cases:
115
3 Functions
 1: Only energies (in a.u.) and occupation numbers will be loaded, while strengths and FWHMs
of all orbitals will be automatically set to 1.0 and 0.25 a.u., respectively
 2: The same as 1, but you also have to specify strength and FWHM for each orbital as columns
3 and 4 because in this case they will also be loaded. If the strength of an orbital is set to k,
then the broadened curve from this orbital will be normalized to k rather than 1 (default value)
 3: The same as 1, but energy unit is eV.
 4: The same as 2, but energy unit is eV.
3.12.3 Options for plotting DOS and basic usage
When you entered the interface of plotting DOS, you will see following options. Notice that
options -1 and 7 only appear when the input file contains basis function information, that is if you
want to draw PDOS and OPDOS graph, you have to use .mwfn/.fch/.molden/.gms file as input.
Also note that in this interface, you can input s to save current status (plotting settings, fragment
definition and orbital information) to a specific file, and you can input l to load status from a specific
file, so that you can quickly replot a map without redoing the settings.
-5 Customize energy levels, occupations, strengths and FWHMs for specific MOs: By this
option, you can manually set energies, occupations, strengths and FWHMs for specific orbitals. For
example, if you would like only to plot DOS for a few MOs, you can set strengths of the other MOs
to zero (by default, strengths for all MOs are 1.0).
-4 Show all orbital information: Print information of all orbitals on screen.
-3 Export energy levels, strengths, FWHMs to plain text file: Export energy, strength and
FWHM of each orbital to orginfo.txt in current directory, this file complies the format introduced in
last section, so can be directly used as input file.
-2 Define MO fragments for MO-PDOS: The "MO-PDOS" is a special kind of PDOS, which
is used to reveal DOS contributed by different sets of MOs (rather than atoms or basis functions),
the DOS curves and discrete lines corresponding to different sets of MOs are drawn using different
colors. This option is used to define different sets of MOs involved in the "MO-PDOS" plot. See
Section 4.10.5 for example.
-1 Define fragments for PDOS/OPDOS: You can define up to 10 fragments in this option.
After you entered this option, the information of present fragments are shown on screen. You can
input x to define fragment x, and input -x to unset fragment x, or input i,j to exchange the definition
of fragment i and j. To leave this interface, input 0. The PDOS will be plotted for all of the fragments
defined in this option. OPDOS will be drawn only when both fragments 1 and 2 have been defined.
DOS graph will contain the line and curve of “PDOS frag 1” only when fragment 1 is defined.
0 Draw TDOS graph!: After selecting this, TDOS map will be immediately shown on screen.
If any fragment has been defined, the texts in this option will become “TDOS+PDOS”; if both
fragments 1 and 2 have been defined, the texts will become “TDOS+PDOS+OPDOS”.
1 Select broadening function: Select which broadening function will be used, you can select
Lorentzian, Gaussian or Pseudo-Voigt function. Default is Gaussian.
2 Set energy range: This option is used to set the lower limit, upper limit and step between
labels of X-axis.
3 Set full width at half maximum (FWHM): As the title says.
4 Set scale ratio for DOS curve: If this option is set to k, then height of all curves will multiply
116
3 Functions
k (in full energy range).

5 Set Gaussian-weighting coefficient: This option sets wgauss, which is mentioned in last
section. This option only appears when Pseudo-Voigt function is chosen.
6 Choose orbital spin: This option appears only when the loaded file contains basis function
information and the wavefunction is unrestricted. This option determines which set of orbitals (alpha,
beta, or both alpha and beta) will be taken into account.
7 Set the method for calculating PDOS: Mulliken (default), SCPA, Hirshfeld and Becke
methods are supported for calculating orbital compositions used in plotting PDOS, via this option
you can select one of them. Hirshfeld and Becke methods are more robust than Mulliken and SCPA
methods, especially for unoccupied MOs, unfortunately they are more expensive, and in this case
OPDOS cannot be plotted (i.e. the fragments can only be defined as a set of atoms). It is very
important to note that when diffuse functions are presented, the PDOS maps plotted based on
Mulliken or SCPA methods will be meaningless! If you diffuse functions cannot be ignored due to
special reasons, Hirshfeld or Becke method have to be employed.
8 Switch unit between a.u. and eV: You can switch the unit for X-axis of the DOS map via
this option. eV is more commonly used.
9 Toggle using line height to show orbital degeneracy: If you want to show degeneracy of
orbitals in the DOS map, you can choose this option to enable this effect, then you will be asked to
input threshold of energy difference for determining degeneracy. This feature can be used when
plotting TDOS and MO-PDOS, but cannot be utilized when plotting PDOS.
Once you choose option 0 in the DOS module, Multiwfn starts to calculate data and then DOS
graph pops up. You can see there is a vertical dash line, which highlights position of HOMO level.
Note that some people believe this is Fermi energy, which, however is an ill-defined concept for
isolated systems, any energy that  EHOMO and < ELUMO may be regarded as "Fermi energy".
After closing the graph, a post-processing menu appears on the screen, it contains many options,
which are self-explanatory and can be used to adjust various plotting parameters. When the
parameters have been changed, you can choose "1 Show graph again" to check the effect.
Worthnotingly, there is an option named “Set scale factor of Y-axis for OPDOS”, if the value is set
to k and the range of left-axis (for TDOS/PDOS) is set to e.g. [-3.5, 2.0], then the range of right-axis
(for OPDOS) will become [-3.5*k, 2.0*k]. The reason why Multiwfn uses double axis is because
the magnitude of OPDOS is generally much smaller than TDOS and PDOS. You can also choose to
export the DOS map as graphical file in current folder, or export X-Y data set of DOS to plain text
files so that you can reproduce the graph by third-part softwares, such as Origin. By choosing option
0 you can return to last interface, the quality of DOS graph can be gradually improved by repeating
the adjustments until you are satisfied.
The absolute value of the DOS curves does not have evident practical meaning, only relative
height of the curves at different energy regions is interesting. Hence, after you have properly set
each plotting parameter, you can choose "13 Toggle showing labels and ticks on Y-axis" in the post-
processing menu to switch its status to "No".
Very detailed examples of plotting TDOS, PDOS and OPDOS are given in Section 4.10.1,
while Section 4.10.3 illustrates how to plot DOS for open-shell systems via Multiwfn in combination
with Origin software to get better effect.
117
3 Functions
3.12.4 Local DOS
There is a special kind of DOS named local DOS (LDOS), which is also known as spatial DOS.
LDOS curve for a given point r is evaluated as follows:
LDOS(r, E ) = i2 (r ) f ( E −  i )
i
It is worth to note that in some first-principle books and literatures, the word "Local DOS" in fact refers to the
PDOS introduced in Section 3.12.1, do not be confused!
To plot LDOS map, choose option "10 Draw local DOS for a point" after you entered main
function 10, you will be prompted to input the coordinate of the point r.
One can also plot LDOS for a set of points placed evenly in a line as color-filled map, the X-
axis correspond to energy, while the Y-axis corresponds to coordinate in the line relative to starting
point. To plot this kind of LDOS map, choose "11 Draw local DOS along a line", you will be
prompted to input the coordinate of starting point and end point defining the line, as well as the
number of points consisting of the line.
In the DOS module, the options controlling FWHM, energy unit, energy range and scale ratio
affect the resulting LDOS graphs.
Note that fragment definition does not affect the result of LDOS, i.e. the LDOS always
corresponds to total DOS. However, if you want to separate angular moment contribution to LDOS,
you can use subfunction 25 of main function 6 to set coefficient of unwanted GTF in all MOs to
zero, then they will not contribute to LDOS.
An example of plotting LDOS can be found in Section 4.10.2.
3.12.4 Photoelectron spectrum
The plotting of photoelectron spectrum (PES) spectrum based on (generalized) Koopmans'

theorem) is closely related to plotting TDOS. Since PES is a kind of spectrum under frequent studies,
a special interface in the DOS module is provided for easily generating theoretically simulated PES
spectrum.
Theory
The position of the peaks in the PES spectrum reflects the energy difference between various
N-1 states and the original N-electron state. If a system is neutral, the system is usually in vibrational
ground state of neutral electonic state. After ionization of an electrons, the system can be in different
vibrational states of the cationic state. Therefore, the PES has a fine structure, which reflects the
vibration coupling effect. However, in order to simplify the problem, we often ignore the quantum
effect of nuclear motion, and the ionization is assumed to be a vertical process starting from
minimum point of potential energy surface of initial state. At this time, the peak positions in the PES
are equivalent to the vertical ionization energy (VIP) of the electrons of different shells of present
system. Clearly, as long as we optimize the system and then calculate the VIPs, we can simulate the
PES.
The 1st VIP corresponds to the energy needed for ionizing out the electron of outermost shell,
118
3 Functions
it is usually calculated as E(N-1) - E(N) for a system containing N electrons at minimum point
geometry of potential energy surface of the N-electron state; the E signifies electronic energy. The
VIP corresponding to inner electrons can also be theoretically calculated, but special methods are
needed, such as OVGF, IP-EOM-CC, ADC and so on.
The easiest way to plot PES is based on the Koopmans' theorem, which stated that the
ionization energy of an electron is equal to the negative value of orbital energy of corresponding
shell. Note that this is only an approximate relationship, it fully ignores the electron correlation and
orbital relaxation effect. Under the Koopmans' theorem, the simulated PES simply corresponds to
the TDOS curve broadened by all occupied MOs, however before the broadening the sign of orbital
energies should be reversed (corresponding to "electronic binding energy").
Because of the fact that Koopmans' theorem does not work well for most popular DFT
functionals (except for some special ones, such as QTP17), the negative values of MO energies
deviate from actual VIP evidently, leading to poor PES compared to the experimental one.
Fortunately, there is a so-called generalized Koopmans' theorem, if it is applied to theoretical
simulation of PES, it essentially corresponds to adding a shift value to all electronic binding energies
before PES plotting, the shift value is defined as 1st VIP + E(HOMO). After taking the shift into
account, the first peak of simulated PES will exactly correspond to the 1st VIP, which does not
deviate from the first peak of experimental PES evidently as long as the DFT functionals and basis
set are properly chosen, the geometry has been substantially optimized, and the electronic state to
be computed corresponds to actual ground state.
Usage
The PES plotting interface is embedded in the DOS module as option 12, it allows user to
extremely easily plot PES spectrum based on (generalized) Koopmans' theorem. Any kind of input
files that used for plotting DOS can also be employed for PES plotting purpose.
If you want to plot the PES based on Koopmans' theorem, simply select option 1 after entering
the PES plotting interface. If you want to adapt generalized Koopmans' theorem, you should choose
option 3 to set a shift value before plotting.
There are many parameters and options in the PES plotting interface similar to those in the
DOS plotting interface, such as range and step of X and Y axes, FWHM, if showing discrete lines
and so on, however most parameters are not shared by the two interfaces. Note that only eV unit
and Gaussian broadening function can be employed for PES plotting. If present wavefunction is an
unrestricted open-shell one, the type of orbital spin is not distinguished.
If you want to adjust relative height between the PES curve and the discrete lines corresponding
to various binding energy levels, you can use "11 Set scale ratio for PES curve" to set a scale ratio.
The smaller the ratio, the lower the curve.
If nothing is shown after choosing option 1 to plot the PES spectrum, you should:
 Select option "-2 Show all binding energy level information" and then check if values of
binding energies are correct.
 Check if range of X-axis has been properly set. To plot PES, there must be at least one
binding energy occurs within the current range of X-axis.
By default, all orbitals have identical strength (1.0) and FWHM (0.2 eV), the latter can be
directly set in the PES interface via option 6. If you want to adjust strength and FWHM for individual
orbitals, you can select "-3 Export occupied MO energies, strengths and FWHMs to plain text file",
then a file PESinfo.txt in a format as described in Section 3.12.2 will be exported to current folder.
119
3 Functions
You can then manually adjust strengths and FWHMs of certain orbitals in this file, and then use this
file as input file to plot PES, so that the simulated PES could be more close to the experimental one.
The absolute value of the PES curve does not have practical meaning. Hence, after you have
properly set range of Y-axis, you can choose "13 Toggle showing labels and ticks on Y-axis" to
switch its status to "No".
An example of plotting PES is given in Section 4.10.4.
Information needed: For PDOS, OPDOS and local DOS, see Section 3.12.2. For TDOS and
PES spectrum, use .mwfn/.fch/.molden/.gms file, or Gaussian output file with pop=full, or plain text
file in a format as described in Section 3.12.2.
3.13 Plotting IR, Raman, UV-Vis, ECD, VCD, ROA and

NMR spectra (11)
Although spectrum plotting is beyond the scope of wavefunction analysis, due to this function
is very useful for theoretical spectrum research, the spectrum plotting function is also included in
Multiwfn. Multiwfn is able to plot IR (Infrared), normal Raman / Pre-resonance Raman, UV-Vis
(ultraviolet-visible), ECD (electronic circular dichroism), VCD (vibrational circular dichroism) and
Raman optical activity (ROA) spectra, see Sections 3.13.1~3.13.4. Many examples of plotting
various kinds of spectra are given in Section 4.11.
Multiwfn is also able to plot NMR spectrum, see Section 3.13.5 for introduction and Section
4.11.10 for examples.
3.13.1 Theory
For comparing theoretical results with experimental spectrum, the discrete lines corresponding
to each transition mode have to be broadened to emulate the situation in real world, the commonly
used broadening function for vibrational spectra (IR, Raman, VCD and ROA) is Lorentzian function,
while for electronic spectra (UV-Vis and ECD), Gaussian function is often used, see Section 3.12.1
for details. Unlike DOS graph, in which the strengths for each energy level are always simply set to
1.0 (namely broadened curves from each energy level are normalized to 1.0), the strengths of
transitions are very important data for plotting spectrum.
The unit of transition energy of IR, Raman, VCD and ROA spectra is cm-1 in common, for UV-
Vis and ECD spectra all of eV (1 eV=8.06551000 cm-1), nm and 1000 cm-1 are common units. The
strength data of each transition mode outputted by quantum chemistry programs is proportional to
the area under the broadened curve. Below we discuss some details of the six types of spectra
supported by Multiwfn respectively.
IR: The frequently used unit of molar absorptivity , namely L(molcm)-1, can be rewritten as
L 1000cm 3 1000cm 2
= =
mol  cm mol  cm mol
120
3 Functions

Since IR intensity  =  ( ) d , where  is transition frequency, the unit of  should be
1000 cm2 1000 cm 0.01 km

(cm−1 ) = =
mol mol mol
However km/mol is the more commonly used unit for IR intensity, therefore if IR intensity of a
vibrational mode is p km/mol, then the broadened curve from which should be normalized to 100*p
(in the other words, the area under the curve is 100*p). Sometimes the unit esu2cm2 is used for IR
intensity, the relationship with km/mol is 1 esu2cm2 = 2.5066 km/mol.
Below is an example of IR spectrum plotted by Multiwfn. Notice that the left axis corresponds
to the curve (artificially broadened data), the right axis corresponds to the discrete lines (original
transition data).
7480.21 473.59
6707.25 424.65
Molar absorption coefficient (L/mol/cm)
5934.30 375.72
5161.34 326.78
IR intensities (km/Mol)
4388.39 277.84
3615.43 228.90
2842.48 179.96
2069.52 131.03
1296.57 82.09
523.61 33.15
-249.34 -15.79
0.00 401.37 802.74 1204.12 1605.49 2006.86 2408.23 2809.60 3210.98 3612.35 4013.72
Frequencies (cm**-1)
Raman (normal or pre-resonance): Raman spectrum measures intensity of scattered light.

Beware that for a specific vibration mode i, its Raman activity Si and Raman intensity Ii are two
different quantities. Raman activity is a intrinsic property of each molecular vibrational mode, while
Raman intensity is directly related to experimental Raman spectrum, and its value is dependent of
the choice of wavenumber of incident light 0 as well as temperature. The conversion relationship
can be found in some literatures, e.g. J. Mol. Struct., 702, 9 (2004):
C ( 0 − vi ) 4 Si  hc i 
Ii = Bi = 1 − exp  − 
 i Bi  kT 
where C is a suitably chosen common normalization factor for all peak intensities, i is vibrational
frequency. Only the Raman spectrum broadened based on Raman intensities is strictly comparable
to the experimental one; however, as the peak position of Raman activity and intensity are identical,
and for a specific vibrational mode its intensity is proportional to its activity, the Raman spectrum
broadened from Raman activities is also useful in some sense.
Almost all quantum chemistry codes only output Raman activities, therefore by default,
121
3 Functions
Multiwfn emulates the Raman spectrum in terms of broadening Raman activities. However you can
also let Multiwfn to first convert the activities to intensities via option 19 (v0 and T are provided by
users) based on above formula, then the Raman spectrum will be obtained by broadening Raman
intensities. The normalization coefficient C is fixed to 10-12 in Multiwfn (which is unimportant since
what we are interested in is the shape rather than the absolute height of Raman spectrum). The
integral of the peak broadened by one unit of Raman activity or intensity is equal to 1.
UV-Vis: In theoretical chemistry field, oscillator strength is always used for representing
transition strength in UV-Vis spectrum, however this is a dimensionless quantity and the precise
relationship with molar absorptivity have not been built. However, there is a well-known relation
between theoretical data and experimental spectra: If the unit 1000 cm-1 and L/mol/cm are used for
X-axis and Y-axis respectively, then the area under the curve that broadened from per unit oscillator
strength should be 1/4.32*106. Equivalently, if eV is used as X-axis unit, the value should be
1/4.32/8.0655*106=28700. By this relation, UV-Vis spectrum can be simulated by theoretical data.
Fluorescence (or phosphorescence, as long as you can obtain oscillator strength by means of
spin-orbit coupling calculation) spectrum can also be plotted with exactly the same manner as UV-
Vis, the only difference is that according to Kasha's rule, you should only take the first singlet excited
state into account.
ECD：The significance of rotatory strengths in ECD spectrum is analogous to oscillator
strengths in UV-Vis spectrum, each electron transition mode corresponds to a rotatory strength. If
the rotatory strengths are broadened, after somewhat scaling and shifting, the resultant curve will be
comparable to experimental ECD spectrum. The integral of the peak broadened by one unit of
rotatory strength is equal to 1. In quantum chemistry programs, such as Gaussian, rotatory strengths
can be calculated in length representation or in velocity representation, in the former case the
strengths are origin-dependent, while in the latter case the strengths are origin-independent. For
complete basis set situation, the results in the two representations converge to the same values.
Commonly velocity representation is recommended to be used.
VCD: VCD is akin to ECD, each vibrational mode has a rotatory strength. The integral area
under the curve ∆𝜀 = 𝜀L − 𝜀R corresponding to a vibrational mode is proportional to its rotatory
strength. Therefore after broadening rotatory strengths, the curve shape will be comparable to
experimental VCD spectrum.
ROA: The ROA spectum measures the difference between scattering intensity of right and left
circularly polarized light
( 0 − vi )4 Ai  hc i 
ROA intensity  I iR − I iL  Bi = 1 − exp  − 
 i Bi  kT 
The ROA strength data outputted by Gaussian ROA task in fact is the Ai term, which should be
converted to actual ROA intensity according to above equation. The Ai term is dependent of
frequency of incident light. The integral of the peak broadened by one unit of ROA intensity/strength
is equal to 1. There are several different forms of ROA, including ROA SCP(180), ROA SCP(90),
ROA DCP(180). The 90 and 180 denote the angle between incident light and scattered light. The
SCP (scattered circular polarization) means the incident light is linearly polarized light while the
scattered light is circularly polarized light; DCP (dual circular polarization) corresponds to the case
that both incident and scattered lights are circularly polarized light. The ROA SCP(180) is
commonly employed, and it is also known as SCP backscattered ROA.
122
3 Functions
The ROA task of Gaussian also simultaneously outputs frequency-dependent Raman strength,
which corresponds to the Ri term of below equation
( 0 − vi )4 Ri  hc i 
Raman intensity  I iR + I iL  Bi = 1 − exp  − 
 i Bi  kT 
correspondingly, there are also Raman SCP(180), Raman SCP(90) and Raman DCP(180) data.
Since the current used incident light should be far from electron excitation energy, such Raman
spectrum is called as far from resonance Raman.
3.13.2 Input file
Only the input files mentioned in this section are supported by Multiwfn for plotting spectrum
purpose. DO NOT use such as .fch, .molden and .wfn as input file, evidently the data needed for
plotting spectrum are not recorded in these files.
1 Gaussian output file

• IR spectrum: Use output file of freq task as input. If you use Gaussian 09 D.01 or later
revisions and meantime freq=anharm keyword was specified to carry out anharmonic analysis,
Multiwfn will prompt you to choose if loading anharmonic frequencies and IR intensities instead of
the harmonic ones.
• Raman spectrum: Use output file of freq=raman task as input. If you would like to plot pre-
resonance Raman spectrum, you should at the same time use CPHF=rdfreq keyword and write the
frequencies of the incident lights after a blank line under the geometry specification, e.g. 300nm
400nm 500nm. If you hope to plot anharmonic Raman spectrum, use freq(raman,anharm) keywords,
then Multiwfn will prompt you to choose if loading anharmonic frequencies and Raman activities
instead of the harmonic ones.
• VCD spectrum: Use output file of freq=VCD task as input. If you hope to plot anharmonic
Raman spectrum (supported by Gaussian since G16), use freq(VCD,anharm) keywords, then
Multiwfn will prompt you to choose if loading anharmonic frequencies and rotatory strengths
instead of the harmonic ones.
For anharmonic IR, Raman and VCD spectra, you can choose if only loading anharmonic
fundamental data, or simultaneously loading anharmonic overtone band or combination band data.
• UV-Vis or ECD spectrum: Use output file of TDDFT, TDHF, CIS, ZINDO or EOM-CCSD
task as input, no any additional keywords are required. The rotatory strengths in both length and
velocity representation (the data under "R(length)" and "R(velocity)", respectively) can be chosen
to be loaded by Multiwfn. Output file of optimization task for excited state can also be used, only
the last output for transition information will be loaded by Multiwfn, therefore the resulting
spectrum corresponds to final geometry.
• ROA spectrum: Use output file of freq=ROA task as input. The frequencies of incident lights
should be specified after a blank line under the geometry specification, e.g. 0.02, 0.03, 0.04, 0.05,
the unit is default to a.u.; or write e.g. 500nm 520nm 550nm.
2 ORCA output file

• IR spectrum: Use freq keyword.
• Raman spectrum: Use keywords like below
! b3lyp def2-SVP numfreq
123
3 Functions
%elprop Polar 1 end

• UV-Vis and ECD spectrum: Use keywords like below (CIS, TDHF and ZINDO calculations
are also supported)
! b3lyp def2-SVP
%tddft
nroots 20
TDA false
end
Notice that for TDDFT calculation, ORCA by default uses Tamm-Dancoff Approximation
(TDA), in this case the oscillator strength and rotatory strengths are evidently not as good as that
produced by TDDFT, therefore in above example TDA false is used to make ORCA use standard
TDDFT formalism. Also it is worth noting that ORCA employs length representation for rotatory
strengths in ECD.
Output file of sTDA or sTD-DFT task of ORCA (see below for details) is also supported by
Multiwfn for plotting UV-Vis or ECD spectrum purpose. Below is example keywords:
! PBE0 def2-SVP def2/J RIJCOSX
%maxcore 6000
%pal nprocs 36 end
%tddft
Mode sTDDFT
Ethresh 10.0
maxcore 6000
end
Since ORCA 4.1 spin-orbit coupling (SOC) effect can be taken into account during TDDFT
calculation, and all data needed for plotting SOC corrected UV-Vis and ECD spectra are
automatically outputted. Therefore, when output file of SOC-TDDFT task is loaded into Multiwfn,
when you enter UV-Vis and ECD plotting option, you can select if loading SOC corrected data
instead of the one without SOC consideration. It is very easy to conduct SOC-TDDFT, simply
adding dosoc true in %tddft section, for example:
%tddft nroots=40 TDA false dosoc true end
• VCD and ROA spectra: Currently not supported by ORCA.
3 Grimme's sTDA output file

The sTDA proposed by Grimme in J. Chem. Phys., 138, 244104 (2013) is a method
approximately solve the TDDFT equation and thus reduces the computational cost of the electron
excitation part of TDDFT calculation by about two or three orders of magnitude. The correpsonding
sTDA program can be freely downloaded at
https://www.chemie.uni-bonn.de/pctc/mulliken-center/software/stda/stda
Although the Grimme's sTDA code has already been implanted into ORCA program, the
tda.dat file outputted by standalone sTDA code can also be used as input file of Multiwfn for plotting
UV-Vis or ECD purpose.
When ECD is to be plotted, you can choose which representation of rotatory strength will be
used. The length and velocity representations have been mentioned above, while the mixed-form of
representation, which is recommended in the sTDA original paper, is defined as RM= RV  fL / fV,
where fL and fV are length and velocity representations of oscillator, respectively, and RV is the
124
3 Functions
velocity representation of rotatory strength.
4 Grimme's xtb output file

The xtb program written by Grimme is mainly used to carry out GFN-xTB calculation (J. Chem.
Theory Comput., 13, 1989 (2017) and J. Chem. Theory Comput., 15, 1652 (2019)), which may be
viewed as a semiempirical variant of DFT method. It is not only robust but also rather fast, it can be
conveniently applied to systems consisted of hundreds of atoms. Note that although the accuracy of
xtb frequency has been verified to be basically reasonable, the quality of IR intensities outputted by
xtb is not good (according to my experiences). xtb program can be freely obtained via
https://www.chemie.uni-bonn.de/pctc/mulliken-center/software/xtb/xtb.
By using following command, xtb will optimize the structure in test.xyz and then perform
frequency analysis: xtb test.xyz -ohess |tee test.out. Then the output file test.out will contain
harmonic frequencies and IR intensities, it can be used as input file of Multiwfn for plotting IR
spectrum purpose.
Note that xtb outputs 3N modes, where N is the number of atoms. When plotting spectrum,
Multiwfn will ask you how many lowest modes will be removed to discard the overall movement
modes, you should input 5 and 6 for linear and non-linear system, respectively.
5 Plain text file

For generality, Multiwfn supports plain text file as input, you can extract transition data from
output files of quantum chemistry packages other than Gaussian/ORCA/sTDA, and then fill them
into a file according to the format shown below
numdata inptype
energy strength [FWHM]  For transition 1
...
energy strength [FWHM]  For transition numdata
where numdata denotes how many entries in this file. If inptype is set to 1, then only energy and
strength will be read, and FWHMs for all transitions will be automatically set. If inptype is set to 2,
then FWHMs will be read too. The transitions should be sorted according to the energies from low
to high. The unit of both energy and FWHM should be in cm-1 for IR, Raman, VCD and ROA spectra,
in eV for UV-Vis and ECD spectra. The unit of strength should be in km/mol, Å4/amu, cgs (10-40
erg-esu-cm/Gauss), 10-44 esu2 cm2, 104 K for IR, Raman, ECD, VCD, ROA spectra, respectively
(oscillator strength of UV-Vis is dimensionless).
An example of the plain text file is shown below (for IR)
6 2
81.32920 0.72170 8.0
417.97970 3.58980 8.0
544.67320 21.06430 8.0
583.12940 41.33960 8.0
678.66900 91.47940 8.0
867.37410 2.94480 8.0
125
3 Functions
3.13.3 Usage and options
After boot up Multiwfn, first input path of Gaussian/ORCA/sTDA/xtb output file or the plain
text file containing transition data, then enter main function 11, and you will be asked to select the
type of spectrum (for pre-resonance Raman and ROA, you also need to choose the frequency of
interest), after that you will see below options. The meaning of some options may be different for
various types of spectra.
Note that in this interface you can input s anytime to save current plotting settings to a file, or
input l to load plotting settings from a file, so that you can quickly recover plotting status.
-4 Set format of saving graphical file: This is used to choose the format of the graphical file
exported by option 1. Commonly, using vector format such as ps, pdf and svg is recommended.
-2 Export transition data to plain text file: Output energies, intensities and FWHMs of all
transitions to transinfo.txt in current directory, this file fully complies with the format introduced in
last section, so it can be directly used as input file.
-1 Show transition data: Print energy and intensity data of all transitions on screen.
0 Plot spectrum: Plot spectrum right now! The spectrum will be shown on screen, and
meantime the minima and maxima of the spectrum curve will be shown in console window.
1 Save graphical file of the spectrum in current folder: As the title says.
2 Export X-Y data set of lines and curves to plain text file: Export X-Y data set of lines and
broadened spectrum to spectrum_line.txt and spectrum_curve.txt in current directory, respectively,
you can replot the curve and discrete line graph directly by the two files via external program, such
as Origin.
3 Set lower and upper limit of X-axis: As the title says. The stepsize between ticks also needs
to be inputted. By default the range of X-axis is adjusted automatically according to minimum and
maximum transition energies.
4 Set left Y-axis: Set starting value, ending value and step size for left Y-axis. By default the
range of Y-axis is adjusted automatically according to the maximum peak.
5 Set right Y-axis: Like option 4, but for right Y-axis.
In many cases, after you adjusting setting for Y-axis at one side, the position of zero point of
this Y-axis deviates from that of Y-axis at another side, which makes the graph weird; To address
this problem, options 4 and 5 enable you to choose if correspondingly adjusting range of Y-axis at
another side. If you input y, then lower/upper limit and stepsize of Y-axis at another side will be
proportionally scaled to make zero point of left and right Y-axes are in the same horizontal line.
6 Select broadening function: Gaussian, Lorentzian and Pseudo-Voigt function can be
selected for broadening discrete lines as curves.
7 Set scale ratio for curve: If the value is set to k, then the height of curve will multiply k in
full range. For Raman, ECD, VCD and ROA spectra the default value is 1.0, for IR the value is 100,
for UV-Vis spectrum an empirical value 28700.0 is used when energy unit is eV or nm, when unit
is 1000 cm-1 the value 1/(4.32*10-6) is used.
8 Input full width at half maximum (FWHM): As the title says.
9 Toggle showing discrete lines: Choose if show the discrete lines corresponding to transitions
on the spectrum graph.
10 Switch the unit of infrared intensity / Set the unit of excitation energy: For IR spectrum,
switch the unit of IR intensities between km/mol (default) and esu2*cm2. For UV-Vis and ECD
126
3 Functions
spectra, choose the unit of energies between eV, nm and 1000 cm-1.
11 Set Gaussian-weighting coefficient: Sets wgauss, which is mentioned in Section 3.12.1, this
option only appears when Pseudo-Voigt function is chosen.
12 Set shift value in X: If the value is set to k, then the final curve and discrete lines will be
shifted by k in X direction.
13 Set colors of curve and discrete lines: As the title says.
14 Set scale factor for transition energies (or vibrational frequenices): The selected
transition energies (or vibrational frequencies) will be scaled by a given factor when drawing
spectrum. This is option is mainly used to apply frequency scaling factor onto the theoretically
calculated harmonic frequencies.
15 Output contribution of individual transitions to the spectrum: If select this option, user
will be prompted to input a criterion (e.g. k), then not only the total spectrum (as option 2), but also
the contributions from the individual transitions whose absolute value of strength larger than k will
be outputted to spectrum_curve.txt in current folder. This feature is particularly useful for identifying
the nature of total spectrum.
16 Set status of showing labels of spectrum minima and maxima: You will enter an interface,
in which there are many options used to set how to show spectrum minima and maxima on the
plotted map, they are self-explanatory.
17 Other plotting settings: Contains various trivial plotting settings, such as if showing
dashed grid lines, if showing labels on axes, etc.
18 Toggle weighting spectrum of each system: See explanation in next section.
19 Convert Raman (or ROA) activities to intensities: As mentioned earlier, this option is
used to convert Raman (or ROA) activities to intensities. After that, the Raman (or ROA) spectrum
plotted by option 0 will be the one broadened based on Raman (or ROA) intensities rather than
Raman (or ROA) activities. The wavenumber of incident light and temperature are inputted by users.
20 Modify strengths: You can select some transitions and change their strengths (depending
on spectrum type) to specific value. This option is very useful if you would like to plot fluorescence
spectrum, in this case you need to set oscillator strength of all transitions except for the lowest
singlet excitation to zero (Kasha's rule).
21 Set status of showing weighted curve and curves of individual systems: As shown in the
next section, Multiwfn is able to plot weighted spectrum according to given weights of multiple
systems. This option controls if plotting weighted curves and curves of various systems.
22 Set thickness of curves/lines/texts/axes: As the title says.
23 Set status of showing spikes to indicate transition levels: Via this option, you can plot
one or multiple sets of spikes at bottom of the spectrum to indicate position of different sets of
transitions. If some transitions are degenerate, the height of spikes can be used to reflect degree of
degenerate. See Section 4.11.9 for illustration of this option. This option is not available when
multiple systems are considered in the plotting simultaneously.
Note that the actual number of points constituting the spectrum is controlled by "num1Dpoints"
parameter in settings.ini. The default value is commonly large enough and thus need not to be
adjusted.
127
3 Functions
3.13.4 Plotting multiple systems together and weighted spectrum
In Multiwfn, it is possible to simultaneously plot spectrum for multiple files, and meantime
taking weights into account. This feature is quite useful for obtaining actual spectrum of flexible
molecule with many thermally accessible conformations.
For example, there is a molecule containing four accessible conformations, their distribution
ratio have been determined according to Boltzmann's method based on calculated free energies. If
you want to plot its weighted spectrum and spectrum of each conformation, you should write a plain
text file named multiple.txt (other file name cannot be recognized by Multiwfn), the content is:
boltz\Excit\a.out 0.6046
boltz\Excit\b.out 0.1950
boltz\Excit\c.out 0.1686
boltz\Excit\d.out 0.0317
where the first and second column correspond to path of input file and weight of each conformation.
Different type of input file can be presented together in this file. If you use this file as input file,
after you entered spectrum plotting module, Multiwfn will load data from these files in turn. When
plotting spectrum, spectrum of all the four conformations will be separately calculated and drawn
with different colors. In addition, the weighted spectrum is plotted on the graph as thick red curve,
it is simply evaluated as follows:
weighted spectrum = 0.6046a + 0.1950b + 0.1686c + 0.0317d
In the graph you can also observe many black discrete lines, these are collection of discrete lines
corresponding to different conformations. Notice that their heights have already been multiplied by
corresponding weights.
If before plotting spectrum you have chosen option 18 once, then the spectrum of various
conformations shown on graph are the weighted ones; in other words, these curves represent
contribution of each conformation to actual spectrum (thick red curve). In this case, the color of
weighted discrete lines are no longer all black but in accordance with color of curves, so that user
can easily recognize correspondence between curve and discrete among different systems.
If you simply want to simultaneously plot spectrum for multiple systems but do not want to
draw weighted spectrum, then the second column of multiple.txt should not be weights, but
customized legends, for example
boltz\Excit\a.out Molecular A
boltz\Excit\b.out Molecular B
boltz\Excit\c.out Molecular C
boltz\Excit\d.out Molecular D
Then plot spectrum as usual, curve of all the four systems will be shown together, and the legends
will be "Molecular A", "Molecular B", etc.
When the legend is completely composed of numbers, the legend will be regarded as weight. If the legend have
to be a number, you should add $ in front of the legend to let Multiwfn know it is legend rather than weight. For
example, $50 will be recognized as legend of “50”.
If you are using Linux platform and some file paths in the multiple.txt contains / symbol or
space, you should add double quotation marks at the two ends of the file path, so that the path could
be properly loaded.
128
3 Functions
3.13.5 Plotting NMR spectrum
Theory
Many quantum chemistry programs are able to compute magnetic shielding tensors  at nuclei,
usually only the average of its diagonal terms, namely isotropic magnetic shielding value iso, is of
interest, because NMR spectrum is commonly determined in solvent environment and the solute is
able to freely rotate.
The chemical shift  is calculated as follows
 ref − 
=   ref − 
1− ref
where ref is shielding value of reference substance (tetramethylsilane for 13C and 1H NMR), while
the  is shielding value of the sample to be determined. The two values should be evaluated at
exactly the same calculation setting (theoretical method, basis set, solvation model...).
Another way of determining chemical shift is scaling method, see http://cheshirenmr.info for
detail. Briefly speaking, this method evaluate  in a very simple way:
 = a  + b
where the a and b are prefitted slope and intercept parameters based on a training set for specific
calculation level. For example, if the optimization is conducted under vaccum using B3LYP/6-31G*
while shielding value is calculated using B3LYP/6-31G* under chloroform environment represented
by SMD solvation model, then a = -1.0157 and b = 32.2109 for 1H NMR, while a = -0.9449 and b
= 188.4418 for 13C NMR.
If there are methyl groups in the system, the  of its three hydrogens should be averaged before
generating NMR spectrum, since methyl groups can freely rotate in actual environment due to its
extremely low rotation barrier. In addition, if there are multiple thermally accessible conformers and
they can interconvert with each other easily, the weighted average shielding value should be
calculated for each atom based on weights of the conformers:
 A =  pi iA
i
where i is conformer index and A is atom index. pi stands for weight of conformer i, which can be
calculated according to Boltzmann distribution based on relative free energies between the
conformers.
With  of all atoms in hand, one can plot discrete line map of NMR, the X-axis of the spikes
in the map corresponds to  while Y-axis corresponds to degeneracy, which is N if maximal spacing
between N chemical shifts is less than specific threshold (e.g. 0.05 ppm), for other cases the
degeneracy is 1.0.
Actual NMR spectrum with finite peak width can be generated by broadening the spikes by
Lorentzian function, the full width at half maximum (FWHM) is a key parameter of controlling peak
shape. In this curve map, the peak height corresponds to strength of NMR signal.
Input file
Output file of NMR task of Gaussian and ORCA programs are directly supported as input file
for plotting NMR spectrum.
• Example of NMR task of Gaussian
# B972/def2tzvp scrf=solvent=chloroform NMR
129
3 Functions
Title Card Required
0 1
N -0.14557000 1.69364200 0.17479300
H -1.18363600 1.59873700 0.51496300
[...ignored]
• Example of NMR task of ORCA

! B3LYP/G 6-31G* NMR cpcm(chloroform) RIJK autoaux
* xyz 0 1
C 0. 1.14218 0.72212
C 0. 1.19872 -0.67302
[...ignored]
Plain text file recording atom names and shielding values can also be used as input file for
plotting NMR spectrum, see examples\spectra\NMR\general.txt for example. Thus in principle
Multiwfn can plot NMR for any program that can yield shielding values.
Usage
After loading input file, entering main function 11 and selecting NMR, you will see the NMR
plotting interface. Then if you select option 0, NMR spectrum will be displayed, and raw data will
be shown on console window. Since most options in this interface are self-explanatory, only a few
noteworthy options are mentioned below.
• -2 Export NMR data to a plain text file: This option exports shielding values to
NMRdata.txt in current folder. If you have used option 7 to convert shielding values to chemical
shifts, then both of them will be exported.
• 6 Choose the element considered in plotting: Via this option you can choose the element
to be considered. For example, if you want to plot 1H NMR spectrum or examine relevant data, you
should select this option and input H.
• 7 Set how to determine chemical shifts: By default, the X-axis in the plotted NMR spectrum
corresponds to magnetic shielding value, while if want to make chemical shift  as X-axis, you
should select this option prior to plotting. In this option you can select the way of determine :
(1) Taking difference between shielding value of current system and reference system. You can
manually input reference value or directly use built-in data.
(2) Using scaling method to determine the . you can manually input the fitted slope and
intercept, or directly use built-in data.
There are several sets of built-in data, they correspond to the levels very suitable for calculating
.
• 10 Average shielding values of specific atoms: For example, usually the  of the hydrogens
in the same methyl group should be averaged, you can use this function to realize this purpose.
• 11 Set strength of specific atoms: By default, every atom has strength of 1.0, namely
contributes to degeneracy by 1.0. You can use this option to modify strength of specific atoms. For
example, if you set strength of atoms 3, 6, 7 to 0, then these atoms will disappear in the plotted
spectrum.
• 16 Change setting of labelling atoms: In order to distinguish the atoms in the NMR spectrum,
Multiwfn labels atomic indices on the plotted peaks. You can use this option to control details of
130
3 Functions
labelling, such as position, size, color, content of the labels and so on.
Plotting conformation weighted spectrum and spectrum containing multiple systems

Conformation weighted NMR spectrum can be plotted by using plain text file as input (the
name must contain "multiple", for example, valine_multiple.txt is a valid name), each line contains
path of an input file and corresponding conformation weight, for example:
D:\valine\conf1.out 0.43
If you simply want to plot NMR spectrum of multiple systems together, specifying file path
and legend in each line, for example
/lovelive/nico.out Isomer 1
/lovelive/nozomi.out Isomer 2
Then after entering NMR plotting interface, you can plot the spectrum via exactly the same
steps as plotting single system. Note that all systems must have the same number of atoms.
Special notes
To save the NMR spectrum as graphical file, you can choose option 1. It is highly suggested to
use pdf format, because it is a vectorial format, the spectrum can be losslessly zoomed in and out,
and the lines and texts look very smooth. The format can be chosen by option -3, and the default
format can be changed by "graphformat" in settings.ini.
After adjusting various plotting settings, you can input s in the interface to save plotting settings
to a specific text file, so that when you want to replot the NMR spectrum with exactly the same
effect in the future, you can input l to directly restore plotting settings from the text file. Note that
any operation on plotted data must be manually redone everytime (e.g. averaging shielding values,
converting shielding values to chemical shifts, etc.).
Some examples of plotting NMR spectrum are given in Section 4.11.10.
3.14 Topology analysis (2)
3.14.1 Theory
The topology analysis module of Multiwfn is quite flexible, powerful and general, it can be
applied to any real space function supported by Multiwfn as long as the the variation of the function
is smooth and thus its gradient is contiguous. All commonly studied functions, such as electron
density, its Laplacian, orbital wavefunction, spin density, ELF, LOL, ESP... can be analyzed by this
function. The default real space function to be studied is electron density, it can be altered by option
-11. Notice that once option -11 is chosen, all previous topology analysis results will be clean.
The topology analysis technique proposed by Bader was firstly used for analyzing electron
density in "atoms in molecules" (AIM) theory, which is also known as "the quantum theory of atoms
in molecules" (QTAIM), this technique has also been extended to other real space functions, e.g. the
first topology analysis research of ELF for small molecules is given by Silvi and Savin, see Nature,
371, 683 (1994). In topology analysis language, the points at where gradient norm of function value
131
3 Functions
is zero (except at infinity) are called as critical points (CPs), CPs can be classified into four types
according to how many eigenvalues of Hessian matrix of real space function are negative.
(3,-3): All three eigenvalues of Hessian matrix of function are negative, namely the local
maximum. For electron density analysis and for heavy atoms, the position of (3,-3) are nearly
identical to nuclear positions, hence (3,-3) is also called nuclear critical point (NCP). Generally the
number of (3,-3) is equal to the number of atoms, only in rarely cases the former can be more than
(e.g. Li2) or less than (e.g. KrH+) the latter.
(3,-1): Two eigenvalues of Hessian matrix of function are negative, namely the second-order
saddle point. For electron density analysis, (3,-1) generally appears between attractive atom pairs
and hence commonly called as bond critical point (BCP). The value of real space functions at BCP
have great significance, for example the value of  and the sign of ∇2 𝜌 at BCP are closely related
to bonding strength and bonding type respectively for analogous bonds (The Quantum Theory of
Atoms in Molecules-From Solid State to DNA and Drug Design, p11); I have demonstrated that the
 at BCP can be reliably used to predict hydrogen bond binding energies (J. Comput. Chem, 40,
2868 (2019)); local information entropy at BCP is a good indicator of aromaticity (Phys. Chem.
Chem. Phys., 12, 4742 (2020)).
(3,+1): Only one eigenvalue of Hessian matrix of function is negative, namely first-order
saddle point (like transition state in potential energy surface). For electron density analysis, (3,+1)
generally appears in the center of ring system and displays steric effect, hence (3,+1) is often named
as ring critical point (RCP).
(3,+3): None of eigenvalues of Hessian matrix of function are negative, namely the local
minimum. For electron density analysis, (3,+3) generally appears in the center of cage system (e.g.
pyramid P4 molecule), hence is often referred to as cage critical point (CCP).
The positions of CPs are searched by Newton method, one need to assign an initial guess point,
then the Newton iteration always converge to the CP that is closest to the guess point. By assigning
different guesses and doing iteration for each of them, all CPs could be found. Once searches of CPs
are finished, one should use Poincaré-Hopf relationship to verify if all CPs may have been found,
the relationship states that (for isolated system)
n(3,-3) – n(3,-1) + n(3,+1) – n(3,+3) = 1
If the relationship is unsatisfied, then some of CPs must be missing, you may need to try to
search those CPs by different guesses. However even if the relationship is satisfied, it does not
necessarily mean that all CPs have been found. Notice that the function spaces of ELF/LOL and
Laplacian of  are much more complex than , it is very difficult to locate all CPs for these functions,
especially for middle and large system, so, you can stop trying for searching CPs once all CPs that
you are interested in have been found.
The maximal gradient path linking BCP and associated two local maxima of density is termed
as “bond path”, which reveals atomic interaction path for all kinds of bonding. The collection of
bond paths is known as molecular graph, which provides an unambiguous definition of molecular
structure. Bond path can be straight line or curve, obviously for the latter case the length of bond
path is longer than the sum of the distances between BCP and associated two (3,-3) CPs.
Let us see an example. In the complex shown below, the imidazole plane is vertical to
magnesium porphyrin plane, the nitrogen in imidazole coordinated to magnesium. Magenta, orange
and yellow spheres correspond to (3,-3), (3,-1) and (3,+1) critical points, brown lines denote bond
paths.
132
3 Functions
The topology paths for other real space functions can be generated too in Multiwfn, they are
very helpful to clarify intrinsic relationship between CPs.
About degenerate CP
Appearance of degenerate CPs in general system is very rare, so I mention it in the final of this
section. The so-called degenerate CP is the CP have one or two zero eigenvalues of Hessian matrix.
These CPs are very instable, slight perturbation of molecular geometry can break them into other
type CPs. However, degenerate CPs are commonly occur in axisymmetric system. For example, in
fluorine atom at groundstate, one of p orbital is singly occupied while other two are doubly occupied,
topology analysis on LOL gives below picture:
The two (3,-3) correspond to the two valves of the singly occupied p orbital, while the locally
maximum of electron localization arise from the other two p orbitals are represented by a circle of
CPs. Ostensibly, the CPs in the circle are either (3,-1) or (3,+1) type, but in fact they are the same
type degenerate CP, in which one eigenvalue is zero. Simply because the reasons of numerical
convergence of the eigenvalue, they are formally recorded as different CP types in Multiwfn.
3.14.2 Search critical points
Searching modes
In Multiwfn, several modes are provided to assign guessing points for searching CPs:
133
3 Functions
• (1) Search CPs from given starting points: If you already know approximate position of CPs
or you are able to reasonably guess where CPs may appear, then this mode is very suitable for you,
the Newton iteration will start from the provided coordinates. In this mode, you can directly input
coordinate of a starting point, or input two atomic indices, then their midpoint will be taken as
starting point. You can also provide a .pdb or .pqr file, the atom coordinates in it will be treated as
starting points in turn. In addition, a .txt file containing all starting points can also be provided, the
format should look like follows, each line cooresponds to X, Y, Z coordinate of a starting point in
Bohr, the format is free
1.2 3.10 -0.4
0.477 5.2 5.92
-3.26 3.3 0.5
...
• (2) Search CPs from nuclear positions: This mode uses all nuclear positions in turn as starting
points, this mode is very suitable for searching all (3,-3) for , and those (3,-3) CPs in innermost
region of atom for ELF/LOL and Laplacian of . For electron density analysis, if you located n CPs
after using this mode while there are more than n atoms in your system, it generally implies that
Newton method has missed some (3,-3) CPs.
• (3) Search CPs from midpoint of atom pairs: This mode uses midpoint of all atom pairs in
turn as starting points. This mode is very suitable for searching all (3,-1) CPs for .
• (4) Search CPs from triangle center of three atoms: Like mode 3, but use triangle center of
all combinations of three atoms in your system. Suitable for searching all (3,+1).
• (5) Search CPs from pyramid center of four atoms: Like mode 3, but use pyramid center of
all combinations of four atoms in your system. For electron density analysis, if you have tried mode
2 and mode 3 but still cannot find certain (3,+3), this mode is worth to try. This mode is more
expensive than mode 3 since more combinations will be considered.
• (6) Search CPs from a batch of points within a sphere: You need to set sphere center, radius
and the number of points first, then select option 0, the specified number (not the exact number you
have set) of points will randomly distributed in the sphere as starting points for Newton method.
The sphere center can be defined in very flexible way by option 2~6. If you choose option -1, then
the search starts and each atom center will be taken as sphere center in turn, assume that there are n
atoms and the number of points in sphere you set is m, then about n*m points will be used as initial
guesses. If you choose option -2, then you will be prompted to input indices of atoms, the nuclei of
these atoms will be used as sphere center in turn.
This mode is very appropriate for searching the CPs that are difficult to be located by other
modes. For ELF, LOL and Laplacian, it is highly recommended to use option -1 in this mode
to try to locate all possible CPs. Note that each time you carry out the search the positions of
starting points are different, if some interesting CPs were not found in previous searches, try to
launch the search again and again to locate missing CPs.
Beware that the index of the CPs located by searching mode 6 may be different in each time of
execution!
CP searching parameters
The default searching parameters are appropriate for most systems, however in some cases you
have to adjust parameters manually to ensure expected CPs could be found. These parameters can
be set in the option "-1 Set CP searching parameters".
134
3 Functions
(1) Set maximal iterations: If the number of Newton iteration exceeds this value before
converged to a CP or before Hessian matrix became singular (that is the iteration cannot continue),
then the Newton iteration will be terminated.
(2) Set scale factor of stepsize: The default value is 1.0, that is using original form of Newton
method. Sometimes reduce the stepsize is benefit for locate CPs. For example the Uracil.wfn in
"example" folder, by using mode 2 with default parameter the BCP linking N6 and H12 cannot be
found, if you reduce the stepsize to 0.5, then the problem will be solved.
(3)/(4) Set criteria for gradient norm / displacement convergence: If both gradient norm in
current position and displacement in last step is smaller than the two values, then the iteration will
stop and current position will be regarded as a critical point.
Notice that for searching mode 2, which is usually employed for locating NCPs, the convergence criterion of
gradient norm is temporarily removed, since electron density peak is quite sharp at nuclear position for very heavy
atoms, the iteration can hardly converge if gradient norm criterion is taken into account.
(5) Minimal distance between CPs: If an iteration converged to a CP, however the distance
between this CP and any CP that has been found is smaller than this value, then the CP just found
will not be considered as a new CP, and hence discarded.
(6) Skip search if distance between atoms is longer than the sum of their vdW radius multiplied
by: In CP searching mode 3, 4 and 5, if the distance of any two atoms that involved in the
combination is longer than the sum of their vdW radii multiplied this value then current search will
be skipped. The purpose of this option is to reduce the number of searches and hence computational
cost of huge system. For example the imidazole--magnesium porphyrin complex mentioned earlier,
there are 49*48/2=1176 atom pairs, hence if this cutoff strategy is not employed when you choose
searching mode 2, Multiwfn will try as many as 1176 searches. While if this strategy is used, since
only closely related atom pairs will be considered, only 274 searches are needed be performed, it
can be found that their results are identical.
(7) If print details of CP searching procedure: This option is present only in serial mode. User
can select output level of details of CP searching procedure. This option is mainly used for
debugging.
(8) Criteria for determining if Hessian matrix is singular: If the absolute value of determinant
of Hessian matrix is lower than this value, then the Hessian matrix will be regarded as singular and
terminate search. Too large of this value may lead some CPs be ommitted, while too small of this
value may cause numerical instability and occurence of CPs in the regions far from system. Default
value is appropriate for most cases.
(9) Set value range for reserving CPs: By making use of this option, during the CP searching,
only the CP with value within user-defined range will be reserved, and thus unnecessary CPs could
be ignored. By default all CPs are reserved. Via this option you can for example only search the CPs
in weak interaction region (corresponding to low electron density).
(10) Set the atoms to be considered in searching modes 2, 3, 4, 5: If you only want to search
CPs in a local molecular region, you can use this option to define the fragment, then only the atoms
in this fragment will be taken into account when using searching modes 2, 3, 4, 5.
3.14.3 Generate topology paths
Path generating methods

Once CPs have been found, you can choose to generate topology paths, which connect various
CPs. In Multiwfn, the paths are essentially represented as a bunch of points evenly distributed in a
135
3 Functions
curve. Using options 8 and 9, two kinds of paths can be generated:

(1) Generate the paths connecting (3,-3) and (3,-1) CPs: Arithmetically, firstly displacing the
coordinate of each (3,-1) forward and backward respectively along the eigenvector that corresponds
to the positive eigenvalue of Hessian, and then go uphill following the gradient vector until
encounter a (3,-3), the resulting trajectories constitute the bond paths.
(2) Generate the paths connecting (3,+1) and (3,+3) CPs: Like above, the difference is that the
starting points are (3,+1), and their coordinates are firstly moved forward and backward along the
eigenvector corresponding to the negative eigenvalue of Hessian, and then go downhill following
the gradient vector.
Path generating parameters

In the option "-2 Set path generating parameters", there are some suboptions used to adjust
parameters for generating paths.
(1) Maximum number of points of a path: In the generation of each path, if the number of steps
reached this value before encountering a CP, then the trajectory will be discarded. For generating
very long paths, default value may need to be enlarged.
(2) Stepsize: The space between neighbour points that constitutes the paths. For paths with
large curvature, sometimes they cannot be generated under default stepsize, you need to properly
decrease the stepsize and try to regenerate paths. If you set maximum number of points and stepsize
of paths as m and n respectively, then the maximum length of paths will not longer than mn.
(3) Stop generation if distance to any CP is smaller than: During the generation of paths, if the
distance between current point and a located CP is found to be smaller than this value, then the path
will be regarded as connected to the CP.
3.14.4 Generate interbasin surfaces
The interbasin surfaces (IBS) generated by Multiwfn actually consist of a bunch of paths
derived from (3,-1) CPs, these surfaces divide the whole space into respective region for each (3,-
3) CPs. By the function "10 Add or delete interbasin surfaces", you can generate, delete and check
interbasin surfaces. Notice that before generating IBS, generation of CPs should be completed first,
and at lease one (3,-1) must be found.
If you want to generate the IBS from the (3,-1) CP with index of 15, then simply input 15 (you
will find it is useful to visualize CPs by function 0 first to get the CP index). To delete this surface,
input -15 (negative sign means "delete"). If there is no IBS presented and you hope to generate all
IBSs, input 0. To delete all already generated IBS, also input 0. A list of generated IBS can be printed
by inputting the letter l If you need to export the paths of a specific IBS (e.g. corresponding to the
(3,-1) CP with index of 4) to external file, input o 4, then the coordinates of all paths derived from
the CP4 will be saved to surpath.txt in current folder. Input letter q can return to upper level menu.
Parameters for generating IBSs can be adjusted by the option "-3 Set interbasin surface
generating parameters". Enlarging number of paths in each IBS or lowering stepsize will make IBSs
looks more smooth. The length of paths in IBSs, or say the area of IBSs, is proportional to product
of stepsize and number of points in each IBS path. Notice that once the parameters are changed, all
generated IBSs will be lost.
There are three ways to portray IBSs in Multiwfn, which can be controlled by "isurfstyle" in
settings.ini. Way 1 represents the IBS directly by the paths derived from corresponding (3,-1) CP;
136
3 Functions
By Way 2 IBSs are shown by solid surfaces, this is default style; Way 3 is uncommonly used, you
can try it by yourself.
3.14.5 Visualize, analyze, modify and export results
Visualize results (option 0)

If you select the function “0 Print and visualize all generated CPs, paths and interbasin
surfaces”, a GUI will pop up, you can control if showing each type of CPs, paths, interbasin surfaces,
molecular structure, labels etc. Meanwhile, a summary of found CPs and paths are printed in the
command-line window. The satisfaction of Poincaré-Hopf relationship is also checked.
Evaluate CP properties (option 7)

You can obtain values of all real space functions supported by Multiwfn (as well as gradient
and Hessian matrix of the selected real space function) at a given CP or at all CPs by function “7
Show real space function values at specific CP or all CPs”. Note that electrostatic potential is the
most expensive one among all of the real space functions, if you are not interested in it, you can set
"ishowptESP" parameter in settings.ini to 0 to skip calculation of electrostatic potential.
In this function, you can also decompose a selected real space function at a given CP into
contributions from a range of orbitals. See corresponding example in Section 4.2.4.
Measure geometry (option -9)

Distance, angle and dihedral angle between atoms and CPs can be conveniently measured by
using the option "-9 Measure distances, angles and dihedral angles between CPs or atoms".
Manipulate CPs (option -4)

In the option “-4 Modify or export CPs”, you can print, add, delete and export CPs. In the
suboption 2, the CPs can be deleted according to various condition, including: (1) CP type (2) CP
index range (3) Distance to specific molecular fragment (thus you can remove CPs in irrelevant
regions) (4) Range of electron density (using this feature you can solely delete CPs with low density
and thus remove the CPs in weak interaction regions, or delete CPs with high density and thus
remove those in the chemical bond regions).
The positions and types of all found CPs can be saved to a formatted text file CPs.txt in current
folder by suboption 4. The information of CPs can also be loaded from an external formatted text
file by suboption 5 (the found CP at current session will be clean), notice that the file format must
be identical to the one outputted by suboption 4.
All CPs can be exported as .pdb file by suboption 6, so that CPs can be conveniently visualized
137
3 Functions
by external visualization softwares such as VMD. In this file, element C/N/O/F correspond to (3,-
3)/(3,-1)/(3,+1)/(3,+3) respectively.
Manipulate topology paths (option -5)

In the option “-5 Modify or print detail or export paths, or plot property along a path”, you can
print summary of generated paths and print coordinate of all points in a specific path via suboptions
1 and 2, respectively.
By suboption 4 you can export out the detail information of all paths to paths.txt in current
folder. By suboption 5, the paths can be imported from an external file, the file format must be
identical to the one outputted by suboption 4.
By suboption 6, all points in all paths can be exported to a pdb file in current folder, so that
paths can be conveniently visualized by external visualization softwares.
By suboption 7, value of real space functions along selected topology paths can be plotted as
curve map or exported as plain text file. You can also take the data correpsonding to "Dist." and
"Value" column as X and Y axes respectively to plot curve graph by third-part plotting softwares.
Paths can be manually deleted via suboption 3 by directly inputting their indices. Suboption 8
is also used to delete paths, it is designed for only retaining bond paths (and corresponding BCPs)
connecting two fragments but removing all other bond paths, so that one can more easily study
interfragment interactions via AIM method. The usefulness of this option will be illustrated in the
example in Section 4.2.6.
3.14.6 Calculate the aromaticity indices based on topology properties

of electron density
If the real space function you selected is electron density, options 20 and 21 will be visible,
they are utilities used to analyze aromaticity.
Shannon aromaticity index

Option 20 is used to calculate Shannon aromaticity (SA) index, see Phys. Chem. Chem. Phys.,
12, 4742 (2010) for detail. The formula can be briefly written as
N
 (rBCPi )
SA = ln( N ) −  (− pi ln pi ) pi = N
i
  (r
i
BCPi )
In above formula, N is the total number of the BCPs in the ring you want to study aromaticity, rBCP
is the position of BCP. In option 20, you need to input N and the index of these CPs in turn, then
Shannon aromaticity index will be printed immediately. The smaller the SA index, the more
aromatic is the molecule. The range of 0.003 < SA < 0.005 is chosen as the boundary of
aromaticity/antiaromaticity in original paper.
Curvature of electron density perpendicular to ring plane at RCP

In Can. J. Chem., 75, 1174 (1997), the authors showed that electron density at RCP is closely
related to aromaticity of corresponding ring. The larger the density, the stronger the aromaticity.
They also pointed out that the curvature of electron density perpendicular to the ring plane at the
138
3 Functions
RCP has more significant correlation with the ring aromaticity. The more negative the curvature, the
stronger the aromaticity. Assume that the ring is strictly perpendicular to a Cartesian axis, for
example, a ring perpendicular to Z axis (viz. the ring is in XY plane), then the curvature is just ZZ
component of Hessian matrix of electron density at the RCP, which can be directly obtained by using
option 7 to print out properties of a given RCP. However, if the ring is distorted, or is not exactly
perpendicular to any Cartesian axis, you should use option 21 to calculate the curvature. In option
21, you need to input the index of the RCP (or directly input coordinate of a point), and then input
indices of three atoms, their nuclei position will be used to define the ring plane. Then, the electron
density, the gradient and the curvature of electron density perpendicular to the ring plane at the RCP
will be outputted on screen. In addition, unit normal vector, the coordinates of the two points above
and below 1 Å of the RCP in normal direction will be outputted together, which can be taken as the
point used to calculate NICS(1).
Examples of various kinds of topology analysis can be found in Section 4.2.
3.15 Quantitative analysis of molecular surface (12)
Quantitative analysis of molecular surface is a powerful tool, it has a lot of practical

applications, such as predicting reactive sites, predicting molecular properties, interpreting
intermolecular weak interaction. The theory and numerical algorithm involved in present module
have been detailedly described in my paper J. Mol. Graph. Model., 38, 314 (2012). In the next two
sections, these two aspects will only be briefly introduced.
3.15.1 Theory
In Multiwfn, in principle, distribution of any real space function on molecular surface (or the
surface defined by isosurface of a certain function) can be quantitatively studied. Electrostatic
potential and average local ionization energy on molecular vdW surface are particularly useful,
therefore they will be discussed detailed in this section. Same kind of quantitative analyses could
also be applied to other real space functions, such as user-defined functions, electron delocalization
range function (EDR) and even Fukui function and dual descriptor.
(1) Electrostatic potential on vdW surface

Molecular electrostatic potential (ESP), V(r), has been widely used for prediction of
nucleophilic and electrophilic sites, as well as molecular recognition mode for a long time, the
theoretical basis is that molecules always tend to approach each other in a complementary manner
of ESP. These analyses of ESP are common performed on molecular van der Waals (vdW) surface.
Although the definition of such a surface is arbitrary, most people prone to take the 0.001 a.u.
isosurface of electron density as vdW surface, since this definition reflects specific electron structure
features of a molecule, such as lone pairs and π electrons, this is also what the definition used in our
analyses.
The analysis of ESP on vdW surface has been further quantified to extract more information.
139
3 Functions
It is shown that the strength and orientation of weak interaction, including such as hydrogen bonding,
dihydrogen bonding and halogen bonding, can be well predicted and explained by analyzing the
magnitude and positions of minima and maxima on the surface. Politzer and coworkers (J. Mol.
Struct. (THEOCHEM), 307, 55 (1994)) have defined a set of molecular descriptors based on ESP
on vdW surface, which are taken as independent variables of general interaction properties function
(GIPF). GIPF successfully connects distribution of ESP on vdW surface and many condensed phase
properties, including density, boiling point, surface tension, heats of vaporization and sublimation,
LogP, impact sensitivity, diffusion constant, viscosity, solubility, solvation energy and so on. Below
I enumerate and brief these descriptors.
A+ and A− indicate the surface area in which the ESP has positive and negative value,
respectively. Total surface area A is the sum of them.
+ −
V S and V S denote average of positive and negative ESP on vdW surface respectively
m n
+ −
V S = (1 / m)V (ri ) V S = (1 / n)V (r j )
i =1 j =1
where i and j are indices of sampling points in positive and negative regions respectively. The S
subscript means "Surface". The average of ESP over the entire surface is
t
V S = (1 / t )V (rk )
k =1
 is the average deviation over the surface, which is viewed as an indicator of internal charge
separation:
t
 = (1 / t ) V ( rk ) − V S
k =1
The total ESP variance can be written as the sum of positive and negative parts:
m n
+ −
 tot
2
=  +2 +  −2 = (1 / m) [V (ri ) − V S ]2 + (1 / n) [V (r j ) − V S ]2
i =1 j =1
The variance reflects the variability of ESP. The larger the 𝜎+2 and 𝜎−2 , the more tendency that the
molecule interacts with other molecules by positive and negative ESP regions respectively.
Degree of charge balance is defined as
 +2 −2
=
( tot
2 2
)
When 𝜎+2 equals to 𝜎−2 ,  attains its maximum value of 0.250. The closer the  is to 0.250, the
more possible that the molecule can interact to others through positive and negative region with
similar extent.
2
The product of 𝜎tot and  is a very useful quantity too, a large value of 𝜎tot
2
is an indicative
of a molecule that has relatively strong tendencies to interact with others of its own kind
electrostatically.
In order to quantify molecular polarity, I defined a quantity named "molecular polarity index"
(MPI), which is closely related to the  index.
140
3 Functions
t
MPI = (1 / t ) V (rk )  (1 / A)  | V (r) | d S
k =1 S
where S denotes molecular surface. My test for some representative molecules showed that MPI is
a quite reliable index of measuring molecular polarity, the larger the index, the higher the polarity.
In Multiwfn, above mentioned surface descriptors can be calculated not only on the whole vdW
surface, but also on the subregions corresponding to atoms or user-defined fragments. The detail of
the theory is to be published, and thus not documented here at the moment. In addition, these surface
descriptors can be calculated for other kinds of real space functions.
(2) Some practical applications of GIPF descriptors

·Predicting heat of vaporization and heat of sublimation
A practical application of above GIPF descriptors is presented in J. Phys. Chem. A, 110, 1005
(2006). The authors showed that for a series of molecules containing C, H, N and O elements, the
heat of vaporization can be well evaluated as
H vap = a A + b  tot
2
+c
with least-squares fit coefficients a = 2.130, b = 0.930 and c = -17.844. The heat of sublimation can
be predicted as
H sub = aA2 + b  tot

2
+c
with a = 0.000267, b = 1.650087 and c = 2.966078. In above equations the surface area A is in Å2,
2
Hsub is in kcal/mol, 𝜎tot is in (kcal/mol)2. Note that the coefficients are more or less dependent on
the calculation level used. The author used B3LYP/6-31G* to optimize geometry and B3LYP/6-
311++G(2df,2p) to calculate ESP.
In Int. J. Quantum Chem., 105, 341 (2005), Politzer et al. proposed another equations for
predicting Hvap and Hsub in standard state, their equations can be used for molecules containing
C, H, O, N, F, Cl, S:
H vap = 1.3556 A + 1.1760  tot
2
− 10.4331
H sub = 4.4307  10−4 A2 + 2.0599  tot
2
− 2.4825
2
The Hvap and Hsub are in kcal/mol, A is in Å2 and 𝜎tot is in (kcal/mol)2. B3PW91/6-31G** was
employed in their study. The mean absolute error of Hvap and Hsub were found to be 2.0 kcal/mol
and 2.8 kcal/mol, respectively.
·Predicting density of molecular crystal
Another typical application of statistical data of ESP on vdW surface is predicting crystal
density of organic molecules containing C, H, N and O elements. The crystal density is a crucial
property of energetic compounds. In J. Phys. Chem. A, 111, 10874 (2007), it was shown that the
density can be estimated by  = M / Vm, where M is molecular mass and Vm is molecular vdW
volume defined by  = 0.001 a.u. isosurface; for ionic crystal (e.g. ammonium azide), M and Vm
correspond the sum of mass and volume of the cation and anion comprising a formula unit of the
compound. Although the relationship is quite simple, it indeed works well for most neutral species,
but the error is evidently larger for ionic species. In order to improve the prediction accuracy for
neutral ones, in Mol. Phys., 107, 2095 (2009), the authors introduced GIPF descriptors into the
formula:
141
3 Functions
M
 = +  ( tot
2
)+
Vm
At B3PW91/6-31G** level, the fitted coefficients are  = 0.9183,  = 0.0028 and  = 0.0443. This
formula has proven to have improved accuracy, because intermolecular electrostatic interaction is
somewhat effectively taken into considered. In a succeeding paper Mol. Phys., 108, 1391 (2010),
the author showed that the crystal density of ionic compounds can be estimated much better than 
= M / Vm if GIPF descriptors are introduced:
M  V +S (cation)   V −S (anion) 
 = +  + + +
Vm  A(cation)   A(anion)
−

   
with least-squares fit coefficients  = 1.0260,  = 0.0514,  = 0.0419 and  = 0.0227 at B3PW91/6-
+ + + +
31G** level. In the equation, V S (cation) and A(cation) denote the V S and A of the cation;
− − − −
V S (anion) and A(anion) denote the V S and A of the anion. For 30 test cases, the average
absolute error is merely 0.033 g/cm3.
Noticed that above relationships are only appropriate for small organic compounds containing
C, H, N, O elements, the error are significantly larger for other kinds of systems.
·Predicting boiling point
In J. Phys. Chem., 97, 9369 (1993), it is shown that boiling point can be predicted as
Tbp = A +   tot
2
+
where  = 2.736,  = 33.31,  = -72.05 were fitted at HF/STO-5G*//HF/STO-3G* level. This paper
also showed equations for predicting critical temperature, volume and pressure.
·Predicting solvation free energy
In J. Phys. Chem. A, 103, 1853 (1999), the prediction equation for solvation free energy is
presented as (Vmin denotes the ESP value at its global minimum in the whole space):
Gsolv (kJ/mol) = 0.17201Vmin − 2.6412  10 −5 (VS,max − VS,min ) 3
− −
+ 0.051892 A− V S + 9704.2 /( A− V S ) + 46.827
·Predicting pKb
In J. Chem. Inf. Model. (2020) DOI: 10.1021/acs.jcim.9b01173, the authors showed that pKb
of amino groups can be nicely estimated by fitting equation, for example, primary amine cases:
p Kb = 0.4952 VS,min + 24.5880
where VS,min is in kcal/mol and should be calculated at B97XD/cc-pVDZ level. Prediction accuracy
of this equation is fairly satisfactory, the R2 is as high as 0.9519 with mean absolute error of merely
0.12. Similar fitted equations are also available for secondary amines, tertiary amines and tertiary
amines.
·Predicting other properties
In addition, the equations used to predict heat of fusion, surface tension and crystal/liquid
density can be found in J. Phys. Chem., 99, 12081 (1995), the equations used to predict lattice energy
for ionic crystal containing NH4+, K+ and Na+ are given in J. Phys. Chem. A, 102, 1018 (1998). More
142
3 Functions
formulae used to predict physical properties of organic molecules based on GIPF descriptors are
summarized in Table 3 of J. Mol. Struct. (THEOCHEM), 425, 107 (1998). GIPF also has many
important uses in study of biochemical systems, see Int. J. Quantum Chem., 85, 676 (2001) for a
review.
(3) Other mapped functions: Average local ionization energy and so on

Average local ionization energy, 𝐼 ,̅ has attracted more and more attentions, see corresponding
part of Section 2.6 for brief introduction. This function has many uses, for example, reproducing
atomic shell structure, measuring electronegativity, quantifying local polarizability and hardness.
However, the most important use maybe the prediction of reactivity according to function value on
vdW surface, 𝐼𝑆̅ . Lower value of 𝐼𝑆̅ indicates that the electron at r are more weakly bounded,
therefore more likely r is the site of electrophilic or radical attack. Many studies have shown that
the global minimum of 𝐼 ̅ on vdW surface exactly locates the experimental reaction site, while
relative magnitude of 𝐼 ̅ at corresponding reaction site in homologues correlates well with relative
reactivity. The interested users are recommended to take a look at J. Mol. Model, 16, 1731 (2010)
and Chapter 8 of the book Theoretical Aspects of Chemical Reactivity (2007).
Local electron affinity EAL is a quantity very similar to 𝐼 ,̅ the only difference is that the MOs
under consideration is not all occupied ones, but all unoccupied ones. It was shown that EAL on
molecular surface is useful for analyzing nuclophilic attacking, for detail see J. Mol. Model., 9, 342
(2003).
Quantitative analysis on molecular surface for Fukui function and orbital overlap distance
function D(r) have also been proven to be fairly useful, see Sections 4.12.4 and 4.12.8 for example,
respectively.
3.15.2 Numerical algorithm
3.15.2.1 Analysis on the whole molecular surface

In summary, in the common task of quantitative analysis of molecular surface, what we need
to obtain are minima and maxima of a selected real space function (i.e. mapped function) on vdW
+
surface (or isosurface of a specific real space function), as well as quantitative indices such as V S ,
−
V S , V S ,  ,  +2 ,  −2 and so on. Here I briefly describe how these properties are computed in
Multiwfn, the basic steps are given below.
1. Grid data of electron density enclosed the entire molecular space is compute. The smaller
the grid spacing is used, the more accurate result you will get, however the more vertices will be
generated in next step and therefore you will wait longer time in step 3.
2. Marching Tetrahedra algorithm is performed by making use of the grid data generated above,
this step generally does not cost much computational time. The volume enclosed by the isosurface
is computed at the same time. This step generates vertices representing the isosurface, along with
their connectivity. Each neighbouring three vertices constitute a triangle (will be referred to as facet
below). Below example is a water molecule, vertices (red points) and connectivity (black lines) are
portrayed:
143
3 Functions
3. Since computing ESP is time consuming, in order too cut down overall computational time,
Multiwfn eliminates redundant points automatically. Specifically, if the distance between two points
is smaller than a specific value, one of the points will be eliminated, and the other point will be
moved to their average position. In above graph, the aggregated points such as those inside blue
circles will be finally merged to one point.
4. Calculate mapped function (ESP, 𝐼 ̅ and so on) at each vertex on the isosurface. For ESP,
this is the most time consuming step; however for such as 𝐼 ̅ and EAL, this step can be finished
immediately.
5. Locate and then output minima and maxima of the mapped function on the surface by
making use of connectivity. If the mapped function value at a vertex is both lower (larger) than that
at its first-shell neighbours and second-shell neighbours, then this vertex will be regarded as surface
minimum (maximum).
+ −
6. Compute and output V S , V S , V S ,  ,  +2 ,  −2 ,  tot
2
, ,  tot , P, as well as vdW
2
volume, area of total vdW surface, the area where the mapped function is positive and where is
negative. As an example, V S is computed as
N
V S = (1 / Atot ) Ai Fi
i =1
where N is the total number of facets, Atot is the sum of area of all facets, Ai is the area of facet i, Fi
is the ESP value (or value of other kind of mapped function) of facet i, which is calculated as the
average of ESP at the three vertices composing the facet.
3.15.2.2 Analysis on local molecular surface

In order to unveil more chemically useful information from distribution of mapped functionon
vdW surface, Multiwfn supports three kinds of analyses for local vdW surface, as described below.
144
3 Functions
(1) Analysis of local molecular surface corresponding to various atoms

In this mode the whole molecular surface will be first decomposed to local surfaces
corresponding to individual atoms, and then all surface properties are calculated for these atomic
surfaces. This function is very helpful to study atomic properties. See Section 4.12.3 for example.
Note that atomic local regions on molecular surface cannot be uniquely defined. The rule
employed by Multiwfn is shown as follows, which is simple and has clear physical meaning. For
any given point r on the molecular surface, Multiwfn calculates weight w for all atoms via
| r − rA |
wA = 1 −
RA
where A denotes atomic index, rA and RA are coordinate and radius of atom A, respectively. The
surface point r is attributed to the atom that has the maximal weight. It can be seen that at the
position of nucleus of A, wA reaches its maximal value 1.0. wA linearly attenuates with increase of
the distance |r-rA|, it attains zero when the distance just equals to corresponding atomic radius. The
larger radius the atom has, the slower the weight attenuates, therefore the above definition of weight
makes larger sized atoms have higher possibility to cover broader range on molecular surface.
In Multiwfn it is also possible to perform analysis of local molecular surface corresponding to
a user-defined fragment, the region of the fragment is simply the sum of the regions of all the atoms
that constitute it.
(2) Analysis of local molecular surface around specific surface extreme

This kind of analysis is mainly used for measuring area and graphically revealing region of -
hole and -hole, see Section 4.12.10 for practical example, but it can also be used for other analysis
purposes if the mapped function is not chosen to be electrostatic potential.
In this mode, you need to select a surface extreme and set a criterion. When you select a surface
maximum (minimum), if a surface vertex is directly or indirectly connected to the surface extreme
and value of mapped function at this vertex is higher (lower) than the criterion, the vertex will be
selected, all the selected vertices collectively define the local molecular surface. In order to make
you intuitively understand how this mode defines the local surface around a selected surface extreme,
an graphical illustration is provided below, in this map one-dimension ESP distribution is used to
abstractly represent two-dimensions distribution of ESP on molecular surface.
(3) Analysis of local molecular surface based on Basin-like partition
145
3 Functions
Basin partition generally refers to partitioning the whole molecular space into individual local
spaces by employing zero-flux surfaces of gradient of a real space function as basin boundaries,
each basin contains a maximum, see Section 3.20.1 for detailed introduction. The same idea may
also be employed for partitioning the molecular surface. When this technique is adapted in
combination with ESP, some chemically useful information could be gained, see Section 4.12.11 for
practical example.
The algorithm of applying basin partition to molecular surface is proposed by me (to be
published). The main purpose is to determine the attribution relationship between surface vertices
and surface extrema. Because the mapped function may have both positive and negative values,
absolute value of the mapped function is firstly taken, then all surface vertices are considered in
turn. Iteration is performed for each surface vertex; in each step, the vertex temporarily moves
towards the neighbouring vertex having largest value, namely climbing up hill. The iteration
continues until the vertex reached a surface extreme, which is just the extreme that the surface vertex
finally attributed to. Note that after usual quantitative molecular surface analysis, the connection
relationship between various surface vertices is already known, if vertex B is directly connected to
A, then B will be the neighbouring vertex of A. To make you better understand central idea of this
method, a figure is shown below
In this graph, the black curve represents the original ESP distribution, the cyan and black curves
collectively correspond to the absolute value of the ESP. The region enclosed by blue bracket
corresponds to the local surface corresponding to minimum 1, while the two pink brackets
respectively correspond to the two local surfaces corresponding to maxima 1 and 2.
For the above mentioned modes (2) and (3), if all the three vertices of a surface facet are
selected, this surface facet will be regarded as attributed to the selected local surface. Then the area
as well as average of mapped function of the selected local surface can be straightfowardly evaluated.
3.15.3 Parameters and options
You will see below options in the main interface of quantitative analysis of molecular surface.
0 Start analysis now!: When this option is selected, the analysis boots up. All steps described
in last section will be implemented sequentially.
6 Start analysis without considering mapped function: This option also starts analysis like
option 0, however calculation and analysis of mapped function are skipped. This option is useful if
146
3 Functions
you are only interested in e.g. volume and area of molecular surface while value of mapped function
is not of interest.
1 The isovalue of electron density used to define molecular surface: Default value is 0.001,
corresponding to the most frequently used definition of vdW surface. In general it is not
recommended to adjust this value.
2 Select mapped function: The mapped function to be studied can be selected by this option.
Such as ESP, 𝐼 ,̅ EAL and user-defined function (Section 2.7) are directly supported and can be
automatically computed by internal code of Multiwfn during the surface analysis. Alternatively, if
you intend to load value of mapped functions at all surface vertices from an external file, you should
choose "0 Function from external file", in this case you may analyze broader type of mapped
functions (e.g. Fukui function), see description of option 5 given below for more detail.
Note that by default the ESP is evaluated based on wavefunction, this process may be quite time-consuming for
large systems. However, if you choose "Electrostatic potential from atomic charge", then Multiwfn will evaluate
ESP based on atomic charges, which are loaded from a .chg file, the computational time will be reduced by several
orders. You can use main function 7 to calculate atomic charge and produce the .chg file, or you can write the .chg
file manually, see corresponding part of Section 2.5 on the .chg file format. Note that the analysis result will be
reasonable only when the method used to generate atomic charges can reproduce ESP well (e.g. CHELPG, MK and
ADCH methods). Also note that in some cases ESP generated by atomic charges differ significantly from the ESP
generated based on wavefunction, see J. Chem. Theory Comput., 10, 4488 (2014) for comprehensive discussion.
3 Spacing of grid points for generating molecular surface: This setting defines the spacing
of electron density grid data, see step 1 introduced in the last section. The spacing directly
determines the accuracy and computational cost of the analysis. Default value is suitable for general
cases. Largening this value can reduce computational time evidently, however if this value is not
small enough, the vertices on the isosurface will be spare, this may cause missing or erroneous
locating of some extrema. In general, the results under the default spacing are accurate and reliable.
If you find some extrema were not located under default spacing, try to decrease spacing and rerun
the task.
4 Advanced options: Suboptions in this option is not needed to be frequently adjusted by

normal users.
(1) The ratio of vdW radius used to extend spatial region of grid data: The role of this
parameter is exactly identical to the parameter k introduced in Section 3.100.3. Enlarging this value
will lead larger spatial extension of grid data of electron density around molecule. If isovalue of
electron density is set to a lower value than default, or the system is negatively charged, you may
need to enlarge this parameter to ensure that the isosurface will not be truncated.
(2) Toggle if eliminating redundant vertices: If this option is switched to "No", then the
elimination of redundant vertices (step 3 described in last section) will be skipped, and you will
waste vast time to calculate mapped function at those meaningless vertices. If this option is switched
to "Yes", you will be prompted to input a distance criterion for merging adjacent vertices. Commonly,
0.4~0.5 times of grid spacing is recommended to be used as the criterion.
(3) Number of bisections before linear interpolation: Simply speaking, the larger the value,
the more exactly the isosurface (corresponding to vdW surface) can be generated. Enlarging this
value will bring additional cost in step 2. The generated isosurface under default value is exact
enough in general. You can descrease it to 2 even to 1 to save computational time, however decrease
it to 0 will frequently lead to false surface extrema.
5 Loading mapped function values from external file
147
3 Functions
By choosing proper suboption in this option, during surface analysis, the value of mapped
function at surface vertices can be loaded from external file rather than directly calculated by
Multiwfn. This option has two uses: (1) Reduce overall analysis cost (2) Analyze a special function
that cannot be calculated by Multiwfn, or the analyze a a function requiring mathematical operations
and thus not directly available in present module (e.g. dual descriptor).
This option has four suboptions:
(0) Do not load mapped function but directly calculate by Multiwfn: This is default case.
(1) Load mapped function at all surface vertices from plain text file: If this option is
selected, then after generation of molecular surface, coordinate of all surface vertices (in Bohr) will
be automatically exported to surfptpos.txt in current folder. Then you can use your favourite program
to calculate mapped function value at these points, and write the values as the fourth column in this
file (in free format, unit is in a.u.). For example
1324 // The first line is the total number of points
-1.6652369 -0.5480503 -0.2554867 -0.0196978306
-1.6835983 -0.5563165 -0.1342924 -0.0242275610
-1.6977125 -0.5530614 -0.0099311 -0.0287667191
-1.7013207 -0.5536310 0.1194197 -0.0330361826
-1.6954099 -0.5523031 0.2547866 -0.0371580132
....
Then input the path of this file (the name can still be surfptpos.txt), Multiwfn will directly read the
values. An exemplificative application of this option is given in Section 4.12.4.
Hint: If you will analyze a system twice or more times, and want to avoid calculating mapped
function values every time for saving time, at the first time you analyze the system, you can select
option 7 at post-processing interface to export coordinates and corresponding mapped function
values of surface vertices to vtx.txt in current folder. At next time you analyze the system, if you
choose this option, and input the path of the vtx.txt during the surface analysis, then mapped function
values will be directly loaded rather than re-calculated (See Section 4.12.1 for example).
(2) Similar to 1, but specific for the case of using cubegen utility of Gaussian: A file named
cubegenpt.txt will be generated in current folder after generation of molecular surface. This file is
very similar to surfptpos.txt, the difference is that in this file the first line is not presented, and
coordinate unit is in Å. Based on this file, you can make use of cubegen utility of Gaussian to
calculate mapped function at all surface vertices. After that, input the path of cubegen output file,
the data will be loaded by Multiwfn.
(3) Interpolate mapped function from an external cube file: After generation of molecular
surface, a template cube file named template.cub will be generated in current folder. Then you will
be prompted to input the path of a cube file representating the mapped function you are interested
in, the grid setting of this cube file must be exactly identical to template.cub. The mapped function
values at surface vertices will be evaluated by interpolation from the cube file you provided.
3.15.4 Options at post-processing stage
Once the all calculations of surface analysis are finished, a summary will be printed on screen.
Meanwhile below options will appear on screen used to check, adjust and export results.
-3 Visualize the surface: By this option you can directly visualize the isosurface analyzed.
148
3 Functions
-2 Export the grid data to surf.cub in current folder: The grid data used to generated
isosurface will be exported to the cube file surf.cub in current folder.
-1 Return to upper level menu
0 View molecular structure, surface minima and maxima: A GUI window will pop up if
this option is chosen. Red and blue spheres represent the position of maxima and minima. All
widgets are self explanatory and hence not be referred here.
1 Export surface extrema as plain text file: This option exports value of mapped function
and X,Y,Z coordinates of surface extrema to surfanalysis.txt in current folder.
2 Export surface extrema as pdb file: This option outputs surface extrema to surfanalysis.pdb
in current folder. The B-factor column records mapped function value (the actually used unit is
shown on screen).
3 Discard surface minima in certain value range: If the mapped function value at a surface
minima is between the lower and upper limit inputted by user, then this minimum will be discarded
and cannot be recovered. This option is useful to screen the minima with too large value.
4 Discard surface maxima in certain value range: If the mapped function value at a surface
maximum is between the lower and upper limit inputted by user, then this maximum will be
discarded and cannot be recovered. This option is useful to screen the maxima with too small value.
5 Export present molecule as pdb format file: This option outputs structure of present system
to a specified pdb file. Since pdb is a widely supported format, in conjunction with the output by
option 2, surface extrema can be conveniently analyzed in external visualization softwares such as
VMD.
6 Export all surface vertices to vtx.pdb in current folder: This option outputs surface
vertices to vtx.pdb file in current folder, mapped function values are written to B-factor field (the
149
3 Functions
actually used unit is shown on screen). This option is mainly used to check validity of isosurface
polygonization and visualize distribution of mapped function on molecular surface.
There is a hidden option 66, which not only outputs surface vertices to vtx.pdb, but also outputs connectivity
into CONECT field of this file. If you would like to visualize connectivity based on the vtx.pdb in VMD program,
please consult my blog article "Setting connectivity of atoms according to CONECT field in VMD"
(http://sobereva.com/121, in Chinese)
7 Export all surface vertices to vtx.txt in current folder: Namely outputting all reserved
surface vertices to a plain text file named vtx.txt in current folder, including vertex X/Y/Z
coordinates in Bohr and mapped function values in various units.
8 Export all surface vertices and surface extrema as vtx.pqr and extrema.pqr: This option
exports all surface vertices and surface extrema as vtx.pqr and extrema.pqr in current folder,
respectively. The column originally designed for recording atomic charge in this format (i.e. the
third last column) is used to record mapped function value in a.u. Because recording precision in
this case is relatively high, if the value of mapped function is too small to be properly recorded as
B-factor field of .pdb file (e.g. Fukui function on vdW surface), obviously you should use this option
to replace options 2 and 6.
9 Output surface area in specific value range of mapped function
By this option, one can gain the knowledge of the distribution of molecular surface area in
various range of mapped function. First one needs to input the index range of the atoms in
consideration, then input overall range, interval and the unit. For example, one sequentially inputs
2,6-9, -45,50 and 10 and 3, the statistic then is applied on the local molecular surface corresponding
to atoms 2, 6, 7, 8 and 9, the output looks like below:
Begin End Center Area %
-50.0000 -40.0000 -45.0000 4.8764 1.5171
-40.0000 -30.0000 -35.0000 28.0413 8.7242
-30.0000 -20.0000 -25.0000 23.2699 7.2397
-20.0000 -10.0000 -15.0000 17.6022 5.4764
-10.0000 0.0000 -5.0000 61.4759 19.1263
0.0000 10.0000 5.0000 72.6197 22.5933
10.0000 20.0000 15.0000 55.2707 17.1957
20.0000 30.0000 25.0000 53.9060 16.7712
30.0000 40.0000 35.0000 2.8590 0.8895
40.0000 50.0000 45.0000 1.5000 0.4667
Sum: 321.4212 100.0000
where "begin" and "end" is the lower and upper limit of local value range, respectively. "Center" is
their average value. Area is in Å2, "%" denotes the proportion of the area in overall molecular surface
area.
10 Output the closest and farthest distance between the surface and a point
In this option, after defining a point (you can define a nuclear position or geometry center as
the point, you can also directly input the coordinate of the point), the closest and farthest distance
between the molecular surface and the point will be outputted. These two quantities have two main
uses:
(1) In atoms in molecules (AIM) theory, for systems in gas phase, the vdW isosurface is defined
as the ρ=0.001 a.u. isosurface. The closest distance between a nucleus and the surface can be
regarded as non-bonded atomic radius. For a noncovalently interacting atom pair AB, The difference
between the length of A-B and the sum of their non-bonded radii is termed as mutual penetration
150
3 Functions
distance. In general, the larger the distance is, the stronger the interaction will be.
(2) The farthest distance between molecular surface and geometry center can be viewed as a
definition of molecular radius. Of course, the concept of molecular radius is only meaningful for
sphere-like molecules.
If you input f, Multiwfn will output the farthest distance between all surface points. This can
be regarded as a definition of molecular diameter.
11 Output surface properties of each atom
This option is used to realize the analysis of local molecular surface corresponding to various
atoms, as introduced in Section 3.15.2.2. After outputting the surface properties, the user can select
if outputting the surface facets to locsurf.pdb in current folder. If choosing "y", then in the outputted
pdb file, each atom corresponds to a surface facet, and the B-factor field records the attribution of
the surface facets, e.g. a facet having B-factor of 11.00 means the facet belongs to the local surface
of atom 11. If you load the pdb file into VMD and set "Coloring Method" as "Beta", then you can
directly visualize how the whole molecular surface is decomposed to atomic surfaces.
12 Output surface properties of specific fragment
Similar to function 11, but user can define a fragment, the surface properties will only be
calculated on the local surface corresponding to this fragment, so that one can study fragment
properties according to the local surface descriptors. Also, you can choose to output the surface
facets to locsurf.pdb in current folder, in which the atom having B-factor of 1 and 0 means
corresponding surface facet belongs to and does not belong to the local surface of the fragment you
defined, respectively.
13 Calculate grid data of mapped function and export it to mapfunc.cub
For example, if the mapped function you selected before the quantitative surface analysis is
ALIE, then if you select this option in post-processing menu, grid data of ALIE will be calculated
and exported to mapfunc.cub in current folder, the grid setting is the same as the one employed in
the quantitative surface analysis. Based on the surf.cub exported by option -2 and the mapfunc.cub,
you can plot color-mapped isosurface map via VMD program. Section 4.12.6 illustrates value of
this option.
14 Calculate area of the region around a specific surface extreme
This option is used to realize the analysis "(2) Analysis of local molecular surface around
specific surface extreme" described in Section 3.15.2.2. The area as well as averaged mapped
function value of the local surface region will be printed on screen, and meantime a file selsurf.pqr
will be exported to current folder, you can load it into VMD program to visualize the selected local
surface region (drawing as "Points" method and coloring according to "Charge" property are
recommended, the charge column in this file correspond to the mapped function values in a.u.).
15 Basin-like partition of surface and calculate areas
This option is used to realize the analysis "(3) Analysis of local molecular surface based on
Basin-like partition" described in Section 3.15.2.2. The number of surface vertices of each surface
basin, the area of each surface basin, as well as average value of mapped function on each surface
basin are outputted. At the meantime, surfbasin.pdb is exported to current folder and it contains all
surface vertices, B-factor corresponds to index of the extreme that the vertex attributed to.
19 Discard some surface extrema by inputting indices
Sometimes due to numerical noise or other reasons, there are some unwanted surface extrema.
In this case you can use this option to conveniently remove them by inputting their indices.
151
3 Functions
Abundant examples of various kinds of quantitative molecular surface analyses can be found
in Section 4.12.
3.15.5 Special topic: Hirshfeld and Becke surface analyses
The quantitative molecular surface analysis module is also able to carry out Hirshfeld and
Becke surface analysis, present section is devoted to introduce this point.
Theory of Hirshfeld and Becke surface analyses

Hirshfeld surface analysis was first proposed in Chem. Phys. Lett., 267, 215 (1997) and
comprehensively reviewed in CrystEngComm, 11, 19 (2009). This method focus on analyzing the
so-called Hirshfeld surface to reveal weak interactions between monomers in complex or in
molecular crystal.
Hirshfeld surface in fact is a kind of inter-fragment (or inter-monomer) surface, which is
defined based on the concept of Hirshfeld weight. Probably Hirshfeld surface is the most reasonable
way to define inter-fragment surface.
Atomic Hirshfeld weighting function of an atom is expressed as
 A0 (r )
w AHirsh (r ) =
  B0 (r)
B
where  A0 denotes the density of atom A in free-state. Summing up weight of all atoms in a
fragment yields Hirshfeld weight of this fragment
wPHirsh (r ) =  wAHirsh (r )
AP
Hirshfeld surface of fragment P is just the isosurface of wP

Hirsh
= 0.5 .
Motivated by Hirshfeld surface, I proposed Becke surface, which replaces Hirshfeld weight
with Becke weight (see Section 3.18.0 for introduction of Becke weight), only geometry and atomic
radii are required to construct Becke surface. Generally the shape of Becke surface and Hirshfeld
surface are comparable, but I prefer to use Becke surface, mostly because Becke surface does not
rely on atomic density and thus somewhat easier to be constructed.
To intuitively illustrate Hirshfeld/Becke surface, acetic acid dimer in 2D case is taken as
example here
152
3 Functions
Becke weighting function of the monomer at left side of the graph is represented by color, going
from red to purple corresponds to the weight varying from 1.0 to 0.0. The black line, which is the
contour line of 0.5, is just its Becke surface. Evidently, Becke surface very elegantly partitioned the
whole space into two monomer regions, the difference of atomic size is properly and automatically
taken into account. The Becke surface in this case is an open surface, the surface extends to infinity;
while if the monomer is completely buried, such as in molecular crystal or metal-organic framework
environment, then its Becke surface will be a close surface and encloses all of its nuclei, just like
common molecular surface.
If we map specific real space functions on Becke/Hirshfeld surface and study their distributions,
just like quantitative analysis on molecular surface, we can gain many important information about
intermolecular interaction. There are three real space functions very useful for this purpose
d i − ri vdW d e − revdW
(1) Normalized contact distance d norm = + , where di (de) is the
ri vdW revdW
distance from a point on the surface to the nearest nucleus inside (outside) the surface, 𝑟𝑖vdW and
𝑟𝑒vdW denote vdW radius of the corresponding two atoms. Small value of dnorm indicates close
intermolecular contact and implies evident interaction.
(2) Electron density. If electron density is large in some local region of Becke/Hirshfeld surface,
obviously the intermolecular interactions crossing these regions must be strong. The usefulness of
electron density is similar to dnorm, but the former is more physically meaningful. While dnorm also
has its own advantage, namely only geometry and vdW radii information are needed.
(3) sign(λ2), see corresponding part Section 2.6 for detailed explanation. This function can
not only exhibit interaction strength but also reveal interaction type.
Below is urea crystal, the isosurface represents Hirshfeld surface of the central urea, and dnorm
is the mapped function. Red parts correspond to small dnorm and thus exhibit close contact, which
mainly originates from H-bond interaction.
153
3 Functions
Fingerprint plot
The so-called "fingerprint plot" defined in the framework of Hirshfeld surface analysis is very
useful for investigating the noncovalent interactions in molecular crystals. X and Y axes in this plot
correspond to di and de, respectively. The plot is created by binning (di, de) pairs in certain intervals
and colouring each bin of the resulting 2D histogram as a function of the number of surface points
in that bin, ranging from purple (few points) through green to red (many points). Black region means
there are not surface points. If you are confused, see the fingerprint plot in Section 4.12.6 for
example. The usefulness of fingerprint plot is demonstrated in page 24 and 25 of CrystEngComm,
11, 19 (2009). Although in Multiwfn fingerprint plot can also be drawn in Becke surface analysis,
the plot in this case is not as useful as that in Hirshfeld surface analysis.
The Hirshfeld surface in fact can be viewed as the contact surface between the atoms in the
Hirshfeld fragment you defined and all of the other atoms. The remarkable flexibility of Multiwfn
allows the overall contact surface to be decomposed to various local contact surfaces and draw the
corresponding fingerprint plots. For example, in Multiwfn, one can draw fingerprint plot for the
local contact surface between the nitrogen in the central urea and the hydrogens in specific
peripheral ureas. In this case, all points in the contact surface simultaneously satisfy two conditions:
(1) In the Hirshfeld fragment (viz. the central urea), the atom closest to the surface point is the
nitrogen (2) Among all atoms that do not belong to the Hirshfeld fragment, the atom closest to the
surface point is one of hydrogens in specific peripheral ureas. Undoubtedly, fingerprint plot of local
contact surface greatly facilitates one to study the noncovalent interaction in local region due to the
contact between specific atom sets.
Usage
The procedure to perform Becke/Hirshfeld surface analysis is similar to usual quantitative
molecular surface analysis. After you entered main function 12, choose option 1 and select Becke
or Hirshfeld surface, then input the index of the atoms in the fragment (referred to as Hirshfeld
fragment below), and use option 2 to select mapped function. After that choose option 0 to start
calculation. Quantitative data on the surface such as average value and standard deviation will be
outputted, and surface extrema will be located. Then via corresponding options you can visualize
surface minima/maxima, export result and so on, all options in the post-processing menu (except
154
3 Functions
one, see below) have already been introduced in Section 3.15.4 and thus are not described again
here. If you use option 6 to export the surface vertices to .pdb file, based on which you can also plot
color-mapped Hirshfeld/Becke surface by means of VMD program, the drawing method should be
set to "Points" and coloring method should be set to "Beta" (viz. B-factor in the pdb file, which
corresponds to mapped function value in the present context).
Becke surface analysis and Hirshfeld surface analysis commonly work equally well, the
computational cost of the latter is lower.
Current Multiwfn does not support .cif file. So, if you use Multiwfn to perform Hirshfeld/Becke
surface analysis to study intermolecular interaction in molecular crystals, you commonly need to
use some third-part softwares to open the corresponding .cif file and then properly dig out a cluster
of molecules.
In order to draw fingerprint plot, the mapped function must be set to dnorm, which is default.
Then after the Hirshfeld/Becke surface analysis has finished, in the post-processing menu you will
see an option titled "20 Fingerprint plot analysis", after entering it you will see a menu. If you select
option 0, the fingerprint analysis will be started. By default the fingerprint analysis is performed for
the overall Hirshfeld surface. If you intend to study fingerprint plot in local contact region, you
should use options 1 and 2 to properly define the "inside atoms" set and "outside atoms" set, only
the contact surface between these two sets will be taken into account in the fingerprint plot analysis.
The "inside atoms" set must be a subset of Hirshfeld fragment, while any atom in the "outside atoms"
set should not belong to the Hirshfeld fragment. In the options 1 and 2 you will be asked to input
two filter conditions, their intersection defines the set. Condition 1 corresponds to the atom index
range, the condition 2 corresponds to element. For example, if you input 1,3-6 as condition 1 and
Cl as condition 2, then Cl atoms will be defined as the set if their indices belong to 1,3-6.
After carrying out the fingerprint map analysis, the area of the contact surface will be shown
on screen, you can regard this data as actual interaction area between the sets you defined. Then, in
the post-processing menu, by options 1~3 you can draw the fingerprint plot, save the plot, and export
the original data of the plot (namely the number of surface points in each bin) to plain text file. By
options 3 and 4 you can adjust color scale and the range of X/Y axes, respectively. By option 5 you
can export the surface points corresponding to the current fingerprint plot to finger.pdb in current
folder. In addition, by options 10 and 11, you can draw scatter map for the surface points between
di and de , and export di and de values of all surface points to plain text file.
An example of Becke surface analysis is given in Section 4.12.5. An example of Hirshfeld

analysis and drawing fingerprint plot is given in Section 4.12.6, this example also illustrates how to
easily plot pretty color-mapped Hirshfeld surface via VMD program.
Information needed: Depending on the real space function used to define the surface and that
mapped on the surface. At least atom coordinates must be provided. (For local electron affinity,
virtual MOs must be presented, hence such as .mwfn, .fch, .molden and .gms should be used)
155
3 Functions
3.16 Processing grid data (13)
If Gaussian-type cube file (.cub) or DMol3 grid file (.grd) was loaded when Multiwfn boots
up, or grid data has been generated by main function 5 or other functions, a set of grid data will
presented in memory (which will be referred to as "present grid data" below), then this module will
be available. If grid data has not been presented in memory but you choose this main function, you
can also directly load grid data from an external file.
In this module, you can visualize present grid data, extract data in a specified plane, perform
mathmetical algorithm, set value in specified range by corresponding options and so on. These
options will be described below.
3.16.0 Visualize isosurface of present grid data (-2)
Visualize isosurface of present grid data in a GUI window, this is useful to check validity of
the grid data updated by some functions (e.g. function 11)
3.16.1 Output present grid data to Gaussian cube file (0)
If you choose this function, present grid data (may be has updated by using function 11, 13, 14,
15) along with atom information will be outputted to a cube file.
3.16.2 Output all data points with value and coordinate (1)
By this function, all present grid data will be outputted to output.txt in current folder, the first
three columns correspond to X,Y,Z value (in Å), the last column is data value.
3.16.3 Output data points in a XY/YZ/XZ plane (2, 3, 4)
By these functions, the grid data in the XY/YZ/XZ plane with specified Z/X/Y value will be
outputted to output.txt in current folder, which is a plain text file, you can load it to visualization
softwares such as sigmaplot and then plot plane graphs. Since grid data is discretely distributed, the
actual outputted plane is the one nearest to your input Z/X/Y value.
Please read program prompts for the meaning of each column in output file.
3.16.4 Output average data of XY/YZ/XZ planes in a range of Z/X/Y

(5,6,7)
By these functions, the average grid data in some XY/YZ/XZ planes whose Z/X/Y coordinate
are in specified range will be outputted to output.txt in current folder. The column 1/2/3/4 correspond
to X,Y,Z,value respectively, unit is Å.
3.16.5 Output data points in a plane defined three atom indices or three
points (8,9)
By these two functions, the data in an arbitrary plane can be outputted to plain text file.
However if the plane you are interested in is XY/YZ/XZ plane, you should use function 2,3,4 instead
156
3 Functions
respectively. You can define the plane by inputting three atom indices or by inputting three points.
You need to input tolerance distance, the data points whose distance to the plane are short than
this value will be outputted. In general it is recommended to input 0 to use default value.
Then if you want to project the data points to XY plane so that you can load them into some
visualization softwares and then plot them as plane graph, you can input 1 to tell program to do that.
You will find Z values of all points in the output file are zero.
3.16.6 Output data points in specified value range (10)
Like function 2, but only the data points whose value are in specified range will be outputted.
If you input both lower and upper limit of value as k, then the data between k-abs(k)*0.03 and
k+abs(k)*0.03 will be outputted.
3.16.7 Grid data calculation (11)
In this function, you can perform algorithm for present grid data by corresponding options,
then the grid data will be updated, and then you can use such as function -2 to visualize the updated
grid data, use function 0 to output the updated grid data as cube file or extract data in a plane by
function 2~9, etc.
Supported operations are shown below, where A means value of present grid data, B means
value at corresponding point in the cube file that will be loaded. C means the updated value at
corresponding point.
 1 Add a constant e.g. A+0.1=C
 2 Add a grid file i.e. A+B=C
 3 Subtract a constant e.g. A-0.1=C
 4 Subtract a grid file i.e. A-B=C
 5 Multiplied by a constant e.g. A*0.1=C
 6 Multiplied by a grid file i.e. A*B=C
 7 Divided by a constant e.g. A/5.2=C
 8 Divided by a grid file i.e. A/B=C
 9 Exponentiation e.g. A1.3=C
 10 Square sum with a grid file i.e. A2+B2=C
 11 Square subtract with a grid file i.e. A2-B2=C
 12 Get average with a grid file i.e. (A+B)/2=C
 13 Get absolute value i.e. |A|=C
 14 Get exponential value with base 10 i.e. 10A=C
 15 Get logarithm with base 10 i.e. log10(A)=C
 16 Get natural exponential value i.e. eA=C
 17 Get natural logarithm i.e. ln(A)=C
 18 Add a grid file multiplied by a value i.e. A+0.4*B=C
 19 The same as 6, but multiplied by a weighting function at the same time. The weighting
function is defined as min(|A|,|B|)/max(|A|,|B|). So, at any point, the more the magnitude
of the data in A deviates from the counterpart in B, the severely the result will be punished.
 20 Multiplied by a coordinate variable: This option multiplies all grid data by one of
selected coordinate variables X, Y and Z.
157
3 Functions
 21 Get minimal value with another function i.e. min(A,B)

 22 Get min(|A|,|B|)
If the operation you selected involves a constant, you will be prompted to input its value; If
involves another cube or Dmol3 .grd file, you will be prompted to input its filename, of which the
origin point, translation vectors and data points in each dimensions must be identical to the grid data
presented in memory.
3.16.8 Map values of a cube file to specified isosurface of present grid

data (12)
The function is especially useful if you have a electron density cube file and corresponding
ESP cube file, you can obtain ESP values of the points laying on the vdW surface, which may be
defined as the isosurface with electron density isovalue of 0.001. (Note that main function 12 can
realize the same goal, meanwhile the accuracy is higher)
You need to input a isovalue to define the isosurface of present grid data, assume that you input
p, and then input deviation in percentage, referred to as k here, then the data points whose values
are between p+abs(p)*0.01*k and p-abs(p)*0.01*k will be regarded as isosurface points.
Subsequently, you need to input the filename of another cube file (should has identical grid setting
as present grid data), the value in this cube file of those isosurface points will be exported to
output.txt in current folder, along with X/Y/Z coordinates.
3.16.9 Set value of the grid points that far away from / close to some
atoms (13)
By this function, the value of grid points beyond or within scaled vdW region of a molecular
fragment can be set to a specific value. This is very useful for screening uninteresting region when
showing isosurface, namely setting value of this region to a very large value (very positive or very
negative, according to the character of the grid data).
You need to input a scale factor for vdW radius, then input expected value. After that, you need
to specify fragment, you can either directly input atomic indices (e.g. 3,5,1-15,20), or input filename
of a plain text file, in which a molecular fragment is defined as atomic list, below is an example of
the file:
3
1 3 4
where 3 means there are three atoms in this fragment, 1, 3, 4 are corresponding atom indices.
Then all grid points that beyond the region occupied by scaled vdW spheres of the fragment
atoms will be set to specific value.
If the scale factor of vdW sphere is set to a negative value, e.g. -1.3, then all grid points that
within the scaled vdW surface of the fragment will be set to the specific value.
An example of using this function is given in Section 4.13.4.1.
158
3 Functions
3.16.10 Set value of the grid points outside overlap region of two
fragments (14)
This function is similar to function 13, but only the grid points outside superposition region of
scaled vdW regions of two fragments will be set to a specified value. You can either directly input
atomic indices of the two fragments, or prepare two files containing atom lists for the two fragments,
the format is the same as function 13.
This function is very useful if you are only interested in studying isosurfaces between two
fragments, all isosurfaces outside this region can be screened by setting grid data value to very large.
An illustrative example is given in Section 4.13.4.2.
3.16.11 If data value is within certain range, set it to a specified value

(15)
You need to input lower and upper limit value and a expected value, if any value in present
grid data is within the range you inputted, its value will be set to the expected value.
3.16.12 Scale data range (16)
By this function, the value of present grid data can be linearly scaled to certain range. You need
to input original data range, assumed that you inputted 0.5,1.7, and you inputted -10,10 as new data
range, then all the value of present grid data that higher than 1.7 will be set to 1.7, all the value lower
than 0.5 will be set to 0.5. After that, the value between 0.5 and 1.7 will be linearly scaled to -10,10.
It may be more clear if the algorithm is expressed as pseudo-code:
where (value>0.5) value=0.5
where (value<1.7) value=1.7
all value = all value - 0.5
ratiofac = [10 - (-10)] / (1.7 - 0.5) = 20/1.2
all value = all value * ratiofac
all value = all value + (-10)
3.16.13 Show statistic data of the points in specific spatial and value
range (17)
This function can output statistic data of the points in specific spatial and value range. If user
do not want to impose any constraint (namely the statistic data is for all data points), input 1. If
constraint is needed to be imposed, user should input 2, and then input lower and upper limit of
X,Y,Z coordinates and value in turn.
The minimum and maximum value, average, root mean square, standard deviation, volume,
sum, integral and barycenter position of the data points satisfied the constraints will be outputted.
The positive, negative and total barycenter are computed respectively as
159
3 Functions
+ +
R + =  ri f (ri ) /  f (ri )
i i
− −
R − =  r j f (r j ) /  f ( r j )
j j
all all
R tot =  rk f (rk ) /  f (rk )
k k
where f is the data value, r denotes coordinate vector, the indices i, j and k run over positive, negative
and all grid points respectively.
3.16.14 Calculate and plot integral curve in X/Y/Z direction (18)
Integral curve is defined as below (e.g. in Z direction)

z ' ++
I ( z' ) =    p( x, y, z)dxdydz
zini − − 
Local integral curve is defined as (e.g. in Z direction)

++
I L ( z) =   p( x, y, z)dxdy
− − 
Evidently
z'
I ( z ') = I
zini
L ( z )dz
In Multiwfn, I and IL curves are evaluated based on cubic numerical integration of grid data.
User first needs to choose which direction will be integrated, and then input the lower and upper
limit of the coordinate in this direction. Then integral curve and local integral curve will be
calculated within this range. After that, by corresponding options, the graph of the curves can be
plotted or saved, the curve data can be exported to current folder as plain text file.
Assume that user has chosen Z direction, and the lower and upper limit were set to -5 and 10,
respectively, then the spatial range of the curve generated by Multiwfn will be z=[-5,10], and the zini
in above formula will be -5. If user input the letter a (denotes "all"), then the minimal and maximal
value of Z coordinate of current grid data will regarded as lower and upper limit, respectively.
Worthnotingly, if the integrand is chosen as electron density difference, then the integral curve
sometimes is known as "charge displacement curve" and useful in discussion of charge transfer, see
J. Am. Chem. Soc., 130, 1048 for example. If you want to obtain such a curve, before entering
present function, you should calculate grid data of electron density difference, or directly load the
grid data from external file (e.g. cube file).
A practical example is given in Section 4.13.6.
Information needed: Grid data, atom coordinates
160
3 Functions
3.17 Adaptive natural density partitioning (AdNDP)

analysis (14)
3.17.1 Theory
Famous NBO analysis developed by Weinhold and coworkers is able to recover up to 3-centers
2-electrons (3c-2e) orbitals from density matrix (e.g. by using "3cbond" keyword in NBO program).
Adaptive natural density partitioning (AdNDP), which was proposed by Zubarev and Boldyrev in
year 2008 (Phys. Chem. Chem. Phys., 10, 5207), may be viewed as a natural extension of NBO
analysis aiming for locating N>3 centers orbitals. AdNDP has been extensively used to study
electronic structure characteristic of widespread of cluster systems, by googling "AdNDP" you can
find many related literatures.
Canonical molecular orbitals (CMOs) are generally highly delocalized, often lacking of
chemical significances; While 2c or 3c NBOs are substantially localized, for highly conjugated
system resonant description is often needed (otherwise large non-Lewis composition will occur, that
means current system is inappropriate to be portrayed by single set of NBOs), this somewhat
conflicts with modern quantum chemistry concepts and obscures delocalization natural of electrons
in conjugated system. AdNDP orbitals seamlessly bridged CMOs and NBOs, AdNDP bonding
patterns avoid resonant description and are always consistent with the point symmetry of the
molecule.
The basic idea of AdNDP to generate multi-center orbitals is very similar to NBO analysis, that
is constructing proper sub-block of density matrix in natural atomic orbital (NAO) basis and then
diagonalize it, the eigenvalues and eigenvectors correspond to occupation number and orbital
wavefunction respectively. For example, we want to generate all possible 4-centers orbitals for atom
A,B,C,D, we first pick out corresponding sub-blocks and then combine them to together:
 PA, A PA,B PA,C PA,D 
P PB ,B PB ,C PB ,D 
= 
B, A
P ( A,B ,C ,D )
 PC , A PC ,B PC ,C PC ,D 
 
 PD , A PD ,B PD ,C PD ,D 
After diagonalization of P(A,B,C,D), if one or more eigenvalues exceeded the predefined threshold,
which is commonly set to close to 2.0 (e.g. 1.7), then corresponding orbitals will be regarded as
candidate 4c-2e bonds. Completely identical strategy can be used to generate orbitals with higher
number of centers.
Indeed, the orbital generating process of AdNDP is quite easy once atom combination is
determined, however the searching process of final Nc-2e orbitals in entire system is complicated,
manual inspections and operations are necessary. AdNDP approach has large ambiguity, it is
possible that the searching process carried out by different peoples finally results in different AdNDP
pattern, I think this is the most serious limitation of current AdNDP approach. So, AdNDP is never
a black box, before using it users must have preliminary understanding of the searching process of
the AdNDP implemented in Multiwfn.
Before the search, densities from core-type NAOs are automatically eliminated from the
density matrix, since they have no any contribution to bonding. After that, 1-center orbitals (lone
161
3 Functions
pair), 2-centers orbitals, 3-centers orbitals, 4-centers orbitals ... will be searched in turn, until
residual density (trace of density matrix) is close to zero. The search could be exhaustive, that means
when searching N-centers orbitals, Multiwfn will construct and diagonalize M!/(M-N)!/N! sub-
blocks of density matrix, where M is the total number of atoms. All of the orbitals whose occupation
numbers are larger than the threshold will be added to candidate orbital list. For large system, the
searching process may be very time-consuming or even forbidden, for example, exhaustive search
of 10-centers orbitals in the system with 30 atoms needs to construct and diagonalize 30045015 sub-
blocks of density matrix! This is very difficult to be finished in personal computer, for such case,
user-directed searching is necessary. In Multiwfn you can customly define a search list, then the
exhaustive search will only apply to the atoms in the search list, so that the amount of computation
would be greatly reduced. You can also directly let Multiwfn construct and diagonalize sub-block
of density matrix for specified atom combination. Note that user-directed searching has relatively
high requirements of skill and experience on users.
Once the search of N-centers orbitals is finished, we will get a list containing candidate N-
centers orbitals. We need to pick some of them out as final N-centers AdNDP orbitals. Commonly,
one or more orbitals with the highest occupation numbers will be picked out. Notice that, since some
densities are simultaneously shared by multiple candidate orbitals, if we directly pick out several
candidate orbitals with the largest occupation at one time, the electrons may be overcounted. To
avoid this problem, assume that K orbitals with the highest occupation numbers obviously overlap
with some other candidate orbitals meanwhile there is no evident overlapping between the K orbitals,
we should first pick out K orbitals as final AdNDP orbitals, then Multiwfn will automatically deplete
their density from the density matrix and then reconstruct and diagonalize the corresponding sub-
blocks of density matrix for remained candidate orbitals to update their shapes and occupation
numbers. If there are still some candidate orbitals with occupation numbers close to 2.0, you may
consider picking them out, then remained orbitals will be updated again. Such process may be
repeated several times until there is no orbitals have high occupation numbers. After that, you can
start to search N+1 centers orbitals.
The general requirements of AdNDP analysis are that: The final residual density (corresponds
to non-Lewis composition in NBO analysis) should as low as possible; the occupation numbers of
each AdNDP orbital should as close to 2.0 as possible; the number of centers of AdNDP orbitals
should as less as possible; the resulting orbitals must be consistent with molecular symmetry.
However, there is no unique rule on how to search orbitals and pick out candidate orbitals as
AdNDP orbitals. For example, one can first search 5-centers orbitals before completing search of 3-
centers orbitals, and one can also directly search 6-centers orbitals after the search of 2-centers
orbitals has finished. The sequence of picking out candidate orbitals is also not necessarily always
in accordance to magnitude of occupation numbers. The final AdNDP pattern obtained by different
operations may be different, how to do AdNDP analysis is largely dependent on users themselves.
Actually some molecules may have two or even more equally reasonable AdNDP patterns,
sometimes it is difficult to discriminate which pattern is the best. I have confidence to say that some
AdNDP patterns presented in publicated papers are not the optimal ones. The experience of using
AdNDP approach can be gradually accumulated in practices and during reading related literatures.
AdNDP is very insensitive to basis set quality as NBO analysis, 6-31G* is enough to produce
accurate results for main group elements in the first few rows. Over enhancing basis set quality will
not improve AdNDP analysis results but only lead to increase of the computational burden in
162
3 Functions
diagonalization step, since the size of sub-block of density matrix is directly determined by size of
basis set.
Multiwfn offers capacity of evaluating AdNDP orbital energies. You need to provide a file
containing Fock (or Kohn-Sham) matrix in original basis functions. The Fock matrix can be obtained
from output of Gaussian or other programs. The energy of AdNDP orbital is corresponding diagonal
term of Fock matrix in AdNDP orbital representation. Specifically, Multiwfn performs below
representation transform:
FAdNDP = CT FAO C C = X AONAO c

where FAO is the Fock matrix in original basis function that loaded from user-provided file, C(r,i)
corresponds to coefficient of basis function r in AdNDP orbital i. c(s,i) corresponds to coefficient of
NAO s in AdNDP orbital i. XAONAO is transformation matrix between original basis function and
NAO, i.e. X(t,s) is coefficient of basis function t in NAO s. Energy of AdNDP orbital j is simply
FAdNDP(j,j), which is expectation value of Fock operator of AdNDP orbital wavefunction.
3.17.2 Input file
The output file of NBO program containing density matrix in NAO basis (DMNAO) can be
used as input file for AdNDP analysis. If you also need to visualize AdNDP orbitals or export orbitals
as cube files, .fch file must be provided, meanwhile transformation matrix between NAO and
original basis functions (AONAO) must be presented in the NBO output file.
Assume that you are a Gaussian user, in order to obtain an Gaussian output file containing all
information needed by Multiwfn to perform the AdNDP analysis and visualization, you should write
a Gaussian input file of single point task with pop=nboread keyword in route section, and write
$NBO AONAO DMNAO $END after molecular geometry section with a blank line as separator.
Then run the input file by Gaussian and then convert .chk file to .fch format by formchk utility.
The Gaussian output file (not .fch file) should be used as the initial input file when Multiwfn
boots up. Once you entered AdNDP module, Multiwfn will load NAO information and DMNAO
matrix from this file. If then you choose corresponding options to visualize or export orbitals,
AONAO matrix will be loaded and the program will prompt you to input the path of the .fch file
(if .fch is in the same folder and has identical name as the Gaussian output file, then the .fch will be
automatically loaded).
Multiwfn is also compatible with the output files of stand-alone NBO program (GENNBO), of
course you have to add DMNAO keywords in $NBO section in .47 file. In this case it is impossible
to visualize AdNDP orbitals.
Formally, AdNDP approach is also applicable to open-shell systems; of course, the occupation
threshold should be divided by 2. When you enter AdNDP module, Multiwfn will ask you which
density matrix should be used, the so-called total density matrix is the sum of alpha and beta density
matrix.
Notice that if after you entered AdNDP module Multiwfn suddenly crashes, and the basis set
you used contains diffuse functions, you can try to use another basis set without diffuse functions.
This problem is caused by the bug in NBO 3.1 module, namely in rarely cases the DMNAO output
may be slightly problematic if diffuse functions present. Since AdNDP analysis is quite insensitive
to diffuse functions, they can be safely removed without any loss of accuracy.
If you want to obtain AdNDP orbital energies, Fock matrix corresponding to present system at
163
3 Functions
the same calculation level must be provided in lower-triangular sequence in a plain text file, namely:
F(1,1) F(2,1) F(2,2) F(3,1) F(3,2) F(3,3) ... F(nbasis,nbasis), where nbasis is the total number of
basis functions. The format is free. If you are a Gaussian user, you can add archive file=XXX
keyword between $NBO ... $END, then in the resulting XXX.47 file, search $FOCK and copy all
data between $FOCK ... $END to a plain text file, then this file can be directly used to provide Fock
matrix to Multiwfn (In fact, Multiwfn is also able to automatically locate and read the $FOCK field
when the file name has .47 suffix).
3.17.3 Options
All of the options involved in AdNDP module are introduced below, some options are invisible
in certain cases. If current candidate orbital list is not empty, then all candidate orbitals will always
be printed on screen in front of the menu (except when you select option 5 or 13), the candidate
orbital indices are determined according to occupation numbers. The number of residual valence
electrons of all atoms in the search list is always printed at the upper of the menu, this value
decreases with gradually picking out candidate orbitals as final AdNDP orbitals. If this value is very
low (e.g. lower than 1.4), it is suggested that new Nc-2e AdNDP orbitals will be impossible to be
found between the atoms in the search list.
-10 Return to main menu: Once you choose this option, you will return to main menu,
meantime all results of AdNDP analysis will be lost. Hence the status of AdNDP module can be
reseted by choosing this option and then re-entering the module.
-2 Various other settings and functions: This options have several unimportant subfunctions
and settings. The option "Set maximum number of candidate orbitals to be printed" is worth to
mention, it is used to set how many candidate orbitals will be printed on screen, proper choice of
the threshold can avoid excessive output when very large number of candiates are found.
-1 Define exhaustive search list: In this option, one can customly define search list, exhaustive
search (option 2) will only apply to the atoms in the search list. All commands in this defining
interface are self explanatory. Notice that default search list includes all atoms of the molecule.
0 Pick out some candidate orbitals and update occupations of others: Users need to input
two numbers, e.g. i,j, then all candidate orbitals from i to j will be picked out and saved as final
AdNDP orbitals. If user only input one number, e.g. k, then k candidate orbitals with the largest
occupation numbers will be picked out. After that, the eigenvectors (orbital shape) and eigenvalues
(occupation numbers) of remained candidate orbitals will be updated as mentioned earlier.
1 Perform orbitals search for a specific atom combination: Users need to input indices of
some atoms, e.g. 3,4,5,8,9, then sub-block of density matrix for atom 3,4,5,8,9 will be constructed
and diagonalized, all resultant eigenvectors will be added to candidate orbitals list, meantime all
previous candidate orbitals will be removed. There is no limit on the number of inputted atoms.
2 Perform exhaustive search of N-centers orbitals within the search list: N atoms will be
selected out from the search list in an exhaustive manner, assume that the search list contains M
atoms, then totally M!/(M-N)!/N! atom combinations will be formed. For each combination,
corresponding sub-block of density matrix will be constructed and diagonalized, all eigenvectors
with eigenvalues larger than user-defined threshold will be add to candidate orbital list. Old
candidate orbital list will be cleaned.
3 Set the number of centers in the next exhaustive search: Namely set the value N in option
2. Once exhaustive search of N-centers orbitals is finished, N will be automatically increased by one.
164
3 Functions
4 Set occupation threshold in the next exhaustive search: Namely set the threshold used in
option 2.
5 Show information of AdNDP orbitals: Print occupation numbers and involved atoms of all
saved AdNDP orbitals.
6 Delete some AdNDP orbitals: Input two numbers, e.g. i, j, then saved AdNDP orbitals from
i to j will be removed.
7 Visualize AdNDP orbitals and molecular geometry: The path of corresponding .fch file
will be prompted to be inputted, after loading some necessary information from the file, a GUI
window will pop up and molecular geometry will be shown. Isosurfaces of AdNDP orbitals can be
plotted by clicking corresponding numbers in the right-bottom list
8 Visualize candidate orbitals and molecular geometry: Analogous to option 7, but used to
visualize isosurfaces of candidate orbitals. It is useful to visualize the isosurfaces before picking out
some candidate orbitals as final AdNDP orbitals.
9 Export some AdNDP orbitals to Gaussian-type cube files: User need to choose grid setting
and then input index range, e.g. 2-4, then wavefunction value of AdNDP orbitals 2, 3, 4 will be
calculated and exported to AdNDPorb0002.cub, AdNDPorb0003.cub and AdNDPorb0004.cub in
current folder, respectively. They are Gaussian-type cube files and can be visualized by many
softwares such as VMD.
10 Export some candidate orbitals to Gaussian-type cube files: Analogous to option 9, but
used to export cube files for candidate orbitals.
11/12 Save/Load current density matrix and AdNDP orbital list: Option 11 is used to save
current density matrix and AdNDP orbital list in memory temporarily, when density matrix and
AdNDP orbital list is changed, you can choose option 12 to recover previous state.
13 Show residual density distributions on the atoms in the search list: After choosing this
option, population number of each atom in the search list will be calculated according to present
density matrix and then printed out. If some neighboring atoms have large population number, it is
suggested that multi-center orbitals with high occupation number may appear on these atoms; while
the atoms with low population number often can be ignored in the following searching process.
Thus this option is very helpful for setting up user-directed searching.
14 Export AdNDP orbitals as .molden file: Via this option, all picked AdNDP orbitals will
be exported as AdNDP.molden in current folder (see Section 2.5 for introduction of .molden format).
By using this file as input file, you can perform various kinds of analyses for AdNDP orbitals (e.g.
orbital composition analysis by main function 8, plotting plane map via main function 4). Note that
if there are N basis functions and M AdNDP orbitals have been picked out, then the first M orbitals
in the AdNDP.molden will correspond to the AdNDP orbitals, while the other N-M orbitals in this
file are meaningless and can be simply ignored.
15 Evaluate and output composition of AdNDP orbitals: This option is used to calculate
orbital composition of picked AdNDP orbitals by natural atomic orbital (NAO) method, which has
been introduced in Section 3.10.4.
16 Evaluate and output energy of AdNDP orbitals: This function is used to evaluate energy
of AdNDP orbitals that have already been picked out. Multiwfn will prompt you to input the path of
the file containing Fock matrix in original basis functions, the elements of the matrix should be
recorded in lower-triangular sequence, the NBO .47 file containing $FOCK field can also be directly
used as input file. Then after a simple transformation, orbital energies are immediately outputted.
165
3 Functions
AdNDP analysis is relatively complicated and not a blackbox, please follow the examples in
Section 4.14 before using this module to analyze your systems.
Information needed: NBO output file (with AONAO DMNAO keywords), .fch file (only
needed when visualizing and exporting cube file for AdNDP orbitals, or exporting AdNDP orbitals
as .molden file), plain text file (containing Fock matrix. Only needed if you want to gain orbital
energies)
3.18 Fuzzy atomic space analysis (15)
3.18.0 Basic concepts
Before introducing each individual function, here I first introduce some basic concepts of fuzzy
atomic space.
Atomic space is the local space attributed to specific atom in the whole three-dimension
molecular space. Below we will express atomic space as weighing function w. The methods used to
partition the whole space into atomic spaces can be classified to two categories:
1 Discrete partition methods: The two representative methods are Bader's partition (also
known as AIM partition) and Voronoi partition. They partition molecular space discretely, so any
point can be attributed to only one atom, in other words,
 w A (r ) = 1 if r   A

 w A (r ) = 0 if r   A
where ΩA is atomic space of atom A.
2 Fuzzy partition methods: The representative methods include Hirshfeld, Becke, Hirshfeld-
I and ISA. They partition molecular space contiguously, atomic spaces overlap with each other, any
point may simultaneously attributed to many atoms to different extent, and the weights are
normalized to unity. In other words, below two conditions hold for all atoms and any point
0  w A (r )  1 A
w
B
B (r ) = 1
The most significant advantage of fuzzy partition may be that the integration of real space
function in fuzzy atomic space is much easier than in discrete atomic space. By using Becke's
numerical DFT integration scheme (J. Chem. Phys., 88, 2547 (1988)), high accuracy of integration
in fuzzy atomic space can be achieved for most real space functions at the expense of relatively low
computation effort. In the fuzzy atomic space analysis module of Multiwfn, all integrations are
realized by this scheme. The more integration points are used, the higher integration accuracy can
be reached, one can adjust the number of points by "radpot" and "sphpot" parameter in settings.ini.
In fuzzy atomic space analysis module of Multiwfn, one can obtain many properties that based
166
3 Functions
on fuzzy atomic spaces. Currently, the most widely used definitions of fuzzy atomic spaces, namely
Hirshfeld, Hirshfeld-I and Becke are supported, they are introduced below. One can choose which
fuzzy atomic spaces will be used by option -1.
Hirshfeld atomic space: In Theor. Chim. Acta (Berl.), 44, 129 (1977), Hirshfeld defined the
atomic space as
 Afree (r − R A )
w Hirsh
(r ) =
  Bfree (r − R A )
A
B
free
where R is coordinate of nucleus, ρ denotes spherically averaged atomic electron density in free-
state.
In option -1, you will found two options "Hirshfeld" and "Hirshfeld*". The former uses
atomic .wfn files to calculate the weights, they must be provided by yourself or let Multiwfn
automatically invoke Gaussian to generate them, see Secion 3.7.3 for detail. The latter evaluates the
weights directly based on built-in radial atomic densities and thus is more convenient, detail can be
found in Appendix 3. I strongly suggest using "Hirshfeld*" instead of "Hirshfeld".
Hirshfeld-I (HI) atomic space: This is a well-known extension of Hirshfeld method, it was
proposed in J. Chem. Phys., 126, 144111 (2007). Commonly the atomic space defined by HI is more
physically meaningful than that of Hirshfeld, since it can respond actual molecular environment.
Unfortunately, HI is much more expensive than Hirshfeld due to its iterative nature. Details of
Hirshfeld-I and its implementation in Multiwfn have been introduced in Section 3.9.13 and thus will
not be repeated here. When you choose HI in option -1, Multiwfn will first perform regular HI
iterations (If you are confused by the operations, please consult the example of computing HI
charges in Section 4.7.4). After HI atomic spaces have converged, you can do subsequent analyses.
Becke atomic space: First, consider a function p
p(d ) = (3 / 2)d − (1 / 2)d 3

which can be iterated many times
f1 ( d ) = p( d )
f 2 ( d ) = p[ p( d )]
f 3 ( d ) = p{ p[ p( d )]}
...
Then define a function s
sk (t ) = (1 / 2)[1 − f k (t )]
The plot of sk versus to t is
167
3 Functions
1.0
0.9 k=1
k=2
0.8 k=3
0.7 k=4
k=5
0.6
sk(t)
0.5
0.4
0.3
0.2
0.1
0.0
-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
t
From above graph it can be seen that sk gradually reduces from 1 to 0 with t varying from -1 to
1. The larger the k is, the sharper the curve becomes. The weighting function of Becke atomic space
is based on simple transformation of sk, for details please consult original paper J. Chem. Phys., 88,
2547 (1988).
PA (r )
w ABecke (r ) =
 PB (r)
B
PA (r ) =  sk ( AB (r ))  AB (r ) =  AB (r ) + a AB (1 −  AB (r ) 2 )
B A
u AB  AB − 1 R Acov
a AB = u AB =  AB =
u AB − 1
2
 AB + 1 RBcov
a AB = −0.5 if a AB  −0.5

a AB = 0.5 if a AB  0.5
rA − rB
 AB (r ) = R AB = R A − R B rA = r − R A rB = r − R B
R AB
where R stands for coordinates of nucleus. Rcov denotes covalent radius.
The number of iterations, namely k value, can be set by option -3. The default value (3) is
appropriate for most cases. The definition of the covalent radius used to generate Becke atomic
space can be chosen by option -2. Through corresponding suboptions, one can directly select a set
of built-in radii (CSD radii, modified CSD radii, Pyykkö radii, Suresh radii, Hugo radii), load radii
information from external plain text file (the format required is described in the program prompts),
or modify current radii by manual input.
The origin paper of CSD radii is Dalton Trans., 2008, 2832, these radii were deduced from
statistic of Cambridge Structural Database (CSD) for the elements with atomic numbers up to 96.
Pyykkö radii was defined in Chem. Eur. J., 15, 186 (2008), which covers the entire periodic table,
Groups 1–18, Z=1–118. Suresh radii was proposed in J. Phys. Chem. A, 105, 5940 (2001), which is
based on theoretically calculated geometries of H3C-EHn, the defined radii cover most of main group
and transition elements in periodic table. Hugo radii was proposed in Chem. Phys. Lett., 480, 127
168
3 Functions
(2009), which has clear physical meaning and is based on atomic ionization energy. Notice that
Hugo radii for hydrogen is rather large (even larger than Kr by 0.01 Bohr).
I found it is inappropriate to directly use any covalent radii definition shown above to define
Becke's atomic space. The covalent radii of metal elements in IA and IIA groups are always large,
e.g. CSD radius of lithium is 1.28 Å. While covalent radii of elements in such as VIIA group are
always small, e.g. CSD radius of fluorine is only 0.58 Å. For main groups, the elements with small
(large) covalent radius generally have large (small) electronegativity. So, in molecule environment,
the atoms with small (large) covalent radius prefer to withdraw (donate) electrons to expand (shrink)
their effective size, this behavior makes actual radii of main group elements in each row equalized.
In order to faithfully reflect this behavior, I defined the so-called "modified CSD radii", namely the
CSD radii of all main group elements (except for the first row) are replaced by CSD radii of the IVA
group element in corresponding row, while transition elements still use their original CSD radii. The
modified CSD radii are the default radii definition for Becke's atomic space.
The Becke atomic space of carbon in acetamide constructed by default parameters is illustrated
below
3.18.1 Integration of a real space function in fuzzy atomic spaces (1)
This function is used to integrate real space function P in atomic spaces
I A =  wA (r ) P(r )dr
A
For example, if P is chosen as electron density, then IA will be the electron population number
of atom A.
P may be also chosen as the real space functions involving coordinates of two electrons, such
as exchange-correlation density and source function. For this case, the coordinate of reference point
can be set by option -10 (this is equivalent to set "refxyz" in settings.ini). If you have carried out
topology analysis, you can also use -11 to set a critical point as reference point, this is especially
convenient for studying source function (for which bond critical point is usually set as reference
169
3 Functions
point).
The "% of sum" and "% of sum abs" in output is defined as ( I A / I
B
B )  100% and
( I A /  I B )  100% , respectively.
B
By default, all atomic spaces will be integrated. If you only need integral value of certain atoms,
you can use option -5 to define the atom list.
Special note: This option uses single-center integration grid to integrate each atom, commonly this is no problem.
However, if you choose to use Hirshfeld or Hirshfeld-I partition, and the real space function to be integrated varies
very fast around nuclei (e.g. Laplacian of electron density), then this option is unable to give accurate result. You
should use option 101 (invisible in the interface) instead, which employs molecular integration grid to integrate every
Hirshfeld atoms, the result is always very accurate.
3.18.2 Integration of a real space function in overlap spaces (8)
This function is used to integrate specified real space function P in overlap spaces between
atom pairs
I AB =  wA (r ) wB (r ) P(r ) d r
A
For example, if P is chosen as electron density, then IAB will be the number of electrons shared
by atom A and B. P may be also chosen as the real space functions involving coordinates of two
electrons.
Integrations of positive and negative parts of P are outputted separately. Meanwhile, sum of
diagonal elements I
A
AA , sum of non-diagonal elements I
A B A
AB and sum of all elements
 I
A B
AB for positve and negative parts are also outputted together.
Currently only the fuzzy atomic space defined by Becke can be used together with this function.
3.18.3 Calculate atomic and molecular multipole moments (2)
This function is used to evaluate atomic and molecular monopole, dipole, quadrupole moments
and octopole moments. All units in the output are in a.u.
In below formulae, superscript A means an atom named A. x, y and z are the components of
electron coordinate r relative to nuclear coordinate R.
x = rx − RxA y = ry − R yA z = rz − RzA
and r2 = x2 + y2 + z2.
Atomic monopole moment due to electrons is just negative of electron population number
pA = −  wA (r )  (r) d r
A
Atomic charges are outputted together, namely qA = pA + ZA, where Z denotes nuclear charge.
Atomic dipole moment is useful to measure polarization of electron distribution around the
atom, which is defined as
170
3 Functions
  xA   x
 A
μ =   y  = −   y  wA (r )  (r ) d r
A
A 
  zA   z 
 
Its magnitude, or say its norm is
μ A = ( xA )2 + (  yA )2 + ( zA )2
Multiwfn also outputs the contribution of present atom to total molecular dipole moment,
which is evaluated as qAR + A.
Cartesian form of atomic quadrupole moment tensor is defined as (see Section 1.8.7 of book
The Quantum Theory of Atoms in Molecules-From Solid State to DNA and Drug Design).
  xx
A
 xy
A
 xzA  3x 2 − r 2 3xy 3xz 
  1  
Θ A =   Ayx  Ayy  Ayz  = −   3 yx 3y2 − r2 3 yz  wA (r )  (r ) d r
2 A
 zxA zyA zzA   3zx 3zy 3z 2 − r 2 
 
whose magnitude can be calculated as
Θ A = (2 / 3) [( xx
A 2
) + ( Ayy ) 2 + ( zzA ) 2 ]
Atomic quadrupole moments in Cartesian form can be used to exhibit the deviation of the
𝐴 𝐴
distribution from spherical symmetry. Specifically, 𝑥𝑥 <0 (𝑥𝑥 >0) indicates that the electron
density of atom A is elongated (contracted) along X axis. If the atomic electron density has spherical
symmetry, then Θxx = Θyy = Θzz. Noticeably, the Cartesian quadruple moment tensor Θ given here is
in traceless form, that means the condition Θxx + Θyy + Θzz = 0 always holds.
The atomic quadrupole and octopole moments in spherical harmonic form are also outputted.
The general expression of multipole moments in spherical harmonic form is
QlA,m = −  RlA,m (r) wA (r)  (r) d r

A
All of the five components of quadrupole moment in spherical harmonic form correspond to
R2,0 = (3z 2 − r 2 ) / 2
R2,−1 = 3 yz R2,1 = 3xz
R2,−2 = 3xy R2,2 = ( 3 / 2)( x 2 − y 2 )
All of the 7 components of octopole moment in spherical harmonic form correspond to

R3,0 = (1 / 2)(5 z 2 − 3r 2 ) z
R3,−1 = 3 / 8(5z 2 − r 2 ) y R3,1 = 3 / 8(5 z 2 − r 2 ) x
R3,−2 = 15 xyz R3,2 = ( 15 / 2)( x 2 − y 2 ) z
R3,−3 = 5 / 8(3x 2 − y 2 ) y R3,3 = 5 / 8( x 2 − 3 y 2 ) x
The magnitude of multipole moments in spherical harmonic form is calculated as
171
3 Functions
QlA =  (Q
m
A 2
)
l ,m
At the end of the output, the total number of electrons, molecular dipole moment and its
magnitude are outputted. Molecular dipole moment is calculated as the sum of all contributions
from atomic dipole moments and atomic charges (i.e. the sum of all "Contribution to molecular
dipole moment" terms in the output information)
μ mol =  ( q AR A + μ A )
A
In addition, Multiwfn outputs molecular quadruple and octopole moments in Cartesian form and
spherical harmonic form. For example, molecular quadruple moment of Θxy is evaluated as
3 
xy =  
2 A
RxA RyAZ A −  xy  (r ) d r 

By default, atomic multipole moments for all atoms are evaluated, and multipole moments of
the whole system are printed at the end. If you only need multipole moments of some atoms, you
can use option -5 to define an atom list, in this case the multipole moments of selected atoms will
be calculated and outputted. In addition, via this feature you can calculate multipole moments of a
molecule in a molecular complexes, or calculate multipole moments of a fragment in a molecule,
because in this case the "Molecular dipole and multipole moments" printed at the end of output are
only contributed by the atoms in the defined list. See example in Section 4.15.3 for illustration of
use of this feature.
PS: If your purpose is only calculating electric dipole and multipole moments for the whole system, it is best to
use the function described in Section 3.300.5, it is significantly faster and more accurate since it calculates them
analytically.
3.18.4 Calculate atomic overlap matrix (3)
This function is used to calculate atomic overlap matrix (AOM) for molecular orbitals or
natural orbitals in atomic spaces, the AOM will be outputted to AOM.txt in current folder. The
element of AOM is defined as
Sij ( A) =  i (r) j (r)d r

A
where i and j are orbital indices. Notice that the highest virtual orbitals will not be taken into account.
For example, present system has 10 orbitals in total, 7,8,9,10 are not occupied, and user has set the
occupation number of orbital 3 to zero by option 26 in main function 6, then the dimension of each
AOM outputted by Multiwfn will be (6,6), corresponding to the overlap integral between the first 6
orbitals in each atomic space.
Since orbitals are orthonormal in whole space, in principle, summing up AOMs for all atoms
(corresponding to integrating in whole space) should yield an identity matrix
SUM =  S( A) = I
A
Of course, this condition does not strictly hold, because the integration is performed numerically as
mentioned earlier. The deviation of SUM to identity matrix is a good measure of numerical
integration accuracy
172
3 Functions
 SUM i, j − Ii , j
Error = i j
N atom
Multiwfn automatically calculates and outputs the "Error" value. If "Error" is not small enough,
e.g. >0.001, then you may want to improve the integration accuracy by increasing "radpot" and
"sphpot" in settings.ini.
For unrestricted HF/DFT or unrestricted post-HF wavefunction, AOM for  and β orbitals will
be outputted respectively.
3.18.5 Calculate localization and delocalization index (4)
For open-shell systems, the LI () and DI () are calculated for each spin of electrons
respectively. Below only the expression of LI and DI for  electrons is given. For β electrons, just
replacing  with β, similarly hereinafter. The electrons in atomic space A that can delocalize to
atomic space B is computed as
  ( A → B) = −   XC,tot (r1 , r2 ) d r1 d r2
A B
where ГXC is exchange-correlation density; if you are not familiar with it, please consult the
discussion in part 17 of Section 2.6. The electrons in atomic space B that can delocalize to atomic
space A is
  ( B → A) = −   XC,tot (r1 , r2 ) d r1 d r2
B A
Clearly, above two terms are identical in value, therefore we define DI between A and B as below,
it measures the total number of  electrons shared by atom A and B
  ( A, B) =   ( A → B) +   ( B → A) = −2  XC,tot (r1, r2 )d r1 d r2

A B
The LI measures the number of  electrons localized in an atom. Note that this quantity is not
additive.
  ( A) = −   XC,tot (r1 , r2 ) d r1 d r2 =   ( A, A) / 2
A A
The relationship between LI, DI and the population number of electrons in atomic space is
given below, the physical meaning is that the sum of  electrons of atom A that localized in atom A
and that delocalized to other regions is the total number of  electrons of space A.
  ( A) + (1 / 2)    ( A, B)    ( A) +    ( A → B) =
B A B A
= − 
 ,tot
XC (r1 , r2 ) d r1 d r2 =   (r ) d r = N A

A A
Using the approximate expression of ГXC, the DI and LI can be explicitly written as
  ( A, B ) = 2 i j Sij ( A) Sij ( B )
i j
 ( A) =  i j Sij ( A) Sij ( A)

i j
173
3 Functions
Total DI and LI are the summation of  part and β part
 ( A, B) =   ( A, B) +   ( A, B)
 ( A) =  ( A) +  ( A)
Since in closed-shell case   ( A, B) =   ( A, B) , one can evaluate total DI as
m n
 ( A, B) = 2  ( A, B) = 2  2 Smn ( A) Smn ( B) = 2 mn Smn ( A) Smn ( B)
m n 2 2 m n
where m and n denote closed-shell natural orbitals. Similarly, the total LI for closed-shell cases is
 ( A) =  mn S mn ( A) S mn ( A)
m n
For closed-shell systems, it is argued that the value of total DI is a quantitative measure of the
number of electron pairs shared between two atoms. For example, total (A,B)=1.0 implies a pair of
electron (an  and a  electrons) is shared between atom A and B. (In fact, this is strictly true only
for nonpolar bonds such as H-H bond in H2. In polar bonds, the DI must be lower than formal bond,
because what total DI actually reflects is the effective number of electron pairs shared by two atoms
and thus somewhat reflects covalency. Note that the value of DI is very sensitive the definition of
atomic space employed)
For single-determinant wavefunction, because of integer occupation number of orbitals, DI and

LI can be simplified as
occ occ
  ( A, B ) = 2 Sij ( A) Sij ( B )
i j
occ occ
 ( A, B ) = 4 Smn ( A) Smn ( B )
m n
occ occ
 ( A) =  Sij ( A) Sij ( A)
i j
occ occ
 ( A) = 2 Smn ( A) Smn ( A)
m n
Conventionally, LI and DI are calculated in AIM atomic space (also called as AIM basin).
While in fuzzy atomic space analysis module of Multiwfn, they are calculated in fuzzy atomic space,
the physical nature is the same. According to the discussion presented in J. Phys. Chem. A, 109,
9904 (2005) (compare Eq. 13 and Eq. 18), the DI calculated in fuzzy atomic space is just the so-
called fuzzy bond order, which was defined by Mayer in Chem. Phys. Lett., 383, 368 (2004).
For closed-shell system, atomic valence can be calculated as the sum of its fuzzy bond orders
V ( A) =  ( A, B)
B A
In Multiwfn, before calculating LI and DI, AOM is calculated first automatically, this is the
most time-consuming step. For open-shell systems, the LI and DI for  and β electrons, as well as
for all electrons are outputted respectively. Notice that the diagonal terms of DI matrix are calculated
as the sum of corresponding off-diagonal row (or column) elements. For closed-shell system, as
174
3 Functions
stated above, they correspond to atomic valence.
For strictly planar molecules, because overlap integral of σ orbital and π orbital is exactly zero
in atomic space, the contributions from σ and π electrons to DI can be exactly decomposed as DI-σ
and DI-π
  ( A, B ) =   ( A, B ) +   ( A, B )
 
  ( A, B ) = 2 i j Sij ( A) Sij ( B )
i j
 
  ( A, B ) = 2 i j Sij ( A) Sij ( B )
i j
Similarly, LI can be decomposed as LI-σ and LI-π. Summing up corresponding off-diagonal

elements in DI-σ and DI-π matrix gives σ-atomic valence and π-atomic valence, respectively. If you
want to compute DI/LI-σ (DI/LI-π), before the DI/LI calculation, you should set the occupation
numbers of all π orbitals (σ orbitals) to zero by subfunction 26 of main function 6.
By the way, in some literatures, especially the ones written by Bernard Silvi, the variance of
electronic fluctuation in atomic space σ2(A) and the covariance of fluctuation of electron pair
between two atomic spaces cov(A,B) are discussed. They are not directly outputted by Multiwfn,
because there is a very simple relationship correlates σ2(A), cov(A,B) and DI(A,B), thus you can
calculate them quite easily, see Chem. Rev., 105, 3911 (2005) for derivation
cov( A, B ) = − ( A, B ) / 2
 2 ( A) = N A −  ( A) = −  cov( A, B ) =  ( A, B) / 2
B A B A
where NA is the electron population number in A. As mentioned above, the diagonal terms of the DI
matrix outputted by Multiwfn are calculated as the sum of off-diagonal elements in the
corresponding row (or column), hence you can simply obtain σ2 by dividing corresponding diagonal
term of DI matrix by two.
A quantity closely related to σ2 is the relative fluctuation parameter introduced by Bader, which
indicates the electronic fluctuations for a given atomic space relative to its electron population, you
can calculate it manually if you want
F ( A) =  2 ( A) / N A
Alternatively, you can calculate below value to measure the proportion of the electrons localized in
the atomic space
l ( A) =  ( A) / N A
3.18.6 Calculate PDI (5)
Para-delocalization index (PDI) is a quantity used to measure aromaticity of six-membered

rings. PDI was first proposed in Chem. Eur. J., 9, 400 (2003), also see Chem. Rev., 105, 3911 (2005)
for more discussion. PDI is essentially the averaged para-delocalization index (para-DI) in six-
membered rings.
175
3 Functions
 (1,4) +  (2,5) +  (3,6)

PDI =
3
The basic idea behind PDI is that Bader and coworkers reported that DI in benzene is greater
for para-related than for meta-related carbon atoms. Obviously, the larger the PDI, the larger the
delocalization, and the stronger the aromaticity. The main limitation of the definition of PDI is that
it can only be used to study aromaticity of six-membered rings, and it was shown that PDI is
inappropriate for the cases when the ring plane has an out-plane distortion.
In Multiwfn, before calculating PDI, AOM and DI are first calculated automatically. Then you
will be prompted to input the indices of the atoms in the ring that you are interested in, the input
order must be consistent with atom connectivity.
PDI currently is only available for closed-shell systems, although theoretically it may be
possible to be extended to open-shell cases.
Note that for completely planar systems, since DI can be decomposed to  and π parts, PDI
can also be separated as PDI- and PDI-π to individually study  aromaticity and π aromaticity. In
order to calculate PDI- (PDI-π), before enter present module, you should first manually set
occupation number of all MOs except for π () MOs to zero (or you can utilize option 22 in main
function 100 to do this step, which will be much more convenient).
3.18.7 Calculate FLU and FLU-π (6,7)
Aromatic fluctuation index (FLU) was proposed in J. Chem. Phys., 122, 014109 (2005), also
see Chem. Rev., 105, 3911 (2005) for more discussion. Like PDI, FLU is an aromaticity index based
on DI, but can be used to study rings with any number of atoms. The FLU index was constructed by
following the HOMA philosophy (see Section 3.100.13), i.e. measuring divergences (DI differences
for each single pair bonded) from aromatic molecules chosen as a reference. FLU is defined as
below
 2
1 ring  V ( B)    ( A, B) −  ref ( A, B) 
FLU =    
n A− B  V ( A)    ref ( A, B) 
where the summation runs over all adjacent pairs of atoms around the ring, n is equal to the number
of atoms in the ring, ref is the reference DI value, which is precalculated parameter.  is used to
ensure the ratio of atomic valences is greater than one
 1 V ( B )  V ( A)
=
− 1 V ( B )  V ( A)
The first factor in the formula of FLU penalizes those with highly localized electrons, while
the second factor measures the relative divergence with respect to a typical aromatic system.
Obviously, lower FLU corresponds to stronger aromaticity.
The dependence on reference value is one of main weakness of FLU. The default ref in
Multiwfn for C-C, C-N, B-N are 1.468, 1.566 and 1.260 respectively, they are obtained from
calculation of benzene, pyridine and borazine respectively under HF/6-31G* (geometry is optimized
at the same level. Becke's atomic space with modified CSD radii and with sharpness parameter k=3
is used to derive ref). Users can modify or add ref through option -4.
176
3 Functions
The original paper of FLU also defined FLU-π, which is based on DI-π and π-atomic valence
 2
1 ring  V ( B)    ( A, B) −  avg  
FLU =      
n A− B  V ( A)    avg  

where π is the average value of the DI-π for the bonded atom pairs in the ring, and the other symbols
denote the aforementioned quantities calculated using π-orbitals only. The advantage of FLU-π over
FLU is that FLU-π does not rely on predefined reference DI value, while the disadvantage is that
FLU-π can only be exactly calculated for planar molecules.
Akin to FLU, the lower the FLU-π, the stronger aromatic the ring. If FLU-π is equal to zero,
that means DI-π is completely equalized in the ring. The reasonableness to measure aromaticity by
FLU-π is that aromaticity for most aromatic molecules are almost purely contributed by π electrons,
rather than σ electrons.
In fuzzy atomic space analysis module of Multiwfn, PDI, FLU and FLU-π are calculated in
fuzzy atomic spaces. In J. Phys. Chem. A, 110, 5108 (2006), the authors showed that the correlation
between the PDI, FLU and FLU-π calculated in fuzzy atomic space and the ones calculated in AIM
atomic space is excellent.
In Multiwfn, before calculating FLU and FLU-π, AOM will be calculated automatically. If you
are calculating FLU-π, you will be prompted to input the indices of π orbitals, you can find out their
indices by checking isosurface of all orbitals by main function 0. Then DI or DI-π matrix will be
generated. Next, you should input the indices of the atoms in the ring, the input order must be
consistent with atom connectivity. Besides FLU or FLU-π value, the contributions from each bonded
atom pair are outputted too.
FLU and FLU-π are only available for closed-shell system in Multiwfn. It is not well known
whether FLU and FLU-π are also applicable for open-shell systems.
3.18.8 Calculate condensed linear response kernel (9)
Linear response kernel (LRK) is an important concept defined in DFT framework, which can
be written as
  2E    (r1 ) 
 (r1 , r2 ) =   =  
  (r1 ) (r2 )  N   (r2 )  N
This quantity reflects the impact of the perturbation of external potential at r2 on the electron density
at r1, which may also be regarded as the magnitude coupling between electron at r1 and r2.
In Multiwfn, LRK is evaluated by an approximation form based on second-order perturbation
theory (see Eq.3 of Phys. Chem. Chem. Phys., 14, 3960 (2012))
 i* (r1 ) j (r1 ) *j (r2 ) i (r2 )

 (r1 , r2 )  4  
where φ is molecular orbital, ε stands for MO energy. Note that this approximation form is only
applicable to HF/DFT closed-shell systems, therefore present function only works for HF/DFT
177
3 Functions
closed-shell systems.
Condensed linear response kernel (CLRK) is calculated as
Sij ( A) S ji ( B)
 A,B =    (r1 , r2 ) d r1 d r2 = 4  
A B
where A and S(A) denote fuzzy atomic space and atomic overlap matrix for atom A, similar for atom
B. In Phys. Chem. Chem. Phys., 15, 2882 (2013), it was shown that CLRK is useful for investigation
of aromaticity and anti-aromaticity.
Present function is used to calculate CLRK between all atom pairs in current system, and the
result will be ouputted as a matrix. Due to evaluation of LRK requires virtual MO information, in
current version .mwfn/.fch/.molden/.gms file must be used as input file.
Note that CLRK can be decomposed to orbital contribution, e.g. for MO i
Sij ( A) S ji ( B )
 A( i,)B = 4 
jvir i −  j
For instances, assume that you want to evaluate the contribution from MO 3,4,7, then before
calculating CLRK, you should enter main function 6 and use option 26 to set occupation number of
all MOs except for 3,4,7 to zero. (Note that the virtual MOs used to calculate LRK will automatically
still be the original virtual MOs, rather than the ones after modification of MO occupation numbers.)
3.18.9 Calculate para linear response index (10)
The definition of para linear response index (PLR) has an analogy to PDI, the only difference
is that DI is replaced by CLRK
1,4 +  2,5 +  3,6

PLR( A, B) =
3
In Phys. Chem. Chem. Phys., 14, 3960 (2012), the authors argued that PLR is as useful as PDI
in quantitatively measuring aromaticity, and it is found that the linear relationship between PLR and
PDI is as high as R2=0.96.
Present function is used to calculate PLR. Multiwfn will first calculate CLRK, and then you
should input the indices of the atoms consituting the ring in quesiton, e.g. 3,5,6,7,9,2. The input
order must be consistent with atom connectivity. Then PLR will be immediately outputted on screen.
PLR is only applicable to HF/DFT closed-shell systems, and currently .mwfn/.fch/.molden/.gms
must be used as input file.
Note that for completely planar systems, PLR can be exactly separated as PLR- and PLR-π
to individually study  aromaticity and π aromaticity. In order to calculate PLR- (PLR-π), before
enter present module, you should first manually set occupation number of all MOs except for π ()
MOs to zero (or you can utilize option 22 in main function 100 to do this step, which will be much
more convenient).
178
3 Functions
3.18.10 Calculate multi-center delocalization index (11)
n-center multi-center DI is calculated as

 ( A, B, C...H ) = 2n−1   Sij ( A) S jk ( B) Skl (C )  S qi ( H )
i j k q
where i, j, k... only cycle occupied orbitals. The normalized form of multi-center DI is defined as
1/n, and may be compared between rings with different number of members.
Currently this function is only available for single-determinant closed-shell wavefunctions, and
supports up to 10 centers. Note that for relatively large size of systems, calculating multi-center DI
for more than 6 centers may be quite time-consuming.
3.18.11 Calculate information-theoretic aromaticity index (12)
In ACS Omega, 3, 18370 (2018) it is shown that arithmetic mean of some information-theoretic
quantities of the atoms constituting a ring has good linear relationship with other widely accepted
aromaticity indices, such as HOMA and aromatic stabilization energy (ASE). It is thus clear that the
arithmetic mean may be used as index for measuring aromaticity, although this point needs to be
further explored.
The information-theoretic aromaticity index, namely the above-mentioned arithmetic mean can
be calculated via subfunction 12 of fuzzy analysis module. After entering this function, you should
choose the way of defining atomic information-theoretic quantity, three choices are currently
available:

Atomic Shannon entropy: sS ( A) = −  (r ) ln  (r ) wA (r ) d r

Atomic Fisher information: iF ( A) = |  (r ) |2 /  (r ) wA (r ) d r

Atomic GBP entropy: sGBP ( A) = (3 / 2)  (r ){ + ln[t (r ) / tTF (r )]}wA (r ) d r
Essentially, the three quantities correspond to the integral of user-defined functions 50, 51 and 54 in
fuzzy atomic space. In this function, you also need to input the index of the atoms in the ring. Once
calculation of the selected quantity for all atoms in the ring is finished, the average will be shown,
and it can be regarded as an aromaticity index.
Before using this function, you can firstly select the way of defining atomic space. In the
original paper, Hirshfeld partition was employed, while the default partition method of the fuzzy
analysis module is Becke.
Information needed by fuzzy analysis module: GTFs, atom coordinates
179
3 Functions
3.19 Charge decomposition analysis and plotting orbital

interaction diagram (16)
3.19.1 Theory
The charge decomposition analysis (CDA) proposed by Dapprich and Frenking (J. Phys.
Chem., 99, 9352 (1995)) is used to provide deep insight on how charges are transferred between
fragments in a complex to achieve charge equilibrium. The idea of CDA is based on fragment orbital
(FO), which denotes the molecular orbital (MO) of fragment in its isolated state. Besides, once the
compositions of FOs in MOs of complex are obtained, the orbital interaction diagram can be directly
plotted, which allows one visually and directly understand how orbitals of fragments are mixed to
form orbitals of complex.
For simplicity, in this section we assume that the complex consists of only two fragments. The
CDA can also be straightforwardly employed for more than two fragments cases.
Fragment orbitals
Consider we are studying a complex AB, NA basis functions are located in the atoms of
fragment A, NB basis functions in fragment B, then each MO of complex will be linearly expanded
by NA+NB basis functions, and meanwhile, the complex has NA+NB MOs. By using the same basis
set, and maintaining the same geometry as in complex, if we calculate the two fragments respectively,
we can obtain NA MOs of fragment A, and NB MOs of fragment B, they are collectively called as
fragment orbital (FO). We can take these FOs as basis functions to linearly expand the MOs of
complex. Since the dimension (the number of basis functions) is still NA+NB, the expansion is exact.
In other words, we equivalently transformed the basis.
Charge decomposition analysis

In the original paper of CDA, the authors defined three terms:
occ vir
di =  C
mA nB
i m ,i C n ,i S m ,n
vir occ
bi =  C
mA nB
i m ,i C n ,i S m ,n
occ occ
ri =  C
mA nB
i m ,i Cn ,i S m ,n
where i and η are index and occupation number of MO of complex, respectively.
S m,n =   m (r ) n (r )dr is overlap integral between FO m and FO n. Note that though the NA and
NB FOs are respectively orthonormalization sets, the NA set are in common not normal to the NB
set, so S is not an identity matrix. Cm,i denotes the coefficient of FO m in MO i of complex. The
superscript "vir" and "occ" mean virtual (viz. unoccupied) and occupied, respectively.
The term di denotes the amount of electron donated from fragment A to B via MO i of complex;
similarly, the term bi denotes the electron back donated from B to A. In fact, iCm,iCn,i Sm,n can
be regarded as the half of overlap population between FO m and n in MO i. Hence, the difference
180
3 Functions
between term d and b is that which fragment provides its electrons from its occupied FOs to virtual
FOs of another fragment. The term r reveals closed-shell interaction between two occupied FOs in
different fragments; positive value of ri means that owing to MO i, the electrons of the two fragments
are accumulated in their overlap region and shows bonding character, while negative value indicates
that the electrons are depleted from the overlap region and thus reflecting electron repulsive effect.
The sum of all ri terms is in general negative, because overall interaction between filled orbitals are
generally repulsive. r is also known as "repulsion polarization" term
Beware that although the CDA formulae given in original paper are correct, by carefully
inspecting the data, I found the d, b and r terms in the examples presented in the original paper are
erroneous (the data should be divided by two).
Generalization of CDA
The original definition of CDA has two drawbacks. First, it is only applicable to closed-shell
cases (namely, complex and each fragment must be closed-shell) and hence unable to be used when
the two fragments are bound by covalent bonding. Second, in post-HF calculations, though the MOs
of complex can be replaced by natural orbitals (NOs), the FOs can only be produced by HF or DFT
calculation, because occupation numbers of FOs are not explicitly considered in the original CDA
formulae.
To address the limitations of the original definition, in my paper J. Adv. Phys. Chem., 4, 111-
124 (2015) (http://dx.doi.org/10.12677/JAPC.2015.44013) I proposed a generalized form of CDA,
which is the form used in CDA module of Multiwfn:
 mFO −  nFO
ti = i
mA nB  ref
Cm ,i Cn ,i S m ,n
min( mFO , nFO )

ri =  2
mA nB  ref
i Cm,i Cn ,i S m,n
In the generalized CDA, orbitals of complex and fragments can be produced either by HF/DFT or
by post-HF method, corresponding to MOs and NOs, respectively. mFO stands for occupation
number of FO m. For open-shell cases, ηref is 1.0, CDA will be performed for alpha spin and beta
spin separately; for the former, i denotes alpha orbital of complex, m and n run over all alpha FOs;
for the latter, i denotes beta orbital of complex, m and n run over all beta FOs. For closed-shell cases,
ηref is 2.0, m and n run over space orbitals. min() is the function used to extract minimum from two
values. During calculation of t, if the values are only accumulated for the cases mFO  nFO , then
the resulting t is d; if only for the cases mFO  nFO , then t will be b.
For the situations when the original CDA is applicable, the b and d calculated by the generalized
form are exactly identical to the ones obtained via original definition; while r will be exactly twice
of the one produced via original definition. The reason why the factor 2 is introduced into the
generalized form of r is because after doing so, r has more clear physical meaning, namely it equals
to overlap population (also known as Mulliken bond order) between the occupied FOs in the two
fragments.
Because the CDA has been generalized, below, FO will stand for MO or NO of fragment,
"complex orbital" will denote MO or NO of complex. The orbitals can either be spin-space orbital
181
3 Functions
(for open-shell) or space orbital (for closed-shell).

It is clear that the d, b and r terms can be decomposed into FO pair contributions, this kind of
decomposition is supported by Multiwfn and greatly faciliates analysis of the nature of FO
interaction.
Composition of FO in complex orbital and orbital interaction diagram

By using the methods discussed in Section 3.10, the composition of FO in complex orbital can
be calculated. In common, Mulliken method is the best choice for this purpose, the composition of
FO m in complex orbital i is calculated as
 
m,i =  Cm2 ,i +  Cm,i Cn ,i Sm,n   100%
 n m 
Although using SCPA method to calculate composition is more convenient (since overlap matrix is
not involved), if i is a high-lying virtual orbital, the calculated composition is not very reasonable.
Note that due to a well-known drawback of Mulliken method, sometimes negative contributions
occur, since the values are often not large, you can simply ignored them.
From Θ, we can clearly understand how each complex orbital is formed by mixing FOs of the
two fragments. Furthermore, one can plot orbital interaction diagram to visually and intuitively
study the relationship between complex orbitals and FOs, namely plot a bar for each complex orbital
and FO according to its energy, and then check each Θ to determine how to link the bars, e.g. if the
value of Θm,i is larger than 5%, then the two bars corresponding to FO m and complex orbital i will
be linked. Consequently, from the graph we will directly know that FO m has important contribution
to complex orbital i.
Extended charge decomposition analysis (ECDA)

The difference between the total number of donation and back donation electrons, that is d - b,
may be regarded as the net transferred electrons. However, in J. Am. Chem. Soc., 128, 278 (2006),
the authors argued that this viewpoint is not correct, because b and d terms not only represent charge
transfer effect (CT), but also electron polarization effect (PL); the latter describes the adjustment of
electron distribution within the fragment, which is caused by mixing virtual and occupied FOs of
the same fragment during formation of the complex, and should be excluded in the calculation of
the amount of net transferred electrons. In this paper they proposed extended charge decomposition
analysis (ECDA) method, by which they argued that the number of net transferred electrons can be
calculated more reasonably.
In ECDA viewpoint, four terms can be defined as follows
1. PL(A) + CT(A→B) = The sum of compositions of occupied FOs of fragment A in all virtual
orbitals of complex, multiplied by Occ
2. PL(A) + CT(B→A) = The sum of compositions of virtual FOs of fragment A in all occupied
3. PL(B) + CT(B→A) = The sum of compositions of occupied FOs of fragment B in all virtual
4. PL(B) + CT(A→B) = The sum of compositions of virtual FOs of fragment B in all occupied
where Occ is 1.0 and 2.0 for open-shell and closed-shell cases, respectively.
After the four terms are calculated, the number of net transferred electrons from A to B can be
directly obtained as
182
3 Functions
CT(A→B) - CT(B→A) = [ PL(A) + CT(A→B) ] - [ PL(A) + CT(B→A) ]

The composition of complex orbitals can be calculated by various methods, leading to different
ECDA result. The method used in Multiwfn is identical to the one in ECDA original paper, namely
Mulliken method.
Note that ECDA can neither be applied to post-HF calculation nor the system consisting more
than two fragments.
According to the name, ECDA is as if an extension of CDA, however in my personal opinion,
ECDA is irrelevant to CDA, their basic ideas are quite different, and thus the amount of net
transferred electron calculated by ECDA is not comparable with the d - b produced by CDA at all.
In addition, though ECDA is realized in Multiwfn, I do not think this is a useful method. The most
remarkable feature of CDA is that the electron transfer can be decomposed to contribution of
complex orbitals, however ECDA is incapable to do this; ECDA can only reveal how many electrons
is transferred between two fragments, but this quantity actually can be obtained by a more
straightforward approach, namely calculating the fragment charge by summing up all atomic
charges in the fragment, and then subtracting it by the net charge of the fragment in its isolated state.
3.19.2 Input file
CDA analysis can be carried out as long as you have file containing basis function information
for complex and all fragments. As described in Section 2.5, .mwfn, .fch, .gms, .molden can be used
as input file in this case.
Below requirements on input files should be noticed:
(1) The method and basis set employed in the calculation of complex and fragments must be
the same, otherwise the result will be meaningless. For example, assume that you are preparing input
files of CDA analysis for a transition metal complex, if mixed basis set is employed for complex
calculation (e.g. Lanl2TZ for metal and 6-31G* for ligands), then in the fragment calculations,
Lanl2TZ must be used for the metal fragment, and 6-31G* must be used for the ligand fragment(s).
(2) The coordinate of each fragment in their wavefunction files must be exactly identical to the
coordinate in the complex wavefunction file.
In order to guarantee this, it is best to first optimize the complex (and meantime obtain its wavefunction file),
and then directly extract coordinates of various fragments from the optimized complex geometry and then write them
as individual input files of single point task (after calculating them you will obtain wavefunction files of the
fragments). Clearly, the fragments should never be optimized, otherwise their coordinates will become inconsistent
with the complex coordinate.
(3) The sequence of atoms in fragments and in complex must be identical, that means the actual
atom sequence in the complex can be retrieved by successively combining the atoms in fragment 1,
2, 3 ...
(4) Avoid employing diffusion functions whenever possible! According to my experiences,
when diffusion basis functions are presented, the CDA results are often unreasonable or even
completely meaningless.
"examples\CDA\COBH3_ORCA" folder contains example .molden files generated by ORCA

3.0.1 for performing CDA analysis of COBH3 system.
Special notes for Gaussian users

For Gaussian users, below points should be noted.
• Since Gaussian automatically puts the system in standard orientation, in order to satisfy the
above requirement (2), nosymm keyword should be used in the calculation to avoid this treatment.
183
3 Functions
• If you intent to use mixed basis set, and in particular Pople type basis set is involved (such as
6-31G*), it is best to specify 5d keyword in the fragment calculations to force Gaussian to use
spherical-harmonic type of basis functions for d shell.
• By default Gaussian automatically eliminates linearly dependant basis functions, hence in
some cases (usually when diffuse functions are employed), the number of basis functions is not
equal to the number of orbitals, in this case CDA cannot be performed. Using IOp(3/32=2) can
avoid this problem.
• If you employed post-HF for complex and fragment calculations, you need to follow the steps
described at the beginning of Chapter 4 to save natural orbitals into .fch file.
Gaussian output file of single point task can also be used for CDA analysis. nosymm pop=full
must be employed in all cases, and IOp(3/33=1) should also be specified in the complex calculation.
Molecular geometry must be given in Cartesian coordinate. If you want to perform CDA analysis at
post-HF level, do not use pop=full but use density pop=NO and density pop=NOAB for closed-shell
and open-shell cases, respectively, so that coefficients of natural orbitals can be outputted.
3.19.3 Usage
After booting up Multiwfn, you should first input the path of the file of complex, and then enter
the CDA module. Next, you should set the number of fragments, and then input the path of the file
for each fragment in turn.
For open-shell fragments, you will be prompted to choose if flipping its electron spin. If you
select "y", then orbital information of its alpha and beta orbitals will be exchanged. The reason for
introduction of this step is clear: for example, we want to use CDA to decompose electron transfer
between fragment CH3 and NH2 in CH3NH2; CH3NH2 has 9 alpha and 9 beta electrons. However,
during calculation via quantum chemistry codes, both CH3 and NH2 will be regarded as having 5
alpha and 4 beta electrons. Therefore, in order to carry out CDA, we have to flip electron spin of
either CH3 or NH2, otherwise the total numbers of alpha and beta electrons in the two fragments,
namely 5+5=10 and 4+4=8, respectively, will be unequal to those of the complex, namely 9 and 9.
Once the loading is finished, Multiwfn starts to calculate some data. If only two fragments are
defined, CDA and ECDA result will be directly shown. Then you will see a menu:
-2 Switch output destination (for options 0 and 1): By default the options 0 and 1 output
results on screen; if you select this option once, then their results will be outputted to CDA.txt in
current folder.
0 Print CDA result and ECDA result: Input the index of two fragments, then the CDA and
ECDA analysis result between them will be outputted.
1 Print full CDA result: If you select this option, the CDA result for all complex orbitals will
be outputted. By default, Multiwfn does not output CDA result for the complex orbitals lying higher
than LUMO+5, because the number of such orbitals is too large, whereas their contributions to d, b
and r terms are often completely negligible due to their occupation numbers are often quite small.
(For HF/DFT wavefunctions, the occupation numbers of the orbitals lying higher than HOMO are
exactly zero, and thus have no contribution to d, b and r terms at all. So for this case this option is
meaningless)
2 Show fragment orbital contributions to specific complex orbital: If you input x, then the
composition of complex orbital x will be outputted (for open-shell cases, the xth alpha and the xth
184
3 Functions
beta complex orbital will be outputted respectively). By default only the FOs having contribution
1% will be shown, but this threshold can be altered by "compthresCDA" parameter in settings.ini.
If you want to obtain composition of a fragment orbital in all complex orbitals, you can input
for example 1,6, which means orbital 6 of fragment 1 is selected.
3 Export coefficient matrix of complex orbitals in fragment orbital basis: The coefficient
matrix corresponding to all FOs in all complex orbitals will be outputted to coFO.txt in current
folder.
4 Export overlap matrix between fragment orbitals: The overlap matrix between all FOs
will be outputted to ovlpint.txt in current folder.
6 Decompose complex orbital contribution to CDA: You need to input index of a complex
orbital and set threshold for printing, if contribution of a pair of fragmental orbitals to any of d, b
and r term of this complex orbital is larger than the threshold then the contribution value will be
shown. This greatly faciliates analysis of interaction between fragment orbitals.
5 Plot orbital interaction diagram: If you select this option, you will enter a new menu, in
which by corresponding options you can plot and save orbital interaction diagram and adjust plotting
parameters, such as size of labels, energy range (namely Y-axis range) and the criterion for linking
bars. The orbital interaction diagram plotted under default settings looks like this:
In above graph, occupied and virtual orbitals are represented as solid and dashed bars,
respectively, the vertical positions are determined by their energies. The bars at left and right sides
correspond to the FOs of the two fragments you selected; the bars in the middle correspond to
complex orbitals. Orbital indices are labelled by blue texts. If two or more labels occur in the same
bar, that means these orbitals are degenerate in energy. If composition of a FO in a complex orbital
is larger than specific criterion, then the corresponding two bars will be connected by red line, so
that simply by viewing the diagram one can directly understand the complex orbitals are constructed
by mainly mixing which FOs. The compositions are labelled in the center of the lines.
185
3 Functions
By default, all FOs and complex MOs are plotted, and if the contribution of a FO of fragment
A or B to a complex MO is larger than 10% then they will be connected. For large systems, usually
there are too many bars and linking lines in the diagram, and it is hence difficult to identify the
orbital interaction mode based on the diagram. In these cases, you should properly use the option "4
Set the rule for connecting and drawing orbital bars" to manually set up the rule for connecting and
plotting the orbital bars. See the prompt shown on the screen on how to use this option.
Sometimes the difference between orbital energies of the two fragments in the orbital
interaction diagram is too large and thus hinders one to analyze the diagram, so you may want to
equalize their energies. In this case, you can use the option "12 Set orbital energy shifting value" to
set the value used to shifting orbital energies of the complex or the two fragments presented at the
left and right side of the diagram; their orbital energies will be agumented by the given value.
Some notes
If complex or any fragment is an open-shell system, then CDA will be performed separately
for alpha and beta electrons.
Infinite number of fragments can defined; however, if more than two fragments are defined,
ECDA analysis will be unavailable.
ECDA analysis and orbital interaction diagram are not available if complex or any fragment
was calculated by multiconfiguration method (e.g. MP2, double-hybrid functional), since orbital
energy of natural orbitals is not undefined.
Some examples of CDA are given in Section 4.16.
3.20 Basin analysis (17)
3.20.1 Theory
The concept of basin was first introduced by Bader in his atom in molecular (AIM) theory,
after that, this concept was transplant to the analysis of ELF by Savin and Silvi. In fact, basin can
be defined for any real space function, such as molecular orbital, electron density difference,
electrostatic potential and even Fukui function.
A real space function in general has one or more maxima, which are referred to as attractors or
(3,-3) critical points. Each basin is a subspace of the whole space, and uniquely contains an attractor.
The basins are separated with each other by interbasin surfaces, which are essentially the zero-flux
surface of the real space functions; mathmatically, such surfaces consist of all points {r} satisfying
f (r)  n(r) = 0 , where n(r) stands for the unit normal vector of the surface at position r.
Below figure illustrates the basins of total electron density of NH2COH, such figure is often
involved in AIM analysis. Brown points are attractors; in this case they correspond to maxima of
total electron density and thus are very close to nuclei. Blue bolded lines correspond to interbasin
surfaces, they dissect the whole molecular space into basins, which in present case are also known
as AIM atomic space or AIM basin. Grey lines are gradient paths of total electron density, they are
emanated from attractors.
186
3 Functions
Multiwfn is able to generate basins for any supported real space functions, such as electron
density, ELF, LOL, electrostatic potential. Alternatively, one can let Multiwfn load the grid file
generated by third-part programs (e.g. cube file or .grd file) and then generate basins for the real
space function recorded in the grid data.
After the basins were generated, some analyses based on the basins can be conducted to extract
information of chemical interest. The three most useful types of basin analyses are supported by
Multiwfn and are briefed below. All of them require performing integration in the basin.
(1) Study integral of a real space function in the basins. For example, one can first use electron
density to define the basins, and then integrate electron density in the basins to acquire electron
population numbers in the basins; this quantity is known as AIM atomic population number. Another
example, one can first partition the space into basins according to ELF function, and then calculate
integral of spin density in the basins.
(2) Study electric multipole moments in the basins. In Multiwfn electric monopole, dipole and
quadrupole moments can be calculated in the basins. This is very useful to gain a deep and
quantitative insight on how the electrons are distributed in different regions of a system. Based on
these data, one can clearly understand such as how lone pair electrons contribute to molecular dipole
moment (see J. Comput. Chem., 29, 1440 (2008) for example, in which ELF basins are used), how
the electron density around an atom is polarized as another molecule approaching.
(3) Study localization index and delocalization index in the basins. The former one measures
how many electrons are localized in a basin in average, while the latter one is a quantitative measure
of the number of electrons delocalized (or say shared) between two basins. They have been very
detailedly introduced in Section 3.18.5 and hence will not be reintroduced here. The only difference
relative to Section 3.18.5 is that the range of integration in our current consideration is basins rather
than fuzzy atomic spaces.
3.20.2 Numerical aspects
This section I will talk about many numerical aspects involved in basin analysis, so that you
can understand how the basin analysis module of Multiwfn works.
Algorithms for generating basins

The algorithms used to generate/integrate basins can be classified into three categories:
187
3 Functions
(1) Analytical methods. Due to the popularity and importance of AIM theory, and the complex
shape of AIM basins, since long time ago numerous methods have been proposed to attempt to speed
up integration efficiency and increase integration accuracy for AIM basins, e.g. J. Comput. Chem.,
21, 1040 (2000) and J. Phys. Chem. A, 115, 13169 (2011). All of these methods are analytical, and
they are employed by almost all old or classical AIM programs, such as AIMPAC, AIM2000,
Morphy and AIMAll. This class of methods suffers many serious disadvantages: 1. The algorithm
is very complicated 2. High computational cost 3. Only applicable to the analysis of total electron
density 4. Before generating basins, (3,-3) and (3,-1) critical points must be first located in some
ways. The only advantage of these methods is that the integration accuracy is very satisfactory.
(2) Grid-based methods. This class of methods relies on cubic (or rectangle) grid data. Since
the publication of TopMoD program, which aims to analyze ELF basin and for the first time uses a
grid-based methods, grid-based methods continue to be proposed and incurred more and more
attention. Grid-based methods solved all of the drawbacks in analytical methods; they are easy to
be coded, the computational cost is relative low, and more important, they are suitable for any type
of real space function. Unfortunately these methods are not free of shortcomings, the central one is
that the integration accuracy is highly dependent on the quality of grid data; to obtain a high accuracy
of integral value the grid spacing must be small enough.
(3) Mixed analytical and grid-based method. The only instance in this class of method is
described in J. Comput. Chem., 30, 1082 (2009), in which Becke's multicenter integration scheme
is used to integrate basins. This method is faster than analytical method, and more accurate than
grid-based method, but it is only suitable to analyze total electron density, thus the application range
is seriously limited. Since the basin analysis module of Multiwfn focuses on universality, this
method is not currently implemented.
Currently the best grid-based method is near-grid method (J. Phys.: Condens. Matter, 21,
084204 (2009)), which is employed in Multiwfn as default method to generate basins. Multiwfn
also supports on-grid method (Comput .Mat. Sci., 36, 354 (2006)), which is the predecessor of near-
grid method.
Basic steps of generating basins and locating attractors in Multiwfn

In Multiwfn, generating basin and locating attractor are carried out simultaneously; the basic
process can be summarized as follows: Given a grid data, all grids (except for the grids at box
boundary) are cycled in turn. From each grid, a trajectory is evolved continuously in the direction
of maximal gradient. Each step of movement is restricted to its neighbour grid, and the gradient in
each grid is evaluated via one direction finite difference by using the function values of it and its
neighbour grids. There are four circumstances to terminate the evolvement of present trajectory and
then cycle the next grid (a) The trajectory reached a grid, in which the gradients in all directions is
equal or less than zero; such a grid will be regarded as an attractor, meanwhile all grids contained
in the trajectory will be assigned to this attractor. (b) The trajectory reached a grid that has already
been assigned; all of the grids contained in the trajectory will be assigned to the same attractor. This
treatment greatly reduced computational expense. (c) The upper limit of step number is reached (d)
The trajectory reached the grids at box boundary; all of the grids in the trajectory will be marked as
"travelled to boundary" status.
After all grids have been cycled, we obtained position of all attractors covered by the spatial
scope of the grid data. Each set of grids that assigned to a same attractor collectively constitutes the
basin corresponding to the attractor. Interbasin grids will then be detected, according to the criterion
188
3 Functions
that if any neighbour grid belongs to different attractor. Note that sometimes after finished above
processes, some grids may remain unassigned, and some grids may have the status "travelled to
boundary". Such grids are often far away from atoms and thus unimportant, in general you can
safely ignore them.
The only difference between on-grid method and near-grid method is that in the latter one, the
so-called "correction step" is introduced to continuously calibrate the evolvement direction of the
trajectory, which greatly eliminates the artificiality problem of interbasin surfaces due to on-grid
method, and in turn makes the basin integration more close to actual values. Near-grid method can
be supplemented by a refinement step for interbasin surfaces at the final stage, so that the partition
of basin can be even more accurate, this additional step is not very time-consuming.
For ELF (and may be other real space functions), if sphere- or ring-shape attractors exist in the
system, in corresponding regions a large number of attractors having basically the same value will
be found. Below two examples illustrated the existence of the ring-shape ELF attractor encircling
N-C bond of HCN and the sphere-shape ELF attractor encompassing the Ar atom.
Based on nearest neighbour algorithm, Multiwfn automatically checks the distances and the relative
value differences between all pairs of located attractors, and then clusters some attractors together.
The resulting attractors thus have multiple member attractors, and will be referred to as "degenerate
attractors" later. For example, the left graph and right graph given below displayed the attractor
indices before and after clustering, respectively. As you can see, after clustering, all of the attractors
constituting the ring-shape ELF attractor now have the same index, suggesting that the
corresponding basins have been merged together as basin #2.
For certain cases, in the regions far beyond the systems, some artifical or physically
meaningless attractors are located. These attractors have very low function value and thus are called
as "insignificant" attractors; in constrast, other attractors can be called as "significant" attractors.
The presence of insignificant attractors is because in corresponding regions the function values are
rather small, hence the function behavior is not very definitive, even a very slight fluctuation of
function value or trivial numerical perturbation is enough to result in new attractors. When
insignificant attractors are detected, Multiwfn will prompt you to select how to deal with them, there
are three options: (1) Do nothing, namely do not remove these attractors (2) Remove these attractors
meanwhile set corresponding grids as unassigned grids (3) Remove these attractors meanwhile
assign corresponding grids to the nearest significant attractors. The third option is generally
recommended.
189
3 Functions
If the real space function simultaneously possesses positive and negative parts, e.g. electrostatic
potential, Multiwfn will automatically invert the sign of negative part of the grid data and then locate
attractors and generate basins as usual, after that the sign will be retrieved. Therefore, for negative
parts, the "attractors" located by Multiwfn actually are "repulsors".
Integration of basins
Once the basins have been generated by grid-based method, the integral of a real space function
f in a basin P can be readily evaluated as F =  f (r )  dV , where i is grid index, dV is volume
iP
i
differential element. Evidently, the finer the grid data (smaller grid spacing) is used, the more
accurate the integral values will be, and correspondingly, the more computational effort must be
paid.
Since the arrangement of grids in general is not in coincident with system symmetry, one should
not expect that the integral values always satisfy system symmetry well, unless very fine grid is
used.
Note that different functions have different requirement on the quality of grid data. For the
same grid spacing, the faster the function varies, the lower the integration accuracy will be. Due to
Laplacian of electron density, source function and kinetic density function etc. fluctuate violently
near nuclei, it is impossible to directly integrate these functions in AIM basins solely by grid-based
method at satisfactory accuracy (since only uniform grids are used, which is incapable to represent
nuclear region well enough). In order to tackle this difficulty, an integration method based on mixed
atomic-center and uniform grids is supported by Multiwfn, in which the regions close to nuclei are
integrated by atomic-center grids, while the other regions are integrated by uniform grids. This
method is able to integrate AIM basin for any real space function, including the ones having
complicated behavior at generally acceptable accuracy without evident additional computational
cost with respect to grid-based method. In order to further improve the integration accuracy for AIM
basins, Multiwfn also enables one to exactly refine the assignment of the grids at basin boundary.
An exact steepest ascent trajectory is emitted from each boundary grid, and terminates when reaches
an attractor. Since there are often very large number of boundary grids and each steepest ascent step
requires evaluating gradient of electron density once, this process is time-consuming. In Multiwfn
these gradients can also be approximately evaluated by trilinear interpolation based on pre-
calculated grid data of gradient, 1/3~1/2 of overall time cost of the AIM basin integration can be
saved. Note that if the mixed type of atomic-center and uniform grids is employed, and especially
when the exact refinement process of boundary basin is enabled at the same time, then the
requirement on quality grid will be markedly lowered, usually accurate result can be obtained at
"Medium-quality grid" level (grid spacing=0.1 Bohr).
It is worth noting that even though basin analysis module of Multiwfn can be used to acquire
position of attractors, the accuracy is directly limited by the quality of grid data, since the attractors
will be located on a grid point. Moreover, this module is unable to be used to search other types of
critical points (attractor corresponds to (3,-3) type critical point); So if you want to obtain accurate
coordinate and value for all types of critical points, you should make use of main function 2, namely
topology analysis function.
190
3 Functions
3.20.3 Usage
The basin analysis module in Multiwfn is very powerful, fast and flexible. Basins can be
generated and visualized for any real space function, meanwhile any real space function can be
integrated in the generated basins. Electric multipole moments and localization index in the basins
and delocalization index between basins can be calculated.
Basic steps
To perform basins analysis in Multiwfn, there are five basic steps you need to do
(1) Load input file into Multiwfn and then enter main function 17.
(2) Select option 1 to generate basins and locate attractors. Before doing this, you can change
the method used to generate basins via option -1, or adjust the parameters for clustering attractors
via option -6.
(3) Visualize attractors and basins by option 0. This step is optional.
(4) Perform the analyses you want. Integral of a real space function in generated basins, electric
multipole moments in the basins and localization/delocalization index of the basins can be
calculated by option 2 (or 7), 3 (or 8) and 4, respectively.
You can export the generated basins as cube files by option -5, export attractors as .pdb file by
option -4, check the coordinate of the located attractors by option -3, measure the distances, angles
and dihedrals angles between attractors or atoms by option -2. Also you can use suboption 3 in
option -6 to manually merge specified basins as a single one. If you want to regenerate basins and
locate attractors for other real space functions or under new settings, you can select option 1 again.
About input file and grid data

Grid data must be obtained first before generating basins. In option 1 of basin analysis module,
you can let Multiwfn calculate the grid data. However, if a grid data has already been stored in
memory, you can directly choose to use it. There are two circumstances in which grid data will be
stored in memory: (1) The input file contains grid data, e.g. .cub and .grd file (2) You have used
main function 5 or 13 to calculate grid data before entering main function 17. This design makes
Multiwfn rather flexible: you can use Multiwfn to generate basins for a real space function recorded
in a .cub or .grd file, which may be outputted by a third-part program; and you can also first use
main function 5 or 17 to generate grid data for electron density difference, Fukui function, dual
descriptor and so on, and then generate basins for them.
In the function of basin integration, namely option 2, the value of integrand at each grid must
be available so that the integral can be evaluated. These values can come from three ways, you can
choose any one: (1) Let Multiwfn directly calculate them (2) Load and use the grid data in a .cub/.grd
file, note that this operation will not overwrite the grid data already stored in memory (3) Use the
grid data already stored in memory. Beware that the grid setting of the grid data stored in memory
or the grid data recorded in a .cub/.grd must be exactly in coincidence with that of the grid data used
to generate basin, otherwise the integral is meaningless.
Note that if you have made Multiwfn calculate grid data for a real space function, then this grid
data will be stored in memory and hence you can directly use it in the basin integration step.
Visualization
In option 0, you can visualize attractors and basins, for example
191
3 Functions
Most of buttons are self explanatory, please feel free to play with them. You can plot a basin
by choosing corresponding index in the basin list at right-bottom corner. By default, only the grids
at the basin boundary are shown. If you want to plot the whole basin region, "Show basin interior"
check box should be selected. If you only want to visualize the region of the basin enclosed by  =
0.001 a.u. isosurface (common definition of vdW surface), you should select "Set basin drawing
method" - "rho>0.001 region only".
At the end of the basin list, the "Unas" entry means unassigned grids during basin generation,
while the "Boun" term corresponds to the grids travelled to box boundary, you can select them to
plot corresponding grids. Some basins are very small and very close to nucleus, such as core-type
basin of ELF, they are often screened by molecular structure; if you want to inspect these basins,
you should deselect "Show molecule" check box.
Light green spheres denote attractors, and corresponding basins are colored as green. If the real
space function simultaneously has positive and negative parts, the "attractors" in negative regions
(in fact they are repulsors) will be shown as light blue spheres, and corresponding basin will be
portrayed as blue color.
If a set of attractors have been clustered as a single one, although all of them will be displayed
in the GUI, the index labels shown above them will be the same, which is the index of corresponding
degenerate attractor.
After using option 1 to locate the attractors, if you visualize isosurface by the GUI of main
function 5 or 13, you can also see the attractors and their labels. This design faciliates comparative
study of attractors and isosurfaces.
192
3 Functions
Detail of various options

All of the options involved in basin analysis module are described below; some options have
been partially discussed above.
 -6 This option controls the parameter used in attractor clustering: If the parameters in
suboption 1 and 2 are set as A and B, respectively, that means for any two attractors, if between
them the relative value difference is smaller than A, meanwhile their interval is less than
B*sqrt(dx2+dy2+dz2), where dx, dy, dz are grid spacings in X,Y,Z, then they will be clustered together.
If you want to nullify the automatic clustering step after generating basins, simply set the parameter
in suboption 2 to 0.
After basins have been generated, you can also use suboption 3 to manually cluster specified
attractors.
 -5 Export basins as cube file: The selected basins will be outputted to basinXXXX.cub files
in current folder, where XXXX is the index of the basin. Via this option, you may use other
visualization software such as VMD to render the basins by drawing them as isosurface map with
isovalue of 0.5, see corresponding description and illustration in Section 4.17.
In this option you can also input a, then all basins will be collectively exported to basin.cub in
current folder, in which grid value corresponds to the basin index that the grid belongs to.
 -4 Export attractors as pdb/pqr/txt file: Via this option, you can export all found attractors as
one of below three files:
attractors.pdb: In this file, the atom indices correspond to the attractor indices before clustering,
while the residue indices correspond to the attractor indices after clustering.
attractors.pqr: The same as pdb, however, the "charge" column in this file corresponds to
function value at the attractor
attractors.txt: Each line contains X, Y, Z coordinate (in Bohr) and function value of a attractor
 -3 Show information of attractors:: The coordinate and value of all attractors will be
outputted on screen. For degenerate attractors, the shown coordinates and values are the average
ones of their member attractors, the coordinate and the value of all of the member attractors will
also be individually shown on screen. This option also asks you if printing attractor information
after sorting the value from most negative to most positive.
 -2 Measure distances, angles and dihedral angles between attractors or atoms: As the title
says. Please follow the prompts shown on the screen.
 -1 Select the method for generating basins: Three methods are available now: (1) On-grid
method (2) near-grid method (3) near-grid method with boundary refinement step. Note that the
near-grid method used in Multiwfn has been adapted by me, and thus became more robost. In general
I recommend method 3, and this is also the default method.
 0 Visualize attractors and basins: This option has already been introduced above.
 1 Generate basins and locate attractors: Choose a real space function, and select a grid setting,
then Multiwfn will calculate grid data, and then generate basins meanwhile locate attractors. After
that, as mentioned earlier, if there are some attractors having similar value and closely placed then
they will be automatically clustered. If a grid data has been stored in memory, you can directly
choose to use this grid data to avoid calculating new grid data. You can use this option to regenerate
basins again and again.
Multiwfn provided a lot of ways to define the grid setting. If you want to study the basins in
the whole system, in commonly "medium quality grid" is recommended for qualitative analysis
193
3 Functions
purposes; while if you want to obtain higher integration accuracy, "high quality grid" is in general
recommended. If you only need to study the basins in a local region of the whole system, you can
make the spatial scope of the grid data only cover this region; for this purpose, I suggest you to use
the option "Input center coordinate, grid spacing and box length", by which you can easily define
the spatial scope of the grid data.
 2 Integrate real space functions in the basins: Select a real space function, or select the grid
data stored in memory, or choose to use external .cub/.grd file, then the real space function you
selected or the real space function recorded in the grid data will be integrated in the basins that have
been generated by option 1, the integral in each basin and basin volumes will be outputted. If there
are some unassigned grids or the grids travelled to box boundary, their integrals will also be shown.
If the basins you analyzed are AIM basins, it is highly recommended to use option 7 instead to
gain much better integration accuracy.
 3 Calculate electric multipole moments in the basins: Please consult Section 3.18.3. The
differences relative to Section 3.18.3 are that the ranges of integration in our current consideration
are basins rather than fuzzy atomic spaces, and nuclear positions should be replaced by attractor
positions. The multipole moments are outputted up to quadrupole, while octopole and higher orders
are not supported, because based on grid integration they are difficult be evaluated accurately.
If the basins you analyzed are AIM basins, it is highly recommended to use option 8 instead to
gain much better integration accuracy.
 4 Calculate localization index (LI) and delocalization index (DI) for the basins: Please
consult Section 3.18.5. The only difference relative to Section 3.18.5 is that the range of integration
in our current consideration is basins rather than fuzzy atomic spaces.
 5 Calculate and output orbital overlap matrix in all basins (also known as basin overlap
matrix, BOM) to BOM.txt in current folder: The (i,j) element of BOM of basin K is defined as
Sij ( K ) =  i (r) j (r)d r , where K denotes the region of basin K, i corresponds to occupied
K
molecular orbital i. All virtual orbitals are not taken into account.
Below options are available only when the real space function used to define the basins is
chosen as electron density
 6 Calculate and output orbital overlap matrix in all atoms (also known as atomic overlap
matrix, AOM) to AOM.txt in current folder: The definition is identical to BOM, the only difference
is that the atom index is used instead of basin index. Note that the basins corresponding to non-
nuclear attractors are not outputted.
 7 Integrate real space functions in AIM basins with mixed type of grids: This option is
specific for integrating AIM basins; there are three different ways to realize it:
Hint: Accuracy: (2)≥(3)>>(1). Time spent: (2)>(3)>>(1). Memory requirement: (3)>(2)=(1)
(1) Integrate a specific function with atomic-center + uniform grids: Atomic-center grids are
mainly used to integrate the real space function near nuclei, while uniform grids are mainly used for
other regions. The accuracy is much better than solely using uniform grids (namely option 2 in basin
analysis module).
(2) The same as (1), but with exact refinement of basin boundary: Compared to option (2),
when integrating the grids at basin boundary, the assignment of these grids will be exactly refined
by steepest ascent scheme, hence the result will be more accurate than using option (1),
unfortunately the refinement step is time consuming.
(3) The same as (2), but with approximate refinement of basin boundary: This is an
194
3 Functions
approximate version of option (2). The gradient of electron density used in refinement step will not
be evaluated exactly, but obtained by trilinear interpolation of pre-calculated grid data of gradient.
Although evaluating the grid data of gradient is also time consuming, the overall time cost is lower
than option (2), and the loss of accuracy is trival.
After you used option (2) or (3) once, the assignment of the boundary grids will be updated
permanently, that means then if you use option (1), the result will be identical to (2) or (3).
Not only the basin volumes are outputted along with the integrals, the basin volumes with
electron density > 0.001 are also outputted, which can be regarded as atomic sizes.
 8 Calculate electric multipole moments in AIM basins with mixed type of grids: Similar to
option 3, but use mixed atomic-center and uniform grids to calculate electric multipole moments to
gain better accuracy without bringing evident additional computational cost. This option is only
applicable to AIM basins.
When using options 7 and 8, if effective core potential (ECP) is used for an atom and meantime you did not
allow Multiwfn to supply EDF information (see "isupplyEDF" parameter in settings.ini), then many attractors will
occur at valence region of the atom; hence before entering options 7 and 8, you have to manually cluster these
attractors into one by suboption 3 in option -6.
 9 Obtain atomic contribution to population of external basins: The external basins means the
basin defined by a cube file named basin.cub in current folder, in which the grid value corresponds
to basin index. This option aims to obtain atomic contribution to population of ELF bond basins (or
other type of basins) based on AIM partition. Please check Section 4.17.7 for example.
 10 Calculate high ELF localization domain population and volume (HELP, HELV): This
option is used to calculate the HELP and HELV, which were defined in ChemPhysChem, 14, 3714
(2013) to characterize lone pair electrons. This option appears only when the function used to
partition basin is ELF. See Section 4.17.8 for example on using this option to calculate HELP and
HELV.
 11 Calculate orbital compositions contributed by various basins: Via this option, you can
calculate contribution to specific orbital contributed by AIM basins or other kinds of basins, such as
ELF basins. See Section 4.8.6 for example.
For options 3, 4, 5, 7 and 8, if the input file you used does not contain GTF information
(e.g. .cub file is used as input file and you directly use the grid data carried by it to generate basins),
then Multiwfn will prompt you to input a new file, which should contain GTF information of present
system, you can use for example mwfn/.wfn/.wfx/.fch/.molden/.gms as input.
Many examples of this module can be found in Section 4.18.
Information needed: GTFs or grid data loaded from external file (e.g. .cub), atom coordinates
3.21 Electron excitation analysis (18)
3.21.A Basic information about electron excitation analysis module
1 Overview
Main function 18 contains a lot of subfunctions aiming for electron excitation analysis, namely
195
3 Functions
characterizing the electron excitation in various ways. All functions in this category fully support
single-reference methods (i.e. reference wavefunction for generating excited state wavefunction is
single Slater-determinant), including TDDFT, TDA-DFT, CIS and TDHF, while ZINDO is also
supported by transition density matrix plotting function. Other kinds of methods for excited state
problems such as EOM-CCSD, LR-CC2/3, CASSCF, CASPT2 and MRCI are not formally
supported. Examples of some of these electron excitation analysis functions are provided in Section
4.18.
Both closed-shell and open-shell systems are fully supported by all kinds of electron excitation
analyses of Multiwfn.
2 Basic knowledges about single-reference methods

Excited state wavefunction (exc) of CIS and TDA-DFT methods can be represented as
occ vir
 exc =  wia  ia   wia  ia
i →a i a
where i and a respectively run over all occupied and all virtual MOs, similarly hereafterin. 𝑎𝑖 is
the configuration state wavefunction corresponding to moving an electron from originally occupied
MO i to virtual MO a. w is known as configuration coefficient. The electron excitation in CIS or
TDA-DFT framework therefore can be represented as linear combination of orbital pair transitions.
The weighting coefficients w satisfy this normalization condition:
(w )
i →a
a 2
i = 1.0
Clearly, the i→a orbital pair transition has contribution of 100% × (𝑤𝑖𝑎 )2 to the electron
excitation.
While for TDHF and TDDFT, excited state wavefunction also contains so-called de-excitation
part:
exc =  wia  ia +  wia  ia
i →a i a
where w and w' correspond to configuration coefficient of excitation and de-excitation, respectively.
In this cases, the normalization condition becomes:
( w ) − ( w )
i →a
a 2
i
i a
i
a 2
= 1.0
The MOs used for CIS/TDHF and TDA-DFT/TDDFT are yielded by HF and DFT calculation
for ground state of present system, respectively. The Slater determinant consisted of the occupied
MOs, namely the ground state wavefunction, is known as reference state. If the reference state is
closed-shell, then  and  MOs are exactly matched with each other, and thus → orbital
transitions have one-to-one correspondence with → orbital transitions; in this situation, only one
set of orbital transition is recorded, and correspondingly, the configuration coefficients are
normalized to 0.5 instead of 1.0.
3 Input files
Input file of almost all electron excitation analysis functions are basically the same, except for
subfunction 3 (Analyzing charge transfer based on density difference grid data). Two kinds of input
files are needed:
(1) A file containing basis function and molecular orbital information.
196
3 Functions
mwfn, .fch/.fchk, .molden and .gms files produced by excited state calculation (e.g. TDDFT) can be
directly used. This file should be loaded when Multiwfn boots up.
(2) A file containing configuration coefficients of excited states. The path of this kind of file
should be inputted when you enter corresponding analysis function, Multiwfn will load
configuration coefficients from this file. There are three cases, as shown below:
 Gaussian users: Output file (.out or .log) of CIS, TDHF, TDDFT and TDA-DFT tasks can
be used. Both single point and optimization tasks are supported; for the latter case, Multiwfn
analyzes electronic excitation at the final geometry. Since by default Gaussian only outputs the
configuration coefficients whose absoluate value is larger than 0.1, In order to achieve acceptable
accuracy, you must add IOp(9/40=4) keyword in the route section so that all configuration
coefficients whose magnitude larger than 0.0001 will be printed (If the calculation in Multiwfn is
found to be too expensive, using IOp(9/40=3) instead is also generally acceptable).
 ORCA users: Output file of CIS and TDA-DFT tasks can be used. Beware that TPrint
keyword should be used within %cis or %tddft, otherwise only very small amount of configuration
coefficients will be printed. TPrint x means outputting configurations whose contribution to excited
state larger than x*100%. Typically, I suggest using TPrint 1E-8. Since contribution is calculated as
square of configuration coefficient, TPrint 1E-8 simply corresponds to outputting configurations
who have absolute value of coefficients larger than 1E-4, the effect is identical to IOp(9/40=4) in
Gaussian. Below is an example input:
! PBE0 def2-SVP nopop
%tddft
nroots 8
tprint 1E-8
end
IMPORTANT NOTE: It is also possible to use ORCA TDDFT/TDHF output file, but the analysis result may
be unreliable or even fully wrong!!! Because in TD task, ORCA only prints configuration contributions (which are
given as sum of excitation and de-excitation contributions) but does not print configuration coefficients for excitation
and de-excitation respectively. In this case, Multiwfn automatically generates configuration coefficients by
calculating square root of the configuration contributions. This treatment is sometimes reasonable, however when
de-excitation is significant, the configuration coefficients yielded in this manner must be nonsense; in addition, even
if de-excitation is completely zero, the result may still be incorrect, because actual configuration coefficients may
either be positive or negative, while the sign evidently cannot be determined from configuration contributions.
 ORCA users using sTDA or sTDDFT: They are approximations of regular TDA and TDDFT,
respectively. In ORCA, their calculations are based on DFT MOs. Once the single point task of DFT
has finished, the excited states will be calculated by sTDA/sTDDFT with negligible cost. To carry
out these calculations, use keywords like follows (see ORCA manual for details)
! wB97X-D3 def2-SV(P) def2/J RIJCOSX nopop
%tddft
Mode sTDDFT //the sTDDFT may also be changed to sTDA
Ethresh 7.0
PThresh 1e-4
PTLimit 30
maxcore 6000
end
It is important to note that, at least for ORCA 4.2, only the three largest configuration
coefficients are printed (unfortunately, TPrint does not work for sTDA/sTDDFT calculation),
197
3 Functions
therefore often the normalization condition of configuration coefficients is violated evidently, in this
case the analysis result is unreliable or even fully misleading! So, please take care of the "Deviation
to expected normalization value" shown on screen after loading selected excited state.
 General cases: You can also use plain text file as the input file. The format of transition
information should be completely identical to Gaussian output, for instance:
Excited State 1 1 5.7945 // Label, index, multiplicity and excitation energy
(eV)
5 -> 6 0.70642 // MO pairs and configuration coefficients
// Use a blank line to separate each excited state
Excited State 2 1 7.8943
5 -> 7 0.63860
5 -> 8 0.30006
Excited State 3 1 7.8943
5 -> 7 -0.30006
5 -> 8 0.63860
4 <- 8 0.01000
Evidently, the above mentioned two kinds of files must correspond to the same geometry and
same calculation level. For example, if the MOs in the .fch were produced at B3LYP/6-31G* level
while the Gaussian output file corresponds to the TDDFT task carried out at PBE0/6-31G* level,
the analysis results will be completely meaningless.
 Special case for GAMESS-US and Firefly users: If you are a user of Firefly or GAMESS-
US program, you do not need to separately provide two kinds of files as mentioned above for
electron excitation analysis. If the input file used for Multiwfn is TDDFT output file with .gms suffix,
the Multiwfn will not only load basis function and molecular orbital information from this file when
Multiwfn boots up, but also load configuration coefficients of excited states when performing
electron excitation analysis. The H2CO_TDDFT_Firefly.gms and H2CO_TDDFT_GAMESS.gms in
"examples\excit" folder are example file of TDDFT output file of Firefly and GAMESS-US,
respectively.
For Firefly user, you should decrease the "PRTTOL" parameter in $TDDFT so that more
configuration coefficients could be printed.
For GAMESS-US, output file of CIS and TDA-DFT are not supported. In addition, there are
no option used to control the printing threshold of configuration coefficients, therefore some
analysis result may be not very accurate because some configurations, which have nonnegligible
contributions, may be ignored.
Output file of excited state optimization and frequency tasks of GAMESS-US and Firefly is
not supported.
3.21.0 Check, modify and export configuration coefficients of an

excitation (-1)
This function allows one to check, modify and export coefficient of configuration coefficients.
198
3 Functions
The input files needed by this function have been detailedly described at the beginning of Section
3.21, namely you should load a file containing basis function information when Multiwfn boots up,
then load a file containing configuration coefficients of excited states. The summary of all
recognized excited states will be printed on screen, you should select one of them, the configuration
coefficients of orbital pairs involved in this electron excitation will be loaded. After that, up to 10
orbital pairs that have largest absolute contribution to the excitation are automatically shown. Then
in the newly appeared menu, you can find below options:
1 Set coefficient of an orbital pair: You can use this option to replace the loaded configuration
coefficients of an orbital pair with inputted value.
2 Set coefficient for specific range of orbital pairs: This option is used to replace a batch of
loaded configuration coefficients with inputted values. You should input index range of the occupied
MOs and virtual MOs corresponding to the orbital transitions.
After manually modifying coefficients use above two options, in order to make the
modification affect following electron excitation analysis, you should use option -3 to export the
modified coefficients as plain text file, and then use this file as the second kind of input file for
electron excitation analyses.
-1 Retrieve original coefficient of all orbital pairs: If configuration coefficients have been
manually modified by above two options, you can select this option to retrieve the coefficients to
the original loaded values.
-2 Print coefficient (and contribution to excitation) of some orbital pairs: You can use this
option to print configuration coefficients whose absolute value are larger than specific value,
meantime the corresponding contributions to the electron excitation are shown together. Via this
option you can easily find out which orbital pair transition has crucial contribution to the electron
excitation.
-3 Export current excitation information to a plain text file: Basic information and
configuration coefficients of currently selected excited state can be exported to a specific plain text
file. This file can then be employed as the second kind of input file for various electron excitation
analysis functions of Multiwfn, e.g. hole-electron analysis and NTO analysis, and then the analysis
result will correspond to the modified configuration coefficients.
Obviously, if you set coefficient of some orbital pairs to zero, then their contributions to the
quantities you studied will be completely ignored; while if you have cleaned all coefficients except
for a specific orbital pair, then the resulting quantities will only reveal characters of this orbital
transition.
The example given in Section 4.18.10 utilized this function.
3.21.1 Analyze and visualize hole-electron distribution, transition

density, and transition electric/magnetic dipole moment density (1)
This very powerful module is used to analyze and visualize hole-electron distribution,
transition density and transition electric/magnetic dipole moment density. Moreover, hole and
electron can be decomposed to orbital pair contributions as well as atom and fragment contributions;
furthermore, the atom/fragment contributions can be directly plotted as heat map for visual
199
3 Functions
inspection.
3.21.1.1 Theory
There are many knowledge points involved in this module, they will be described below first.
Theory 1: Real space representation of hole and electron

Process of single-electron excitation can be described as "an electron leaves hole and goes to
electron", the "hole" and "electron" can be defined in different ways. If an excitation can be perfectly
described as HOMO→LUMO transition, then hole and electron could be simply represented by
HOMO and LUMO, respectively. However, in most practical cases, the single orbital pair
representation is not suitable, excitations have to be represented as transition of multiple MO pairs
with corresponding weighting coefficients.
How to represent hole and electron distributions when there is no single dominant MO pair
transition? One way is using natural transition orbital (NTO) analysis, as introduced in Section
3.21.6. Unfortunately, in many cases, even the MOs have been transformed to NTOs, there is still
no single NTO pair that has dominating contribution. The best representation of hole and electron
may the one introduced in this Section, the idea was originally proposed by me and my collaborator
Cheng Zhong in 2013. Although the paper detailedly introducing this method has not been published,
if this theory is involved in your study, please cite my this work: Carbon, 165, 461-467 (2020) DOI:
10.1016/j.carbon.2020.05.023, in which hole-electron analysis is utilized and briefly described.
It can be shown that density distribution of hole and electron can be perfectly defined as
 hole(r ) =  (loc)
hole
(r ) +  (cross)
hole
(r ) =  ( wia ) 2 i (r )i (r ) +   w w  (r) (r)
a
i
a
j i j
i →a i →a j i →a
 ele (r ) =  (loc)
ele
(r ) +  (cross)
ele
(r ) =  ( wia ) 2  a (r ) a (r ) +  w w  a
i
b
i a (r ) b (r )
i →a i → a i →b  a
note that the notions used here:

occ vir occ occ vir
  
i →a i a

i →a j i →a
 
i j i a
where  denotes MO wavefunction. "loc" and "cross" stand for the contribution of local term and
cross term to the hole and electron distribution, respectively. Note that the definition of hole and
electron given above is in density form rather than wavefunction form, hence the hole and electron
do not have phase (If you really need phase information of hole and electron, you should resort on
NTO analysis, see Section 3.21.6).
Due to the orthonormality of MOs and the fact that the sum of square of all configuration
coefficients is 1.0, it is clear that
 (r) d r = 1  (r) d r = 1
hole ele
This is an important property that any reasonable definition of hole and electron distribution should
satisfy, it indicates that one electron is excited.
The overlap function between hole and electron distribution can be defined as
Sm (r) = min[  hole(r),  ele (r)]

namely taking the minimal value of hole and ele everywhere. Another function for measuring the
overlap is
200
3 Functions
Sr (r) =  hole (r)  ele (r)

It is evident that Sr is always equal or larger than Sm. Both the two definitions are reasonable, but I
prefer to use Sr, since its graphical effect is better and its mathematical meaning is more clear.
The charge density difference (CDD) between excited state and ground state can be easily
evaluated as
 (r) =  ele (r) −  hole(r)

NOTE: Beware that if you are a Gaussian user, the  calculated in this way is obviously different
to the  produced via substracting excited state density by ground state density, unless you
specified keyword density=rhoci when generating .wfn/wfx file of excited state. Because by default
the excited state density exported to .wfn/wfx file by Gaussian is relaxed density rather than
unrelaxed density (which is directly constructed by MOs and excited state configuration
coefficients). In other words, unrelaxed excited state density can be simply written as
 excited (r) =  ground(r) −  hole(r) +  ele (r) , while deriving relaxed excited state density requires
employing the very complicated "Z-vector" method.
After generalization, above definitions of hole and electron can also be applied to TDHF and
TDDFT cases, where de-excitations must be taken into account. The generalized local terms are
 (loc)
hole
=  ( wia ) 2  i −  ( wia ) 2  i
i →a i a
 ele
(loc) =  ( w )  a −  ( wia ) 2  a
a 2
i
i →a i a
where i=|i|2stands for electron density of orbital i, w' denotes configuration coefficient of de-
excitation. The generalized cross terms are
 (cross)
hole
=  w w   −   w w  
a
i
a
j i j i
a
j
a
i j
i →a j i →a i a j i a
 (cross)
ele
=  w w   −   w w  
a
i
b
i a b i
a
i
b
a b
i →a i →b  a i a i b  a
If combining local and cross terms together, the expressions of hole and electrons could be
simply expressed as
 hole = w w  
i, j →a
a
i
a
j i j −  w w  
i, j a
i
a
j
a
i j
 ele = w w  
i → a ,b
a
i
b
i a b −  w w  
i  a ,b
i
a
i
b
a b
Theory 2: Contribution of MOs, basis functions, atoms and fragments to hole and
electron
In order to investigate which MOs have significant contributions to hole and electron, I defined
the contribution of occupied MO to hole and contribution of virtual MO to electron as follows
ihole = [( wia )2 − ( wia )2 ] a = [( wi ) − ( wi ) ]
ele a 2
a 2
a i
Below normalization conditions are held evidently:
201
3 Functions

i
hole
i =1 
a
ele
a =1
Contribution to hole/electron by an atom can be derived as follows. Considering the

normalization condition of the hole (de-excitation part is temporarily ignored for simplicity)
 
   w w   dr = 1
a a
i j i j 
i, j→a 
 
   w w     dr = 1
a a
C C
i j ,i ,j
 
i, j→a 
 w w 
i, j→a  
C C     dr = 1
a
i
a
j ,i ,j
 w w 
i, j→a  
a
C C
i
a
j ,i ,j S  , = 1
where  denotes basis function, S and C are overlap matrix and coefficient matrix, respectively. If
we employ Mulliken-like method to partition the 
 
C C ,i ,j S , term as atomic contributions,
then we can define contribution of atom A to hole in below form
 
hole
A = ww a
i
a 1
j 2
   C ,iC , j S , +  C ,iC , j S , 
 
i, j →a   A    A 
We can simiarly apply above treatment on de-excitation part of hole as well as electron. The
actual equations used to evaluate atomic contribution to hole and electron are
   
hole
A = ww a
i
a 1
j 2
   Tij, +  Tij,  −  wia wja
 
1
2
   Tij, +  Tij, 
 
i, j →a   A    A  i, j a   A    A 
   
ele
A = ww a
i
b 1
i 2
   Tab, +  Tab,  −  wia wib 12    Tab, +  Tab, 
   
i → a ,b   A    A  i  a ,b   A    A 
where T is an intermediate matrix for facilitating calculation, it is defined as Tij, = C ,iC , j S , .

Fragment contributions to hole and electron can be simply evaluated by summing up atomic
contributions
hole
frag = 
A frag
hole
A ele
frag = 
A frag
ele
A
Contribution of a basis functions  to hole and electron can be defined as

   
hole
 = ww
i, j →a
a
i
a 1
j 2   Tij, +  Tij,   −  wia wja 12   Tij, +  Tij,  
   i, j a   
   
ele
 = ww
i → a ,b
a
i
b 1
i 2   Tab, +  Tab,   −  wia wib 12   Tab, +  Tab,  
   i  a ,b   
In order to significantly save computational time, Multiwfn ignores all terms if magnitude of product of
corresponding two configuration coefficients is less than 0.001. The loss of accuracy due to this trick is negligible.
Furthermore, we can define contribution of atom and fragment to charge density difference as
202
3 Functions
CDD
A = ele
A − A
hole
CDD
frag =  frag −  frag
ele hole
Overlap between hole and electron in atom and fragment spaces are defined as geometry
average of their contributions:
ovlp
A = ele
A A
hole
ovlp
frag = ele
frag frag
hole
Notice that the overlap in this form is not additive, namely  A + ovlp  ovlp
ovlp
B AB .
Some atomic contributions derived in above way may be small negative values in certain
situations, this is due to the theoretical shortcoming of Mulliken type of partition, obviously in this
case the overlap between hole and electron in corresponding atomic spaces cannot be evaluated, so
Multiwfn automatically sets the overlap values to zero. The Mulliken type of partition is well-known
incompatible with diffuse functions, therefore basis set containing diffuse functions must not be
employed in the excited state calculation if you intend to evaluate atomic contributions to hole and
electron in above way.
Due to the extreme flexibility of Multiwfn, there are also other ways of determining atomic
contributions to hole and electron, however you have to manually evaluate them. For example, if
you want to employ fuzzy partition for hole and electron, you should first export cube file of hole
or electron, then set "iuserfunc" in settings.ini to -1 (in this case the user-defined function will
correspond to the interpolated function based on the grid data), then load hole or electron cube file
into Multiwfn, use subfunction 1 of main function 15 to integrate user-defined function in each
atomic fuzzy space. Although this way is more time-consuming, it is quite robust and completely
compatible with diffuse function, and no unphysical negative value will occur. If in the main
function 15, you first select option -4 to define a fragment and then use subfunction 1 to integrate
user-defined function, then the sum of results of all atoms will correspond to the fragment
contribution.
Theory 3: Quantitative characterization of hole and electron distribution in the whole

space
The overall distribution of hole and electron can be quantitatively characterized in following
ways, they are quite useful for identifying type of electron excitations.
To characterize overlapping extent of hole and electron, Sm index and Sr index are defined as
follows (Sr must be equal or larger than Sm index)
Sm index =  Sm (r ) d r   min[  hole (r ),  ele (r )]d r
Sr index =  Sr (r ) d r    hole (r )  ele (r ) d r
Centroid can be calculated to reveal most representative position of hole and electron
distribution. For example, X coordinate of centroid of electron is written as
X ele =  x ele (r ) d r
where x is X component of position vector r.

The charge transfer (CT) length in X/Y/Z can be measured by distance between centroid of
hole and electron in corresponding directions:
203
3 Functions
Dx = X ele − X hole Dy = Yele − Yhole Dz = Zele − Z hole

The total magnitude of CT length is referred to as D index:
D index =| D | ( Dx )2 + ( Dy )2 + ( Dz )2
It is noteworthy that the variation of dipole moment of excited state (corresponding to
unrelaxed density) with respect to ground state in X, Y and Z can be simply calculated as
x = −( X ele − X hole) y = −(Yele − Yhole) z = −( Zele − Z hole)
The RMSD of hole and electron can be used to characterize their extent of spatial distribution.
For example, X component of RMSD of hole is expressed as
 hole,x =  (x − X hole )2  hole(r )dr

The |hole| and |ele| are referred to as hole and ele indices, they measure overall RMSD of hole and
electron, respectively.
The difference between RMSD of electron and hole in X/Y/Z direction can be measured via
, while overall difference can be measured via  index
  =  ele,  −  hole,  = {x, y, z}
 index =| σele | − | σ hole |
H measures average degree of spatial extension of hole and electron distribution in X/Y/Z
direction, HCT is that in CT direction, and H index is an overall measure
H  = ( ele,  +  hole, ) / 2  = {x, y, z}
H CT =| H  uCT |
H index = (| σ ele | + | σ hole |) / 2
where uCT is unit vector in CT direction and can be straightforwardly derived using centroid of hole
and electron.
t index is designed to measure separation degree of hole and electron in CT direction:
t index = D index − H CT
If t index<0, it implies that hole and electron is not substantially separated due to CT. Clear
separation of hole and electron distributions must correspond to evidently positive t index.
The hole delocalization index (HDI) and electron delocalization index (EDI) are defined as
follows
HDI = 100   [
hole
(r )]2 d r
EDI = 100   [
ele
(r )]2 d r
It is found that the smaller the HDI (EDI), the larger the spatial delocalization of hole (electron); in
other words, the more evenly distributed throughout the system. HDI and EDI are pretty useful in
quantifying breadth of spatial distribution (although |hole| and |ele| can also reveal this point, they
are not suitable when hole or electron are concentrated in multiple areas).
204
3 Functions
There are often many nodes or complicated fluctuations in hole and electron distributions. In
order to make visual study of hole and electron easier, Chole and Cele functions are defined as follows.
The function behavior of Chole and Cele is similar to Gaussian function, they are highly smooth
functions, the value asymptotically approaches zero from centroid of hole/electron.
 ( x − X ele ) 2 ( y − Yele ) 2 ( z − Z ele ) 2 

Cele (r ) = Aele exp  − − − 
 2 2
2 2
2 2 
 ele, x ele, y ele, z 
 ( x − X hole) 2 ( y − Yhole) 2 ( z − Z hole) 2 
Chole(r ) = Ahole exp  − − − 
 2 2
2 2
2 2 
 hole,x hole,y hole,z 
The factor A is introduced so that Chole and Cele are normalized.
In fact, the definition of RMSD, Chole, Cele, H and t indices introduced above was motivated by
J. Chem. Theory Comput., 7, 2498 (2011), these quantities were originally used to analyze electron
excitation based on density difference, but I found all of them work well under the framework of
hole-electron analysis. Also note that many details of these indices have been modified when
introduced to hole-electron analysis framework.
The above defined quantitative indices could be used for distinguishing type of electron
excitation. My empirical rule is summaried as follows, it should be suitable for most cases.
Index
Excitation type
D Sr t ∆
LE small medium~large <0 small
Single direction CT large ? ? ?
Centrosymmetric CT small ? <0 large
Rydberg small Small <0 large
In the table, three kinds of excitations are involved:
• Local excitation (LE): The hole and electron occupy similar spatial region.
• Charge-transfer excitation (CT): The spatial separation of hole and electron is large, leading
to evident displacement of charge density. The CT may be single directional or multiple directional
(centrosymmetric CT is a special case of the latter).
• Rydberg excitation: Electron mainly consists of very diffuse MOs, therefore the overlap
between electron and hole must be small. This type of excitation in general does not lead to
prominent long-range displacement of charge density.
Theory 4: Transition density matrix and transition density

(One-electron, spinless) transition density matrix between excited state and ground state of an
N-electron system in real space representation is defined as follows (real type of wavefunctions is
assumed, so complex conjugation sign is omitted)
T (r; r' )  T (r1 ; r1 ' ) =   0 (x1 , x 2 ,  x N ) exc (x1 ' , x 2 ,  x N )d 1dx 2 dx 3 dx N
where 0 is Slater-determinant of ground state wavefunction. x is spin-space coordinate,  stands

for spin coordinate. The T is called as matrix because it has two continuous indices.
For excited state wavefunction generated by single-reference methods, after expanding exc
and applying Slater-Condon rule, it can be easily shown that T can be explicity written as
205
3 Functions
T (r; r' ) =  wiai (r ) a (r' )

i a
If we only take the diagonal terms of the transition density matrix, then we obtain transition
density
T (r ) =  wiai (r ) a (r )
i a
T(r) can be studied as a common real space function, for example, visualized in terms of isosurface
map. Assuming that there is only one dominant orbital transition, for example, HOMO→LUMO,
then T(r) is simply HOMO(r)LUMO(r). Therefore, it is easy to understand, if a region has large
magnitude of transition density, the hole and electron must be strongly coupled in this region; while
if a region has small distribution of T(r), then overlap between hole and electron in this area should
be insignificant. Clearly, T(r) is a useful function for characterizing underlying nature of electron
excitation, and its main distribution characteristics is closely related to the Sr(r) function.
Note that due to the orthonormality of MOs, integral of T(r) over the whole space is exactly
zero. If the excited state and ground state correspond to different spin states, due to the
orthonormality of spin coordinate, T(r;r') must be a zero matrix, and T(r) is correspondingly zero
everywhere. However, notice that only spatial part of T(r) is taken into account when Multiwfn
evaluates it, therefore you are still able to study T(r) for e.g. S0→T1 excitation.
Theory 5: Transition electric/magnetic dipole moment density

Note that there are many kinds of transition dipole moment, including transition electric dipole
moment, transition magnetic dipole moment, transition velocity dipole moment and so on. The word
"transition dipole moment" commonly refers to transition electric dipole moment.
X, Y and Z components of transition electric dipole moment density can be written as negative
of product of X, Y and Z coordinate variables and transition density, respectively:
Tx (r) = − xT (r) Ty (r) = − yT (r) Tz (r) = − zT (r)
Integrating transition electric dipole moment density over the whole space yields transition
dipole moment D
Dx =  Tx (r )dr Dy =  Ty (r )dr Dz =  Tz (r )dr
Obviously, one can conveniently study contribution to transition electric dipole moment of various
molecular regions by plotting transition electric dipole moment density.
Next, we look at transition magnetic dipole moment. The operator for magnetic dipole moment
due to movement of electrons is the angular momentum operator L (see e.g. Theoret. Chim. Acta, 6,
341 (1966))
L = −i (r  ) = îLx + ˆjLy + kˆ Lz
           
= −i î  y − z  + ˆj  z − x  + kˆ  x − y 
  z y   x z   y x 
where i, j, k are unity vectors in X, Y and Z directions, respectively. Therefore, the X component of
transition magnetic dipole moment can be explicitly defined as below. Noticed that in order to
provide a real value I ignored the imaginary and negative signs simultaneously; the symbol "←"
denotes de-excitation MO pairs in TDHF/TDDFT formalism.
206
3 Functions
  exc    
M x = 0 y −z  =  wia i y − z  a −  wjb  j y − z b
z y i →a z y j b z y
My and Mz can be defined similarly.
We can define transition magnetic dipole moment density component mi(r) by considering the

relationship M i = mi (r )dr i = x, y, z , so that distribution of transition magnetic dipole
moment can be visualized in terms of e.g. isosurface map. Explicit expression of mi(r) of X
component is given below, Y and Z components can be defined similarly.
       
mx (r ) =  wiai (r )  y a (r ) − z a (r ) −  wjb j (r )  y b (r ) − z b (r )
i →a  z y  j b  z y 
Theory 6: Coulomb attraction between hole and electron (exciton binding energy)
The "electron" of course carries negative charge, while "hole" can be regarded as carrying
positive charge, therefore formally there is a Coulomb attractive energy between them, its negative
value is known as exciton binding energy, which is a positive value. This term can be calculated via
simple Coulomb formula (in atomic unit form):
 hole(r1 )  ele (r2 )

EC =   d r1 d r2
| r1 − r2 |
Some discussions about the exciton binding energy can be found in e.g. J. Chem. Phys., 143, 244905
(2015) and J. Phys. Chem. C, 121, 17088 (2017).
Note that the exciton binding energy calculated in above form is different to the exciton binding energy defined
in another form, namely EC=(IP-EA)-Eoptical gap (see Mater. Horiz., 1, 17 (2014) for more details), because electronic
correlation and orbital relaxation effects are involved in practical electron ionization and electron affinity processes;
moreover, in fact there is an exchange term in EC (though it is negligible when separation of hole and electron is
significant). All of these factors are ignored in the evaluation of exciton binding energy in Multiwfn.
In Multiwfn, above integral is directly calculated based on evenly distributed grid data of hole
and electron. Notice that although the code has been substantially optimized and parallelized, the
computational cost is still high, therefore you need to wait patiently during calculation. The cost is
formally proportionally to square of the number of grids; therefore, the cost of medium quality grid
will be higher than low quality grid by one order of magnitude.
3.21.1.2 Usage and Functions

The input files needed by present module have been detailedly described at the beginning of
Section 3.21, namely you should load a file containing basis function information when Multiwfn
boots up, and then load a file containing configuration coefficient information of excited states. The
summary of recognized excited states will be printed on screen, you should select the excited state
that you want to carry out aforementioned analyses. Each time only one state can be analyzed, if
you want to analyze another state, you should exit this function, then enter again and select another
state.
Present module has many functions, they will be described below in turn.
Function 1: Visualize and analyze hole, electron and transition density and so on
After you enter this function, you are requested to set up grid data, then grid data will be
calculated for hole distribution, electron distribution, overlap of hole and electron, transition density,
207
3 Functions
transition electric/magnetic dipole moment density, charge density difference and Cele/Chole
functions.
After calculation of grid data is finished, various quantities introduced in Section 3.21.1.1 will
be evaluated based on the evenly distributed grid data and then shown on screen, their meanings
should be very easy to understand. The outputted transition electric/magnetic dipole moment is
calculated by integrating grid data of transition dipole moment density, the value should be very
close to the one directly outputted by quantum chemistry program. The ideal value of the integral
of hole or electron over the whole space is 1.0, while for transition density the ideal value is 0. If
the actual outputted values deviate too far from expected values, then the printed t index, H index,
D index, Sm index and so on may be unreliable. There are three reasons may lead to this problem:
(1) The grid quality is too poor. Higher number of grid points should be used
(2) The spatial extent of the grid data is too narrow, you should enlarge extension distance so
that the grid data could cover broader regions
(3) You forgot to use the IOp(9/40=x) option mentioned at the beginning of Section 3.21, as a
result, only very small number of configuration coefficients are loaded
In post-processing menu, grid data of hole, electron, transition density, Sm/Sr and so on can be
directly visualized as isosurface map, or be exported as cube file in current folder by corresponding
options. You can also choose corresponding option to calculate Coulomb attractive energy between
hole and electron distribution, notice that this calculation is expensive even if you only choose low
quality grid.
By default, the transition magnetic dipole moment density is not evaluated because it is less
important than the transition electric dipole moment density. If you want to calculate it, select option
-1 before entering this function.
For large systems, if computational cost for grid data is too high and you only need to
qualitatively examine isosurface map of hole, electron, transition density and so on, in Gaussian you
can safely use IOp(9/40=3) instead of IOp(9/40=4), so that smaller number of configurations will
be taken into account.
Function 2: Show molecular orbital contribution to hole and electron distribution

You only need to input printing threshold, then contribution of MO to hole and electron
distribution will be shown. This function is very useful to identify which MOs have significant
contribution to hole and electron. Below is an output example:
MO 126, Occ: 2.00000 Hole: 0.29664 Electron: 0.00000
Sum of hole: 1.00000 Sum of electron: 1.00000
Function 3: Show atom or fragment contribution to hole and electron and plot the
contributions as heat map
After you enter this function, many quantities mentioned in "Theory 2" of Section 3.21.1.1 will
be printed on screen, below is an output example. Mulliken type of paritition is used to derive the
208
3 Functions
atomic contributions.
Contribution of each non-hydrogen atom to hole and electron:
1(C ) Hole: 1.37 % Electron: 8.96 % Overlap: 3.50 % Diff.: 7.59 %
2(C ) Hole: 11.86 % Electron: 0.74 % Overlap: 2.97 % Diff.: -11.11 %
...[ignored]
14(N ) Hole: 0.18 % Electron: 23.80 % Overlap: 2.07 % Diff.: 23.62 %
15(O ) Hole: 3.07 % Electron: 17.23 % Overlap: 7.28 % Diff.: 14.16 %
16(O ) Hole: 3.07 % Electron: 17.23 % Overlap: 7.28 % Diff.: 14.16 %
In the output, the "Overlap" is simply the geometry average of "Hole" and "Electron", while "Diff."
is obtained by substracting "Hole" from "Electron". Since hydrogens commonly do not participate
in electronic excitions of interest, by default hydrogens are ignored, but you can choose "Toggle if
taking hydrogens into account" option to switch status.
If you need contribution of molecular fragments to above mentioned quantities, you can select
option "-1 Load fragment definition" and then input the number of fragments and atomic index of
each fragment in turn. Fragment definition can also be loaded from an external plain text file, in
whcih each fragment occupies a line, for example
1,3,6-10,12
2,4,5
11
13-15
This example totally defines four fragments, the first fragment consists of atom 1,3,6,7,8,9,10,12.
Once defining fragments is completed, contribution of the fragments to various quantities will be
immediately printed on screen.
Composition of atom/fragment in hole and electron, as well as hole-electron overlap in various
atom/fragment spaces can be plotted as heat map, so that their distribution character can be very
vividly exhibited. Below is an example, the color correspond to function value, while abscissa
corresponds to atom index.
From the graph, you can immediately recognize that this is a local excitation, since most part of
both hole and electron are distributed on the fragment consisted of atom 1~14. In particular, atoms
7 and 8 are the atoms that contribute most to this electron excitation. If you load fragment definition
before plotting, then the abscissa of the heat map will correspond to fragment index. In the menu,
there are also options used to adjust color scale, ratio of the map and stepsize between labels in X
axis.
An very detailed example of this hole-electron module is given in Section 4.18.1. Example of
analyzing transition density and transition dipole moment density using this module is given in
Section 4.18.2.1. More discussion and examples can be found from my blog article "Using Multiwfn
to perform hole-electron analysis to fully investigate electron excitation character" (in Chinese,
209
3 Functions
http://sobereva.com/434).
Information needed: See beginning of Section 3.21.
3.21.2 Plot atom/fragment transition matrix of various kinds as heat

map (2)
This function is used to plot atom transition matrix (ATM) of various kinds as heat map (color-
filled matrix map). The ATM refers to any kind of atom based matrix that represents electron
transition information between two states. For example, it may correspond to the atom based
transition density matrix (see below), the atom-atom charge transfer matrix, the atom transition
dipole moment matrix and so on. In this function, the ATM can also be further transformed to
fragment transition matrix (FTM) and then plotted as heat map.
Although this function can also plot heat map for other matrices, the major purpose of
developing this function is plotting atom or fragment based transition density matrix, therefore I
will first introduce theories related to transition density matrix.
Theories about transition density matrix (TDM)

Below, the word "TDM" refers to the transition density matrix in basis function representation.
The TDM between ground state and an excited state can be calculated as (de-excitation transitions
have been ignored for simplicity)
occ vir
 =  wi Ci Ca
Ptran a
i a
where Ci denotes the expansion coefficient of basis function  in MO i. It is worth to note in passing
that the TDM in real space representation, which is introduced in Section 3.21.1.1, can be
constructed easily via TDM in basis function representation ( stands for basis function):
T (r; r' ) =  Ptran
   (r )  (r ' )
 
The off-diagonal elements of TDM essentially represent the coupling between various basis
functions during electron excitation. Assume there are only two basis functions and meantime the
excitation can be perfectly represented as i→a MO transition, then the TDM could be explicitly
written as below form (notice that the index of the elements has been rearranged according to
convention of TDM heat map)
1,2 2,2 C1iC2 a C2iC2 a 

P tran =  =
1,1 2,1  C1iC1a C2iC1a 
If magnitude of off-diagonal element P1tran

,2 is large, it implies that basis functions 1 and 2
significantly participate in occupied orbital i and virtual orbital a, respectively. More generally, we
may say that basis functions 1 and 2 have large contribution to hole and electron, respectively, in
this case the two basis functions are strongly coupled during the excitation. The diagonal terms are
also meaningful, if element Ptran

,  has large magnitude, it implies that basis function  must
210
3 Functions
simultaneously have large contribution to both hole and electron.

Since TDM in general is not a symmetric matrix, in order to make certain discussions easier,
some literatures employ below symmetrized form
 + P
Ptran tran
 =
Ptran
2
The TDM can be contracted to atom based form according to correspondence between basis
functions and atoms, it will be symbolized as p. In Multiwfn, below construction ways are available:.
Way 1 : p AB =  ( P 
 
 A B
tran 2
)
Way 2 : p AB =  ( P 
 
 A B
tran 2
)
Way 3 : p AB = | P 
 
 A B
tran
|
Way 4 : p AA = | P 
A
tran
|, p Atran B =  ( P 
 
 A B
tran 2
)
where  and  denote the basis functions centered at atom A and on B, respectively. Both original
form and symmetrized form of TDM could be employed here.
If way 1 is employed, the p will correspond to the matrix of so-called correlated electron-hole
probability diagram (CEHPD), its (A,B) element was interpreted as the probability of simultaneously
finding a hole in atom A and an electron in atom B (this interpretation is not strictly true in general
cases). See J. Chem. Phys., 113, 10002 (2000) and J. Am. Chem. Soc., 129, 14257 (2007) for
example, in which the authors used Ptran

 obtained at ZINDO level.
If the p are constructed in way 2, 3 or 4, the resulting matrix may be referred to as atom
transition density matrix. For example, the way 4 has been employed in Chem. Rev., 102, 3171
(2002). However, according my experiences, using way 2 or 3 is more preferred, since I found that
the diagonal terms obtained in way 4 is often too large compared to the off-diagonal terms.
Assume that the TDM used to construct p was not symmetrized, the general structure of the
resulting p could be expressed in below form
1, N 2, N  N , N 
     
p  electron  
 1,2 2,2  N ,2 
 
 1,1 2,1  N ,1 
hole
In complete analogy with the discussion about TDM, the physical meaning of the matrix elements
of p can be roughly understood as follows, irrespective of the choice of the specific way of
constructing the p:
 Diagonal terms: If (A,A) is large, it implies that atom A has large contribution to both hole and
electron, therefore the electron excitation should result in evident charge reorganization within
atom A
 Off-diagonal terms: If (A,B) is large, then atom A should have large contribution to hole and
211
3 Functions
meantime atom B should have large contribution to electron, implying that electron excitation
leads to CT from A to B
The "hole" and "electron" mentioned above are highly abstract concepts, although they have the same physical
meaning as the those defined in the hole-electron analysis (Section 3.21.1), one cannot expect that the pattern of the
p defined in any one of above ways is always very close to the atom-atom charge transfer matrix, which is much
more strictly defined and more meaningful.
If symmetrized form of TDM was used to bulid p, then CT directional information will not be
reflected by p. In this case, if off-diagonal term (A,B)=(B,A) is large, then we can simply say that
coherence between atoms A and B is strong during the electron excitation, in other words, charge
transfer occurs between atoms A and B.
The heat map of p is particularly useful for analyzing large-size and highly conjugated
molecules. Commonly hydrogens are omitted in the plot to make the map compact, since hydrogens
rarely participate in electron excitation of chemical interest.
If fragments are defined, the p (or other kinds of atom transition matrix) can further be
contracted to fragment based form:
pRS =  p
Afrag R Bfrag S
AB
This form is very convenient when one wishes to study role of various fragments in electron
excitation.
Input files
Since there are different types of atom transition matrix, and the matrix can be passed to
Multiwfn in different ways, there are several circumstances as shown below, you should use proper
input files. The file that should be loaded when Multiwfn boots up is always the file containing basis
function information, and it should corresponds to another file that needed to be loaded when you
enter present function.
(1) Plotting heat map of p in usual way
You should load a file containing configuration coefficient information of excited states when
you enter this function (see beginning of Section 3.21). Then Multiwfn will automatically generate
TDM between ground state and you selected excited state, and at the same time you can choose if
symmetrizing the resulting TDM in aforementioned way.
(2) Plotting heat map of p based on the TDM recorded in Gaussian output file
You should load Gaussian output file of electron excitation task when you enter this function.
The keywords density=transition=x IOp(6/8=3) must be specified in Gaussian input file, so that
TDM between ground state and excited state x can be printed in output file by Link 601 of Gaussian.
Via this way, not only the TDM of CIS/TDHF/TDA-DFT/TDDFT can be plotted, but also the TDM
generated by the EOM-CCSD and semi-empirical ZINDO method can be plotted.
Note 1: The TDM outputted by Gaussian is in aforementioned symmetrized form.
Note 2: If your ground state is singlet state while you used such as TD=triplet to request Gaussian to compute
triplet excited states, then the outputted TDM will be exactly zero due to spin forbidden, and thus Multiwfn is unable
to plot corresponding TDM map. However, it is possible to draw spatial part of the singlet-triplet TDM. To do this,
you should let Multiwfn itself to generate TDM, see (1).
Note 3: If the basis set you used contains diffuse basis functions, in rare cases, the TDM outputted by Gaussian
is incorrect, and thus the resulting heat map will be useless.
In summary, if the method you are using is not ZINDO, do not let Multiwfn to load TDM directly from Gaussian
output file.
(3) Plotting heat map of p based on the TDM recorded in a plain text file
You should load a file named tdmat.txt when you enter this function, Multiwfn will read TDM
from this file. Commonly, the tdmat.txt is generated by subfunction 9 of main function 18 (see
212
3 Functions
Section 3.21.9 for detail), which can not only generate TDM between ground state and an excited
state, but can also generate TDM between two excited states. An example file has been provided as
examples\excit\tdmat.txt.
For above three cases, you can choose the way used to contract the TDM to the p.
(4) Plotting atom transition dipole moment matrix
You should load a file named one of AAtrdip.txt, AAtrdipX.txt, AAtrdipY.txt, AAtrdipZ.txt when
you enter this function, Multiwfn will read atom transition dipole moment matrix from this file.
Commonly, they are generated by subfunction 11 of main function 18 (see Section 3.21.11 for detail).
By plotting heat map of these matrices, one can easily recognize which atoms and which interatomic
couplings notably affect transition dipole moment.
(5) Plotting atom-atom charge transfer matrix
You should load a file named atmCTmat.txt when you enter this function, Multiwfn will read
atom-atom charge transfer matrix from this file. Commonly, the atmCTmat.txt is generated by
subfunction 8 of main function 18 (see Section 3.21.8 for detail). By plotting heat map of this kind
of matrix, charge transfers between various atoms or fragments as well as charge reorganization
sites can be intuitively recognized.
Hint: In fact, you can also make the tdmat.txt or AAtrdip.txt/atmCTmat.txt contain other kind of matrices so that
they can be plotted as heat map via present module. For example, you can export bond order matrix as bndmat.txt
using corresponding subfunction in main function 9, then rename it as atmCTmat.txt and delete the first line from it,
then if you load this file into Multiwfn when entering present module, the plotted heat map will correspond to the
bond order matrix.
Usage
After loading all needed files and generating all needed data as mentioned above, you will enter
the interface for plotting heat map of the atom transition matrix (ATM). There are several options:
Option 0: Showing heat map of ATM on screen. By default, labels in abscissa and ordinate of
this map correspond to indices of non-hydrogen atoms.
Option 1: The same as option 0, but save the heat map as graphical file in current folder.
Option 3: Exporting the ATM as matrix.txt in current folder, so that it can then be conveniently
plotted by some third-party tools such as Origin and Sigmaplot.
Option 4: Switching the status that if hydrogens will be included in the heat map
Option 5: Changing upper and lower limits of color scale. By default they are automatically set
to maximal and minimal elements of the ATM, respectively.
Option 6: Changing the number of interpolation steps between grid data. If you want to make
the graph look smooth, it should be set to a large value (the default 10 is already quite large); if the
value is set to 1, then interpolation will not be performed, in this case each square grid in the map
exactly corresponds to a matrix element.
Option 7: Setting interval between labels in abscissa and ordinate.
Option 8: Determining if performing normalization. If the status is switched to "Yes", then
normalization factor will be applied so that the sum of all elements of ATM is equal to unity.
If you select option "-1 Define fragments", fragment definition can be directly inputted or be
loaded from a plain text file, which should look like below, each fragment occupies a line:
1,3,6-10,12
2,4,5
11
13-15
213
3 Functions
Then Multiwfn will contract the atom transition matrix to fragment transition matrix (FTM). After
that, the matrix to be plotted or exported in present module will be FTM instead of ATM
An example of plotting and studying p matrix is given in Section 4.18.2.2; example of

analyzing transition dipole moment matrix is given in Section 4.18.2.3; example of plotting atom-
atom charge transfer matrix is given in Section 4.18.8.
3.21.3 Analyze charge-transfer based on density difference grid data (3)
Theory
In the paper J. Chem. Theory Comput., 7, 2498 (2011), the authors proposed a method for
analyzing charge-transfer (CT) during electron transition, present function fully implements this
analysis method. It is also probable that this method can be used to study CT in other processes,
such as formation of molecular complex. In the original paper, the author only discussed the cases
when charge-transfer is in one-dimension, while in Multiwfn this scheme has been generalized to
three-dimension case. In addition, some quantities introduced below are not proposed in the original
paper but proposed by me, definition of some quantities in the original paper have also been
modified by me to make the analysis more meaningful.
The electron density variation between excited state (EX) and ground state (GS) is
 (r) =  EX (r) − GS (r)

Notice that the geometry used in calculating EX and GS must be identical, otherwise the resulting
 will be meaningless. Therefore, present function can only be used to characterize "vertical"
process.  can be divided into positive and negative parts, namely + and −. Of course, the integral
of + and -− over the whole space should be equal. If evident unequality is observed, that means
the error in numerical integral is unneglectable, and higher quality of grid (i.e. larger number of grid
points) is required. Even though what you analyzed is single-electron excitation, the magnitude of
+ and − as well as their integrals over the whole space in principle can also be larger than 1.0, this
is because excitation of an electron must lead to reorganization of distribution of the rest of electrons,
which also make contribution to .
The transferred charge qCT is the magnitude of the integral of + and − over the whole space.
It is important to correctly recognize the physical meaning of this quantity. qCT only corresponds to
the total amount of charge whose distribution is perturbed during electron excitation, it does not
correspond to net charge transfer from one fragment to another fragment (e.g. from donor group to
acceptor group)
The barycenter of positive and negative parts of  can be computed as
R + =  r + (r ) d r /   + (r ) d r
R − =  r − ( r ) d r /   − ( r ) d r
The Cartesian component coordinates of R+ will be referred to as X + ,Y+ ,Z + below, while that of
R− will be referred to as X − , Y− , Z − .
The distance between the two barycenters measures the CT length, its three Cartesian
214
3 Functions
components:
Dx = X + − X − Dy = Y+ − Y− Dz = Z + − Z −
The D index is defined as ( Dx )2 + ( Dy )2 + ( Dz )2 | R + − R − | , which characterizes total CT
length.
The dipole moment variation caused by electron excitation can be evaluated as
 X = ( X + − X − )qCT Y = (Y+ − Y− )qCT Z = ( Z + − Z − )qCT

The RMSDs of distribution of + and − in each direction are defined as
  (r)( '−  )
2
dr
 a , =
a a
  (r ) d r
a
where a={+,-}, ’={x,y,z}, ={X, Y, Z}. x, y and z are Cartesian components of position vector r.
For example, +,y can be explicitly written as
 +, y =  + (r )( y − Y+ ) 2 d r
 + (r ) d r
The difference between RMSD of + and − in X/Y/Z direction can be measured via , while
overall difference can be measured via the  index
  =  + , −  −,  = {x, y, z}
 index =| σ + | − | σ − |
It is noteworthy that D index is zero for exactly centrosymmetric systems, therefore, it is useless for
discussing CT problem of such kind of system. However,  index is often useful to identify this
type of excitation, since in this case  index must be large because diffuseness extent of + is much
higher than −.
C+ and C− functions are defined aiming for visualizing CT more intuitively than . Their
structures are similar to Gaussian function, the value asymptotically approaches zero from the
centroid of the function.
 ( x − X + ) 2 ( y − Y+ ) 2 ( z − Z + ) 2 
C + (r ) = A+ exp  − − − 
 2  2
2 2
2 2 
 +, x +, y +, z 
 ( x − X − ) 2 ( y − Y− ) 2 ( z − Z − ) 2 
C − (r ) = A− exp  − − − 
 2  2
2 2
2 2 
 −, x −, y −, z 
The normalization factor A is introduced so that the integrals of C+ and C− over the whole space are
equal to that of + and −, respectively.
H measures average degree of spatial extension of − and + in X/Y/Z direction, HCT is that in
CT direction, and H index is an overall measure:
215
3 Functions
H  = ( + , +  −, ) / 2  = {x, y, z}
H CT =| H  uCT |
H index = (| σ + | + | σ − |) / 2
where uCT is unit vector in CT direction and can be straightforwardly derived using centroid of −
and +.
t index measures separation degree of + and −:
t index = D index − H CT
If t index<0, it implies that − and + are not substantially separated due to CT. Clear separation of
− and + distributions must correspond to evidently positive t index.
I defined another quantity to measure overlapping extent between C+ and C−:
S+− =  C+ (r ) / A+ C− (r) / A− d r
If the value equals to 1, that means the two functions are completely superposed, else if the value
equals to zero, it indicates that the distribution of them are completely separated. This index is
dimensionless.
Usage
Because all numerical integrals mentioned above are computed based on evenly distributed
grid data, user needs to generate grid data of  by using custom operation of main function 5, see
Section 3.7.1, or load a file (e.g. cube file) containing grid data of density difference when Multiwfn
boots up. After that, enter subfunction 3 of main function 18, all aforementioned quantities will be
shown on screen immediately. The "Overlap integral between C+ and C-" term is the S+− introduced
above. In the post-processing menu, user can choose to visualize C+ and C−, or export grid data for
the two functions to cube file in current folder.
Information needed: Grid data of electron density difference
3.21.4 Calculate ∆r index to measure charge-transfer length (4)
Theory
In the paper J. Chem. Theory Comput., 9, 3118 (2013), r index was proposed to measure CT
length during electron excitation. The r can be expressed as
r =  ria
i,a
where ri is contribution of orbital transition between i and a to the r index:

a
( Kia )2
ria =  a r  a − i r i
 i(
i, a
K a 2
)
216
3 Functions
The index i and a run over all occupied and virtual MOs, respectively.  is orbital wavefunction.
Assume that the method you used to calculate electron excitation is CIS or the TDDFT under Tamm-
a
Dancoff approximation, then K i is simply the configuration coefficient corresponding to
excitation of i→a. While if the method you used is TDHF or TDDFT, then Ki = wi + wi , where
a a a
wia and wia denote the configuration coefficient corresponding to excitation of i→a and de-
excitation of ia, respectively.
r is especially useful for diagnosing when certain classes of DFT functionals are failure for
TDDFT purpose. When r is large, pure functionals such as BLYP and PBE, and the hybrid
functionals with low Hartree-Fock exchange composition such as B3LYP and PBE0, will not work
well. In this case, long-range corrected functionals should be employed; for instance, CAM-B3LYP,
LC-ωPBE and ωB97X.
It is worth to mention that if an electron excitation can be perfectly represented by one pair of
MO transition, then the r index and D index defined in hole-electron analysis framework will be
exactly identical in principle:
r =  a r  a − i r i   r  a (r ) d r −  r i (r ) d r
2 2
D index = D =  r ele (r ) d r −  r hole (r ) d r =  r  a (r ) d r −  r i (r ) d r

2 2
However, their values outputted by Multiwfn should be marginally different, since they are
evaluated based on different numerical integration algorithms.
Usage
boots up, and then load a file containing configuration coefficient information of excited states.
After entering present function (subfunction 4 of main function 18), you will be prompted to
select the excited states for which the r will be calculated, then the results will be printed on screen
immediately.
If you only selected one state, then Multiwfn will ask you to choose if decomposing the r into
orbital pair contributions. If you inputted e.g. 0.01, then orbital pairs which have contribution to r
larger than 0.01 will be printed. From the output, you can easily identify which orbital pairs have
significant contribution to charge-transfer of electron excitation.
An example of present function is provided as Section 4.18.4.
3.21.5 Calculate transition electric dipole moments between all states

and electric dipole moment of each state (5)
This function is used to calculate transition electric dipole moment between all states, including
217
3 Functions
the ones between ground state and excited states, as well as the ones between excited states. This
function is also able to print electric dipole moment for each state. For two states i and j, the
transition electric dipole moment is defined as i|-r|j; when i=j, this quantity corresponds to
electric dipole moment of this state that contributed by electrons.
The input files needed by present function have been detailedly described at the beginning of
boots up, and then load a file containing configuration coefficient information of excited states. If
you need very accurate transition dipole moments, you should use IOp(9/40=5) keyword of
Gaussian or TPrint 1E-10 keyword of ORCA to make the program print as much configuration
coefficients as possible.
After you enter present function, summary of all excitations will be printed. "Normalization"
should be as close as possible to expected value (0.5 and 1.0 for closed- and open-shell reference
state, respectively). If the deviation is large, then the resulting transition dipole moments must have
large error, and you must make your quantum chemistry program output more configuration
coefficients.
Then you will be prompted to select a task, there are four choices:
(1) Output transition electric dipole moments between all states (including both ground state
and excited states) to screen
(2) The same as (1), but output to transdipmom.txt in current folder.
(3) Generate input file of SOS module of Multiwfn as SOS.txt in current folder. Then if you
use the SOS.txt as input file, you can use SOS module to evaluate (hyper)polarizability, see Section
3.200.8 for detail.
(4) Output electric dipole moment of each excited state to dipmom.txt in current folder. Note
that both electronic and nuclear contribution to the value are taken into account.
The calculation process of all the four tasks consists of three stages:
Stage 1: Calculate dipole moment integrals between all basis functions
Stage 2: Calculate dipole moment integrals between all MOs
Stage 3: Calculate dipole moment integrals between all excited states.
If the output file of quantum chemistry program includes both singlet and triplet excited states,
for example, you used TD(50-50) keyword in Gaussian and reference state is closed-shell, only
aforementioned tasks (1) and (2) are available, and transition dipole moment of all singlet-singlet
pairs (including ground state) and triplet-triplet pairs will be calculated by Multiwfn and outputted
separately, while singlet-triplet pairs are ignored because due to spin-forbidden the result must be
zero. In addition, excitation energies between S0 and all excited states are printed at the end of
output. This function is of great importance if you want to use PySOC code to calculate spin-orbit
coupling matrix element, see my blog article "Using Gaussian+PySOC to calculate spin-orbit
coupling matrix element under TDDFT" (http://sobereva.com/411, in Chinese) for detail.
Below is a simple example of calculating dipole moment of each state. The corresponding
Gaussian input file for generating the .fch and .out files is examples\excit\4-nitroaniline.gjf. Boot up
Muliwfn and input
examples\excit\4-nitroaniline.fch
18 // Electronic excitation analysis
218
3 Functions
5 // Present function
examples\excit\4-nitroaniline.out
4 // Obtain dipole moment of each state
Then you will have dipmom.txt in current folder, the content is:
Note: The dipole moments shown below include both nuclear charge and electronic
contributions
Ground state dipole moment in X,Y,Z: 0.326322 -2.792165 -0.000000 a.u.
Excited state dipole moments (a.u.):
State X Y Z exc.(eV) exc.(nm)
1 0.334929 -1.219854 -0.000000 4.0557 305.92
2 0.251666 -7.797482 -0.000000 4.2762 290.14
3 0.334065 -1.439663 -0.000000 4.5846 270.62
Hint: There is a parameter "maxloadexc" in settings.ini, if this value is not 0 (default) and the
actual number of excited states is higher than this value, then only the first "maxloadexc" excited
states will be loaded and subjected to transition electric dipole moment calculation.
3.21.6 Generate natural transition orbitals (NTOs) (6)
Theory
This function is used to generate natural transition orbitals (NTOs). NTO was proposed in J.
Chem. Phys., 118, 4775 (2003), it has become a very popular and useful way to analyze character
electron excitation obtained by single-reference methods.
Transition of electronic state is often not predominated by only one MO pair, in many cases
multiple MO pair transitions simultaneously have non-negligible contributions, which can be
evaluated as square of corresponding configuration coefficient. This fact brings great hindrance of
analyzing electron excitation character by simply visualizing related MOs. The NTO method aims
to relieve this difficulty, it separately performs unitary transformation for occupied MOs and virtual
MOs, so that only one or very few number of orbital pairs have dominant contributions.
The basic procedure of yielding NTOs is outlined below:
(1) Generating transition density matrix in MO basis (T). Assume that the system has nocc
occupied MOs and nvir virtual MOs, then T has dimension of (nocc, nvir), its (i,l) element is simply
constructed as
Ti ,l = wia
a
where i<nocc, l<nvir and a=l+nocc. wi stands for configuration coefficient corresponding to i→a
orbital transition. Note that for TD formalism, there may be some de-excitations, their configuration
coefficients are simply ignored in constructing the T.
(2) Generating temporary matrix for occupied and virtual orbitals, respectively
219
3 Functions
Tocc = TTT Tvir = TT T

Evidently, both Tocc and Tvir are square matrices, their dimension are nocc and nvir, respectively.
(3) Diagonalizing Tocc and Tvir to obtain eigenvalues and eigenvectors
−1
U occ ToccU occ = Λ occ U −vir1 TvirU vir = Λ vir
(4) The diagonal terms of occ and vir are eigenvalues of occupied and virtual NTOs,
respectively. For the former, the eigenvalues are commonly sorted from low to high, while for the
latter, the eigenvalues are commonly sorted from high to low. A NTO pair consists of an occupied
NTO and a virtual NTO sharing the same eigenvalue. Eigenvalue of a NTO pair multiplied by 100
is just its percentage contribution to the electron excitation.
For CIS and TDA-DFT, the range of eigenvalue must be 0.0~1.0. However, in the TDHF and
TDDFT cases, due to presence of de-excitations, which is not explicitly considered in the NTO
analysis, it is possible that a NTO pair has eigenvalue slightly larger than 1.0, in this situation you
can simply treat it as 1.0 (However, if the value is much larger than 1.0, the TDDFT result may be
unreliable, and I suggest you use TDA-DFT instead).
(5) MOs are transformed to NTOs via unitary transformation matrix U
NTO
Cocc = Cocc
MO
U occ NTO
Cvir = CMO
vir U vir
MO MO
where Cocc and C vir are coefficient matrix of occupied MOs and virtual MOs in original basis
functions, respectively; their columns correspond to different MOs. The counterpart matrices with
NTO superscript denote coefficient matrix of occupied and virtual NTOs.
Implementation and Usage

After you enter present function, you should input the index of the electron excitation to be
studied, then Multiwfn will load corresponding configuration coefficients and generate NTOs
according to the equations shown above, and then output eigenvalues of occupied and virtual NTOs.
Next, you can choose whether exporting the NTOs to .fch or .molden file. If you choose to export,
then you can use Multiwfn to load the newly generated .fch or .molden file to visualize the NTOs,
analyze NTO orbital composition and so on (note that in this case the data in orbital energy field in
fact is NTO eigenvalues).
Present function works for both restricted and unrestricted reference states; for the latter, the
result of Alpha part and Beta part are calculated and printed separately, you only need to pay
attention to the NTO pairs having largest eigenvalues (e.g. the largest eigenvalue of Alpha part is
0.03, while the largest eigenvalue of Beta part of 0.95, that means this electron excitation is
dominated by transition of the Beta NTO pair)
According to my experiences, NTO analysis often works equally well as the hole-electron
analysis introduced in Section 3.21.1, namely both of them are able to avoid necessity of inspecting
many MOs when discussing electron excitation. An additional advantage of NTO analysis over hole-
electron analysis is that the orbital phase information is retained; however, in some cases NTO
analysis completely fails, namely even after transformation from MO to NTO representation, there
220
3 Functions
is still no dominant orbital pair transition. Clearly in this case you have to resort to hole-electron
analysis.
An example of generating and analyzing NTOs is provided in Section 4.18.6.
Information needed: See beginning of Section 3.21
3.21.7 Calculate ghost-hunter index (7)
In J. Comput. Chem., 38, 2151 (2017), Adamo et al. proposed an index used to diagnose if an
excited state yielded by TDDFT calculation is a ghost state. The ghost states are artificial low-lying
CT states resulting from evidently incorrect asymptotic behavior of exchange potential of pure DFT
functionals or the hybrid functionals with low Hartree-Fock exchange composition. Since the ghost
states are unreal states, they should be ignored when discussing electron excitation problems. If
there are many ghost states which significantly polluted electronic spectrum, then DFT functionals
with relatively high HF exchange composition (e.g. M06-2X and BH&HLYP) or long-range
corrected functional (e.g. B97X and CAM-B3LYP) must be used to get rid of them. The ghost-
hunter index (MAC) aims to present a quantitative criterion to judge if an excited state should be
regarded as a ghost state.
The ghost-hunter index in its original paper is defined as
wia ( − i −  a ) 1
M AC =  −
i,a  wi
i,a
a
DCT
where i and a denote occupied and virtual MOs, respectively. The summation loops all TDDFT
configurations, the w is configuration coefficient.  stands for MO energy. DCT is charge-transfer
index (see below). MAC is essentially an approximate estimation of lower limit of CT excitation
energy.
In my opinion, above definition of MAC is somewhat peculiar, since it employs wia / w
i,a
a
i
as weight of i→a transition rather than ( wia ) 2 / (w

i,a
a 2
i) as expected. In addition, it does not
consider excitation and de-excitation separately (i.e. the summation simply loops all configurations
irrespective of excitation and de-excitation). Taking these problems into account, I defined an
another form of MAC:
( wia ) 2 ( − i −  a ) 1
M AC = −
i →a  ( wi )
i →a
a 2
DCT
In which all de-excitations are ignored in the summation.

In the original paper, it is argued that if calculated excitation energy is lower than MAC, then
the excited state should be regarded as ghost state, and the result should be dealt with caution.
However, according to my personal experience, this criterion is often too strong.
In Multiwfn, the MAC is not calculated in subfunction 7 of main function 18. To calculate it,
you should perform usual hole-electron analysis (see introduction in Section 3.21.1 and example in
Section 4.18.1). Once calculation of grid data of hole and electron is finished, Multiwfn
221
3 Functions
automatically prints the MAC index as well as its two terms (their difference corresponds to the MAC
index). The MAC index is calculated and shown in two forms, the definition 1 corresponds to the
formula in original paper, while the definition 2 corresponds to the formula defined by me, as shown
above.
Beware that in Multiwfn, the DCT used in the MAC expression is evaluated as distance between
centroid of electron and hole distributions, it is more or less different to the DCT used in MAC original
paper (referred to as DCT' below), which is calculated as centroid distance between positive and
negative parts of density difference between relaxed excited state density and ground state density.
The DCT used by Multiwfn is not only reasonable enough, but also much cheaper than DCT' (since
evaluating TDDFT relaxed density for large system is fairly expensive). However, if you really want
to calculate MAC index based on DCT', you should get the first term of MAC via electron-hole module,
and then calculate the second term (i.e. DCT') via subfunction 3 of main function 18 (see the example
given in Section 4.18.3), and then manually get their difference.
It is worth to mention that the MAC index is only applicable to one-dimension CT problem, if
the CT takes place in multiple directions, then this index is incapable and must be properly
generalized.
Calculation of MAC is involved in the example in Section 4.18.1.
3.21.8 Calculate interfragment charge transfer in electron excitation

via IFCT method (8)
Theory
Interfragment charge transfer is a very important phenomenon in electron excitation process. I
devised an albeit simple but quite useful way of evaluating amount of interfragment charge transfer
between any number of fragments, the method is described below (to be published publicly). This
method will be referred to as IFCT (InterFragment Charge Transfer).
The IFCT method contains three steps:
(1) Calculating atomic contribution to hole and electron (see introduction of the concept of hole
and electron in Section 3.21.1.1)
(2) Calculating fragment contributions to hole and electron by summing up atomic
contributions
(3) Constructing interfragment charge transfer matrix Q. Its (R,S) element corresponds to the
electron transfer from fragment R to fragment S during the excitation:
QR, S = R, holeS ,ele
where R,hole and S,ele denote contribution of fragment R to hole and contribution of fragment S to
electron, respectively. Above formula is very easy to comprehend, it essentially assumes that
electron transfer from R to S is proportional to both composition of R in hole (where electron leaves)
and composition of S in electron (where electron goes).
222
3 Functions
Then three additional useful quantities could be defined:
·Electron net transferred from fragments S to R: pS → R = QS , R − QR, S
·Variation of electron population of fragment R: pR = p

S R
S →R = (Q
S R
S,R − QR , S )
·Intrafragment electron redistribution of fragment R: QR,R
By the way, it is easy to show that variation of electron population of a fragment evaluated in
above way is quite reasonable:
pR =  (Q
S R
S,R − QR , S ) =  (
S R
 R , ele −  R , holeS ,ele )
S , hole
=  R ,ele  S , hole −  R , hole  S ,ele

S R S R
=  R ,ele (1 −  R , hole) −  R , hole(1 −  R , ele )

=  R ,ele −  R , hole
It is easy to comprehend that this is a quite reasonable way of evaluating variation of electron
population of fragment R, and thus well demonstrated reasonabless of the interfragment charge
analysis formalism introduced above.
In addition, it is worth to note that sum of amount of interfragment transferred electrons and
amount of intrafragment redistribution electrons exactly equals to unity, reflecting the fact that only
one electron is excited:
 Q
R S R
R,S +  QS , S
S
=  QR , S
R S
=   R , holeS ,ele
R S
=   R , hole  S , ele
R S
=1
Usage
The input files needed by the IFCT analysis have been detailedly described at the beginning of
After entering this function, you need to choose the excited state to be studied, then input the
total number of fragments, after that you should define each fragment in turn by inputting atomic
indices. If you prefer to load fragment definition from a plain text file, you can input 0 when
Multiwfn let you set the total number of fragments, then you can input the path of the file containing
fragment defintion, the file format should look like below, definition of each fragment occupies a
line:
1,3,6-10,12
2,4,5
11
13-15
223
3 Functions
If in this step you input -1, Multiwfn will not carry out regular IFCT analysis but export a file named
atmCTmat.txt in current folder, this file records atom-atom charge transfer matrix, whose element
is defined as QA, B =  A, holeB,ele . If you input path of this file after entering the function used to
plot atom/fragment transition matrix (see Section 3.21.2), this matrix could be plotted as heat map
so that you can visually study its matrix elements.
Once definition of fragments is completed, Multiwfn will calculate and print contribution of
all defined fragments to hole and electron, as well as amount of electron transfer between fragments.
In addition, net electron transfer as well as variation of electron population of each fragment are
also printed. Below is an output instance:
Variation of population number of fragment 1: -0.25313
Variation of population number of fragment 3: 0.48423
Intrafragment electron redistribution of fragment 1: 0.00334
Transferred electrons between fragments:
1 -> 2: 0.11977 1 <- 2: 0.00874 Net 1 -> 2: 0.11103
1 -> 3: 0.14299 1 <- 3: 0.00089 Net 1 -> 3: 0.14210
2 -> 3: 0.36476 2 <- 3: 0.02263 Net 2 -> 3: 0.34213
By default, Mulliken-like method is automatically employed to calculate composition of hole

and electron (see "Theory 2" of Section 3.21.1.1). This method is fast, however diffuse functions
must not be used in this case, otherwise the result may be misleading. If you prefer to use the more
robust Becke partition to evaluate composition of hole and electron, you should first use hole-
analysis module to yield cube file of hole and electron, then if hole.cub and electron.cub can be
found in current folder when entering IFCT module, Multiwfn will ask you if carrying out IFCT
analysis for the hole and electron recorded in these two cube files based on Becke partition.
An example of IFCT analysis is given in Section 4.18.8.
3.21.9 Generate and export transition density matrix (9)
According to MO expansion coefficients and configuration coefficients, transition density

matrix (TDM) in basis function representation can be constructed by this function. Two kinds of
TDMs can be generated:
(1) TDM between ground state and a selected excited state K:
224
3 Functions
occ vir
Ptran =  wiK,aCiC a
i a
where w corresponds to coefficient of the configurations involved in the excitation, Ci denotes the
expansion coefficient of basis function  in MO i. Excitation and de-excitation cases are not
distinguished.
(2) TDM between two selected excited states K and L:
occ vir occ vir
PKL =  wiK,a  w Lj ,bViajb
i a j b
 CaC b (i = j , a  b )
 −C C (i  j , a = b )

Viajb =  i  j
 P − CiC i + CaC a (i = j , a = b )
 0 (i  j , a  b )
where P is density matrix of ground state. For TD case, the V between excitation and de-excitation
configurations is simply ignored. In addition, when calculating V between de-excitation
configurations, it is replaced with -V.
Once generation of TDM has been finished, you can choose if symmetrizing the TDM. There
are two ways
 = ( P  + P ) / 2
Ptran tran tran
• Way 1:
 = ( P  + P ) / 2
• Way 2: Ptran tran tran
The way 1 is reasonable and should be used in common case. However, it should be noted that the
TDM generated by Gaussian program corresponds to the one symmetrized by way 2, therefore you
should choose way 2 if you want the resulting TDM follows convention of Gaussian.
The generated matrix will be outputted to tdmat.txt in current folder. You can also choose to
output TDM.fch in current folder, whose “Total SCF Density” field will correspond to TDM (this
file is useful if you would like to calculate TrEsp type of atomic transition charges by making use
of cubegen utility in Gaussian, see Section 4.A.9 for detail).
Note that when ground state and excited state have different spin multiplicities, due to the
orthonormality of spin coordinate, although in principle the transition density matrix should be zero,
the outputted matrix is not, because Multiwfn only takes spatial part of the MOs into account during
constructing the matrix.
The example in Section 4.18.2.4 and Section 4.18.9 utilized present function.
Appendix: Derivation of the formula of evaluating TDM between two excited states
The TDM between two excited states K and L in real space representation is
225
3 Functions
T KL (r; r ') =   K (r, r2 , , rN ) L (r ', r2 , , rN )d r2 d rN
where the excited state wavefunctions are represented by linear combination of singly excited Slater
determinants
occ vir occ vir
 =  w 
K K
i ,a
a
i  =  wLj ,b bj
L
i a j b
We have
occ vir occ vir
T KL (r; r ') =  wiK,a  wLj ,b  ia (r, r2 , , rN ) bj (r ', r2 , , rN ) d r2 d rN
i a j b
̂ = ∑𝑖 ℎ𝑖 , integral between two

Slater-Condon rule shows that for an single-electron operator ℵ
𝑎 ̂ 𝑏
singly excited determinants, namely ⟨𝑖 |ℵ|𝑗 ⟩, satisfies
=0 (i  j , a  b )
= ahb (i = j , a  b )
=− j h i (i  j , a = b )
N
= p h p − i h i + a h a (i = j , a = b)
p
Without performing the integral and view the h operator as 1, based on the above relations we
have
 0 (i  j , a  b )

  wi ,a wi ,b a (r ) b (r ') (i = j, a  b)
K L

i a b
T KL (r; r ') =  −  wiK,a w Lj ,a j (r )i (r ') (i  j, a = b)

 i j a
 L 
N

  wi ,a wi ,a   p (r ) p (r ') − i (r )i (r ') +  a (r ) a (r ')  (i = j, a = b)
K
 i a  p 
Given that
T KL (r; r ') =  PKL   (r )  (r ')
 
 k (r ) =  C  k   ( r )

where  is basis function, we can finally reach the formula or evaluating PKL shown earlier in this
section. For example, in the case of i=j, ab:
T KL (r; r ') =  wiK,a wiL,b a (r )b (r ')
i a b
=  wiK,a wiL,b  CaC b   (r )  (r ')

i a b  
P KL
=  w w CaC b K
i ,a
L
i ,b
i a b
226
3 Functions
3.21.10 Decompose transition dipole moment as molecular orbital pair

contributions (10)
Oscillator strength (f) of an electron excitation directly relates to the integral area of the
corresponding absorption peak. f has direct relationship with transition electric dipole moment D (in
atomic unit)
2
f = E  ( Dx + Dy + Dz )
2 2 2
3
where E denotes the transition energy between the two electronic states. Clearly, transition electric
dipole moment is a crucial quantity of electron excitations.
The three Cartesian components of transition electric dipole moment between ground state and
an excited state can be calculated as follows
Dx = − wia i x  a Dy = − wia i y  a Dz = − wia i z  a
i,a i,a i,a
where i and a loop over all occupied and virtual MOs, respectively. w is configuration coefficient
(irrespective of excitation and de-excitation here),  denotes molecular orbital, x/y/z stands for
Cartesian components of position vector. It is clear that the transition dipole moment can be
straightforwardly decomposed into contribution of various MO pairs. Via such a decomposition,
one can easily study why some excitations have relatively large oscillator strength and thus have
strong absorption, and why some excitations only have small oscillator strength and thus they are
difficult to be observed in electronic spectrum.
boots up, and then load a file containing configuration coefficient information of excited states when
you enter this function.
In this function, you will be prompted to choose the excited state for which the transition
electric dipole moment will be decomposed as MO pairs, then a menu appears. You can select
corresponding option to make Multiwfn output contribution of every MO pair to transdip.txt in
current folder, or let Multiwfn sort the MO pairs according to their absolute contribution to certain
component of transition electric dipole moment and then output the first few or dozens of terms, so
that you can immediately capture the most important orbital pair transitions. In addition, you can
ask the program to only output terms that contribute more than a given threshold.
Below is an output example, the terms are sorted according to absolute contribution to Y
component of transition electric dipole moment and the first five terms are requested to be printed:
#Pair Coefficient Transition dipole X/Y/Z (a.u.)
1082 36 -> 37 0.697700 -0.046386 -2.938793 -0.000000
1055 35 -> 38 0.096590 -0.001156 0.278865 -0.000000
2141 36 <- 37 -0.056630 0.003765 0.238532 0.000000
1083 36 -> 39 -0.047340 0.001138 0.065117 -0.000000
2118 35 <- 38 0.020870 -0.000250 0.060254 -0.000000
An example of this function is given in Section 4.18.10.
227
3 Functions
3.21.11 Decompose transition dipole moment as basis function and

atom contributions (11)
This function is used to decompose the total transition electric or magnetic dipole moment
between ground state and a selected excited state into contributions from each basis function and
atom. Therefore, you can infer which part of the system has significant impact on excitation
properties such as oscillator strength. The result will be exported to trdipcontri.txt.
There are many possible ways to realize the decomposition. In this function Mulliken-like
partition is employed due to its simplicity. For example, the contribution of basis function μ to Z
component of transition electric dipole moment is evaluated as follows
   − z   + [ P 
Dz(  ) = Ptran   − z  + Ptran  − z   ] / 2
tran
 
and the contribution of basis function μ to Z component of transition magnetic dipole moment
 
M z(  ) = Ptran
  x −y  +
y x
     
  P  tran
 x
y
−y
x
 + Ptran  x − y
y x
  / 2
 
  
Then the contribution from an atom is simply the sum of the contribution from the basis functions
belonging to it.
Since Mulliken partition is incompatible with diffuse functions, hence the decomposition result
is unreliable if diffuse functions are presented in the basis set you used. In this case, the best way to
study contribution from various atoms is visualizing the transition dipole moment density (see
Section 3.21.1).
This function also asks you if outputting atom transition dipole moment matrix, if you choose
y, then X, Y, Z components of the matrix will be exported to AAtrdipX.txt, AAtrdipY.txt, AAtrdipZ.txt
in current folder, respectively, the matrix elements are defined as follows (I take atom-atom
contribution matrix of transition electric dipole moment as example, the matrix for transition
magnetic dipole moment is defined similarly and thus not explicitly shown here)
DAX, B =  P 
 
 A B
tran
 − x 
DYA, B =  P 
 
 A B
tran
 − y 
DAZ, B =  P 
 
 A B
tran
 − z 
For example, the term DAX, B corresponds to joint contribution of A-B atom pair to X component of
transition dipole moment, the sum of all elements of DX equals to X component of transition dipole
moment of current system. Total transition dipole moment matrix (sum of square of X, Y, Z) is
exported as AAtrdip.txt in current folder. All of these .txt files can be directly plotted as colored
matrix map (heat map) by atom transition matrix plotting module (see Section 3.21.2 for detail).
228
3 Functions
you enter this function.
The example of Section 4.18.11 utilized this function.
3.21.12 Calculate Mulliken atomic transition charges (12)
This function is used to calculate atomic transition charges, which is useful for studying
Coulomb coupling between ground state and excited state (excitonic coupling) of two molecules,
see e.g. J. Phys. Chem. B, 110, 17268 (2006) and Photosynth. Res., 111, 47 (2012).
Transition population of a basis function μ derived by Mulliken method is
tran  +  S   ( P  + P ) / 2
= Ptran tran tran

 
So, the Mulliken atomic transition charge of atom A should be −  

 A
tran
. Sum of all atomic
transition charges must be zero because the total number of electrons keeps unchanged during
electron excitation.
you enter this function. After that, you should choose the excited state for which the Mulliken
transition charges will be calculated. Then the result will be outputted to atmtrchg.chg file in current
folder, the format of this kind of file has been introduced in Section 2.5, the last column of this file
corresponds to the transition charges.
Below is an example of the calculation. Boot up Multiwfn and input

examples\excit\N-phenylpyrrole.fch
18 // Electron excitation analysis
12 // Calculate Mulliken transition charges
N-phenylpyrrole.out
3 // Study the transition from ground state to the third excited state
Then you will find atmtrchg.chg in current folder.
Note that in Multiwfn it is also possible to calculate the TrEsp (transition charge from
electrostatic potential) introduced in J. Phys. Chem. B, 110, 17268 (2006), which is derived by ESP
fitting method based on transition density. See Section 4.A.9 on how to do this. For studying
excitonic coupling purpose, TrEsp should work better than Mulliken atomic transition charge, but
for large systems, cost of evaluating the former is is significantly higher than the latter.
229
3 Functions
3.21.13 Generate natural orbitals of specific excited states (13)
This function is used to generate natural orbitals (NOs) for a batch of selected excited states,
and then export the NOs as .molden file. After that, if you want to perform wavefunction analysis
for an excited state, you can simply load corresponding .molden file. Of course, you can also
calculate e.g. density difference between two excited states using corresponding two .molden files
via custom operation feature of main functions 3, 4 and 5. This function supports both closed-shell
and open-shell reference state.
To use this function, you should load a file containing basis function information, and then load
a file containing configuration coefficients when you enter this file, see beginning of Section 3.21
for detail. After that, you will be prompted to input the indices of the excited state for which NOs
will be generated. For each selected excited state, the program will do below steps:
(1) Generating density matrix of excited state:
PES = PGS +  ( wia )2 ( −Pi + Pa ) +  ( wia )2 ( −Pa + Pi )
i →a i a
where PGS is density matrix of ground state wavefunction, The matrix like Pi is density matrix
constructed solely by orbital i, it can be evaluated as below, where Ci is column vector of expansion
coefficients of orbital i
Pi = Ci CiT
Note that the density matrix constructed in this way corresponds to unrelaxed density.
(2) Diagonalizing the PES to yield NOs. Each NO is an eigenvector of PES, the accompanying
eigenvalue is occupation number of the NO.
(3) Exporting information of basis function and NOs to .molden file. If the excited state you
selected is 2, then they will be exported as NO_00002.molden in current folder.
The example given in Section 4.18.13 fully utilized this function.
3.21.14 Calculate Λ index to characterize electron excitation (14)
Theory
In the paper J. Chem. Phys., 128, 044118 (2008),  index was proposed to distinguish types of
electron excitations. The form of the  index and the r index (see Section 3.21.4) is very similar.
 index can be expressed as
 =  ai
i,a
where  i is contribution of MO transition between i and a to the  index:

a
( Kia )2
 ( Kia )2  i a
 =
a
i  (r )  ( r ) d r
i, a
230
3 Functions
All quantities involved in above expression are identical those in r index. The integral corresponds
to overlap extent of the MO i and a, it is calculated numerically via Becke's mult-center grid-based
integration approach. The default grid is a good compromise between cost and accuracy; if you want
to change it, you can set "iautogrid" in settings.ini to 0 and then specify "radpot" and "sphpot" in
settings.ini as your expected values. Note that calculation cost of the  index is by far higher than
the r index, because of the above numerical integral step is expensive, especially for large systems.
The theoretical lower and upper limits of the  index are 0.0 and 1.0, respectively; the former
(latter) corresponds to the case that hole and electron are completely separated (perfectly
overlapped).
As the r index, the  index is useful for distinguishing type of electron excitations. Notice
that their intrinsic characteristics are different, the r index is essentially an indicator of
configuration weighted orbital separation distance, while  index reflects configuration weighted
orbital overlapping extent. In some sense, the physical nature of r and  indices are similar to the
D and Sr indices defined in hole-electron analysis framework, respectively (see Section 3.21.1.1),
however I believe that the D and Sr indices are more reasonable, since their physical meanings are
more clear and couplings between different configurations are fully taken into account. Therefore,
without special reasons, using D and Sr indices is more recommended.
It is worth to mention that if an electron excitation can be perfectly represented by one pair of
MO transition, then the  index and Sr index defined in hole-electron analysis framework will be
exactly identical in principle:
 =   i (r )  a ( r ) d r
Sr =   hole (r )  ele (r ) d r =  i (r )  a (r ) d r =  i (r )  a (r) d r

2 2
However, their values outputted by Multiwfn should be marginally different, since they are
evaluated based on different numerical integration algorithms.
Usage
After entering present function (subfunction 14 of main function 18), the matrix containing
overlap integral between norms of all occupied and unoccupied MOs will be evaluated first, then
you will be prompted to select the excited states for which the  will be calculated, then the results
will be printed on screen immediately.
If you only selected one state, then Multiwfn will ask you to choose if decomposing the  into
orbital pair contributions. If you inputted e.g. 0.01, then orbital pairs which have contribution to 
larger than 0.01 will be printed.
An example of present function is provided as Section 4.18.4.
231
3 Functions
3.21.15 Print major MO transitions in all excited states
This is a useful function used to show major MO transitions for all excited states, so that you
can quickly recognize basic characteristics of various excited states in terms of MOs.
Below is an example. Boot up Multiwfn and input
examples\excit\D-pi-A.out // Output file of TDDFT task of Gaussian
15 // The present function
You can see below information immediately, including excitation energy, spin multiplicity,
notable MO transitions and their contributions of each excited state.
# 1 3.9069 eV 317.35 nm f= 0.01880 Spin multiplicity= 1:
H-4 -> L 81.9%, H-4 -> L+2 12.1%
H -> L 86.0%, H-3 -> L 5.3%
H-6 -> L 85.3%, H-6 -> L+2 11.9%
H-2 -> L 54.5%, H -> L+1 27.6%, H-3 -> L+1 6.4%
H -> L+3 57.3%, H-2 -> L 17.0%, H-1 -> L+2 8.8%, H-1 -> L 8.0%
From above output, for example, we can find HOMO-4 → LUMO transition contributes 81.9% to
the excitation from ground state to S1 state.
For open-shell cases, orbital spins are explicitly indicated. For example, Ha-4 means
HOMOalpha-4.
By default, only MO transitions with contribution larger than 5% are printed. The printing
threshold corresponds to 10 times of "compthres" parameter in settings.ini.
You can use output file of ZINDO/CIS/TDHF/TDA-DFT/TDDFT task of Gaussian, ORCA,
GAMESS-US/Firefly as input file. Unlike most functions in main function 18, the file containing
basis function information is not needed in the present function.
3.22 Orbital localization analysis (19)
Theory and algorithm overview

Canonical molecular orbitals (CMOs) often show strong delocalization character and thus do
not convey useful information about chemical bonding. There are many ways to localize MOs, the
most popular ones are Foster-Boys (FB) localization, Edmiston–Ruedenberg (ER) localization and
Pipek–Mezey (PM) localization. The NLMO method supported in NBO program is also a kind of
orbital localization algorithm. The resulting orbitals from these methods are known as localized
molecular orbitals (LMOs). Both the LMOs and CMOs are orthonormal sets and have identical
dimension, they can be transformed to each other via unitary transformation.
The FB is the oldest orbital localization method, it was proposed in Rev. Mod. Phys., 32, 300
(1960). This method minimizes below quantity, so that the spatial distribution range of all orbitals
become as small as possible
232
3 Functions
 Boys
=    i (r1 )(r1 − r2 )2 i (r2 ) d r1 d r2 i =| i |2
i
The FB method is very popular and widely used, so it is supported by Multiwfn.

The ER localization proposed in Rev. Mod. Phys., 35, 457 (1963) is also a well-known method,
it localizes orbitals via maximizing below quantity (orbital self-repulsion integral)
1
 =    i (r1 ) i (r2 ) d r1 d r2
ER
i | r1 − r2 |
The ER method is highly deprecated, because it needs evaluation of two-electrons integral, which
is very complicated; furthermore, transformation of the integral from AO basis to MO basis is very
expensive. Although in a few papers some people argue that ER method has better physical meaning
and the computational cost can be considerably reduced via introducing resolution-of-identity
technique, I never think there is any convincing reason to employ ER method instead of FB, so ER
method is not supported by Multiwfn.
The most popular orbital localization method is PM. The essence of PM localization is to
maximize below quantity, so that distribution range of all orbitals can be shrinked as much as
possible
P =  ( p iA ) 2
i A
In the original paper of PM method J. Chem. Phys., 90, 4916 (1989), pAi corresponds to Mulliken
population of atom A in MO i. While in J. Chem. Theory Comput., 10, 642 (2014), it is shown that
other population methods such as Löwdin, Hirshfeld, Becke, AIM can also be combinedly used with
PM and obtaining reasonable result. Currently PM method based on Mulliken and Löwdin
populations are supported in Multiwfn.
The maximation or minimization of above mentioned quantities can be done via Jacob sweep
algorithm, which is employed in the original paper of the orbital localization methods and still be
prevalently used until now. This method has lower efficiency than the later developed sophisticated
methods such as unitary optimization and trust region; however, the Jacob sweep is very simple and
works for most cases, in particular when only occupied orbitals are needed to be localized, therefore
this algorithm is employed in Multiwfn.
The working equation of PM localization based on various population methods are largely
identical, see Eq. 9 of J. Comput. Chem., 14, 736 (1993), they only differ in the definition of the
term Q, which is needed to be computed in each iteration of Jacob sweep for every orbital pair. For
Mulliken population, the term corresponding to orbitals i and j for atom A is
Q Aij = 1
2 [C C
A 
i j + Ci Cj ]S  
where  and  correspond to basis function index and the latter cycles all basis functions. While for
Löwdin population, because the basis functions have been orthonormalized, the term Q is simplified
as
Q Aij =  Ci Cj
A
PM-Löwdin is seemingly much cheaper than PM-Mulliken; however, if properly programming, the
233
3 Functions
cost of two methods are basically identical, because the Q for Mulliken case can be reformulated as
Q Aij = 1
2 [C (SC) 
A
j i + Ci (SC) j ]
If the SC matrix is calculated and stored in memory before Jacob sweep and updated frequently
during iteration, the summation for index  can be completely ignored. In fact, because symmetric
diagonalization step of Löwdin method is time-consuming for large system, the overall cost of PM-
Mulliken is generally lower than PM-Löwdin, therefore PM-Mulliken is taken as the default
localization method. Regarding the FB method, because it involves transformation of dipole
moment integrals from AO to MO basis, which requires large amount of arithmetic operations, it
can only be employed for much smaller system compared to PM-Mulliken and PM-Löwdin.
Usage
The input file must contain basis function information, thus you can use
e.g. .mwfn, .fch, .molden or .gms as input file. This function only works for restricted and
unrestricted SCF wavefunction.
Due to the robust and very low cost of PM-Mulliken method, it is chosen as the default orbital
localization method in Multiwfn. If you want to change to PM-Mulliken or FB methods, use option
"-6 Set localization method".
In the interface, you can use option 1 to choose to localize occupied MOs only, or use option 2
to localize both occupied and unoccupied MOs (the two set of orbitals will be localized separately,
i.e. no mixture between occupied and unoccupied orbitals is allowed). For unrestricted
wavefunctions, the alpha and beta parts are treated separately. Via option 3 you can localize specific
subset of MOs; in other words, only specific MOs are allows to be mixed during localization. This
feature enables you to realize special purposes, for example, obtaining fully or semi- localized MOs
in certain region by selecting proper MOs.
Orbital localization is an iterative process, thus you should set criterion of convergence and
maximum number of cycles. The default values are commonly appropriate and need not to be
modified. The convergence status is printed during iteration, for all localizations methods, the
change of P =  ( p
i A
i 2
A) is used for judging convergence.
Once the localization is converged, orbital composition of all resulting LMOs will be calculated
and major characters of the LMOs are printed. By default, the robust Hirshfeld method is employed
for evaluating orbital compositions (see Section 3.9 for detail), but you can also change to other
methods via option "-9 Set the method for calculating orbital composition" before the orbital
localization.
Finally, the LMOs are exported to new.fch in current folder, and then Multiwfn automatically
loads it, after that you can analyze the localized orbitals in various ways; for example, plotting them
as isosurfaces by main function 0 or performing orbital composition analysis by main function 8. If
you do not want to let Multiwfn automatically load the newly generated new.fch, you can choose
option -3 once to switch the status.
Hints on performing orbital localization analysis

In Multiwfn, the cost of PM-Löwdin and PM-Mulliken methods are proportional to Norb2Nbas,
while cost of FB is proportional to Norb2Nbas2, where Norb and Nbas are the number of orbitals to be
localized and the number of basis functions, respectively. Clearly, FB is much more expensive.
234
3 Functions
Convergence is usually more difficult for unoccupied orbitals than occupied orbitals, more
difficult for large system than small system, and more difficult for FB than PM method.
If you do not have special reasons, you only need to localize occupied orbitals, since only
occupied orbitals carry interesting information about electronic structure. Localization of
unoccupied orbitals is much more time-consuming than localizing occupied orbitals, since the
number of unoccupied orbitals is often very higher when extended basis set is used. Commonly, via
PM-Mulliken/Löwdin method, occupied orbitals can be easily localized for a system containing up
to 200 atoms with medium-sized basis set. While for FB, in general this work can only be realized
for a system containing up to 100 atoms.
If you want to decrease the cost of localization of occupied orbitals, you can choose "-5 If also
localizing core orbitals" once to switch the status from the default "Yes" to "No". Commonly, mix
between valence orbitals and core orbitals is rather weak, therefore ignoring inner-core orbitals
during localization of valence orbitals is safe. However, in rare cases, ignoring the inner-core orbitals
may causes difficulty in convergence.
PM method is not parallelized in Multiwfn, because I found that parallelization does not
improve its speed evidently but sometimes make convergence more difficult. FB method is fully
parallelized, thus using multi-cores CPU will reduce computational cost considerably.
Basis set of 2-Zeta with polarization functions quality (e.g. 6-31G*, def2-SVP) is completely
adequate for orbital localization analysis, using larger quality basis set never leads to detectably
improved result. It should be noticed that both PM-Löwdin and PM-Mulliken methods can not be
used when diffuse functions are presented, because Löwdin and Mulliken populations are
meaningless in this case. In contrast, FB localization is fully compatible with diffuse functions,
though it is evidently more expensive. When the basis set does not contain diffuse functions,
generally the resulting orbitals of all the above mentioned three methods are similar (except that FB
orbitals do not preserve - separation character as PM orbitals).
PS: It is noteworthy that if PM method is used in combination with such as Becke and Hirshfeld populations,
the PM method will become more insensitive to basis set and can be used safely with diffuse functions. However,
this combination is even much more expensive than FB method and thus does not show practical advantage, so this
idea is not implemented into Multiwfn.
Special topic 1: Evaluating LMO energies

If you want to obtain energy of the localized orbitals, you need to choose option “-4 If
calculating and print orbital energies”, and input path of a file containing Fock (or Kohn-Sham,
similarly hereinafter) matrix in original basis functions. The matrix elements should be provided in
lower-triangular form, namely in this sequence: F(1,1) F(2,1) F(2,2) F(3,1) F(3,2) F(3,3) ...
F(nbasis,nbasis), where nbasis is the total number of basis functions, the format is free. The Fock
matrix may be obtained from output of some quantum chemistry codes (In fact, you can ask
Gaussian to produce NBO .47 file, Multiwfn is able to automatically locate and read the $FOCK
field when the file name has .47 suffix). Once Fock matrix is loaded and transformed to localized
orbital representation, the energy of localized orbitals will correspond to diagonal terms of Fock
matrix. Specifically, Multiwfn performs below representation transform:
FLMO = CT FAO C
where FAO is the Fock matrix in original basis functions that loaded from external file, C(r,i)
corresponds to coefficient of basis function r in localized orbital i. Energy of localized orbital j is
simply FLMO(j,j).
235
3 Functions
Special topic 2: Revealing center of LMOs

To facilitate capturing basic distribution character of the generated LMOs, Multiwfn is able to
calculate center position of LMOs and add them as Bq atom (ghost atom) into current system, so
that you can use main function 0 to easily visualize them. The center of LMO is evaluated as follows
R i = i r i
where r is coordinate vector.
To generate the LMO centers, you should choose “-8 If calculating center position and dipole
moment of LMOs” once to switch its status to “Yes”. Then after generating LMOs, exporting .fch
and reloading it, the center positions of the LMOs will be evaluated and added as Bq atoms. The
coordinate of LMO centers as well as the correspondence between LMO indices and Bq indices will
be outputted to LMOcen.txt in current folder, meantime the setting of main function 0 will be set to
the best status for showing LMO centers (as illustrated in Section 4.19.1). Since the newly added
Bq atoms do not have accompanying basis functions, the current wavefunction should not be
subjected to wavefuction analyses, otherwise Multiwfn may crash or the result is completely
meaningless.
Note that if there are multiple bonds and PM localization algorithm is used, the Bq atom
corresponding to the center of -LMO and -LMO of the same bond may overlay with each other.
This can be avoided using FB algorithm instead, because FB represents multiple bond as multiple
banana LMOs, whose center positions are evidently different with each other.
Special topic 3: Dipole moment analysis for occupied LMOs

Once “-8 If calculating center position and dipole moment of LMOs” has been switched to
“Yes”, after performing orbital localization, you will be asked to choose if also performing dipole
moment analysis for occupied LMOs. If you input y, then you will have LMOdip.txt, which contains
dipole moment analysis result for all occupied LMOs. In order to make you correctly understand
the output, below I describe the details.
The contribution of electron of an occupied LMO to dipole moment of the whole system is
Di = i − r i
This vector for all LMOs is outputted as “Contributions of all occupied LMOs to system dipole
moment” in the LMOdip.txt file.
However, this quantity is unable to be directly used to measure polarity of a LMO. Given that
r = (r-rc)+rc, where rc is a fixed point, the above quantity can be rewritten as follows
Di = i − (r − rci ) i − i rci i = − i r − rci i − rci i i = − i r − rci i − rci

we can defined a quantity di, which measures dipole moment of the orbital with respect to the rc:
di = − i r − rci i = Di + rci
If we properly choose the rc for an orbital, then the di may be able to reflect polarity of the LMO.
For each LMO that identified as single-center one, the rc is automatically set to be the position
of the atom having largest contribution to the LMO. Therefore, the di represents deviation of
centroid of the LMO electron distribution with respect to the nuclear position. These {d} are printed
as “Single-center orbital dipole moments (a.u.)” in the LMOdip.txt.
For each LMO that identified as two-center one, assume that the two atoms with largest
236
3 Functions
contributions are A and B, the rc is set to
RB RA
rc = rA + rB
RA + RB RA + RB
where rA and RA are nuclear position and covalent radius of atom A, respectively. Similarly for atom
B. The rc locates at center of the bonding region, therefore the di, which exhibits deviation of
centroid of LMO electron distribution from rc, is capable of revealing the bond polarity. These {d}
are printed as “Two-center orbital dipole moments (a.u.)” in the LMOdip.txt.
For closed-shell case, the printed d are multiplied by a factor of 2, because the LMOs are
doubly occupied. For open-shell case, the d of alpha and beta LMOs are printed separately.
As byproducts, dipole moment of the whole system, as well as nuclear contribution and
electronic contribution are also printed at the beginning of the LMOdip.txt. It is important to notice
that even if there is no more delocalized LMOs, the sum of d of all LMOs is generally unequal to
the dipole moment of the whole system.
BTW: In fact, only when sum of the rc vectors in all d equals to − ∑𝐴 𝑍𝐴 𝐫𝐴, namely exactly
cancels the nuclear contribution to the system dipole moment, the sum of all d will be equal to the
system dipole moment. If one wants to satisfy this point, the (rA+rB)/2 should be employed as rc for
all two-center LMOs, and all LMOs should just correspond to a Lewis structure of current system.
Of course, these conditions are not met in present implementation of LMO analysis (but met in the
“DIPOLE” analysis of NBO theory). Note that if (rA+rB)/2 is taken as rc and meantime you use the
d of two-center LMOs to measure bond polarity, ridiculous result will be obtained, for example you
will find C-H is even much more polar than O-H!
Examples of orbital localization analysis are provided in Section 4.19. The examples of LOBA
method (Section 4.8.4) and Section 4.100.22 also utilized present function.
3.23 Visual study of weak interaction (20)
Visual study of weak interaction has become increasingly popular, and numerous related
analysis methods were put forward. Main function 20 of Multiwfn is a collection of these analyses
methods.
3.23.1 Noncovalent interaction (NCI) anaylsis (1)
The noncovalent interaction (NCI) method, which is also known as reduced density gradient
(RDG) method, is a very popular method for studying weak interaction. The theory of NCI method
is described in its original paper J. Am. Chem. Soc., 132, 6498 (2010). In this section, I will detailedly
introduce the basic idea of this method and illustrate how to realize it in Multiwfn. If you just want
to learn how to plot the color-filled RDG map, you can directly jump to “Part 3” of this section. It
is also strongly recommended to look at this video tutorial: https://youtu.be/e4FpVc9ao48, you will
very quickly learn how to plot various maps related to the NCI analysis.
237
3 Functions
If you can read Chinese, you are also suggested to check my blog article "Visual research of weak interaction
by Multiwfn" (in Chinese, see http://sobereva.com/68) and "Some key points and common problems of carrying out
RDG analysis via Multiwfn+VMD" (in Chinese, see http://sobereva.com/291).
Part 1: Using RDG isosurface to reveal weak interaction regions

How to visualize weak interaction? The first thing is to find a way to distinguish weak
interaction region from other regions. From the table given below we can find that if only the regions
where the value of reduced density gradient (RDG) function is in the range of 0~medium are
preserved, then “Around nuclei” and “Boundary of molecule” regions will be shielded.
Around nuclei Around chemical bond Weak interaction region Boundary of molecule
|(r)| Large 0~Minor 0~Small Very small~Small
(r) Large Medium Small 0~Small
RDG(r) Medium 0~Minor 0~ Medium Medium ~Very large
The definition of the RDG function is shown below, it is essentially a dimensionless form of electron
density gradient norm function
1  (r )
RDG(r ) =
2(3 )2 1/ 3
 (r ) 4 / 3
For remaining regions ("Around chemical bond" and " Weak interaction region"), if we only
keep the region where (r) is small, then only weak interaction region will be revealed.
Now I use phenol dimer to exemplify this idea, we will calculate grid data of RDG function
and visualize it as isosurface. Boot up Multiwfn and input following commands
examples\PhenolDimer.wfn // Any format containing GTF information can be used as input
file, see Section 2.5 for detail
5 // Generate grid data
13 // RDG function
7 // Use middle point of two atoms as center of grid data, this way of defining spatial scope
is very suitable for weak interaction analysis
1,14 // The indices of the two atoms are set to 1 and 14, because from molecular structure
(see below graph) we can estimate that the weak interaction region occurs between C1 and C14
40,40,40 // The weak interaction region is small, so 40*40*40=64000 grid is fine enough
3,3,3 // Set extension distance (buffer distance) in all X/Y/Z directions to 3 Bohr
-1 // Show the isosurface of RDG
Please make sure that the isovalue in the GUI window is set to 0.5, which is suitable for
visualizing weak interaction regions (if the isovalue is too small, then RDG isosurface will be too
thin and thus ugly; if too large, then unwanted “Around nuclei” and “Around chemical bond”
regions will appear). Now you can see below graph in the GUI window:
238
3 Functions
The green isosurface represents the weak interaction region between phenol dimer very clearly.
Notice that by default, RDG function is set to an arbitrarily large value (100.0) where electron
density is larger than or equal to 0.05, so that the isosurfaces in the region “around chemical bond”
can be shielded. The threshold is determined by “RDG_maxrho” parameter in settings.ini file. The
default 0.05 is suitable for visualizing weak interaction regions for most cases. If you do not want
to enable the screening treatment due to special reason, you can set “RDG_maxrho” to 0.
The cubic blue frame in above graph shows spatial scope of the calculated grid data, it is only
visible when “Show data range” is checked in the GUI window. Because the extension distance
from center of the grid data was set to 3.0 Bohr, the side length is 2*3=6 Bohr.
Part 2: Discriminating weak interaction types by filling color to RDG isosurfaces

In Bader’s AIM theory, appearance of a (3,-1) type of critical point (CP) usually implies that
electron density is locally aggregated, it commonly appears on bond path or between the atoms
which have attractive interaction. (3,+1) type of CP often implies that electron density is locally
depleted and exhibits steric effect, it generally occurs at center of a ring. The criterion for
distinguishing (3,-1) and (3,+1) CPs is the second largest eigenvalue of Hessian matrix of electron
density (referred to as 2 below). If 2 exceeds zero, then the CP is (3,+1), else it is (3,-1). Besides,
the strength of weak interaction has positive correlation with electron density  in corresponding
region. Van der Waals interaction regions always have very small , while the regions corresponding
to strong steric effect or evident attractive weak interaction (e.g. H-bond, Halogen bond) always
have relatively large . So we can define a real space function sign(2), namely the product of sign
of 2 and . If we use different colors to represent value of this function according to below color
bar, and map it on RDG isosurfaces, we can not only know where weak interaction occurs, but also
intuitively capture the type of the interaction.
239
3 Functions
High-resolution of the above labelled color bar is examples\RGB_bar.png, you can directly embed it into figures
of your paper.
Current Multiwfn does not support plotting color-filled isosurface graph, however, we can use
Multiwfn to generate cube file for sign(2) and RDG, and then use plotting script of VMD to draw
such map. VMD is one of the best visualization tool and can be freely downloaded at
http://www.ks.uiuc.edu/Research/vmd. Here I illustrate how to do this for phenol dimer by using
subfunction 1 of main function 20. This time we do not only want to study the weak interaction
region between the two monomers, but also want to examine the steric effect within in aromatic ring
of phenol, therefore the spatial scope of grid data should cover the entire dimer.
Boot up Multiwfn and input following commands
examples\PhenolDimer.wfn
20 // Visual study of weak interaction
1 // NCI analysis
-10 // Set extension distance in all directions with respect to molecular boundary
0 // Because weak interaction regions only appear in internal region of present system, we do
not need to leave a buffer region at system boundary, so we set the extension distance to 0 Bohr
2 // Medium quality grid (about 512000 points). Because the spatial scope of grid data is
evidently larger than last example, we need more grid points than last example, otherwise the the
RDG isosurfaces will look discrete
I first discuss the characteristics of various kinds of regions via scatter graph, I think it will be
helpful to understand the nature and idea of the NCI method. Select option -1 in the post-processing
menu, a scatter graph immediately pops up (you can also select option 1 to export this graph as file):
In the graph, the X-axis and Y-axis correspond to sign(2) and RDG functions, respectively;
each point in the graph corresponds to a grid point in 3D space. There are four spikes, the points at
their peaks are just the approximate CP positions in AIM theory. If you draw a horizontal line on the
graph as below, then the segments crossing the spikes just correspond to the points used to construct
240
3 Functions
the RDG isosurfaces. Hence, the NCI analysis method can be regarded as an extension of the AIM
theory for visual study. The spikes can be classified into three types, I marked them by blue, green
and red circles, as shown above.
Then close the scatter map, select option 3 to export grid data of sign(2) and RDG as
func1.cub and func2.cub in current directory, respectively, then copy the two files as well as the
RDGfill.vmd file in “examples” folder to VMD installation directory. The RDGfill.vmd is a plotting
script of VMD written by me. Boot up VMD, select "file"-"Load state", choose RDGfill.vmd
(alternatively, you can directly input source RDGfill.vmd in console window), you will see below
graph below in OpenGL window.
The default RDG isosurface is 0.5, the color range is -0.035 to 0.02. You can manually edit the
RDGfill.vmd to change the default settings, the current values are suitable for general cases. (for
present example, in VMD I also suggest selecting "Display"-"Light 3" to enable lighting with index
of 3, otherwise some parts of the isosurfaces will be relatively dark and the color cannot be faithfully
exhibited)
From the color-filled RDG isosurface, we can identify different types of regions by simply
examining their colors. Recall the color scale bar I showed previously, the more blue implies the
stronger attractive interaction; in current graph it can be seen that the elliptical slab between oxygen
and hydrogen atoms shows light blue color, so we can conclude that there is a hydrogen bond, but
not very strong. The interaction region marked by green circle can be identified as vdW interaction
region, because the mapped color is green or light brown, which shows that the electron density in
this region is low. Obviously, the regions at the center of the two rings correspond to strong steric
interaction, since they are filled by red.
Part 3: Summary of general steps for generating color-filled RDG map

Above I have talked a lot about the NCI analysis. In order to make you clearly and quickly
understand how to plot the sign(2) mapped RDG isosurface graph using Multiwfn, below I present
the minimum steps to do this, which are suitable for most cases.
Boot up Multiwfn and input
xxx.wfn (or wfx/mwfn/fch/molden... file) // Load input file
241
3 Functions

1 // NCI analysis
3 // Please properly define the grid points at this step. “High quality grid” is usually adequate
for small and medium sized system
3 // Export func1.cub and func2.cub
Move the two .cub files and examples\RDGfill.vmd to VMD folder. Boot up VMD and input
source RDGfill.vmd in console window, then you will see the graph you need.
Part 4: On the grid setting

Here I talk more about grid setting for computing grid data of RDG and sign(2), because this
point significantly influences computational cost and quality of the resulting RDG isosurface map.
The total time spent in the calculation is linearly proportional to the total number of grid points,
and the quality of RDG isosurface is highly dependent on grid spacing. The smaller the grid spacing,
the smoother the resulting isosurfaces. Too large grid spacing will result in severe jaggies at the
edges or hole at interal regions of the isosurfaces. It is easy to comprehend, if the box size (i.e.
spatial range of grid data) keeps fixed, then the higher number of grid points you set, the smaller the
grid spacing will be. Clearly, the box should be properly defined, its spatial range should not be too
broad, otherwise the grid spacing will be large and thus lowers the graph quality; it should also not
be too narrow, otherwise the interesting RDG isosurfaces may be truncated. The best practice is to
make the box just enclose the interesting region. Then, if you can afford high computational cost,
you can use as large number of points as possible to improve the final isosurface quality.
Note that the "low/medium/high quality grid" options in the interface of setting up grid are
relative to small or medium sized systems. If the system is huge and you have to employ large box,
even "high quality grid" will correspond to relatively large grid spacing, and thus the graphical
quality is not satisfactory. In this case, you should use the option "4 Input the number of points or
grid spacing in X,Y,Z, covering whole system" and manually input a reasonable grid spacing value.
Below is an illustration of various grid setting for the phenol dimer system, the value denotes
grid spacing. From this plot you can intuitively understand how grid spacing influences the result.
242
3 Functions
Part 5: Some worth mentioning points about NCI analysis

• Choice of level for generating wavefunction: It is absolutely unnecessary to use large basis
set to carry the NCI analysis. Using moderate size of basis set such as def2-SVP or 6-31G** is
completely adequate for NCI analysis, further enlarging the basis set is just waste of time. Regarding
the choice of theoretical method, using popular DFT functionals such as B3LYP or M06-2X to yield
wavefunction is adequate. Although post-HF density is known to be more accurate than DFT density,
the improvement in electron density quality can hardly be detected in the final NCI analysis result.
You may have known that the computational levels such as B3LYP/6-31G* perform quite poor for
weak interactions, however, it never means that using electron density produced at this level is
insufficient for NCI analysis, because electron density is never as sensitive as interaction energy to
calculation level, and there is not strictly positive relationship between the quality of calculated
interaction energy and electron density.
A frequently encountered annoying problem is that unexpected RDG isosurfaces occurred
around interesting regions and thus polluted the NCI graph, this makes visual analysis of weak
interaction at interesting regions difficult. For example, there is a system consisted of three
molecules, we only want to study weak interaction between molecules 1 and 2; however, in the
actual generated NCI graph, you may find unwanted isosurfaces corresponding to interactions
between 1-3 and 2-3 as well as those corresponding to intramolecular interactions also occur. To
screen the uninteresting isosurfaces, you can try to use the methods described in Section 4.13.4;
alternatively, you can consider to use IGM method instead, which can be completely free of this
problem as long as you properly define fragment, see Section 3.23.5 for introduction.
• Domain analysis for RDG: Multiwfn is capable of integrating any real space function within
isosurface defined by any real space function, this is known as “domain analysis”. Therefore you
can calculate such as volume and number of electrons enclosed within an isosurface of RDG (or
other related functions such as DORI and IGM) to try to discuss weak interactions at quantitative
level. Introduction of this kind of analysis is provided in Section 3.200.14, illustrative examples are
given in Section 4.200.14.
• NCI analysis for huge systems: If you want to apply the NCI analysis to very large systems
(e.g. more than 300 atoms), commonly the cost will be extremely high and thus not computationally
feasible. One of the best solutions is using promolecular version of NCI analysis or IGM analysis
instead, please check Section 3.23.2 and 3.23.5, respectively. Another solution is using Grimme's
xtb code to rapidly calculate the system using semi-empirical variant of DFT, and then using the
resulting .molden file as input file to perform the NCI analysis, the result should be better than the
promolecular NCI result.
• Averaged NCI: If you want to study interaction between a molecule with environmental
atoms during molecular dynamics process, the average NCI method should be used instead of
performing NCI analysis only for single structure, please check Section 3.23.3 for detail.
• NCI+AIM map: It is also possible to simultaneously plot AIM critical points and bond paths
in the color-filled RDG map, so that more information about weak interactions could be revealed,
below map is an example provided by a Multiwfn user [email protected]. The way of plotting
this kind of map is exemplified in Section 4.20.1 and illustrated as part 4 of this video:
https://youtu.be/e4FpVc9ao48.
243
3 Functions
• NCI+ELF map: As introduced in Section 2.6 and illustrated in relevant examples in Sections
4.4 and 4.5, the ELF (electron localization function) is very useful function for exhibiting character
of chemical bonds. Clearly, plotting NCI map and ELF isosurfaces together can convey more
information. Part 5 of this video tutorial illustrated how to realize this by Multiwfn in combination
with VMD: https://youtu.be/e4FpVc9ao48.
Special skill 1: Generating color mapped scatter map

It is also possible to map color to scatter map to facilitate identification of correspondence
between spikes and RDG isosurfaces. A plotting script of gnuplot program (http://www.gnuplot.info)
has been provided as examples\scripts\RDGscatter.gnu, which can realize this purpose. First, select
the option "2 Output scatter points to output.txt in current folder" in the post-processing menu to
exported output.txt in current folder, then move it and the RDGscatter.gnu into the folder containing
gnuplot executable file, then in this folder run command: gnuplot RDGscatter.gnu. After a while
you will obtain RDGscatter.ps in this folder, this is a graphic file of postscript format, you can open
it using such as Acrobat, photoshop or Irfanview (ghostscript must be installed in the machine). The
graph should look like this:
244
3 Functions
The default color range in this plotting script is from -0.035 to 0.02, if you intend to correlate this
map with RDG color-filled map, you should ensure that the color scale setting in RDGscatter.gnu
and RDGfill.vmd are completely identical.
If you find difficulty in reproducing this map, please follow part 2 of this video tutorial:
https://youtu.be/e4FpVc9ao48.
Special skill 2: Interactively set RDG value where sign(2) is in specific range
Multiwfn allows you to interactively set RDG value where sign(2) is in specified value range,
using this feature you can easily screen unwanted regions. Here I continue the phenol dimer example
described in “Part 2” and illustate how to screen RDG isosurface corresponding to the H-bond from
the graph. From the original scatter map, we find that the H-bond region corresponds to sign(2)
range of -0.035 ~ -0.015, therefore we can input below command in post-processing menu
-2 // Set RDG value where sign(2) in within given data range
-0.035,-0.015 // The lower and upper limit of sign(2)
100 // Set RDG value in these regions to an arbitrarily large value to screen RDG isosurface
Then, if you select option -1 to plot the scatter map again, you will see
245
3 Functions
Clearly the spike corresponding to H-bond no longer exists. We can also export the cube files and
use VMD redraw the color-filled map, as shown below, the H-bond RDG isosurface has indeed
disappeared.
Notice that the original grid data cannot be retrieved once modified as exemplified above.
Information needed: Atom coordinates, GTFs
3.23.2 NCI analysis based on promolecular density (2)
Generating wavefunction and calculating grid data of RDG and sign(2) for large system are
very time-consuming, which greatly hinders application range of NCI analysis method. Fortunately,
it is shown that the NCI analysis based on promolecular density is also reasonable in general. The
so-called promolecular density is the electron density approximately constructed by superposing
electron density of atoms in their free-state, this is known as “Promolecular approximation”. High
quality free-state atomic electron density for almost all elements in periodic table are predetermined
and built-in, hence NCI analysis based on promolecular density can be in principle used for any
system in Multiwfn.
246
3 Functions
To carry out NCI analysis under promolecular approximation, just choose subfunction 2 in
main function 20, all operation steps are completely identical to regular NCI analysis. Since only
atom coordinate information is required for constructing promolecular density, any input file
containing atomic coordinate information can be used as input file, such as the popular .pdb and .xyz
formats.
The VMD plotting script for NCI analysis based on promolecular density is offered as
examples\RDGfill_pro.vmd, which is slightly different to examples\RDGfill.vmd in the default
setting of color scale and isovalue.
By default, RDG value is automatically set to 100.0 where  is larger than 0.1 when
promolecular approximation is used, this threshold may not be suitable for certain circumstances.
You can manually change the threshold by “RDGprodens_maxrho” in settings.ini.
Example of performing NCI analysis based on promolecular density is given in Section 4.20.1
and illustrated as part 3 of this video: https://youtu.be/e4FpVc9ao48.
3.23.3 Averaged NCI analysis (NCI analysis for multiple frames. 3)
Theory
In J. Chem. Theory Comput., 9, 2226 (2013), the NCI method described in last sections is
extended to analyzing fluctuation environment (e.g. molecular dynamics trajectory), resulting in the
averaged NCI (aNCI) method. Present function aims at realizing the aNCI analysis.
The only difference between aNCI and the original NCI method is that in the former, the
electron density  and its gradient norm || are not calculated for only one geometry, but for
multiple frames in a trajectory file, then get average (namely  and  ). Therefore, the
isosurface of averaged reduced density gradient
1 |  (r ) |
aRDG(r ) =
2(3 ) 2 1/ 3
 (r )
4/3
can be directly used to reveal the averaged weak interaction regions for a dynamics process.
Similarly, in order to exhibit averaged weak interaction type, in aNCI method, the λ2 term in
Sign(λ2) function is obtained as the second largest eigenvalue of the averaged electron density
Hessian matrix computed through out the dynamical trajectory.
aNCI method also defines a new quantity named thermal fluctuation index (TFI) to reveal the
stability of weak interaction
std [  (r )]
TFI(r ) =
 (r )
whose numerator is standard deviation of electron density in the dynamical trajectory, which can be
calculated as
247
3 Functions
 [  (r ) −  ( r ) ]
i
2
std [  (r )] = i
n
where n is the number of frames in consideration, i is the density calculated based on the geometry
of frame i. After mapping TFI on the isosurface of aNCI, the stability of each weak interaction region
can be clearly identified by visually examining the colors.
The quality of aNCI graph directly depends on the number of frames that taken into account.
Small number of frames, for example 50 frames, can only leads to inaccurate and very unsmooth
isosurface graph. In general, at least 500 frames are should to be used to generate aNCI graph.
Usage
Firstly, note that since calculating electron density based on wavefunction for large number of
geometries is very expensive, promolecular approximation is forced to be used in the aNCI analysis
function of Multiwfn. This approximation is highly reasonable and always works well.
The trajectory stored in .xyz file is allowed as input file. You can use such as VMD program to
convert other format of trajectory files to .xyz trajectory file.
PS: The structure of a multiple frame .xyz file looks like below
[Number of atoms in frame 1]
[Element, x, y and z of atom 1 in frame 1]
...
[Element, x, y and z of atom n in frame 1]
...
[Element, x, y and z of atom n in frame 2]
...
In all of the frames, the coordinate of the interesting molecule should be fixed. For example, if
you want to study the weak interaction between solvents and a benzene molecule, then the position
of the benzene must be fixed through out the whole trajectory. Commonly, the geometry center of
interesting molecule is fixed at box center.
After you enter present function, you will be prompted to input the frame range to be analyzed,
for example inputting 140,450 means the frame from 140 to 450 will be used in the aNCI analysis.
Then you need to set up grid, the spatial range of the box should properly enclose the molecule of
interest. After that, averaged electron density, averaged density gradient and averaged density
Hessian will be calculated for each frame, you should wait patiently. Once the calculation is finished,
you can use corresponding options to draw scatter graph between averaged NCI and averaged
sign(λ2), output scatter points, export their cube files, etc. Thermal fluctuation index can also be
calculated and export to cube file.
Information needed: Multiple frames of atom coordinates
3.23.4 Density Overlap Regions Indicator (DORI) analysis (5)
Sometimes ELF and RDG are used in combination to simultaneously investigate covalent and
248
3 Functions
noncovalent interactions, see J. Chem. Theory Comput., 8, 3993 (2012) for example. Is it possible
to study both kinds of interactions by single real space function? The answer is yes. In J. Chem.
Theory Comput., 10, 3745 (2014) the authors proposed a function named density overlap regions
indicator (DORI), it was found that if properly choosing an isovalue, both covalent and noncovalent
interaction regions can be exhibited by DORI isosurface, and sign(2) can also be mapped on to
DORI isosurfaces to faciliate analysis of the nature of interactions.
The expression of DORI is
DORI(r) =  (r) /[1 +  (r)]
where  (r) = [( (r) /  (r)) 2 ]2 /[ (r) /  (r)]6 . The value range of DORI is [0,1].
To plot the sign(λ2) mapped DORI isosurface map, enter subfunction 5 of main function 20,
the subsequent operations are exactly identical to NCI analysis. After exporting the grid data of
sign(λ2) and DORI as cube files by option 3 in post-processing menu, you can then copy them
along with examples\DORIfill.vmd to VMD folder, then boot up VMD and execute the plotting
script DORIfill.vmd to plot the color-filled isosurface map.
Personally I do not recommend to use DORI, because the interaction region indicator (IRI)
introduced in Section 3.23.8 is not only simpler than DORI and thus the computational cost is lower,
but also the graphical effect is much better.
An example of performing DORI analysis is given in Section 4.20.5
3.23.5 Independent Gradient Model (IGM) analysis based on

promolecular density (10)
Preface
In Phys. Chem. Chem. Phys., 19, 17928 (2017), Hénon et al. proposed a very useful way for
visually studying interfragment and intrafragment interactions, it is named as Independent Gradient
Model (IGM). Note that currently IGM has three versions:
(1) IGM based on promolecular density. This is the original version of IGM proposed in 2017,
the function introduced in this section implements this form of IGM, only molecular structure is
needed in the analysis.
(2) IGM based on gradient-based partitioning (GBP). This version was proposed in
ChemPhysChem, 19, 724 (2018) and requires actual molecular electron density. This is not
supported by Multiwfn.
(3) IGM based on Hirshfeld partition (IGMH). This version was proposed by me (to be
published). The function described in Section 3.23.6 implements this form. The IGMH is more
expensive than IGM and meantime wavefunction must be provided in input file, the advantage of
IGMH is that the result is more meaningful and the graphical effect is better than IGM.
Idea of IGM
Full introduction of IGM can be found in the original paper, below I outline the essential idea
249
3 Functions
of this method from my personal point of view.

Let us first look at a very simple system, the H2 molecule. The atomic density in free-state of
each atom along the molecular axis is shown as below
From above graph one notices that the gradient of atomic density of the two atoms in the interatomic
region have opposite signs. For example, at the position of X=1.2, the density gradient of H1 is
negative, while that of H2 is positive. Therefore, in the gradient of promolecular density (the curve
g in below map), the contribution from the two atoms largely cancel with each other in the region
between the two atoms. Note that at the midpoint of the two hydrogens, g is exactly zero, such point
corresponds to bond critical point (BCP) in AIM theory under promolecular density.
In above map, the gIGM is IGM type of gradient, it is calculated as sum of absolute value of
density gradient of each atom in their free-states; in other words, phase is completely ignored and
thus the density gradient originating from various atoms never cancel with each other. Due to this
feature, gIGM is upper limit of g.
The g function is defined as the difference between gIGM and g, it is plotted as deep blue curve
in above map. It can be seen that g is non-zero in the atom interaction region, and has maximum
value at the BCP position. Clearly, g could be used to reveal interaction regions like reduced density
gradient (RDG). In addition, as will be illustrated in the examples in Section 4.20.10, magnitude of
g in interaction region has close relationship with interaction strength.
In general case, gIGM and g may be defined as follows
g (r ) =   (r)
i
i g IGM (r ) =  abs[ (r)]
i
i
g (r ) = g IGM (r ) − g (r )
250
3 Functions
where abs( ) is the operator to take absolute value for each component of a vector. The i stands for
spherically averaged density of atom i in its free state. The atomic density covering the whole
periodic table is bulit-in data in Multiwfn, see Appendix 3 for detail.
Based on the idea of gIGM and g, the IGM method also defines ginter and gintra aiming to study
interfragment and intrafragment interactions, respectively
g inter (r ) =   (r)
A iA
i
 
g IGM,inter (r ) =  abs   (r)
A iA
i
g inter
(r ) = g IGM,inter
(r ) − g inter (r )
g intra (r ) = g (r ) − g inter (r )
where A and i are index of fragments and atoms, respectively. The fragment can be arbitrarily
defined according to character of actual system and research purpose. (The above expression of
ginter and gintra should be their most general forms and are employed in Multiwfn, they are not
explicitly given in the IGM original paper)
The idea of ginter is easy to understand from above formula. One first calculate density gradient
in usual way as ginter, and then calculate the gIGM,inter, which ignores cancellation effect of density
gradient of various fragments due to possible different phases, then the difference between gIGM,inter
and ginter, namely ginter, must be able to reveal the interaction between the fragments. The g reveals
all kinds of interactions in present system, irrespective of the type is interfragment or intrafragment.
Therefore, if ginter is subtracted from g, the remaining part, namely gintra, must be capable of
revealing intrafragment interactions.
In Section 3.23.1, it is shown that interaction region and interaction type can be simultaneously
exhibited by plotting RDG isosurface map colored by sign(2) function. Similarly, if sign(2)
function is mapped on ginter and gintra isosurfaces using various colors, the type and position of
inter- and intra-fragment interactions could also be vividly revealed.
Quantifying atom pair contributions

I defined atom pair g index to quantify the contribution of atom pair to interaction between
two fragments (A and B),
 gi , j =  [ giIGM
, j (r ) − gi , j (r )]d r i  A, j  B
where
, j (r ) = abs[g i (r )] + abs[g j (r )]
giIGM
g i , j ( r ) = g i (r ) + g j (r )
I also defined atom g index to quantitatively measure importance of atom to interfragment
interaction
 gi =   gi , j i  A j  B
j
When plotting molecule structure, if atoms are colored according to the atom g indices, then the
relative importance between various atoms to interfragment interaction can be immediately captured.
Inspired by the IBSI (intrinsic bond strength index) introduced in Section 3.11.9, I defined
251
3 Functions
IBSIW (IBSI for weak interaction) as follows
 gi , j
IBSIW (i, j ) = 100 
(d i , j )2
where di,j is distance between atoms i and j in Å. My preliminary test showed that IBSIW has
somewhat better ability to distinguish interaction strengths. Clearly, the larger the IBSIW, the
stronger the interaction. Since in Multiwfn the g index is given in a.u., the formal unit of IBSIW
should be a.u./Å2.
Why using IGM instead of NCI?

According to my viewpoint and experiences, the advantage of IGM method over the popular
NCI method can be summarized as follows:
·The inter- and intra-fragment interactions can be studied individually and thus mutual
interference is avoided
·Calculation of the functions defined by IGM method is rather fast and only dependent of
geometry, thus the method can be applied to broad range of systems. (Note that NCI also has
promolecular approximation version)
·As will be shown in the examples in Section 4.20.10, the isosurface graph given by the IGM
method is more plump than the NCI map, and thus the IGM map has low requirement on grid
spacing. In contrast, NCI graph is prone to ugly jaggies and holes when the grid points are sparse.
·Contribution of atoms and atomic pairs to interfragment interaction can be quantified, and
the former can be vividly rendered on molecular structure, these features make identification of
"hot" atoms easy.
·The value of g function in interaction region directly reflects interaction strength. In
particular, g at bond critical point of AIM theory is a good quantitative indicator of strength of
corresponding interaction.
Usage of IGM analysis function in Multiwfn

Using Multiwfn to carry out IGM analysis is extremely easy and flexible. First, you should
load a file containing atom coordinates. The most commonly used formats such as .xyz, .pdb
and .mol are all supported by Multiwfn (of course, any wavefunction file such as .wfn and .fch can
also be used). Notice that the geometry must have been optimized using proper theoretical level,
otherwise the IGM result may be misleading.
IGM module is subfunction 10 of main function 20, after you enter this module, you should
define fragments. The definition of fragment is quite flexible, you can define any number of
fragments (at least one fragment). No atom should be simultaneously shared by two or more
fragments. The union set of defined fragments is not forced to be equal to the whole system, only
the atoms in the defined fragments will be finally taken into calculation.
Next, you need to setting up grid, it is better to make the box just enclose the region where the
interesting interactions may occur. Some advices about setting grid are given in Section 3.23.1.
After that, Multiwfn starts to calculate grid data of sign(2), g, ginter and gintra. Assume that
you have defined n fragments, then the calculated ginter will correspond to interaction between all
the n fragments, while gintra will correspond to intrafragment interactions within all the n fragments.
The larger number of atoms in the fragments, the higher the overall cost.
252
3 Functions
The options in post-processing menu are self explanatory, I briefly describe them here:
-1: Suboptions 1, 2 and 3 of this option are used to draw scatter map of g, ginter, gintra vs.
sign(2), respectively. While suboption 4 is used to draw ginter and gintra vs. sign(2)
simultaneously with different colors. As shown in the original paper of IGM, this kind of map is
useful for discussing details about interactions (recall that RDG vs. sign(2) scatter map is
frequently involved in NCI analysis). If you want to directly save the scatter map in current folder
as graphic file, use option 1. If the default axis range is not appropriate, use option -2 or -3 to adjust.
2: If you want to draw scatter map using third-part softwares such as Origin and gnuplot, use
this option to export data of g, ginter, gintra and sign(2) to plain text in current folder. Meaning
of each column of this file is shown on screen.
3: Output grid data of sign(2), g, ginter and gintra to cube file in current folder. After
exporting the cube files, you can use IGM_inter.vmd and IGM_intra.vmd scripts in "examples"
folder to draw color-filled ginter and gintra isosurfaces map in VMD, respectively. See examples of
Section 4.20.10.
4: This option is used to directly visualize isosurface of sign(2), g, ginter or gintra in
Multiwfn.
5: This option is used to set gintra to zero where sign(2) is not within specified value range.
By this option uninteresting regions could be screened from gintra scatter and isosurface maps. For
example, we merely want to study weak intrafragment interactions, then we can input the range
corresponding to relatively small value of sign(2). (The aim of this option resembles
RDG_maxrho parameter used in NCI analysis)
6: This option is used to evaluate atom and atom pair g indices. If you have defined more than
two fragments, here you need to choose two fragments for which the g indices will be calculated.
Multiwfn will compute g grid data of every atom pair between the two fragments and calculate
integral of g function to derive the g indices. The integrals are calculated using Becke's multi-
center integration method, there are several choices of integration grids, the better the grid, the more
accurate the result, but the higher the cost. Once the calculation is complete, atmdg.txt will be
outputted to current folder, which records all atom and atom pair  indices, the values are sorted
from high to low. Then the program asks you if also outputting atmdg.pdb in current folder, which
contains all atoms in present system. The "Beta" field of this pdb file corresponds to ten times of
atom g index. If you load this file into VMD visualization program and color the atoms according
to Beta field, then relative importance of various atoms to interfragment interactions can be
intuitively identified.
At the same time of outputting atmdg.txt, IBSIW indices for atom and atom pairs are also
exported to IBSIW.txt in current folder.
7&8: These two options are used to set value of g and ginter functions respectively if the
sign(2) at corresponding grid is out of specific range. Clearly, these options are useful when you
want to screen unwanted region from IGM isosurface map. For example, you only want to visualize
isosurface of ginter where sign(2) is within -0.04 ~ -0.025, then you can enter option 8, input -
0.04,-0.025 and then input 0 to set ginter of these grids to zero; next, you can plot the updated scatter
map or export cube files to visualize IGM map in VMD.
If your input file contains GTF or basis function information (e.g. mwfn, .wfn, .fch, .molden),
when carrying out IGM analysis, Multiwfn will let you choose the type of the sign(2) to be used,
253
3 Functions
the first one is that based on actual electron density, the second one is that based on promolecular
density. Using the former should give more meaningful result, however, calculation the cost for the
former is evidently higher than that of the latter (if your input file only contains atomic coordinate
information, the latter one is always used).
Several examples of IGM analysis are given in Section 4.20.10. More discussions and instances
about IGM method can be found in my blog article "Investigating intermolecular weak interactions
via Independent Gradient Model (IGM)" (in Chinese, http://sobereva.com/407).
3.23.6 IGM analysis based on Hirshfeld partition of molecular density

(IGMH) (11)
As shown in Section 3.23.5, the original version of IGM is calculated purely based on density
of atoms in their free states, namely promolecular approximation is used. A different form of IGM
was proposed by me in 2020 and named as "IGM based on Hirshfeld partition of molecular density"
(IGMH). Everything of IGMH is identical to IGM except for the way of calculating gradient of
atomic density.
In IGMH, the atomic density is defined based on Hirshfeld partition, namely 𝜌𝐴 (𝐫) = 𝜌𝑤𝐴 (𝐫),
where  is the electron density of the whole system calculated based on wavefunction as usual, the
Hirshfeld weighting function of atom A is expressed as
 Apro  Apro
wA = =
 pro   Bpro
B
pro pro
where 𝜌𝐴 is spherically averaged electron density of atom A in its free state, 𝜌 corresponds
pro
to promoleular density, the index B loops over all atoms. The derivative of 𝜌𝐴 with respect to
Cartesian coordinate can be derived as follows:
   
wA    Apro   Apro 1 pro   1 
= = + A
x x    Bpro  x   Bpro x    Bpro 
 B  B  B 
 pro
1 A pro
 B
pro
1  A pro
A pro
 Bpro
2  2 
= A −  −
x   Bpro  pro 
x  pro x (  pro ) B x
  B 
B
B
 B 
  (  wA )  w    pro  Apro  Bpro
2 
 A= = wA +  A = wA + pro A −
x x x x x  x (  pro ) B x
The derivative with respect to coordinate y and z is similar.
Based on the gradient of atomic density calculated in the above way, the g and other functions
as well as the atom and atom pair g indices can be calculated in exactly the same way as IGM.
In the IGMH analysis, the sign(2) function is also calculated based on actual electron density.
Clearly IGMH is much more expensive than IGM, since both gradient and Hessian of actual electron
254
3 Functions
density must be evaluated.

The advantages of IGMH over IGM are two:
(1) The graphical effect of isosurface map is better. The isosurface of the g function defined
in IGM is too fat and sometimes unreasonable; in contrast, the shape of the g function calculated
in terms of IGMH is thinner and thus easier to examine and compare.
It is worth to note that the isosurface of g in IGMH is close to the isosurface of reduced density gradient (RDG),
which is employed in the NCI method (see Section 3.23.1). The advantage of the former is that the isosurface looks
smoother, and unsightly jagged edges in the RDG isosurface is highly avoided.
(2) The physical meaning of IGMH is more rigorous than IGM, since all factors that influence
distribution of electron density during formation of the system has been naturally taken into account.
Due to above reasons, using IGMH instead of IGM is always recommended if the system is
not quite large and thus the computational cost is affordable. It is worth to note that Hénon and
coworkers proposed a version of IGM based on wavefunction in ChemPhysChem, 19, 724 (2018),
however, in my opinion, the IGMH is a better choice, since it is defined in an evidently more elegant
and understandable way.
The use of the present function, namely subfunction 11 of main function 20, is exactly the same
as the IGM function, see Section 3.23.5 for introduction of various options.
Example of using IGMH analysis is given in Section 4.20.11.
Information needed: Atom coordinates, GTF information
3.23.7 Visualization of van der Waals potential (6)
Please check my this paper for full description of idea, implementation and application of the
van der Waals (vdW) potential analysis: Tian Lu, Qinxue Chen, ChemRxiv (2020) DOI:
10.26434/chemrxiv.12148572. Below I only briefly introduce this function in a nutshell.
Recall that the vdW interaction energy between two atoms A and B is usually expressed in
terms of Lennard-Jones potential in below form
12 6
 R0   R0 
E vdW
AB =E repul
AB +E disp
AB =  AB  AB  − 2 AB  AB 
 rAB   rAB 
where the potential well  and equilibrum distance R0 are dependent of atom types. The two
components of EvdW, namely Erepul and Edisp, correspond to exchange-repulsion and dispersion
interaction, respectively.
I define vdW potential of a chemical system as follows
 RAB 0

12
  RAB 0
 
6
V vdW
(r ) = V repul
(r ) + V disp
(r ) =   AB   +   −2 AB   
A  | RA − r | A 
  | RA − r | 

where B can be regarded as probe atom. The Vrepul and Vdisp denote repulsion and dispersion
potentials, respectively.
In the implementation in Multiwfn, the vdW parameters from UFF forcefield are employed,
this is because the elements supported by UFF almost cover the whole periodic table (H~Lr), and
255
3 Functions
the parameters are only dependent of element, thus the problem in assigning atom types is fully
avoided. The element index of the probe atom can be set by "ivdwprobe" in settings.ini, the default
is carbon (i.e. ivdwprobe=6). If the "ivdwprobe" is set to 0, then program will ask you to input
element name of probe atom when entering this function.
The vdW potential can be easily calculated by subfunction 6 of main function 20, the unit of
result is kcal/mol. In this function, you need to first select grid setting, then grid data of VvdW, Vrepul
and Vdisp will be calculated, then you can visualize their isosurfaces or export them as cube files by
corresponding options.
Note that the VvdW, Vrepul and Vdisp also directly correspond to user-defined functions, 92, 93
and 94, respectively.
Example of visualization and analysis of VvdW is given in Section 4.20.6.
3.23.8 Interaction region indicator (IRI) analysis (4)
The interaction region indicator (IRI) proposed by me (to be published) is defined as follows
IRI = [ (r)]a / (|  | +b)

where a and b correspond to "uservar" and "uservar2" in settings.ini, respectively. If "uservar" is 0,
a will be the recommended value 1.1, while if "uservar2" is 0, then b will be the recommended value
0.0005.
IRI is able to reveal both chemical bond regions and weak interaction regions by its isosurface
(usually isovalue of 0.9~1.1 is used), this point is very similar with the DORI introduced in Section
3.23.4. Indeed, the isosurface maps of the IRI and DORI are often comparable, however IRI is
always preferred over DORI due to two evident advantages:
(1) The definition of IRI is much simpler, only electron density and its gradient are needed,
while DORI also requires Hessian of electron density. Clearly, evaluation of IRI is thus evidently
cheaper than DORI.
(2) The graphical effect of IRI isosurface is much better than DORI. In weak interaction regions
the DORI isosurfaces usually appear to be composed of two layers and thus quite ugly.
Like NCI, IGM and DORI analyses, sign(2) can also be mapped to IRI isosurfaces to visually
distinguish the nature of interactions. To do so, you should
(1) Enter subfunction 4 of main function 20
(2) Select proper grid setting, then in the post-process menu export the grid data to func1.cub
and func2.cub
(3) Copy the two cube files as well as examples\IRIfill.vmd to VMD folder
(4) Boot up VMD and input source IRIfill.vmd in the console window to run the script.
If you are only interested in IRI itself and want to study its distribution, for example, plotting
it as plane map by main function 4, you should set "iuserfunc" in settings.ini to 99, then the user-
defined function will correspond to IRI. Then you can study it via various functions in Multiwfn,
you just need to select "User-defined function" when Multiwfn asks you to choose the real space
256
3 Functions
function to be investigated.
Example of IRI analysis is given in Section 4.20.4.
Information needed: Atom coordinates, GTF information
3.24 Energy decomposition analysis (21)
3.24.1 Energy decomposition analysis based on molecular forcefield

(EDA-FF)
Energy decomposition analysis (EDA) is an important method to reveal the nature of

interaction. Most EDA analysis methods are based on wavefunction, they are accurate, rigorous and
the results are meaningful, unfortunately they are often too expensive for large systems. Present
module is designed for analyzing intramolecular and intermolecular weak interactions based on
classical molecular force field (FF), this method could be referred to as EDA-FF, the computational
cost is negligible for systems consisting of hundreds of atoms, and it can even be applied to systems
with thousands of atoms. Due to limitation of FF, this module evidently cannot be used to discuss
the natural of chemical bond interactions. The word "weak interaction" in this section refers to the
interatomic interaction separated by more than three bonds.
Theory
There are many popular FFs for molecular systems. The major ingredients of weak interactions
are van der Waals (vdW) interaction and electrostatic interaction, most FFs represent them by mean
of pairwise potential, as shown below.
·Electrostatic interation energy between atoms A and B (atomic unit is used):
q A qB
ele
E AB =
rAB
where q is atomic charge and rAB is distance between A and B.
·vdW interaction energy between atoms A and B:
vdW
E AB = E AB
rep
+ E AB
disp
12 6
 R0   R0 
E rep
AB =  AB  AB  E disp
AB = −2 AB  AB 
 rAB   rAB 
where Erep represents repulsive interaction due to Pauli repulsion effect (also known as exchange-
repulsion), while Edisp is attractive dispersion interaction. The AB is well depth of interatomic vdW
interaction potential, while R0AB is vdW nonbond distance. When rAB=RAB0, the interaction energy
just corresponds to well depth.
The parameters  and R0 are provided by FFs, and the values are commonly defined for each
atom type. The interatomic parameters used in practical calculation are commonly evaluated as
geometric average or arithmetic average of atomic parameters. For example, in UFF forcefield,
257
3 Functions
below mixing rule is used
 AB =  A B 0
RAB = RA0 RB0
While for some other FFs such as AMBER and GAFF, atomic  and R*are defined, the employed
mixing rule is
 AB =  A B 0
RAB = RA + RB
where R* is known as atomic nonbond radius or atomic vdW radius.
Given the interatomic interaction terms, evaluating various physical components of
interfragment interaction energy is straightfoward:
EIJele =  E AB
ele
EIJrep =  E AB
rep
EIJdisp =  E AB
disp
AI BJ AI BJ AI BJ
The present module is mainly used to evaluate above three terms between two or more user-
defined fragments, many useful quantities can be obtained at the meantime.
Usage
The basic steps of using this module are:
(1) Prepare a "molecule list" file, which contains paths of "molecule type" files. Detailed
description is given later.
(2) Load a file containing geometry information of the whole system. Evidently, many formats
supported by Multiwfn could be used, for example, .xyz, .mol, .pdb, .fch and so on.
(3) Enter subfunction 1 of main function 21.
(4) Use option 3 to load molecule list file. This step is used to assign atomic charge and type
for each atom.
(5) Use option 2 to define fragments. Infinite number of fragments may be defined, any atom
should not simultaneously appear in two or more fragments. If you want to study interaction between
two fragments in the same molecule, any atom pair between the two fragments should be separated
by at least three bonds (otherwise the interaction will no longer belong to the scope of weak
interaction). Note that the order of the steps (4) and (5) could be exchanged.
(6) Choose option 1 to start analysis, then interaction energy components between each
fragment pair, as well as atomic contributions will be printed. Before the analysis, if you want to
check whether the charges and types have been properly assigned, you can choose option 4.
There are other options you can select before calculation:

·Option -1: This is used to choose the FF employed in the calculation. Currently, AMBER99
& GAFF (default) and UFF are supports, their difference is that the built-in atomic vdW parameters
and the mixing rule used in the calculation are different.
·Option -2: This option can choose the operator in calculating electrostatic interactions. By
default 1/r operator is employed, while using this option you can change it to 1/r2. Obviously, the
attenuation of Eele calculated based on 1/r2 with respect to interaction distance r is much faster than
the default case, this is why some studies employ this simple strategy to effectively exhibit water
environment, since it is well-known that polar solvents such as water can significantly shield
electrostatic interaction strength due to its large dielectric constant.
·Option -3: If you choose this option once to switch the status to "Yes", then after calculation,
all interatomic interaction energy terms including their physical components will be exported to
interatm.txt in current folder.
258
3 Functions
·Option -4: In the standard .pqr format, the last two columns are specific for storing atomic
vdW radii and atomic charges. If you choose this option once to switch the status to "Yes", then
during calculation, atmint_tot.pqr, atmint_ele.pqr, atmint_rep.pqr, atmint_disp.pqr and
atmint_vdW.pqr will be outputted in current folder, their last columns record atomic contribution to
total, electrostatic, repulsion, dispersion and vdW (i.e. repulsion+dispersion) interaction energies,
respectively. In the popular VMD visualization program, you can load one of these .pqr file and
color the atoms according to "atomic charges" data, then the atom colors will vividly exhibit
contribution of each atom to corresponding kind of interaction energy.
Next, I introduce the rule of writing the molecular list file. The content of the file should look
like this:
C:\mol1\phenol.txt 1
C:\mol2\H2O.txt 4
C:\HCl.txt 2
This example file implies that, in the geometry information provided by the file loaded when
Multiwfn boots up, the recording sequence is: one phenol molecule, four H2O molecules and two
HCl molecules. The three .txt files contain atom types (case sensitive) and charges of respective
molecule. For example, below is the content of the C:\mol2\H2O.txt, which records information of
the atoms in H2O molecule, the OW and HW are atom types of AMBER force field.
OW -0.728713
HW 0.364427
HW 0.364286
The first column is atom type corresponding to the FF you currently choose, the second column
corresponds to atomic charge. Notice that the atom sequence in this file must be exactly identical to
the geometry information loaded when Multiwfn boots up. If the forcefield employed in analysis is
UFF, then this file should only contain atomic charge, thus there should be only one column (because
for each element, all relevant atom types share the same UFF vdW parameters, therefore users do
not need to define atom types).
·About atom types: Detailed description of atom types can be found in original paper of
corresponding forcefields. For AMBER, see Table 1 of J. Am. Chem. Soc., 117, 5179 (1995). For
GAFF, see Table 1 of J. Comput. Chem., 25, 1157 (2004); you can also consult the AMBER99.txt
and GAFF.txt in "examples\EDA\EDA_FF" folder for description of atom types. In general, you can
manually find appropriate atom type for each atom in current system according to its actual chemical
environment. However, if you feel this process is troublesome, you can use third-part programs to
help you to identify atom types and construct the molecular files. For example, GaussView can
automatically assign AMBER atom types (enter "Atom List", click the icon with large organge "M"
symbol, double click head of "AMBER Type" column, select "File"-"Export Data", then extract the
data corresponding to "AMBER Type" column), while Antechamber utility in AmberTools package
is able to assign GAFF atom types. Note that AMBER and GAFF atom types can be mixed together
in the same molecule file, since these two forcefields are completely compatible with each other,
the AMBER and GAFF atom types use upper and lower case, respectively.
Note: The atom types assigned by GaussView are always in upper case, however, some atom types of
AMBER99 are lower case, e.g. Br. Clearly, you should manually make modification before loading the file into
Multiwfn. If you are confused, take a look at examples\EDA\EDA_FF\AMBER99.txt.
·On the choice of forcefield: For organic type of systems, commonly I suggest using
259
3 Functions
AMBER/GAFF forcefield to conduct the analysis, the result should be reasonable and chemically
meaningful. In fact, since vdW parameters of GAFF are directly inherited from AMBER, commonly
there should be no different between using GAFF and AMBER atom types. Evidently, the geometry
used in the analysis should be firstly optimized under reasonable level. Using UFF is generally
deprecated, since I found that when UFF is employed, the total interfragment interaction energy is
usually positive due to overestimation of Erep, even if the geometry has already been substantially
optimized with appropriate quantum chemistry method. A way to solve this problem is using the
geometry optimized by UFF itself (many programs can do this, such as Gaussian and OpenBabel),
however the resulting geometry for weakly interacting molecular dimer or multimer is often not
quite good. The unique advantage of UFF is that it covers almost entire periodic table. Considering
this, I designed a trick in Multiwfn: If you are using AMBER/GAFF, when atom type is written as
UF, then UFF vdW parameter will be employed. This treatment greatly extends the application
scope of AMBER and GAFF.
·On the choice of atomic charges: The atomic charges used for energy decomposition
analysis should be able to reproduce electrostatic potential (ESP) around molecular vdW surface
well. Commonly I suggest to use CHELPG atomic charge, which is obtained via ESP fitting process
and can be directly calculated via Multiwfn, see Section 3.9.10 for introduction and Section 4.7.1
for example. Note that if a type of molecule appears more than once in current system with
significantly different conformations, given that ESP fitting charges of each monomer may be very
different to others, it is suggested to treat these replicas as different types of molecules, so that
atomic charges can be individually assigned. If some monomers of the system is to large to calculate
their ESP fitting atomic charges, you may change to EEM atomic charges using the parameters fitted
for reproducing ESP fitting charges, see Section 3.9.15 for introduction. The computational cost of
EEM charges is negligible for a system even composed of hundreds of atoms, since the calculation
is purely based on molecular geometry information and empirical parameters.
Examples of this module are provided in Section 4.21.1.
Information needed: Atom coordinates and special files containing atomic charges/types
3.24.2 Shubin Liu's energy decomposition
Theory
In J. Chem. Phys., 126, 244103 (2007), the author Shubin Liu proposed an idea of energy
decomposition, which will be referred to as EDA-SBL below. In this method, the total molecular
energy is decomposed as
E = Esteric + Eelectrostatic + Equantum
The steric term is simply the energy derived by Weizsäcker kinetic functional, which corresponds
to the exact kinetic energy under assumption that the elctrons in present system are non-interacting
bosons:
Esteric = TW =  (r) / [8 (r)]

2
260
3 Functions
The electrostatic term is the sum of all classical Coulomb interactions of the particles in the system:
 (r1 )  (r2 ) ZA Z Z
Eelectrostatic = EJ + EN −E + EN −N =   dr1dr2 −   (r ) dr +  A B
r12 A | r − RA | A B RAB
Finally, the quantum term is the energy purely caused by quantum effect:
Equantum = Epauli + EXC
where the EXC is exchange-correlation energy, the Epauli = TS − TW is Pauli kinetic energy, in
which the TS stands for total kinetic energy of non-interacting electron model and can be computed
as the sum of kinetic energy of all occupied molecular orbitals. The Equantum essentially exhibits
electronic correlation effect as well as influence of Pauli exclusion principle on electronic kinetic
energy under non-interacting particle assumption.
The EDA-SBL method has been employed in many research papers, such as the ones shown
below, which are suggested to read to if you want to understand how this method can be used to
study practical chemical problems: J. Phys. Chem. A, 117, 962 (2013), J. Chem. Phys., 133, 114110
(2010), Phys. Chem. Chem. Phys., 17, 27052 (2015), J. Phys. Chem. A, 119, 8216 (2015), Chem.
Phys. Lett., 687, 131 (2017).
Usage
The Multiwfn itself is unable to evaluate all terms in the EDA-SBL method, Multiwfn needs
to read relevant information from Gaussian output file. The way of using Gaussian to perform EDA-
SBL analysis is summarized as follows:
(1) Manually create a special Gaussian input file of single point task based on optimized
geometry
(2) Run the input file by Gaussian and get output file as well as fch/fchk file
(3) Boot up Multiwfn and load the fch/fchk file, then enter subfunction 2 of main function 21
(4) Input the path of the Gaussian output file
Then Multiwfn calculates the Esteric term, and prints all the three energy components defined
by the EDA-SBL method. Other intermediate terms involved in the EDA-SBL terms are also
simultaneously given, such as Pauli kinetic energy, nuclear-electronic Coulomb attraction energy
and so on.
The special Gaussian input file should be coincident with following format, the geometry has
been optimized using appropriate level.
%chk=H2O.chk
# B3LYP/6-31G* ExtraLinks=L608
Optimized water
0 1
O 0.00000000 0.00000000 0.11930801
H 0.00000000 0.75895306 -0.47723204
H 0.00000000 -0.75895306 -0.47723204
261
3 Functions
-5
The DFT functional and basis set can be arbitrarily chosen, the "ExtraLinks=L608" must be
specified so that Gaussian can break total energy into various components and print them to output
file. After calculation, the resulting H2O.chk should be converted to fch/fchk file using formchk
utility. As you can see, there is a value "-5" at the end of the input file, this value should be specified
according to the DFT functional you actually used, you can find corresponding value by consulting
IOp(3/74) in Gaussian IOps reference. The value for commonly employed hybrid functionals are:
-58 (B97XD), -55 (M06-2X), -54 (M06), -53 (M06L), -40 (CAM-B3LYP), -13 (PBE0), -5
(B3LYP), -6 (B3PW91).
An alternative way of finding the corresponding value of IOp(3/74) for present DFT functional is carrying out
a simple calculation and check the value of IOp(3/74) automatically shown at the beginning of the output file, for
example, the output file using B3LYP/6-31G* level will contain a line "3/5=1,6=6,7=1,11=2,16=1,25=1,30=1,
74=-5/1,2,3;", showing that the IOp(3/74) is -5. More explanation of ExtraLinks=L608 can be found in
http://gaussian.com/faq1/.
A practical example of EDA-SBL analysis is given in Section 4.21.2.
3.24.5 Simple energy decomposition analysis
Please check the example give in Section 4.100.8. Currently to realize this function you need
to manually use subfunction 8 of main function 100.
3.100 Other functions, part 1 (100)
Since Multiwfn has too many functions, some of the functions are relatively "small" compared
to main functions, and some functions are not closely related to wavefunctions analysis, these
functions are classified as "other functions". Because the number of subfunctions in "other
functions" are huge, "other functions" are splitted as part 1 (main function 100) and part 2 (main
function 200). Part 1 will be described below, and part 2 will be intorduced in Section 3.200.
3.100.1 Draw scatter graph between two functions and generate their
cube files
This function allows grid data of two functions to be generated at the same time with sharing
grid setting, you can choose to export their cube files, view their isosurfaces and plot scatter graph
between them.
After you entered this function, select two real space functions that you are interested in, for
example you want to analyze real space function 16 and 14, you should input 16,14 (the first and
the second function will be referred to as functions 1 and 2 respectively below). Then select a mode
to set up grid points. After that Multiwfn starts the calculation of grid data for them. Once the
calculation is finished, Pearson correlation coefficient of the two functions in all grid points is
printed and a menu appears on screen, all options are self explanatory. If you choose -1 to draw
scatter graph, a graph like this will pop up immediately:
262
3 Functions
Each scatter point corresponds to each grid point, the position in X-axis and Y-axis corresponds
to the value of function 1 and function 2 at this point, respectively. Multiwfn determines the range
of axes automatically according to the minimum and maximum value, sometimes you have to use
option 4 and 5 to reset the range by yourself, otherwise barely points can be seen in the graph. The
size of points can be adjusted by “symbolsize” in settings.ini. The graph can be saved to current
directory by option 1. The X-Y data set of the points can be exported to output.txt in current directory
by option 2.
Option -2 and -3 set the value of function 2 where the value of function 1 is within or without
of a specific range respectively. Notice that the data once modified cannot be retrieved again.
Special usage: If you already have cube files of the two functions to be studied (referred to as
func1.cub and func2.cub, respectively) and you want to directly use the functions in post-processing
menu (e.g. plotting scatter map, modifying values), you should follow these steps: Input the path of
func1.cub after booting up Multiwfn, then enter main function 100 and select subfunction 1, then
input 0,0 when selecting real space functions, then input the path of func2.cub, after that the grid
data of the two cube files will be directly taken as function 1 and function 2. Notice that, of course,
the grid setting of func1.cub and func2.cub must be exactly identical.
3.100.2 Export various files or generate input file of quantum chemistry

programs
Exporting various kind of files

This function can be used to output current structure to .pdb, .xyz and .cml files. Wavefunction
information can be exported as .wfn or .wfx file when GTF information is available, and can be
exported as .molden, .fch, GENNBO input file (.47), old Molekel input file (.mkl) when basis
function information is presented. Clearly, Multiwfn can be used as a file format converter,
e.g. .mwfn/.wfn/.wfx/.fch/.molden/.gms ... → .pdb/.xyz/.cml/.gjf/.wfn/.wfx/.molden/.fch/.47 ... In
addition, it is worth to note that one can use Gaussian or other codes to generate .fch or .molden file,
263
3 Functions
then use Multiwfn to convert it to .mkl file, and then use orca_2mkl test -gbw to convert test.mkl to
test.gbw, so that ORCA can use wavefunction generated by other codes as initial guess.
If current input file is .chg format (see Section 2.5 for details), you can use option 2 to convert
it to .pqr file. The .pqr and .pdb formats are very similar, the major differences is that the former has
additional two columns to record atomic charges and atomic radii. In the resulting .pqr file, the
atomic charges are identical to that in .chg file, while the atomic radii column corresponds to Bondi
vdW radii. The .pqr can be directly loaded into VMD program, the atoms can be colored according
to atomic charges (if the fourth column of .chg file records other atomic information such as atomic
spin populations, in VMD the atoms can also be colored according to spin populations). This is very
useful for intuitively exhibiting atomic properties, see Section 4.A.10 for illustrations.
When there is a set of grid data in memory (may be loaded from e.g. .cub file or calculated by
e.g. main function 5), via option 30 you can export it to .vti file, which is a format supported by the
well-known and freely available volume data visualizer ParaView (https://www.paraview.org), also
at the meantime Multiwfn asks you if also exporting a .cml file recording current system with Bohr
as unit, if you choose to export it, you can use ParaView to load it so that grid data and molecule
structure can be simultaneously plotted.
Generating input file of quantum chemistry programs

This function is also able to yield basic input file for a batch of known quantum chemistry
codes based on current structure, net charge and multiplicity, including Gaussian, GAMESS-US,
ORCA, MOPAC, Dalton, MRCC, Molpro, NWChem, PSI, CFOUR, Molcas. The task type is single
point except for MOPAC.
For generating Gaussian or GAMESS-US input file, if basis function information is presented,
you can select if writing the orbital expansion coefficients to their input files so that the present
wavefunction can be used as initial SCF guess. Notice that you must then manually specify the basis
set in the input file as the one originally used to yield present wavefunction, otherwise the Gaussian
or GAMESS-US task must be failed.
Because ORCA is very popular, fast and has numerous important functions, while its keywords
are not as easy to specifiy as Gaussian, therefore a special interface is provided for generating ORCA
input file, in which commonly employed ORCA calculation levels can be selected, the generated
keywords are the most appropriate and efficient ones for corresponding level. Very detailed
description can be found in "On the function of generating ORCA input file in Multiwfn "(in Chinese,
http://sobereva.com/490), also there is an practical example "Simulating UV-Vis and ECD spectra
using ORCA and Multiwfn" (http://sobereva.com/485). The video "Study geometry, vibration, IR
spectrum and orbitals based on ORCA program and other codes" (https://youtu.be/tiTmTbtbtig) also
utilized this function.
Information needed: Atom coordinates, GTFs (for exporting .wfn/.wfx), basis functions (for
exporting .mwfn/.fch/.molden/.47)
3.100.3 Calculate molecular van der Waals volume
In this function Monte Carlo method is used for evaluating van der Waals (vdW) volume of
present system, two definitions of vdW region are provided: (1) The superposition of vdW sphere
264
3 Functions
of atoms. This definition is not very accurate, because electron effect is not taken into consideration,
but the speed of evaluation is very fast and wavefunction information is not required. (2) The region
encompassed by certain isosurface of electron density, the isovalue of 0.001 is suitable for isolated
system, while 0.002 is more suitable for molecules in condensed phase. One can choose the
definition by "MCvommethod" in settings.ini.
The principle of the Monte Carlo procedure is very simple: If we define a box (volume is L)
which is able to hold the entire system, and let N particles randomly distributed in the box, if n
particles are presented in the vdW region, then the vdW volume of present system is n/N*L. Of
course, the result improves with the increase of N. In Multiwfn, you need to define N by input a
number i, the relationship is N=100*2i, for small molecular when i=9 the accuracy is generally
acceptable, for large system you may need to increase i gradually until the result variation between
i and i+1 is small enough to be acceptable as converged. For definition 2 of vdW region, you also
need to input the isovalue of density, and the factor k used to define the box, see below illustration,
where Rvdw is vdW radius. If k is too small, then the vdW region may be truncated, however if k is
too large, more points are needed to maintain enough accuracy. For isovalue of 0.001, k=1.7 is
recommended.
Information needed: atom coordinates (for definition 1), GTFs (for definition 2)
3.100.4 Integrate a function over the whole space
This is a very useful and powerful function for integrating selected real space function in the
whole space. The numerical integration method used here is based on the one proposed by Becke in
the paper J. Chem. Phys., 88, 2547 (1988) for integrating DFT functional, which is also suitable for
any real space function, but notice that the function must be smooth and converges to zero at infinite
asymptotically. The accuracy is determined by the number of integration points, you can adjust the
number of radial and angular integration points by “radpot” and “sphpot” parameters in settings.ini,
respectively. The integrand can be selected from built-in functions, and you can also write new
function by yourself as user-defined function, see Sections 2.6 and 2.7.
Examination of difference of a real space function between two wavefunction files
265
3 Functions
By the way, if you would like to examine difference between two wavefunction files for a real
space function, you can select function -4 or -5 (hidden functions) in main function 100, the function
to be integrated will be [f(file1)-f(file2)]2 or |f(file1)-f(file2)|, respectively, where f is the real space
function you will select, file1 is the wavefunction file loaded when Multiwfn boots up, and file2 is
the wavefunction file you will choose in this function. In this function you can also set criterion of
electron density and thus let Multiwfn only evaluate the difference for low density region (i.e.
ignoring core region). In addition, after calculation, Multiwfn automatically exports grid data of
file1 as a plain text file in current file, the file name directly reflects current calculation condition.
For example, the file name a.wfn_003_0075_0434 implies that file1 is a.wfn, the 3rd real space
function was selected, the radial and angular integration points are 75 and 434, respectively. In later
studies, if all calculation conditions match with the file name (the file must be placed in current
folder), then Multiwfn will directly load data of file1 from this file rather than recalculate them to
reduce computational time.
Evaluation of spherically symmetric average ELF / LOL

In J. Comput. Chem., 38, 2258 (2017), the authors proposed that the optimal  parameter of
range-separated DFT functionals can be determined by means of below quantity:
 ELF(r)r ELF(r) d r
2
rELF =
 ELF(r) ELF(r) d r
In Multiwfn, this quantity will be automatically calculated and outputted if you select ELF as the
integrand in present function. The rELF is outputted as “spherically symmetric average ELF”, the
numerator and denominator in the root sign are also outputted together.
In J. Phys. Chem. C, 123, 4407 (2019), the LOL-tuning is proposed, in which the “spherically
symmetric average LOL” is involved:
 LOL(r)r LOL(r) d r
2
rLOL =
 LOL(r) LOL(r) d r
The rLOL along with its numerator and denominator in the root sign will be printed if LOL is selected
in present function.
An example is given Section 4.100.4.
Information needed: Atom coordinates, GTFs.
3.100.5 Show overlap integral between alpha and beta orbitals
For unrestricted wavefunctions, orthonormalization condition does not in general hold between
alpha and beta orbitals. This function computes the overlap matrix between alpha and beta orbitals
, j =  i (r ) j (r ) d r
Si  
The diagonal elements are useful for evaluating the matching degree of corresponding spin orbital
266
3 Functions
pairs, evident deviation to 1 indicates that spin polarization is remarkable.

In present function, there are two options, option 1 calculates the full overlap matrix, the
diagonal elements will be printed on screen and the whole overlap matrix can be selected to output
to ovlpmat.txt in current folder; in addition, the maximum pairing between Alpha and Beta orbitals
are shown. This calculation may be time-consuming for large system. Option 2 only calculates and
prints the diagonal elements, this is always fast.
Since the expectation of S2 operator for single determinant (SD) wavefunction can be easily
derived from the overlap matrix, if option 1 is selected, Multiwfn also outputs this quantity:
N N 
 S 2  SD = S 2 Exact + N  −  Si
2
,j
i j
where N and N are the number of alpha and beta electrons.  S  Exact is the exact value of
2
square of total spin angular momentum
N − N   N − N  
S 2
Exact = S ( S + 1) =  + 1
2  2 
Information needed: GTFs
3.100.6 Monitor SCF convergence process of Gaussian
Difficulty in SCF convergence is an annoying problem that often encountered in daily work,
monitoring the convergence is important for finding proper solutions. Multiwfn can monitor SCF
process by using the output file of Gaussian as input file. Notice that #P has to be specified in the
route section, otherwise no intermediate information of SCF process will be recorded in output file.
When you entered this function (subfunction 6 of main function 100), all information of
previous steps and the thresholds of convergence are printed on screen, such as
Step# RMSDP Conv? MaxDP Conv? DE Conv?
8 3.51D-06 NO 5.15D-05 NO -8.43D-08 YES
9 1.37D-06 NO 9.11D-06 NO -2.17D-09 YES
10 3.03D-07 NO 2.91D-06 NO -3.75D-10 YES
11 3.12D-08 NO 4.76D-07 YES -1.07D-11 YES
12 7.69D-09 YES 5.72D-08 YES -1.56D-13 YES
Goal 1.00D-08 1.00D-06 1.00D-06
SCF done!
Meanwhile a window pops up, which contains curves that corresponding to convergence process of
energy, maximum value and RMS variation of density matrix. After you close the window, you can
print the information and draw the curve graphs again in specific step range by choosing
corresponding options, the Y-axis is adjusted automatically according to the data range.
If the SCF task is work in progress, that is output file is updated constantly, every time you
choose to print and draw the convergence process, the Gaussian output file will be reloaded, so what
you see is always the newest information. For monitoring a time-consuming SCF process, I suggest
you keep the interface on until the SCF task is finished, during this period you choose option 2 every
267
3 Functions
so often to show the latest 5 steps and analyze convergence trend.

In the graph, gray dashed line shows the zero position of Y-axis, the red dashed line shows the
threshold of convergence. The picture below shows the last 10 SCF steps of a system. If “Done”
appears in the rightmost, that means corresponding property has already converged, here all three
terms are marked by “Done”, so the entire SCF process has finished.
This function is also compatible with keyword SCF=QC and SCF=XQC, but not with
SCF=DM.
3.100.8 Generate Gaussian input file with initial guess combined from
fragment wavefunctions
This function is used to combine several fragment wavefunctions to an initial guess

wavefunction, there are three main uses:
1 Generate high quality initial guess wavefunction for complex: If you already have
converged wavefunctions for each fragment, and the interaction between fragments is not very
strong, by using the combined wavefunction as initial guess the SCF process of complex will
converge faster.
2 Perform simple energy decomposition: The total energy variation of forming a complex
can be decomposed as
Etot = E complex −  Eifrag = ( Eels + Eex ) + Eorb = Esteric + Epolar
i
where Eels is electrostatic interaction term, normally negative if the two fragments are neutural; EEx
is exchange repulsion term, which comes from the Pauli repulsion effect and is invariably positive.
For convenience, it is customary to combine these two terms as steric term (Esteric).
268
3 Functions
Eorb in above formula is orbital interaction term, and sometimes also known as induction term
or polarization term. Eorb arises from the mix of occupied MOs and virtual MOs. If the combined
wavefunction is used as initial guess for complex, then Eorb can be evaluated by subtracting the the
first SCF iteration energy from the last SCF iteration energy.
Eorb = ESCF, last − ESCF,1st
Note that ESCF,last = E complex , obviously we can write out below relationship
Esteric = Eels + Eex = Etot − Eorb = ESCF,1st −  Eifrag

i
By the way, if the complex you studied concerns evident dispersion interaction (vdW
interaction), there are two possible ways to evaluate the dispersion energy component:
(1) Use Hartree-Fock to calculate interaction energy first ( E tot ), then use MP2 (or better post-HF
HF
method) to calculate interaction energy again ( E tot ), then

MP 2
Edisp = Etot
MP 2
− Etot
HF
. This
relationship comes from the fact that dispersion energy is completely missing in HF energy.
(2) Use HF or the DFT functionals that completely failed to represent dispersion energy to calculate
interaction energy (e.g. B3LYP and BLYP), then use Grimme's DFT-D3 program
(http://www.thch.uni-bonn.de/tc/index.php?section=downloads&) with corresponding parameter to
evaluate DFT-D3 dispersion correlation to interaction energy, which can be simply regarded as the
dispersion component in total interaction energy. If you do not know how to do this and you can
read Chinese, you may consult the post in my blog: http://sobereva.com/210.
An example of the simple energy decomposition is given as Section 4.100.8.
3 Modelling antiferromagnetic coupling system: I exemplify this concept and show you how
to use the function by a representative antiferromagnetic coupling system -- Mn2O2(NH3)8,
The ground state is singlet, while the two Mn atoms have opposite spin and each Mn atom has high
spin. Obviously, restricted closed-shell calculation is not suitable for this system, unrestricted
calculation is required, however, the default initial guess is non symmetry-broken state, therefore
the converged unrestricted wavefunction returns to restricted closed-shell wavefunction. In order to
make the wavefunction converges to expected state, we have to compute wavefunction for four
fragments separately and then combine them by Multiwfn to construct a proper symmetry-broken
initial guess. The four fragments should be defined as
Fragment 1: Mn(NH3)4 at left side. Charge=+2, sextet.
Fragment 2: Mn(NH3)4 at right side. Charge=+2, sextet.
Fragment 3: One of bridge oxygen atoms. Charge=-2, singlet.
Fragment 4: Another bridge oxygen atom. Charge=-2, singlet.
269
3 Functions
Notice that nosymm and pop=full keywords must be specified in the calculation of each
fragment. Assuming the output files are frag1.out, frag2.out, frag3.out and frag4.out, respectively,
let Multiwfn load frag1.out first after boot up, then select function 100 and subfunction 8 to enter
present function, input 4 to tell Multiwfn there are four fragments in total; since frag1.out has
already been loaded, you only need to input the path (including filename) of frag2.out, frag3.out
and frag4.out in turn. After that a Gaussian input file named new.gjf will be outputted in current
directory. Notice that everytime you input a fragment, Multiwfn asks you if flip its spin, only for
fragment 2 you should choose y, that is make the spin direction of unpaired electrons down (by
default the spin is in up direction) to exactly counteract the opposite spin in fragment 1, so that
multiplicity of complex is 1.
From the comment of new.gjf (the texts behind exclamation mark), you can know clearly how
the MOs of complex are combined from MOs of fragments. For example, a two-fragment system,
one of complex MOs in new.gjf is
! Alpha orbital: 12 Occ: 1.000000 from fragment 2
0.00000E+00 0.00000E+00 0.00000E+00 0.00000E+00 0.00000E+00
0.00000E+00 0.00000E+00 0.00000E+00 0.18850E-01 -0.53690E-01
-0.74180E-01 0.48861E+00 0.16080E+00 -0.12897E+00 -0.12897E+00
0.47150E-01 -0.25269E+00 0.39361E+00 -0.48811E+00 0.36123E+00
We already know there are 8 basis functions in fragment 1 and 12 basis functions in fragment 2, and
this complex MO comes from fragment 2, so the first 8 data (highlighted) are zero and only the last
12 data have values (the same as corresponding MO coefficients in Gaussian output file of fragment
2).
If you used diffuse functions and encounter problem at Link401 when running .new by
Gaussian, you can add IOp(3/32=2) keyword and retry.
3.100.9 Evaluate interatomic connectivity and atomic coordination

number
In the original paper of DFT-D3 (J. Chem. Phys., 132, 154104 (2010)), the authors argued that
the coordination number (CN) of an atom A can be approximately expressed as
1
CN A =  1 + exp{−16  [(4 / 3)( R
B A + RB ) / rAB − 1]}
A
where R is Pyykkö covalent radius from Chem. Eur. J., 15, 186 (2009), and rAB is distance between
A and B.
According this idea, in present module the interatomic connectivity index (I) between A and B
is determined as follows:
1
I AB ( rAB ) =
1 + exp{−16  [(4 / 3)( RA + RB ) / rAB − 1]}
Assume that RA+RB=2.0, then the function could be plotted as follows. The I value equals to
0.995 when rAB=RA+RB
270
3 Functions
Note that the I should not be utilized as an indicator of bond order, it does not have capability
of discriminating bonding type and strength.
Present module outputs I between each pair of atoms, the printing threshold can be inputted by
user. Commonly, when I is close to 1.0, it implies that the two atoms are bonded, while if it is close
to 0.0, then they may be regarded as not binded by chemical bond. The nearest integer of I, namely
nint(I), is also outputted for facilitating examination of the result.
For each atom (e.g. atom A), the I
B A
AB is printed as "Sum of connectivity", while
 nint(I
B A
AB ) is printed as "Sum of integer connectivity". The former and the latter may be
regarded as raw and actual coordination number, respectively.

Finally, you can choose if exporting all I values as matrix to connmat.txt in current folder.
Below is output example of ethyne at equilibrium geometry:

1C --- 2C : 0.99951 Nearest integer: 1
1C --- 3H : 0.99974 Nearest integer: 1
2C --- 4H : 0.99974 Nearest integer: 1
1 C Sum of connectivity: 2.0016 Sum of integer connectivity: 2
2 C Sum of connectivity: 2.0016 Sum of integer connectivity: 2
3 H Sum of connectivity: 1.0021 Sum of integer connectivity: 1
4 H Sum of connectivity: 1.0021 Sum of integer connectivity: 1
Information needed: atom coordinates
3.100.11 Calculate overlap and centroid distance between two orbitals
This function is used to calculate overlap and centroid distance between two orbitals, this is
useful for many purposes, e.g. analyzing charge transfer during electron excitation. You need to
input index of two orbitals, then X, Y, Z of centroid of the orbitals will be calculated as follows
X i =  | i (r) |2 x d r Yi =  | i (r ) |2 y d r Zi =  | i (r) |2 z d r
271
3 Functions
then the centroid distance between orbital i and j is calculated as
Dij = ( X i − X j ) 2 + (Yi − Y j ) 2 + ( Z i − Z j ) 2
Present function also calculates overlap degree of the two orbitals, below two quantities are
calculated respectively (while directly calculating overlap integral of two orbital wavefunctions is
clearly meaningless, since it must be zero due to orthonormalization condition):
 |  (r) ||  (r) | d r  |  (r ) | |  ( r ) |
2 2
i j i j dr
The integrals shown above are not calculated analytically but numerically via Becke's grid-
based integration approach. The integration grid can be set by "radpot" and "sphpot" in settings.ini,
the default values are high enough, and you may want to somewhat decrease them to reduced
computational cost for large system, especially when you want to study many orbital pairs.
When the calculation is finished, Multiwfn will ask you whether or not add the two centroids
as two additional dummy atoms (the symbol is Bq). If you choose y, then you can go to main
function 0 to visualize corresponding orbital isosurfaces by transparent or mesh style to examine
correspondence between centroid position and orbital shape.
3.100.12 Perform biorthogonalization between alpha and beta orbitals
Introduction
It is well known that for wavefunctions generated by unrestricted open-shell calculations (UHF
or UKS), the alpha and beta orbitals are often evidently mismatch with each other, this phenomenon
makes analysis of orbitals difficult, because one must simultaneously consider two set of orbitals.
Although restricted open-shell (ROHF or ROKS) calculation does not have this problem, the total
electronic energy, orbital energy and electron distribution is not as accurate as unrestricted open-
shell calculation.
Present function is used to perform biorthogonalization between alpha and beta orbitals for
unrestricted open-shell wavefunction with spin multiplicity higher than 1. Original alpha and beta
molecular orbitals will be respectively transformed to a set of new orbitals. Although finally there
are still two sets of orbitals, their wavefunctions have matched with each other almost perfectly,
therefore then you only need to discuss one set of orbitals.
The so-called biorthogonalization mentioned here specifically refers to simultaneously
satisfying two conditions: (1) For each set of spin orbitals, they are orthonormal with themselves (2)
Alpha orbitals are orthonormal with beta orbitals. Without applying the biorthogonalization, the
UHF/UKS orbitals only satisfy the first condition.
Algorithm details
The biorthogonalization is realized via singular value decomposition (SVD) technique. The
overlap integral matrix O between alpha and beta orbitals is first constructed, and then SVD is
applied to decompose it as O=UV , where  is a diagonal matrix, the diagonal elements are referred
†
to as "singular values", which essentially correspond to the overlap integrals between the orbitals
after the biorthogonalization transformation, under normal situations they should be very close to
272
3 Functions
1.0. The column matrix U (V) corresponds to the transformation matrix between the original orbitals
and the new orbitals of alpha (beta) spin. The coefficient matrix of the newly generated
biorthogonalized orbitals can be obtained as

Cbiortho = UC original
 
Cbiortho = VC original
Since U and V are unitary matrices, such a transformation does not affect observable quantities of
current system.
Notice that the biorthogonalization transformation in fact is not done for all orbitals at once,
because this will lead to mix between occupied and virtual orbitals and thus result in change of
observable properties. In Multiwfn, the transformation is successively carried out via below three
steps. The total number of orbitals of each spin will be denoted as ntot, the numbers of alpha and
beta electrons will be denoted as n and n, respectively. n>n is assumed.
(1) Biorthogonalization between all occupied alpha orbitals (1~n) and all occupied beta
orbitals (1~n). This step makes each resulting occupied beta orbital paired with a resulting alpha
orbital.
(2) Biorthogonalization between alpha orbitals (n+1~n) and all virtual beta orbitals
(n+1~ntot). This step makes each resulting "singly occupied" alpha orbital paired with a resulting
beta virtual orbital
(3) Biorthogonalization between all alpha virtual orbitals (n+1~ntot) and the virtual beta
orbitals that have not been paired (n+1~ntot).
After these three steps of tranformation, one-to-one pairing between all alpha and beta orbitals
should be nearly perfectly satisfied, difference of orbital wavefunction distribution of alpha orbital
i and the beta orbital with the same index should be negligible. Note that in biorthogonalization
steps 2 and 3, the utilized overlap integral matrix O should be reconstructed based on the coefficient
matrices updated at the last step.
The biorthogonalized orbitals are not eigenfunctions of Fock operator (or Kohn-Sham operator,
similarly hereinafter) as molecular orbitals, however their energies can be evaluated as expectation
value of Fock operator. Specifically, if you request Multiwfn to evaluate the orbital energies,
Multiwfn performs below representation transform:
Fbiortho = CTFAOC
where FAO is the Fock matrix in original basis functions that loaded from external file, C(,i)
corresponds to coefficient of basis function  in biorthogonalized orbital i. Energy of
biorthogonalized orbital j is simply Fbiortho(j,j). Once generation of orbital energies has done,
Multiwfn is able to order the orbitals according to their energies. Notice that the energy used in the
ordering process is average of energy of alpha orbital and its beta counterpart, and the three batch
of orbitals (1~n), (n+1~n) and (n+1~ntot) are ordered individually (hence for example, index of
an orbital originally in the second batch must still be higher than any orbital in the first batch after
ordering).
Since the number of unoccupied MOs is generally much higher than the number of occupied
MOs, while one often only has interest in occupied biorthogonalized orbitals, therefore Multiwfn
provides an option to skip the biorthgonalization between unoccupied MOs, namely skipping the
step (3) shown above. In this case the alpha and beta orbitals in the range of (n+1~ntot) will be
273
3 Functions
meaningless and you should not then study them.
Usage
After booting up Multiwfn, simply loading a file containing basis function information
(e.g. .mwfn, .fch, .molden, .gms) that generated by UHF or UKS calculation, then go to subfunction
12 of main function 100, Multiwfn will biorthgonalize the alpha and beta orbitals.
Next, you can choose whether or not performing biorthogonalization between unoccupied MOs,
as well as if generating energies of the biorthogonalized orbitals. For the last option, if you choose
y, you then need to input path of a file containing Fock (or Kohn-Sham) matrix in original basis
functions. The matrix elements should be provided in lower-triangular form, namely in this sequence:
F(1,1) F(2,1) F(2,2) F(3,1) F(3,2) F(3,3) ... F(nbasis,nbasis), where nbasis is the total number of
basis functions, the format is free. The Fock matrix may be obtained from output of some quantum
chemistry codes (In fact, you can ask Gaussian to produce NBO .47 file, Multiwfn is able to
automatically read the $FOCK field when the file name has .47 suffix). After that, you can choose
if ordering the biorthogonalized orbitals in the way mentioned earlier.
After that, two files are automatically exported to current folder
• biortho.txt: This file contains singular values and occupancy of the biorthogonalized orbitals.
If energies of the orbitals have been generated, they will also be written into this file.
• biortho.fch: This file contains biorthogonalized alpha and beta orbitals. If you did not request
Multiwfn to generate their energies, then in this file the “orbital energies” information will
correspond to their singular values; while if their energies have been generated, then these energies
will be written into this .fch file as “orbital energies” information.
Finally, if you input y, then the biortho.fch will be loaded directly, so that then you can directly
visualize the newly generated orbitals via main function 0 or analyze them by various functions; if
you input n, then the file that loaded when Multiwfn boots up will be reloaded to recover to the
initial state.
Application of the biorthogonalization transformation for a practical system is illustrated in

Section 4.100.12.
Information needed: Basis functions, plain text file containing Fock/KS matrix (optional)
3.100.13 Calculate HOMA and Bird aromaticity index
HOMA index
HOMA is one of the most popular indexes for measuring aromaticity. This quantity was
originally proposed in Tetrahedron Lett., 13, 3839 (1972), and then the generalized form was given
in J. Chem. Inf. Comput. Sci., 33, 70 (1993). The generalized HOMA can be written as (notice that
the HOMA formula has been incorrectly cited by numerous literatures)
i , j
HOMA = 1 −  (R ref − R i , j )2
i N
where N is the total number of the atoms considered, j denotes the atom next to atom i,  and RRef
are pre-calculated constants given in original paper for each type of atom pair. If HOMA equals to
274
3 Functions
1, that means length of each bond is identical to optimal value Rref and thus the ring is fully aromatic.
While if HOMA equals to 0, that means the ring is completely nonaromatic. If HOMA is significant
negative value, then the ring shows anti-aromaticity characteristic.
The inventor of HOMA developes the HOMA parameters in the following way, see Chem. Inf.
Comput. Sci., 33, 70 (1993) for detail
Rref = ( Rs + wRd ) / (1 + w)
2
=
( Rs − Rref ) + ( Rd − Rref )2
2
where Rs and Rd are experimental bond lengths of single bond and double bond, respectively. w=kd/ks,
where ks and kd are force constants of single and double bonds, respectively. Usually w is assumed
to be 2.0 (special case also exists, such as w=4.2 for BN bond). For example, the Rref and 
parameters for CO bond was derived based on the experimental lengths of C-O and C=O bonds in
formic acid with assumption of w=2.0.
HOMA can be calculated by subfunction 13 in main function 100. When you choose option 0,
Multiwfn will prompt you to input the indices of the atoms in the local system, for example,
2,3,4,5,6,7 (assume that there are six atoms in the ring. The input order must be consistent with atom
connectivity), then HOMA value and contributions from each atom pair will be immediately
outputted on the screen. For example, thiophene optimized under MP2/6-311+G**, the output is
Atom pair Contribution Bond length(Angstrom)
1(C ) -- 2(C ): -0.001852 1.382006
2(C ) -- 3(C ): -0.056708 1.421170
3(C ) -- 4(C ): -0.001852 1.382006
4(C ) -- 5(S ): -0.023886 1.712627
5(S ) -- 1(C ): -0.023886 1.712627
HOMA value is 0.891817
Since 0.891817 is close to 1, the HOMA analysis suggests that thiophene has prominent aromaticity.
The source of built-in  and RRef parameters are shown below:
• CC, CN, CO, CP, CS, NN, NO: Chem. Inf. Comput. Sci., 33, 70 (1993)
• BN: Tetrahedron, 54, 14913 (1998)
• BC: Struct. Chem., 23, 595 (2012)
The built-in parameters can be modified or supplemented by user via option 1.
Bird index
Bird index (Tetrahedron, 41, 1409 (1985)) is another geometry-based quantity aimed at
measuring aromaticity, and can be calculated by option 2 of subfunction 13 in main function 100.
The formula is
I = 100[1 − (V / VK )]
where
100  (N i. j − N )2
a
V= i
N i. j = −b
N n Ri , j
In the formula, i cycles all of the bonds in the ring, j denotes the atom next to atom i. n is the total
275
3 Functions
number of the bonds considered. N denotes Gordy bond order, 𝑁 ̅ is the average value of the N
values. Ri,j is bond length. a and b are predefined parameters respectively for each type of bonds.
VK is pre-determined reference V, for five and six-membered rings the value is 35 and 33.2,
respectively. The more the Bird index close to 100, the stronger the aromaticity is.
Available a and b parameters include C-C, C-N, C-O, C-S, N-O and N-N, they are taken from
Tetrahedron, 57, 5715 (2001), the B-N parameter was obtained from Tetrahedron, 54, 14913 (1998).
For other type of bonds user should provide corresponding parameter by option 3. By option 4 user
can adjust or add VK parameter.
Corresponding example of this function is provided in Section 4.100.13.
3.100.14 Calculate LOLIPOP (LOL Integrated Pi Over Plane)
In the paper Chem. Commun., 48, 9239 (2012), the authors proposed a quantity named
LOLIPOP (Localized Orbital Locator Integrated Pi Over Plane) to measure π-stacking ability of
aromatic systems, they argued that a ring with smaller LOLIPOP value has stronger π-depletion
(namely lower π-delocalization), and hence shows stronger π-stacking ability.
LOLIPOP is defined as definite integral of LOL-π (the LOL purely contributed by π-orbitals)
from a distance of 0.5Å away from the molecular plane. Only points with LOL-π>0.55 are taken
into account. The integration is made in a cylindrical region perpendicular to the molecular plane,
with a radius of 1.94Å corresponding to the average between the C and H ring radii in benzene.
In Multiwfn, function 14 in main function 100 is designed for calculating LOLIPOP, the default
value 0.5Å and 1.94Å mentioned above can be changed by option 4 and 3 respectively. Before
starting the calculation, you have to choose which orbitals are π orbitals by option 3, you can find
out π orbitals by visualizing orbital isosurfaces via main function 0. The calculation can be triggered
by option 0, you need to input the indices of the atoms in the ring in accordance with the atom
connectivity, then grid data of LOL around the ring will be evaluated, then the LOLIPOP value is
computed by numerical integration. Of course, smaller grid spacing gives rise to higher integration
accuracy, but brings severer computational requirement on evaluating LOL. The default spacing of
grid data is 0.08 Bohr, this value is fine enough in general.
3.100.15 Calculate intermolecular orbital overlap
This function is used to calculate orbital overlap integral between two molecules, namely
Siintmol
,j =  imonomer1 (r) monomer
j
2
(r)dr
where i and l are molecular orbital indices of monomer 1 and monomer 2, respectively. This integral
276
3 Functions
is useful in discussions of intermolecular charge transfer, e.g. J. Phys. Chem. B, 106, 2093 (2002).
Below three files are required for evaluating the integral, any kind of file containing basis
function can be used as the input files.
(1) Wavefunction file of dimer
(2) Wavefunction file of monomer 1
(3) Wavefunction file of monomer 2
To calculate the integral, after booting up Multiwfn, the file (1) should be loaded first. After
entering present module, the paths of files (2) and (3) should be inputted in turn. Then you can input
intmol
such as 8,15 to obtain 𝑆8,15 . If you input the letter o, then the entire Sintmol matrix will be outputted
to ovlpint.txt in current folder.
Notice that the atomic coordinates in files (2) and (3) must be in accordance with those in file
(1), and the atomic sequence should be identical. The basis set used for the three calculations must
be the same.
Special case:
If you are a Gaussian user, in addition to using .fch as input file, you can also use Gaussian output files for
present module. The three files in this case should be
(1) Gaussian output file of dimer, IOp(3/33=1) nosymm guess=only should be specified in route section.
Multiwfn will read overlap matrix from this file.
(2) Gaussian output file of monomer 1, nosymm pop=full should be specified in route section. Multiwfn will
read orbital coefficients of monomer 1 from this file.
(3) Gaussian output file of monomer 2, nosymm pop=full should be specified in route section. Multiwfn will
read orbital coefficients of monomer 2 from this file.
3.100.16 Calculate various quantities in conceptual density functional

theory (CDFT)
The conceptual density functional theory (CDFT) originally developed by Robert Parr is a
theory framework aiming for unraveling reactivity of chemical systems. CDFT contains numerous
concepts and quantities, some of them can be used to predict favorable reactive sites and reactive
character, and some of them can compare reactivity among different chemical species. Due to the
high popularity and important role of CDFT in quantum chemistry, as well as there are so many
meaningful relevant quantities, I believe it is very useful to develope a module to calculate all
commonly investigated quantities involved in CDFT with minimal steps.
In the Part 1 of this section, I will briefly describe the definition of all quantities that can be
studied via this module; then in the Part 2, I will show how to use this module. This module is able
to calculate the so-called orbital-weighted quantities, which will be specifically described in Part 3.
Note that aside from CDFT, Multiwfn also supports many other methods for revealing reactive
sites, see Section 4.A.4 for overview.
1 Theory
To yield all below quantities, electronic energy (E) and electron density of N, N+1 and N-1
electron states must be available. Commonly N refers to the number of electrons carried by a
chemical system at its most stable status. Geometry optimized for N electrons state is employed for
277
3 Functions
all calculations.
➢ Global indices
• Vertical ionization potential (VIP): E(N-1) − E(N)
• Vertical electron affinity (VEA): E(N) − E(N+1)
• Mulliken electronegativity (): (VIP+VEA)/2
• Chemical potential (): −
• Hardness (): VIP−VEA, which is also equivalent to fundamental gap. See J. Am. Chem.
Soc., 105, 7512 (1983). Note that according to the convention employed by many CDFT papers, the
prefix of 1/2 in original definition of  is dropped.
• Softness (S): 1/. See Proc. Nati. Acad. Sci., 82, 6723 (1985)
• Electrophilicity index (): 2/(2). See J. Am. Chem. Soc., 121, 1922 (1999)
• Nucleophilicity index (NNu): EHOMO(Nu) − EHOMO(TCE), where Nu denotes nucleophile, TCE
denotes tetracyanoethylene, whose HOMO energy is almost the lowest one among all organic
molecules and therefore it is chosen as reference system. See J. Org. Chem., 73, 4615 (2008).
➢ Real space functions

• Fukui functions f(r) and dual descriptor f(r): See Section 4.5.4 for detailed introduction
• Local softness: s+(r)=Sf +(r), s−(r)=Sf −(r), s0(r)=Sf 0(r) for nucleophilic, electrophilic, radical
attacks, respectively, where f(r) is Fukui function of corresponding type. See Proc. Nati. Acad. Sci.,
82, 6723 (1985)
• Local electrophilicity index: 𝜔loc (𝐫) = 𝜔𝑓 + (𝐫)
loc
• Local nucleophilicity index: 𝑁Nu (𝐫) = 𝑁Nu 𝑓 − (𝐫)
➢ Atom indices
• Condensed Fukui function (fA) and dual descriptor (fA): See Section 4.7.3 for detailed
introduction
• Condensed local softness
For nucleophilic attack: 𝑠𝐴+ = 𝑆𝑓𝐴+
For electrophilic attack: 𝑠𝐴− = 𝑆𝑓𝐴−
For radical attack attack: 𝑠𝐴0 = 𝑆𝑓𝐴0
where fA is condensed Fukui function of corresponding type.
• Relative electrophilicity index: 𝑠𝐴+ /𝑠𝐴− , see J. Phys. Chem. A, 102, 3746 (1998)
• Relative nucleophilicity index: 𝑠𝐴− /𝑠𝐴+, see J. Phys. Chem. A, 102, 3746 (1998)
• Condensed local electrophilicity index: 𝜔 𝐴 = 𝜔𝑓𝐴+
𝐴
• Condensed local nucleophilicity index: 𝑁Nu = 𝑁Nu 𝑓𝐴−
➢ cubic electrophilicity index

The electrophilicity index cubic introduced in J. Phys. Chem. A, 124, 2090 (2020) is somewhat
special, it is the only quantity that also relies on N-2 electron states. It includes higher-order term
than the aforementioned electrophilicity index . Its definition is
  
cubic =   1 + 2  
 3 
In practice, it is calculated as
278
3 Functions
( cubic )2  cubic 
cubic = 1 + 
2 cubic 
2cubic  3(cubic ) 
cubic = (1 / 6)( −2 VEA − 5VIP+ VIP2 )
cubic = VIP− VEA
 cubic = 2 VIP− VIP2 − VEA
where VIP2 is the second vertical ionization potential and defined as E(N-2) - E(N-1).
𝐴
Correspondingly, there is a cubic form of condensed local electrophilicity index 𝜔cubic = 𝜔cubic 𝑓𝐴+ .
𝐴
In J. Phys. Chem. A, 124, 2090 (2020) is shown that 𝜔cubic value of halogen atom (which behaves
as Lewis acid due to its -hole) in halogen-bond dimers R-X...NH3 has excellent correlation with
calculated binding energies (however, note that they employed AIM partition for atomic spaces
rather than the Hirshfeld partition utilized in the present module).
2 Usage
After entering the present module, namely subfunction 16 of main function 100, you will see
a menu, in which the options 2 and 3 are used to calculate above quantities. Before using them, you
must provide N.wfn, N-1.wfn and N+1.wfn in current folder, they contain wavefunction and
electronic energy of N, N-1 and N+1 states respectively for present system, the geometries must be
the same and correspond to the optimized geometry of N state. The calculation level of the three
files must also be the same.
Option 2: Used to calculate all aforementioned global indices and atomic indices, the result
will be exported to CDFT.txt in current folder. Because as mentioned in Section 4.7.3, Hirshfeld
method is an ideal choice for calculating condensed Fukui functions and may be other relevant
atomic indices, therefore Hirshfeld charges are automatically calculated and used for evaluation of
all atomic indices. Nucleophilicity index as well as its local version are dependent of HOMO energy
of TCE, which should be calculated using the same level for present system, notice that these indices
printed in present module simply employ the EHOMO(TCE) = -0.335198 Hartree calculated at the
commonly used B3LYP/6-31G* level (Clearly, if you want to get more reliable result and your
current calculation level is not B3LYP/6-31G*, you should calculate EHOMO(TCE) by yourself and
then manually evaluate these indices)
Option 3: Used to calculate grid data of various types of Fukui function and dual descriptor,
then their isosurface maps can be directly visualized, the grid data can be exported to cube files in
current folder. In this option you can set a value to be multiplied to the calculated grid data. For
example, if you set the value to the global softness outputted by option 2, then the f −(r) Fukui
function will correspond to s−(r).
For using present module, the file loaded after booting up Multiwfn is relatively arbitrary, the
only requirement is that the atomic information in this file is identical to the system under study.
Generation of .wfn files

You can manually prepare the three .wfn files used by options 2 and 3, alternatively, you can
use option 1 of present module to automatically realize the preparation work.
After choosing option 1, you will be prompted to input Gaussian keywords for single point
task, then Multiwfn asks you to input charge and spin multiplicity for N, N+1 and N-1 states in turn,
then Gaussian single point input files N.gjf, N+1.gjf and N-1.gjf will be generated in current folder
279
3 Functions
(the geometry in these files correspond to the geometry in the input file of Multiwfn). Then, you can
manually use Gaussian to run them to obtain N.wfn, N+1.wfn and N-1.wfn, or if Gaussian has been
installed on your computer, you can directly let Multiwfn to invoke Gaussian to calculate them (in
this case the "gaupath" parameter in settings.ini must have been set to actual path of Gaussian
executable file), after calculations the three .wfn files will appear in current folder.
Sometimes we need to use mixed basis set, in this case you should prepared a file named
basis.txt in current folder, which records the definition of basis set (may be also accompanied by
pseudopotential definition). If the inputted keyword contains "gen" or "genecp", then the content of
basis.txt will be automatically appended to the end of the generated .gjf files.
Multiwfn is also able to generate input file of ORCA for producing the three .wfn files. You
should choose option -2 and select ORCA, then option 1 will be able to generate N.inp, N-1.inp and
N+1.inp, after running them by ORCA manually, you will have N.wfn, N-1.wfn and N+1.wfn.
To use present module to study large organic systems, commonly I suggest using B3LYP/6-
31G* level, because this level is inexpensive, while the quality of the yielded quantities is already
satisfactory.
Note on calculating cubic

By default Multiwfn does not calculate cubic. If you need it, you should first select option -1
to switch its status to "Yes". Then you can use option 1 to help you to prepare .wfn file for N, N+1,
N-1, N-2 electron states, or you manually provide them. Then after selecting option 2, the resulting
CDFT.txt file will contain condensed local cubic, global cubic as well as VIP2.
An example of using present module to calculate various CDFT quantities for phenol is
provided in Section 4.100.16.1.
3 Special topic: Orbital-weighted Fukui function and dual descriptor
Theory
The originally defined Fukui function and dual descriptor do not work well when frontier
molecular orbitals are (quasi-)degenerate. For example, when HOMO and HOMO-1 have very
similar or exactly identical energies, the Fukui function f − may be unable to give meaningful result
or the result is fully misleading; in addition, when the system shows point group symmetry, such as
C60 fullerene, the distribution of f − is usually not in consistent with molecular symmetry, this is an
apparently unexpected observation.
In order to address these problems, in J. Comput. Chem., 38, 481 (2017), the authors proposed
orbital-weighted Fukui function, and in J. Phys. Chem. A, 123, 10556 (2019), they further proposed
orbital-weighted dual descriptor, they are summarized below (the 𝑓𝑤0 is defined by me)

exp[−(  −i ) 2 ]
f w+ (r ) =  wi | i (r ) |2 wi = 
i = LUMO
i = LUMO
 exp[ −(  −i ) 2 ]
HOMO
exp[−(  −i ) 2 ]
f w− (r ) =  wi | i (r ) |2 wi = HOMO
i
 i
exp[−(  −i ) 2 ]
+ −
f (r ) = [ f (r ) + f (r )] / 2
0
w w w
f w (r ) = f w+ (r ) − f w− (r )
280
3 Functions
where i and i are energy and wavefunction of orbital i;  is chemical potential and approximately
calculated as (EHOMO+ELUMO)/2 in the above formulae. The  is an adjustable parameter, in principle
its best value is the one able to make the functions have ideal predictability of local reactivity.
Clearly the most appropriate  is dependent of practical system, usually 0.1 Hartree is a worth-
trying guess. If you find the orbital-weighted functions under this value do not work well, you can
try to properly change it and redo calculations.
Compared to the frozen orbital approximation form of f −, namely f −(r)=|(r)|2, the advantage
of 𝑓𝑤− is that it takes all orbitals into account with different weights. From the expression it can be
seen that the closer a low-lying orbital energy is to the HOMO energy, the greater its weight.
Evidently degenerate orbitals share the same weight. The Gaussian function involved in the formula
behaves as a decay function, the larger the , the higher the contribution of low-lying orbitals to the
𝑓𝑤− . When energy difference between HOMO and HOMO-1 is significant, there will be no reason
to use 𝑓𝑤− instead of f −. The situation is similar for 𝑓𝑤+ , 𝑓𝑤0 and ∆𝑓𝑤 .
Usage
Since orbital-weighted functions involve virtual orbitals, you should use .mwfn, .fch, .molden
or .gms file as input file. Commonly, the geometry in the input file should correspond to the
optimized geometry of N-electron state; however, it is also possible to study them for
nonequilibrium structure, such as a point in intrinsic reaction coordinate (IRC).
Only closed-shell single-determinant wavefunction is acceptable. Diffuse functions should not
be used if you intend to calculate 𝑓𝑤+ , 𝑓𝑤0 and ∆𝑓𝑤 , since they utilize virtural orbitals, whose
chemical meaning may be severely broken when diffuse functions are employed.
In the subfunction 16 of main function 100, four options are related to the orbital-weighted
calculation:
• Option 4: Set the  parameter used in the subsequent orbital-weighted calculations
• Option 5: Print the highest 10 weights (i.e. the {w} in the aforementioned formulae) involved
in the orbital-weighted calculations. This option is useful to check if current  parameter is
reasonable and helps user to better understand how the orbital-weighted method works
• Option 6: Calculating condensed 𝑓𝑤+ , 𝑓𝑤− , 𝑓𝑤0 and 𝑓𝑤 values, in other words, calculating
integration of these functions in Hirshfeld atomic spaces. The result is useful in quantitatively
examing net amount of these functions at various atoms. The default radial and angular integration
points are usually fine enough, if you find the sum of condensed 𝑓𝑤+ or 𝑓𝑤− deviates from 1.0
evidently, you should set "iautointgrid" parameter in settings.ini to 0 and then properly enlarge
"radpot" and "sphpot" parameters.
• Option 7: Calculating grid data of 𝑓𝑤+ , 𝑓𝑤− , 𝑓𝑤0 and 𝑓𝑤 functions, then you can directly
visualize their isosurfaces or export them as cube files so that you can render them via third-part
softwares such as VMD and ChimeraX.
Examples of using this module to calculate orbital-weighted Fukui function and orbital-
weighted dual descriptor are provided in Section 4.100.16.2.
3.100.18 Yoshizawa's electron transport route analysis
Theory
This function is used to analyze electron transport route and is mainly based on Yoshizawa's
281
3 Functions
formula (Acc. Chem. Res., 45, 1612 (2012)), At the Fermi energy, the matrix elements of the zeroth
(0)𝑅/𝐴
Green's function, 𝐺𝑟𝑠 , which describes the propagation of a tunneling electron from site r to
site s through the orbitals in a molecular part, can be written as follows:
Crk Csk*
Grs( 0) R/A ( E F ) = 
k E F −  k  i
where Crk is the kth MO coefficient at site r, asterisk on the MO coefficient indicates a complex
conjugate, εk is the kth MO energy, and η is an infinitesimal number determined by a relationship
between the local density of states and the imaginary part of Green's function. By this formula, if
two electrodes are connected to site r and s, by this formula we can readily predict the transmittion
probability between the two sites. From the expression of the denoiminator, it can be seen that
HOMO and LUMO have the largest contribution to G. If only HOMO and LUMO are taken into
account, the formula can be explicitly written as
Cr HOMO Cs*HOMO C C*
Grs( 0) R/A ( E F ) = + r LUMO s LUMO
E F −  HOMO  i E F −  LUMO  i
This formula allows one to visually examine the possibility of electron transmission between site r
and s by means of observing molecular orbital diagram. If both site r and s have large C in magnitude,
and their relative phase in HOMO and in LUMO is different, then the magnitude of G will be large,
indicating that transmission between r and s will be favourable. For examples, see Acc. Chem. Res.,
45, 1612 (2012). Note that G may be positive and negative, but only absolute value of G is
meaningful for discussing transmission.
In Multiwfn, iη term in G is always ignored, since this is an infinitesimal number. The
coefficient C comes from the output of "NAOMO" keyword in NBO program. In almost all organic
conductors, σ orbitals have little contribution to conductance, therefore present function only takes
π orbitals into account, and the molecular plane must be parallel in XY or YZ or XZ plane.
Input file
To run this kind of analysis, there are two choices on the input file
(1) Only using Gaussian output file
In the route section, pop=nboread must be specified, and $NBO NAOMO $END must be
written at the last line, for example:
#P b3lyp/6-31g* pop=nboread
b3lyp/6-31g* opted
0 1
[Molecular geometry field]
$NBO NAOMO $END
(2) Load a file containing basis function information when Multiwfn boots up, then after
entering present function, load a NBO output file containing NAOMO information (namely the
NAOMO keyword has been passed to NBO program. If you are using GENNBO, you can add
NAOMO into $NBO…$END field of .47 file).
282
3 Functions
Usage
After you enter this function, you need to select which plane is the one your molecular plane
parallel to. Then program will load file and find out the atoms having expected π atomic orbital
(restrictly speaking, the "Val"-type p natural atomic orbitals that perpenticular to the chosen plane).
The coefficient of these atomic orbitals will be used to compute G.
Then you will see a menu, the options are explained below:
-4 Set distance criterion: Input lower and upper limits. Then in option 2 and 3, the route whose
distance exceeds this criterion will not be shown.
-3 Set value criterion: In option 1, the MO whose contribution to G is less than this criterion
will not be shown. In option 2 and 3, the route whose |G| is smaller than this criterion will not be
shown.
-2 Set Fermi energy level: Namely set EF in the Yoshizawa's formula.
-1 Select the range of MOs to be considered: Namely set the MO range of the summation in
the Yoshizawa's formula.
0 View molecular structure: As the title says.
1 Output detail of electron transport probability between two atoms: You need to input
index for two atoms, they will be regarded as site r and s, then Grs (the value behind "Total value")
will be outputted; meanwhile program also outputs the contribution from each MO, the distance
between the two atoms, and the calculated Grs for the case when only HOMO and LUMO are taken
into the summation.
2 Output and rank all electron transport routes in the system: All routes in current system
will be tested, if the route simultaneously fulfills the G and distance criteria set by -3 and -4, then
the involved atoms, G and distance of the route will be printed. The routes are ranked by aboslute
value of G.
3 Output and rank all electron transport routes for an atom: Similar to option 2, but only
consider the routes involving specific atom.
Information needed: As mentioned above
3.100.19 Generate promolecular .wfn file from fragment wavefunctions
This function is used to generate promolecular wavefunction in .wfn format based on fragment
wavefunctions. In the promolecular state, the electron distribution simply comes from the
superposition of the electron distribution of the related fragments; in other words, in this state the
electron transfer and polarization between the fragments have not started. Each promolecular orbital
directly corresponds to one of fragment orbitals.
The input file can be in any wavefunction format, such as .mwfn, .wfn, .wfx, .fch, .molden
and .gms, any kind of wavefunction is supported. Infinite number of fragments are supported.
Assume that the promolecular wavefunction consists of N fragments, after boot up Multiwfn,
you should make Multiwfn load the first fragment wavefunction file, and then go to this function,
input the total number of fragments (N), then input the path of the wavefunction files corresponding
to the other fragments in turn. The promolecule .wfn file will be generated in current folder as
promol.wfn.
For SCF wavefunctions, the MOs are sorted according to orbital energies from low to high,
283
3 Functions
virtual MOs are not recorded. For post-HF wavefunctions, the natural orbitals are sorted according
to orbital occupation from high to low.
If any fragment is unrestricted open-shell, then the promolecular wavefunction will be regarded
as unrestricted open-shell type, namely the alpha and beta orbitals will separately occur in the
resultant .wfn file. In this case, for each unrestricted open-shell fragment Multiwfn will ask you if
flipping the spin of its orbitals; if you choose yes, then the energies and occupations of all alpha and
all beta orbitals will be exchanged; this treatment is particularly important if you want to get the
promolecular wavefunction combined from free-radical fragments.
Please always carefully check the resultant .wfn file to examine its reasonableness before using
it.
Some examples are given in Section 4.100.19.
3.100.20 Calculate Hellmann-Feynman forces
The Hellmann-Feynman (H-F) force is the actual force acting on nuclei in a quantum system,
which is exerted by electron density and other nuclei:
Hˆ  (r )(R A − r ) Z (R − R B )
FA = −   = FAele + FAnuc = − Z A  dr + Z A  B A
R A | RA −r | 3
B A | R A − R B |
3
Present function calculates and prints total H-F force as well as the contribution from electron
density and other nuclei respectively.
In practice, notice that the H-F forces calculated as above based on the electronic wavefunction
and nuclear information recorded in the wavefunction file are generally not equivalent to the forces
computed by quantum chemistry program at current calculation level (FQC), which may be
expressed as:
E Hˆ
FAQC = − =−   − 2  / R A Hˆ 
R A R A
The partial derivative of wavefunction with respect to nuclear coordinate involves partial derivative
of orbital coefficients, configuration state coefficients and basis functions with respect to RA.
Therefore, only for the fully variational wavefunctions such as HF, DFT and MCSCF with basis
functions independent of nuclear coordinates (e.g. plane wave), the second term on the r.h.s. of
above equation is vanishing, and then the H-F forces just equals to the force at current calculation
level.
3.100.21 Calculate properties based on geometry information for

specific atoms
Input index range of some atoms (or input all to select the whole system), then many properties
284
3 Functions
based on geometry information will be calculated for them, including

(1) Mass
(2) Geometry center, center of mass, center of nuclear charges
(3) Radius of gyration
(4) Sum of nuclear charges and dipole moments from nuclear charges
(5) The atom having minimum/maximum coordinate in X/Y/Z
(6) Minimum and maximum distance
(7) Moments of inertia tensor, principal axes and principal moments of inertia
(8) Rotational constant
(9) Electrostatic interaction energy between nuclear charges
Note: If the input file is .chg, then "nuclear charges" mentioned above will stand for atomic
charges.
 m (r − r )
i i c
2
Radius of gyration is computed as i

, where r is the coordinate of nucleus,
m j
j
rc is mass center, m is atomic mass.

The moments of inertia tensor, which is a symmetry matrix, is calculated as
 
I XZ  
mi ( yi2 + zi2 ) −  mi xi yi −  mi xi zi 
 I XX I XY
 i i i

I =  I YX I YY I YZ  =  −  mi yi xi i i i i
m ( x 2
+ z 2
) − i mi yi zi 
 
 I ZX I ZZ   
i
I ZY


− i mi zi xi −  mi zi yi
i

i
mi ( xi2 + yi2 )

In the output, "The moments of inertia relative to X,Y,Z axes" correspond to the three diagonal terms
of I.
The three eigenvectors of I are known as principal axes, they are outputted as the three columns
of the matrix with the title "***** Principal axes (each column vector) *****". The corresponding
three eigenvalues of I are the moments of inertia relative to principal axes.
If the moments of inertia are given in amuÅ2, then the rotational constants in GHz can be
directly evaluated; for example, the rotational constant relative to Z axis is h / (8 I ZZ  )  10 ,
2 11
where h is Planck constant, and =1.66053878*10-27 is a factor used to convert amu to kg.
If you input size in this function, the system will be properly rotated to make its three principal
axes respectively parallel to the three Cartesian axes. Then, according to position of boundary atoms
and vdW radii proposed by Bondi (J. Phys. Chem., 68, 441 (1964)), the radius, diameter, length,
width and height of present system are calculated and printed. The length, width and height can be
visualized as a box enclosing the system in a GUI window, and the system can be exported as
new.pdb file in current folder, which can be further visualized in VMD program. An example is
provided in Section 4.100.21.
If you input dist in this function, you will be asked to define two fragments, then minimum and
maximum distances between the fragments, as well as distances between their geometry centers or
285
3 Functions
between their mass centers, will be outputted.
3.100.22 Detect π orbitals, set occupation numbers and calculate π

composition
You are highly encouraged to read this paper Tian Lu, Qinxue Chen, Theor. Chem. Acc., 139, 25 (2020), in
which the algorithm of this module is very detailedly described and many interesting research examples are given.
If this module is involved in your work, please not only cite Multiwfn original paper but also cite this paper, thanks!
This function is designed for automatically identifying π orbitals, setting occupation numbers
and evaluating π composition of a given set of orbitals. This function is crucially useful for studying
π electrons, e.g. calculating ELF-π. After you entered present function (subfunction 2 of main
function 100), you will see an interface, you should properly select an option according to the type
of your orbitals, the methods for detecting delocalized and localized orbitals are completely different.
Detecting π orbitals for delocalized orbitals

If your input file contains delocalized orbitals, e.g. molecular orbital (MO), natural orbital (NO)
and natural transition orbitals (NTO), and meantime all atoms are in XY or YZ or XZ plane, you
should select "0 Orbitals are in delocalized form". The program will automatically find the actual
molecular plane, then during orbital detecting process, indices, occupation numbers and energies of
identified π orbitals are shown on screen. After that, you can choose option 1 to set the occupation
numbers of these π orbitals to zero, or choose option 2 to set the occupation number of all other
orbitals to zero. Option 3 is similar to option 1, but only valence π orbitals are taken into account.
Option 4 is used to clean occupation number for all orbitals except for valence π orbitals.
For a not exactly planar system,  and  orbitals cannot be strictly separated or defined. Sometimes, local region
of a system is planar, but other atoms are not exactly in the same plane, in this case it is also possible to use present
mode to detect -like orbitals that delocalize on the local planar region. For instance, for toluene, you want to obtain
index of all -like orbitals delocalized on the six-membered ring. Before single point calculation of this system, you
should make the ring exactly parallel to the XY plane, then after calculation and loading the resulting wavefunction
file into Multiwfn, you should use option -4 of main function 6 to remove all GTFs on the methyl group. After that,
enter present function (subfunction 2 of main function 100) and choose option 0. In this case the program will ask
you to manually choose a plane because the program is unable to automatically determine the plane, you should
select the option corresponding to XY plane, and then input two tolerances. Finally, three -like orbitals will be
successfully identified.
Detecting π orbitals for orbitals in localized form

If your input file contains orbitals in localized form (localized molecular orbital (LMO) is
commonly employed, but others such as natural bond orbital are also supported), you should select
option "-1 Orbitals are in localized form", you will enter an interface, in which if you select "0
Detect pi orbitals and then set occupation numbers", the  orbitals will be detected, and then you
can set occupation number for the  orbitals or other orbitals.
The algorithm employed for this purpose was proposed by me (to be published). First,
Multiwfn calculates orbital composition of present orbital and finds out the atom having largest
contribution and the one having second largest contribution (they will be referred to as atoms A and
B, respectively). If contribution of A is larger than a given threshold (default is 85%), then the orbital
will be regarded as single-center orbital and thus be skipped. If the orbital is not singly centered,
orbital density, namely |(r)|2, at two probe points 0.7RA+0.3RB and 0.3RA+0.7RB, will be
calculated, where RA and RB are coordinates of atoms A and B, respectively. If orbital density at
286
3 Functions
both points is smaller than a given threshold (default is 0.01 a.u.), the orbital will be finally identified
as π orbital. The idea of this identification method and the reason of introducing the density threshold
is easy to understand: If an ideal π orbital forms between two atoms, since π orbital have a nodal
plane along the bond, the orbital density at the two probe points on the linking line between the two
atoms should be exactly zero. However, when the two atoms are not in an exactly planar local region,
since  and  orbitals are not strictly separable due to unavoidable - mixing, the orbital density
along the linking line must not be completely vanished, therefore a threshold is set to tolerate this
circumstance.
The interface has many options. By options 1 and 2, you can set the two thresholds used in the
detecting algorithm described above. By option 3, you can choose the range of orbitals for which
the  ones will be identified. By option "5 Set constraint of atom range", you can set constraint of
atom range by inputting atom indices. For example, if you inputted 2,4-7,9, then only the orbitals in
which both the two atoms with largest contributions are in the range of 2,4,5,6,7,9 may be finally
identified as π orbitals. Clearly, you can use this feature to identify all π type of LMOs lying at an
interesting region, e.g. conjugated ring.
The default way of calculating the orbital compositions used for identifying the  orbitals is
"Mulliken+SCPA", namely Mulliken method is used for occupied ones while SCPA method is
employed for unoccupied ones, notice that in this case diffuse functions should never be used,
otherwise the resulting compositions will be useless. The reason why SCPA is employed for
unoccupied orbitals is that Mulliken method often gives physically meaningless compositions for
them (e.g. atom with contribution larger than 100% or smaller than 0%). Via option "6 Set the
method for calculating orbital composition", you can also choose other method for computing the
compositions. Hirshfeld or Becke method is fully compatible with diffuse functions, they should be
used if you have to employ diffuse functions when generating wavefunction. However, these two
methods are relative more expensive and according to my test, they sometimes result in wrong
identification of unoccupied  orbitals (one main reason is that their results do not faithfully reflect
orbital nodal character). More information about orbital composition evaluation can be found in
Section 3.10.
It is worth to note that only the LMOs yielded by Pipek-Mezey orbital localization method
show separation character of  and  and thus could be used in for present function. Although
Multiwfn also supports Foster-Boys localization, the  and  characters are mixed together and
result in banana type of bonds, and thus the resulting LMOs cannot be used in conjection with
present function.
Evaluating π composition for given orbitals

After selecting the "-1 Orbitals are in localized form" mentioned above, if you choose "-1
Detect pi orbitals and then evaluate pi composition for orbitals in another file", then the π type of
LMOs will be detected in aforementioned way first, then you will be asked to input path of a
wavefunction file, whose geometry and the number of basis functions must be exactly identical to
present wavefunction.π composition of the orbitals in this file will then be evaluated based on the
detected π LMOs. For example, the π composition of orbital i is evaluated as:
287
3 Functions
 LMO
i = 100%  c j
2
j ,i
c j ,i = j i =  C ,iC , j   
 
where C is orbital coefficient matrix,  stands for basis function. You will be asked to input a
threshold, only the orbitals having π composition larger than this threshold will be printed on screen.
For unrestricted wavefunction, the π composition of alpha and beta orbitals are evaluated
respectively based on alpha and beta π type of LMOs.
Note that after this analysis, the wavefunction in memory will correspond to the file you newly
loaded.
There is a very important option "3 Switch the orbitals in consideration", which determines
which LMOs will be employed for computing  compositions. This option has below two status,
you need to properly choose the status according to the orbitals you want to study.
• "Occupied localized orbitals": This is default status, only occupied π LMOs will be detected
and employed for evaluating  compositions. If you only want to study  compositions of occupied
MOs, occupied biorthogonalized orbitals or occupied NTOs, then this status is appropriate. because
it can be easily demonstrated that these orbitals are only contributed by occupied LMOs. In this case
only occupied LMOs are needed to be obtained first.
• "All localized orbitals": All π LMOs will be detected and used for evaluating π compositions,
clearly both occupied and unoccupied LMOs must be available prior to the analysis. This status can
be used to evaluate π compositions for unoccupied MOs/NTOs/biorthogonalized orbitals, which are
only contributed by unoccupied LMOs. If you want to evaluate π compositions for natural orbitals,
you also need to switch to this status, because these orbitals can be contributed by both occupied
and unoccupied LMOs.
The use of this function for orbitals in localized form is illustrated in Section 4.100.22. Sections
4.4.9 and 4.5.3 also involves this function, but for the case of delocalized orbitals.
Information needed: GTFs (delocalized orbital case), basis functions (localized orbital case),
atom coordinates
3.100.23 Fit function distribution to atomic value
This function is very similar to the function used to fit ESP charge (see Section 3.9.10 and
3.9.11), but the real space function to be fitted is not limited to ESP, for example you can fit average
local ionization energy or even Fukui function distributed on molecular surface to atomic values.
After select option 1, and then select a real space function, the function value will be calculated
on the fitting points and then fit to atomic value by least-squares method, namely minimizing below
error function
F ( p1 , p2 ... p N ) =  [V (ri ) − V (ri )]
2
where i denotes the index of fitting points, p is atomic value, V is the value of real space function,
while V' is the function value evaluated by atomic value, which is defined as
288
3 Functions
pA
V (ri ) = 
A rA,i
where A denotes atom index, rA,i corresponds to the distance between nucleus of atom A and the
fitting point i.
The error of fitting are measured by RMSE and RRMSE
 V (r ) − V (r )
i i
2
 V (r ) − V (r )
i i
2
RMSE = i
RRMSE = i
N  V (r )
i
i
2
where N denotes the number of fitting points.

By default all fitting centers are placed at nuclei, you can also load additional fitting centers
from external file by option -2, see Section 3.9.10 for the format.
The default fitting points are exactly the ones used by Merz-Kollman method (see Section
3.9.11). You can use option 3 to customize the number of layers, use option 2 to adjust the density
of the fitting points distributed on the surface, and use option 4 to set the scale factor of the vdW
radii used for constructing each layer. (e.g. If you set two layers, and their scale factors are set to 1.0
and 1.2 respectively, then the first and the second layer will be constructed by superposing the vdW
radii of all atoms multiplied by 1.0 and 1.2, respectively)
No constraint on the total value (viz. ∑𝑖 𝑝𝑖 ) is applied by default. However, you can use option
5 to set a constraint on the total value. Evidently, if the constraint is set to the net charge of your
system, and you choose ESP as fitting function, then the result will be identical to MK charges.
The fitting points can be directly defined via external plain text file by option -1 (and thus will
not be constructed by superposing vdW spheres), see Section 3.9.10 for the file format. If the file
contains calculated function values of all fitting points and meanwhile the first line of this file (the
number of points) is set as negative value, then the real space function at each fitting point will not
be evaluated by Multiwfn but loaded from this file directly.
3.100.24 Obtain NICSZZ value for non-planar or tilted system
Introduction
Nucleus-independent chemical shift (NICS) is a very popular index used to measure
aromaticity. In many literatures, such as Org. Lett., 8, 863 (2006), It was shown that NICS(0)ZZ or
NICS(1)ZZ is a better index than the original definition of NICS, which is current known as NICS(0).
For exactly planar systems, if the system plane is parallel to XY plane, then NICS(0)ZZ means
the ZZ component of magnetic shielding tensor at ring center. The only different from NICS(1)ZZ to
NICS(0)ZZ is that the calculated point is not ring center, but the point above (or below) 1 Å of the
plane from ring center. Note that the definition of ring center is highly arbitrary, the original
definition uses geometry center, while some people use center of mass, and some researchers
recommend using ring critical point (RCP) of AIM theory as ring center, for example WIREs
Comput. Mol. Sci., 3, 105 (2013). (Personally, I think using RCP is the best choice)
If the ring of interest is skewed, not exactly planar or tilted, calculation of NICSZZ is somewhat
289
3 Functions
cumbersome, because one cannot directly acquire the component of magnetic shielding tensor
perpendicular to the plane from output file of quantum chemistry programs. Moreover, for
NICS(1)ZZ, it is difficult to properly set the position to be calculated. Present function is designed
to solve these difficulties.
In this function, the component of magnetic shielding tensor perpendicular to a given ring is
calculated as ⊥=uTu, where  is magnetic shielding tensor, u is column unit vector perpendicular
to the ring, and uT is transpose of u.
Steps for obtaining NICS(1)ZZ

If you want to calculate NICS(1)ZZ for a non-planar system, you should follow below steps:
(1) Use Multiwfn to open a file containing atomic coordinates of your system
(e.g. .xyz/.pdb/.mol/.wfn/.mwfn/.fch/.molden...)
(2) Determine ring center. You can use topology analysis module (main function 2) to locate
RCP, or use subfunction 21 in main function 100 to obtain geometry center or center of mass.
(3) Enter subfunction 24 of main function 100 (namely the present function), input the ring
center you just obtained, and input index of a series of atoms to fit the ring plane. Commonly the
inputted atoms should be all atoms in the ring of interest. Then the coordinate of the points above
and below 1Å of the ring plane from the ring center will be outputted.
Hint 1: The unit normal vector perpendicular to the ring plane is also outputted by Multiwfn, by which you can
easily derive the position used to calculate such as NICS(2), NICS(3.5)...
Hint 2: Step (2) can be skipped if you simply want to use geometry center and all atoms in the ring are selected
to fit the plane, because as mentioned in the prompts on screen, if you directly press ENTER button when Multiwfn
asks you to input the ring center, then it will be automatically determined as the geometry center of the atoms you
selected for fitting the ring plane.
(4) Use any of the two points obtained in last step to calculate magnetic shielding tensor at
corresponding position by quantum chemistry program
(5) Input all components of the magnetic shielding tensor in Multiwfn according to the output
of your quantum chemistry program. Then the negative value of "The shielding value normal to the
plane" outputted by Multiwfn is just NICS(1)ZZ.
Note that if the system is not symmetric with respect to the plane, in fact the NICS(1)ZZ and
NICS(-1)ZZ are different. To obtain the NICS(1)ZZ in common sense, you can take their average
when appropriate.
Steps for obtaining NICS(0)ZZ

For calculating NICS(0)ZZ, the process is more simple:
(1) Identical to the step 1 shown above
(2) Identical to the step 2 shown above
(3) Use the ring center you obtained in step (2) to calculate the magnetic shielding tensor at
this position by quantum chemistry program.
(4) Enter subfunction 24 of main function 100, input ring center, and input index of a series of
atoms to fit the ring plane. Then input all components of the magnetic shielding tensor according to
the output of your quantum chemistry program. Then the negative value of "The shielding value
normal to the plane" outputted by Multiwfn is just NICS(0)ZZ.
There is a blog article illustrating this function: http://sobereva.com/261 (in Chinese).
290
3 Functions
3.200.1 Calculate core-valence bifurcation (CVB) index and related

quantities
(1) Theory of CVB index

The idea of the so-called core-valence bifurcation (CVB) index was firstly proposed in Theor.
Chem. Acc., 104, 13 (2000), this index was defined based on electron localization function (ELF)
and mainly used to distinguish strength of various kind of hydrogen bonds (H-bonds). For a H-bond
of typical form (D-HA, where D=donor, H=hydrogen, A=acceptor), this index is expressed as:
CVB index = ELF(C-V) – ELF(DH-A)
where ELF(C-V) corresponds to the ELF bifurcation value between ELF core domain and valence
domain, while the ELF(DH-A) stands for the ELF value at bifurcation point between V(D,H) and
V(A).
In the above-mentioned Theor. Chem. Acc. paper, the authors examined many H-bond dimers
composing of HF and various kinds of monomers, it was found that the CVB index has good linear
relationship with H-bond binding energy. In some succeeding papers, such as Struct. Chem., 16, 203
(2005) and J. Phys. Chem. A, 115, 10078 (2011), this point has been further confirmed, and in the
former it was pointed out at CVB index “is positive in the case of weak complexes and negative in
stronger ones”. In addition, in Chem. Rev., 111, 2597 (2011) the author stated that CVB index “is
positive for weak hydrogen bond, and it decreases if the strength of this interaction increases; usually
this index is negative for strong hydrogen bonds”.
I found the ELF(C-V) and ELF(DH-A) themselves sometimes have better linear relationship
with H-bond binding energy than CVB index, therefore I suggest you also examine this point in
your practical studies when you intend to use CVB index.
(2) Manual evaluation of CVB index

Below I will show how to manually evaluate the two terms involved in the CVB index. HFHF
dimer is taken as example, the wavefunction file is provided as examples\HF_HF.wfn, it was
generated at B3LYP-D3(BJ)/def2-TZVP level, the optimization was also conducted at this level.
The D, H, A atoms in this system correspond to F2, H1, F3, respectively
The ELF(DH-A) term is defined unambiguously in the original paper. The topology analysis
module of Multiwfn is able to locate bifurcation points of ELF, namely (3,-1) critical points of ELF.
Then you can check ELF value of the bifurcation point lying between the hydrogen and acceptor
atom (Section 4.2.2 illustrated how to perform topology analysis for LOL. ELF can be analyzed in
similar way). However, topology analysis of ELF is time-consuming for large system. Considering
the fact that the actual ELF bifurcation point between V(D,H) and V(A) is almost exactly lying on
the straight line linking H and A, it is better to use main function 3 of Multiwfn to plot a ELF curve
map from the H to A and then directly read the value of corresponding minimum. Below is a
screenshot of ELF topology analysis result for the HFHF dimer
291
3 Functions
The purple and orange spheres are (3,-3) and (3,-1) type of ELF critical points (CPs), respectively.
The (3,-1) CP pointed by the arrow corresponds to the aforementioned ELF bifurcation point
between V(D,H) and V(A), its ELF value was found to be 0.06487, which is just the ELF(DH-A)
of present system.
As can be seen from the above graph, the red linking line basically crosses the center of the
orange sphere, this is why the ELF(DH-A) can also be approximately evaluated based on the ELF
curve map between H1 and F3. The curve map plotted using main function 3 is shown below
The minimum highlighted by the arrow is 0.06482, which is very close to the value 0.06487 obtained
based on the expensive ELF topology analysis. This observation well demonstrates the
reasonableness of employing ELF curve map between H and A to estimate the ELF(DH-A).
As regards ELF(C-V), its definition is fairly ambiguous. Since in the original paper of CVB
index the authors did not explicitly and clearly explain how this quantity should be evaluated,
different papers often employ different rule to calculate it, leading to serious confusion in existing
literatures. For example, in the CVB original paper, namely Theor. Chem. Acc., 104, 13 (2000), it
seems that the ELF(C-V) was determined as maximal value at all ELF minima dissecting core and
valence shell on the ELF curve between D and A atoms. However, in the subsequent paper Struct.
Chem., 16, 203 (2005) written by the same author, I found the ELF(C-V) is seemingly calculated as
the ELF value at one of exactly located ELF bifurcation points connecting core and valence basin
of donor atom (while acceptor atom is seemingly ignored).
In my viewpoint, the best definition of ELF(C-V) should be the ELF value at the minimum
dissecting core and valence shell of donor atom on the ELF curve between D and H. Again taking
the HFHF (H4-F3H1-F2) dimer as example, the ELF curve plotted between F2 and H1 is:
292
3 Functions
Namely ELF(C-V) = 0.0936. Hence, the CVB index for the HFHF system should be 0.0936 −
0.0648 = 0.0288. This value is very different to the counterpart (-0.006) in Table 2 of Theor. Chem.
Acc., 104, 13 (2000), because the calculation levels are different, the ways of obtaining ELF(C-V)
are different, and the sign of the data in this paper was erroneously reversed.
(3) Calculating CVB index in full automatic way

In order to simplifying the calculation of CVB index in above mentioned way as much as
possible, Multiwfn provides a function used to calculate this index in full automatic way. Still taking
the HFHF dimer as example, boot up Multiwfn and input below commands
examples\HF_HF.wfn
200 // Other function, part 2
1 // Calculate CVB index and related quantities
2,1,3 // Index of donor atom, hydrogen and acceptor atom of the H-bond, respectively
The result is
Core-valence bifurcation value at donor, ELF(C-V,D): 0.0936
Distance between corresponding minimum and the hydrogen: 0.743 Angstrom
Core-valence bifurcation value at acceptor, ELF(C-V,A): 0.1408
Bifurcation value at H-bond, ELF(DH-A): 0.0648
The CVB index, namely ELF(C-V,D) - ELF(DH-A): 0.028768
The result is completely identical to that we calculated manually. The outputted ELF(CV,A) is
useless in current context, but some users may be interested in it.
In order to make you better understand how the CVB index is automatically calculated in
Multiwfn, here I explain the implementation detail. After the user inputted index of D, H and A
atoms, the ELF curves corresponding to D-H and H-A are calculated in turn, and then the
ELF(CV,D), ELF(DH-A) and ELF(CV-A) are automatically identified from the curve data, as
illustrated below
293
3 Functions
(4) Special case: Calculating CVB index for some very strong H-bonds
In principle, the CVB index calculation protocol described above works for most kinds of
systems that have typical H-bond, both intermolecular and intramolecular H-bonds can be analyzed
in the same way. However, for some very strong H-bonds, whose hydrogen is lying at midpoint
between two heavy atoms, such as H2OH+OH2, this protocol is no longer valid because its ELF
curve does not show typical feature, as shown below (since the O-H-O angle in this system is close
to 180, only one plot is needed):
It can be seen that the V(D,H) has bifurcated as V(O) and V(H). In this case you should evaluate
CVB index manually by plotting ELF curve maps, and the ELF(DH-A) in the standard CVB index
expression should be replaced with the ELF value at the local minimum between the V(H) and V(O)
in the curve map.
Below is a more complicated case, F-HO-H, you also need to manually evaluate the CVB
index by plotting ELF curve map. The ELF curve map shown below was plotted between the F and
O (the FHO is almost linear, therefore only one plot is needed), as can be seen the ELF is
unsymmetric with respect to the central hydrogen:
294
3 Functions
This system can be regarded as having two H-bonds, the H-bond binding energy of O-H...F and
O...H-F must be very different. For the former, CVB index = ELF(B) - ELF(C), while for the latter,
CVB index = ELF(D) - ELF(A). This is because when discussing the system for example as O-H...F,
the O and F behave as donor and acceptor atoms, respectively. Therefore the minimum of point C
between H and F should be regarded as the ELF(DH-A), while the minimum of point B should be
viewed as ELF(C-V,D).
(5) Special case: H-bond acceptor is not a single atom

Acceptor of some H-bonds is not a single atom. For example, the acceptor of HFethylene is
the  region of ethylene. This kind of H-bond is known as -hydrogen bond. In this case, you also
have to manually calculate the CVB index.
The wavefunction file of the HFethylene has been provided as examples\C2H4_HF.wfn, its
geometry is shown below.
For this system, you can obtain ELF(C-V,D) by plotting ELF curve map between F7 and H8
and read the ELF value at minimum, the value will found to be 0.0944.
ELF(DH-A) of this system can be obtained via ELF topology analysis. To do this, we input
below commands in Multiwfn:
2 // Topology analysis
-11 // Select the real space function to be analyzed
9 // ELF
6 // Search critical points by randomly distribute initial guesses within a sphere
4 // Set the sphere center as geometry center of three atoms
1,4,8 // Center of C1, C4 and H8 will be set as the sphere center
0 // Start searching (the sphere radius, the number of guessing points can be set by
295
3 Functions
corresponding options in the interface)

-9 // Return
0 // Visualize topology analysis result
Now you can see below graph. Clearly, the critical point 5 corresponds to the bifurcation point
between V(D,H) and the basin of  electron.
Close the GUI, select option 7, and then input 5 to check properties of the critical point 5, you will
find its ELF value is 0.1241, which is just the ELF(DH-A) of this H-bond. Hence, the CVB index
of this system is 0.0944 - 0.1241 = -0.0297.
In fact, since this system has high symmetry, you can also obtain ELF value of the critical point 5 by simply
plotting ELF curve map between H8 and midpoint of C1-C4.
3.200.2 Calculate atomic and bond dipole moments in Hilbert space
This function is used to calculate atomic and bond dipole moments directly based on basis
functions (viz. in Hilbert space). You can also consult Section 12.3.2 of the book Ideas of Quantum
Chemistry (L. Piela, 2007).
Theory
In the formalism of basis functions, the system dipole moment can be expressed as follows
μ = μ nuc + μ ele =  Z AR A −  Pi , j   i | r |  j 
A i j
where Z and R are charge and coordinate of nuclei. P is density matrix,   i | r |  j  is dipole
moment integral between basis function i and j.
The system dipole moment can be decomposed as the sum of single-atom terms and atom pair
terms
μ =  μ tot
A +  μ AB =  (μ A +μ A + μ A ) +  (μ AB + μ AB )
tot nuc pop dip pop dip
A A B A A A B A
The expression and physical meaning of the five terms are
296
3 Functions
μ nuc
A : Dipole moment due to nuclear charge
A = Z AR A
μ nuc
μ pop
A : Dipole moment due to the electron population number localized on single atom (Notice
that this is different to the electron population number calculated by Mulliken or similar methods,
because the overlap population numbers have not been absorbed into respective atoms)
μ pop
A = − p loc
A RA A =  Pi , j   i |  j 
p loc
iA jA
μ dip
A : Atomic dipole moment, which reflects the electron dipole moment around an atom. rA
is the coordinate variable with respect to nucleus A
A = −  Pi , j   i | rA |  j 
μ dip where rA = r − R A
iA jA
= − Pi , j   i | r |  j  − μ pop
A
iA jA
μ pop
AB : Dipole moment due to the overlap population between atom A and B
AB = − p AB R AB
μ pop p AB = 2 Pi , j   i |  j  R AB = (R A + R B ) / 2
iA jB
μ dip
AB : Bond dipole moment, which somewhat reflects the electron dipole moment around
geometry center of corresponding two atoms. Of course, if A and B are not close to each other, then
this term will be very small, and thus inappropriate to be called as bond dipole moment.
AB = −2 Pi , j   i | rAB |  j 
μ dip where rAB = r − R AB
iA jB
= −2 Pi , j   i | r |  j  − μ pop
AB
iA jB
By means of Mulliken-type partition, the bond dipole moments can be incorporated into atomic
dipole moments, so that the system dipole moment can be written as the sum of single center terms
μ =  (μ nuc pop + μAdip )
A +μ A
A
where μA
pop
is the dipole moment due to the Mulliken population number of atom A
μApop = − p AMul R A p AMul =  Pi , j   i |  j 

B iB jB
and μA
dip
is the atomic overall dipole moment of atom A
μAdip = − Pi , j   i | rA |  j  = − Pi , j   i | r |  j  − μ Mul

A
B iB jB B iB jB
Note that the B index in above formulae runs over all atoms.
Usage
297
3 Functions
The input file must contain basis function information (e.g. .mwfn, .fch, .molden and .gms).
After you enter present function, you can choose option 1 to output information of a specific
A ; atomic dipole moment, μ A ;

p loc dip
atom, including: Atomic local population number,
contribution to system dipole moment due to nuclear charge, μ A ; contribution to system dipole
nuc
moment due to electron, μ A + μ A ; contribution to system dipole moment, μ A + μ A + μ A .

dip pop nuc dip pop
You can also choose option 2 to output information between specific atom pair, including: bond
population number, μ AB ; bond dipole moment, μ AB ; contribution to system dipole moment,

pop dip
μ pop
AB + μ AB .
dip
If choose 3, atomic overall dipole moment and related information of selected atoms will be
Mul
outputted, including: Atomic Mulliken population number, p A ; atomic overall dipole moment,
μAdip ; contribution to system dipole moment due to nuclear charge, μ nuc

A ; contribution to system
dipole moment due to electron, μA + μA ; contribution to system dipole moment, μ A + μA +
dip pop nuc dip
μApop .
If you choose option 10, then X/Y/Z components of electron dipole moment matrix will be
outputted to dipmatx.txt, dipmaty.txt and dipmatz.txt in current folder, respectively. For example, the
(i, j) element of Z component of electron dipole moment matrix corresponds to
−  Pi , j   i | z |  j 
i j
3.200.3 Generate cube file for multiple orbital wavefunctions
By this function, grid data of multiple orbital wavefunctions can be calculated and then
exported to a single cube file or separate cube files at the same time.
After you entered this function, you need to first select the orbitals you are interested in (e.g.
3,5,9-17), then define grid setting, then choose the scheme to export the grid data. If you select
scheme 1, then grid data of all orbitals you selected will be outputted to orbital.cub in current folder.
Lots of visualization programs, including VMD and Multiwfn, support the cube file containing
multiple sets of grid data. If you select scheme 2, then the grid data will be exported as separate files,
for example orb000003.cub, orb000005.cub, orb000009.cub, etc. The number in the filename
corresponds to orbital index.
For restricted and unrestricted single-determinant wavefunctions, in this function you can
select orbital based on HOMO and LUMO. For example, h-3 means HOMO-3, l+2 corresponds to
LUMO+2. See prompt on screen for more examples.
298
3 Functions
3.200.4 Generate iso-chemical shielding surfaces (ICSS) and related

quantities
Theory
Nuclear independent chemical shielding (NICS) is commonly studied at some special points
(e.g. ring center), and in some papers NICS is investigated by scanning its value in a line (1D) or in
a plane (2D). The so-called iso-chemical shielding surface (ICSS) actually is the isosurface of
negative of NICS, which clearly exhibits the distribution of NICS in 3D space, and thus presents a
very intuitive picture on aromaticity.
Present function is used to generate grid data and visualize isotropic ICSS, anisotropic ICSS,
ICSSXX, ICSSYY and ICSSZZ, they essentially correspond to the isosurface of negative of NICS,
NICSani, NICSXX, NICSYY and NICSZZ, respectively. By the way, at a given point, NICSani is defined
as 3 - (1 + 2)/2, where  denotes the eigenvalue of magnetic shielding tensor ranked from small
to large (viz. 3 is the largest one).
The original paper of ICSS is J. Chem. Soc. Perkin Trans. 2, 2001, 1893. While ICSSani,
ICSSXX, ICSSYY and ICSSZZ were proposed by me. I believe for planar systems, the component
form of ICSS must be more meaningful and useful than ICSS, just like NICSZZ has conspicuous
advantage over NICS. ICSSani is useful to reveal the anisotropic character of NICS in different
regions.
If ICSSZZ is utilized in your work, please cite Carbon, 165, 468 (2020), which is one of my
works using ICSSZZ.
Usage
Multiwfn itself is incapable of calculating magnetic shielding tensor and thus requires Gaussian
to do that. The general steps of performing ICSS analysis is shown below
(1) Prepare a Gaussian input file for the system under study, %chk have to be explicitly
specified. The geometry should have been optimized. The keywords in this file will be used for
preparing Gaussian input file of NMR task. For example, see examples\ICSS\anthracene.gjf.
(2) After booting up Multiwfn, load the .gjf file, then enter subfunction 4 of main function 200.
(3) Set up grid by following the prompt. Beware that even using medium quality grid may be
fairly time-consuming. Hence low quality grid is in general recommended for medium-size system.
(4) Input n, namely do not skip steps 5.
(5) Many input files of Gaussian NMR task are generated in current folder, they are named as
NICS0001.gjf, NICS0002.gjf... It is recommend to manually check one of them to ensure the format
and keywords are correct.
In these files, each Bq atom corresponds to a grid point. In the NMR task Gaussian will output magnetic
shielding tensor at each Bq along with that at each nuclei. By default 8000 atoms (the real ones + Bq) are presented
in each input file, but this can be altered via "NICSnptlim" parameter in settings.ini. The reason why separate files
rather than a single file are generated is because Gaussian cannot run properly if the number of Bq atoms is too large
due to over-consume of memory, also there is upper limit on the total number of atoms in each Gaussian run. You
can try to set "NICSnptlim" to a larger value if you have large physical memory, this may reduce overall cost of
ICSS analysis (however "NICSnptlim" should also not be too large, for example, the total computational cost of
"NICSnptlim=10000" is even higher than "NICSnptlim=1000"!).
299
3 Functions
Note that for G09 D.01 and E.01, due to a bug in memory allocation when using the default Harris initial guess,
you should always add "guess=huckel" to route section of template .gjf file, otherwise the NICSnptlim has to be set
to a very small value (e.g. 1000) to make Gaussian work; in this case the overall cost of ICSS calculations is often
quite high. For other Gaussian versions, this keyword should not be added. If error occurs in Link 401 module when
"guess=huckel" is specified, try to use "guess=core" instead. For G16, the guess keyword is not needed.
(6) Run all of the Gaussian input files generated at last step to yield output files. The
NICS0001.gjf must be run prior to any other ones. It is best to keep Multiwfn running (If you have
terminated it, reboot Multiwfn and repeat steps 2 and 3 with exactly the same setting and input y at
step 4).
Hint: You can make use of the script examples\runall.sh (for Linux) or examples\runall.bat (for Windows),
which invokes Gaussian to run all .gjf files in current folder to yield output files with the same name but with .out
suffix.
(7) Input the path of the folder containing Gaussian output files yielded in the last step. Then
Multiwfn will load the magnetic shielding tensors from the NICS0001.out, NICS0002.out ... in this
folder.
(8) Select the property you are interested in.
(9) Visualize isosurface or export the grid data to cube file by corresponding option. For
example, in step 8 you selected "ZZ component", then the isosurface and the grid data will
correspond to ICSSZZ. You can also select "-1 Load another ICSS form" to study other forms.
Notice that if this is not the first time you analyze your system and you already have Gaussian
output files of NMR task of present system in hand, you can start from step 2 and input y in step 4
to bypass steps 5 and 6. In this case, the grid setting selected in step 3 must exactly accord with the
that originally used in generating the Gaussian output files of NMR task.
3.200.5 Plot radial distribution function for a real space function
This function is used to plot radial distribution function (RDF) for a real space function
RDF ( r ) =  f ( r, )r 2 d 
where r is radial distance from sphere center, and  denotes angular coordinate in a sphere layer.
The integration curve of RDF can also be plotted
r' r'
I ( r ') =  RDF ( r ) d r =   f (r, )r
2
d d r
rlow rlow
Clearly, if rlow is set to 0 (viz. sphere center), then I() will be the integral of f over the whole space.
In present function, one can choose the real space function to be studied, set the position of
sphere center, set the lower and upper limit to be calculated and plotted, set the number of points in
radial and angular parts. The larger the number of points, the more accurate the integration curve.
After the parameters have been properly set, selecting option 0 to start the calculation, then you
will see a new menu, in which you can plot RDF and its integration curves, save the graph or export
the corresponding original data. In this menu you can also find an option used to export spherically
averaged function (f sph), it correlates with RDF via below relationshiop
RDF ( r )
f sph ( r ) =
4 r 2
300
3 Functions
3.200.6 Analyze correspondence between orbitals in two wavefunctions
Theory
This function is primarily used to analyze correspondence between the orbitals in two
wavefunctions. The two sets of orbitals can be produced under different basis sets, by different
theoretical methods, at different external environments, in different electronic states, or at slightly
different geometries. The two sets of orbitals can also be different types, for example the first set of
orbitals are canical MOs produced by Hartree-Fock calculation, while the second set of orbitals are
natural orbitals produced by post-HF calculation.
The orbitals {i} in present wavefunction (the wavefunction loaded when Multiwfn boots up)
can be represented as linear combination of the orbitals {j} in another wavefunction (the
wavefunction you specified after entering present module), i.e.
i =  Ci , j j where Ci , j = i j   i (r) j (r)d r

j
Once we have the overlap integral, we immediately know how j is associated with i. The
contribution from orbital j to orbital i is simply the square of overlap integral, namely <i|j>2100%.
The present function is able to compute the C matrix as well as the contributions, so that you can
easily make clear the relationship between the two sets of orbitals.
Usage
After you enter this function, first you need to input the orbital range to be considered for
present wavefunction (istart1~iend1), then input the path of the second wavefunction and the orbital
range to be considered (istart2~iend2). After that the overlap matrix between istart1~iend1 and
istart2~iend2 will be calculated by Becke's multi-center numerical integration scheme. Then you
will see the five largest contributions from istart2~iend2 to each orbital in istart1~iend1.
If you want to obtain all coefficients (as well as the corresponding contributions) of
istart2~iend2 in a specific orbital among istart1~iend1, you can then directly input the index of the
orbital.
If an orbital (i) in present wavefunction can be exactly expanded as linear combination of
istart2~iend2, then the normalization condition must be satisifed:
iend 2
i  100% = 100%
2
j
j =istart 2
From the Multiwfn output you can find the maximum deviation to normalization condition. If the
value is zero, that means all orbitals in istart1~iend1 can be exactly represented by the orbitals in
istart2~iend2.
Note that the atomic coordinate of present wavefunction and that of the second wavefunction
are not necessarily identical, the two wavefunctions can even correspond to different molecules.
However if the difference of the distribution scope of the atomic coordinates in the two
wavefunctions is large, the integration accuracy must be low and the result is not reliable.
301
3 Functions
Commonly the default integration grid is fine enough, i.e. 30 radial points, 302 angular points
with "radcut=15". If you wish to improve the accuracy, you should set "iautointgrid" in settings.ini
to 0, then you can define "radpot", "sphpot" and "radcut" in settings.ini; increasing their values will
result in better integration accuracy.
The computational cost of this function directly depends on the number of orbitals in
consideration; so if your system contains very large number of orbitals, do not choose all orbitals at
once.
Special usage: Evaluating overlap integrals and superpositions between two sets of
orbitals
Present function can also be used to evaluate overlap between orbitals, the orbitals may come
from the same wavefunction, or come from two different wavefunctions.
After entering the interface of present function, if you want to obtain all overlap integrals (i.e.
all <i|j>) between the above mentioned orbitals istart1~iend1 in the first wavefunction and orbitals
istart2~iend2 in the second wavefunction, simply input -1, then these integrals will be outputted to
convmat.txt in current folder.
If what you need is not common overlap integral between orbital wavefunctions but overlap
integral between norm of orbital wavefunctions, which is useful for measuring orbital superposition
and expressed as ∫|𝜑𝑖 (𝐫)||𝜑𝑗 (𝐫)|d𝐫, you should input -2 in the interface of present function, then
all these integrals between the orbitals istart1~iend1 and istart2~iend2 will be outputted to
Snormmat.txt in current folder. Similarly, if what you need is ∫|𝜑𝑖 (𝐫)|2|𝜑𝑗 (𝐫)|2 d𝐫, you should
input -3 in the interface, then the result will be outputted to Snorm2mat.txt.
In fact, the present function can somewhat equivalently realize the functions introduced in
Sections 3.100.5, 3.100.11 and 3.100.15, but the output format and main purpose are different.
Some usage examples of this function are given in Section 4.200.6.
3.200.7 Parse output of (hyper)polarizability task of Gaussian and

evaluate relevant quantities
The output of (hyper)polarizability task of Gaussian (polar keyword) is difficult to understand,

at least for beginners. This function is used to parse these outputs and then print them in a more
readable format, and at the same time some quantities relating to (hyper)polarizability are outputted.
Currently this function is formally compatible with Gaussian 09 and 16.
Basic concepts and theory backgrounds

Energy of a system can be written as Taylor expansion with respect to uniform external electric
field F
E 1 2E 1 3E 1 4E

E ( F) = E ( 0) + F+ F2 + F3 + F4 + 
F F=0 2 F 2 F =0
6 F 3 F =0
24 F 4 F =0
1 2 1 3 1 1 1
 E ( 0) − μ 0 F − αF − βF − γF4 − δF 5 − εF 6 
2 6 24 120 720
302
3 Functions
E 2E 3E 4E

μ0 = − α=− β=− γ=−
F F=0 F 2 F =0
F 3 F =0
F 4 F =0
where 0 is permanent dipole moment, which is a vector;  is polarizability, which is a matrix

(second rank tensor);  is first hyperpolarizability, which is a third rank tensor and known as second-
order nonlinear optical response (NLO) coefficient;  is second hyperpolarizability, which is a fourth
rank tensor and known as third-order NLO coefficient. The higher terms such as  and  are very
unimportant and thus rarely discussed. The (hyper)polarizability tensors are directly correlated to
the frequency of external field F. If F has zero-frequency (static electric field), then the
(hyper)polarizabilities are known as static or frequency-independent ones. The dynamic or
frequency-dependent (hyper)polarizabilities correspond to those at external electromagnetic fields
with non-zero frequency.
• Polarizability ()
Dipole moment of a system in uniform electric field can be written as
E
μ=− = μ 0 + αF
 + (
1 / 2)βF 2 + (1 / 6) γF3 + 
F μ1
   
μ 2 μ 3
The linear response of dipole moment with respect to F, namely the 1 term, can be explicitly
written as below
  x   xx  xy  xz   Fx 
 
μ1 = α  F    y  =  yx  yy  yz   Fy 
   
  z   zx  zy  zz   Fz 
The polarizability  is a symmetric matrix rather than a scalar, implying the difference of
polarizability in different directions. In order to facilitate comparison of overall polarizability
between various systems, it is convenient to define the isotropic average polarizability
  = Tr(α) / 3 = ( xx +  yy +  zz ) / 3
Anisotropy of polarizability can be defined in various ways:

Definition 1: see Chem. Phys., 410, 90 (2013) for example
 = [( xx −  yy ) 2 + ( xx −  zz ) 2 + ( yy −  zz ) 2 + 6( xy2 +  xz2 +  yz2 )] / 2

Definition 2: This definition is the most commonly use one, see J. Chem. Phys., 98, 3022 (1993)
for example
 = [( xx −  yy ) 2 + ( xx −  zz ) 2 + ( yy −  zz ) 2 ] / 2
Definition 3: {} stand for eigenvalues of  ranking from small to large
 =  3 − (1 +  2 ) / 2
• First hyperpolarizability ()

First hyperpolarizability  is a third rank tensor that can be described by a 333 matrix.
Gaussian is capable of calculating both static and dynamic . For the latter case, dc-Pockels form
(-;,0) and SHG form (-2;,) can be evaluated.
It is worth to note that for i,j,k(-;,0), the i and j index can be freely exchanged, while for
303
3 Functions
i,j,k(-2;,), the j and k index can be freely exchanged, therefore the 3*3*3=27 components can
be reduced to 18 unique ones. For static case, viz. i,j,k(0;0,0), since all of the three indices are
exchangeable (known as Kleinman's symmetry, e.g. xyy=yxy=yyx …), only 10 components are
unique. The Kleinman's symmetry is only approximately applied to dynamic  at low-frequency
external field.
The  value in one of the three Cartesian axes can be calculated by the general equation
 i = (1 / 3) (  ijj +  jji +  jij ) i, j = {x, y, z}
j
The magnitude of  is defined as
 tot =  x2 +  y2 +  z2
prj (also known as vec) is the projection of  on dipole moment vector , it can be sampled
experimentally (electric field induced second harmonic generation (EFISHG) experiment). || is the
 component in the direction of 
i  i
 prj =  || = (3 / 5) prj
i ||
Some people prefer to discuss the perpendicular and parallel components of  with respect to
Z axis, they are defined respectively as
 ⊥( Z ) = (1 / 5) (2 zjj − 3 jzj + 2 jjz ) ||( Z ) = (3 / 5)  z
j
For static case, we can explicitly write out  in x, y and z directions as

 x =  xxx +  xyy +  xzz
 y =  yxx +  yyy +  yzz
 z =  zxx +  zyy +  zzz
and  ⊥( Z ) = (1 / 5) z .
• Second hyperpolarizability ()

Second hyperpolarizability  is a fourth rank tensor of 3333 form. Gaussian can calculate
its static limit form (0;0,0,0); while for dynamic case, Gaussian is capable of calculating its EOKO
(Electro-optic Kerr effect) form (-;,0,0) and SHG form (-2;,,0).
The i components of  is defined as
 i = (1 / 15) ( ijji +  ijij +  iijj ) i, j = {x, y, z}
j
The total magnitude of  is measured as  tot =  x2 +  y2 +  z2 .
There are two definitions of average of , as shown below. Definition 1 is more common, and
it is equivalent to ||
def 1:   =  x +  y +  z
def 2 :   = (1 / 5)( xxxx +  yyyy +  zzzz +  xxyy +  xxzz +  yyzz +  yyxx +  zzxx +  zzyy )
The _|_ is defined as
304
3 Functions
 ⊥ = (1 / 15) (2 ijji −  iijj ) i, j = {x, y, z}

i j
Most of above mentioned equations about  can be found in Chapter 5 of Reviews in

Computational Chemistry, Vol. 12 (1998).
• Hyper-Rayleigh scattering (HRS) and depolarization ratio (DR)

The Hyper-Rayleigh Scattering (HRS) technique was developed as an alternative method to
EFISHG for the measurement of molecular hyperpolarizabilities. HRS can be used to directly
measure  of all molecules, including nonpolar molecules, which cannot be studied by EFISHG.
See Acc. Chem. Res., 31, 675 (1998) for introduction.
According to intensity of incident light at given frequency () and that of scattered light with
doubled frequency (2) detected at 90 angle, the HRS could be determined, which correlates
components of frequency-dependent  tensor as follows. See Phys. Chem. Chem. Phys., 10, 6223
(2008) for more details.
 HRS ( −2; , ) =  ZZZ

2
+  XZZ
2
where (index  cycles {x,y,z} in turn, similar for  and )

1 4 2 4
 ZZZ
2
= 
7 

2
+  
35  
2
+    +   
35   35  
4 1 4 1
+ 
35  
  +  
35  
2
+ 
105   
  +   
105   
4 2 4
+ 
105   
  + 
105   

2
+   
105   
1 4 2 8
 XZZ
2
= 
35 

2
+ 
105  
  −    +
35  
 
105  
2
3 2 1 2
+ 
35  
 −    +
35  

35   
  −   
105   
2 2 2
− 
105   
  + 
35   

2
−   
105   
the  means  and meantime . the brackets indicate an averaging over all possible
molecular orientations. If assume that Kleinman's symmetry is approximately applicable, the above
equations could be reduced to a more simple form, see Eq. 6 of J. Chem. Phys., 136, 024506 (2012).
The associated depolarization ratio (DR) is defined as
 ZZZ
2
DR =
 XZZ
2
Molecules with Td point group have DR exactly equals to 1.5. If the half of wavelength of incident
light is close to absorption band of the current system calculated at same level using TDDFT, the
printed DR may be lower than 1.5 due to SHG resonance.
There are some relevant quantities about HRS could be studied, as shown below, see also J.
Chem. Phys., 136, 024506 (2012) for more information. Note that for consistency, the ZXX in the
equations of this paper have been replaced with XZZ, they are numerically identical in the present
context.
305
3 Functions
2 〉 2 〉
The 〈𝛽𝑍𝑍𝑍 and 〈𝛽𝑋𝑍𝑍 can be regarded as contributed by two components, dipolar (J=1) and
octupolar (J=3):
9 6
 ZZZ
2
= |  J =1 |2 + |  J =3 |2
45 105
1 4
 XZZ
2
= |  J =1 |2 + |  J =3 |2
45 105
clearly the two components can be evaluated as follows
|  J =1 |= 6  ZZZ
2
− 9  XZZ
2
|  J =3 |= 1
2 ( −7  ZZZ
2
+ 63  XZZ
2
)
The nonlinear anisotropy parameter is evaluated as  = |J=3| / |J=1|.
The dipolar relative contribution to , namely (J=1), and octupolar relative contribution to ,
namely (J=3), are defined as below. Clearly (J=1)+(J=3)=1
(  J =1 ) = 1 / (1 +  )
(  J =3 ) =  / (1 +  )
For small molecules, the one with larger dipole moment tends to have larger (J=1), higher
DR and lower , while the one with smaller dipole moment tends to have larger (J=3), lower DR
and higher .
Assuming a general elliptically polarized incident light propagating along the X direction, with
a state of polarization characterized by two angles (, δ), the intensity of the harmonic light scattered
at 90° along the Y direction and vertically (V) polarized along the Z-axis are given by Bersohn’s
expression (the phase retardation  is assumed to be /2)
I 2V   XZZ
2
cos4  +  ZZZ
2
sin4  + sin2  cos2  (  ZXZ +  ZZX )2 − 2 ZZZ  XZZ
where (  ZXZ +  ZZX )2 − 2 ZZZ  XZZ = 7  XZZ

2
−  ZZZ
2
,  is polarization angle of the
incident light beam.

According to theoretically calculated SHG form of  tensor, all above mentioned quantities
2𝜔
could be readily predicted. The variation of 𝐼V with respect to  could be scanned and plotted as
curve map.
Input file and usage

In this function, Multiwfn outputs all components of dipole moment, polarizability and 1st/2nd
hyperpolarizability (if available) with explicit labels, as well as all of their relevant quantities
introduced above, such as isotropic polarizability, polarizability anisotropy, hyperpolarizability in
three axes (x, y, z), magnitude of hyperpolarizability (tot) and so on. The  and  components
printed by Multiwfn correspond to standard orientation, while the printed  components correspond
to input orientation.
The polar keyword in Gaussian is specific for calculating ,  and  based on analytic
derivatives (by means of coupled-perturbed SCF equation) or numerical derivatives (by means of
finite field treatment). Notice that in the Gaussian input file you must specify #P, otherwise Multiwfn
306
3 Functions
cannot properly parse relevant information.
After you enter present function of Multiwfn, you should select actual case of your Gaussian
(hyper)polarizability calculation, so that Multiwfn can successfully parse the outputted information
and show valuable data for you. As can be seen in the menu, there are seven options corresponding
to different situations:
(1) polar keyword + methods supporting analytic 3-order derivatives (HF/DFT/Semi-empirical
methods)
(2) polar keyword + methods supporting analytic 2-order derivatives (e.g. MP2)
(3) polar=Cubic keyword + methods supporting analytic 2-order derivatives
(4) polar keyword + methods supporting analytic 1-order derivatives (CISD, QCISD, CCSD,
MP3, MP4(SDQ), etc.)
(5) polar=DoubleNumer (equivalent to Polar=EnOnly) keyword + methods supporting
analytic 1-order derivatives
(6) polar keyword + methods only supporting energy calculation (CCSD(T), QCISD(T),
MP4(SDTQ), MP5, etc.)
(7) polar=gamma keyword + methods supporting analytic 3-order derivatives (HF/DFT/Semi-
empirical methods)
All options print polarizability and relevant data, only options (1), (3) and (5) also print first
hyperpolarizability, only (7) also prints second hyperpolarizability.
For case (1), if CPHF=RdFreq is specified along with polar or you used polar=DCSHG, and
meantime the external field frequencies (e.g. 0.05 0.07 0.1 or 532 nm 680 nm) are provided after
molecular geometry with a blank line in front of it, Gaussian will calculate and output frequency-
dependent (hyper)polarizabilities along with static (hyper)polarizability. The CPHF=RdFreq polar
case only evaluates (-;,0) values, while the polar=DCSHG case evaluates both (-;,0) and
(-2;,) values. For the case of (7), Gaussian calculates (0;0,0,0), (-;,0,0) and (-2;,,0),
and you must provide external field frequencies at the end of input file.
For cases (1) and (7), by default Multiwfn only parses static (hyper)polarizability. If you wish
to parse the frequency-dependent ones instead of the static one, before selecting option 1 or 7 to
start parsing, you should select option “-1 Toggle loading frequency-dependent result for options 1
and 7” first. Then after starting parsing, user can choose the result at which frequency will be parsed.
Note that in the case (1) if you choose to parse (-2;,), you must employ polar=DCSHG
keyword in Gaussian input file.
The quantities related to hyper-Rayleigh scattering (HRS) experiment mentioned above are
also automatically printed when you request Multiwfn to parse frequency-dependent (-2;,)
2𝜔
based on output file of polar=DCSHG. After that, you can also let Multiwfn scan 𝐼V versus ,
then the generated HRS_angle.txt could be plotted using Origin and so on.
It is noteworthy that, it is well known that the sign of all hyperpolarizability components
outputted by Gaussian are wrong and should be multiplied by -1, Multiwfn automatically accounts
for this problem.
Before carrying out parsing, via option -3 of interface of present function, you can choose the
unit in the output. Atomic unit, SI unit and esu unit can be chosen. The conversion factors are
SI esu
 1 a.u. 8.4783510-30 Cm 2.54175 10-18 esu
307
3 Functions
 1 a.u. 1.648810-41 C2m2J-1 1.481910-25 esu

 1 a.u. 3.2063610-53 C3m3J-2 8.6392210-33 esu
 1 a.u. 6.2353810-65 C4m4J-3 5.0367010-40 esu
The polarizability  is often expressed in terms of "polarizability volume" ('), which has
volume unit.  (1 a.u.)=' (0.14818470 Å3).
An example is given in Section 4.200.7. More discussion and examples about this function can
be found in my blog article "Using Multiwfn to analyze polarizability and hyperpolarizability
outputted by Gaussian" (http://sobereva.com/231, in Chinese)
3.200.8 Study (hyper)polarizability by sum-over-states (SOS) method

and two- or three-level model analyses
This function is used to calculate polarizability, first, second, and third hyperpolarizabilities
based on the well-known sum-over-states (SOS) method, as described below. In addition, the
popular two-level model analysis for the first hyperpolarizability as well as its extension (three-level
model) can also be realized in this module, as introduced in Section 3.200.8.2.
3.200.8.1 Calculation of (hyper)polarizability
A brief survey of the theories for evaluating (hyper)polarizability

Some basic concepts of (hyper)polarizability are introduced in Section 3.200.7. There are a few
different ways to calculate (hyper)polarizability, including derivative method, sum-over-states (SOS)
and response method
(1) Derivative method: This is the most straightforward and commonly used one. The
derivatives needed by static (hyper)polarizability can be evaluated analytically by means of coupled-
perturbed SCF (CPSCF) equation; specifically, CPHF for HF and CPKS for KS-DFT. These
derivatives can also be evaluated numerically by means of finite difference technique, which is also
known as finite field (FF) method. Evidently FF is much slower and not as accurate as CPSCF,
however it is still useful, because high-order of analytic derivatives, especially the ones at
sophisticated post-HF levels, are not widely supported by many quantum chemistry programs due
to the difficulties in coding. When all requested derivatives are available analytically, derivative
method will be very efficient. The frequency-dependent variant of CPSCF equation enables the
derivative method to evaluate dynamic (hyper)polarizability, but there is no way to evaluate
dynamic (hyper)polarizability in terms of FF treatment. The polar keyword in Gaussian, as
discussed carefully in Section 3.200.7, corresponds to this derivative method.
(2) SOS method: This method for evaluating static and dynamic (hyper)polarizability is
relatively inefficient, because in principle it involves a sum over all excited states (in practical
applications, taking 60~120 lowest states into account is often enough), while determination of a
large number of excited states is usually quite time comsuming in ab initio cases (e.g. CIS and
TDDFT), especially for large system (e.g. > 40 atoms). Due to the high computational cost, SOS is
generally not recommended for evaluation of (hyper)polarizability when derivative method can be
carried out analytically. The only advantages of SOS may be that the contribution from different
states can be separated and discussed respectively, and when transition dipole moments between
different excited states are available in hand, the (hyper)polarizability at different frequencies can
308
3 Functions
be evaluated rather rapidly. Worthnotingly, the SOS based on the cheap semi-empirical ZINDO
calculation (SOS/ZINDO) is very popular for evaluating (hyper)polarizability of large system.
(3) Response method: This method is specific for dynamic (hyper)polarizability and also
known as propagator method. TDHF and TDDFT are its two practical realizations. This method is
not prevalently supported by mainstream quantum chemistry codes.
Working equations of SOS method

The explicit SOS equations for evaluating polarizability and 1st/2nd/3rd hyperpolarizability
can be found in J. Chem. Phys., 99, 3738 (1993), the idea was originally proposed by Orr and Ward
in Mol. Phys., 20, 512 (1971).
The equations for polarizability  and first hyperpolarizability  are (all units are in a.u.)
 0Ai iB0 0Bi iA0  ˆ 0Ai iB0

 AB ( − ;  ) =   + = P[ A( − ), B( )]
i 0   i −   i +   i 0  i − 
0Ai ijB  Cj0

 ABC ( − ; 1 , 2 ) = Pˆ [ A( − ), B(1 ), C (2 )]
i 0 j 0 (  i −  )(  j − 2 )
where
ijA = i ˆ A j ijA = ijA − 00A  ij  =  i
i
A,B,C... denote one of directions {x,y,z};  is energy of external fields, =0 corresponds to static
electric field; i stands for excitation energy of state i with respect to ground state 0. P̂ is
permutation operator, for  and  evidently there are 2!=2 and 3!=6 permutations, respectively. ijA
is A component of transition dipole moment between state i and j; when i=j the term simply
corresponds to electric dipole moment of state i. ̂ is dipole moment operator, e.g. ̂ x  − x .
The SOS equation for second hyperpolarizability  is

 ABCD ( − ; 1 , 2 , 3 ) = Pˆ [ A( − ), B(1 ), C (2 ), D(3 )]( I −  II )
0Ai ijB  Cjk  kD0
 I = 
i 0 j 0 k 0 (  i −  )(  j − 2 − 3 )(  k − 3 )
0Ai iB0 0Cj  Dj0

 II = 
i 0 j 0 (  i −  )(  i − 1 )(  j − 3 )
The SOS equation for third hyperpolarizability  is
309
3 Functions
 ABCDE ( − ; 1 , 2 , 3 , 4 ) = Pˆ [ A( − ), B(1 ), C (2 ), D(3 ), E (4 )]( I −  II −  III )

0Ai ijB  Cjk  klD lE0
 = 
I
l , j ,k ,l (  i −  )(  j −  + 1 )(  k − 3 − 4 )(  l − 4 )
( 0 )
0Ai iB0 0Cj  Djk  kE0  1 1  1 1 

 = (1 / 2) 
II
 +  + 
i , j ,k (  j + 2 )(  k − 4 )   i −   i − 1   j − 3 − 4  k + 2 + 3 

( 0 )
0Ai iB0 0Cj  Djk  kE0  1 1 

 III
= (1 / 2)   + 
i , j ,k (  i −  )(  i − 1 )   (  j − 3 − 4 )(  k − 4 ) (  j + 2 )(  k + 2 + 3 ) 
( 0 )
Input file
Two kinds of input files could be used:
• Plain text file containing excitation energies and transition dipole moments for all involved
states. Polarizability, first, second and third hyperpolarizabilities can be calculated in this case.
Below format should be satisfied (assume a very simple case, only 2 excited states).
2 // The number of excited states
1 1.1 // Excited state 1, its index and excitation energy (eV)
2 3.2
0 0 0.845 0.2 0.4 // Electric dipole moment of ground in X,Y,Z (a.u.)
0 1 0.231 0.3 0.7 // Transition dipole moment between ground and excited state 1
0 2 0.112 0.564 0.21
1 1 0.021 0.465 0.0 // Electric dipole moment of excited state 1
1 2 0.001 0.3 0.11 // Transition dipole moment between excited states 1 and 2
2 2 0.432 0.14 0.42
You can directly utilize the function introduced in Section 3.21.5 to generate such a plain text
file based on output file of electron excitation task of Gaussian or other codes.
If merely polarizability is the quantity of interest, only the content before the line "1 1" is
needed to be provided, all other contents can be ommited; in this case, the number of excited states
should be written as a negative number (-2 in above case) to tell Multiwfn do not to load them.
• Gaussian output file of common CIS, TDHF, TDDFT or ZINDO task. Since Gaussian does
not output all transition dipole moments needed by SOS hyperpolarizability calculation, only
polarizability will be calculated by Multiwfn in this case. In order to obtain accurate polarizability,
the number of calculated states should be large enough. If nstates keyword is specified to a very
large value, e.g. 1000000, then all states will be calculated. #P is suggested to be used, since the
excitation energy will then be printed in a higher precision format.
Usage
After you entered this function you will see a menu, there are three kinds of functions:
• Options 1~4: Used to calculate , ,  and  at given frequencies, respectively. User needs to
input frequency of each external field. The inputted frequencies may be negative. For example, to
compute hyperpolarizability (-(0.25-0.32);0.25,-0.32), one should input 0.25,-0.32 in option 2. The
default unit is a.u., if you prefer to input the frequencies in nm, you should add corresponding suffix,
for example, 182.25,-142.385 nm.
310
3 Functions
Since calculation of  and especially  is often time-consuming, in these cases users will be
prompted to input the number of states in consideration, smaller number leads to lower cost, but too
small number may give rise to poor result.
• Options 5~7: Used to study the variation of ,  and  with respect to the number of states in
consideration. User needs to input frequency of each external field. For  and , the number of
states taken into account ranges from 1 to all states loaded, the stepsize is 1. While for , since the
computational cost may be quite high, users are allowed to customly defined the ending value and
stepsize. The result will be outputted to plain text file in current folder, the meaning of each column
is clearly indicated in command-line window.
• Options 15~17: Used to study the variation of the ,  and  with respect to frequency of
external fields. For , users need to input initial value, ending value and stepsize of external field
frequencies. For  and , users should write a plain text file, each row corresponds to a pair of
frequency (in a.u.) to be calculated. Multiwfn will prompt users to input the path of the file. Below
is an example file used to study how (-0;0,,-) varies as  goes from 0 to 0.2 a.u. with stepsize
of 0.02
0.0 0.0 0.0
0.0 0.02 -0.02
0.0 0.04 -0.04
...[ignored]
0.0 0.2 -0.2
Since the computational cost for evaluating  may be quite high, in this case users are allowed to set
the number of states in consideration. The result will be outputted to plain text files in current folder,
the meaning of each column is clearly indicated in command-line window.
• Option 19: This option is used to scan both 1 and 2 of (-(1+2);1,2). You only need
to input initial frequency, ending frequency and number of steps for 1 and 2. Then after a while,
 at different 1 and 2 frequencies will be outputted to plain text files in current folder, the meaning
of each column is clearly indicated in command-line window. Then you can use third-part software
to plot relif map of " vs. 1,2".
Multiwfn not only outputs the tensor of (hyper)polarizability, but also outputs many related
quantities, such as anisotropy, magnitude and the component along Z axis. The quantities involving
,  and  have been introduced in Section 3.200.7.
An example is given in Section 4.200.8.1. More discussion and examples about this function
can be found in my blog article "Using Multiwfn to calculate polarizability and hyperpolarizability
based on sum-over-states (SOS) method" (http://sobereva.com/232, in Chinese)
3.200.8.2 Two-level and three-level model analyses for hyperpolarizability
Theory
From the SOS expression of , it is clear that magnitude of  is completely determined by
character of excited states. Clearly it is a useful idea to interpret the nature of difference in  between
different systems from excited state point of view. Indeed, this analysis has been prevalently
employed in literatures, such as one of my works J. Comput. Chem., 38, 1574 (2017). Let us see
how to derive such an analysis model.
Recall the SOS formula for 
311
3 Functions
0Ai ijB  Cj0

 ABC ( − ; 1 , 2 ) = Pˆ [ A( − ), B(1 ), C (2 )]
i 0 j 0 (  i −  )(  j − 2 )
Assume that only the ZZZ component is of our interest and we only focus on static limit case
(=0), the equation simplifies to
0Zi ijZ  Zj0

 SOS
= 6
i  j
ZZZ
i 0 j 0
Given that ijA = ijA − 00A  ij , when i=j, this terms corresponds to variation of dipole moment
between excited state i and ground state, namely iiA = iiA − 00A = iZ ; while if ij , this term
ijA = ijA corresponds to transition dipole moment between excited state i and j.
𝐴 𝐴 SOS
With the fact that 𝜇𝑖𝑗 = 𝜇𝑗𝑖 , the 𝛽𝑍𝑍𝑍 shown above can be written as sum of contribution of
individual excited states and cross term contribution between various excited states:
( 0Zi )2 iZ 0i 0 j ij Z Z Z
 SOS
= 6 +  12
i2 i  j
ZZZ
i i j i
The two-level model is very popular, it assumes that the ZZZ is dominated by ground state and
only one excited state:
( 0Zi )2 iZ
 ZZZ
SOS
6
i2
The excited state i is usually referred to as crucial state and commonly corresponds to the lowest-
lying one with large oscillator strength (strictly speaking, in the present context, the crucial state
should refer to the lowest-lying one with large Z component of transition dipole moment, however,
the crucial state determined in this way is commonly identical to that determined according to
oscillator strength).
Often the two-level model is equivalently expressed in terms of oscillator strength:
 ZZZ
SOS
= 9iZ fi Z / i3
where fi = (2 / 3)i ( 0i )

Z Z 2
is Z component of oscillator strength. Furthermore, with assumption
that only Z component of transition dipole moment and variation of dipole moment are relatively
prominent, we have  SOS  i fi / i3 . Obviously, now one can easily analyze the source of
difference of  between various systems by comparing the ∆𝜇𝑖𝑍 , fi and i terms.
Occasionally, there is no well-defined crucial state. For example, both the 1st and 2nd excited
𝑍
states have large 𝜇0𝑖 , while their energy separation is marginal (nearly degenerate), in this case we
should not simply ignore either one, the two excited states should be simultaneously taken into
account, I define this model as three-level model:
( 0Zi )2 iZ 0Zi 0Zj ijZ ( 0Zj )2  Zj

 SOS
=6 + 12 +6
i2 i  j  2j
ZZZ
312
3 Functions
Usage
In the SOS module (subfunction 8 of main function 200), the suboption 20 is used to carry out
the two- and three-level model analyses, all involved terms in the models will be reported so that
you can easily compare them among different systems.
After entering this function, if you only input index of one excited state, then two-level model
analysis will be carried out, if indices of two excited states are inputted, then three-level model
analysis will be performed.
The component (X, Y or Z) of  to be analyzed is defined by users. For example, you already
know that the total  of your system is fully dominated by ZZZ and you want to explain why other
analogous systems have different magnitude of , you should choose Z direction to carry out the
analyze.
The input file of this function is completely identical to the one used for SOS calculation of
first hyperpolarizability, as described in last section.
An example is given in Section 4.200.8.2.
3.200.9 Calculate average bond length and average coordinate number
This function is used to calculate average bond length between two elements and average
coordinate number. This function is particularly useful for analyzing structure character of atom
clusters, for example the Ge12Au cluster shown below (the structure file is provided as
examples\Ge12Au.pdb). By using this function, we can immediately obtain the average Ge-Ge bond
length and average Au-Ge bond length, as well as average coordinate number of Ge due to Ge-Au
or Ge-Ge bonds, or of Au due to Ge-Au bonds. A nice application of this kind of analysis on Al
clusters can be found in J. Chem. Phys., 111, 1890 (1999).
The average bond length is defined as follows
1
R =  Rij
nb i  j
where Rij is the distance between atom i and j, only the terms smaller than or equal to a given distance
cutoff (e.g. 2.2 Å) will be regarded as bonds and thus be taken into the summation. nb is the total
number of bonds.
The average coordinate number is calculated as follows
313
3 Functions
1
CN =  Ni
n i
where Ni is the number of bonds surrounding the atom i, n is the total number of atoms.
After you entered this function, you need to input two elements, for example Ge,Au, and input
a distance cutoff, for example 3.2, then the Ge-Au contacts  3.2 Å will be regarded as Ge-Au bonds
and the average bond length will be calculated, the minimum and maximum bond lengths will also
be outputted. After that, if you select y, the average coordinate number of Ge due to Ge-Au bonds
will be shown.
3.200.10 Output various kinds of integral between orbitals
This function is used to calculate electric/magnetic dipole moment integral, velocity integral,
kinetic energy integral and overlap integral between orbitals, advanced users may recognize the
significance of these data. In the case of a range of orbitals, the results are exported to orbint.txt in
current folder, the first and second columns correspond to the index of the two orbitals; In the case
of a pair of orbitals, the result is directly printed on screen.
The electric dipole moment integral vector between two orbitals is defined as
μij = μ ji =  i | −r |  j 
The magnetic dipole moment integral vector between two orbitals is calculated as (more detail
can be found in Section 3.21.1.1. The negative sign is ignored)
Mij = i   i | r   |  j 
The velocity integral vector between two orbitals is evaluated as (the negative sign is ignored)
vij = i  i |  |  j 
Worthnotingly, due to the Hermitian of the operators, we have

M ii = 0 M ij = Mji = −M ji
v ii = 0 v ij = v ji = − v ji
Note that the imaginary sign is not explicitly shown in the output.
The kinetic energy and overlap integrals between two orbitals are respectively evaluated as
Kij = −(1 / 2)  i | 2 |  j  Sij = i |  j 
If you need to calculate Coulomb or exchange integral between two orbitals, you should use
the function described in Section 3.200.17.
314
3 Functions
3.200.11 Calculate center, the first and second moments of a real space
function
This function is used to calculate center, integral value, the first and second moments of a
specific real space function for present system.
Theory
The center of a real space function f is defined as
rc =
 r  f (r)dr
 f (r)dr
where the denominator is the integral of the function over the whole space.
The first moment is a vector and is evaluated as
x   x
μ =  y =   y  f (r )dr
 
   
  z   z 
The second moment is a matrix and defined as
  xx  xy  xz  x2 xy xz 
   
Θ =  yx  yy  yz  =   yx y2 yz  f (r )dr
  zx  zy  zz   zx zy z 2 
 
where x, y, z are the coordinate components relative to rc.
If its eigenvalues {} are sorted from low to high, then the anisotropy of  can be calculated
1 +  2 +  3
as  3 − (1 +  2 ) / 2 . The radius of gyration can be calculated as .
 f (r )dr
Usage
This function employs Becke's multicenter integration method for evaluating above mentioned
quantities. The accuracy is determined by radial points and angular points, which can be set by
"radpot" and "sphpot" in settings.ini, respectively.
The option 1 calculates and outputs the integral, the first and second moments (relative to rc),
and the corresponding anisotropy and radius of gyration of the real space function selected by option
3. The center (rc) should be properly first set by option 4. Of course, commonly rc is initially
unknown, the default value is (0,0,0), you may use option 2 to evaluate the actual center of a real
space function and take it as the center for the succeeding calculations triggered by option 1.
For example, to evaluate the first and second moments of spin density (relative to the center of
spin density), after you entered this function you should input
3 // Select real space function
5 // Spin density
2 // Calculate center of spin density
y // Take the resultant center as the one for evaluating the first and second moments
1 // Evaluate the first and second moments
When using option 1, if the function to be studied is chosen as electron density, then the nuclear
315
3 Functions
quadrupole and molecular quadrupole moment tensors will also be outputted, the latter can be
straightforwardly obtained by subtracting the former by the second moment of electron density.
3.200.12 Calculate energy index (EI) or bond polarity index (BPI)
This function is used to calculate energy index (EI) or bond polarity index (BPI), which were
defined in J. Phys. Chem., 94, 5602 (1990).
The EI for atom A in a molecule is defined as follows
val
   i i i, A
EI A = i
val
 
i
i i, A
where i,A denotes composition of atom A in MO i. i and i are occupation and energy of MO i,
respectively. The summation runs over valence MOs. In fact, the denominator is simply the number
of valence electrons of atom A, and the numerator corresponds to total energy of its valence electrons.
Therefore, EIA can be regarded as average energy per valence electron of atom A. In the original
paper of EI, Mulliken method was used to compute the atomic contribution to MOs, thus this method
is also employed in present implementation of EI, though other methods such as Hirshfeld partition
should work equally well or even better. (Note that since Mulliken method is used, which is
incompatible with diffuse functions, the use of diffuse basis functions must be avoided!)
The BPI between atoms A and B in a molecule is defined as
BPI AB = ( EI A − EI ref
A ) − ( EI B − EI B )
ref
where EIref is reference EI value derived from calculation of homonuclear species. For example, you
ref ref
study BPICN for H3C-NH2, then EI C is computed as EIC in ethane, and EI N is computed as
EIN in H2N-NH2. The larger magnitude of BPIAB implies higher bond polarity of the A-B bond.
Group electronegativity is evaluated as negative of EIX for corresponding radical, X is the
attaching atom. For example, to obtain group electronegativity for -CH3 group, you should calculate
-EIC for CH3 radical.
This function of Multiwfn is used to calculate EI for specific atom in present system, all the R,
RO and U types of HF/DFT wavefunction are supported. Multiwfn automatically detects the number
of inner-core electrons and determines which MOs are the valence ones and thus should be taken
into account.
316
3 Functions
3.200.13 Evaluate orbital contributions to density difference or other

grid data
Theory
This function is mainly designed to evaluate contribution of each of selected orbitals to a given
density difference, , so that you can clearly understand which orbital(s) are main contributor(s)
of change in electron density distribution. A similar idea has been employed in J. Mol. Model., 24,
25 (2018) to study contribution of various NBO orbitals to Fukui function (a special kind of electron
density, see Section 4.5.4) to better unravel its chemical meaning. Below, the theory and algorithm
used in the present function are outlined.
 is able to be approximately represented as linear combination of probability density of
orbitals, which is norm of corresponding orbital wavefunction
   pi | i (r ) |2
i
What we need to obtain is the optimal value of {p} for expanding the . The pi can be regarded as
contribution of orbital i to the . The {p} could be derived via least-squares method by minimizing
the difference between  and  p |  (r ) |
i
i i
2
over the whole space, at the meantime the sum of
{p} could be constrained to a given value P via Lagrangian multiplier technique. The error function
to be minimized in the actual implementation is
2
   
F =    (r ) −  pi | i (r) |2  d r +    pi − P 
 i   i 
In Multiwfn, the integral is treated as numerical integration based on evenly distributed grids, that
is
2
   
F = V    (r ) −  pi | i (r) |2  +    pi − P 
  i   i 
where  is index of grid point, V is grid volume.
Clearly, below conditions should be satisfied to determine the optimal {p}
F
= 0 i = {1, 2,3 }
pi
p i
i =P
more explicitly,
F
pi
=0   p  |  (r ) | |  (r ) |
j
j i
2
j
2
+  =  | i (r ) |2  (r )

obviously the working equation to determine {p} should be
317
3 Functions
 A1,1 A1, N 1  p1   B1 
    
   =  
 AN ,1 AN , N 1  pN   BN 
    
 1 1 1 0     P 
Ai , j = Aj ,i =  | i (r ) |2 |  j (r ) |2

Bi =  | i (r ) |2  (r )

The fitting error reported by Multiwfn is estimated using the following formula
  (r) −  p |  (r) |
2
definition1: i i dr
i
2
 
definition 2 :    (r ) −  pi | i (r ) |2  d r
 i 
In principle, this algorithm works for any kind of  and orbital. For example, the  may
corresponds to Fukui function, density variation during electron excitation and so on. The orbital
could be localized molecular orbital (LMO), NBO, MO, etc. The contribution p can be positive or
negative, the sign reflects phase of participation of orbital density in the . Notice that the choice
of orbital range is highly arbitrary while evidently affects the result. For example, the range can
include all orbitals of a certain type, or only include occupied ones.
Usage
Below is the common procedure to derive contribution of a set of orbitals to  via the present
function
(1) Use Multiwfn or other codes to generate cube file of . The procedure of calculating grid
data of  has been substantially illustrated in many sections of present manual, for example,
Section 4.5.4 (Fukui function and dual descriptor) and Section 4.18.3 ( corresponding to electron
excitation).
(2) Load a file containing orbitals of interest into Multiwfn, then enter subfunction 13 of main
function 200.
(3) Input the path a cube file containing  to make Multiwfn load it.
(4) Use option 1 to set constraint on the sum of contributions, namely the P value in above
equations. The P is default to 1.0.
(5) Select option 0, then set the range of orbitals to be taken into account. After calculation,
contribution (p) of all chosen orbitals will be shown on screen.
(6) In the post-processing menu, you can choose to visualize isosurface of , fitted density
 p |  (r ) |
i
i i
2
or their difference  −  p |  (r) |
i
i i
2
so that you can visually examine the
fitting quality. The fitted density can also be exported as cube file.
In Multiwfn, there are two modes to deal with the grid data of orbitals during construction of
the A matrix and B vector. You can switch the mode via option 2.
• Memory based (default): Grid data of density of all chosen orbitals are automatically
calculated first and recorded in memory, in this case the calculation is fairly fast however the
318
3 Functions
requirement on available memory is very high when large range of orbitals is selected and the grid
quality is relatively high. This mode is strongly recommended to use if Multiwfn does not crash due
to insufficient memory.
• Cube file based: Grid data of density of all chosen orbitals are automatically calculated and
saved as individual cube files in current folder as rho_xxxxx.cub, where xxxxx is orbital index. The
file will be loaded when corresponding orbital is used to construct the A matrix and B vector. The
speed of this mode is by far slower than the "memory based" mode, but the advantage is that memory
consumption is almost negligible.
Note that the grid setting of the automatically calculated orbital densities is set by the program
to exactly identical to that of the loaded  grid data.
It is important to note that the present function is general, flexible and not necessarily limited
to study orbital contributions to . For example, if the provided cube file contains grid data of 
rather than , then the present function will yield contribution of selected orbitals to  (of course,
before doing this, the P should be properly set to make the resulting contribution values meaningful.
If you are not sure how to set it, you can simply remove the constraint)
Practical analysis examples of this function are given in Section 4.200.13.
3.200.14 Domain analysis (obtaining properties within isosurfaces of a

function)
This function is used to integrate specific real space functions in domains. The domain refers
to individual spatial regions enclosed by isosurfaces of a given real space function. For example,
you can integrate electron density within various domains defined by isosurfaces of reduced density
gradient (RDG) to study strength of weak interaction (In fact, if this module is flexibly used, much
more things can be realized. For example, visualizing and obtaining volume of molecular cavity).
Below is basic procedure of using this module:
(1) Use option 2 and 3 to set the way to define the domains. For example, you selected RDG by
option 2 and input <0.5 in option 3, then the region where RDG is less than 0.5 will be integrated.
(2) Choose option 1 and properly define grid, then Multiwfn starts to calculate the grid data for the
real space function you selected and identifies domains that statisfied the criterion you set. (If you
already have grid data in memory, you may also choose option -1 to directly use it rather than
calculate new grid data)
(3) Once calculation in last step is finished, Multiwfn prints total number of grids in each domain.
In very simple case, from these information you may directly infer which domain is the one you
want to study, while for common case, you need to use option "3 Visualize domains" to visualize
domains in a GUI window, in which you can select domain at the right-bottom list and check its
profile, each green point on the graph corresponds to a grid in the domain. Once you found the
domain you are interested in, close GUI and choose "1 Perform integration for a domain", then
select the real space function to be integrated in the domain, after that the integral value, domain
volume and average/maximum/minimum value of the integrand in the domain will be outputted. In
addition, minimum and maximum X/Y/Z of grids belonging to the domain, as well as span distance
319
3 Functions
in X/Y/Z will also be outputted. You can also select "2 Perform integration for all domains" to obtain
integral values for all domains. (Hint: If what you need is just domain volume, you should choose
the user-defined function as integrand, which by default is 1.0 everywhere and thus integrating this
function does not take any computational time).
In addition, some regions you are interested in may be identified as separate domains, to study
the property of the regions more conveniently, you can choose "-1 Merge specific domains" to merge
selected domains are single domain, so that you do not need to manually sum up their integral values.
If you wish to visualize domains in third-part program such as VMD, there are two ways:
 Select "10 Export a domain as domain.cub file in current folder" and input index of the domain
of interest, then Multiwfn will export the domain as domain.cub, in which the grid point
belonging and not belonging to the domain have value of 1 and 0, respectively (value of
boundary grids are also outputted as 0 to guarantee that the isosurfaces always look closed).
After that, you can load the cube file into visualization program to visualize isosurface using
isovalue of 0.5.
 Select "11 Export boundary grids of a domain to domain.pdb file in current folder" and input
index of the domain of interest, then the resulting .pdb file will contain particles, each one
corresponds to a boundary grid. You can directly drag this file into VMD and render the
particles as spheres to visualize domain.
By the way, in the post-processing menu there is an option "5 Calculate q_bind index for a
domain", this is used to calculate the qbind index defined in J. Phys. Chem. A, 115, 12983 (2011), in
which it was demonstrated that for hydrogen-bond dimer, the scan curve of qbind index well mimics
to actual potential energy curve. This index for a domain is defined as:
qatt =   n (r ) d r attractive effect

2 ( r )0
qrep =   n (r ) d r repulsive effect

2 ( r )0
qbind = −( qatt − qrep )

where 2(r) is the second largest eigenvalue of electron density Hessian matrix at r, its sign can be
utilized to discriminate interaction type. Since the paper showed that n=4/3 gives best correlation
between qbind and actual potential curve, in Multiwfn the n parameter is chosen as 4/3. Note that the
paper used isosurface of RDG=0.6 when calculating this index. More negative of qbind may imply
more stablized interaction.
The method used to integrate domains in present module is even-grid integration method. In
other words, the integration value of a real space function for a domain is simply the sum of the real
space function value of the grids constituting the domain multiplied by grid volume. Therefore, the
accuracy of integration result is directly affected by the quality of grid you set.
Illustrative application examples of present module is given in Section 4.200.14.
Information needed: Atom coordinates, GTFs (depends on the choice of real space function)
320
3 Functions
3.200.15 Calculate electron correlation index
The total, dynamic and nondynamic electron correlation indices proposed by Matito et al. in
Phys. Chem. Chem. Phys., 18, 24015 (2016) are useful indicator of measuring magnitude of electron
correlation in present system.
Dynamic and nondynamic electron correlation indices (ID and IND) are defined as
ID = 1
4 {[ (1 −  )]
i
i i
1/ 2
− 2i (1 − i )}
I ND = 1
2  (1 −  )
i
i i
Where i denotes index of natural spin orbital,  is corresponding occupation number. Note that in
some cases,  may be marginally larger than 1.0 or negative, Multiwfn automatically set it to 1.0
and 0.0 respectively to make the calculation feasible.
Total electron correlation index defined is
I T = I D + I ND = 1
4 [ (1 −  )]
i
i i
1/ 2
| i (r ) |2
Present function is used to calculate all the three electron correlation indices. Any wavefunction
file carrying occupation number of natural orbitals may be used as input file, e.g. .mwfn, .wfn, .wfx
and .molden files. An example is given in Section 4.A.6.
Note that Matito et al. also proposed local version of the three functions to characterize electron
correlation in local regions, Multiwfn is also able to study them, see Section 4.A.6 for example.
3.200.16 Generate natural orbitals, natural spin orbitals and spin

natural orbitals based on the density matrix in .fch/.fchk file
In .fch (or .fchk) file, density matrix is always recorded. For example, if you carried out a MP2
task for an open-shell system with Gaussian keywords "# MP2/cc-pVTZ density", then the
resulting .fch file will have below four fields recording corresponding type of density matrix:
Total SCF Density, Spin SCF Density, Total MP2 Density, Spin MP2 Density
While for a closed-shell system, if the keyword used is "# TD PBE1PBE/6-311G* density", then
the resulting .fch file will contain below type of density matrix:
Total SCF Density, Total SCF Density, Total CI Rho(1) Density, Total CI Density
If you do not know which kinds of density matrix are recorded in the .fch file, simply search
"Density" in the file.
Various kinds of natural orbitals can be obtained via diagonalization of proper type of density
matrix:
Natural orbitals (NOs): Diagonalizing total density matrix. The occupation is from 0.0 to 2.0.
This type of NOs is also known as spatial NOs, and specifically, unrestricted natural orbital (UNO)
for unrestricted wavefunctions
Alpha and beta natural orbitals (collectively known as natural spin orbitals, NSOs):
Diagonalizing alpha and beta density matrix, respectively. The occupation is from 0.0 to 1.0.
Spin natural orbitals (SNOs): Diagonalizing spin density matrix (i.e. Difference between
321
3 Functions
alpha and beta density matrix). The occupation is from -1.0 to 1.0. SNO with positive (negative)
occupation represent distribution of unpaired alpha (beta) electrons.
Using present function, you can obtain any set of above mentioned types of NOs. For example,
you want to obtain SNOs of triplet water at CCSD/cc-pVDZ level, you can run below Gaussian
input file:
%chk=C:\CCSD_water_m3.chk
#p CCSD/cc-pVDZ density
test
0 3
O 0.00000000 0.00000000 0.11930801
H 0.00000000 0.75895306 -0.47723204
H 0.00000000 -0.75895306 -0.47723204
Convert the .chk file to .fch, then boot up Multiwfn and input
C:\CCSD_water_m3.fch
200
16
CC // Meaning that we want to analyze coupled-cluster density matrix. You can also input
SCF here to analyze Hartree-Fock density matrix
3 // Generate SNOs (if the system is closed-shell, this selection will not occur, since only
NOs can be generated in this case)
Now the basis function information in memory has been updated to SNOs. If then you want to
visualize SNOs, or to perform real space function analysis (e.g. analyzing orbital composition of
SNOs via Hirshfeld partition), you should choose y to export wavefunction information to
new.molden file in current folder, and then program will automatically load it. After that, all
following analysis will correspond to SNOs.
One of my blog articles detailedly discussed and presented analysis example of SNOs: "The
way of generating natural orbitals based on fch file in Multiwfn and analysis instances about excited
state wavefunctions and spin natural orbitals" (in Chinese) http://sobereva.com/403.
Note: Once .molden file containing SNOs is loaded into Multiwfn, the system will be regarded
as open-shell and there will be the same number of alpha and beta orbitals, only the former
correspond to SNOs, while the latter are completely meaningless and you should simply ignore them.
This function works well for .fch/.fchk files produced by Gaussian and PSI4, but may or may
not be compatible with other programs.
If this function is used in combination with PSI4, you can analyze wavefunction as high as
CCSD(T) level, please check Section 4.A.8 for detail.
The example in Section 4.18.9 utilized this function to generate natural orbitals for transition
density matrix.
Information needed: .fch/.fchk file
322
3 Functions
3.200.17 Calculate Coulomb and exchange integral between two

orbitals
Theory and implementation

Coulomb (ii|jj) and exchange integral (ij|ji) are the two most important integrals in quantum
chemistry. This function is used to calculate them between two selected orbitals i and j, their
expressions are:
i (r1 )i (r1 ) j (r2 ) j (r2 )
(ii | jj ) =   d r1 d r2
r12
i (r1 ) j (r1 ) j (r2 )i (r2 )
(ij | ji ) =   d r1 d r2
r12
This function is applicable for any kind of orbital, such as molecular orbitals, localized
molecular orbitals, natural transition orbitals and so on.
Currently, the integral is calculated based on uniform grid (i.e. evenly placed grid):
 2j (rl )
(ii | jj ) = ( d x d y d z ) 2

k
i
2
(rk ) 
l k rl − rk
i (rl ) j (rl )
(ij | ji ) = ( d x d y d z ) 2 i (rk ) j (rk ) 
k l k rl − rk
where k and l are indices of grid; dx, dy and dz are grid spacing in X, Y and Z directions, respectively.
This function is fairly time-consuming. For a given system, the smaller the spacing, the higher
the computational cost and better the accuracy. If you do not know if the grid spacing currently
employed is small enough, you can make a convergence test, namely gradually decreasing the
spacing and check when the value is converged.
There are two kinds of input files could be used:
• A file containing orbital wavefunction. If you use such as .mwfn, .fch or .molden as input file
when Multiwfn boots up, after entering present function, you will be asked to input two orbital
indices and choose a grid setting, then grid data of wavefunction will be automatically calculated
for them.
• Two cube files containing orbital wavefunction. You should input the cube file of the first
orbital when Multiwfn boots up, and after entering present function, input cube file of another orbital.
The cube file can be generated by any quantum chemistry code (also including main function 5 of
Multiwfn).
In the interface of present function, you can set truncation value for Coulomb (J) integral and
exchange integral (K) respectively prior to the calculation. The innermost summation of Coulomb
and exchange integral will be ignored if i2 (rk )   J and | i (rk ) j (rk ) |  K , respectively.
Clearly, if the truncation values are properly set, the cost could be significantly reduced while
keeping accuracy almost unchanged. Commonly using the default value is suggested.
Note that if you need to calculate one-electron orbital integrals, you should use the function
described in Section 3.200.10.
Example
323
3 Functions
Here I use water molecule as example to illustrate calculation of the two kinds of integrals.
examples\H2O_iijj.fch // Containing molecular orbitals at HF/6-31G* level
200 // Other functions (Part 2)
17 // Calculate Coulomb and exchange integrals between two orbitals
4,10 // The two orbitals are selected to be MO4 and MO10
1 // Low quality grid (corresponding to grid spacing of 0.2 Bohr)
1 // Calculate Coulomb integral with default truncation level. The result is 0.615700
3 // Calculate exchange integral with default truncation level. The result is 0.122246
The exact value of (ii|jj) and (ij|ji) computed by analytic integral are 0.623256 and 0.129893,
respectively, clearly accuracy of our values calculated based on numerical integration is basically
satisfactory. If you employ better grid, for example spacing of 0.1 Bohr (corresponding to "medium
qualtiy grid"), the accuracy will be further noticeably improved (0.62143 and 0.12793, respectively),
but the cost will be eight times higher, note that the cost is inversely proportional to cube of the grid
spacing.
3.200.18 Calculate bond length/order alternation (BLA/BOA) and

angle/dihedral alternation
Theory
In conjugated polymers, the atom and bond properties in the conjugation chain show alternant
character. The bond length alternation (BLA) is an important quantity in the study of this kind of
systems. To calculate BLA, the atom sequence in the conjugated chain should be given. For example,
the atom sequence is given as 3-5-6-9-10-12, the bond 1 is thus 3-5, the bond 2 is 5-6, etc. The BLA
in this case is calculated as
BLA = (R5-6+R9-10)/2 − (R3-5+R6-9+R10-12)/3
where R is bond length. More generally, the BLA is defined as below (see Eq. 7.6 of Handbook of
Thiophene-based Materials: Applications in Organic Electronics and Photonics)
BLA = average length of even bonds − average length of odd bonds
Smaller magnitude of BLA implies better electron conjugation along the selected path. This quantity
has been frequently employed in literatures, see J. Chem. Phys., 136, 094904 (2012) for research
example and http://photonicswiki.org/index.php?title=Structure-Property_Relationships for a
comprehensive review about the relationship between BLA and various molecular properties.
Bond order is a quantity closely related to bond length, thus the bond order alternation (BOA)
is also a quantity as useful as BLA. The only difference between BOA and BLA is that the bond
length in the latter is replaced with bond order. Compared to the BLA, the BOA exhibits the bond
alternation character from electronic structure aspect rather than simply from geometric aspect. As
fully introduced in Section 3.11, there is no unique definition of bond order. The Mayer bond order
is very suitable for evaluating BOA since it is quite general, cheap and its magnitude is close to
formal bond order.
The bond angle alternation and dihedral alternation are also frequently studied, Multiwfn is
able to calculate variation of bond angles and dihedrals along the chain.
Usage
324
3 Functions
To use this function, an atom sequence must be defined, this is quite easy. After entering the
present function, you will be asked to input the indices of the atoms that make up the sequence, the
order is completely arbitrary. Then you need to input index of the beginning atom and ending atom
in the sequence. After that, based on these information and interatomic connectivity, Multiwfn
automatically identifies the actual atom sequence and prints it on the screen, you are suggest to
briefly check it to ensure the sequence is correct. Then, for each bond in the atom sequence,
Multiwfn prints its index, corresponding atom indices, bond length and Mayer bond order, then
outputs BLA and BOA values. The bond data are also exported to current folder as bondalter.txt so
that you can import it to data plotting tools such as Origin to plot "bond length vs. bond index" and
"bond order vs. bond index" curve maps. Finally, if you want to study bond angle and dihedral
alternation along the sequence, you can also let Multiwfn to output them.
The present function is also able to be used to study above mentioned properties for a closed
path (e.g. a ring), the index of ending atom in this case should be identical to the beginning atom.
If your input file contains both atom information and basis function information, such
as .mwfn, .fch and .molden files, both bond lengths and bond orders will be outputted, as stated
above. However, if your input file only contains atom information, such as .xyz, .mol2 and .pdb
files, then bond order information will not be calculated and printed.
If the input file contains interatomic connectivity, such as .mol and .mol2 format, the
connectivity matrix will be directly loaded from it. For other file formats, the connectivity is guessed
based on atom coordinate and atomic radii. If present function does not properly work, using .mol
or .mol2 file with correct connectivity as input file is recommended.
Since Mayer bond order is incompatible with diffuse functions, employing diffuse functions
must be avoided when generating wavefunction.
An example of calculating BLA/BOA and plotting "bond length/order vs. bond index" map is
given in Section 4.200.18.
Information needed: Atom coordinates, basis function (optional)
3.200.20 Bond order density (BOD) and natural adaptive orbital

(NAdO) analyses
1 Preface
The concept of delocalization index (DI) has been detailedly introduced in Section 3.18.5. The
DI between two regions is closely related to the electronic correlation between the two regions.
Essentially, Mayer bond order and fuzzy bond order are DI calculated based on atomic spaces
defined in terms of Hilbert partition and fuzzy partition.
DI is a value. If it can be visualized, then it will be quite helpful in understanding its nature and
interatomic interaction. In J. Phys. Chem. A, 124, 339 (2020), the author proposed a real space
function named bond order density (BOD), its integral over the whole space is just DI, therefore
BOD directly reveals local contribution to DI. Clearly BOD must be a useful function in
characterizing chemical bonds. The natural adaptive orbital (NAdO) is a kind of orbital closely
related to BOD, it can exhibit source of DI from orbital picture. Below I detailedly describe all
325
3 Functions
details about BOD and NAdO.

Note that the NAdO has no any relationship with the adaptive natural density partitioning (AdNDP) orbital
introduced in Section 3.17.
2 Theory of BOD
In order to fully understand underlying idea of BOA, it is crucial to first familiar yourself with
some related concepts.
• nth-order cumulant density
The nth-order cumulant density 𝜌C𝑛 (𝐫1 , 𝐫2 ⋯ 𝐫𝑛 ) was detailedly introduced in Comput. Theor.
Chem., 1003, 71 (2013), it represents the part of nth-order reduced density 𝜌𝑛 (𝐫1 , 𝐫2 ⋯ 𝐫𝑛 ) that
cannot be expressed in terms of lower orders of reduced density, and thus provides an appropriate
measure of the n-electrons correlation existing in the system. Explicit expression of 𝜌C1 , 𝜌C2, and
𝜌C3 are given below (expressions of others orders can be found in the Comput. Theor. Chem. paper).
 C1 (r ) =  (r )
 C2 (r1 , r2 ) =  (r1 )  (r2 ) −  2 (r1 , r2 )
 C3 (r1 , r2 , r3 ) =  (r1 )  (r2 )  (r3 ) + (1 / 2)  3 (r1, r2 , r3 )
− (1 / 2)[  (r1 )  2 (r2 , r3 ) +  (r2 )  2 (r1, r3 ) +  (r3 )  2 (r1, r2 )]
where 𝜌2 (𝐫1, 𝐫2 ) corresponds to the pair density  introduced in Section 2.6.

𝜌C𝑛 has an important feature
Cn−1 (r1 rn−1 ) =  Cn (r1 rn )d rn
as a consequence,
  d rn = N
n
C (r1 rn )d r1
where N is the total number of electrons.
• n-center population and DI

n-center population is defined as follows
N ( A, B n) =   
n
C (r1 rn )d r1 d r2 d rn
A B n
The subscript of the integral denotes the integration region, usually it corresponds to atomic space.
After properly normalization, the n-center population can be named as n-center delocalization index
to quantify multi-center delocalization extent.
It is important to note that 𝜌C2 just corresponds to the negative of the well-known exchange-
correlation density XC, whose integral directly defines DI ():
 ( A, B) = −2  XC (r1, r2 )d r1 d r2  2  C2 (r1, r2 )d r1 d r2

A B A B
Extensive introduction of DI can be found in Section 3.18.5. Evidently, (A,B) essentially

corresponds to the 2-center population (only differs by a factor of 2).
• Definition of BOD
The one-electron function BOD between regions A and B is defined as
BOD AB (r) = 2 AB (r)
326
3 Functions
where
 AB (r) =   C3 (r1, r2 , r3 )d r2 d r3

A B
For closed-shell cases, the working equations for single-determinant wavefunctions (and thus
without explicit representation of Coulomb correlation in the wavefunction) are
occ occ
 AB (r ) = i (r ) DiAB
, j  j (r )
i j
D AB
= S( A)S( B ) + S( B )S( A)
where i and j are doubly occupied spatial orbitals,  is orbital wavefunction. S(A) usually denotes
atomic overlap matrix (AOM) of  spin in atom A, it can also denotes basin overlap matrix (BOM)
of  spin in basin A.
There are important relationships correlating the BOD with DI and localization index (LI, )
 BOD AB (r ) d r =  ( A, B)
(1 / 2)  BOD AA (r ) d r =  ( A)
Obviously, BOD is able to reveal contribution of every spatial position to DI and LI.
For unrestricted open-shell single-determinant wavefunctions,  and  spins should be

separately taken into account:
BOD AB (r) = BODAB (r) + BODAB (r)

The working equation of  spin is
occ occ
BODAB (r ) = 2  AB (r ) = i (r ) Di, j, AB j (r )
i j
 , AB
D = S ( A)S ( B ) + S ( B )S ( A)
 
Si, j ( A) =  i (r ) j (r ) d r i, j  
A
Relevant relationships:
 BOD

AB (r ) d r =   ( A, B)
(1 / 2)  BOD AA (r ) d r =   ( A)
In fact the relationships can be easy demonstrated, given that (with consideration of the orbital
orthogonality condition)
occ occ occ
 BOD AB (r) d r =  Di , j  i (r) j (r) d r =  Di ,i

  , AB   , AB
i j i

occ
Di, j, AB =   Si,k ( A) Sk, j ( B ) + Si,k ( B ) Sk, j ( A) 
k
we have (note that S is a symmetric matrix)
327
3 Functions
occ occ
 BOD AB (r) d r =   Si ,k ( A) Sk ,i ( B) + Si ,k ( B) Sk ,i ( A) 
    
i k
occ occ
=   Si,k ( A) Si,k ( B ) + Si,k ( A) Si,k ( B ) 
i k
occ occ
= 2 Si,k ( A) Si,k ( B )
i k
which corresponds to the expression of   given in Section 3.18.5.
In principle the BOD can be applied to multiconfiguration wavefunctions, however currently

Multiwfn only supports BOD analysis for single-determinant wavefunctions.
3 Natural adaptive orbital (NAdO)

The BOD of  spin can also be expressed in terms of natural adaptive orbitals (NAdOs, ) of
 spin:
 AB (r) = i (r)nAB,ii (r)
i
,𝑖
The 𝑛𝐴𝐵 is eigenvalue of NAdO i in  spin. Evidently, if the eigenvalues are viewed as occupation
numbers, the BOD will be equivalent to the electron density calculated based on the NAdOs.
The NAdOs between regions A and B can be easily constructed. First, diagonalizing DAB to
obtain eigenvalue matrix n and eigenvector matrix U
U−1D ABU = n
MOs can then be transformed to NAdOs via the unitary transformation matrix U
CNAdO = Cocc
MO
U
MO NAdO
where 𝐂occ and 𝐂occ are coefficient matrix of occupied MOs and NAdOs in basis functions,
respectively, and different columns correspond to different orbitals. Assume that there are m
occupied MOs, then both of them have m columns.
Note that for unrestricted wavefunctions, the  and  NAdOs are generated in above way
separately based on  and  occupied MOs, respectively.
4 Usage of BOD analysis module

Before performing BOD analysis, you should use fuzzy atomic space analysis module to export
AOM to AOM.txt, or use basin analysis module to export BOM to BOM.txt. Then, enter subfunction
20 of main function 200, select corresponding option to load either AOM or BOM from a specific
plain text file, input two atomic indices or two basin indices, then NAdOs will be generated and
exported to NAdOs.mwfn in current folder, in which occupation number corresponds to NAdO
eigenvalues (hence the sum of occupation numbers just corresponds to DI).
Next, if you want to directly examine BOD and NAdOs, you should select "y" to load the
NAdOs.mwfn, then you can visualize NAdOs via e.g. main function 0. Note that as mentioned above,
electron density corresponds to BOD currently; therefore, for example, if you want to plot isosurface
of BOD, you can use main function 5 to calculate and plot electron density, the resulting map will
correspond to BOD isosurface.
328
3 Functions
Examples of using BOD and NAdO to analyze practical chemical systems are given in Section
4.200.20.
Information needed: Atom coordinates, basis function.
3.300.1 Viewing free regions and calculating free volume in a box
This function is dedicated to visualize free regions in the box of molecular dynamics simulation
or in experimental crystal cell of systems containing pores, such as porous materials and coal. The
volume of free region can also be outputted.
Algorithm
This function calculates below two types of grid data, their isosurfaces are used to graphically
exhibit free regions, in other words, regions that not occupied by atoms.
(1) Primitive grid data
Value of all grid points is originally set to 1, then all grid points are looped, if distance between
a grid and any atom is smaller than van der Waals (vdW) radius of this atom, then this grid point
will have value of 0, meaning occupied.
The volume of free region, namely free volume, is calculated as the total number of occupied
grid points multiplied by grid volume.
(2) Smoothed grid data
Value of a grid point i is calculated as follows. In this way, atoms are broadened as Gaussian
functions to make variation of the grid data smooth
 2  
f i = max  1 −  e −|R A −ri | /(2c )  , 0 
2
 A  
c = FWHM/ (2 2 ln 2)
where RA is coordinate of atom A, ri is position of point i, FWHM is full width at half maximum of
Gaussian function and defined as two times of atomic vdW radius. Value of 0 and 1 denotes this
grid is fully occupied and unoccupied, respectively; while value between 0 and 1 shows the grid is
partially occupied, the higher the value, the lower the occupancy. In order to visualize free region
based on this smoothed grid data, commonly isovalue should be set to 0.5.
Usage
You should load a file containing atom coordinate information as input file, then enter
subfunction 1 of main function 300. After that you can find a menu, the options are described below
• 1 Set grid and start calculation: After selecting this option, Multiwfn will ask you to input
grid spacing and box lengths in X, Y, Z. If the input file is .pdb format and CRYST1 field is present
to record box lengths, these values will be automatically employed. Next, Multiwfn starts to
calculate the grid data mentioned above, and free volume is outputted in this process. In the post-
processing menu, you can directly visualize isosurface to examine free regions or export grid data
as cube file.
329
3 Functions
• 2 Toggle considering periodic boundary condition: When status of this option is "Yes",
periodic images of the atoms in the box will be taken into account during calculating grid data. This
treatment makes calculation much more expensive but results in much more realistic grid data.
• 3 Toggle calculating smoothed grid data of free regions: By default Multiwfn generates both
primitive grid data and smoothed grid data. If status of this option is set to "No", then only the
former will be generated, and the cost will thus be reduced.
• 4 Toggle the way of setting up grid: By default, Multiwfn assumes that the origin of the box
is (0,0,0) and actual box lengths are available, and the latter will be provided by user after selecting
option 1. If these two conditions are not met, you should select this option once to change status to
"Defining grid using normal interface", then after selecting option 0 you will find an interface like
main function 5 used to define grid setting.
• 5 Toggle making isosurface closed at boundary: When status is set to "Yes", if there are
isosurfaces at box boundary, the isosurfaces will look totally closed. If the status is switched to "No",
then the isosurfaces will looks open at boundary.
• 6 Set scale factor of vdW radii for calculating free volume and primitive free region: If value
of this option is not the default 1.0 rather than set to a value x, then the vdW radii used for calculating
primitive grid data will be scaled by x. Clearly this option affects isosurface of primitive grid data
and outputted free volume.
Note that current function is only able to deal with rectangle box, while other types, such as
triclinic box, are not supported.
The efficiency of the code of this function is high, and it is possible to use this function to study
a system containing even tens of thousands of atoms. When you find the calculation cost is too high
to afford, you should consider below solutions:
(1) Using better computer. The code of this function is fully parallelized, therefore the more
the CPU cores, the lower the computational time.
(2) Using larger grid spacing. Usually the default grid spacing of 0.3 Å is a good balance
between accuracy and cost. When the cost is found to be too high for present system, you can try to
increase grid spacing to e.g. 0.5 Å or even 0.7 Å to reduce the number of points to be calculated.
Clearly, the larger the spacing, the poorer the isosurface and worse the accuracy of the outputted
free volume.
(3) Do not take periodic boundary condition into account. The cost will be reduced by one
order of magnitude, however the isosurfaces at some boundary regions will become artificial, and
free volume will even be fully misleading.
An example of using this module is given in Section 4.300.1.
3.300.2 Fitting atomic radial density as linear combination of multiple

STOs or GTFs
This function is used to fit atomic radial density as linear combination of multiple 1s-type Slater
type orbitals (STOs) or S-type of Gaussian type functions (GTFs), so that the atomic radial density
330
3 Functions
then can be evaluated analytically. The fitting involves many technical details, which will be
introduced in Section 3.300.2.1, then the usage of the present module will be described in Section
4.300.2.2. Practical examples of using this module to performing fitting will be given in Section
4.300.2.
3.300.2.1 Algorithm and technical details
Basic idea
The purpose of this function is performing below fitting
 ( r )   fit ( r ) r
  ci e − r
 iSTO
 fit ( r ) 
  ci e
− r 2
iGTF
where {c} are coefficients to be fitted, {} are exponents to be fitted, r is radial distance to nuclear
position.
Fitting type
To realize the fitting, a set of fitting points should be defined. The fitting essentially
corresponds to minimizing the least-square residual that measures overall error between actual
density and fitted density at the fitting points. There are three ways to define the residual, which
correspond to different fitting types:
  
2
(1) Minimizing absolute error: i − ifit 
i
2
  − ifit 
(2) Minimizing relative error:   i
i  i 
i 4 ri2 ( i − ifit )

2
(3) Minimizing error of radial distribution function (RDF):
In above formulae, {i} are fitting points placed at different radial distances, ri denotes radial distance
of point i. i is sphericalized (i.e. spherically averaged) electron density calculated based on loaded
wavefunction file at point i. In Multiwfn, 170 Lebedev angular grid points are used to perform the
spherical average.
Commonly, fitting type (3) is preferred over others, because the density fitted in this way can
reproduce actual density over the entire range. Fitted density corresponding to fitting type (1) can
only well represent the region very close to nucleus, since electron density in this region is
significantly larger than other regions. The density fitted by type (3) usually represents valence and
tail regions well, but has a poor description in the region very close to nucleus.
Fitting functions
The STO and GTF, which are most important functions in quantum chemistry calculation, are
supported in the present module as fitting functions.
The fitting quality is quite sensitive to the number of fitting functions. Clearly, in principle, the
more the fitting functions, the less the fitting error and the higher the time cost during fitting.
In Multiwfn, the least-square type of fitting is conducted by means of Levenberg-Marquardt
331
3 Functions
(LM) algorithm, which is an iterative method to minimize the residual by gradually optimizing
parameters until convergence tolerance is reached. It is important to note that this algorithm
essentially is a local minimization method, therefore, the resulting fitted coefficients and exponents
may be dependent of initial guess.
The coefficients of the fitting functions should always be fitted, however, the exponents can
either be fitted together or be fixed at initial guess to diminish dimension of parameter space and
thus make the convergence easier. Clearly, for a given number of fitting functions, if their exponents
are variable during fitting, the fitting quality in principle should be better than the case that the
exponents are fixed.
When number of fitting functions are large (e.g. >20), fitting exponents along with coefficients
is usually deprecated, because in this case the convergence is often quite difficult, the cost is fairly
high, and sometimes the routine for performing LM algorithm does not work normally.
Removing redundant fitting functions

In order to achieve good fitting quality, usually dozens of fitting functions with small to large
fixed exponents should be employed, these exponents are often generated in an even-tempered way,
namely i=abi. For example, i=0.052(i-1), i=1, 2, 3 ... 30. In this case, sometimes one or more
fitting functions are redundant because their exponents are too large or too small, leading to severe
numerical problems. In order to identify and remove the harmful redundant fitting functions
possessing unreasonable exponents, the following three rules are employed in Multiwfn by default
when GTF is chosen as fitting function:
(1) Steep fitting functions with relatively insignificant coefficient (>105 and meantime c<5)
and flat fitting functions with negligible coefficient (<3 and c<10-4) are deleted.
(2) Apparently, fitted density should be positive everywhere. If fitted density at a detecting
point is negative, then the fitting function having maximum negative contribution to this point will
be deleted. The detecting points include all fitting points as well as midpoints between neighbouring
evenly placed fitting points. Usually too large exponent tends to result in negative value in the region
very close to nucleus, while too small exponent tends to lead to negative value in tail region.
(3) Variation of electron density should decrease monotonically with respect to increase of
radial distance. This requirement is checked using double dense grid in the range spanned by first
half of fitting points; if it is violated, then the fitting function with largest exponent will be deleted,
because this problem is usually incurred by too large exponent.
Each time only one redundant is deleted. After deletion, Multiwfn will redo the fitting. The
fitting repeats until no redundant fitting function can be found.
Scaling coefficients
Integral of fitted density over the whole space may deviate from actual number of electrons
(Nelec), clearly this breaks physical meaning of fitted density and makes it useless in many scenarios.
In order to solve this problem, the fitted coefficients should be scaled by a factor:
N elec
=
 4 r  (r ) d r
2 fit
In Multiwfn, the integral in the denominator is evaluated in terms of Gaussian quadrature of

Chebyshev second kind with 100 integration points.
Number and stepsize of fitting points
332
3 Functions
The common interesting atomic radial region is r=0~4 Å, therefore fitting points should
sufficiently cover this region. The smaller spacing between fitting points, the higher the fitting
quality. By default Multiwfn employs quite fine grid, which has grid spacing of 0.001 Å, and
correspondingly, the default number of grid points is 4000. The default setting of fitting points is
quite appropriate and should not be altered without special reasons. Indeed, properly enlarging grid
spacing and correspondingly decrease number of fitting points can save computational time,
however, given that calculation of electron density for single atom is usually inexpensive, enlarging
grid spacing is not a good idea.
Commonly, using evenly placed fitting points with small spacing is fully adequate to reach
satisfactory fitting quality, however, Multiwfn also supports adding second kind Gauss-Chebyshev
points into fitting points. This kind of points samples very heavily around nucleus, but sparsely
samples in long range; in other words, the grid spacing is positively correlated with radial distance.
Evidently, taking second kind Gauss-Chebyshev points into account will more or less improve
fitting quality in the region very close to nucleus. However, since the improvement is not remarkable
as long as spacing of evenly placed grid is small enough (e.g. the default 0.001 Å), this kind of
points are not included in the fitting by default.
Examining fitting quality

After fitting, it is crucial to examine fitting quality to ensure that the outputted coefficients and
exponents are really reliable and meaningful in practice. There are several ways to do the check:
(1) Check root mean square error (RMSE), which is defined as
N
 ( i − ifit )2
RMSE = i
N
where N is the number of fitting points. The lower the RMSE, the better the overall fitting quality.
You can also use this quantity to compare fitting accuracy between various fitting settings.
(2) Pearson correlation coefficient (r) and r2 between fitted density and actual density at fitting
points. The more the value close to 1, the better the fitting quality.
(3) Visually compare actual density and fitted density curves. Using logarithmic scaling is
suggested. The two curves should be close with each other at least in the high electron density region.
Also you can quantitatively check difference and relative error between actual density and fitted
density at fitting points.
(4) Check integral of fitted density. The integral should be evaluated under different number of
quadrature points, e.g. 60, 80 ... 300. If the fitted coefficients and exponents are indeed reasonable,
in all cases the integral should be very close to actual number of electrons in the atom.
(5) Check fitted density in broad range with double or triple dense grid compared to fitting
points. No negative density should be found and the fitted density should vary monotonically.
3.300.2.2 Usage
Common steps of fitting

To use the present function to performing fitting, commonly you should do below steps
(1) Load wavefunction file of an atom. The file should contain at least GTF information
(e.g. .mwfn/.wfn/.fch/.molden), the atom must be at original point.
(2) Enter subfunction 2 of main function 300.
333
3 Functions
(3) Choose option 3 and select a way to define initial guess, then briefly examine the initial
guess printed on screen, then return to upper level of menu.
(4) Choose option 1 to start fitting.
(5) Carefully check information printed on screen, then you can use proper options to further
check quality of the fitting.
Options before fitting

Meaning and details of the options in the interface of present module are described below
• Start fitting: After selecting this option, Multiwfn will do following things (some procedures
may be skipped as requested by users)
 Print initial guess of fitting functions
 Calculate sphericalized radial atomic density at fitting points
 Use Levenberg-Marquardt algorithm to optimize coefficients/exponents of fitting functions.
In this procedure some redundant fitting functions may be deleted
 Sort fitting functions according to their exponents
 Calculate integral of fitted density and correspondingly scale coefficients
 Print final coefficients and exponents of fitting functions
 Print error statistics
Then you will see a menu used to check fitting quality or export data, the options will be
described later.
• Switch type of fitting functions: You can use this option to switch type of fitting function
between STO and GTF
• Check or set initial guess of coefficients and exponents: After entering this option,
information of current fitting functions will be shown first. By default no fitting functions are set
and thus you should use one of below options to define them prior to starting fitting:
 1 Load initial guess from text file: Via this option, coefficients and exponents will be loaded
from a given plain text file, whose format should look like below:
4.0 1000
2.0 100
1.0 1.5
This file defines three fitting functions, the first and second columns are initial coefficients and
exponents. Clearly, this is the most flexible way of defining number and initial parameters of fitting
functions.
 2 Set initial guess as "crude fitting by a few STOs with variable exponents": This option aims
at performing crude fitting, because only a few STOs are employed. Exponents will be optimized
during fitting to make fitting quality better. Specifically, 1, 2, 4, 6 STOs are employed when the
element is in the first, second, third&fourth and latter rows of periodic table, respectively. The initial
parameters of fitting functions set by this option are commonly reasonable, however, the fitting may
be occasionally failed due to inappropriate initial parameters, in this case you should try to use
option 1 to load manually provided parameters instead.
 3 Set initial guess as "fine fitting by 30 GTFs with fixed exponents": If you want to accurately
fit actual density over the entire range, using fitting functions set by this option is usually the best
choice. 30 GTFs will be employed, and in order to guarantee numerical stability and significantly
reduce fitting cost, only coefficients will be fitted, while exponents will be kept at the initial values,
which are generated as i=0.052(i-1), where integer i varies from 1 to 30.
334
3 Functions
 4 Set initial guess as "fine fitting by 10 GTFs with variable exponents": If you hope to use
limited GTFs to gain acceptable fitting quality, then this option is worth to consider.
 5 Set initial guess as "fine fitting by 15 GTFs with variable exponents": This option is similar
to 4, but will reach higher fitting accuracy since more GTF is employed. Even though the exponents
are variable, the fitting quality of this option is not as perfect as option 3, and meantime the fitting
cost is higher, and there is a danger that the iteration of Levenberg-Marquardt algorithm cannot get
converged.
• Set fitting tolerance: The smaller the value, the better the numerical accuracy of the fitting
while the higher the fitting cost. Commonly the default value is able to guarantee high numerical
accuracy.
• "Set number of evenly placed fitting points" and "Set spacing between fitting points": The
default number of evenly placed points (4000) is large enough and the default spacing (0.001 Å) is
fine enough, the corresponding fitting points distribute from r=0.001 Å to r=4.0 Å. You can properly
adjust these two options if you want to include points in longer range into fitting, or if you want to
reduce fitting cost (the time cost is basically proportional to number of fitting points).
• Toggle scaling coefficients to actual number of electrons: When status of this option is "Yes"
(default), then after calculating integral of fitted density, the coefficients of the fitting functions will
be scaled in the way described in last section. Performing scaling is always recommended.
• Select fitting type: You can use this option to choose how to perform the fitting, including
"Minimizing absolute error", "Minimizing relative error" and "Minimizing radial distribution
function (RDF) error ", which have been introduced in the last section. The third one is usually the
most recommended, thus it is the default.
• Toggle fixing exponents: When status of this option is set to "Yes", Multiwfn will optimize
both coefficients and exponents of fitting functions during fitting procedure. If you want to keep
exponents at their initial guess, this option should be switched to "No".
• Toggle sorting functions according to exponents: When status of this option is "Yes" (default),
Multiwfn will reorder the fitting functions so that their exponents are ranked from low to high.
• Toggle removing redundant fitting functions: When status of this option is "Yes" (default),
the redundant fitting functions will be automatically deleted during fitting, as described in the last
section. This is important to guarantee numerical stability and reasonabless of the result and thus
commonly should be enabled.
• Set number of second kind Gauss-Chebyshev fitting points: As described in the last section,
if you want to add the points heavily sampling the region close to nucleus into fitting, evenly
distributed points and some second kind Gauss-Chebyshev points can be combined together as the
actually adopted fitting points. This option controls the number of employed second kind Gauss-
Chebyshev points. When the value is set to 0 (default), this kind of points will not be employed.
Options after fitting

In the menu that appears after fitting is complete, there are many options used to examine or
export the result, also there are some options used to check fitting quality. Since most of them are
self-explainatory, only a few are described here:
• Visualize actual density and fitted density curves using logarithmic scaling: This option is
quite useful in visually examining fitting quality, the blue and black curves show fitted density and
actual density, respectively. Clearly, the closer the two curves, the better the fitting quality. From
this map you can also understand which regions are nicely or badly fitted.
335
3 Functions
• Export fitted density from 0 to 10 Angstrom with double dense grid to fitdens.txt in current
folder: This option is useful in checking quality of fitted density over broad range. You can use this
option to export fitdens.txt file, from which if you find the fitted density varies as expected (e.g. no
negative value, decays smoothly and monotonically), then the fitted density should be reliable and
usable.
• Check integral of fitted density: This option employs various numbers of quadrature points
(from 40 to 300 with step of 20) of Gaussian integral to calculate radial integral of fitted density. If
the result are all close to actual number of electrons, that means the fitted density should be reliable.
• Check fitted density at a given radial distance: Via this option you can check fitted density at
specific radial distances by inputting their values. This is also a useful way of rationalizing fitted
density.
Practical examples of using this module to performing fitting are given in Section 4.300.2.
3.300.3 Visualize (hyper)polarizability via unit sphere and vector

representations
If you are not familiar with (hyper)polarizability, please check Section 3.200.7 first to gain
basic knowledges. In this section, the unit sphere representation will be introduced, it was proposed
in J. Comput. Chem., 32,1128 (2011) to intuitively represent first-order hyperpolarizability tensor,
while I also extended this idea to polarizability and second-order hyperpolarizability.
Theory
Recall the relationship between molecular dipole moment and external field
μ = μ0 + α  F + (1/ 2)β  F  F + (1/ 6) γ  F  F  F +
The  is known as first order hyperpolarizability tensor, the component ABC is proportional to the
magnitude of induced dipole moment in direction A caused by combination of two incident electric
fields respectively in directions B and C.
In the unit sphere representation, effective dipole vector is defined as
βeff ( ,  ) = β  e( ,  )  e( ,  )

where  and  are angles of spherical polar coordinate, e(,) is unit vector normal to the sphere
surface. More specifically, the components of eff can be explicitly written as
ieff =  i , j ,k ek e j i, j, k = {x, y, z}
j k
The orientation and length of eff(,) vector respectively reflect the direction and magnitude
of induced dipole moment caused by combination of two incident electric fields exerted in the
direction of (,). If eff is calculated at every vertex of a sphere surface enclosing the molecule, one
can clearly and vividly understand the response of molecular dipole moment with respect to external
electric field exerted in various directions. The original paper only employs this representation to
second harmonic generation (SHG) type of , in fact it can also be applied to other kinds of ,
336
3 Functions
including both static and dynamic ones (in the latter case, the exerted external field with varying
strength comes from incident electromagnetic wave, and its direction is perpendicular to the
propagation direction of the electromagnetic wave).
Based on the same idea of eff, I defined below quantities
αeff ( ,  ) = α  e( ,  )
γ eff ( ,  ) = γ  e( ,  )  e( ,  )  e( ,  )
The eff(,) vector reflects the direction and magnitude of induced dipole moment caused by
the incident electric field exerted in the direction of (,), while the eff(,) vector reflects the
direction and magnitude of induced dipole moment caused by combination of three incident electric
field simultaneously exerted in the direction of (,). Similarly, by plotting the eff(,) or eff(,)
at every vertex on a sphere, one obtains unit sphere representation map of  and , respectively.
The so-called vector representation corresponds to plotting (x, y, z) vector as an arrow, the
components are defined as
i = (1 / 3) ( ijj +  jji +  jij ) i, j = {x, y, z}
j
This representation is quite simple, it can approximately represent major character of ,

however, anisotropy character cannot be explicitly exhibited in this representation. For example, x
shows collective effect of xxx, xyy, xzz... components, where the xyy and xzz describe anisotropy
response character since the exerted two electric fields are not colinear with the induced dipole
moment resulting from their combination effect.
Usage
Multiwfn is able to perform unit sphere representation analysis for ,  and , namely
generating plotting script of VMD software (http://www.ks.uiuc.edu/Research/vmd/) based on
loaded (hyper)polarizability tensor. In addition, plotting script corresponding to vector
representation can also be generated for .
After booting up Multiwfn, you should load a file containing atom information for the molecule
under study. For example, .xyz, .pdb and .fch can be used, see Section 2.6. The atom information
will be used to determine proper radius of the sphere involved in the unit sphere representation.
After enter present module (subfunction 3 of main function 300), you can use many options to
adjust parameters of unit sphere and vector representations, such as scale factor of arrow length,
radius of arrow and so on, they are fully self-explanatory. By choosing options 1 or 2 or 3, Multiwfn
will respectively load  or  or  tensor from a specific file (see below), then VMD plotting script
corresponding to unit sphere representation will be generated in current folder (alpha.tcl, beta.tcl
and gamma.tcl, respectively). If option 2 is selected, plotting script for vector representation will
also be generated (beta_vec.tcl). Then, using VMD to run the scripts, the corresponding map will
be immediately obtained.
It is worth to mentioning there is an option "-8 Toggle making longest arrow on sphere has specific length". If
you select it once to switch its status to "Yes", then after selecting option 1 or 2 or 3, you will be asked to input the
expected length of longest arrow on the sphere. Via this option, you can make map plotted by VMD for systems
having very different magnitude of (hyper)polarizability easily comparable.
Preparation of the file containing (hyper)polarizability tensor

The file containing  or  or  tensor can be directly generated by subfunction 7 of main
function 200 by extracting corresponding data from output file of "polar" task of Gaussian. In that
function, you should choose option "-4 Export (hyper)polarizability as .txt file after parsing" once
337
3 Functions
to switch its status to "Yes", then after parsing data via corresponding option,  will be exported to
alpha.txt,  will be exported to beta.txt, and  will be exported to gamma.txt in current folder, they
are what you need in the present function.
The files containing (hyper)polarizability can also be manually prepared, in this case the data
can be generated by quantum chemistry codes other than Gaussian. The format of the file is free,
the sequence of the tensor components is shown as follows (represented in Fortran grammar)
• Polarizability: (((i,j),j=1,3),i=1,3)
• First-order hyperpolarizability: ((((i,j,k),k=1,3),j=1,3),i=1,3)
• Second-order hyperpolarizability: (((((i,j,k,l),l=1,3),k=1,3),j=1,3),i=1,3)
where the cycle of index i is the slowest. For example, below is a file recording  tensor (highlighted
texts are comments):
3.62370000E+001 XX
-2.20999000E+000 XY
0.00000000E+000 XZ
-2.20999000E+000 YX
3.91836000E+001 YY
0.00000000E+000 YZ
0.00000000E+000 ZX
0.00000000E+000 ZY
2.54054000E+001 ZZ
Example of using this function to study practical molecules are given in Section 4.300.2.
3.300.4 Simulating scanning tunneling microscope (STM) image
Theory
Scanning tunneling microscopy (STM) is a quite common experimental technique to image
chemical systems at atomic level, also it has close relationship with electronic structure of the
sample, see wiki page for more information. In the STM imaging process, a conducting tip is put
over the sample and meantime bias voltage (V) is applied between them. Due to quantum tunneling
effect, tunneling current (I) can form between the tip and sample with appropriate distance
separation (usually 4~7 Å) and V. At different positions, I is different due to different interaction
between the tip and sample, in essence the observed STM image characterizes the function I(r). The
STM experiment can be illustrated by the following figure
338
3 Functions
There are two modes of STM:

• Constant height mode: The z coordinate of the tip is fixed at a given value while x and y are
scanned. The two-dimension STM image in this case corresponds to the I(x,y) function.
• Constant current mode: Two-dimension scanning of x and y is performed, and for each (x,y),
the z coordinate of the tip is gradually adjusted until finding the z position (zc) where the I equals to
a specific value. The STM image in this mode at a given current corresponds to the zc(x,y) function.
Evidently, generating STM image of constant current mode is more time-consuming than constant
height mode, since additional dimension (z) needs to be taken into account.
Despite that there are many ways to computationally simulate the STM image, the only popular
one is the model derived by Tersoff and Hamann, see Phys. Rev. B, 31, 805 (1985) and Chapter 6 of
book Introduction to Scanning Tunneling Microscopy (2ed, Julian Chen, 2008). In principle, to
determine the I one must know realistic character of the tip, which, however, is usually unknown.
The key point of the Tersoff-Hamann model is replacing the tip as a point probe, and hence the
formula for deriving I is largely simplified (Tersoff and Hamann also discussed the case that the tip
is locally spherical with specific radius, however this situation is not considered here). The original
Tersoff-Hamann model corresponds to small V and low temperature limits, and it is only applicable
to periodic systems; if finite V is explicitly taken into account, the I for an isolated system can be
expressed as
EF + eV → EF
I (r )   i
| i (r ) |2 (V  0)
EF → EF + eV
I (r )   a
|  a (r ) |2 (V  0)
where i denotes occupied MOs whose energy is between EF+eV and EF, a denotes unoccupied MOs
whose energy is between EF and EF+eV. The EF is Fermi level, which is not well defined for isolated
system, but usually it is taken as average of EHOMO and ELUMO. The e is elementary charge, which is
a positive value (1.602E-19 C). When V is positive, electrons in the tip tunnel into empty states of
the sample; for a negative V, electrons tunnel out from occupied states of the sample into the tip.
Note: Some documents use -eV rather than +eV, this is because the e in these documents corresponds to the
charge carried by an electron rather than elementary charge. Although the above formulae are usually referred to as
Tersoff-Hamann model, strictly speaking, it is not the form proposed by Tersoff and Hamann.
339
3 Functions
The above formulae show that the I(r) is positively proportional to the local density-of-states
(LDOS) at r contributed by the MOs between EF+eV and EF when V<0 or between EF and EF+eV
when V>0 (note that the LDOS in this context is defined differently to that introduced in Section
3.12.4), therefore, if orbital wavefunctions are available and thus LDOS can be calculated, the STM
image can be obtained straightforwardly. When discussing the simulated STM image in this way, it
is suggested to write the unit of the I as that of LDOS, which corresponds to a.u. in Multiwfn.
The STM image simulated in above way cannot be rigorously compared with the experimental
one, since the actual character of STM tip is not taken into account but simplified as a point (owing
to this, the simulated STM image has infinite high resolution). Also note that the magnitude of I is
unable to be determined using the present model due to introduction of approximations, hence only
the relative difference of I between different positions is meaningful and can be discussed.
Usage
To simulate STM image in Multiwfn, using the files containing basis function information is
recommended, such as .mwfn, .fch and .molden. Using the formats only containing GTF
information such as .wfn is also acceptable, however, in this case the STM image with V>0 cannot
be simulated, since they only contain occupied MOs information.
Since the STM simulation is based on molecular orbitals, the theoretical method only supports
HF and KS-DFT (except for double-hybrid functionals), while multiconfigurational methods such
as MP2 and CASSCF are not supported. Restricted closed-shell, restricted open-shell and
unrestricted open-shell are all supported.
The subfunction 4 of main function 300 corresponds to the STM simulation function. After
entering this function, please carefully read prompt on screen, which informs you how the default
Fermi level and bias voltage are determined. In the interface, you can customize EF, V, number of
grids, spatial range of the STM. The mode of STM image can also be selected, both constant height
(default) and constant current modes are available. After making settings appropriate, you can
choose option 0 to start calculation of the data used to plot STM map, the MOs considered in the
calculation will be shown on screen so that you can easily examine if EF and V have been properly
defined. Note that the computational cost is fully determined by the number of grids, the more the
grids, the smoother the image, while higher the cost.
The way of simulating STM image of constant height and constant current modes is somewhat
different, as described below:
• Constant height mode: Before calculation, you should properly set range of X and Y, by
default they are determined by extending the position of boundary atoms by 3 Bohr. The default Z
position of the plane to be calculated is 0.7 Å higher than the top atom (i.e. the atom with maximal
Z coordinate). After selecting option 0 to calculate current at every point in the two-dimension plane,
you can find an interface for plotting the STM image, all options are self-explanatory and thus will
not be described here (the options are quite similar with those in the post-processing menu of main
function 4). By default, the lower limit of color scale is 0, while upper limit is automatically set to
maximal I in the plane data.
• Constant current mode: You should first choose option 1 once to switch the mode from the
default constant height mode to constant current mode. Then you should properly define the
calculation range in X, Y and Z. The default X and Y ranges are identical to the constant height
mode, the default lower and upper limits of Z are 0.7 and 2.5 Å higher than the top atom, respectively.
After selecting option 0, Multiwfn will start calculation of 3D grid data of I, then you can use
340
3 Functions
corresponding options to visualize isosurface map of tunneling current, export the grid data as cube
file, or visualize 2D STM image. For the latter case, you need to input the value of tunneling current,
then Multiwfn will estimate the z value (zc) where I equals to the value at every (x,y) point, the
resulting plane data zc(x,y) can then be directly plotted as plane map via corresponding options on
screen.
Examples of simulating STM images are given in Section 4.300.4.
3.300.5 Calculate electric dipole, quadrupole and octopole moments of

present system analytically
This function calculates electric dipole, quadrupole and octopole moments of present system
based on analytically evaluated integrals. Note that the function described in Section 3.18.3 is also
able to calculate these information, however it calculates the data numerically, and thus it is
evidently slower and its accuracy is marginally lower than the present function.
The information printed by this function is shown below.
Dipole moment:
 x  X A  x
 
μ =   y  =  q A Y A −   y   (r ) d r
 
   
 z  A 
 Z 
  z 
  A
where XA, YA and ZA are the three Cartesian coordinates of atom A. qA is nuclear charge of atom A.
x, y and z are the three Cartesian coordinates of electron position.
Quadrupole moment (standard Cartesian form):

  xx  xy  xz  X AX A X AYA X AZ A   x2 xy xz 
   
Θ =   yx  yy  yz  =  qA  YA X A YAYA YA Z A  −   yx y 2
yz   (r ) d r
 
 zx zy zz 
A
 Z A X A Z AYA Z A Z A   zx zy z 2 
 
Quadrupole moment (traceless Cartesian form):
3 xx − rr 3 xy 3 xz 
 
Θ =  3 yx
tr
3 yy − rr 3 yz 
 3zx 3zy 3zz − rr 

where rr = (xx)2 + (yy)2 + (zz)2.
Magnitude of quadrupole moment: Θ = (2 / 3) [(trxx )2 + (tryy )2 + (trzz )2 ] .

tr
Note that since quadrupole moment in Cartesian form is a symmetric matrix, only 6
components are unique.
Quadrupole moment (spherical harmonic form):
341
3 Functions
Q2,0 = (3 zz − rr ) / 2

Q2,−1 = 3 yz Q2,1 = 3 xz
Q2,−2 = 3 xy Q2,2 = ( 3 / 2)( xx −  yy )
2
Magnitude of Q2: Q2 =  (Q
m =−2
2,m )2 .
Octopole moment in Cartesian form is a third-order tensor with 333=27 components,

however only 10 elements are unique. The XYZ component is calculated as follows, the other
components can be calculated similarly
xyz =  qA X AYAZ A −  xyz  (r ) d r
A
Octopole moment in spherical harmonic form:

Q3,0 = (1 / 2)(5zzz − 3rrz )
Q3,−1 = 3 / 8(5 zzy − rry ) Q3,1 = 3 / 8(5 zzx − rrx )
Q3,−2 = 15 xyz Q3,2 = ( 15 / 2)(  xxz −  yyz )
Q3,−3 = 5 / 8(3 xxy −  yyy ) Q3,3 = 5 / 8(  xxx − 3 yyx )
where rrx = (xxx ) + ( yyx ) + (zzx ) , similarly for rry and rrz.
2 2 2
3
Magnitude of Q3: Q3 =  (Q
m =−3
3,m )2 .
A simple example of this function is given in Section 4.300.5.
Information needed: Atom coordinates, GTF information or basis functions
342
4 Tutorials and Examples
Prologue and generation of input files
Welcome to use Multiwfn! If you have not read "ALL USERS MUST READ" at page 2 of this
manual, please read it first. If you encountered any problem in using Multiwfn, please free feel to
post topic on Multiwfn forum.
Before getting start, I first show you how to generate various kind of input files.
Notice that different functions in Multiwfn require different type of input file, see Section 2.5
for explanation. briefly speaking, for any analysis that solely based on real space function, you can
use .wfn or .wfx as input file. However, many functions require basis function information, in these
cases you have to use .mwfn, .fch/fchk, .molden or .gms file as input file. Since these files contain
richer information than .wfn/wfx file (i.e. basis functions and virtual orbitals), in principle for any
function that requires .wfn/wfx file as input file, you can also use .mwfn/fch/molden/gms instead.
A few functions in Multiwfn (e.g. AdNDP and ICSS analysis) rely on some special files,
requirements on the input files for these situations are clearly indicated in corresponding section in
Chapter 3.
Generating .wfn and .wfx files

• Gaussian: Write out=wfn in route section, leave a blank line after molecular coordinate
section and write the destination path of .wfn file, e.g. C:\otoboku\H2O.wfn (you can consult
H2O.gjf in “examples” folder), then run this file. If the task terminates normally, H2O.wfn will
appear in "C:\otoboku" folder.
If you would like to generate .wfx file in Gaussian (supported since G09 B.01), simply write
out=wfx instead of out=wfn in route section.
If you use MCSCF in Gaussian, in order to generate and export natural orbitals to .wfn, you
should also use pop=no keyword. If you are using Gaussian older than G09 C.01, please carefully
read below information:
If the theoretical method is post-HF type, you have to also add “density” keyword in route section to use current
density, otherwise what outputted to .wfn file will still be HF orbitals. If you are using TDDFT or CIS and you want
to export natural orbitals corresponding to excited state wavefunction, you also need to specify "density" keyword.
For CCD/CCSD, QCISD or MP2/3, MP4SDQ tasks based on unrestricted HF reference state, only when
“pop=NOAB” keyword is also specified then natural spin orbitals rather than spatial natural orbitals will be saved
to the .wfn file. TD, CI and MCSCF tasks of Gaussian can not produce natural spin orbitals.
If the Gaussian you are using is older than G09 B.01, be aware that there is a serious bug, if your task is restricted
open-shell (ROHF and RODFT), the occupation numbers of singly occupied orbitals in .wfn file will erroneously be
2.0, you have to open the file by text editor, locate the last entry "OCC NO =", and then manually change the value
behind it to 1. 0000000.
• GAMESS-US: Add AIMPAC=.TRUE. in $CONTRL section. After the task is finished, the
generated .dat file in the folder defined by $SCR environment variable (see rungms script) will
contain wavefunction information with the same format as .wfn file, extract the content between "-
---- TOP OF INPUT FILE FOR BADER'S AIMPAC PROGRAM -----" and "----- END OF INPUT
FILE FOR BADER'S AIMPAC PROGRAM -----" and save them to a new file with “.wfn” suffix.
• ORCA: .wfn and .wfx files can be generated simply using aim keyword in the input file, or
using the command orca_2aim XXX to convert XXX.gbw to XXX.wfn and XXX.wfx. At least for
343
ORCA 4.1, .wfn and .wfx files cannot be generated when ECP is used. Using .molden file as input
file of Multiwfn is always more recommended.
As regards the method of outputting .wfn files in other quantum chemistry packages, please
consult corresponding manuals.
Generating .fch file

• Gaussian: First run a Gaussian task with e.g. "%chk=test.chk" to yield binary checkpoint file
test.chk file, then run command formchk test.chk to convert test.chk to test.fch.
Note: There is no any difference between .fch and .fchk formats. The former and the latter are the default
extensions of formatted checkpoint file of Windows and Linux version of Gaussian, respectively. You can use either
of them as input file of Multiwfn.
When post-HF task is performed, the orbitals and occupations recorded in Gaussian .fch file
by default are the HF ones, hence the Multiwfn analysis results are identical to HF. Similarly, under
default case, analysis results based on .fch file produced by TDDFT task are identical to ground
state DFT wavefunction. To analyze wavefunction for post-HF wavefunction or TDDFT excited
state wavefunction, analysis should be done using natural orbitals (NOs) at corresponding level,
there are two ways to yield them:
(1) Make Multiwfn generate natural orbitals (or spin natural orbitals, natural spin orbitals) by
using subfunction 16 of main function 200. See Section 3.200.16 for detail. This way is
recommended since it is very convenient.
(2) Make Gaussian write natural orbitals into .fch file. You should first perform post-HF or
TDDFT task with “density” keyword, and then rerun the task only with "guess
(save,only,naturalorbitals) chkbasis" in route section. Note that Gaussian fills orbital occupation
numbers into orbital energy field in .fch file, hence you should write “saveNO” in the first line
of .fch file to let Multiwfn know this behavior. Beware that if what you performed is open-shell
post-HF calculation, even above process is unable to correctly store natural spin orbitals into .fch
file. Generally, I strongly recommend using .wfn/.wfx file to view natural orbitals and analyze real
space functions for post-HF wavefunctions.
For MCSCF calculation, you should load the resulting .fch file and use subfunction 16 of main
function 200 to generate .molden file containing NOs at MCSCF level, and then use this .molden
file as input file. Since for MCSCF the alpha and beta orbitals cannot be generated separately, for
systems with spin multiplicity larger than 1, you must manually open the .fch file, set "Number of
beta electrons" to the same value as "Number of alpha electrons" to make Multiwfn recognize that
there is only one set of orbitals in the input file.
• Q-Chem: Write GUI 2 in $rem field, after task has finished, you will find the resulting .fchk
file in current folder. Beware that if you are using quite old version (maybe 5.0) of Q-Chem, before
loading the .fchk files into Multiwfn, you must set “ifchprog” in settings.ini to 2.
• PSI4: The .fchk file produced by currently latest version (not older versions) of PSI4 is
compatible with Multiwfn. The examples\psi4_fch.inp is an example file of generating .fchk file at
B3LYP/6-31G** level. In Section 4.A.8, I also show how to analyze post-HF wavefunction based
on .fchk file of PSI4.
Generating .molden file

 ORCA: Using the command orca_2mkl XX -molden to convert XX.gbw to Molden input file
XX.molden.input. You do not need to then manually change the suffix from .molden.input
to .molden, since the former can also be recognized by Multiwfn. You can also set
"orca_2mklpath" in settings.ini to actual path of orca_2mkl executable file in ORCA folder,
344
then Multiwfn will be able to directly load .gbw file.

 Molpro: Adding such as put,molden,ltwd.molden at the last line of your input file, the Molden
input file ltwd.molden will be produced after finishing the calculation.
 Dalton: The program automatically outputs .molden file when calculation is finished. The file
is molden.inp in .tar.gz package. This file can be directly loaded without changing suffix.
 NWChem: An example input file is provided as examples\NWChem_molden.nw. After running
it, the .molden file will be generated in current folder. Notice that spherical harmonic basis
functions must be employed (i.e. the "spherical" keyword) and "noautosym" keyword must be
employed when the system has symmetry of point group.
 MRCC: Once the calculation is normally finished, a file named MOLDEN will be generated in
current folder. Then rename it to make it possess .molden suffix. An example is given in Section
4.A.8.
 xtb: Run xtb with "--molden" option, then molden.input will be generated in current folder.
 Other programs: Please consult corresponding manuals.
When pseudo-potential is used and you need to do some analyses relating to nuclear charges,
do not forget to manually change atomic indices in the .molden file as nuclear charges, see Section
2.5 for details.
Note: Currently, only the Molden input file generated by Molpro, ORCA, xtb, Dalton, NWChem, MRCC,
deMon2k, BDF programs are formally supported. If the file is generated by other programs, the result may or may
not be correct, because the files produced by numerous programs are non-standard or problematic. Fortunately,
molden2aim utility is able to deal with the Molden input files produced by wider scope of programs and can output
standardized Molden input files, which is then able to be used as Multiwfn input file. See Section 5.1 for detail.
Generating .gms file

GAMESS-US and Firefly (old name is PC-GAMESS) output file can also be used as Multiwfn
input file, you need to change its suffix as .gms so that Multiwfn can recognize it. Currently, I can
only guarantee that output file of HF/DFT calculation with default NPRINT option can be normally
loaded by Multiwfn. A sample of input and output files of GAMESS-US are provided as
GAMESS_US.inp and UKS_cc-pVDZ.gms in examples folder, respectively.
Now let us start! Note that the examples in 4.x section are relevant to main function x, therefore
you can quickly find the examples you needed. Section 4.A includes special topics and advanced
tutorials, in which more than one functions and some advanced skills may be involved, such as
studying aromaticity and weak interactions. You can find almost all of the files involved in these
examples in “examples” folder. All the texts behind // are comments and should not be inputted as
command. Tutorials in this chapter only cover basic applications of Multiwfn, if you want to learn
the usage of more functions and more options, please read corresponding sections in Chapter 3 and
play with the options that not mentioned in the examples.
For most examples, I take .fch or .wfn as format of input file, however it never means these
functions can only accept these two formats! If you have read Section 2.5, you must know how to
properly choose format of input file for different functions.
PS1: If you would like to analyze wavefunction higher than CCSD level, reading Section 4.A.8
is suggested, you will need PSI4 or MRCC program.
PS2: In chapter 4, many my blog articles written in Chinese are involved, they often contain
extended discussion and more examples. If you cannot read Chinese, you can try to use Google
translator (For example, you can install Mozilla Firefox add-on called "Google translator for
Firefox". After a successful installation, you will find an icon "T" in the firefox toolbar. Now open
345
the desired weblink, click the "T" icon and you will find the entire text in the desired language).
PS3: If you can read Chinese, reading these three articles will be highly helpful: "Tips for
getting start with Multiwfn" (http://sobereva.com/167), “Multiwfn FAQ” (http://sobereva.com/452)
and "The significance, functions and uses of multifunctional wavefunction analysis program
Multiwfn" (http://sobereva.com/184).
4.0 View orbitals and structure
In this section, I will first introduce how to use the built-in interface for visualizing various
kinds of orbitals, then in Section 4.0.3, I will show how to use Multiwfn in combination with VMD
to easily and quickly plot state-of-art orbital graphs.
4.0.1 Viewing molecular orbitals of cycloheptatriene
Boot up Multiwfn, input examples\cycloheptatriene.fch and press ENTER button, then select
main function 0, a GUI window will pop up, meanwhile information of all atom coordinates along
with basic information of featured molecular orbitals are printed on Multiwfn console window.
You can rotate molecule, zoom in/out, adjust bonding threshold, save the graph as image file
and so on via corresponding widgets.
The numbers in the right-bottom list are orbital indices, you can view orbital isosurface by
selecting corresponding index, or directly inputting the orbital index in the text box and then press
ENTER button. Note that if your system is unrestricted open-shell, to select beta orbital you should
input negative index, for example, -5 corresponds to the 5th beta orbital. Plot two orbitals
simultaneously in this window is also possible, as illustrated in Section 4.0.2.
The green and blue isosurfaces correspond to positive and negative regions, respectively. The
isovalue can be adjusted by dragging slide bar. Isosurface style and colors and be altered by
346
corresponding suboptions in "Isosur#1 style". The quality of the isosurface can be set by the options
in "Isosur. quality".
By selecting "Orbital info."-"Show all", energy, occupation number and type of all orbitals will
be shown in Multiwfn console window. If you do not want too many high-lying virtual MOs are
shown, you can choose "Show up to LUMO+10" or "Show occupied orbitals". If irreducible
reprenstations are recorded in the loaded .mwfn/molden/gms file, then they will be shown as the
last column.
In the "Other settings" and "Tools" of the menu bar, there are many useful options, please play
with them, and when you are confused, see Section 3.2 for explanation. The "Tools - Batch plotting
orbitals" is quite worthnoting, via this tool you can very conveniently save a lot of selected orbitals
to respective image file in current folder, see https://youtu.be/SHwrQhqBHZ0 for video illustration.
To close the window, click “RETURN” button. More detailed explanation about this interface
can be found in See Section 3.2.
Note: Visualizing isosurface of Rydberg orbitals by main function 0 of Multiwfn is also possible, however,
since they show very diffuse character, in order to avoid truncating of isosurfaces, you should select "Other settings"-
"Set extension distance" in the menu, and then input a relatively large value, for example, 12 (the unit is Bohr), then
select the orbital to visualize it. The default value of extension distance is controlled by "Aug3D" in settings.ini. An
example of visualizing Rydberg orbitals is given in Part 2 of Section 4.200.5.
Tip: Recommendation of the steps for obtaining pretty orbital isosurface graph
• Enter main function 0, select the orbital to be visualized, properly set isovalue
• Click "Show Labels" to disable axis
• Properly adjust viewpoint
• Properly change the size of atomic labels. Note that type of the labels can be changed via
"Set atomic label type" in "Other settings" at menu bar
• If the rendering effect of the isosurface is not quite good, use "Set lighting" in "Other settings"
to adjust lightings.
• Select "Isosur. quality" in the menu bar, set to "high quality" (medium sized system) or "very
high quality" (large system).
• Click "Save picture". Use such as Irfanview or Photoshop program to open it, shrink the size
of the image file to 50% (in this process resample will be automatically done, making anti-aliasing
effect effectively realized), then properly crop the graph.
Below is an example obtained via above steps, the quality is pretty good
It is suggested to use "face+mesh" drawing style instead of the default "solid face", since in
this case the saved picture will be more stereoscopic.
347
4.0.2 Viewing natural bond orbitals (NBO) of ethanol
There are two ways to view NBOs, if you are a Gaussian user, way 2 may be more convenient,
however if you also need to view natural hybrid orbital (NHO) or natural atomic orbital (NAO) or
some other types of orbitals generated by NBO program, you have to use way 1.
Way 1: Using NBO plot files

The common way is to generate NBO plot files (.31~.40) and load them into Multiwfn. To
generate these files by Gaussian, you should add pop=nboread in route section, that means the
keywords of NBO at the end of input file will be passed to NBO module (Link 607 in Gaussian),
then add for example $NBO plot file=C:\NH2COH $END at the end of the input file with a blank
line before it, you can refer to the NH2COH_NBO.gjf in “example” directory. Run the input file by
Gaussian, you will find that NH2COH.31, NH2COH.32 ... NH2COH.41 have been generated in C:\
folder. The NH2COH.31 and NH2COH.37 have already been provided in “example” folder. Now
boot up Multiwfn and input following commands
examples\NH2COH.31 // .31 file contains necessary basis function information for plotting
examples\NH2COH.37 // .37 file contains NBO information. .32~.40 files correspond to
PNAO/NAO/PNHO/NHO/PNBO/NBO/PNLMO/NLMO/MO respectively. Hint: You can only
input 37, because in present example the .37 and the .31 file share the same name
0 // Enter the GUI
You can choose corresponding NBO orbital from right-bottom list to view the isosurface.
Multiwfn is also capable to plot two orbitals simultaneously, for instance, here we will plot NBO 12
and NBO 56, which correspond to occupied lone pair of nitrogen atom and unoccupied anti-π bond
between carbon and oxygen atoms respectively. Firstly, we choose 12 from the orbital list to plot
NBO 12, and then click "Show+Sel. isosur#2", after that we click 56 in the list, you will see both of
NBO 12 and NBO 56 are shown. The yellow-green and purple parts of NBO 56 (isosurface#2)
correspond to positive and negative parts, respectively.
348
It is somewhat difficult to study overlapping extent between the two orbitals from the solid
face graph, so we choose "Use mesh" in "Isosur#1 style" and the counterpart in "Isosur#2 style" to
make the two isosurfaces represented as mesh, see below. (Please also try "transparent face" style)
Now the overlapping extent become distinct, it is quite clear that NBO 12 substantially overlapped
with NBO 56, the resulting strong delocalization is one of the main reasons why the second-order
perturbation energy between them is very large (~60 kcal/mol). In Section 4.4.5, you will learn how
to obtain contour map for the two orbitals.
Notice that for unrestricted calculations, .32 and .33 files outputted by NBO 3.1 module in Gaussian are
incorrect -- the title parts are missing, which will lead to strange result, you should fix them by consulting other plot
files such as .34, it is very easy.
Regarding the ways to pass the keywords for generating NBO plot files to NBO module in
other quantum chemistry packages, please consult corresponding manual. You can also use stand-
alone version of NBO program (GENNBO) to generate NBO plot files, an input file (.47) is needed
to be prepared first. To generate it, you should load a file containing basis function information into
Multiwfn, then enter main function 100, select subfunction 2, then choose corresponding option to
export .47 file. After that, manually add plot keyword between “$NBO” and “$END” in the .47 file.
Then if you use GENNBO program to run the .47 file, you will get NBO plot files.
Way 2: Using .fch file as NBO information carrier

Gaussian provides a keyword pop=saveNBO, if you add it in your Gaussian input file, NBOs
will be saved to checkpoint file instead of MOs. You can use corresponding .fch file as Multiwfn
input file to view NBOs. If theoretical level of the task is HF or DFT, you should add saveNBOene
in the first line of the .fch file; if post-HF method is used and density keyword has also been specified,
you should add saveNBOocc in the first line of the .fch file, in this case Multiwfn will do some
special treatments internally. However, if your aim is just viewing NBOs in main function 0, you
can ignore this step.
Beware that when Gaussian storing the NBOs to checkpoint file, they may be automatically
reordered. For example, you may see the information like below in the Gaussian output file:
Reordering of NBOs for storage: 7 8 3 1 2 4 6 5 9 38 ...
That means the 1st, 2nd, 3rd, 4th ... orbitals in the .chk/.fch file in fact correspond to the 7th, 8th,
3rd, 1st ... NBOs generated by the NBO module, respectively.
It is worth to note that if you use Multiwfn in combination with VMD, you can plot very pretty
NBO isosurface maps, see my blog article "Using Multiwfn to plot NBO and related orbitals" (in
Chinese, http://sobereva.com/134) for detail. Below is a map plotted by a Multiwfn user in his work
J. Mol. Graph. Model., 59, 31 (2015).
349
4.0.3 Using Multiwfn+VMD to rapidly plot high-quality orbital

isosurface map
Note: Chinese version of this tutorial is http://sobereva.com/447 and http://sobereva.com/449.
Video illustration corresponding to this section is available at https://youtu.be/-3TXfdO8H7s, please never
forget to look at it!!!
Prologue
If one employs Multiwfn to export cube file for the orbitals of interest, and then render them
as isosurface map in VMD (http://www.ks.uiuc.edu/Research/vmd/), very ideal orbital isosurface
map can be obtained, the procedure has been detailedly described in my blog article "Using
Multiwfn to visualize molecular orbitals" (in Chinese, http://sobereva.com/269). However, the
procedure introduced in this article is somewhat lengthy, many manual operations are needed. In
order to simplify the procedure as much as possible, here I show how to use scripts to very easily
and quickly draw high-quality orbital isosurface map by combinely using Multiwfn and VMD. In
this section I only illustrate how to plot MOs, but the same procedure can also be applied for plotting
other kinds of orbitals, however you need to properly modify the inputstream file (see below). If
you do not known how to run Multiwfn in silent mode, I suggest you read Section 5.2 first so that
you can better understand this section. Here I assume you are using Windows system, for Linux
platform you should manually write corresponding script. The VMD program I used here is version
1.9.3.
Preparation work
Copying showorb.bat and showorb.txt from "examples\scripts" to the folder containing
Multiwfn executable file.
The showorb.bat is a Windows batch process file, it is used to invoke Multiwfn to calculate
grid data of wavefunction for selected orbitals and then move the exported cube files to VMD folder.
You should manually edit this file to make the input file path corresponds to the actual path of input
file, and then replace the VMD folder in this file with actual VMD folder in your machine.
The showorb.txt is inputstream file, each line corresponds a command needed to be inputted in
the Multiwfn interactive interface. You should manually set the third line as indices of the orbitals
you want to plot, for example, 10,20-23,28-30.
The showorb.vmd in "examples\scripts" is a VMD plotting script, you should copy it to VMD
folder, and then add source showorb.vmd to the end of the vmd.rc file in VMD folder, so that the
script will be automatically executed when VMD boots up. This script defines three customized
commands:
·orb i: Used to load cube file of orbital i and show it as isosurfaces. The default isovalue is
0.05, you can change it by editing showorb.vmd
350
·orbiso x: Used to change the isovalue to x.

·orbclean: Used to delete all orbital cube files in VMD folder.
Example
Here we plot MOs for examples\excit\D-pi-A.fchk. Make sure that all prepraration work has
done, then edit the showorb.bat, replace the default input file 1.fch with examples\excit\D-pi-A.fchk,
and ensure that the actual VMD folder has been properly specified in this file. Then open
showorb.txt, set the third line as 54-59, so that we can visualize MOs from 54 to 59. Then double-
click the showorb.bat, Multiwfn will be invoked to load input file, calculate and export grid data of
wavefunction for the selected orbital. For e.g. orbital 54, the exported file will be named
orb000054.cub. All the orbital cube files are then automatically moved to the VMD folder. After
that, boot up VMD, input orb 56 in VMD console window, then orb000056.cub will be loaded into
VMD and drawn as isosurfaces:
In above map, positive and negative phases are represented as red and blue colors, respectively.
"Glossy" material is used by default. If you want to change the color or material, you should enter
"Graphics" - "Representation" and modify corresponding options. You can also change default color
and material by modifying the showorb.vmd script.
If then you want to visualize another orbital, for example MO54, then simply input orb 54 in
the VMD console window.
If you want to change the isovalue to e.g. 0.02, simply input orbiso 0.02.
If after visualization, you want to clean all orbital cube files in the VMD folder, just input
orbclean, then all orb??????.cub files will be deleted.
By default, "medium quality grid" (about 512000 points) is used to calculate orbital
wavefunction, this is adequate for small and medium sized systems. However, for large systems,
such as those consisted of one hundered of atoms or more, you must employ higher number of grid
points. If you want to change the default grid to "high quality grid", you should set the fourth line
in showorb.txt to 3. In addition, as mentioned in Section 4.0.1, for visualizing Rydberg orbitals you
must increase extension distance of grid data. To do so, you should add
-10
12
between the third and fourth lines of showorb.txt, then extension distance will be increased from
default value to 12 Bohr.
Plot state-of-the-art orbital isosurface map

If you want to obtain even better quality of orbital isosurface graph, just follow below
procedure:
351
(1) Plot an orbital in VMD as described above

(2) Copy all content in examples\VMDrender.txt to VMD console window to modify the
plotting setting
(3) In VMD, select "File" - "Render" - "Tachyon", click "Start Rendering". Then vmdscene.dat
will appear in VMD folder, it is input file of the Tachyon render.
(4) Copy the examples\VMDrender_full.bat to VMD folder
(5) Double click the VMDrender_full.bat, then the Tachyon render (tachyon_WIN32.exe) in
VMD folder will be invoked to carry out render. After a while, full.bmp appears in the VMD folder,
it is the produced image file.
The rendered image of MO56 of examples\excit\D-pi-A.fchk is shown below, the graph looks
extremely good!
The rendering time is fairly long for large systems. For saving time, you can use the
VMDrender_noshadow.bat instead of the VMDrender_full.bat, in this case no shadow effect will be
observed in the resulting graph, while the rendering cost is correspondingly reduced.
Sometimes, especially for large system, the shadows casted by the transparent orbital
isosurfaces make the graph look too dark, you can manually add -shadow_filter_off argument in
the .bat file to disable this kind of shadow during rendering.
4.1 Calculate properties at a point
4.1.1 Show all properties of triplet water at a given point
In this example I illustrate how to calculate a wide variety of real space functions at a given
point for triplet water. Boot up Multiwfn and input below commands
examples\H2O_m3ub3lyp.wfn
1 // Main function function 1, show properties at a point
0.2,2.1,2 // X, Y, Z coordinate of the point
1 // The unit of inputted coordinate is Bohr
Now all real space functions supported by Multiwfn at this point are printed along with
components of electron density gradient/Laplacian, Hessian matrix and its eigenvalues/eigenvectors.
If you are unable to fully understand the output, please read Sections 2.6 and 2.7 carefully, all terms
in the output are very detailed described.
Density of all electrons: 0.4598301528E-02
352
Density of Alpha electrons: 0.2861566387E-02
Density of Beta electrons: 0.1736735141E-02
Spin density of electrons: 0.1124831246E-02
Lagrangian kinetic energy G(r): 0.3365319167E-02
Hamiltonian kinetic energy K(r): 0.1088761528E-03
Potential energy density V(r): -0.3474195320E-02
Energy density: -0.1088761528E-03
Laplacian of electron density: 0.1302577206E-01
Electron localization function (ELF): 0.1998328717E+00
Localized orbital locator (LOL): 0.1008002781E+00
Local information entropy: 0.3533635333E-02
Reduced density gradient (RDG): 0.2033111359E+01
Reduced density gradient with promolecular approximation: 0.2294831921E+01
Sign(lambda2)*rho: -0.4598301528E-02
Sign(lambda2)*rho with promolecular approximation: -0.3918852312E-02
Corr. hole for alpha, ref.: 0.00000 0.00000 0.00000 : -0.1251859403E-03
Source function, ref.: 0.00000 0.00000 0.00000 : -0.3565867942E-03
Wavefunction value for orbital 1 : 0.1536978161E-03
Average local ionization energy: 0.4664637535E+00
User-defined real space function: 0.1000000000E+01
ESP from nuclear charges: 0.3453377860E+01
ESP from electrons: -0.3439063818E+01
Total ESP: 0.1431404144E-01 a.u. ( 0.3895049E+00 eV, 0.8982204E+01 kcal/mol)
Note: Below information are for electron density
Components of gradient in x/y/z are:
-0.7919856828E-03 -0.6903543769E-02 -0.6651181972E-02
Norm of gradient is: 0.9618959378E-02
Components of Laplacian in x/y/z are:
-0.3809549089E-02 0.1052804857E-01 0.6307272576E-02
Total: 0.1302577206E-01
Hessian matrix:
-0.3809549089E-02 0.1394394193E-02 0.1197923973E-02
0.1394394193E-02 0.1052804857E-01 0.1008387143E-01
0.1197923973E-02 0.1008387143E-01 0.6307272576E-02
Eigenvalues of Hessian: -0.3959672207E-02 0.1886890004E-01 -0.1883455778E-02
Eigenvectors(columns) of Hessian:
0.9964397434E+00 0.8076452238E-01 0.2418531975E-01
-0.4735415689E-01 0.7734993430E+00 -0.6320255930E+00
-0.6975257409E-01 0.6286301443E+00 0.7745700228E+00
Determinant of Hessian: 0.1407217564D-06
353
Ellipticity of electron density: 1.102344
All data are expressed in scientific notation, the value behind E is exponent, e.g.
0.6307272576E-02 corresponds to 0.006307272576.
In the line of "Corr. hole (correlation hole)" and "Source function", the so-called "ref" is the
position of reference point, which is determined by "refxyz" parameter in settings.ini.
By default, the outputted wavefunction value corresponds to orbital 1, you can input for
example o6 to choose orbital 6.
By default, the components of gradient and Laplacian as well as Hessian and its
eigenvalue/eigenvectors are for electron density. You can input such as f10 to choose the real space
function with index of 10 (namely ELF), after that all of these quantities will be for ELF. If you
want to inquire indices of all available real space functions, input allf.
You can continue to input other coordinates, when you want to return to upper level menu,
input q; If you want to exit the program, press “CTRL+C” button or directly close commond-line
window.
4.1.2 Calculate ESP at nuclear positions to evaluate interaction strength

of H2O∙∙∙HF
This is an advanced example, if you are not interested in weak interactions you may skip this section.
In J. Phys. Chem. A, 118, 1697 (2014), Mohan and Suresh studied a batch of electrostatic
dominated interacting systems, including hydrogen, halogen and dihydrogen bonds, all of them
belong to electron donor-acceptor interactions, where donor stands for electron-rich moiety (Lewis
base), while acceptor is electron-deficient moiety (Lewis acid). They fitted a surpringly good linear
equation to correlate Vn index with interaction energy (Enb) for all kinds of interactions, the R2 is
as high as 0.9762. Their results can be summarized as following graph
For an electrostatic dominated complex, assume that we can obtain Vn, then according to
the equation shown in above graph, we can easily predict the interaction energies as
354
Enb = −89.2857  Vn − 0.125

The Vn is defined based on ESP at nuclear positions
Vn = Vn −D − Vn −A = (Vn −D' − Vn −D ) − (Vn −A' − Vn −A )

where Vn-D' is the ESP at nuclear position of donor atom in complex environment, but the
contribution due to nucleus of this donor atom is ignored. The only difference between Vn-D and Vn-
D'is that the former is calculated in monomer state, therefore Vn-D = Vn-D' - Vn-D can be regarded as
the change in ESP at nuclear position of donor atom due to presence of another molecule, which
directly reflects strength of intermolecular interaction. The definition of Vn-A' and Vn-A are identical
to Vn-D' and Vn-D, respectively, but they are calculated for acceptor atom.
In this example, we calculate Vn for H2O∙∙∙HF and check if the interaction energy predicted
based on Vn is really closed to the accurately calculated interaction energy. In this complex the
oxygen of H2O is electron donor atom and hydrogen of HF is electron acceptor atom. Because the
equation presented by Mohan and Suresh was fitted for specific calculation level, in order to
properly use their equation, the calculation level we employed here is identical to them. The .wfn
files used below were produced at MP4(SDQ)/aug-cc-pVTZ level at MP2/6-311++G** optimized
geometries, these .wfn files and the corresponding Gaussian input files can be found in
"examples\Vn" folder.
Note that if you are using relatively old revision of G09 and post-HF method is employed, "density" keyword
is indispensable, otherwise the density in the resultant .wfn file will correspond to Hartree-Fock density. Besides, in
G09 and G16, density cannot be produced at MP4 level, so we use MP4(SDQ) keyword instead (MP4 keyword is
default to MP4(SDTQ), which is more accurate and but much expensive than MP4(SDQ)).
First we calculate Vn-A' and Vn-D'. Boot up Multiwfn and input
examples\Vn\H2O-HF.wfn
1 // Calculate properties at a point
a1 // Nuclear position of atom 1
From the output you can see
Total ESP without contribution from nuclear charge of atom 1:
-0.2228775074E+02 a.u. ( -0.6064805E+03 eV, -0.1398579E+05 kcal/mol)
That means Vn-D' is -22.2877 a.u. Then input a5, you will find Vn-A' is -0.9608 a.u.
Next we calculate Vn-D. Reboot up Multiwfn and input below commands
?H2O.wfn // The symbol ? means the folder of the file we last time loaded
1
a1 // In H2O.wfn oxygen is atom 1
We find Vn-D is -22.3339 a.u. Then we calculate Vn-A. Reboot Multiwfn and input
?HF.wfn
1
a2 // In HF.wfn hydrogen is atom 2
The Vn-A is found to be -0.9136 a.u.
The Vn is thus -22.2877-(-22.3339) - [-0.9608-(-0.9136)] = 0.0462 + 0.0472 = 0.0933 a.u.
Using the equation mentioned earlier, the interaction energy can be approximately predicted as
-89.28570.0933-0.125 = -8.45 kcal/mol, this value is quite close to the accurate interaction energies
(-8.31 kcal/mol) obtained by Mohan and Suresh at MP4/aug-cc-pVTZ level with Counterpoise
correction.
355
Generating wavefunction at MP4(SDQ)/aug-cc-pVTZ is quite time consuming even for small complex such as
the system we studied here, thus it is important to find a calculation level that significantly saves computational time
but without too much sacrifice in accuracy. For present system, based on the MP2/6-311++G** geometry, I tried
using several levels to evaluate the Vn:
B3LYP/6-311+G**: 0.1021 a.u.
MP2/cc-pVTZ: 0.1052 a.u.
MP2/aug-cc-pVTZ: 0.0955 a.u.
B3LYP/aug-cc-pVTZ: 0.0985 a.u.
MP2/aug-cc-pVDZ: 0.0939 a.u.
B3LYP/aug-cc-pVDZ: 0.0980 a.u.
The Vn produced at MP2/aug-cc-pVDZ (0.0939) is very close to the value we obtained above at
MP4(SDQ)/aug-cc-pVTZ (0.0933), while the computational cost is reduced by factors of two. So, in practical studies,
using MP2/aug-cc-pVDZ level to evaluate Vn is a very ideal choice.
4.2 Topology analysis
Multiwfn is able to perform topology analysis for any real space functions, such as electron
density, its Laplacian function, ELF, LOL, orbital wavefunctions, spin density, electrostatic potential
and so on. Four kinds of critical points (CPs) can be located and real space function values at these
points can be easily obtained; topology paths linking CPs and interbasin surfaces can be generated.
There are also many additional capacities, see Section 3.14 for details. Below I will present some
practical applications to illustrate how to use this powerful module.
4.2.1 Atoms in molecules (AIM) and aromaticity analysis for 2-

pyridoxine 2-aminopyridine
Topology analysis of electron density is a main ingredient of Bader's atoms in molecules (AIM)
theory. In this example we will perform this kind of analysis for 2-pyridoxine 2-aminopyridine
complex.

examples\2-pyridoxine_2-aminopyridine.wfn // Assume that the input file is in a subdirectory
of current directory, we can only input relative path rather than entire absolute path
Then we search all critical points (CPs) by inputting below commands
2 // Use nuclear positions as initial guesses, generally used to search (3,-3) CPs
3 // Use midpoint of each atom pair in turn as initial guesses. Generally all (3,-1) CPs could
be found, some (3,+1) or (3,+3) may also be found at the same time
356
The search of CPs is very fast. After that input 0, the positions and types of all found CPs will
be printed in the command-line window, at the end of the output the number of each type of CPs is
shown:
(3,-3): 25, (3,-1): 27, (3,+1): 3, (3,+3): 0
25 - 27 + 3 - 0 = 1
The second line shows that Poincaré-Hopf relationship has been satisfied, that means all CPs may
have been found. If this relationship is unsatisfied, then some CPs must be missing. From the GUI
that popped up, we can see all expected CPs are presented, hence we can confirm that all CPs have
been found.
Click "RETURN" button in the GUI window and input 8 to generate topology paths (in present
context they correspond to "bond paths"), select option 0 again to view CPs and paths. After slight
adjustment of plot settings, the graph looks like below (Click "CP labels" at the right side of the
GUI if CPs indices are not shown):
Magenta, orange and yellow spheres correspond to (3,-3), (3,-1) and (3,+1) critical points,
brown lines denote bond paths. The indices of CPs are labelled by cyan texts. It can be seen that
indices 53 and 38 correspond to the bond critical point (BCP) of N-HO and N-HN hydrogen
bonds (H-bonds), respectively. It is worth to note that the label color of the CPs can be changed via
“Set label color” option in the menu bar.
Now close the GUI window.
The topology analysis module provides many analysis options, for example, let us measure the
distance between CP30 and the nuclear position of H25. Select option -9, then input c30 a25, the
result is 5.877601 Å. Then we measure the angle between C14-N13-H12, namely input a14 a13
a12, the result is 120.297432 degree. Now, we input q to return.
Evaluate H-bond binding energy

J. Comput. Chem., 40, 2868 (2019) DOI: 10.1002/jcc.26068 is a very important paper about
H-bond, a broad range of H-bond systems were subjected to thorough investigation and deep
analysis. In this work I and my collaborators proposed two extremely useful and important equations
for predicting H-bond binding energy (BE) based on electron density at BCP corresponding to H-
bond.
𝐵𝐸 ≈ −223.08 × 𝜌(𝐫BCP ) + 0.7423 (for neutral H−bond)
𝐵𝐸 ≈ −332.34 × 𝜌(𝐫BCP ) − 1.0661 (for charged H−bond)
357
where the unit of  is a.u., the unit of BE is kcal/mol. It was shown that these formulae are not only
reliable but also universal. The first equation is suitable for present complexes, here we employ it
to predict the BE of the N23-H25O1 H-bond in our system.
Choose option 7 and then input the index of corresponding BCP, namely 53, you will see values
of a lot of real space functions at this point are shown:
CP Position: 0.44887255865472 3.56434324597741 -0.10652884364257
CP type: (3,-1)
Density of all electrons: 0.3129478049E-01
Density of Alpha electrons: 0.1564739024E-01
Density of Beta electrons: 0.1564739024E-01
Spin density of electrons: 0.0000000000E+00
Lagrangian kinetic energy G(r): 0.2530207716E-01
Hamiltonian kinetic energy K(r): 0.8463666362E-03
Potential energy density V(r): -0.2614844379E-01
Energy density: -0.8463666362E-03
Laplacian of electron density: 0.9782284209E-01
Electron localization function (ELF): 0.1105388527E+00
... (Ignored)
The output indicates that the (r) at this BCP is 0.03129 a.u., therefore the H-bond BE could be
evaluated as BE = -223.08*0.03129+0.7423 = -6.2 kcal/mol = -26.1 kJ/mol.
It is also worth to note that in Chem. Phys. Lett., 285, 170 (1998), Espinosa and coworkers
stated that for H-bond of X-HO (X=C,N,O) type, the BE could be estimated as
BE=V(rBCP)/2
As shown in above output, V(r) at the BCP of N23-H25O1 is -0.026148, thus the BE could be
predicted to be -0.026148/2*2625.5 = -34.3 kJ/mol, which notably differs from the -26 kJ/mol using
the prediction equation proposed in the J. Comput. Chem. (2019) paper. Which one is more accurate?
As rigorously demonstrated in the J. Comput. Chem. article, the popular V(rBCP)/2 equation in fact
has an evidently larger error and thus cannot be recommended; in other words, the BE of -26.1
kJ/mol should be more reliable.
Evaluating aromaticity based on CP properties

In this part we use two special options in the topology analysis module to evaluate aromaticity.
If you are not interested in this topic you can skip.
First we use information entropy method to examine if the aminopyridine (the monomer at the
left side of above graph) in the dimer can be regarded as an aromatic molecule. This method was
proposed in Phys. Chem. Chem. Phys., 12, 4742 (2010) and is based on electron density at BCPs of
a ring, see Section 3.14.6 for detail. First, we select option 20, and then input indices of the BCPs
in the ring, namely 44,42,32,29,31,40, the outputted Shannon aromaticity index (SA) is 0.000812.
The smaller the SA index, the more aromatic is the ring. In origin paper, 0.003 < SA < 0.005 is
chosen as the boundary of aromaticity/antiaromaticity. Since our result is much smaller than 0.003,
we can conclude that aminopyridine is an aromatic molecule. The SA for 2-pyridoxine (the
monomer at the right side of above graph) is 0.000865, hence shows slightly weaker aromaticity
than aminopyridine.
Next, we calculate the curvature of electron density perpendicular to ring plane at RCP. In Can.
J. Chem., 75, 1174 (1997), it was shown that more negative curvature implies stronger aromaticity.
358
We first calculate the curvature for aminopyridine. Select option 21, input the index of the RCP
(namely 36), and then input three atoms to define the ring plane, here we input 15,13,17 (also, you
can input such as 14,16,18 and 13,15,18, because all of them correspond to the same plane). From
the output we find the curvature is -0.0187. Then we calculate the curvature for 2-pyridoxine,
namely input 41 and 2,7,4, the result is -0.0164. Comparison between the two curvatures again
shows that aminopyridine has stronger aromaticity. In Multiwfn, aromaticity can also be measured
in many other schemes, such as HOMA, FLU, PDI, ELF-π and multicenter bond order, they are
collectively discussed in Section 4.A.3.
Generating interbasin surfaces

Interbasin surfaces (IBS) dissect the whole molecular space into individual basins, each IBS
actually is a bunch of gradient paths derived from a (3,-1) CP. Now we generate IBS corresponding
to the (3,-1) with index of 53, 38 and 37. Choose function 10, and input
53 // Generate the IBS corresponding to the (3,-1) CP with index of 33, the same as below.
You may need to wait a few seconds for each generation of IBS
38
37
q // Return
Visualize the results by choose function 0, the graph will be shown as below. The three surfaces
are IBS.
In Section 4.20.1, we will use another important weak interaction analysis method NCI to
further study this system.
You may feel that the current Multiwfn GUI for showing CPs and topology paths is somewhat
difficult to use for large system, since the system cannot be rotated completely smoothly, and
sometimes index of interesting CP is difficult to be observed. In Section 4.2.5 I will introduce how
to use the powerful VMD program based on Multiwfn outputs to very easily plot CPs and topology
paths, in this case the graph is very pretty, the perspective is completely controllable, and index of
interesting CPs can be easily found out.
There are two important points regarding AIM topology analysis I would like to mention here,
359
though they are not related to present example.
What should I do if some CPs of electron density are missing?

For small systems, commonly we can check whether all CPs have been located by simply enter
the GUI and visualize the distributions of the CPs. There is also an useful equation named Poincaré-
Hopf relationship. For isolate system, the relationship is
nNCP − nBCP + nRCP − nCCP = 1

If all CPs have been found, this relationship must be satisfied, but the satisfication of this
relationship does not necessarily mean all CPs have been found. If the Poincaré-Hopf relationship
is unsatisfied, then some CPs must be missing.
Sometimes, you may find some expected CPs are not successfully located after searches. There
are two reasons may cause this problem: (1) The position of initial guesses are not close enough to
the CPs (2) The default CP searching parameters are not well-suited for present case. There are some
commonly used ways to solve this problem. More detailed descriptions may be found in Section
3.14.2.
a) If you have tried options 2~5 and some CPs are not located, try to use suboption -1 of option
6. This searching mode is powerful but expensive, which by default places 1000 guessing points
within in a spherical region around every atom. If after repeating this mode several times the missing
CPs are still unable to be located, it is highly possible that the reason is due to the inappropriate
searching parameters rather than the positions of guessing points.
b) If some BCPs are unable to be located, you can enter option -1, set the scale factor of stepsize
to 0.5, and then try again
c) NCPs of very heavy atoms are difficult to be located, because the peak of electron density
at nucleus in these cases are very sharp, thus under default parameters the searching algorithm is
difficult to capture the NCPs. In order to locate them, you can enter option -1, loose the criteria for
gradient-norm and displacement convergences by several factors, and then try to use option to 2
search the NCPs again. If the NCPs are then successfully found, do not forget to recover the
convergence criteria. In fact, since NCP of heavy atoms are almost exactly located at nuelcus
position, you can directly enter option -4 and choose suboption 3 to add NCPs artificially at the
corresponding nuclear positions.
d) If some missing CPs are far away from atoms, for example, the CCP at the center of a very
large cage or tube system, try to enter option -1 and decrease the criterion for determining singularity
of Hessian matrix via option 8 by several factor, and then search the CPs again.
Describing electron density of very heavy atoms

Very heavy atoms (heavier than Kr) bring much more computational burden than light atoms,
and relativistic effect is non-neglectable. There are two different ways to describe them.
• Using pseudo-potential (PS): As mentioned in Section 2.5, if PS is employed but the .wfx file
produced by Gaussian is used as input file, the EDF (electron density functions) field in the .wfx
file by default will be loaded into Multiwfn, which represents the inner-core electron density. For
other type of input files, such as .mwfn, .wfn, .fch, .molden and .gms, by default Multiwfn
automatically loads proper EDF information from built-in EDF library.
When EDF information is provided, all CPs of electron density can be properly located and
artifical CPs will never occur, the bond paths emitting from BCPs can connect to NCPs normally,
all CP properties that solely based on electron density will be reasonable. Although large core PS
360
can be employed without problems in this situation, I still recommend using small core PS, because
the accuracy of resultant CP positions and properties must be better than using large core PS.
Lanl2, cc-pVnZ-PP,
Lanl1 SDD SBKJC
Lan2TZ/08 def2 series
Main groups Large Large L/S (optional) Small Large
Transition metals Large Small L/S (optional) Small Small
If you decide not to utilize EDF information (see "readEDF" and "isupplyEDF" in settings.ini
for detail), evidently it is impossible to find out (3,-3) CP at nuclear position, and accordingly, the
bond paths emitted from BCP will be unable to connect to the nucleus. Instead, you may find (3,+3)
at nucleus position due to the vacancy of inner-core density, and a lot of CPs in different types will
appear around the nucleus, this is because the electron density no longer decreases exponentially
from nucleus, so the topology structure of electron density becomes quite complicated. However,
you can simply ignore those irrelvant CPs but only focus on the BCPs that you are really interested
in.
• Using all-electron basis set with relativistic Hamiltonian: This is the most expensive but most
accurate solution for representing electron structure of heavy atoms. Only considering scalar
relativistic effect is totally enough for AIM analysis. DKH2 Hamiltonian is a very good choice (in
Gaussian, simply using int=DKH2 keyword to employ it, beware that basis set optimized for DK
calculation must be used, e.g. cc-pVDZ-DK).
For more discussions, please consult my post "Some explaination of performing wavefunction
analysis under pseudo-potentials" (in Chinese, http://sobereva.com/156).
4.2.2 LOL topology analysis for acetic acid
A brief of localized orbital locator (LOL) has been given in Section 2.6. In this example, we
will locate CPs and generate topology paths of LOL for acetic acid. With completely identical
procedure, you can also study topology character for electron localization function (ELF) and
Laplacian of electron density.
examples\acetic_acid.wfn
2 // Enter topology analysis module
-11 // Select a real space function
10 // Localized orbital locator (LOL)
Notice that the distribution feature of LOL is much more complex than electron density, so it
is very difficult to locate all of its CPs. Fortunately, in general only a small subset of CPs is what
we are interested in, the search of CPs can be aborted once all interesting CPs have been found.
We first use option "2 Search CPs from nuclear positions" to locate the CPs that very close to
nuclei. However the positions of CPs in other regions are somewhat unpredictable, hence a lot of
initial guessing points have to be randomly scattered around each atom to try to locate those CPs.
Now, input below commands::
6 // In this searching mode, initial guessing points will be randomly scattered in a sphere, the
sphere center, radius, number of points and so on can be defined by users. This time we leave the
default value unchanged
361
-1 // Use each nucleus as sphere center in turn to search CPs. Since there are 8 atoms, and
guessing points in each sphere is 1000, Multiwfn will try to search CPs based on 8*1000 guessing
points. Of course, the more the guessing points you set, the larger probability that all CPs could be
found in this search
-9 // Return to upper menu
0 // Visualize the result
From above map it can be seen that the number of CPs in function space of LOL is very large.
Actually there are still some CPs have not been found in the search. If you repeat the search one
more time or several times, some missing CPs could be found. Since all interesting CPs have been
found currently, repeating the search is unneccessary. In the graph, each purple sphere signifies a
(3,-3) type of CP, which represents locally maximal electron localization. It can be seen that CP15
delineates the covalent bond between the two carbons. CP8 and CP9 correspond to the two C-O
bonds. CPs 7, 57, 12 and 13 correspond to lone pairs of oxygens.
Now choose option 8 to generate the topology paths linking (3,-1) and (3,-3) CPs, then choose
option 0 to visualize the result again. To relieve visual burden, some uninteresting objects can be
screened by properly adjusting corresponding GUI widgets. As you can see from the below graph,
the topology paths clarify the intrinsic connectivity between CPs.
362
4.2.3 Plot real space function along bond path
All real space functions that supported by Multiwfn could be easily plotted along topology
paths. In this example we plot ellipticity of electron density along bond path of boundary C-C bond
of butadiene.
First open settings.ini file and change "iuserfunc" parameter to 30, because the 30th user-
defined function corresponds to electron density ellipticity, see Section 2.7 for detail.
Then boot up Multiwfn and input below commands:
examples\butadiene.fch
2 // Search nuclear critical points from nuclear positions
3 // Search bond critical points from midpoint of atom pairs
8 // Generate bond path
0 // Enter GUI window to visualize result
Clicking "Atom labels" and "Path labels" buttons at right side of the GUI window, then we can
find paths 5 and 6 collectively constitute the bond path of a boundary C-C bond:
363
Clicking "RETURN" to close the window and then input

-5 // Various operations on paths
7 // Calculate and plot specific real space funtion along a path
5,6 // The index of the paths (in fact, you can also equivalently input c13 here)
100 // User-defined function, which corresponds to ellipticity of electron density currently
The curve of electron density ellipticity along the boundary C-C bond path immediately shows
on the screen, the dashed line denotes the position of bond critical point. In the plot, the left and
right corner correspond to CP3 and CP4, respectively. At the same time, the raw data of the curve
are shown on the command-line window and you can copy them out, so that the map can be further
analyzed or reploted in third-part plotting tools such as Origin.
From the graph it is clear that the electron density ellipticity is positive in the middle region of
the bond path, exhibiting the double-bond character of the C-C bond.
You can also plot other real space functions such as ELF and kinetic energy density along bond
paths, please have a try.
4.2.4 Decompose properties at a critical point as orbital contributions
Many real space functions can be exactly or approximately decomposed as orbital

contributions. If occupation number of all orbitals except for an orbital i is set to zero, then the
calculated function value just corresponds to contribution of orbital i.
364
Multiwfn is able to decompose any real space function as orbital contributions at any point,
this feature is supported by both main function 1 and main function 2; in the former the position of
the point can be directly inputted by user, while in the latter the point can be selected as one of found
CPs. In this section I will use 1,3-butadiene as example to illustrate this feature.
We will first check which MOs have evident contribution to BCP corresponding to the
boundary C-C bonds. Boot up Multiwfn and input
2 // Search nuclear critical points
3 // Search BCPs
0 // Visualize CPs
As can be seen, CP13 and CP17 are the BCPs of boundary C-C bonds. Close the GUI and then
input
7 // Show properties at a CP
13d // Decompose properties of CP13
1 // The real space function to be decomposed is electron density
[Press ENTER button] // Take all occupied orbitals into account, but only print ten orbitals
having largest contributions
You will see below output
Contribution from orbital 11 (occ= 2.000000): 0.107469 a.u. ( 31.34% )
Sum of above values: 0.34286855 a.u. ( 100.00% )
Exact value: 0.34286855 a.u.
Clearly, electron density at this BCP is simultaneously contributed by many MOs, the largest
contribution is 31.3%. What will happen if we transform the MOs to localized MOs (LMO)? (see
Section 3.21 for introduction about orbital localization and LMO). To examine this, return to main
365
function menu, then input

19 // Orbital localization
1 // Only localize occupied orbitals
2 // Enter topology analysis function again. We do not need to redo topology analysis, since
all topology information are retained when you exit topology analysis module
7 // Show properties at a CP
13d // Decompose properties of CP13
1 // The real space function to be decomposed is electron density
[Press ENTER button]
You will see
...[ignored]
Sum of above values: 0.34286855 a.u. ( 100.00% )
As can be seen, currently only one orbital, namely LMO11 has remarkable contribution to the
BCP. This is what we expected, since in the LMO framework, each chemical bond is commonly
mainly represented by only one or very few number of LMOs. You can visualize the LMO11 using
main function 0:
It is clear that LMO11 fully corresponds to the boundary C-C -bond, this is why properties such
as electron density of the corresponding BCP is solely dominated by LMO11.
Next, let us check which LMOs have nonnegligible contribution to the point above 1 Bohr of
the BCP corresponding to the boundary C-C bond. The index of the BCP is 13, when you choose
option 0, you can find its coordinate in console window:
Index XYZ Coordinate (Bohr) Type
...[ignored]
11 -2.227945085 -3.985434919 0.000000000 (3,-1)
12 -0.041512516 -3.979859852 0.000000000 (3,-1)
13 -1.131478160 -2.047305989 0.000000000 (3,-1)
...[ignored]
Evidently, the point above 1 Bohr of CP13 should be (-1.131,-2.047,1.0). Input below command
-10 // Return to main menu
1 // Print various properties at a given point
d // Decompose to orbital contributions
-1.131,-2.047,1.0
366
1 // The unit of inputted coordinate is Bohr

1 // Decompose electron density
Then you will see below information
...[ignored]
It can be seen that, not only the LMO11, but also LMO15 has evident contribution. If you
inspect this orbital in main function 0, you will find it corresponds to  bond of the boundary C-C
bond.
Via similar way, we then decompose electron density of the point above 1 Bohr of the BCP
corresponding to the middle C-C bond, the result is
...[others are not shown due to negligible contribution]
Isosurface graph of the three LMOs are shown below.
Although both the LMO 14 and LMO 15 mainly correspond to  bond of the marginal C-C bonds,
they have 10.7% contribution to the position above 1Å of the middle C-C bond, implying that the
middle C-C bond must also have  character, though much weaker than the boundary C-C bonds.
4.2.5 Easily plot high quality AIM topology map in VMD visualization
program based on Multiwfn outputs
Note 1: Chinese version of this tutorial is http://sobereva.com/445.
Note 2: There is a video illustration corresponding to this section: https://youtu.be/mgsnhvWH5SI. I strongly
suggest looking at it!!!!!!!!!!!!
In Section 4.2.1 I have shown how to carry out AIM topology analysis for 2-pyridoxine 2-
aminopyridine complex, in present section I will show how to very easily render the located CPs
367
and generated topology paths in the very powerful and freely available VMD program
(http://www.ks.uiuc.edu/Research/vmd/). The map is very pretty, and you can easily find index of
interesting CPs. The whole process is highly automatic, because it is based on pre-provided
Windows batch process file and VMD plotting script. If you do not know how to run Multiwfn in
silent mode, I strongly you read Section 5.2 first, so that you can fully understand how the batch
process file works. If you manually write a similar shell script, this method can also be realized
under Linux platform.
The script files

We need to do a few preparation works first. Copy AIM.bat and AIM.txt from
"examples\scripts" to the folder containing Multiwfn executable file. Edit the AIM.bat, modify the
default VMD folder to actual VMD folder on your machine. Then copy the VMD plotting script
AIM.vmd to VMD folder, and add proc aim {} {source AIM.vmd} to the end of the vmd.rc file in
VMD folder so that then you can activate this script by simply inputting command aim.
The AIM.txt is an input stream file for running Multiwfn in silent mode, it does below things:
(1) Carrying out standard AIM analysis (locating CPs via options 2,3,4,5 in succession and generating topology
paths)
(2) Exporting the CPs and paths as CPs.pdb and paths.pdb in current folder
(3) Calculate all properties except for ESP at all located CPs and export the result to CPprop.txt in current folder
(4) Exporting the structure of present system as mol.pdb in current folder
If you want to reduce calculation cost of VMD.bat, you can remove the fourth and fifth lines of VMD.txt, in
this case Multiwfn will only try to locate CPs starting from nuclear positions and midpoints of atom pairs. However,
this treatment may cause missing of some CPs.
Example: 2-pyridoxine_2-aminopyridine
In this example, we copy the examples\2-pyridoxine_2-aminopyridine.wfn to the folder
containing Multiwfn executable file, modify the input file name in the VMD.bat as 2-pyridoxine_2-
aminopyridine.wfn, then double-click the VMD.bat, this batch process file will invoke Multiwfn to
carry out analysis according to the commands in VMD.txt for this .wfn file. After a while, CPprop.txt
is generated in this folder, the meantime generated mol.pdb, CPs.pdb and paths.pdb are
automatically moved to the VMD folder.
Now boot up VMD and input command aim in VMD console window, you will immediately
see below graph.
Notice that in order to gain slightly better effect, I used the bulit-in Tachyon render to obtain below graph,
namely selecting "File" - "Render", change to "Tachyon (internal, in-memory rendering)" and click "Start Rendering"
button (The resulting file is in .tga format, you need to use advanced image viewer to view it, such as IrfanView,
which is freely available at https://www.irfanview.com)
It is also possible to simultaneously show molecular structure on the map. Double click the
"D" shown in below screenshot to make it become black:
368
Then the molecular structure will be visible:
Show labels for CPs

The CP indices can be labelled on the graph easily. The AIM.vmd script defined a command
named labcp to do this, the usage is:
labcp [type] [label size] [offset in X] [offset in Y]
The “type” could be “all”, “no”, “3n3”, “3n1”, “3p1”, “3p3”. The “offset in X” and “offset in Y”
are used to define position offset of the labels, they are default to -0.1 and 0.0, respectively.
Examples:
labcp all: Labelling index of all CPs
labcp no: Removing all CP labels
labcp 3n1 1.3: Labelling all (3,-1) CPs with size of 1.3
labcp 3p3 1.8 -0.05 0.1: Labelling all (3,+3) CPs with size of 1.8, the position offset in X and
Y are set to -0.05 and 0.1, respectively
For present system, we input labcp 3n1 1.5 -0.1 0.1 in VMD console window, you will see
Since the labels are "1-index", namely the index starts from 1, therefore a CP with label of 38
just corresponds to the CP 38 in the CPprop.txt. Clearly you can easily examine properties of various
CPs by checking corresponding entry in the CPprop.txt.
Show specific CPs

If you want to hide some type of CPs, you should enter "Graphics" - "Representations", and
369
then switch "Selected Molecule" to "CPs.pdb", as shown below.
Since in the CPs.pdb, the C, N, O, F atoms are used to represent (3,-3), (3,-1), (3,+1), (3,+3)
CPs, respectively, if you want to hide (3,+1) and (3,+3) in the map, you can double click the entries
of "name O" and "name F" to make them invisible.
In addition, for example, among all (3,-1) CPs you only want to show those with index of
3,5,6,7,11, you can select the "name N" entry, and then input name N and serial 3 5 to 7 11 in the
"Selected Atoms" box and then press ENTER button. While if you want to make those CPs hidden,
you should input name N and not {serial 3 5 to 7 11}. Clearly, the selection in VMD is very flexible.
Show specific topology paths

It is also possible to selectively display topology paths. As you can find from the content of
paths.pdb, each path corresponds to a unique residue index, so you can use "resid" property in VMD
to choose which paths are visible or invisible. For example, if you want to hidden topology paths 3
and 7, you should enter "Graphics" - "Representations" panel, choose "paths.pdb" from "Selected
Molecule" drop-down box, then input not resid 3 7 in the "Selected Atoms" box.
If you want to inquire index of a topology path, you should active VMD graphical window,
press button "0" to enter query mode, then click a path in the graphical window, then the "resid:"
shown in the console window is just the path index. If you want to return to rotation model, you
should press button "r".
4.2.6 Topology analysis in special regions: G-C...G-C base pair as an

example
Owing to the extreme flexibility of Multiwfn, it is possible to perform topology analysis only
in some special regions that you are really interested in. In this section, the below G-C...G-C base
pair will be used as instance. The system can be regarded as consisting of four fragments. Its .wfn
file can be downloaded at http://sobereva.com/multiwfn/extrafiles/GCGC.zip.
370
(1) Perform AIM analysis only in weakly interacting regions

First I show how to only locate CPs and generate bond paths in weakly interacting regions,
where electron density must be relatively low. Boot up Multiwfn and input
GCGC.wfn
2 // Search CPs from nuclear positions
-1 // Set CP searching parameters
9 // Set value range for reserving CPs:
0,0.1 // Only CPs with density within 0~0.1 a.u. (i.e. relatively low density) will be reserved
during searching
0 // Return
3 // Search CPs from midpoint of atom pairs
8 // Generating the paths connecting (3,-3) and (3,-1) CPs
Now choose option 0 to visualize the result, see below. The left and right graphs are actually
the same, but molecule structure is hidden in the right graph. It is clear that only BCPs as well as
accompanying bond paths corresponding to weak interactions have been generated, while BCPs and
bond paths corresponding to chemical bonds are not obtained.
If you are not interested in RCPs (ring CPs, yellow spheres) and CCPs (cage CPs, green
spheres), you can easily delete them. Input below commands:
-4 // Modify or export CPs
2 // Delete some CPs
5 // Delete all (3,+1) CPs
6 // Delete all (3,+3) CPs
0 // Return
371
0 // Return
Then if you choose option 0 again to visualize the result, you will find RCPs and CCPs have
disappeared.
(2) Perform AIM analysis only in local spatial region

Now I illustrate how to perform AIM topology analysis only for fragments 1 and 2. Boot up
Multiwfn and input
GCGC.wfn
-1 // Set CP searching parameters
10 // Set the range of atoms considered for searching modes 2, 3, 4, 5
1-29 // The atomic index range of fragments 1 and 2
2 // Set scale factor of stepsize. We modify this because under default setting some BCPs
corresponding to C-H and N-H of present system may be missing
0.5 // Stepsize will scaled by 0.5
0 // Return
Choose option 0 to visualize result, you will see below graph. Clearly, only CPs and bond paths
related to fragments 1 and 2 are generated, this is what we expected.
(3) Only retaining bond paths and corresponding BCPs connecting two specific fragments
Sometimes we only want to study interfragment interaction between two specific fragments
and hope that all irrelevant bond paths and BCPs could be removed to make the graph clearer.
Although you can manually delete undesired BCPs and bond paths manually via corresponding
suboptions in options -1 and -2, respectively, the process is usually tedious. Fortunately, in Multiwfn
there is a special option aiming for realizing this purpose. Below I will illustrate how to only retain
bond paths and corresponding BCPs connecting fragments 1 and 3 while removing all other BCPs
and bond paths.
GCGC.wfn
372

-5 // Manipulate paths
8 // Only retain bond paths (and corresponding BCPs) connecting two specific fragments
while delete all other bond paths and BCPs
1-13 // Atomic indices in fragment 1
30-45 // Atomic indices in fragment 3
y // Also delete BCPs in other bond paths
Then we manually delete all (3,+1) and (3,+3) CPs as described above, and then visualize result,
you will see
Evidently, only the BCPs and bond paths characterizing the stacking interaction between fragments
1 and 3 are retained, the graph looks very clear, this is what we desired.
4.2.7 Topology analysis based on attractors located by basin analysis:

spin density of biradical as an example
Topology analysis module is able to exactly locate various types of CPs, however, if
distribution of a real space function is relatively complicated, such as ELF, orbital wavefunction and
density difference, it is usually difficult to locate all CPs. In contrast, the basin analysis module
guarantees that all minima of negative region and maxima of positive region can be successfully
located, they are collectively known as "attractors", see examples of Section 4.17. However, since
the basin analysis is carried out based on evenly distributed grid, the accuracy of the attractors is
limited, because each attractor corresponds to a grid, while grid spacing is usually larger than 0.05
Bohr, which is several magnitude larger than displacement convergence threshold of topology
analysis. If you are only interested in minima and maxima of a real space function, obviously it is a
good idea to use attractors determined by basin analysis as starting points (initial guessing points)
for locating CPs in topology analysis module, in other words, one can use topology analysis module
to refine the positions of the attractors. The joint use of the two modules ensures that all minima of
negative region and maxima of positive region can be accurately located. In this example, I take
spin density of C4H8 biradical as example to illustrate how to realize this.
The example file is examples\C4H8.wfn. Before conducting basin and topology analyses for
spin density, we can use main function 5 to visualize its isosurface to examine its basic distribution
feature. The isosurface corresponding to spin density = 0.01 a.u. is shown below
373
We perform basin analysis first. Boot up Multiwfn and input

examples\C4H8.wfn
17 // Basin analysis
1 // Generate basins and locate attractors
5 // Spin density
2 // Medium quality grid
0 // Check the located attractors
After slightly modifying plotting settings, you will see
In above map, blue and green spheres correspond to minima of negative part and maxima of positive
part, respectively. Clearly their positions are fully in line with our expectation, which can be inferred
from the isosurface map.
Then close the GUI and input
-4 // Export attractors as pdb/pqr/txt file
3 // Export coordinates and function values of all attractors as attractors.txt
Now we have attractors.txt in current folder, in which the first three columns correspond to X,
Y, Z coordinate of the attractors in Bohr. Now we use them as starting points for topology analysis
of spin density. Reboot Multiwfn and input
examples\C4H8.wfn
-11 // Reselect the real space function to be studied
5 // Spin density
1 // Search CPs from given starting points
4 // Using starting points from a .txt file
attractors.txt
Then from screen you can find all the 16 points in the attractors.txt were employed as starting
374
points in turn, and finally 14 new CPs are found. Then input 0 to return and select option 0 to
visualize the CPs, you will see
In above map, the purple spheres, namely (3,-3) CPs, represent maxima of positive part of spin
density; while the green points, namely (3,+3) CPs, correspond to minima of negative part of spin
density. Although it is hard to detect visually, their positions are indeed more accurate than the
attractors directly yielded by the basin analysis module.
4.3 Output and plot various properties in a line
4.3.1 Plot the spin density curve of triplet formamide along carbon and
oxygen atoms

examples\formamide-m3.wfn
3 // Main function 3, plot real space function along a line
5 // Spin density
1 // Defining the line by nuclear coordinate of two atoms
1,6 // Indices of the two atoms, carbon and oxygen atoms correspond to 1 and 6 in present
example, respectively
The graph shows up immediately:
375
The X-axis corresponds to the position in the line you defined, the dashed line corresponds to
the position of Y=0, the left and right red circles denote the position of the first and the second atoms
you selected. After clicking right mouse button on the graph to close it, a new menu appears on
command-line window, you can save graphic file, adjust the range of Y-axis, export X-Y data points
and replot the graph and so on by corresponding options. By selecting option 6, minimum and
maximum positions can be located:
Local maximum X: 0.753677 Value: 0.20104903D-01
Local minimum X: 1.219518 Value: -0.55161276D-02
Local maximum X: 1.503567 Value: 0.29655008D+00
Local minimum X: 2.145518 Value: -0.70202704D-02
Local maximum X: 3.736194 Value: 0.51047629D-01
Local minimum X: 4.006988 Value: 0.18918295D-01
Local maximum X: 4.184992 Value: 0.24448996D+00
Totally found 3 local minimum, 4 local maximum
Using the same procedure illustrated above, you can plot curve map for any real space function
supported by Multiwfn, please have a try.
4.3.2 Study Fermi hole and Coulomb hole of H2

This is a relatively advanced example, you can skipped this section if you are a newbie of quantum chemistry.
In this example we will plot correlation hole (Fermi hole and Coulomb hole) along the axis of
H2. This is an advanced topic, if you are not familiar with the concept of correlation hole, please
consult the discussion in part 17 of Section 2.6.
Hartree-Fock wavefunction is capable to exhibit Fermi correlation, but Coulomb correlation is
completely ommitted. In this case exact Fermi hole can be calculated and plotted by Multiwfn. If
Coulomb hole is needed to be analyzed, then post-HF wavefunction must be employed. In current
version, Multiwfn is able to evaluate and plot approximate Fermi hole and Coulomb hole for post-
376
HF wavefunctions by Müller approximation, which makes use of natural orbitals to mimic exact
pair density. Although the approximation is introduced, the result is generally at least qualitative
correct.
First, we optimize and generate .wfn file for H2 under CCSD/cc-pVTZ level. CCSD is the
highest level of wavefunction that can be generated by Gaussian program (If higher level of
wavefunction is needed, you must resort on other programs, see Section 4.A.8 for details). The file
has already been provided as examples\H2_CCSD.wfn. The coordinate of H1 is (0.0, 0.0, 0.7016)
Bohr, the coordinate of H2 is (0.0, 0.0, -0.7016) Bohr.
Correlation hole involves coordinates of two electrons, to visually study it we must determine
position of a reference electron. In this example, we set reference point at (0.0,0.0,-0.3) Bohr.
Therefore, we open settings.ini, change refxyz parameter to 0.0,0.0,-0.3.
The correlation hole we first analyzed is Fermi hole (also known as exchange hole), so we
change paircorrtype in settings.ini to 1. Since this is closed-shell system, the results for  or β
electron are exactly the same, while for open-shell system, you should use "pairfunctype" in
settings.ini to select which type of spin electrons will be studied, you can also choose to study
exchange-correlation density or correlation factor by adjusting this parameter.
Now boot up Multiwfn, input following commands
examples\H2_CCSD.wfn
3 // Draw curve map
17 // Correlation hole
1 // Defining the line by nuclear coordinate of two atoms
2,1 // Draw curve graph along H2 and H1
Then you will see
This graph suggests that if we place an  electron at (0.0,0.0,-0.3), then the probability of
finding another  electron around the two nuclei will be significantly decreased by almost identical
extent due to Pauli repulsion between like-spin electrons. In the H-H bonding region, the probability
also obviously decreases. According to Bader's statement "An elecron can go where its hole goes
and, if the Fermi hole is localized, then so is the electron" (p251 in Atoms in molecules - A quantum
377
theory), we can say that if an  electron presents at (0.0,0.0,-0.3), then it must be able to easily
delocalize over the whole H2 molecular space.
Now we study Coulomb hole. Close Multiwfn, change "paircorrtype" to 2, reboot Multiwfn
and then replot the curve graph by using the same procedure, you will see
The red dot at left and right sides highlight position of H2 and H1, respectively. How can I
infer the position of the reference point on the graph? From the prompt "Set extension distance for
mode 1, current: 1.500000 Bohr" we know that the plotting range is extended by 1.5 Bohr in both
sides relative to Z coordinates of H2 and H1, and we know that Z coordinate of reference point is
larger than H2 by about 0.4016 Bohr, so the reference point on above graph is 1.5+0.4016=1.9016
Bohr (If you want to highlight the position of reference point on the graph, you can select "4 Draw
a vertical line at specific X" and input 1.9016, then redraw the graph via option -1). Above graph
reveals that owing to Coulomb repulsion, the probability of finding another electron ( or β) around
the hydrogen closest to the reference electron (namely around H2) is largely decreased. As a
compensation, the probability is increased at backside of H1.
Finally, we study the collective effect of Fermi correlation and Coulomb correlation. Change
"paircorrtype" to 3, and plot the line graph again, you will see
378
This graph is actually the sum of the first two graphs. From this graph it is clear that the
probability decrease of finding an electron ( or β) around the hydrogen closest to reference electron
is more severe than that around another hydrogen.
4.3.3 Study interatomic interaction via PAEM-MO method

This is a relatively advanced example, you can skipped this section if you are a newbie of quantum chemistry.
In J. Comput. Chem., 35, 965 (2014), the authors proposed a method named PAEM-MO to
reveal the nature of interaction between two atoms (covalently or noncovalently bounded). In this
example I will briefly introduce this method and show how to realize PAEM-MO analysis in
Multiwfn.
PAEM (potential acting on one electron in a molecule) refers to the total potential acting on an
electron at point r, and can be written as VPAEM (r ) = −VESP (r) + VXC (r) ; where VESP is
molecular electrostatic potential and has been introduced in part 12 of Section 2.6. -VESP can be
regarded as the classical potential acting on an electron in the system, while the exchange-
correlation (XC) potential VXC represents the important correction to the classical potential due to
quantum effect. VXC has two components, namely correlation potential (VC) and exchange potential
(VXC); in fact only the latter is important, that means even the potential obtained at Hartree-Fock
level is in general a good approximation to exact VXC.
In wavefunction theory, the exchange-correlation potential can be explicitly written as
1 XC (r, r' )

VXC (r ) = 
 (r ) | r − r' |
d r' , where Γ is known as exchange-correlation density, see part 17
of Section 2.6 for detail. In DFT theory, the XC potential directly comes from the variation of
EXC [  (r )]
exchange-correlation functional with respect to electron density, i.e. VXC (r ) = . The
 (r )
379
VXC can be used in Multiwfn in terms of user-defined function. If parameter "iuserfunc" is set to 33,
then VXC will be calculated based on Γ; if it is set to 34, VXC will be evaluated in terms of DFT XC
potential, in this case the computational cost is several times lower than 33. For more information
please check corresponding description in Section 2.7.
The essence of HF and KS-DFT theories is the one-electron eigenvalue equation
hˆ (r) =  (r) where hˆ = −(1 / 2) + VPAEM (r )

2
Solving the equation results in a set of MOs {}, their eigenvalues correspond to the energy of the
electrons running in the orbitals.
According to the viewpoint of PAEM-MO, if the energy of an occupied MO is higher than the
barrier of VPAEM between a pair of atoms, then the electrons in this MO will be able to freely
delocalize over the two atoms, and thus have direct contribution to covalent bonding.
Below I present two very simple examples of using PAEM-MO method to judge the type of
interatomic interaction. More examples and discussions can be found in J. Comput. Chem., 35, 965
(2014).
H-H interaction in hydrogen molecule

First set "iuserfunc" in settings.ini to 33, then user-defined function will be equivalent to the
VXC evaluated based on Γ. Boot up Multiwfn and input below commands
examples\H2.fch // Produced at HF/def2-TZVP level
3 // Plot real space function along a line
100 // User-defined function
0 // Adjust extension size at both sides
3 // 3 Bohr, which is larger than the default value
1 // Use two nuclei to define the line
1,2
Close the graph, then adjust some plotting parameters to make the graph better
11 // Change length unit of the graph to Å
3 // Change range of Y axis
-3,0.1 // From -3.0 a.u. to 0.1 a.u.
10 // Set stepsize of X and Y axes
0.5,0.5
-1 // Replot
You will see below graph, which exhibits the PAEM curve along the H2 axis
380
As you can see, there is a barrier of PAEM around the midpoint of H-H bond. In order to locate
the exact value of the barrier, we close the graph and select 6. From the output we find the maximum
of the curve is -1.903 a.u. (-51.78 eV). Present system only has one doubly occupied MOs, whose
energy (-16.23eV) significantly exceeded the PAEM barrier, implying that the electron is not
confined in any hydrogen atom but substantially contributes to the H-H binding, so the H-H bond
must be covalent.
He-He interaction in He2

Please plot PAEM curve along the He-He axis in He2 with the same steps as above. The
wavefunction file used here is examples\He2.fch, which was produced at HF/def2-TZVP level at
separation of 2.97 Å (a reasonable distance). The resultant graph should look like below
Obviously, the PAEM barrier (-9.78 eV) is higher than that in the case of H2. The energy of the
381
two doubly occupied MOs are -25.03 and -24.91 eV respectively, both of them are lower than the
barrier, therefore the electrons in each He atom are difficult (but not completely impossible due to
tunnel effect) to overcome the barrier to freely delocalize to another He. The He-He interaction thus
should be regarded as noncovalent interaction.
Note that in the original paper of PAEM-MO the PAEM is evaluated based on the expensive
CISD wavefunction, while we merely use HF wavefunction. However, our results are in good
agreement with that at CISD level, showing that correlation potential may be safely neglected in the
study of PAEM.
The interested users are suggested to replot the PAEM with iuserfunc=34 to employ DFT XC
potential in PAEM, you will find the results are very similar to those we obtained earlier.
Worthnotingly, although the PAEM-MO analysis method has clear physical meaning, its many
limitations severely hinders it to be a universal method to distinguish covalent and noncovalent
interactions like ELF or LOL: (1) PAEM-MO analysis does not always present reasonable
conclusion in all cases. For example, PAEM-MO erroneously indicates that the H-bond in water
dimer is covalent interaction. (2) PAEM-MO is difficult to be applied to polyatomic molecules,
since there are often too many occupied orbitals with complex shape. (3) If the two atoms are placed
too close to each other, then PAEM-MO almost always indicates that the interaction is covalent.
4.4 Output and plot various properties in a plane
Main function 4 of Multiwfn is used to plot various kind of plane maps for real space functions.
This module is extremely flexible, it is obviously impossible to demonstrate all possible usages of
this function via limited examples; however, if you carefully follow these examples and attempts to
reproduce the graphs, you will gain enough basic knowledge about plotting plane map using
Multiwfn. Section 3.5 is highly suggested to read, in which many important points about plotting
plane map are introduced.
The video https://youtu.be/E7lAGac3aDM is worth to look, the whole process of reproducing Multiwfn logo
(ELF map of Li6 cluster) is illustrated. This video also shows how to make background of the map transparent.
4.4.1 Illustration of plotting color-filled map and contour line map
4.4.1.1 Plotting electron density of hydrogen cyanide

In this example we plot electron density for hydrogen cyanide as color-filled map and contour
line map. Boot up Multiwfn and input following commands
examples\HCN.wfn
4 // Plot graph in a plane
1 // Electron density
1 // Color-filled map
[Press ENTER button directly] // Use the recommended grid setting, namely 200,200. If you
increase the number of grid points, graph will become finer and smoother, but you have to wait
longer time for calculating data and plotting the graph
2 // XZ plane (Z-axis is the molecular axis of present system, you can confirm this via main
function 0)
0 // The XY plane with Y=0 Bohr
382
After a few seconds the graph pops up
The central regions of carbon and nitrogen are white, suggesting that electron density exceeds
the upper limit of color scale (0.65). Close the graph, then a post-processing menu appears, there
are many options and their meanings are very easy to understand. You can choose corresponding
options to adjust plotting parameters and then use option -1 to replot again, or export X-Y data set
to a plain text file so that you can then plot the graph by external softwares (Sigmaplot, Origin,
matlab, etc.), or save image file in current directory (the graphical format is controlled by
"graphformat" in settings.ini).
Now we slightly improve above graph. Input below commands:
-8 // Change length unit of the graph to Å
-2 // Set stepsize in X, Y, Z axes
1,1,0.1
4 // Enable showing atom labels
1 // Red labels
8 // Enable showing bonds
3 // Blue color for bonds
-1 // Redraw the graph
Now you can see below map on screen
383
In Section 4.6.2 we will plot valence electron density for HCN, you will find valence electron
density conveys much more information than total electron density.
Next, we plot the electron density as contour line map. Repeat above example but select "2
Contour line map" instead of "1 Color-filled map", you will see below graph
There are numerous options in post-processing menu. If option 2 is chosen, the isovalues will
384
be marked on corresponding contour lines. Once option 3 is chosen, you will enter an interface for
setting up contour lines, various parameters such as color, thickness, contour values can be easily
customized, please try to play with them; if you are confused, please consult Section 3.5.4 for more
details.
It is worth to note that the .pdf format is more suitable than the default .png format for contour
line map (and other maps that mainly consist of lines), because .pdf is a vector format, the graph
can be scaled loselessly, and the lines look smoother. In order to change to .pdf format, you should
change "graphformat" in settings.ini to pdf.
4.4.1.2 Plotting electron localization function (ELF) for FOX-7

In this example we plot electron localization function (ELF) for an energetic compound, FOX-
7. The ELF is a very popular function to reveal electron localization character, see Section 2.6 for
introduction. The input file in this section is examples\FOX-7.wfn, if you load it into Multiwfn and
enter main function 0, you will see below structure. We will plot the plane defined by atoms N9, C2,
N12.

examples\FOX-7.wfn
4 // Plot plane map
9 // Electron Localization Function (ELF)
[Press ENTER button directly]
4 // Define the plane by three atoms
9,2,12 // Use nuclear position of the three atoms to define the plane
The graph directly shown on screen is not easy to study. Therefore we close the graph and then
input below command
14 // Brown color
1 // Red color
18 // Change style of atomic labels
3 // Plot both element symbol and atomic index
-1 // Show the graph again
We will see below graph
385
You can find that labels of many atoms are not shown on the map, this is because the vertical
distance between these atoms and the plotting plane is larger than the "disshowlabel" parameter in
settings.ini. If we want to display all atomic labels on the map, we should close the graph and input
below commands:
17 // Set distance threshold for showing atom labels
10 // Enlarge the threshold to 10 Bohr
y // If there is any atom whose vertical distance to the plane exceeds the threshold, its label
will still be shown but using thin text
-1 // Plot the map again to check effect
You will see below map, only interesting part is given
As can be seen, labels of all atoms have been shown, and corresponding bonds are also displayed.
In this map the red color substantially reveals the high electron localization nature in the bonding
386
regions. The degree of electron localization between N and O is not as high as the case of C-N and
C-C bonds, this is a known feature of "charge-shift bond", see Chem. Eur. J., 11, 6358 (2005) for
more information about this point.
The default color transition method is "Rainbow" with white and black colors in the region
where function value is lower and higher than the color scale, respectively. The coloring method
can be changed by user. As an illustration, in the post-processing menu we input
19 // Set color transition
17 // Black-Blue-Cyan
1 // Set lower&upper limit of color scale
0,0.7 // Decrease the upper limit from default value to 0.7 to make color able to better
distinguish ELF in different regions
-1 // Replot
The below graph looks very cool ;-D
Contour lines could be appended to the color-filled map. To make a pretty color-filled map
with contour lines, we input below command
8 // Blue-White-Red
2 // Enable showing contour lines
0,1
-2 // Set stepsize in X,Y,Z axes
2,2,0.1
-1 // Replot
The current graph is shown below, it is quite satisfactory
387
Note in passing that if you want to translate or rotate the plotted object in the graph, in the
interface of defining plotting plane, before choosing option 4 or 5, you should first choose the option
"-1: Set translation and rotation of the map for plane types 4 and 5" and then input translation value
and rotation angle, see Section 3.5.2 for more information.
4.4.2 Shaded surface map with projection effect of electron localization

function (ELF) of monofluoroethane
This section illustrates how to plot shaded surface map with projection. Boot up Multiwfn and
input following commands
examples\C2H5F.wfn
0 // View the molecular structure first to find the plane we are interested in. Suppose that the
C-C-F plane is what we want to plot, record the atomic indices (1, 5, 8), and then click RETURN
button to return to main menu
9 // Electron localization function (ELF)
5 // Shaded surface map with projection effect
[Press ENTER button] // Use recommended grid setting, namely 100,100
0 // Manually set extension distance. If you do not do this, you will found the resulting graph
is somewhat truncated at boundary because the default extension distance is too small for present
case
6 // Set extension distance to 6 Bohr, which is slightly larger than the default value
4 // Define the plotting plane by three atoms
1,5,8 // Indices of the three atoms
Below graph immediately pops up:
388
We can see that C-C and C-F covalent bond regions have high LOL value, rendering high
degree of electron localization in that places. The very low electron localization zones between
valence and inner shells of heavy atoms are revealed by the blue ringlike region around each nucleus.
A lone pair region of fluorine atom is pointed out by the purple arrow.
The graph you have seen in the GUI can be saved to graphic file by option 0 at post-processing
interface. If you find the objects in the exported image are truncated at their edges, you should select
option -1 to re-enter the GUI window, zoom out the graph and then export the picture again.
4.4.3 Plotting plane map without contributions from some atoms
The main purpose of this section is illustrating how to plot plane map of a real space function
without contribution of some atoms. This aim can be realized in two differents ways in Multiwfn,
as respectively exemplified in below two examples.
Example 1: Contour map of electron density Laplacian of uracil without contributions of

two atoms
In main function 6, one can use subfunctions -3 and -4 to delete Gauss type functions (GTFs)
centered at some atoms to remove their contributions to various kinds of analyses that based on real
space function. This feature will be utilized in present example. Since this treatment reduces total
number of GTFs, the computational cost in the subsequent analyses will be lowered.
Boot up Multiwfn and input following content
examples\uracil.wfn
6 // Modify wavefunction
-4 // Discard contribution of some atoms
389
3,4 // All GTFs centered on atoms 3 and 4 will be discarded. In other words, contribution of
atoms 3 and 4 will be removed from current wavefunction
3 // Laplacian function
2 // Contour line map
[Press ENTER button to use default grid setting]
1 // XY plane will be plotted
0 // The Z-position of the XY plane is zero, that is molecular plane
Below is the resulting graph, solid and dashed line correspond to positive and negative regions,
respectively.
From the graph you can see that the contribution from the two carbons have been discarded as
we expected.
By the way, if you are studying a large system but only a local region is of interest, you can
remove GTFs at atoms far from this region to save computational time of real space function
analyses (e.g. topolgy analysis, basin analysis, calculating grid data...).
Example 2: Plane map of LOL only contributed by atoms in the uracil ring
After plotting plane map as usual, one can request the program to plot the map only contributed
by certain fragment. Specifically, Hirshfeld weighting function of the user-defined fragment will be
generated and muliplied to the plane data. This treatment only affects the currently plotted map
while do not influence any further analysis, since this treatment does not modify wavefunction.
Here we use this feature to plot localized orbital locator (LOL) map of uracil that only
contributed by the six atoms in the ring. Boot up Multiwfn and input below commands:
390
examples\uracil.wfn
4 // Plane map
10 // LOL
1 // XY plane will be plotted
0 // Z=0
Close the graph and then input
-9 // Only plot the data around certain atoms
1-6 // The index of the six atoms in the uracil ring
14 // Brown
-1 // Replot
Then you will see
Clearly, the value of LOL at the grids far from the ring atoms have been significantly screened.
Then if you want to restore the original map, you can choose "-9 Recovery original plane data" in
the post-processing menu and then replot.
4.4.4 Contour map of electrostatic potential of chlorine trifluoride
In this example we plot electrostatic potential (ESP) for chlorine trifluoride as contour map.
examples\ClF3.wfn // Generated at B3LYP/6-31G* level
12 // Total electrostatic potential
2 // Draw contour line map
120,120 // Number of grids in each direction
3 // YZ plane
0 // Set X coordinate of the YZ plane to 0
391
Because evaluation of ESP is evidently more time-consuming than other real space functions,
you need to wait for a while.
Hint: For large system and when input file is .fch/fchk format, if you allow Multiwfn to invoke cubegen utility
in Gaussian package to evaluate the ESP values at the plotting points, the overall computational cost could be
evidently reduced, you simply need to set "cubegenpath" in settings.ini to actual path of cubegen. See Section 5.7
for detail.
After the calculation is finished, ESP map pops up. This map is inconvenient to be visually
analyzed, since what we are interested in is often the ESP value on molecular vdW surface, hence
it is better to plot the vdW surface on this map simultaneously. In order to do this, we close the graph
by clicking right mouse button, choose option 15 in post-processing menu, and then choose option
-1 to replot the graph, you will see such a picture. The solid and dashed lines represented the region
having positive and negative value of ESP, respectively.
The bold blue line corresponds to vdW surface (isosurface of electron density=0.001 a.u., as defined
by R. F. W. Bader). From the graph it is clear that chlorine atom is overall positively charged,
because the vdW surface close to the chlorine atom largely intersects solid contour lines. For the
same reason, we can see that the equatorial fluorine atom possesses less electrons than the two axial
fluorine atoms, this point can be further verified when we calculate atomic charges for this molecule
in section 4.7.1.
Plane map of the ESP derived from atomic charges can be directly plotted by Multiwfn too.
First, you need to prepare a plain text file with .chg extension, the first column corresponds to
element name, the 2th, 3th and 4th columns correspond to X,Y and Z cooredinates in Å respectively,
the last column is atomic charge. For example:
Cl 0.000000 0.000000 0.359408 0.529971
F 0.000000 1.726507 0.294501 -0.228394
F 0.000000 0.000000 -1.267884 -0.073185
F 0.000000 -1.726507 0.294501 -0.228394
Boot up Multiwfn as usual, and then use the .chg file as input. The plotting procedure is
392
completely identical to the one given above, except that when Multiwfn prompts you to select real
space function, you should select 8 (ESP from atomic charges) instead of 12.
4.4.5 Contour map of two orbital wavefunctions
Multiwfn is capable of plotting contour map for two orbitals simultaneously. In this section,
we will draw contour map for simultaneously portraying NBO 12 and NBO 56 of NH2COH (recall
Section 4.0.2). The plane we selected is the one perpendicular to molecular plane and passed through
both carbon and nitrogen atoms. As you will see, we need to use a special manner to define such a
plotting plane. The molecule geometry and atomic index are shown as follows.
Boot up Multiwfn and input:

examples\NH2COH.31
37 // Load NH2COH.37
4 // Plot plane graph
4 // Orbital wavefunction
12,56 // The two orbital indices. If you only input one index, then only one orbital will be
plotted
7 // This mode is used to define a plotting plane parallel to a bond and meantime normal to a
plane defined by three atoms
1,4 // The plotting plane is parallel to C1-N4
3,1,4 // The plotting plane is perpendicular to the plane defined by O3-C1-N4
10 // The length of X-axis of the resulting map is 10 Bohr
10 // The length of Y-axis of the resulting map is 10 Bohr
Immediately a graph pops up. We close it by clicking right mouse button, choose option 2 and
input 25 to enable showing isovalue on contour lines, then choose -1 to redraw the graph, we will
see:
393
This contour map is not quite ideal, there are too many contour lines interwined together and
thus confused our vision. The culprits are the contour lines with too small isovalue (magnitude
smaller than 0.01). Since these contour lines are unimportant, we can delete them to make the graph
clearer. Therefore, we close the graph and input
3 // Change contour line setting
4 // Delete some contour lines
1-4 // Delete contour lines 1~4, they respectively correspond to 0.001, 0.002, 0.004, 0.008
4 // Delete some contour lines
28-31 // Delete the four contour lines corresponding to -0.001, -0.002, -0.004, -0.008. For
convenience, you can choose option 6 to export current contour line setting to an external file, when
you use Multiwfn next time you can load present setting directly by choose option 7 in current
interface
15 // Set the drawing style suitable for publication, namely positive and negative parts are
portrayed as red solid lines and blue dashed lines, respectively
1 // Save setting and return to the upper menu
-8 // Change length unit of the graph to Å
-2 // Set stepsize in X and Y axes
1,1 // Stepsize in both X and Y axes are 1.0 Å
-1 // Replot the contour map
2 // Enable showing isovalue on contour lines
30 // Use label size of 30
Now the graph become very clear and informative, the overlapping region of same phase is
very obvious.
394
4.4.6 Gradient+contour map with topology paths of electron density of

hydrogen peroxide
As introduced in Section 3.5.5, critical points, bond paths and interbasin surfaces can also be
plotted on plane map, here I present a simple example. There is a corresponding video illustration
of plotting this kind of map https://youtu.be/gv5FkiFWUY0, you are suggested to look at it.
Gradient map of electron density

First, I show how to plot a normal gradient map of electron density for hydrogen peroxide.
examples\H2O2.fch // Of course, you can also use other type of file as input, as long as the
file contains GTF information
4 // Plotting plane map
6 // Gradient line with/without contour line map
4 // Use three atoms to define the plotting plane
2,1,3 // Define the plane by nuclear positions of atoms 2, 1 and 3
Generating data and plotting gradient map take more computational time relative to other graph
type, however since present system is small and the basis set is only 6-31G*, resulting graph shows
up immediately:
395
This type of graph is very useful in Bader’s AIM analysis. You can also plot gradient+contour
map for any other real space functions supported by Multiwfn. In the post-processing menu, you
can use options 11, 12, 13 and 14 to adjust plotting effect of the gradient lines, they can control the
smoothness, density, color and density of the gradient lines.
Gradient map of electron density with bond paths and critical points of electron density
If you hope critical points and paths also be portrayed on the graph, you need to do topology
analysis as illustrated in Section 4.2.1 prior to plotting the map. Now we input below commands
2 // Topology analysis (by default electron density is the function to be analyzed)
8 // Generating the paths connecting (3,-3) and (3,-1) CPs, namely generating bond paths in
current context
0 // Visually check if all expected CPs and paths have been generated. This step is optional
Then draw gradient line map for electron density via the way described above. The resulting
graph should look like below. Brown, blue, and orange circles denote (3,-3), (3,-1) and (3,+1) critical
points, respectively. Bold dark brown lines depict bond paths.
396
In the post-processing menu, you can enter “4 Set details of plotting critical points and paths” to
adjust settings for showing the critical points and paths.
Hint: The color of different types of critical points can be set by "CP_RGB_2D" in the settings.ini file.
Interbasin paths can also be drawn on the graph. If you have finished the search of CPs in
topology analysis module, after drawing contour/gradient/vector field map, you could find an option
named "Generate and show interbasin paths" in the post-processing stage; select it and replot the
graph, the interbasin paths will be shown on the graph by bold dark blue lines:
397
Note that before generating the interbasin paths, related parameters (stepsize and the number
of iteration) can be set by option "7 Set stepsize and maximal iteration for interbasin path
generation" in the post-processing menu. Larger number of iteration may result in longer interbasin
paths.
Contour line map of Laplacian of electron density with bond paths and CPs of electron
density
Finally, we plot contour line map of Laplacian of electron density, on which the bond paths and
CPs we generated before are also shown. Return to main menu and then input below commands
4 // Plotting plane map
3 // Laplacian of electron density
4 // Use three atoms to define the plotting plane
2,1,3 // Define the plane by nuclear coordinates of atoms 2, 1 and 3
The resulting graph is shown below (only a local region is given)
398
4.4.7 Deformation map of electron density of acetyl chloride
Deformation map of electron density clearly shows variation of electron density distribution
during formation of a molecule, it is defined as subtracting actual molecular electron density by
electron density of all of its constituent atoms in their free-states. Illustrative example of deformation
density analysis can be found from my paper Acta Phys. -Chim. Sin., 34, 503 (2018).
It is a labor work to draw such a graph via custom operation feature since their are so many
atoms in practical chemical systems. Fortunately, Multiwfn provides a special option to realize this
in a highly automatic way. Boot up Multiwfn and input following commands
examples\CH3COCl.wfn
4 // Plot plane map
-2 // Tell Multiwfn you want to draw deformation map, then Multiwfn prepares free-state
atom wavefunctions
B3LYP/6-31G* // The level used to generate atomic wavefunction files by Gaussian, it is the
same as the level used for generating CH3COCl.wfn
D:\study\g09w\g09.exe // The path of executable file of Gaussian (you can also use other
Gaussian version). If you already set correct path in “gaupath” parameter in settings.ini, then
Multiwfn will not ask you to input the path every time
Now Multiwfn starts to invoke Gaussian to calculate atom wavefunctions, then Multiwfn
translates and sphericalizes them internally. These temporary wavefunction files are stored in
“wfntmp” folder in current directory, after you get the expected graph you can delete the folder. Let
us continue to input the remaining commands.
1 // Electron density function
2 // Contour map
1
0 // The XY plane with Z=0 is the plane of acyl chloride
Then the deformation map pops up:
399
As we expected, electron density concentrates towards bonding regions. We also find that the
density distribution around chlorine atom is largely desphericalized, this observation is in line with
hybrid orbital theory, chlorine atom forms somewhat sp3 hybrid state.
You can also plot deformation maps for other functions by choosing corresponding real space
function, though not all of them are meaningful.
If you want to avoid recalculating atomic wavefunction files next time, you can copy the .wfn
files without number suffix (such as “C .wfn“) from “wfntmp” folder to “atomwfn“ folder in current
directory, if Multiwfn finds that all needed atom wavefunctions have already existed in “atomwfn”
folder, then Multiwfn will not invoke Gaussian to calculate them again.
Hint: You can also use genatmwfn.pdb in “examples” directory to generate all atom wavefunctions under
specific basis set in a single run, please consult Section 3.7.3.
The “atomwfn” folder in “examples” directory contains atom wavefunctions (by 6-31G*) for
all first-four row elements, you can directly copy this folder to current directory, after that you will
not need Gaussian again during plotting deformation map.
If your system involves some elements heavier than Kr, you have to manually calculate the
corresponding atomic .wfn files and put them into "atomwfn" folder". More detailed information
about preparing atomic wavefunction files can be found in Section 3.7.3.
4.4.8 Draw difference map of electron density and ELF for water
tetramer with respect to its constituent monomers
In this example I will illustrate how to plot difference map between a system and its constituent
400
fragments for a given real space function. Electron density and ELF will be employed as the function
to be studied.
examples\water_tetramer\wfn\complex.wfn is wavefunction file of optimized water tetramer,
while the water1/2/3/4.wfn in this folder is wavefunction file of each water monomer. The
corresponding Gaussian input files are also provided in the folder. Notice that the monomer
coordinates were directly extracted from the complex coordinate, and nosymm keyword was used
for all files to avoid Gaussian automatically reorientating the molecular geometry during the
calculations. (Bear in mind, density difference map is meaningful only when coordinates of all
fragments are completely consistent with that of the whole complex)
Plotting difference map of electron density

First we plot plane map of electron density difference for the complex with respect to all the
four monomers. Boot up Multiwfn and input
examples\water_tetramer\wfn\complex.wfn
4 // Plot plane map
0 // Custom operation
4 // Four files will be operated on the firstly loaded system
-,examples\water_tetramer\wfn\water1.wfn
2 // Contour map
[Press ENTER to use default grid setting]
4 // Define plane by three atoms
7,10,1
The graph pops up immediately. We can further improve the plotting effect. Close the graph
and input
3 // Change contour line settings
15 // Set line style and width suitable for publication
1 // Save and return
0.2 // 0.2 Bohr
y // If the distance between an atom and the plotting plane is larger than the specified 0.2
Bohr, then the label will be drawn as thin style
0 // Save the plot as graphic file in current folder
The graphic file should look like below
401
In the graph, red solid lines and blue dashed lines correspond to the regions having increased
electron density and decreased electron density during formation of the tetramer, respectively.
In Multiwfn you can also easily plot the density difference in the form of isosurface map by
main function 5, the resulting graph is shown below, the isovalue is 0.003. If you do not know how
to do, please consult Section 4.5.5.
Plotting difference map of ELF

Next, we plot color-filled difference map of ELF. Input below commands
4 // Plot plane map
4 // Four files will be operated on the firstly loaded system
402
9 // ELF
4 // Define plane by three atoms
7,10,1
The graph shown on screen is ugly currently, because the default color scale is inappropriate
for present case. Close the graph and input
-1.5,0.1
4 // Disable showing atomic labels
4 // Enable showing atomic labels again, now you can select label color
3 // Blue labels
You will see
The blue and especially dark blue regions exhibit decrease of ELF in corresponding regions.
This plot shows that during formation of the complex, the electron localization is reduced in the
intermolecular interaction regions, it may be attributed to the consequence of the Pauli repulsion
effect.
4.4.9 Draw LOL-π map for porphyrin to reveal favorable electron

delocalization path
The well-known ELF- is the ELF solely contributed by  electrons. Similarly, LOL- can be
defined as a variant of localized orbital locator (LOL). The features of LOL- are highly analogous
to ELF-, but sometimes LOL- performs better. In this Section I will illustrate how to plot color-
403
filled LOL- plane map above 1.2 Bohr of porphyrin, you will find this map is quite useful for
understanding preferential electron delocalization path, which is closely related to molecular
aromaticity.
The only difference of plotting and analyzing LOL and LOL- is that for the latter case, you
should first set occupation number of all orbitals to zero except for  orbitals. As illustrated below,
this can be automatically done via Multiwfn.
The .fch file of porphyrin calculated at B3LYP/6-31G* level can be downloaded form
http://sobereva.com/multiwfn/extrafiles/porphyrin.rar. Boot up Multiwfn and load the .fch file, then
input below commands:
100 // Other functions
22 // Detect  orbitals
0 // Detect  orbitals for delocalized orbitals of exactly planar system
2 // Set occupation number of all other orbitals to zero
0 // Return to main menu
4 // Plot plane map
10 // LOL
Press ENTER button to use default grids
3 // Plot YZ plane
1.2 // X=1.2 Bohr
Close the graph, then input
0,0.66
7 // Cyan color
2 // Since the distance between the plotting plane and molecular is 1.2 Bohr, to make all
atomic labels shown on the graph, this threshold must be set to a value larger than 1.2 Bohr. Here
we set it to 2.0 Bohr
y
14 // Brown color
-1 // Replot
Now you will see below graph
404
The high LOL- regions (red or orange color regions) clearly reveal the favorable
delocalization path. If you plot current map induced by an external magnetic field vertical to the
molecular plane (using for example AICD or GIMIC methods, see my slideshow for details:
http://sobereva.com/148), you will find the unidirection contiguous induced current is mainly
formed on the favorable delocalization path highlighted by the LOL- function.
4.4.10 Plotting gradient line and vector field map of electrostatic

potential to reveal electric field of LiF
The main purpose of this section is illustrating how to properly plot meaningful gradient line
map and vector field map.
Electric field (F) is defined as negative of 1st-derivative vector with respect to coordinate (i.e.
gradient vector) of electrostatic potential (ESP); therefore, if we plot gradient line or vector field
map of ESP, the F could be vividly exhibited. In this section, I will use an ionic compound LiF as
example.
Gradient line map

examples\LiF.wfn // Generated at B3LYP/6-31G* level
4 // Plot plane map
12 // ESP
6 // Gradient line map
405
0 // Modify extension distance

6 // 6 Bohr (larger than default value)
3 // YZ plane
0 // X=0
Close the graph, then input
15 // Show a contour line to reveal van der Waals surface
10 // Show arrow on the gradient lines
Now you can obtain below map
In above map, the gray gradient lines clearly exhibit direction of electric field everywhere.
Note that since electric field corresponds to negative gradient vector of ESP, the arrows in fact
should be inverted.
Vector field map

Next, we draw a graph of another style to display electric field character. Input below
commands:
4 // Plot plane map
12 // ESP
7 // Vector field map
50,50 // Number of grids in the two dimensions
3 // 3 Bohr (smaller than default value to focus on relatively important region)
3 // YZ plane
0 // X=0
Close the graph shown on the screen and then input
11 // Map color to arrows
406
10 // Set upper limit of absolute value for scaling arrows

0.05
15 // Show a contour line to reveal van der Waals surface
13 // Invert gradient vectors, so that the arrows will correspond to direction of electric field
1 // Disable showing atom labels and reference point
1 // Enable showing atom labels and reference point
3 // Use blue label color
Now replot the map, you will see
In this map, the more red the arrow, the larger the magnitude of the electric field at
corresponding position. We can see that electric field sources from each nucleus, while in the region
relatively far from the two nuclei, the overall direction of the electric field vectors is from the Li
side towards the F side, this is because in this system Li and F carry evidently positive and negative
net charges, respectively. You can also find there is a semicircle shape region at top of the map,
where magnitude of electric field is small or vanished. This region in fact has the most negative
value of ESP and thus behaves as ending point of molecular electric field (indeed, all arrows around
this region point towards this region).
4.5 Generate grid data and view isosurface map
This section contains examples of main function 5 of Multiwfn, all of them need to calculate
grid data. Once the grid data is generated, it can be visualized as isosurface map, or be exported
to .cub file so that it can be rendered by third-part tool such as VMD or be further utilized by other
407
analysis codes.
Note that using VMD to render the cube files generated by Multiwfn using my prepared VMD
script is extremely easy, while the graphical quality is amazingly good, please check Section 4.A.14
on how to realize this.
4.5.1 Electron localization function of chlorine trifluoride
In this section we plot isosurface map of electron localization function (ELF) for chlorine
trifluoride. Boot up Multiwfn and input following commands
examples\ClF3.wfn // Generated at B3LYP/6-31G* level
5 // Generate grid data and view isosurface
9 // Electron localization function (ELF)
2 // Medium quality grid, about 512000 points will be evaluated, this setting is fine enough
for small system but inadequate for medium and especially large system. Please consult Section 3.6
for more knowledge about grid setting.
Now Multiwfn starts to calculate grid data. Once the calculation is finished, Multiwfn outputs
some statistical information. There are many options in the newly appeared menu, you can draw
isosurface map by selecting option -1, after that a GUI window will pop up. Input isovalue of 0.85
in the text box and press ENTER button, you will see below isosurface map
The ELF isosurfaces clearly revealed the lone pair regions of the fluorine and chlorine atoms.
If you want to export the grid data as Gaussian cube file in current directory, you should click
“Return” button to close the GUI window and then choose option 2.
If you use ChimeraX to plot isosurface map based on the ELF cube file exported by Multiwfn,
you will be able to freely set color for various domains in different regions. For example, the below
map is ELF=0.83 isosurface of acetic acid plotted by ChimeraX, you can easily reproduce this map
if you carefully follow this video tutorial: "Plotting electron localization function (ELF) isosurface
using Multiwfn and ChimeraX" (https://youtu.be/vC48iEB8PwI).
408
4.5.2 Laplacian of electron density of 1,3-butadiene
Electron density Laplacian is another useful real space function like ELF and LOL to reveal
electronic structure. Laplacian is not as good as ELF and LOL for highlighting localization region
due to its poor distinguishability. For example, the shell structures of atoms heavier than krypton
cannot be fully illustrated by Laplacian, and if you attempt to use Laplacian to analyze chlorine
trifluoride, you will find the lone pair regions of fluorine atoms are difficult to be identified.
Moreover, the value range of Laplacian is too large, which brings difficulties on visual analysis.
However, for many systems Laplacian of electron density is still useful. In this example we will plot
isosurface map of this function for 1,3-butadiene.
examples\butadiene.fch // Yielded at B3LYP/6-31G** level
5 // Generate grid data and view isosurface
3 // Electron density Laplacian
2 // Medium quality grid (If you want to obtain better graphical effect, choose "high quality
grid" instead)
-1 // View isosurface
In the newly occurred window, change isosurface value from default value to 0.3, then the
green and blue isosurfaces will correspond to isovalue of 0.3 and -0.3 respectively. The current
graph will look like below
409
The presence of blue isosurfaces between C-C and C-H suggests that valence-shell electrons
are strongly concentrated on these regions, this is typical pattern of covalent bonding. If you inspect
the isosurfaces between carbon atoms carefully, you will find that the isosurface between C6-C8 or
C1-C4 is wider than the isosurface between C4-C6, this phenomenon reflects that the two boundary
C-C covalent bonds are stronger than the central one. This conclusion can also be verified by other
wavefunction analysis schemes, for example, Mayer bond order (bond order between C6-C8 is
1.863, while the bond order between C4-C6 is only 1.136).
The isosurface map we obtained in this section is not quite smooth, this is because we only
used medium quality grid. If you choose high quality grid instead, much better isosurface will be
viewed.
4.5.3 Calculate ELF-α and ELF-π to study aromaticity of benzene
ELF-α and ELF-π indices are commonly used α and π aromaticity indices, they are defined as
the ELF value at bifurcation point (i.e. (3,-1) type of CP) of ELF domains that solely contributed
from α orbitals and π orbitals, respectively, see J. Chem. Phys., 120, 1670 (2004) and J. Chem.
Theory Comput., 1, 83 (2005) for detail. These two papers contain very nice examples of employing
ELF- to analyze  delocalization: Theor. Chem. Acc., 139, 25 (2020) and Carbon, 165, 468 (2020).
The theoretical basis of these indices is that the ELF value at bifurcation point measures
interaction between adjointing ELF domains, the larger value means electrons have better
delocalization between these domains. Strong multi-center delocalization is commonly recognized
as nature of aromaticity. It is argued that if ELF-π is larger than 0.70, then the molecule has π
aromaticity. While if the average of ELF-π and ELF-α is larger than 0.70, one can say that the
molecule is global aromatic. In present example, we will calculate ELF-α and ELF-π for benzene.
In order to separate α and π orbitals, we need to know which orbitals are π orbitals first. Boot
up Multiwfn and input following commands
examples\benzene.wfn // Optimized at B3LYP/6-311G* level
0 // View molecular orbitals (MOs)
Now check orbital shape of each MO in turn, we found 17th, 20th and 21th MOs are π orbitals,
as shown below. All other MOs are recognized as α orbitals.
We first calculate ELF-π. The contribution to ELF from α orbitals should be ommited; this can
be realized by setting occupation number of all  orbitals to zero.
410
6 // Enter "Modify & Check wavefunction" interface

26 // Set occupation number for some orbitals
0 // Selecting all orbitals
0 // Set occupation number of all orbitals to zero
17,20,21 // Select MO 17, 20 and 21, namely all π orbitals
2 // Set occupation numbers of MO 17, 20 and 21 to 2.0 (doubly occupied). If you want to
check if occupation numbers have been correctly set, choose option 3. You will find occupation
number of all α orbitals have become zero, namely they will have no contribution to all results
yielded in following calculations
q // Return to last menu
For this system, in fact there is a much more convenient way to set occupation number of all
orbitals except for the π ones to zero. The procedure is: Enter subfunction 22 of main function 100,
select 0, then all π orbitals will be automatically identified, then choose 1 to set occupation number
of all other orbitals to zero (or choose option 3, if heavier elements such as silicon are involved in
present system). Finally, choose 0 to return to main menu. More details about automatic
identification of π orbitals can be found in Section 3.100.22.
Studying ELF-π
There are two ways to study ELF-π, the way 1 is to examine ELF isosurface directly, while the
way 2 is performing topology analysis. Way 1 is more intuitive but less accurate than way 2. Here I
illustrate way 1 first. Generate and view isosurface for ELF by main function 5 as usual (recall
Section 4.5.1. Using High quality grid is recommended). This time the ELF isosurface only reflects
π-electron localization character. By gradually increasing isovalue, you will find that the two circle-
shape ELF domains are bifurcated to twelve sphercial-like domains at about the isovalue of 0.91
(see below graph), implying that ELF-π index of benzene is about 0.91.
Next, let us use way 2 to evaluate ELF-π index again, this way is more rigorous than way 1.
Choose 0 to return to main menu.
-11 // Select real space function
9 // ELF
6 // The guessing points will be scattered around each atom in turn. This searching mode is
the most appropriate one for locating ELF CPs
-1 // Start the CP search
-9 // Return to upper menu
0 // Visualize results. The resulting graph is shown below, two (3,+1) CPs are not shown
411
By comparing this graph with ELF isosurface map, it clear that the (3,-1) CPs (orange) are
bifurcation positions of ELF domains, while (3,-3) CPs (purple) correspond to the maximum points
of the twelves ELF domains. Now we check ELF value at a (3,-1) CPs, we can choose any one,
since they are all equivalent.
7 // Show all properties at a CP
23 // CP23
From the output, we find the ELF value at CP 23, namely ELF-π index of benzene is 0.91247,
this result is in very good agreement with the value 0.913 given in Chem. Phys. Lett., 443, 439
(2007), note that our calculation level is exactly identical to this paper. Evidently, this value exceeds
the criteria (0.70) of π aromaticity, suggesting that benzene has strong π aromaticity.
Studying ELF-
Now we calculate ELF-α for benzene. Reboot Multiwfn and load benzene.wfn, set occupation
number of MO 17, 20 and 21 to zero (it is more convenient to use subfunction 22 in main function
100 to do this). Then generate isosurface for ELF as usual, gradually adjust isovalue, try to find out
at which isovalue the domains corresponding to the α bonds between carbon atoms are bifurcated.
One can finally find that at the isovalue equals to 0.71 the domain are bifurcated, suggesting that
ELF-α index is about 0.71. The bifurcation point is pointed by red arrow in below graphs:
Enter topology analysis module and search ELF CPs, like what we did in way 2 of ELF-π
analysis. You will obtain below graph. For clarity, (3,+1) and (3,+3) CPs are hidden.
412
Comparing positions of CPs with ELF isosurface map, it is clear that the (3,-1) CPs such as CP
48 and 58 correspond to bifurcation points of the α bond domains between the carbon atoms. Check
ELF value at CP 48, we get 0.70907, which is the accurate ELF-α value. Our result agrees well with
the value 0.717 from Chem. Phys. Lett., 443, 439 (2007).
The average of ELF-α and ELF-π is (0.70907+0.91247)/2=0.81077, which is larger than the
criteria of global aromaticity, so benzene possesses global aromatic character.
4.5.4 Use Fukui function and dual descriptor to study favorable site of
electrophilic attack for phenol
Multiwfn supports a batch of methods for predicting the most reactive sites, see Section 4.A.4
for detail. In this section, I will introduce how to realize Fukui function and dual descriptor, which
are the most popular two methods for revealing reactive sites.
IMPORTANT NOTE: In daily research, I strongly suggest you directly use subfunction 16 of
main function 100 to automatically calculate Fukui function and dual descriptor, because the steps
are much more simple than the those described below, and meantime many useful quantities in
conceptual density functional theory can be obtained as byproducts, see Section 3.100.16 for
introduction and 4.100.16.1 for example.
4.5.4.1 Fukui function
Theory
Fukui function is a very important concept in the conceptual density functional theory, it has
been widely used in prediction of reactive sites. Fukui function is defined as follows, see J. Am.
Chem. Soc., 106, 4049 (1984) for original paper and my paper Acta Phys. -Chim. Sin., 30, 628 (2014)
for related discussions and comparison.
  (r ) 
f (r ) = 
 N 
where N is number of electrons in present system, the constant term  in the partial derivative is
413
external potential. Generally, the external potential only comes from nuclear charges, so  can be
simply regarded as nuclear coordinates for isolated chemical system. It is argued that reactive sites
should have larger value of Fukui function than other regions. We cannot directly evaluate the partial
derivative due to the discontinuity when N is integer. Via the finite difference approximation, Fukui
function can be calculated unambiguously for three situations:
Nucleophilic attack: f + (r) =  N +1 (r) −  N (r)   LUMO (r)
Electrophilic attack: f − (r) =  N (r) −  N −1 (r)   HOMO (r)
f + (r ) + f − (r )  N +1 (r ) −  N −1 (r )  HOMO (r ) +  LUMO (r )
Radical attack: f (r ) = = 
0
2 2 2
Preparing wavefunction files

Below we first reveal the reactive sites for electrophilic attack of phenol by means of the Fukui
function f − shown above. The approximate form of Fukui function based on frontier orbitals will
not be used here.
We need to prepare the files needed by calculation of f −. Assume that you are a Gaussian user,
you can use .wfn, .wfx or .fch as input file for present purpose. In this example, we perform
following calculations to generate needed .wfn files, all calculations are conduced at B3LYP/6-31G*
level, which is the lowest acceptable level for producing meaningful result (of course you can use
better level to improve the result):
(1) Optimizing geometry structure of phenol of neutral state. The resulting geometry will be
used for next steps
(2) Performing single point task for phenol of neutral state to yield phenol.wfn (see
examples\phenol.gjf)
(3) Performing single point task for phenol of N-1 state (i.e. cationic state) to yield phenol_N-
1.wfn (see examples\phenol_N-1.gjf)
Notice that the geometry of the N-1 state should not be optimized before performing single
point task of the N-1 state, because the  (nuclear coordinates in this context) is a constant in the
partial derivative of Fukui function. By the way, in fact the step (2) could be omitted if you directly
specifying out=wfn keyword and output path of .wfn file at step (1).
Calculating Fukui functions f −

To study the isosurface of f −, we need to properly use the "custom operation” feature of
Multiwfn (see Section 3.7.1 for detail). Boot up Multiwfn and input following commands:
examples\phenol.wfn // Phenol of neutral state
5 // Calculate grid data
0 // Set custom operation
1 // Only one file will be operated with the file that has been loaded (namely phenol.wfn)
-,examples\phenol_N-1.wfn // “-“ is subtraction sign. Property of the firstly loaded file
(namely phenol.wfn) will be subtracted by corresponding property of phenol_N-1.wfn
Now Multiwfn starts to calculate electron density grid data for phenol.wfn, then calculate that
for phenol_N-1.wfn, and finally get their difference to yield grid data of f −. We choose option -1 to
414
check the isosurface, after adjusting the isovalue to a proper value (0.007), the graph will be
In the map, green and blue isosurface correspond to positive and negative region of f −, respectively.
Clearly, most positive part of f − function is localized on O12, C1, C3, C4 and C5, that means para
and ortho positions of hydroxyl are favourable reactive sites for electrophilic attack, this conclusion
is in agreement with common knowledge, namely hydroxyl group is an ortho-para- director.
Calculating Fukui functions f 0

Next, I take propylene as an example to illustrate how to plot Fukui function for radical attack,
namely f 0 = (N+1 − N-1)/2. Of course, we should yield wavefunction file corresponding to N+1
state and N-1 state. We first optimize geometry of neutral state (examples\propylene\opt_N.gjf), then
use this geometry to perform single point task of N-1 and N+1 states to yield corresponding .fch
files.
examples\propylene\N+1.fch // N+1 electrons state, namely -1 charged state
1 // One file will be operated with propylene-1.fch
-,examples\propylene\N-1.fch // N-1 electrons state, namely +1 charged state
6 // Divide all grid data by a factor
2 // Divided by 2
-1 // Visualize isosurface map
The isosurface map of f 0 = 0.01 is shown below
4.5.4.2 Dual descriptor
Theory
Dual descriptor is another useful function used to reveal reactive sites, see J. Phys. Chem. A,
415
109, 205 (2005) for detail. Formally, the definition of the dual descriptor f has close relationship
with Fukui function:
f (r ) = f + (r ) − f − (r )
= [  N +1 (r ) −  N (r )] − [  N (r ) −  N −1 (r )] =  N +1 (r ) − 2 N (r ) +  N −1 (r )
Worthnotingly, dual descriptor can also be evaluated in terms of spin density  s . Since
 N +1 −  N and  N −  N −1 can be approximated as  Ns +1 and  Ns −1 respectively, it is clear
that f (r )   N +1 (r ) −  N −1 (r ) . Commonly, there is no evident qualitative difference between

s s
the dual descriptor evaluated based on electron density of three states (N+1, N, N-1) and the one
based on spin density of two states (N+1, N-1).
Unlike Fukui function, via f both types of reactive sites can be revealed simultaneously. It is
argued that if f > 0, then the site is favorable for a nucleophilic attack, whereas if f < 0, then the
site is favorable for an electrophilic attack. However, according to my experience, if your aim is to
figure out which ones are more favorable among many potential sites, you do not need to concern
the sign of f, you only need to study which sites have more positive or more negative of f. If the
distribution of f around a site A is more positive than another site B, then one can say A is a more
favorable site for nucleophilic attack than B, and meantime B is a more preferential site for
electrophilic attack than A.
Approximately evaluating dual descriptor based on spin density

Here we calculate dual descriptor for phenol based on spin density of N-1 and N+1 states. Since
we have already calculated phenol_N-1.wfn earlier, now we only need to calculate phenol_N+1.wfn
(this file and corresponding input file phenol_N+1.gjf has been provided in "example" folder). After
that, boot up Multiwfn and input:
examples\phenol_N+1.wfn // N+1 electron system, namely anionic state
1 // Only one file will be operated with the file that has been loaded
-,examples\phenol_N-1.wfn // N-1 electron system, namely cationic state
5 // Electron spin density
-1 // Visualize isosurface of dual descriptor
We gradually change the isovalue so that the dual descriptor at different sites can be clearly
distinguished, we find 0.02 is a proper value, the corresponding isosurface is shown below
416
We can see that in the ring (except for the C4, which cannot participate in reaction), the para-
carbon has evident negative value of f, and meantime the f at the two ortho-carbons is not so
positive as the two meta-carbons, therefore the conclusion of f is identical to Fukui function f −,
namely only para- and ortho- carbons are activated for electrophilic attack by the hydroxyl group.
Exactly evaluating dual descriptor based on electron density

If you would like to evaluate f in its exact form (based on  of three states), you can follow
below steps:
examples\phenol_N+1.wfn // N+1 electron system
3 // Three files will be operated with the file that has been loaded
-,examples\phenol.wfn // N electron system
-,examples\phenol.wfn // N electron system
+,examples\phenol_N-1.wfn // N-1 electron system
-1 // Visualize isosurface of dual descriptor
The graph corresponding to isovalue of 0.01 is shown below
It can be seen that although this map is qualitatively consistent with the f map evaluated based on
spin density, the difference between ortho-carbons and meta-carbons is not so remarkable, showing
that this time f does not have good ability to discriminate preferential sites. So, using exact form
to evaluate f does not necessarily give rise to better result than using spin density to approximately
417
evaluate f !
On the "condensed" Fukui function and dual descriptor

Above we used visualization manner to examine Fukui function and dual descriptor and
obtained the conclusion what we expected. However, visual analysis is somewhat ambiguous and
subjective. Therefore sometimes we hope that the discussions of Fukui function and dual descriptor
can be quantified, namely assigning a value for each atom to exhibit the extent that it can be acted
as reactive site. To do so, one should calculate "condensed" version of Fukui function and dual
descriptor based on population analysis techniques. Since population analysis is exemplified in
Section 4.7, the method for calculating condensed Fukui function and condensed dual descriptor
will be deferred to be introduced as Section 4.7.3. Another scheme to study Fukui function and dual
descriptor is to first partition the whole molecular surface to local surface corresponding to each
atom, and then examine the their average values on these local surfaces. Because this scheme relies
on quantitative molecular surface analysis technique, illustration is deferred to Section 4.12.4.
4.5.5 Plot difference map of electron density to study electron transfer

of imidazole coordinated magnesium porphyrin
In this example, I will show you how to plot fragment electron density difference in Multiwfn.
During coordination between imidazole and magnesium porphyrin, electron transfer and
polarization occur, the variation of electron density can be clearly revealed by subtracting electron
density of imidazole (referred to as NN below) and magnesium porphyrin (referred to as MN below)
in their isolated states from the whole system (referred to as MN-NN below). The geometry of MN-
NN is shown below:
examples\MN-NN.gjf is Gaussian input file of the MN-NN system (geometry has been
optimized), run it by Gaussian after modifying the .wfn output path at the last line, then MN-NN.wfn
will be yielded. Next, respectively delete MN and NN parts from the MN-NN.gjf and properly
modify .wfn output path and then save MN.gjf and NN.gjf (which have already been provided in
"example" folder). Then run them by Gaussian to obtain MN.wfn and NN.wfn. It should be paid
attention that by default, Gaussian always puts the system to standard orientation, which makes the
418
coordinates in MN.wfn and NN.wfn inconsistent with MN-NN.wfn, and thus the density difference
will be meaningless. Therefore, nosymm keyword must be specified in route section to avoid the
automatic adjustment of coordinates. (The MN.wfn, NN.wfn and MN-NN.wfn can also be directly
loaded from here: http://sobereva.com/multiwfn/extrafiles/MN-NN.zip)
Now we generate grid data of electron density difference by Multiwfn. Boot up Multiwfn and
MN-NN.wfn
2 // Two files will be operated with MN-NN.wfn
-,MN.wfn // Will subtract property of MN.wfn from that of MN-NN.wfn
-,NN.wfn // Will subtract property of NN.wfn from that of MN-NN.wfn
1 // The property is selected as electron density
3 // Since present system is relative huge, we need more grid points than normal cases, so we
choose high quality grid
After the calculation is finished, you can choose option -1 and then set isovalue to about 0.001
to visualize the isosurface of the grid data, as shown below.
Plotting isosurface map in VMD

Multiwfn is not very professional at visualization of grid data, for large size of grid data the
visualization speed is relatively slow. You can choose option 2 to export the grid data to cube file
and then visualize it by external tools, such as VMD. Below is the graph generated by VMD (freely
available at http://www.ks.uiuc.edu/Research/vmd/) based on the cube file generated by Multiwfn.
419
The red and blue isosurfaces (+0.0012 and -0.0012 a.u., respectively) represent the region in
which electron density is increased and decreased after NN coordinated to MN, respectively. It is
obvious that electron density is shifted from backside of nitrogen in NN toward magnesium atom to
strengthen the coordination bond. Besides, it can be seen that the appearance of NN does not
perturbe electron density distribution of porphyrin ring remarkably, only slight polarization occurs
on the four coordination nitrogens in MN.
Detailed steps of drawing above graph in VMD: First, drag the cube file density.cub into VMD main window.
Select "Graphics" - "Representations", create a new representation by clicking "Create Rep" button, change the
"drawing method" to "isosurface", set "Draw" to "solid surface", set "Show" to "Isosurface", change the isovalue to
0.0012, set "coloring method" to "ColorID" and choose red. Now the isosurface of positive part of density difference
has been displayed. Then click "Create Rep" button again to create another representation, select blue in "ColorID"
and change the isovalue to -0.0012. If you would like to use white background instead of the default black
background, select Graphics - Colors - Display - Background - 8 white.
In fact, using the VMD plotting script and batch file provided by me, much better effect than
the above map can be obtained with much fewer number of steps. Please be sure to check Section
4.A.14, which illustrate how to use realize this.
Plotting contour map

Next, we plot contour map of electron density difference in the plane defined by atoms 16, 14,
9. Input following commands:
4 // Draw plane graph
0
2
-,MN.wfn
-,NN.wfn
1
4 // Define the plane by three atoms
16,14,9
420
Immediately the contour map pops up. The solid and dashed contour lines exhibit where
electron density is increased and decreased, respectively. The contour lines in the graph are a bit
sparse, so we adjust contour line setting to make the graph looks denser and thus more informative.
3 // Change contour line setting
9 // Generate contour value by geometric series
0.0001,2,30 // Start value, step size and the number of steps, respectively
y // Clean existing contour lines
9
-0.0001,2,30 // Set negative contour lines
n // Append the newly generated contour lines to existing ones
1 // Save setting and return
-1 // Redraw the graph
Below is the final graph, looks nice!
Hint: After completing the definition the contour lines, you can choose option 6 to save the setting to external
plain text file. Next time you can directly load the setting by option 7.
Note that in Multiwfn, plotting difference map for electron density or other real space functions
can be easily extended to more than two fragments cases, see Section 4.4.8 for example.
421
4.5.6 Study electron delocalization range function EDR(r;d) of anionic

water dimer
This section was contributed by Arshad Mehmood and slightly adapted by Tian Lu.
This example shows how to calculate electron delocalization range function EDR(r;d) at user-
defined length scale d for anionic water cluster (H2O)2− (cf. Phys. Chem. Chem. Phys., 17, 18305
(2015)). Using EDR(r;d) we can vividly inspect distribution of solvated electron.
Boot up Multiwfn and input following commands:
examples\solvatedelectron.wfn // Anionic water dimer optimized at B3LYP/6-311++G(2d,2p)
level
20 // EDR(r;d)
11.22 // Input length scale d (Bohr). Here we consider the relatively delocalized solvated
electron at d=11.22 Bohr. Further details are given in J. Chem. Phys., 141, 144104 (2014).
-1 // Show isosurface graph
Now a GUI window pop up. Input isovalue of 0.74 in the “Isosurface value” box and press
ENTER button. The following isosurface will appear.
This figure shows that at length scale of d=11.22 Bohr the solvated electron is between the two
H2O molecules.
4.5.7 Study orbital overlap distance function D(r) of thioformic acid

This section was contributed by Arshad Mehmood and slightly adapted by Tian Lu.
This example will show the calculation procedure of orbital overlap distance function D(r) of
thioformic acid and map it on molecular electron density surface. If you are not familiar with D(r),
you can check entry 21 of Section 2.6 or J. Chem. Theory Comput., 12, 3185 (2016).
examples\ThioformicAcid.wfn // Thioformic acid optimized at B3LYP/6-311++G(2d,2p)
21 // Orbital overlap length function D(r), which maximizes EDR(r;d) with respect to d
Now we need to set input total number, start and increment of EDR exponents αi=1/di2, since
the overlap distance is fit using an even-tempered grid of exponents. The start value is the largest
422
exponent (α1), subsequent exponents are yielded by αi+1/αi = 1/αinc, where αinc is increment. The
default setting (i.e. n=20, α1=2.50, αinc=1.50) suffices for common systems. After selecting the
manual input (option 1) or default setting (option 2), a list of exponents will be appeared, which will
be used in evaluation of D(r)
At this stage Multiwfn starts calculation. Wait until calculation is finished, then choose option
2 to export grid data of D(r) as EDRDmax.cub in current folder. The next step is to generate
molecular density isosurface.
2 // Medium quality grid (The grid setting must be the same as for D(r) calculation)
Then export grid data of electron density in current folder as density.cub by selecting option 2.
Based on the EDRDmax.cub and density.cub, then the D(r) grid data can be mapped on
molecular density isosurface by many visualization programs, such as VMD and GaussView. Below
is the D(r) mapped electron density isosurface (=0.001 a.u.) plotted by GaussView (If you do not
know how to map a real space function using various colors on isosurface of another real space
function based on two cube files via GaussView and VMD, you can consult my blog article
http://sobereva.com/402, in Chinese).
As can be seen from the graph, D(r) plotted between 2.9 Bohr (red) to 3.4 Bohr (blue) clearly
distinguishes the chemically hard oxygen lone pair (red) from the softer sulfur lone pair (blue).
This system will be further studied in Section 4.12.8 by means of quantitative molecular
surface analysis module.
4.6 Modify and check wavefunction
4.6.1 Delete certain Gaussian functions
In subfunction 25 of main function 6, you can set orbital expansion coefficients of Gaussian
423
type functions (GTFs) which satisfied certain conditions. If the coefficients are set to zero, that
means the information of these GTFs are deleted. In this example, we delete all Z-type GTFs of
atom 2,3 and 4 from orbital 23 of phenol and then plot isosurface for this molecular orbital. Boot
examples\phenol.wfn
25 // Set the coefficients of some GTFs that satisfied certain conditions
0,0 // Set the index range of GTFs, only the GTFs satisfied this condition will be reservered
to next step. 0,0 tell Multiwfn the range is “ALL”
2,4 // Only the GTFs attributed to atom 2, 3, 4 will be reserved to next step
Z // Only reserve Z-type GTFs to next step
23,23 // Set lower and upper limit of orbital, if they are identical, then only one orbital is
selected
0 // Set coefficients of selected GTFs in orbital 23 to zero, that is delete their information
0 // Save current wavefunction to new.wfn in current directory
You can choose option 4 and input 23 to check expansion coefficient of orbital 23 to verify if
your operation is correct. Let us compare the isosurfaces of orbital 23 before and after modification.
In order to plot the modified orbital 23, you can close Multiwfn and load new.wfn, or directly select
option -1 to return main menu, and then enter main function 0 to plot its isosurface.
Left side is unmodified state, the molecular plane is parallel to XY plane, so if Z-type GTFs in
some atoms are deleted, the corresponding part of isosurface should disappear, this is what we have
seen at right side (modified state).
4.6.2 Remove contributions from certain orbitals to real space

functions
In Multiwfn, contributions from certain orbitals to real space functions can be removed. From
formulae in Section 2.6 it is clear that if occupation numbers of certain orbitals are set to zero, they
will have no contribution, as if they were not existed. Actually in Section 4.5.3 we have already
used this trick to separate ELF as ELF-α and ELF-π. Notice that some real space functions are not
linear, such as ELF and LOL, so they cannot be computed as the sum of contributions from each
occupied orbitals. In contrast, some real space functions are linear, such as electron density and
kinetic energy density.
In this example, we will remove contributions from the MOs consisting of inner-core atomic
424
orbitals to electron density for hydrogen cyanide, so that we can obtained valence electron density
map. This is known as valence electron density analysis, which was demonstrated to be quite
powerful for analyzing molecular electronic structure. Lots of examples of this kind of analysis can
be found from my paper "Revealing Molecular Electronic Structure via Analysis of Valence
Electron Density" Acta Phys. -Chim. Sin., 34, 503 (2018).
The realization of the valence density analysis is very simple. Inner-core atomic orbitals always
have very low energies, so they can only contribute to the MOs with lowest energies. Hydrogen has
no inner-core atomic orbital, while both carbon and nitrogen have an inner-core atomic orbital,
hence what we need to do is to set occupation numbers of the first two MOs to zero (note that this
is closed-shell wavefunction). Now, boot up Multiwfn, and input following commands
examples\HCN.wfn
6 //Modifying wavefunction
26 // Set occupation numbers
1,2 // Select MO 1 and 2
0 // Set their occupation numbers to zero
q // Return to last menu
-1 // Return to main manu
Then, if we plot color-filled map of electron density as usual by main function 4, we will get
below graph
From the picture, the bonding region of C-N and C-H and be easily identified, and the lone
pair region of nitrogen can be clearly visualized. Note that the small white circles in carbon and
nitrogen centers do not correspond to 1s electrons, but result from the fact that valence atomic
425
orbitals have penetration effect into core region.

For convenience consideration, Multiwfn provides subfunction 34 in main function 6, one can
directly choose it to set occupation numbers of all the MOs that consist of inner-core atomic orbitals
to zero.
It is worth to mention that analysis of valence electron density is never limited to plotting
analysis. For example, using main function 2 and 17, topology analysis and basin analysis can also
be straightforwardly applied to valence electron density, respectively.
4.6.3 Translate and duplicate graphene primitive cell wavefunction to

periodic system
The .fch or .wfn file generated by periodic boundary condition (PBC) calculation of Gaussian
only contains wavefunction of primitive cell, so the analysis results do not show any periodic
character, of course you can enlarge your inputted system to multiple cells before PBC calculation,
but you have to spend much more time for computation. Multiwfn provides a way to convert
primitive cell wavefunction to a large supercell wavefunction, so that in the region you are interested
the real space functions show periodic character.
Note: Beware that constructing supercell wavefunction in this manner is only a crude approximation! Since the
orbital mix between neighbouring primitive cells is ignored, therefore the orbital wavefunctions of supercell cannot
be faithfully reproduced.
In this section I use graphene as example. First, generate .fch (or .wfn) file of graphene
primitive cell. The content of “molecular specification” field in the Gaussian input file is
C 0.000000 0.000000 0.000000
C 0.000000 1.429118 0.000000
TV 2.475315 0.000000 0.000000
TV -1.219952 2.133447 0.000000
In route section, write “#P PBEPBE/3-21g/Auto SCF=Tight”. Use Gaussian to run this input
file and then use formchk to convert the binary checkpoint file to graphene.fch. Then boot up
Multiwfn and input:
examples\graphene.fch
32 // Translate and duplicate primitive cell wavefunction
2.475315 0.000000 0.000000 // Translation vector 1
2 // Unit is Å
3 // Translate and duplicate present system three times in this direction
32 // Notice that current system already have four primitive cells, this time we will translate
and duplicate current system in another direction three times, so the final system will contain 16
primitive cells
-1.219952 2.133447 0.000000 // Translation vector 2
2 // Unit is Å
3 // Translate and duplicate present system three times in this direction
The left part of the picture below is LOL function of primitive cell. After above manipulation,
we recalculate LOL function and then the right graph is obtained (black arrows denote translation
vectors). Apparently, the central region of the extended system shows correct periodic character,
426
however the behavior of boundary region is still incorrect, you can extend the system further to
enlarge “correct” region.
Notice that if nosymm keyword is not specified in PBC calculation, Gaussian may
automatically put the system into standard orientiation, at this time you should not use the translation
vectors in Gaussian input file as the translation vectors for translating and duplicating system in
Multiwfn, but should use the content in “Translation vectors” field of .fch file or “PBC vector”
segment in Gaussian output file.
4.7 Population analysis and atomic charge calculation
4.7.0 Mulliken population analysis on triplet ethanol
In this section I will illustrate how to use Multiwfn to carry out Mulliken analysis, triplet
ethanol is taken as instance. It is worth to note that Mulliken analysis is incompatible with diffuse
functions, if diffuse functions are employed, the analysis result will be meaningless.
examples\ethanol_triplet.fch // Calculated at UB3LYP/6-31G** level based on optimized
singlet structure
7 // Population analysis and atomic charges
5 // Mulliken population analysis
1 // Output Mulliken analysis result. By default the result is outputted on screen, you can also
select "-1 Choose output destination for option 1" to change the output destination to a specified
plain text file
From the output, first you can find population of each basis function:
Population of basis functions:
Basis Type Atom Shell Alpha pop. Beta pop. Total pop. Spin pop.
1 S 1(C ) 1 0.99597 0.99597 1.99193 -0.00000
2 S 1(C ) 2 0.34180 0.34330 0.68510 -0.00150
3 X 1(C ) 3 0.34747 0.35290 0.70037 -0.00543
4 Y 1(C ) 3 0.34869 0.35220 0.70088 -0.00351
427
...
60 Z 8(O ) 30 0.64840 0.09113 0.73953 0.55726
61 S 8(O ) 31 0.30323 0.47550 0.77873 -0.17228
...
Since present system is an open-shell system, not only the total population (i.e. alpha+beta), but also
alpha and beta populations are outputted individually. The spin populations, which equal to
difference between alpha and beta populations, are also printed. The output content is easy to
understand, for example, from the output we can see that there are nearly two electrons located on
the first S basis function of C1 atom, and one of PZ basis functions of O8 atom has large amount of
unpaired electrons (0.557).
Next, we can find population of each basis function shell of each atom:
Population of shells:
Shell Type Atom Alpha pop. Beta pop. Total pop. Spin pop.
1 S 1(C ) 0.99597 0.99597 1.99193 -0.00000
2 S 1(C ) 0.34180 0.34330 0.68510 -0.00150
3 P 1(C ) 1.05533 1.06504 2.12037 -0.00971
4 S 1(C ) 0.30449 0.30842 0.61291 -0.00393
...
28 S 8(O ) 0.99668 0.99637 1.99305 0.00032
29 S 8(O ) 0.51929 0.45809 0.97739 0.06120
30 P 8(O ) 1.71305 1.06458 2.77763 0.64846
...
As you can see, for example, the basis function shell 30, which corresponds to one of P shells of O8,
has unpaired electrons of 0.648 and total population of 2.777.
Next, you can find population of each angular moment atomic orbitals of each atom:
Population of each type of angular moment atomic orbitals:
Atom Type Alpha pop. Beta pop. Total pop. Spin pop.
...
8(O ) s 1.81920 1.92996 3.74916 -0.11076
p 2.49431 1.65216 4.14647 0.84216
d 0.00748 0.00607 0.01355 0.00142
...
Total s 8.35987 7.42761 15.78748 0.93226
p 5.61927 4.56478 10.18405 1.05448
d 0.02087 0.00761 0.02847 0.01326
The output shows that d type of atomic orbitals only have marginal contribution to total population
(0.02847) and spin population (0.01326) of the whole system, since D type of basis functions only
behave as polarization functions for present system. In O8, most unpaired alpha electrons are located
on its p atomic orbitals, while slight unpaired beta electrons are distributed on its s atomic orbitals
(positive and negative value indicate that the unpaired electrons are alpha and beta, respectively).
Finally, we can find atomic populations and atomic charges:
428
Population of atoms:
Atom Alpha pop. Beta pop. Spin pop. Atomic charge
1(C ) 3.15460 3.16849 -0.01388 -0.32309
2(H ) 0.43857 0.43159 0.00698 0.12984
3(H ) 0.43857 0.43159 0.00698 0.12984
4(H ) 0.46846 0.43410 0.03437 0.09744
5(C ) 3.13963 2.95977 0.17986 -0.09940
6(H ) 0.49613 0.36169 0.13444 0.14218
7(H ) 0.49613 0.36169 0.13444 0.14218
8(O ) 4.32100 3.58819 0.73281 0.09082
9(H ) 1.04690 0.26289 0.78401 -0.30979
Total net charge: -0.00000 Total spin electrons: 2.00000
Triplet system has two unpaired electrons, one can see that in the triplet ethanol, most unpaired
electrons (more than 1.5) are located on the hydroxyl group. It is well known that in ground state
ethanol, oxygen atom should have significant negative charge due to very large electronegativity of
oxygen. However, in present system, the oxygen even carries marginal positive charge. This
observation reflects the fact that electronic structure of different electronic states may differ from
each other remarkably.
Mulliken population analysis does not show population information of each atomic orbital,
however, if you first identify correspondence between basis functions and atomic orbitals (see
Section 4.7.6 for details), you can easily obtain population of each atomic orbital by simply
summing up population of corresponding basis functions.
There are several other options in the Mulliken analysis interface, they can help you to gain
deeper insight into electronic population, please play with them by consulting corresponding
explanation in Section 3.9.3.
4.7.1 Calculate Hirshfeld and CHELPG atomic charges as well as

fragment charge for chlorine trifluoride
Calculating Hirshfeld charges

I have introduced the theory of Hirshfeld population in Section 3.9.1, to calculate Hirshfeld
charges for ClF3, input below commands in Multiwfn
examples\ClF3.wfn
1 // Hirshfeld population
Hirshfeld population analysis requires electron density of atoms in their free-states, you need
to choose a method to calculate atomic densities. Selecting 1 to use built-in atomic densities is very
convenient, see Appendix 3 for detail; alternatively, you can select 2 to evaluate atomic densities
based on atomic .wfn files, see Secion 3.7.3 for detail. Here we choose option 1. Now you can see
below output, not only the atomic charges, but also the dipole moment evaluated based on the atomic
charges is printed.
Hirshfeld charge of atom 1(Cl) is 0.523322
Hirshfeld charge of atom 2(F ) is -0.223879
429
Summing up all charges: 0.00001373
Total dipole moment from atomic charges: 0.309313 a.u.
X/Y/Z of dipole moment from atomic charges: 0.000000 -0.000000 0.309313 a.u.
From the result we find the charges of the three fluorine atoms are unequal, the equatorial one
(F3) is -0.075, while the axial ones (F2 and F4) possess more electrons, their charges are thus more
negative, that is -0.224.
As shown on the screen, the sum of all calculated charges is 0.00001373 rather than exactly
zero as we expected, this is due to unavoidable numerical error of space integration. Considering
this, Multiwfn also prints the result after normalization to eliminate the marginal numerical error:
Final atomic charges, after normalization to actual number of electrons
Atom 1(Cl): 0.523317
Atom 2(F ): -0.223882
Atom 3(F ): -0.075553
Atom 4(F ): -0.223882
When using the calculated atomic charges for your research and article, adapting the normalized
charges is recommended, because their sum is exactly identical to net charge of current system.
Finally Multiwfn asks you if exporting the result, if you select y, the element names, atom
coordinates and atomic charges will be outputted to a plain text file with .chg extension, see Section
2.6 of introduction of .chg format. You can use this file as Multiwfn input file and select
“Electrostatic potential from atomic charges” in main function 3, 4, 5 (or other functions) to study
the electrostatic potential derived from Hirshfeld charges.
Calculating CHELPG charges

Next, we calculate CHELPG charge. CHELPG charge has been introduced in Section 3.9.10.
First, select subfunction 12 in the population analysis module, you will see a new menu. In general
you do not need to modify the default options, and you can directly select option 1 to start the
calculation. Since calculation of ESP is time-consuming, for large system you may need to wait for
a while. The result is 0.5772 for Cl, -0.2496 for axial F and -0.0779 for equatorial F. The conclusion
of CHELPG charge is the same as Hirshfeld charge, namely axial F are more negatively charged
than the equatorial one.
Quickly evaluating fragment charge

Fragment charge is defined as sum of charge of atoms constituting a fragment. You can
manually sum up atomic charges to derive fragment charge; however, for large systems this process
must be laborious. In Multiwfn it is possible to directly calculate charge for a fragment. For example,
here we calculate CHELPG charge for the fragment composed by the two axial F atoms. Boot up
Multiwfn and input
examples\ClF3.wfn
7 // Population analysis
-1 // Define fragment
2,4 // Index of the two axial F atoms
12 // CHELPG charge
1 // Start calculation
430
Since the fragment has been defined, Multiwfn not only prints atomic charges, but also prints
fragment charge at the end of all output:
Fragment charge: -0.499331
Hint: Calculation of ESP fitting charges, including CHELPG, MK and so on, requires electrostatic potential
(ESP) at fitting points. The calculation speed of ESP of cubegen utility in Gaussian package is usually faster than
the internal code of Multiwfn. If you have Gaussian installed on your system and the input file is .fch/fchk, it is
recommended to set "cubegenpath" parameter in settings.ini file to actual path of cubegen, so that cubegen could be
automatically invoked by Multiwfn to evaluate the ESP at fitting points. More information about this point can be
found in Section 5.7.
4.7.2 Calculate and compare ADCH atomic charges with Hirshfeld

atomic charges for acetamide
The ADCH (atomic dipole moment corrected Hirshfeld population) charge proposed by me is
an improved version of Hirshfeld charge, it resolved many inherent drawbacks of Hirshfeld charge,
such as poor dipole moment reproducibility, see Section 3.9.9 for brief introduction and J. Theor.
Comput. Chem., 11, 163 (2012) for discussion and comparison. I highly recommend using ADCH
charge to characterize charge distribution. The calculation process of ADCH charges is exactly
identical to the one described in last section, the only difference is that you should select option 11
instead of option 1 in population analysis interface. For example, here we calculate ADCH charges
for CH3CONH2. Boot up Multiwfn and input
examples\CH3CONH2.fch
11 // Calculate ADCH charges
1 // Use built-in atomic densities in free-state
Multiwfn will calculate Hirshfeld charges first, and then perform atomic dipole moment
correction for them to yield ADCH charges. The result is shown below
======= Summary of atomic dipole moment corrected (ADC) charges =======
Atom: 1C Corrected charge: -0.265840 Before: -0.090370
Atom: 2H Corrected charge: 0.096194 Before: 0.037254
Atom: 5C Corrected charge: 0.272281 Before: 0.170596
Atom: 6O Corrected charge: -0.364414 Before: -0.308866
Atom: 7N Corrected charge: -0.677574 Before: -0.159120
Summing up all corrected charges: -0.0000816
Note: The values shown after "Corrected charge" are ADCH charges, the ones afte
r "Before" are Hirshfeld charges
Total dipole from ADC charges (a.u.) 1.4368131 Error: 0.0001385
X/Y/Z of dipole moment from the charge (a.u.) 0.0432390 -1.4253486 0.1759079
It is obvious that for all atoms, the magnitude of ADCH charges are evidently larger than
Hirshfeld charges, the former are in agreement with common chemical senses, while the latter turns
431
out to be too small.

A remarkable feature of ADCH charges is that the molecular dipole moment can be exactly
reproduced. The dipole moment derived from ADCH charges is 1.4368 a.u. (as shown above), which
is exactly identical to the actual dipole moment, namely the one derived based on present electron
density distribution. (The error 0.0001385 comes from trivial numerical aspects and is totally
negligible).
If you scroll up the command-line window, you will find the following information
Total dipole from atomic charges: 1.073849 a.u.
This is the dipole moment derived from Hirshfeld charges, which deviates to actual dipole moment
(1.4368 a.u.) apparently. In fact, for almost all small molecules, Hirshfeld charges always severely
underestimate molecular dipole moments.
Again, it is recommended to employ the atomic charges after normalization (namely the ones
printed under "Final atomic charges" label).
4.7.3 Calculate condensed Fukui function and condensed dual

descriptor
Theory
In Section 4.5.4, I have introduced how to calculate and visualize Fukui function and dual
descriptor. In this section, we will calculate "condensed" version of these two functions, so that the
discussion of the possibility that an atom could act as a reactive site can be upgraded to quantitative
level. Phenol will still be used as example case.
Before calculating phenol, we first derive the expression of condensed Fukui function and
condensed dual descriptor. In the condensed version, atomic population number is used to represent
the amount of electron density distribution around an atom. Recall the definition of Fukui function
f +:
f + (r) =  N +1 (r) −  N (r)

The definition of condensed Fukui function for an atom, say A, can be written as
f A+ = pNA +1 − pNA
where pA is the electron population number of atom A.
Since atomic charge is defined as q = Z − p , where Z is the charge of atomic nuclear, the
A A A
f + can be expresssed as the difference of atomic charges in two states (note that the two Z terms are
cancelled)
f A+ = qNA − qNA +1
By analogous treatments, one can easily formulate other types of condensed Fukui function
Nucleophilic attack : f A+ = qNA − qNA +1
Electrophilic attack : f A− = qNA −1 − qNA
Radical attack : f A0 = ( qNA −1 − qNA +1 ) / 2
432
Similarly, condensed dual descriptor can be written as
f A = f A+ − f A− = (qNA − qNA +1 ) − (qNA −1 − qNA ) = 2qNA − qNA +1 − qNA −1

There are numerous ways to calculate atomic charges, although currently there is no consensus
on which method is the most ideal one to study condensed Fukui function and dual descriptor, but
at least Hirshfeld charge has proven to be a very suitable choice. For example, J. Phys. Chem. A,
106, 3885 (2002), J. Phys. Chem. A, 107, 10428 (2003) and my work J. Phys. Chem. A, 118, 3698
(2014) illustrated Hirshfeld charge can be successfully used to study reactive site in terms of
condensed Fukui function. Furthermore, a comprehensive comparison given in Theor. Chem. Acc.,
138, 124 (2019) demonstrated that Hirshfeld charge may be the best choice of evaluating condensed
Fukui function.
Practical example: Phenol

Here we calculate the condensed Fukui function and dual descriptor for phenol, we still use the
phenol.wfn, phenol_N+1.wfn and phenol_N-1.wfn in "examples" folder, which have already been
utilized in Section 4.5.4
According to the procedure introduced in Section 4.5.4, we calculate Hirshfeld charges for all
carbons in phenol in its N, N+1 and N-1 electrons states, respectively, they are given in below table.
Then, according to the formulae shown above, the condensed f −, f +
and dual descriptor can be
readily calculated, as shown below.
N N-1 N+1 f− f+ Δf
C1 (p) -0.059 0.085 -0.119 0.144 0.060 -0.084
C2 (m) -0.039 0.027 -0.167 0.066 0.128 0.063
C3 (o) -0.060 0.032 -0.187 0.092 0.128 0.036
C4 0.074 0.174 0.022 0.100 0.052 -0.048
C5 (o) -0.073 0.009 -0.196 0.082 0.123 0.040
C6 (m) -0.041 0.034 -0.173 0.075 0.131 0.056
Note: The Hirshfeld charges reported above were estimated based on build-in sphericalized atomic densities.
For f −, the smallest two values occur at C2 and C6, therefore meta atoms are unfavorable sites
for electrophilic attack.
For dual descriptor, the most positive values occur at C2 and C6, suggesting that they are the
most unfavorable sites for electrophilic attack. C1 has a large negative value and hence favored by
electrophilic reactant. Although the two ortho carbons (C3 and C5) have positive value, its
magnitude is not as large as meta carbons, so dual descriptor indicates that ortho carbons are more
possible than meta carbons to be reactive site for electrophilic attack. Our conclusion is completely
in line with that of Section 4.5.4, in which we obtained the conclusion by visual inspecting
isosurface of Fukui function and dual descriptor.
IMPORTANT NOTE: In daily research, I strongly suggest you directly use subfunction 16 of
main function 100 to automatically calculate condensed Fukui function and dual descriptor, because
433
the steps are extremely simple and meantime other useful quantities in conceptual density functional
theory can be printed together, see Section 3.100.16 for introduction and 4.100.16.1 for example.
4.7.4 Illustration of computing Hirshfeld-I atomic charges
Hirshfeld-I (HI) is a more advanced technique to define atomic spaces than its predecessor
(Hirshfeld). Before following the example given below please briefly read Section 3.9.13 to gain
basic knowledge of HI method and its implementation in Multiwfn. It is very important to note that
in order to calculate HI charge, atomic radial densities files (.rad) of various elements in the current
system at different charged states must be available. Commonly, I suggest you directly use the built-
in .rad files in Multiwfn, so that you do not need to generate them before HI calculation. See Section
3.9.13 for detail about this point.
Here we calculate HI charges for CH3COCl. For convenience, we will directly use built-in .rad
files in this example. To do so, we copy "atmrad" folder from "examples" directory to current
directory, then the .rad files in this folder will be employed by Multiwfn in the HI charge calculation.
examples\CH3COCl.wfn // Generated at B3LYP/6-31G* level
15 // Hirshfeld-I method
1 // Start calculation with default settings
Then you will see iteration process
Performing Hirshfeld-I iteration to refine atomic spaces...
Cycle 1
Cycle 2 Maximum change: 0.202864
[ignored]
The "maximum change" denotes the maximum change of HI atomic charges, the iteration
continues until "maximum change" is lower than the threshold, which is 0.0002 by default. After
convergence, Multiwfn prints final HI atomic charges:
Atom 1(C ): -0.64593923
Atom 2(H ): 0.18144874
Atom 3(H ): 0.17983150
Atom 4(H ): 0.18144874
Atom 5(C ): 0.72667580
Atom 6(O ): -0.42009700
Atom 7(Cl): -0.20336855
Then you can choose if outputting these charges to .chg file in current folder. I suggest you
compare above result with Hirshfeld charges, you will find the magnitude of HI charges is much
higher than Hirshfeld charges. This phenomenon is expected, because HI atomic spaces properly
contract or expand with respect to that of neutral state according to actual chemical environment,
hence the size difference of atomic space among various atoms is greatly increased.
434
Letting Multiwfn automatically invoke Gaussian to generate .rad files

In principle, it may be best to generate atomic .rad files at the same level as the current molecule,
since in this case the result has strongest physical meaning. You can directly let Multiwfn to invoke
Gaussian to prepare the .rad files.
Before calculation, you should properly set "gaupath" in settings.ini file to actual Gaussian
executable file. In addition, if "atmrad" folder has existed in current directory and it contains .rad
files of C, H, O, and Cl elements, you should delete them.
examples\CH3COCl.wfn // Generated at B3LYP/6-31G* level
15 // Hirshfeld-I method
1 // Start calculation with default settings
B3LYP/6-31G* // The keyword of Gaussian used to calculate atomic .wfn files
From the prompts shown on screen, you can find that Multiwfn invokes Gaussian to calculate
atomic .wfn files for all elements involved in the present molecule at various charged states. Then
Multiwfn converts atomic .wfn files to .rad files, which record spherically averaged atomic radial
densities. The automatically generated Gaussian input file (.gjf), the resulting Gaussian output file
(.out or .log) and the .rad files are all produced in "atmrad" folder of current folder, you can manually
examine them if you have interesting.
In the current case, the resulting charges are
Atom 1(C ): -0.667920
Atom 2(H ): 0.194766
Atom 3(H ): 0.188545
Atom 4(H ): 0.194766
Atom 5(C ): 0.755439
Atom 6(O ): -0.414041
Atom 7(Cl): -0.251555
As can be seen, the HI charges calculated based on the .rad files generated at B3LYP/6-31G*
level are basically the same as those calculated based on built-in .rad files, therefore commonly I
suggest directly using built-in .rad files since the calculation is easier in this case and Gaussian is
not needed.
If you do not delete the "atmrad" folder or clean it up, then when you recalculate HI charges
for CH3COCl, or calculate a molecule only consisting of C, H, O and Cl elements or some of them,
Multiwfn will directly perform HI calculation based on the existing .rad files in the "atmrad" folder
rather than invoke Gaussian to recalculate them.
Finally, it is noteworthy that the support of Hirshfeld-I in Multiwfn is never limited to

population analysis, this partition method can also be used to calculate orbital composition (main
function 8), and be applied in fuzzy analysis module (main function 15).
4.7.5 Calculating EEM atomic charges for ethanol-water cluster
Please first read Section 3.9.15 to understand basic features of Electronegativity Equalization
Method (EEM) charges before following this example. Here we calculate EEM charges for ethanol-
water cluster, which contains as many as 492 atoms:
435
Evidently, calculating atomic charges by quantum chemistry method for such a large system is
too expensive; however, as you will see, evaluation of EEM charges even for a system composed
by hundreds of atoms is rather easy.
Note that in order to calculate EEM charges in Multiwfn, currently you have to use MDL
molfile (.mol) or .mol2 as input file, because only this file provides atomic connectivity information,
which is needed in the calculation of EEM charges.
Boot up Multiwfn and input below commands
examples\ethanol_water.mol // This is a snapshot of molecular dynamics simulation
17 // EEM charge
You will immediately see
EEM charge of atom 1(O ): -0.658886
EEM charge of atom 2(H ): 0.322520
...
EEM charge of atom 488(C ): -0.071564
EEM charge of atom 491(O ): -0.616013
Electronegativity: 2.454144
The default EEM parameters were fitted by some researchers for reproducing B3LYP/6-31G*
CHELPG charges, therefore, the above EEM charges should be close to CHELPG charges evaluated
at B3LYP/6-31G* level (In fact, for present system, even if calculation of CHELPG charges is
feasible, the result should be much worse than the EEM charges we just obtained. Because it is well-
known that the quality of electrostatic fitting charges is very low for the atoms far from van der
Waals surface, while in present system there are numerous heavily buried atoms).
Note that there are also many other built-in EEM parameters, you can choose them via option
1 before calculation.
436
Note on calculating EEM charge of the system containing  conjugation

It is worth to note if one or more atoms are in  conjugation region, the conjugation must be
represented as Lewis structure in the inputted .mol or .mol2 file, otherwise the calculation cannot
be conducted.
For example, we use GaussView to create an azobenzene molecule:
Save it as .mol file, then use Multiwfn to calculate EEM charge based on this file, you will find
below error:
Error: Multiplicity of atom 1 ( 4) exceeded upper limit ( 3)!
The present EEM parameters do not support such bonding status, or connectivity
in your input file is wrong
To understand the reason, open the .mol file by text editor, you can find below two lines
1 2 4 0 0 0 0
1 6 4 0 0 0 0
which indicates that the bond multiplicity of 1-2 and 1-6 is 4, clearly this is unreasonable, because
formal bond order between two carbons cannot be four! This issue comes from the fact that
GaussView always records a conjugated bond in .mol as a quadruple bond. To solve this problem,
the best way is installing OpenBabel (freely available at http://openbabel.org), then use this
command to convert the previous .mol file to a new .mol file: obabel old.mol -O new.mol. Then if
you use GaussView to open the new.mol, you will find the bonding has fully satisfied Lewis structure:
Now we use Multiwfn to calculate its EEM charges again, you will find below output, which
is quite reasonable:
...ignored
EEM charge of atom 12(N ): -0.3186294612
EEM charge of atom 13(N ): -0.3186294612
437
...ignored
4.7.6 Determining correspondence between basis functions and atomic

orbitals via population analysis
Determine correspondence between basis functions and atomic orbitals is important if one want
to plot PDOS of some atomic orbitals via main function 10, or to evaluate contribution to molecular
orbitals from specific atomic orbitals using main function 8. The correspondence is easy to be
identified if Pople basis set is used. For example, 6-31G* implies using one basis function with
contraction degree of 6 to represent each inner atomic orbital, while each valence atomic orbital is
represented by a basis function with contraction degree of 3 and an uncontracted basis function.
However, for most of other type of basis sets, the correspondence is often difficult to be determined.
Fortunately, as will be illustrated in this section, if one studies total and spin population of basis
function shells via Mulliken population analysis, the correspondence can be unambiguously
identified.
Two typical examples will be given below, more examples and discussions can be found from
my blog article (http://sobereva.com/418, in Chinese). In below text atomic orbitals will be denoted
as lower case (e.g. s, p, d...), while basis functions will be written as upper case (e.g. S, P, D...).
Example 1: cc-pVTZ for sulfur

Sulful atom has configuration of 1s22s22p63s23p4, the ground state is triplet. The
examples\sulfur_cc-pVTZ.fch is the .fch file calculated at B3LYP/cc-pVTZ level by Gaussian16 for
a single sulfur atom at its triplet state. Load this file into Multiwfn, then input
5 // Mulliken analysis
1 // Output Mulliken analysis result
Shell Type Atom Alpha_pop. Beta_pop. Total_pop. Spin_pop.
1 S 1(S ) 0.99997 0.99997 1.99994 -0.00000
2 S 1(S ) 0.94424 0.94387 1.88812 0.00037
3 S 1(S ) 0.60678 0.56547 1.17225 0.04131
4 S 1(S ) 0.12371 0.13260 0.25631 -0.00889
5 S 1(S ) 0.32408 0.35782 0.68189 -0.03374
6 P 1(S ) 2.93141 2.90842 5.83983 0.02299
7 P 1(S ) 1.58628 0.47788 2.06416 1.10840
8 P 1(S ) 0.59303 0.26385 0.85688 0.32918
9 P 1(S ) 0.88863 0.34985 1.23848 0.53879
10 D 1(S ) 0.00064 0.00017 0.00081 0.00048
11 D 1(S ) 0.00058 0.00011 0.00069 0.00046
12 F 1(S ) 0.00065 0.00000 0.00065 0.00064
We want to identify which S basis functions respectively correspond to 1s, 2s and 3s atomic
orbitals, and which P basis function shells respectively correspond to 2p and 3p atomic orbital shells.
All the two unpaired electrons of triplet sulfur atom are distributed on 3p shell, since the sum
of spin population of 7P, 8P and 9P is 1.10840+0.32918+0.53879=1.976, which is nearly equal to
438
two, we can say that these three P shells correspond to 3p shell. The remainder 6P shell clearly
corresponds to 2p shell, this can also be confirmed that its population number is 5.840, which is
close to expected occupation number of 2p shell (6.0).
Then we check the case of S shells. The sum of population number of 3S, 4S and 5S is
1.17225+0.25631+0.68189=2.110, which is close to actual occupation number of 3s atomic orbital
(2.0); considering that occupation number of both 1S and 2S are close to 2.0, it can be concluded
that 1S, 2S and (3S,4S,5S) mainly represent 1s, 2s and 3s atomic orbitals, respectively.
Example 2: def2-TZVP for Au

For Au atom, def2-TZVP is a pseudopotential basis set with Stuttgart small core
pseudopotential, 60 inner electrons are replaced with pseudopotential, therefore only the valence
electrons 5s25p65d106s1 are explicitly represented by the def2-TZVP basis set. The
examples\Au_def2-TZVP.fch is the .fch file calculated at B3LYP/def2-TZVP level by Gaussian16
for a single Au atom at its ground state (doublet state). Load this file into Multiwfn and carry out
population analysis as the last example, you will see
Shell Type Atom Alpha_pop. Beta_pop. Total_pop. Spin_pop.
1 S 1(Au) 0.01764 0.01588 0.03352 0.00175
2 S 1(Au) -0.25037 -0.22638 -0.47675 -0.02399
3 S 1(Au) 0.90648 0.84814 1.75462 0.05834
4 S 1(Au) 0.33047 0.35849 0.68895 -0.02802
5 S 1(Au) 0.60949 0.00436 0.61385 0.60513
6 S 1(Au) 0.38630 -0.00049 0.38581 0.38679
7 P 1(Au) 1.32302 1.32562 2.64864 -0.00260
8 P 1(Au) 1.43496 1.44146 2.87642 -0.00650
9 P 1(Au) 0.24145 0.23250 0.47395 0.00896
10 P 1(Au) 0.00057 0.00042 0.00099 0.00015
11 D 1(Au) 3.11664 3.17522 6.29186 -0.05859
12 D 1(Au) 1.48961 1.45237 2.94198 0.03724
13 D 1(Au) 0.39375 0.37241 0.76616 0.02135
14 F 1(Au) 0.00000 0.00000 0.00000 0.00000
Undoubtedly, all P shells (7P, 8P, 9P, 10P) represent the only p shell (5p), while all D shells
(11D, 12D, 13D) represents the only d shell (5d). Since the sum of population number of 5S and 6S
(i.e. 0.61385+0.38581) is exactly equal to 1.0, and meantime the sum of spin population number of
5S and 6S is also equal to 1.0, it is clear that the 5S and 6S shells collectively represent the 6s atomic
orbital, which has a single unpaired electron. The total electrons in the other four S shells (1S, 2S,
3S, 4S) is almost exactly 2.0, evidently the doubly occupied 5s is mainly represented by them.
4.7.7 Illustration of deriving RESP charges and normal ESP fitting

charges with extra constraints
In this section I will take many examples to substantially illustrate the use of the extremely
powerful and flexible RESP module of Multiwfn, which can very conveniently calculate standard
RESP atomic charges and normal ESP fitting charges with/without charge and equivalence
constraints. Reading Section 3.9.16 is strongly recommended so that you have enough knowledges
439
about the RESP module as well as adequate understanding on the idea of ESP fitting method.
More detailed descriptions and discussions can be found from my blog article "Principle of
RESP charge and its calculation in Multiwfn" (in Chinese, http://sobereva.com/441).
For saving space, only the most important files involved in below examples are provided in
"examples\RESP" folder, while other files, including Gaussian output files and .fch files, can be
downloaded at http://sobereva.com/multiwfn/extrafiles/RESP.zip.
In this section, only example of deriving RESP charges for ground state is given. It is also easy to calculate
RESP charges for excited state. You can follow this example http://sobereva.com/wfnbbs/viewtopic.php?pid=747 if
you are a Gaussian user.
Hint: If you have Gaussian on your machine and input file is .fch/fchk, it is suggested to allow Multiwfn to
invoke cubegen utility in Gaussian package to evaluate ESP at fitting points, usually the computational cost of
calculating ESP fitting charges could be detectably reduced. See Section 5.7 on how to realize this.
4.7.7.1 Example 1: Deriving RESP charges for dopamine in ethanol

environment
In this section I introduce the procedure of calculating standard RESP atomic charges for
dopamine. Ethanol solvent environment is assumed and it will be represented using IEFPCM
implicit solvation model. The structure of dopamine is shown as below.
Commonly, the geometry used for deriving RESP charges should be optimized at reasonable
level. Above geometry was optimized at B3LYP-D3(BJ)/6-311G** level with IEFPCM implicit
solvation model, and it was found to be the most stable geometry of present molecule.
Now, use Gaussian to run examples\RESP\dopamine-single\dopamine.gjf to generate
corresponding .fch file for this geometry. As can be seen in the .gjf file, the keywords are b3lyp/6-
311g(d,p) SCRF=solvent=ethanol, this combination is not expensive while the resulting
wavefunction is completely adequate to yield reliable RESP charges.
dopmaine.fch // The .fch file just yielded
18 // RESP module
1 // Calculate standard RESP charges using two-stage fitting procedure
During the calculation, Multiwfn first sets up atomic radii and determines position of fitting
points, and then calculates ESP values at the fitting points. After that, the first stage of standard
RESP calculation starts, the parameters and conditions employed in this stage can be found from
outputted information:
No charge constraint is imposed in this stage
440
No atom equivalence constraint is imposed in this fitting stage
**** Stage 1: RESP fitting under weak hyperbolic penalty
Convergence criterion: 0.00000100
Hyperbolic restraint strength (a): 0.000500 Tightness (b): 0.100000
Iter: 1 Maximum charge variation: 1.0067306224
Successfully converged!
As you can see, variation of atomic charges converges after 7 cycles in this stage. Then the
second stage starts:
**** Stage 2: RESP fitting under strong hyperbolic penalty
Atoms equivalence constraint imposed in this fitting stage:
Constraint 1: 12(H ) 13(H )
Constraint 2: 14(H ) 15(H )
Fitting objects: sp3 carbons, methyl carbons and hydrogens attached to them
Indices of these atoms:
4C 12H 13H 6C 14H 15H
Convergence criterion: 0.00000100
Hyperbolic restraint strength (a): 0.001000 Tightness (b): 0.100000
Successfully converged!
As indicated in the output, in the second fitting stage, the two hydrogens at each of the two
−CH2− groups are required to be equivalent during the fitting. In addition, charges of only six atoms
are fitted in the stage 2, they are carbons and hydrogens in the two −CH2− groups, while charges of
other atoms keep unchanged at the values yielded in fitting stage 1.
The resulting RESP charges are
Center Charge
1(O ) -0.5406621847
2(O ) -0.5360154461
...[ignored]
12(H ) 0.0727189426
13(H ) 0.0727189426
14(H ) -0.0640826813
15(H ) -0.0640826813
...[ignored]
Sum of charges: 0.000000
441
RMSE: 0.002097 RRMSE: 0.110457
If you examine the charges carefully, you will find all charges are chemically meaningful. The
RMSE and RRMSE are not large, implying that quality of ESP fitting is nice. One can see that
equivalence constraints indeed work, the H12 and H13 share the same charge 0.0727, while both
H14 and H15 have charge of -0.064.
Directly loading fitting points and ESP values from Gaussian output file
As mentioned in Section 3.9.16.2, during calculation of ESP fitting charges in the RESP
module, it is possible to make Multiwfn directly load fitting points and ESP values from Gaussian
output file of pop=MK or pop=CHELPG task. As an illustration, the Gaussian input file of
dopamine for this purpose is provided as examples\RESP\dopamine-single\dopamine_pop_MK.gjf,
use Gaussian to run it, then boots up Multiwfn and input
dopmaine.fch // In present situation this file in fact is only used to provide geometry
information so that Multiwfn can determine atomic connectivity, therefore you can also use other
formats such as .xyz, .pdb and .wfn instead
18 // RESP module
8 // Let Multiwfn directly load fitting points information from Gaussian output file
1 // Calculate standard RESP charges using two-stage procedure
dopamine_pop_MK.out // The Gaussian output file with IOp(6/33=2,6/42=6) pop=MK
keywords
Then the calculation of atomic charges will be completed very quickly, because calculation of
ESP values are avoided. Since the number and positions of fitting points generated by Multiwfn and
those generated by Gaussian pop=MK task are different, current result is slightly different to that
we obtained earlier.
4.7.7.2 Example 2: Taking multiple conformations into account during RESP

charge calculation of dopamine
In this example we still calculate standard RESP charges for dopamine, but multiple
conformations are explicitly considered in the ESP fitting procedure. It was found that there are four
dominating conformations of dopamine in gas phase, the corresponding Gaussian input files of
optimization task at B3LYP-D3(BJ)/6-311G** level have been provided in
examples\RESP\dopamine_4conf folder, run them by Gaussian and then convert the resulting .chk
files to .fch files.
My earlier Gibbs free energy calculations showed that at room temperature, according to
Boltzmann distribution, the population of the four conformers are 8.48%, 2.66%, 48.45% and
40.42%, respectively. Therefore we should write a plain text file named conf.txt (other filenames
are also acceptable) with below content, assuming that all the .fch files have been put into current
folders.
dopamine1.fch 0.0848
The first column is file path of each conformer, while the second column is corresponding weight.
442
Evidently, the sum of all weights must be exactly equal or approximately equal to unity.
dopamine1.fch // In present case, the file loaded at this stage is only used to provide geometry
information that used to determine atomic connectivity, thus you can also use .fch of other
conformers, the result will not be affected
18 // RESP module
-1 // Load conformation list file
conf.txt // Input actual path of this file
1 // Calculate standard RESP charges using the two-stage procedure
The result is
Center Charge
1(O ) -0.5127119135
2(O ) -0.4946153408
...[ignored]
21(H ) 0.4040820785
22(H ) 0.3885435037
Conformer: 1 RMSE: 0.002885 RRMSE: 0.176042
Weighted RMSE: 0.002249 Weighted RRMSE 0.144547
As can be seen, when considering multiple conformations, Multiwfn gives RMSE and RRMSE
for each conformer as well as weighted RMSE and RRMSE. The data shows that current atomic
charges have better ESP reproducibility for conformations 3 and 4 than conformations 1 and 2. The
reason is not difficult to interpret, because the weights of conformations 3 and 4 in conf.txt are
significantly higher than 1 and 2, therefore the fitted charges prone to faithfully represent charge
distribution of conformers 3 and 4.
It is worth to note that if you set weight of conformer 1 in the conf.txt to 1.0 while set that of
other ones to zero, then the outputted statistical error will be
It can be seen that the atomic charges obtained at this time represent ESP of conformer 1 very
well, because the RMSE and RRMSE are small, while ESP reproducibility of conformers 3 and 4,
which have highest probability of occurrence, is no longer quite good. Therefore, current RESP
charges is not ideal for molecular dynamics modeling of dopamine. This observation reflects the
importance of considering multiple conformations for flexible molecules. Indeed, explicit
consideration of multiple conformations in ESP fitting is somewhat troublesome and time-
consuming, if you decide to obtain ESP fitting charges only by single structure, you should at least
use the structure with the lowest free energy as much as possible.
Directly loading fitting points and ESP values from Gaussian output file of each
443
conformer
When considering multiple conformations, the coordinates of fitting points as well as ESP
values can also be directly loaded from Gaussian output files, here I present an example. For present
molecule, the Gaussian input files of pop=MK task corresponding to the four conformers have been
provided in "examples\RESP\dopamine_4conf\ESP" folder, run them by Gaussian to obtain .out
files, then write a plain text file named e.g. confESP.txt with below content, with assumption that
the four .out files have been placed to C:\ directory.
C:\dopamine1_ESP.out 0.0848
After that, load .fch (or other kinds of files) of any conformer into Multiwfn and enter interface
of RESP module, then select
-1 // Load conformation list file
confESP.txt // Input actual path of this file
8 // Make Multiwfn directly load fitting point information from Gaussian output file
1 // Calculate standard RESP charges using the two-stage procedure
Then standard RESP charges will be immediately shown.
4.7.7.3 Example 3: Imposing equivalence constraint in ESP fitting of

Dimethyl phosphate
Calculation of standard ESP charges has been illustrated in above two examples, next I
exemplify how to calculate normal ESP fitting (i.e. one-stage fitting) with equivalence constraint.
Dimethyl phosphate is taken as instance, its structure is shown below
The two methoxy groups of this system are chemically equivalent, and are easily rotated around
O-P bond during molecular dynamic simulation. Therefore, the charges of O5 and O6 should be the
same, the charges of C7 and C11 should be the same, and a total of six hydrogens on the two methyl
groups (H8, H9, H10, H12, H13, H14) should also be the same. However, when only one structure
is taken into account, it is clear that the such expectation in charge distribution cannot be achieved.
Present example uses this system to demonstrate how to calculate the ESP fitting charges that meet
the above equivalence requirements.
We first create a plain text file called e.g. eqvcons.txt, where each row contains indices of the
atoms whose charges will be constrained to be the same. Therefore, the file content corresponding
to current situation should be (in random order)
444
5,6
7,11
8-10,12-14
Run the Gaussian input file of optimization task at B3LYP-D3(BJ)/6-311G** level for present
molecule (examples\RESP\C2H7O4P\C2H7O4P.gjf), then convert the resulting .chk file to .fch.
Next, Boot up Multiwfn and input
C2H7O4P.fch
18 // RESP module
5 // Modify equivalence constraint (Note that for one-stage ESP fitting, by default hydrogens
in each CH2 and CH3 group are constrained to be equivalent)
1 // Load equivalence constraint setting from external plain text file
eqvcons.txt // The file we just created
2 // Start one-stage ESP fitting calculation with constraints
The result is
Center Charge
1(P ) 1.1205246388
2(O ) -0.6229795460
3(O ) -0.5887730856
4(H ) 0.4109873910
5(O ) -0.4034641689
6(O ) -0.4034641689
7(C ) 0.0352980588
8(H ) 0.0694288037
9(H ) 0.0694288037
10(H ) 0.0694288037
11(C ) 0.0352980588
12(H ) 0.0694288037
13(H ) 0.0694288037
14(H ) 0.0694288037
RMSE: 0.002541 RRMSE: 0.136047
Obviously, the result fully satisfies the equivalence constraint we have made, and the atomic
charge values are also very reasonable and chemically significant. If we do not make the customized
constraint but employ the default equivalency setting, the RRMSE will be 0.113024. Although the
equivalence constraint we have made increases the RRMSE, indicating that the ESP reproducibility
is lowered, since the RRMSE does not increase too much, the constraint we currently employed is
within reasonable range.
Note that in the standard two-stage RESP charge calculation, customized charge constraint and equivalence
constraint can also be applied, however they only take effect for the first stage (by default no constraint is employed
in this stage). For the present molecule, if you load equivalence constraint from the above eqvcons.txt and then select
0 to perform two-stage RESP fitting, you will find the O5 and O6 share identical charge in the result, but charges of
C7 and C11 are different, and charges of hydrogens in different methyl groups are also different, this is because the
customized constraints do not take effect for the second stage (according to standard definition of two-stage RESP
fitting, the carbons and hydrogens in the two methyl groups are refitted at the second stage).
445
4.7.7.4 Example 4: Evaluation of atomic charges of aspartic acid residue

with equivalence and charge constraints
This example is more complicated than the previous three ones, because multiple
conformations, equivalence constraints and charge constraints are all involved. After carefully
reading this section, I believe you will deeply feel that the RESP module of Multiwfn is amazingly
flexible.
In this section we will calculate ESP fitting charges for aspartic acid (ASP) residue. The ASP
is one of the most important amino acids in proteins. In general, in order to make electronic structure
of a given residue in quantum chemistry calculation close to that in actual protein environment, the
nitrogen terminal of the residue should be capped by acetyl group (ACE) while carbon terminal
should be capped by N-methyl amide (NME). For present case, this treatment results in a model
system ACE-ASP-NME.
The two most typical secondary structures of proteins are alpha helix and beta-sheet. From the
point of view of the residues that make up them, the difference comes from the phi and psi dihedrals
of the residue backbone. It has been suggested that residue conformations corresponding to both the
secondary structures should be taken into account in the ESP fitting procedure. Also note that the
net charge of the residue segment in the ACE-ASP-NME system must be an integer. Assume that
the proton of the carboxyl group of the ASP side chain has dissociated, the net charge of the ASP
residue should be constrained to be -1.0. In addition, given that the two oxygens of the carboxylate
are chemically equivalent, it is preferable to apply an equivalence constraint to the two oxygens.
The two hydrogens in the CH2 group of the ASP side chain should also be constrained to be
equivalent.
The Gaussian input files of optimization task for the ACE-ASP-NME models corresponding
to alpha helix and beta-sheet have been provided as alpha.gjf and beta.gjf in "examples\RESP\ACE-
ASP-NME" folder. As can be seen in the files, the keywords correspond to B3LYP-D3/6-311G**
level with IEFPCM solvation model to represent water environment. In the optimization, the phi
and psi dihedrals are fixed to their initial values (the dihedrals will vary remarkably during
optmization if they are not frozen). In alpha.gjf, the phi and psi are -90 and -60, respectively,
corresponding to typical case of alpha helix. While in beta.gjf, the two dihedrals are set to -100 and
130, reflecting typical situation of beta-sheet.
Run the two .gjf files by Gaussian, and convert resulting .chk files to .fch format. The two
optimized structures are shown below. The region surrounded by green dashed ellipse is the ASP
residue, the charges of these atoms are what we are interested in. The phi and psi dihedrals
mentioned above correspond to 6-3-1-13 and 1-3-6-19, respectively.
446
We create a plain text file named for example chgcons.txt, in this file each line defines a charge
constraint term. Since we require that the ASP residue has total charge of -1, we should write below
content in this file
1-12 -1
Note that in the RESP module, there is no upper limit on the number of charge constraint terms.
Also note that the indices of the atoms involved in charge constraint are not necessarily contiguous,
for example if you write 1,3-5,8,9-12 1.5, then sum of charges of atoms 1,3,4,5,8,9,10,11,12 will be
constraint to 1.5.
Then we create a plain text file named for example eqvcons.txt, in this file each line defines a
equivalence constraint term. As mentioned earlier, O11 and O12 should be equivalent, H7 and H8
should be equivalent, therefore for present case the content should be
11,12
7,8
Although the hydrogens in the methyl groups at the two ends of the model system are chemically
equivalent, since they are not of our interest, the equivalence constraint setting is ignored.
Next, we write a file named for example conflist.txt, which contains list of .fch files of all
conformers. In present circumstance we hope that the resulting atomic charges can equally well
represent the actual charge distribution of ASP residue in both alpha helix and beta-sheet secondary
structures, therefore weight of both the conformers should be 0.5. Assuming that .fch files have been
placed in D:\ folder, the file content should be
D:\alpha.fch 0.5
D:\beta.fch 0.5
Finally, boot up Multiwfn, load either alpha.fch or beta.fch, then enter RESP module and input
below commands
5 // Modify the equivalence constraint
eqvcons.txt // The equivalence constraint file we created
6 // Set charge constraint
1 // Load charge constraint setting from external plain text file
chgcons.txt // The charge constraint file we created
447
-1 // Load list of conformer and weights from external file

conflist.txt // The conformation list file we created
2 // Start one-stage ESP fitting calculation with constraint
The output is
Center Charge
1(N ) -0.5680297892
2(H ) 0.2986898310
3(C ) 0.2320659798
4(H ) 0.0039518865
5(C ) -0.1872465380
6(C ) 0.5806052594
7(H ) 0.0309424424
8(H ) 0.0309424424
9(C ) 0.7732537802
10(O ) -0.6023381592
11(O ) -0.7964185676
12(O ) -0.7964185676
[ignored...]
Sum of charges: -1.0000000000
Weighted RMSE: 0.002131 Weighted RRMSE 0.017446
The above calculation result is very reasonable, and it can be seen that both the charge
constraint and equivalence constraint work perfectly. Moreover, since the weights of the two
conformations are set to be the same, the RMSE or RRMSE corresponding to the two conformers
have comparable magnitude. Given that the RRMSE is very small, the current fitted charges should
be able to describe the state of ASP residue in various proteins well.
4.7.7.5 Example 5: Example of setting equivalence constraint according to

local or global point group symmetry
1: A small molecule
To obtain ESP fitting charge for below molecule, we should constrain the three fluorine atoms
to have the same charge because they are chemically equivalent. In addition, due to symmetry of
the local geometry of the benzene moiety, the atoms at its two sides should be equivalent, namely
we should constrain H5=H7, H10=H6, C2=C4, C3=C9.
448
Although you can manually create a file containing above equivalent constraints, it is more
convenient to make Multiwfn automatically create the file according to point group symmetry of
local regions of the CF3 group and benzene moiety, as shown below.
CF3benCOCH3.fch
7 // Population analysis and atomic charge calculation
18 // RESP module
5 // Set equivalence constraint
11 // Generate a file containing equivalence constraints according to point group symmetry
of selected regions
Then we need to input atomic indices in each fragment that has local symmetry. In order to
make finding the indices convenient, I suggest using GaussView to open the above .fch file, then
select the fragment as yellow, then enter "Tools" - "Atom Selection" and copy the atomic indices
from the text box to Multiwfn window, as illustrated below
The region colored by yellow has point group of C2v, if we provide the corresponding indices
1-7,9-10 to Multiwfn, then Multiwfn will find symmetrically equivalent atoms and write to
eqvcons_PG.txt in current group (Notice that we should not select the whole benzene moiety,
namely 1-7,9-11, because this fragment has point group of D2h, in this case Multiwfn will also regard
C1 and C11 as symmetrically equivalent atoms).
Now we input 1-7,9-10 in Multiwfn window, then you will see
Detected point group: C2v
Number of symmetry-equivalence classes: 4
Class 1 (C ): 2 atoms
2, 4
3, 9
Class 3 (H ): 2 atoms
5, 7
6, 10
Accept and append to eqvcons_PG.txt in current folder? (y/n)
Clearly, symmetrically equivalent atoms have been correctly identified, therefore we input y to write
the corresponding constraint setting to eqvcons_PG.txt in current folder.
Next, we use this feature to add the three fluorine atoms into the equivalence constraint file. In
the Multiwfn window we input atomic indices of the CF3 group, namely 13-16, then you will see
449
Detected point group: C3v
14, 15, 16
The printed information is obviously correct, therefore we input y. Then input q to exit. Now you
will find the current content of the eqvcons_PG.txt is
2, 4
3, 9
5, 7
6, 10
14, 15, 16
The content is fully in line with our expectation. In fact we can also similarly set the three hydrogens
in the methyl group as equivalent atoms by this interface, however we do not do this because in this
example we will employ two-stage RESP fitting, at the second stage the equivalence constraint is
automatically applied to the three hydrogens.
Subsequently, in the Multiwfn window we input
1 // Load equivalence constraint from external file
eqvcons_PG.txt // The file just generated
1 // Start standard two-stage RESP fitting
The result is
Center Charge
1(C ) 0.0091127275
2(C ) -0.1055999399
3(C ) -0.1400843859
4(C ) -0.1055999399
5(H ) 0.1287865888
6(H ) 0.1361477888
7(H ) 0.1287865888
8(C ) 0.6140083901
9(C ) -0.1400843859
10(H ) 0.1361477888
11(C ) -0.0475485792
12(O ) -0.4669646019
13(C ) 0.4344682195
14(F ) -0.1667205104
15(F ) -0.1667205104
16(F ) -0.1667205104
17(C ) -0.4512569713
18(H ) 0.1232807476
19(H ) 0.1232807476
20(H ) 0.1232807476
RMSE: 0.001518 RRMSE: 0.115139
As can be seen, the charges are very reasonable and fully meet our expectation.
450
2: Coronene
Let us see a molecule containing relatively large number of atoms and having high-order point
group, namely coronene, which has D6h point group.
Because the distribution of ESP fitting points does not satisfy point group symmetry, the
resulting charges thus do not fulfill D6h symmetry. For example, you will find C17 and C18 have
charge of -0.2243 and -0.2169, respectively, however their charges should be identical. Despite the
difference is negligible, it is best to eliminate it. The most ideal way to make the resulting charges
fully satisfy the point group is imposing equivalence constraint according to the symmetry, however
manually writing the constraint file is quite laborious for such a large system, therefore we again
use Multiwfn recognize point group and automatically generate the constraint file.
coronene.fch
7 // Population analysis and atomic charge calculation
18 // RESP module
11 // Generate file containing equivalence constraints according to point group symmetry of
selected region
a // Select the entire system
You will see the equivalent atoms have been correctly identified:
Detected point group: D6h
1, 2, 3, 4, 5, 6
7, 8, 9, 10, 11, 12
13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24
25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36
Then we input below commands
451
y // Write the four classes equivalent constraints to eqvcons_PG.txt in current folder

q // Exit
1 // Load equivalence constraint file
eqvcons_PG.txt
1 // Perform standard two-stage RESP fitting (note that the result is identical to one-stage
fitting, because no atoms will be refitted in the second stage for this molecule)
From the printed result, you can find the atoms in each of the four detected classes are indeed
equivalent. The value of aforementioned charges of C17 and C18 are -0.220866 currently, which is
quite reasonable.
By the way, via suboption 10 in option 5 you can export equivalence constraint of hydrogens
in all CH2 and CH3 to eqvcons_H.txt in current folder. If you combine this file and eqvcons_PG.txt
into a single file, then load it into Multiwfn and perform normal ESP fitting, the two kinds of
equivalence constraint will simultaneously take effect (however, the content of the two sets of
constraint should not be contradict to each other)
It is noteworthy that, sometimes after inputting atomic indices in the aforementioned subfunction 11 of option
5, Multiwfn does not print point group, that means the determination of point group is failed. However, this failure
does not necessary mean that the symmetrically equivalent atoms are not properly identified, therefore if the printed
atomic indices are reasonable, you can still input y to write the constraint to eqvcons_PG.txt. In contrast, if you find
the indices of the symmetrically equivalent atoms are not correctly identified, you should input n to cancel the
writting, and then modify the tolerance for determining point group (for example, inputting t 0.05 means changing
the tolerance to 0.05), after that you can input the atomic indices again and check if the equivalent classes have been
reasonably recognized. The default tolerance is 0.1, when you encounter problem, you can either try to increase it or
decrease it. Also note that the equivalent atoms are always correctly detected if the point group is correctly printed.
4.7.7.6 Example 6: RESP charge calculation with additional fitting centers

Multiwfn is able to calculate RESP charges for additional fitting centers, in other words, some
point charges to be fitted are not necessarily at nuclear positions. Such non-atomic point charge is
valuable in some cases, such as better reproducing ESP around lone pair and -hole regions, this
idea has already been employed in a few forcefields. In this example, I will illustrate how to fit
charge for non-atomic points by taking C18 system as example.
The C18 was very systematically and detailedly studied in my work Carbon, 165, 468 (2020),
Carbon, 165, 461 (2020) and ChemRxiv (2019) DOI: 10.26434/chemrxiv.11320130. The ESP
colored vdW surface map is shown below
452
The minimum structure of C18 has D9h point group, therefore, if we calculate RESP charges as
usual, all resulting atomic charges will be exactly zero. Clearly such atomic centered charges are
completely useless in reproducing ESP around vdW surface for this very special system. In order to
better represent its ESP, it is best to fit some point charges located at midpoint of each C-C bond.
The .fchk file of C18 system corresponding to B97XD/def2-TZVP wavefunction at minimum
point structure can be download here: http://sobereva.com/multiwfn/extrafiles/C18.zip. If you place
additional fitting centers at midpoint of each C-C bond, the situation will correspond to below map,
in which each purple sphere corresponds to an additional fitting center. The Gaussian .gjf file
corresponding to below map has been provided as examples\RESP\C18\C18.gjf.
In this instance, we will simultaneously fit atomic charges and the point charges at midpoint of
the C-C bonds.
Before starting RESP fitting calculation, we need to write a text file containing X, Y, Z
coordinate of all additional fitting centers, and the first line should be the total number of additional
fitting centers, see examples\RESP\C18\fitcen.txt.
In addition, we need to write an equivalence constraint file to ensure that the resulting point
charges in each set are exactly identical: (1) Atomic charges of all carbons (2) All points at midpoint
of short C-C bonds (3) All points at midpoint of long C-C bonds. The constraint file has been
provided as examples\RESP\C18\eqvcons.txt, whose content defined three batches of constraints:
1-18
19,21,23,25,27,29,31,33,35
20,22,24,26,28,30,32,34,36
Note that the index of additional fitting centers is after that of actual atoms, therefore the points
19~36 in the eqvcons.txt correspond to the 18 points defined in the fitcen.txt.
Also note that there is no reason to apply the penalty function defined in RESP method, which
hurts the reproducibility of ESP in the present case, therefore we will disable this treatment, which
is enabled by default.
Now we boot up Multiwfn and input below command

C18.fchk
7 // Atomic charge calculation and population analysis
18 // RESP
4 // Set hyperbolic penalty parameters
453
2 // Set restraint strength (a) for one-stage fitting

0 // Remove effect of penalty function
0 // Return
9 // Load additional fitting centers
examples\RESP\C18\fitcen.txt // The file containing additional fitting centers
5 // Set equivalence constraint in fitting
examples\RESP\C18\eqvcons.txt
2 // Start one-stage ESP fitting calculation with constraints
The result is shown below
Center Charge
1(C ) 0.0642410906
2(C ) 0.0642410906
3(C ) 0.0642410906
...[ignored]
19(X ) -0.5038220210
20(X ) 0.3753398397
21(X ) -0.5038220210
22(X ) 0.3753398397
...[ignored]
RMSE: 0.001058 RRMSE: 0.553294
As you can see, the atomic charge of carbon is 0.064, while the fitted charges at midpoint of
short and long C-C bonds are -0.504 and 0.375, respectively. This observation is in line with the
ESP mapped vdW surface map that given earlier, namely electron is much more heavily
concentrated around the short C-C bond than the long C-C bond.
If we do not specify additional fitting centers in the RESP calculation, you will find the atomic
charges of all carbons are exactly zero, and the ESP reproduction error will be
RMSE: 0.001911 RRMSE: 1.000000
The error is nearly twice as larger as the case having additional fitting centers at midpoint of bonds,
indicating that employing point charges not located at the center of atoms is crucial in faithfully
representing ESP on molecular surface.
There are a few notes about additional fitting centers:

• Additional fitting centers have zero radii, namely they do not affect distribution and number
of fitting points.
• Penalty function in the RESP method also takes effect for additional fitting centers.
• When multiple conformers are taken into account in the fitting, additional fitting centers
should be defined for different conformers, the format of the file defining these points in this case
is described in "Option 9" in Section 3.9.16.2. The distribution of additional fitting centers can be
different for different conformers, but the number must be the same.
• Both equivalence constraint (as illustrated in the present example) and charge constraint work
normally for additional fitting centers.
454
4.7.7.7 Skill 1: Using two times of one-stage fitting to equivalently realize

standard RESP two-stage fitting
In example 1 of this Section, I have illustrated how to derive RESP charges using standard
RESP two-stage fitting procedure. Thanks to the flexiblity of RESP module of Multiwfn, this
“composite procedure” can also be manually realized via two separated one-stage fittings, as
illustrated in this section. After reading this section, I believe you will better understand how to
customize the RESP calculation procedure. Below we will use a very simple molecule methanol as
example, whose .fch file can be found in http://sobereva.com/multiwfn/extrafiles/RESP.zip.
methanol.fch
18 // RESP charge calculation
0 // Remove default equivalence constraint
2 // Using one-stage fitting to derive charges
The result is
Center Charge
1(C ) 0.238915
2(H ) 0.045904
3(H ) -0.018089
4(H ) -0.018089
5(O ) -0.664522
6(H ) 0.415880
They are identical to the charges obtained at the first stage of standard RESP two-stage fitting.
According to definition of standard RESP charge calculation procedure, the charge of the atoms
in hydroxyl group of methanol should keep fixed during the second fitting stage, therefore we create
a file chgcons.txt with below content.
5 -0.664522
6 0.415880
Then input below commands in Multiwfn interface

n // Do not export .chg file
4 // Set hyperbolic penalty parameters
2 // Set restraint strength (a)
0.001 // This value is the one used in the second stage of standard RESP fitting procedure
0 // Return to the upper menu
2 // Constraint hydrogens in CH2 and CH3 groups to be equivalent, as required by the second
stage of standard RESP fitting
6 // Set charge constraint
1 // Load charge constraint setting file
chgcons.txt
2 // Calculate charges by one-stage fitting
The final result is
1(C ) 0.235334
455
2(H ) 0.004436
3(H ) 0.004436
4(H ) 0.004436
5(O ) -0.664522
6(H ) 0.415880
which are completely identical to the charges derived by standard two-stage RESP fitting.
Hint: When the molecule is large, manually editting the chgcons.txt is often cumbersome. In
fact, you can create an empty file named chgcons_stage2.txt in current folder and carry out standard
two-stage RESP fitting, then before performing the second stage fitting, Multiwfn will automatically
export the indices and charges of the atoms whose charges will be kept fixed in the second stage to
this file, so that you will not need to manually write the chgcons.txt file.
4.7.7.8 Skill 2: Quickly obtaining RESP charges from molecular structure file
by only one command
Note: Chinese version of this section is http://sobereva.com/476.
In this section, I will show it is fully possible to use only one command to generate RESP
charges directly from molecular structure file using Linux shell script, the user does not need any
knowledge about quantum chemistry code.
Assume that both Gaussian and Multiwfn have been properly installed on your machine, and
you want to calculate RESP charges for H2O.xyz, which contains geometry of water, what you need
to do is simply:
• Copy RESP.sh from examples\RESP folder to current folder.
• Run chmod +x ./RESP.sh to add executable permission
• Move the H2O.xyz to current folder
• Run ./RESP.sh H2O.xyz 0 1, where 0 and 1 correspond to net charge and spin multiplicity,
respectively.
This script first automatically invokes Gaussian to optimize the geometry at B3LYP-
D3(BJ)/def2-SVP level, then performs single point task at B3LYP-D3(BJ)/def2-TZVP level and
meantime produce ESP data on vdW surface, then converts .chk file to .fchk file via formchk, and
finally, invokes Multiwfn to yield RESP charges in standard manner. After all steps have been
successfully completed, you will find H2O.chg in current folder, whose final column is RESP
charges.
Any input file supported by Multiwfn containing molecular geometry information can be used
as input file for this script, such as .xyz, .mol, .mol2, .pdb, .gjf, .fch and so on. If you do not explicitly
specify net charge and spin multiplicity when booting up the script, the system will default to singlet
neutral system.
Note that you sometimes need to properly modify the RESP.sh before running it. This script by
default invokes Gaussian 09, therefore if you are using other version, you need to replace the "g09"
in this script with "g16". In the optimization and single point calculation, implicit solvation model
is used for representing water environment, thus you should replace the "water" to other solvent
name if you intend to simulate the molecule in other environment. In addition, if you find the def2-
TZVP is too expensive or geometry optimization was found to be difficult to converge in rare cases,
you need to manually change the keywords in this script.
456
4.7.7.9: Special topic: Calculation of RESP2 charges

Note: Much more in-depth discussions about RESP2 charge can be found in my blog article
"Idea of RESP2 atomic charge and its calculation in Multiwfn" (in Chinese,
Definition of RESP2 charge

In solvent environment, charge distribution of solute is evidently polarized by surrounding
solvents. Therefore, if atomic charges of a molecule are to used in molecular dynamics (MD)
simulation with fixed charge forcefield (i.e. non-polarizable forcefield), the polarization effect must
be effectively taken into account into the atomic charges.
In Commun. Chem., 3, 44 (2020), the authors defined RESP2 charge as
q RESP2 = (1 −  )qgas
RESP
+  qwater
RESP
RESP
where  is adjustable parameter, 𝑞gas
RESP
and 𝑞water are RESP charges calculated in gas phase and
in water environment (represented by PCM implicit solvation model), respectively. The authors
employed PW6B95 exchange-correlation functional in combination with aug-cc-pVDZ basis set in
the process of quantum chemistry calculations. It is found that =0.6 leads to lowest overall error in
simulation of various condensed phase properties (=0.5 works equally well). Note that RESP2 with
=0.5, namely RESP20.5, is equivalent to the IPolQ-mod atomic charge defined in J. Comput. Aided
Mol. Des., 28, 277 (2014). These studies showed that =0.5 should be a relatively general and ideal
RESP
choice for evaluating RESP2 charge. Note that directly using 𝑞water for the MD simulation in
aqueous environment leads to worse result for many properties compared to RESP20.6, this is mainly
RESP
because the 𝑞water exaggerates the extent of polarization or does not properly account for cost of
electronic polarization.
In my opinion, the best way of calculating atomic charges used for condensed phase MD
simulation should be the RESP20.5 defined as follows
q RESP2 = 0.5  qgas

RESP
+ 0.5  qsolv
RESP
RESP
where 𝑞solv is RESP charge calculated under actual solvent environment represented by PCM (or
IEFPCM, CPCM, SMD) model. I suggest using B3LYP-D3(BJ) with def2-SVP (or the better one
def-TZVP) for geometry optimization and B3LYP-D3(BJ)/def2-TZVP for the subsequent single
point task calculations in both gas and solvent phases. In principle it is best to perform optimization
under actual solvent environment, however solvent effect on geometry can be safely ignored if the
system is neutral and there is no highly ionic local region.
Example of calculating RESP2 charges

As an example, we calculate RESP20.5 charge via the above recommended way for H2CO in
ethanol environment.
Copy below content into a Gaussian input file and then run it by Gaussian. This task consists
of three steps, namely geometry optimization, single point calculation in gas phase and then in
ethanol phase.
%chk=C:\opt.chk
# B3LYP/TZVP em=GD3BJ opt
457
niconiconi
0 1
C 0.00000000 0.00000000 0.52887991
H 0.00000000 0.93775230 1.12379107
O 0.00000000 0.00000000 -0.67757652
H 0.00000000 -0.93775230 1.12379107
--link1--
%oldchk=C:\opt.chk
%chk=C:\SP_gas.chk
# B3LYP/def2TZVP em=GD3BJ geom=allcheck
--link1--
%oldchk=C:\opt.chk
%chk=C:\SP_solv.chk
# B3LYP/def2TZVP em=GD3BJ scrf=solvent=ethanol geom=allcheck
 Blank line
 Blank line
After calculation, you will obtain SP_gas.chk and SP_solv.chk in C:\ folder. Convert them
to .fch files and then use Multiwfn to calculate RESP charge as usual, you will find the charges in
gas phase is
Center Charge
1(C ) 0.4195430529
2(H ) -0.0050085243
3(O ) -0.4095260044
4(H ) -0.0050085243
while in ethanol phase the result is

Center Charge
1(C ) 0.4625344852
2(H ) 0.0093031373
3(O ) -0.4811407598
4(H ) 0.0093031373
By simply taking average of the above two sets of charges via e.g. Excel, the RESP20.5 charge
will be obtained:
0.441038769
0.002147307
-0.445333382
0.002147307
Using shell script to conveniently calculate RESP20.5 charges

In order to make calculation of RESP2 charge even easier, a Linux script named calcRESP2.sh
is provided in "examples\RESP" folder, it can calculate both RESP and RESP2 charges. Examples
of usage:
458
• Calculation of RESP charges: ./calcRESP.sh gas.fchk

• Calculation of RESP20.5 charges: ./calcRESP.sh gas.fchk solv.fchk
• Calculation of RESP20.7 charges: ./calcRESP.sh gas.fchk solv.fchk 0.7
Since the script invokes Multiwfn, before running it you should make sure that Multiwfn has
been properly installed in your Linux system, see Section 2.1.2 on how to install.
Once running of the script has successfully finished, you will find RESP2.chg in current folder,
the last column corresponds to the resulting RESP2 charges.
Quickly obtaining RESP2 charges from molecular structure file by only one command
In order to make RESP2 charge calculation as easy as possible, I also provides a script named
RESP2.sh in "examples\RESP" folder. Only a file containing (unoptimized) geometry is needed as
input file. This script is very similar to the RESP.sh introduced in Section 4.7.7.8.
Examples of usage:
• Calculating RESP20.5 charges for a singlet neutral molecule for MD simluation in water phase:
./RESP2.sh H2O.pdb
• Calculating RESP20.5 charges for a triplet neutral molecule for MD simluation in water phase:
./RESP2.sh yoshiko.xyz 0 3
• Calculating RESP20.5 charges for a singlet anion for MD simluation in ethanol phase:
./RESP2.sh yohane.mol -1 1 ethanol
As you can find from the examples, the charge and spin multiplicity are default to 0 and 1,
respectively, while the solvent is default to water.
If running the script has successfully finished, you will find a .chg file with identical name as
input file in current folder, the final column corresponds to RESP20.5 charges. In current folder you
can also find gas.chg and solv.chg, they correspond to the RESP charges in gas phase and in solvent
phase.
Specifically, this script do following things in turn, both Gaussian and Multiwfn are invoked in these processes:
(1) Geometry optimization at B3LYP-D3(BJ)/def2-SVP level in solvent environment
(2) Single point task at B3LYP-D3(BJ)/def2-TZVP level in gas phase
(3) Calculating RESP charge corresponding to gas phase
(4) Single point task at B3LYP-D3(BJ)/def2-TZVP level in solvent phase
(5) Calculating RESP charge corresponding to solvent phase
(6) Generating RESP20.5 charge by averaging the result produced by (3) and (5)
4.7.8 Examine electrostatic potential reproducibility of atomic charges
Electrostatic potential (ESP) reproducibility is a crucial property of atomic charges, only

atomic charges having good ESP reproducibility could be employed to reveal intramolecular and
intermolecular electrostatic interactions. It is possible to examine ESP reproducibility of given
atomic charges using the MK and CHELPG charge calculation modules, which have been
introduced in Sections 3.9.10 and 3.9.11, respectively. Here we compare the ability of Hirshfeld and
ADCH charges for reproducing ESP values at Merz-Kollmann ESP fitting points (which are
distributed around molecular van der Waals surface) for CH3CONH2. We first calculate Hirshfeld
charges as usual using examples\CH3CONH2.fch (see Section 4.7.1), then select "y" to export the
atomic charges to CH3CONH2.chg. Then we enter the MK charge calculation module (subfunction
13 of main function 7) and input
-3 // Using atomic charges from a .chg file instead of fitting new charges
CH3CONH2.chg // Atomic charges (i.e. Hirshfeld charges) will be directly loaded from this
459
file
1 // Start calculation. In current case MK charges will not be yielded
The data shown on the screen is
Center Charge
1(C ) -0.090370
2(H ) 0.037254
3(H ) 0.043058
4(H ) 0.048339
5(C ) 0.170596
6(O ) -0.308866
7(N ) -0.159120
8(H ) 0.131920
9(H ) 0.127108
RMSE: 0.006214 RRMSE: 0.310394
These charges are just the Hirshfeld charges loaded from CH3CONH2.chg, the RMSE and RRMSE
measure the ESP reproducibility of the Hirshfeld charges. If you compute MK charges as usual, you
will find the RRMSE will be about 0.05, since as shown above the RRMSE of Hirshfeld charges is
as high as 0.31, it is evident that the ESP reproducibility of Hirshfeld charges is much worse than
MK charges. If you redo the analysis based on the .chg file containing ADCH charges, you will find
the RRMSE is 0.21. Clearly, ADCH charges have evidently lower error in reproducing ESP
compared to Hirshfeld charges.
Studying ESP reproducibility around different atoms or fragment

It is also possible to measure ESP reproducibility on the fitting points corresponding to specific
atom or fragment. By default, the MK points are generated around all atoms in turn and then the
points lying inside the intermost layer are pruned. If only specific atoms are taken into account, the
constructed MK fitting points will only correspond to those atoms. Let us compare ESP
reproducibility of Hirshfeld and ADCH charges around the amino group, only two MK layers with
scale factor of 1.4 and 1.6 will be considered (no special reason, just give an example). Enter the
MK module and input
-3 // Using atomic charges from a .chg file
CH3CONH2.chg // Assume that this file contains Hirshfeld charges
3 // Set number and scale factors of layers of MK fitting points
1.4 // Set scale factor of layer 1
1.6 // Set scale factor of layer 2
q // Setting has finished, now quit
4 // Choose the atoms considered in the construction of fitting points
7-9 // Atomic indices of amino group
You will find below information from the output
RMSE: 0.008745 RRMSE: 0.374584
If we repeat the calculation based on .chg file containing ADCH charge, the output will be
RMSE: 0.003817 RRMSE: 0.163478
Since the RRMSE of ADCH charge (0.163) is by far less than that of Hirshfeld charge (0.374),
460
the ADCH charges have much better ESP reproducibility around the amino group.
Note: The CHELPG module also supports employing fitting points only for specific fragment.
Visualize fitting points and ESP values

If you want to visualize the fitting points corresponding to the amino group, you can select "6
Toggle if exporting fitting points with ESP after the task" once to change the status to "Yes" and
then use option 1 to start calculation. Once calculation is finished, choose 2 to export the fitting
points to ESPfitpt.pqr in current folder. This file can be directly loaded into the famous visualization
tool VMD. If you set the drawing method to "VDW" and change the "Sphere Scale" to 0.8, set
"Coloring Method" to "Charge", then set the color transition mode to "BWR" (Graphics - "Colors"
- "Color Scale"), you will see below graph (molecular structure file is also loaded).
Clearly, the fitting points well correspond to the amino group. The more red (blue), the more
positive (negative) the ESP on the points.
Visualizating reproducibility error of ESP at fitting points

Finally, I would like to mention that the reproduction error of ESP can also be visualized by
combinely using Multiwfn and VMD. Here we check this for MK charges. Load the
examples\CH3CONH2.fch into Multiwfn, enter MK module, choose option 6 once, and then choose
option 1 to start calculation. Once calculation is completed, choose 3 to export ESP fitting points
with ESP reproduction error to ESPerr.pqr in current folder. In this file, the "Charge" column
corresponds to absolute value of difference (in kcal/mol) between the exact ESP and the ESP
evaluated based on current atomic charges (namely MK charges). If you render this file by VMD,
you will see below graph. The color scale has been set to -1.5 to 1.5 (can be set in "Graphics" -
"Representation" - "Trajectory" page), the default color transition "Red-White-Blue" is used,
perspective has been set to orthographic ("Display" - "Orthographic").
461
In this graph, more blue region corresponds to higher ESP reproduction error, while ESP at
white points can be well reproduced by the MK charges (i.e. the absolute error is close to zero). You
can also use this method to visualize ESP reproducibility of other atomic charges (need to use .chg
file, as illustrated eariler).
4.7.9 Calculate PEOE (Gasteiger) charge
In this example we calculate PEOE charge (also known as Gasteiger charge) for a typical
organic system, dopamine. Reading Section 3.9.17 is recommended to gain basic knowledge about
the principle, details and implementation of the PEOE method in Multiwfn.
Since calculation of PEOE charge only requires geometry information, we can use such
as .xyz, .pdb, .mol as input file. Boot up Multiwfn and input
examples\dopamine.xyz
19 // PEOE (Gasteiger) charge
First, the parameters involved in the PEOE calculation are printed:
Determined parameters:
1(O ) numbond= 2 a= 14.180 b= 12.920 c= 1.390 init. q= 0.0000
2(O ) numbond= 2 a= 14.180 b= 12.920 c= 1.390 init. q= 0.0000
3(N ) numbond= 3 a= 11.540 b= 10.820 c= 1.360 init. q= 0.0000
4(C ) numbond= 4 a= 7.980 b= 9.180 c= 1.880 init. q= 0.0000
[ignored...]
Then iteration starts

Max cycles: 50 Charge convergence criterion: 0.00010 Damping factor: 0.500
Cycle 1 Maximum change of charges: 0.312435
[ignored...]
Convergence succeeded after 10 cycles
462
Since the formulae involved in the PEOE method are extremely simple, the iteration is finished
within one second (this is still true even the system consists of several hundreds of atoms!). Then
you can find the PEOE charges:
Atom Charge
1(O ) -0.358163
2(O ) -0.358170
3(N ) -0.330120
4(C ) -0.015405
[ignored...]
4.8 Molecular orbital composition analysis
In this section, I will show how to use various methods to analyze molecular orbital
compositions. The illustrated methods are also applicable to any other type of orbitals, e.g. natural
orbitals, natural transition orbitals (NTOs) and localized MOs (LMOs). Details about orbital
composition analysis can be found in Section 3.10. The pros and cons of different methods are very
detailedly discussed in my paper Acta Chim. Sinica, 69, 2393 (2011, http://sioc-
journal.cn/Jwk_hxxb/CN/abstract/abstract340458.shtml), citation is welcomed. If you can read
Chinese, also you can consult my blog article "On the calculation methods of orbital composition"
(http://sobereva.com/131).
Simply speaking, if your aim is merely obtaining atom compositions in orbitals,
Hirshfeld/Becke method may be the most robust and convenient way, see Section 4.8.3; if you also
would like to obtain atomic orbital composition, then the NAO method exemplified in Section 4.8.2
may be the best choice. The Mulliken method illustrated in Section 4.8.1 also generally works well
but diffuse functions should not be used.
4.8.1 Analyze acetamide by Mulliken method
In this example we employ Mulliken method to first analyze the composition of the 6th
molecular orbital of acetamide, and then analyze which orbitals have main contribution to the
bonding between formamide part and methyl group. Beware that Mulliken method is incompatible
with diffuse functions, if they are involved, you should either choose other orbital composition
methods (e.g. NAO, Hirshfeld...) or remove them from your basis set.
examples\CH3CONH2.fch // You have to use .mwfn/.fch/.molden/.gms file as input for this
type of analysis
8 // Orbital composition analysis
1 // Use Mulliken partition
6 // The orbital index is 6
The composition of basis functions, shells and atoms are printed immediately, see below.
Threshold of absolute value: > 0.50000 % // Only the basis functions with composition
larger than 0.5% will be printed, you can change the threshold by “compthres” parameter
463
in settings.ini.
Orbital: 6 Energy(a.u.): -0.905290 Occ: 2.000000 Type: Alpha&Beta
Basis Type Atom Shell Local Cross term Total
23 S 5(C ) 14 0.44902 % 0.67507 % 1.12409 %
24 X 5(C ) 15 0.31240 % 0.50522 % 0.81762 %
25 Y 5(C ) 15 4.25271 % 5.88221 % 10.13493 %
29 Y 5(C ) 17 0.00777 % -0.61063 % -0.60286 %
38 S 6(O ) 20 3.50037 % 2.65507 % 6.15544 %
42 S 6(O ) 22 3.29488 % 1.71316 % 5.00803 %
53 S 7(N ) 26 15.20411 % 15.32688 % 30.53098 %
57 S 7(N ) 28 16.89006 % 17.40040 % 34.29046 %
61 XX 7(N ) 30 0.00774 % 0.58279 % 0.59053 %
63 ZZ 7(N ) 30 0.02793 % -0.98090 % -0.95297 %
67 S 8(H ) 31 1.27855 % 3.03091 % 4.30946 %
69 S 9(H ) 33 1.52931 % 3.60949 % 5.13880 %
Sum up those listed above: 46.75484 % 49.78967 % 96.54451 %
Sum up all basis functions: 51.95605 % 48.04395 % 100.00000 %
Composition of each shell, threshold of absolute value: > 0.500000 %
Shell 14 Type: S in atom 5(C ) : 1.12409 %
Shell 15 Type: P in atom 5(C ) : 10.95268 %
Shell 17 Type: P in atom 5(C ) : -0.97156 %
Shell 20 Type: S in atom 6(O ) : 6.15544 %
Shell 22 Type: S in atom 6(O ) : 5.00803 %
Shell 26 Type: S in atom 7(N ) : 30.53098 %
Shell 28 Type: S in atom 7(N ) : 34.29046 %
Shell 31 Type: S in atom 8(H ) : 4.30946 %
Shell 33 Type: S in atom 9(H ) : 5.13880 %
Composition of different types of shells (%):
s: 88.193 p: 11.391 d: 0.416 f: 0.000 g: 0.000 h: 0.000
Composition of each atom:
Atom 1(C ) : 1.17249 %
Atom 2(H ) : 0.05445 %
Atom 3(H ) : 0.03212 %
Atom 4(H ) : 0.00817 %
Atom 5(C ) : 11.81245 %
Atom 6(O ) : 11.63274 %
Atom 7(N ) : 65.50085 %
Atom 8(H ) : 4.47022 %
Atom 9(H ) : 5.31651 %
Orbital delocalization index: 46.15
464
The result indicates that nitrogen has primary contribution (65.5%) to orbital 6, and the
contribution consists of two S-shells (30.5% and 34.3%). P-shells of neighbour carbon and S-shells
of oxygen have slight contribution too (both are about 12%). We can check if the result is reasonable
by viewing isosurface (isovalue is set to 0.1 here):
From the graph, the region where the value of orbital wavefunction is large is mainly localized
around nitrogen, and there is no nodal plane, so the orbital wavefunction in this region should be
constructed from s-type orbitals. The isosurface also somewhat intrudes into the region of atom C5
and O6, so they should have small contribution to MO 6, moreover, because there is a nodal plane
in C5, the atomic orbitals of C5 used to form MO 6 should be p-type. Obviously, these conclusions
are in fairly agreement with composition analysis. The advantage of composition analysis is that the
result can be quantified, while by visual study we can only draw qualitative conclusion, for some
complex system we cannot draw even qualitative conclusion.
The " Orbital delocalization index" printed at the end of the output has close relationship with
extent of spatial delocalization of the orbital, this point will be described in Section 4.8.5 in detail.
Now let us find which molecular orbitals have main contribution to the bonding between
formamide part and methyl group. Boot up Multiwfn and input
-1 // Define fragment 1
a 1-4 // Add all basis functions in atom 1, 2, 3, 4 (methyl group) into fragment1
q // Save fragment and return to upper menu
-2 // Define fragment 2
a 5-9 // Add all basis functions in atom 5, 6, 7, 8, 9 (formamide moiety) into fragment 2
q
4 // Print composition of fragment 1 and the cross term between fragment 1 and 2 in all
orbitals by Mulliken analysis. If you only defined fragment 1, then only composition of fragment 1
will be printed
Since amount of the printed information is huge, I only extract cross term composition in all
occupied orbitals:
Cross term between fragment 1 and 2 and their individual parts:
Orb# Type Ene(a.u.) Occ Frag.1 part Frag.2 part Total
1 AB -19.1036 2.00000 -0.0013 % -0.0013 % -0.0026 %
2 AB -14.3492 2.00000 -0.0001 % -0.0001 % -0.0003 %
3 AB -10.2819 2.00000 0.0606 % 0.0606 % 0.1211 %
465
4 AB -10.1845 2.00000 0.0604 % 0.0604 % 0.1209 %
5 AB -1.0376 2.00000 0.3535 % 0.3535 % 0.7070 %
6 AB -0.9053 2.00000 0.7295 % 0.7295 % 1.4590 %
7 AB -0.7387 2.00000 7.9504 % 7.9504 % 15.9008 %
8 AB -0.5884 2.00000 -1.8755 % -1.8755 % -3.7510 %
9 AB -0.5410 2.00000 -0.7506 % -0.7506 % -1.5011 %
10 AB -0.4663 2.00000 3.8098 % 3.8098 % 7.6197 %
11 AB -0.4472 2.00000 5.5270 % 5.5270 % 11.0539 %
12 AB -0.4016 2.00000 -0.3983 % -0.3983 % -0.7965 %
13 AB -0.3961 2.00000 -0.4274 % -0.4274 % -0.8549 %
14 AB -0.3674 2.00000 -5.8826 % -5.8826 % -11.7653 %
15 AB -0.2661 2.00000 0.1192 % 0.1192 % 0.2383 %
16 AB -0.2438 2.00000 -6.0931 % -6.0931 % -12.1863 %
The product of the cross term composition between fragments 1 and 2 in orbital i and
corresponding orbital occupation number is the Mulliken bond order between them contributed by
orbital i. From above information we can see MO 7 and 11 are beneficial to bonding, because the
composition are relative large, while MO 14 and 16 are not conducive for bonding. The isosurfaces
of MO 11 (left side) and 14 (right side) are shown below, it is clear that the result of composition
analysis is reasonable.
As you have seen, using Mulliken method to analyze orbital composition is very convenient.
However, the result of Mulliken method is sensitive to basis set, and is not as robust as NAO method
and Hirshfeld method illustrated below. Especially, do not use Mulliken method when diffuse
functions are involved in your calculation, otherwise the result will be meaningless!
4.8.2 Analyze water by natural atomic orbital method
In this example we analyze molecular orbital composition of water by the natural atomic orbital
(NAO) method discussed in Section 3.10.4. NAO method has much better basis set stability (i.e.
insensitive to the choice of basis set) and stronger theoretical basis than Mulliken or Mulliken-like
methods (such as SCPA and Stout-Politzer).
Note: NAO is never the only way of obtaining composition of atomic orbitals in MOs, you can also use Mulliken
or similar methods (e.g. SCPA) to do that. The correspondence between basis function and atomic orbital can be
identified according to basis set definition or by mean of population analysis, see Section 4.7.6.
Performing NAO method requires MO coefficient matrix in NAO basis, this matrix cannot be
generated by Multiwfn itself, but Multiwfn can utilize the output information containg this matrix
by stand-alone NBO program or NBO module embedded in quantum chemistry softwares. The
466
NBO 3.1 module embedded in Gaussian program is L607. Below is a Gaussian input file for water,
which will output the matrix we needed. Notice that the Gaussian task should be single point task,
do not perform geometry optimization together!
# HF/6-31g* pop=nboread
Title Card Required
0 1
O 0.00000000 0.00000000 0.11472000
H 0.00000000 0.75403100 -0.45888100
H 0.00000000 -0.75403100 -0.45888100
$NBO NAOMO $END
[blank line]
[blank line]
where pop=nboread keyword indicates that the texts enclosed by $NBO and $END, namely
NAOMO, will be passed to NBO module. NAOMO keyword tells NBO module to output MO
coefficient matrix in NAO basis.
Assume that the Gaussian output file is named as H2O_NAOMO.out (can be found in
"example" folder), we start Multiwfn and input:
examples/H2O_NAOMO.out // Note that DO NOT use .fch as input file in current case
8 // Enter orbital composition analysis module
7 // Enter NAO analysis function
You will find the default output mode is "Only show core and valence NAOs". Core and
valence NAOs have one-to-one correspondence with actual atomic orbitals, if the MO to be
analyzed is occupied, in general we only need to concern these NAOs, while Rydberg NAOs can be
ignored. Assume that we want to analyze MO 4, we input
0 // Show orbital composition of specific MO
4 // Analyze MO 4
Below information will appear on screen
Note: All Rydberg NAOs/shells or contributions <= 0.50 % will not be printed
NAO# Center Label Type Composition
2 1(O ) S Val( 2s) 8.573 %
9 1(O ) pz Val( 2p) 84.089 %
16 2(H ) S Val( 1s) 3.542 %
18 3(H ) S Val( 1s) 3.542 %
Condensed NAO terms to shells:
Atom: 1(O ) Shell: 2( 2s Val) 8.573 %
Atom: 1(O ) Shell: 5( 2p Val) 84.089 %
Atom: 2(H ) Shell: 8( 1s Val) 3.542 %
Atom: 3(H ) Shell: 10( 1s Val) 3.542 %
467
Composition of different types of shells (%):
s: 15.761 p: 84.104 d: 0.130 f: 0.000 g: 0.000 h: 0.000
Condensed NAO terms to atoms:
Center Composition
1(O ) 92.899 %
2(H ) 3.548 %
3(H ) 3.548 %
Core composition: 0.031 %
Valence composition: 99.746 %
Rydberg composition: 0.218 %
According to the result, we can say for example, 2pz atomic orbital of oxygen has 84.09%
contribution to MO 4. The contributions from the NAOs listed above (Rydberg composition is not
included in the present example) are also summed up to atom contributions according to which
center they belong to.
Note that the sum of non-Rydberg compositions (i.e. Core + Valence), as shown above, is not
100 % rather than 99.777 %. To make the physical meaning more clear, I personally recommend to
manually perform renormalization for the result. For example, the composition of 2pz should be
84.089 % / 0.99777=84.277 %. Since before and after the renormalization the difference is only
0.188 %, the renormalization is not necessary for current case. Only when the non-Rydberg
composition is nonnegligible (e.g. larger than 2 %), the renormalization is indispensable.
Calculate fragment contribution to specific MOs

Input 0 to return to last menu. Next, we analyze contribution from the NAOs centered at the
two hydrogens to MOs 1~10.
Input -1 to enter the interface for defining fragment. If you input all, then detailed information
of all NAOs will be listed (this step is optional):
NAO# Atom&Index Type Set&Shell Occupancy Energy (a.u.)
1 O 1 S Cor( 1S) 1.99992 -20.39645
2 O 1 S Val( 2S) 1.74644 -1.14691
...[ignored]
15 O 1 dz2 Ryd( 3d) 0.00254 2.02361
16 H 2 S Val( 1S) 0.52321 0.33308
17 H 2 S Ryd( 2S) 0.00086 0.70497
18 H 3 S Val( 1S) 0.52321 0.33308
19 H 3 S Ryd( 2S) 0.00086 0.70497
We input a 2,3, namely adding all NAOs belonging to atoms 2 and 3 to the current fragment.
Then input q to save and quit. From the prompt printed on screen you can find NAOs 16, 17, 18 and
19 are presented in this fragment.
Then select option 1 and input 1-8, the contribution from the four NAOs to MO 1~8 will be
shown as below
Orb.# Core Valence Rydberg Total
468
1 0.000 % 0.119 % 0.002 % 0.121 %
2 0.000 % 18.556 % 0.054 % 18.611 %
3 0.000 % 26.557 % 0.018 % 26.576 %
4 0.000 % 7.084 % 0.012 % 7.096 %
5 0.000 % 0.000 % 0.000 % 0.000 %
6 0.000 % 35.482 % 46.832 % 82.314 %
7 0.000 % 27.558 % 61.538 % 89.096 %
8 0.000 % 32.433 % 29.568 % 62.002 %
Since none of the four NAOs in the fragment is core-type, the Core term is 0 % in the MOs.
Valence and Rydberg terms correspond to the contribution from NAOs 16, 18 and NAOs 17, 19
respectively. NAOs 16 and 18 directly correspond to 1s atomic orbital of H2 and H3, so we can say
that the two hydrogens collectively contribute 26.56 % to MO 3.
The first five MOs are doubly occupied in present system. It is clear that Rydberg NAOs have
very low contribution to the occupied MOs, while their contributions to virtual MOs are significant
and can no longer be ignored. The physical meaning of Rydberg NAOs is difficult to be interpreted,
and these NAOs do not directly reflect atomic orbital characteristics. It is questionable to say that
the two hydrogens contribute either 32.43 % or 62.00 % to MO 8. Although seemingly one can
employ renormalization process to "annihilate" the Rydberg composition, however when Rydberg
composition is too large, e.g. larger than 10 %, this treatment will break meaning of the result. So it
is not generally recommended to use NAO method to analyze atomic contributions to virtual MOs;
for this case, the Hirshfeld and Becke method introduced in Section 3.10.5 and exemplified in the
next section are the best choice.
4.8.3 Analyze acetamide by Hirshfeld and Becke method
In this section, we will first use Hirshfeld method and then Becke method to analyze the MO
composition of acetamide and compare the result with the one obtained by Mulliken method in
Section 4.8.1. Note that Hirshfeld and Becke methods are only capable of analyzing composition of
atom or fragment in orbitals, while the composition of atomic orbitals are impossible to be obtained
by these approaches.
examples\CH3CONH2.fch // You can also use such as .wfn and .wfx file as input. But .wfn
and .wfx files do not contain virtual orbital information!
8 // Use Hirshfeld partition
Hirshfeld analysis requires electron density of atoms in their free-states, you need to choose a
method to calculate atomic densities. Selecting 1 to use built-in atomic densities is very convenient,
see Appendix 3 for detail; alternatively, you can select 2 to evaluate atomic densities based on
atomic .wfn files, see Secion 3.7.3 for detail. Here we choose option 1.
Then Multiwfn initializes the data, for large system you may need to wait for a while. Assume
that you want to analyze MO 6, then simply input 6, the result will be printed on screen, as shown
below. (Because the integrals are evaluated numerically, the sum of all terms will be slightly
deviated to 100%, so Multiwfn automatically normalizes the result.)
Atom 1(C ) : 1.555%
469
Atom 2(H ) : 0.249%
Atom 3(H ) : 0.155%
Atom 4(H ) : 0.037%
Atom 5(C ) : 14.922%
Atom 6(O ) : 12.109%
Atom 7(N ) : 56.339%
Atom 8(H ) : 6.688%
Atom 9(H ) : 7.946%
The composition of C5, O6 and N7 are 14.92%, 12.11% and 56.34%, respectively. This result
is close to the one obtained by Mulliken method (Section 4.8.1), namely 11.81%, 11.63% and
65.50%, respectively. In fact, for occupied MOs, if diffuse basis functions are not employed, in
general Mulliken, NAO and Hirshfeld methods give similar results.
Now let us check the composition of 7N in MO from 14 to 19. We input

-2 // Print atom contribution to a range of orbitals
7 // Atom index
14-19 // Orbital range
You will see:
Orb# Type Ene(a.u.) Occ Composition Population
14 Alpha&Beta -0.3674 2.000 16.072% 0.321446
15 Alpha&Beta -0.2661 2.000 48.508% 0.970161
16 Alpha&Beta -0.2438 2.000 6.792% 0.135840
17 Alpha&Beta 0.0410 0.000 12.378% 0.000000
18 Alpha&Beta 0.0762 0.000 21.755% 0.000000
19 Alpha&Beta 0.1252 0.000 14.512% 0.000000
Population of this atom in these orbitals: 1.427447
where 1.427447 (namely 0.321446+0.970161+0.135840) is the total population number of N7 in

MO 14~19.
PS: If the orbital range you specified is 1~16, namely all occupied MO, then the outputted value 7.1589 will be
the atomic population number of N7, and its Hirshfeld atomic charge is therefore 7.0-7.1589 = -0.1589.
Next, we examine contribution of the amino group to specific orbital. Input below commands:
7-9 // The atoms in the amino group
16 // The index of HOMO
As shown below, you can not only see contribution from all atoms to the orbital, but you can
also find the fragment contribution to the orbital, the 8.703% is simply 6.791%+0.974%+0.938%.
[...ignored]
Atom 6(O ) : 69.178%
Atom 7(N ) : 6.791%
Atom 8(H ) : 0.974%
Atom 9(H ) : 0.938%
Fragment contribution: 8.703%
470
The steps of analyzing orbital composition by Becke method are completely identical to that
of Hirshfeld method. Here we calculate the composition of MO 6. Boot up Multiwfn and input
9 // Use Becke partition
6 // The 6th orbital
The result is
Atom 1(C ) : 1.229%
Atom 2(H ) : 0.085%
Atom 3(H ) : 0.049%
Atom 4(H ) : 0.002%
Atom 5(C ) : 14.902%
Atom 6(O ) : 12.070%
Atom 7(N ) : 60.742%
Atom 8(H ) : 4.929%
Atom 9(H ) : 5.992%
As you can see, the result is highly close to that produced by Hirshfeld method.
In Multiwfn it is also possible to use Hirshfeld-I partition to calculate orbital composition,

however this is not commonly employed, because generating Hirshfeld-I atomic space requires
additional computational cost, while the result is not greatly improved (the orbital composition
computed by Hirshfeld method is already reliable and meaningful enough).
4.8.4 Calculate oxidation state by LOBA method
Please read Section 3.10.100 first to understand basic idea of the LOBA method. This is a
simple and useful method to evaluate oxidation state (OS). In this section I will use two examples
to illustrate the LOBA module in Multiwfn. The used .fch files can be directly loaded at
http://sobereva.com/multiwfn/extrafiles/LOBA.rar.
(1) Fe(CN)63-
First, we use Gaussian to perform regular calculation of this system, the input file is
examples\Fe(CN)6_3-.gjf, please run it by yourself to get Fe(CN)6_3-.fch file. LOBA analysis needs
localized MO (LMO), thus we use Multiwfn to carry out orbital localization. Boot up Multiwfn and
input following commands:
Fe(CN)6_3-.fch
1 // Only localize occupied orbitals, this is enough for LOBA analysis
Now the orbitals recorded in memory has been updated to localized orbitals
0 // Return to main interface
100 // LOBA analysis
50 // Percentage threshold for performing LOBA
Oxidation state of atom 1(Fe) : 3
Oxidation state of atom 2(C ) : 2
471
Oxidation state of atom 8(N ) : -3
The sum of oxidation states: -3
The result looks reasonable and in good agreement with chemical intuition. The sum of all
oxidation states just corresponds to the total net charge of -3. However, the LOBA is not free of
ambiguity, the choice of the threshold is highly arbitrary. If we input 60 instead of 50, we will see
oxidation state of carbon and oxygen become 4 and -1, respectively, and the sum of oxidation states
become 21. Fortunately, the OS of transition metal is never so sensitive to the choice of threshold,
the iron always keeps +3 oxidation state in present case as long as the threshold is not set to very
small or very large. (In my viewpoint, the most appropriate threshold for evaluating OS of transition
metal is 50~60%)
(2) Ferrocene
For this system, we will not only check OS of iron, but also check OS of C5H5 fragment. The
corresponding regular Gaussian input file is examples\Ferrocene.gjf, run it by yourself to obtain
corresponding .fch file, then load it into Multiwfn and perform orbital localization first as shown
above, after that enter LOBA analysis interface and input below commands:
1-5,7-11 // Index of the atoms constituting the C5H5 fragment
50 // Percentage threshold for performing LOBA
The result is
Oxidation state of atom 4(H ) : 1
Oxidation state of atom 5(H ) : 1
Oxidation state of atom 6(Fe) : 2
...[ignored]
The sum of oxidation state: 32
Oxidation state of the fragment: -1
In this system the OS of Fe is +2, which is again reasonable. From the output it is seen that the
OS of individual carbons are not useful, however, the OS of the whole C5H5 fragment is a
meaningful value -1.
Above two examples exhibit usefulness of LOBA method. We should always focus on OS of
472
transition metal or OS of whole ligands, while the OS of individual atom in ligand often does not
make sense.
4.8.5 Quantifying extent of spatial delocalization of orbitals via orbital

delocalization index (ODI)
When Multiwfn outputs composition of various atoms in an orbital, the orbital delocalization
index (ODI) is also printed. The ODI was defined by me, the value for orbital i is expressed as
ODIi = 0.01   ( A,i )2
A
where A,i is composition of atom A in orbital i.

The ODI a useful indicator of quantifying extent of orbital spatial delocalization, the lower
(higher) the ODI, the stronger the orbital delocalization (localization). In this section, I will take a
practical molecule to demonstrate its usefulness and reliability.
If you are familiar with Pipek-Mezey orbital localization method, you can easily understand idea of the ODI.
As a very simple instance, we consider two orbitals. The first one is fully localized on an atom, while another one is
equally distributed on two atoms, then the ODI for the first orbital will be (1002)/100=100, while that for the second
orbital will be (502+502)/100=50. Since the latter is much smaller than the former, the second orbital is much more
delocalized than the first one.

examples\excit\D-pi-A.fchk
1 // Mulliken method
52 // Analyze MO 52
You will find below output, namely the ODI of MO 52 calculated by Mulliken method is 44.28
Similarly, we compute and record ODI for MO 16, MO 53, MO 55, MO 56, MO 62. Note that
only MO62 is a virtual orbital.
For comparison purpose, we also calculate the ODI based on Hirshfeld orbital composition
analysis method. Input below commands
0 // Return
8 // Hirshfeld method
1 // Use built-in atomic density
52 // Analyze MO52
The output is
Similarly, we use the Hirshfeld method to compute and record ODI for MO 16, MO 53, MO
55, MO 56, MO 62.
The isosurfaces of the analyzed MOs under isovalue of 0.04 are summarized below, the red
and blue texts correspond to the ODI calculated by Mulliken and Hirshfeld methods, respectively.
473
By comparing the ODI values and orbital isosurface maps it can be seen that the ODI value is
indeed able to faithfully quantify extent of orbital delocalization. The MO 16 is essentially a core
orbital of a carbon atom, since it is fully localized, the ODI nearly reaches its theoretical upper limit
(100). The MO 52 shows partial delocalization character, the two oxgens mainly and equally
contribute to the orbital, therefore its ODI is not quite high. The MO 55 corresponds to  orbital of
a ring and thus evidently distributes on more than two atoms, this is why its ODI is lower than MO
52. The MO 53 and MO 56 show strong global delocalization character, therefore their ODI values
are quite low. Because their ODI values are comparable, it can be concluded that MO 53 and MO
56 have similar extent of spatial delocalization.
The virtual orbital MO 62 is quite worth to mention, it is essentially a Rydberg orbital and its
main body surrounds the amino group. From the isosurface map it can be seen that MO 62 and MO
55 have comparable delocalization character, the ODI computed by Hirshfeld method for the two
orbitals (18.4 vs. 18.2) is in line with this observation. However, the ODI of MO 62 computed by
Mulliken method is much larger than MO 55, this is totally contrary to reality, showing the fact that
Mulliken (and its variant, SCPA and Stout-Politzer) is usually unreliable in calculating ODI for
virtual orbitals.
Also bear in mind that since Mulliken, SCPA and Stout-Politzer are incompatible with diffuse
functions, they should not be used to evaluate ODI when diffuse functions are present, in this case
you should use e.g. Hirshfeld and NAO methods instead.
In summary, I suggest using Hirshfeld method to compute ODI, it works well for any case.
However, if you only need to analyze occupied orbitals and diffuse function is not employed, you
can also use Mulliken or SCPA method, which is faster than Hirshfeld method for large system.
Calculating ODI for a batch of orbitals

The Hirshfeld, Hirshfeld-I and Becke orbital composition analysis module is able to directly
calculate ODI for a batch of orbitals. For instance, here we calculate ODI for all occupied MOs for
the D--A system we studied above.
474

1 // Use built-in atomic density
-5 // Print ODI for a batch of orbitals
1-56 // The range of occupied MOs
We immediately obtain below result
Orb: 1 Ene(a.u.): -19.246317 Occ: 2.0000 Type: Alpha&Beta ODI: 55.03
[ignored...]
You can copy out the data and plot it as bar map:
100
90
80
70
60
50
ODI
40
30
20
10
0
5 10 15 20 25 30 35 40 45 50 55
MO index
From this graph we can very quickly identify the orbitals showing significant delocalization
character. The first 16 MOs in this system are core orbitals, as can be seen from the map, they are
much more localized than valence orbitals. The graph exhibits that MOs 50, 51 and 52 are also
highly localized, if you inspect their isosurface maps via main function 0, you will find they mainly
localize over the nitro group.
ODI for a fragment

In order to measure orbital delocalization extent on a specific fragment, I defined fragment
ODI:
2
 
ODI = 0.01    A,i 
frag
i
Afrag  pi 
pi = 0.01    A,i
Afrag
where p is normalization factor to account for the difference of total amount of orbital distribution
on different fragments. If the fragment contains all atoms, then the ODIfrag will be identical to the
aforementioned ODI.
Clearly, fragment ODI is very useful if you want to quantitatively compare orbital
475
delocalization for a fragment shared by analogues. Currently, only Hirshfeld, Hirshfeld-I and Becke
orbital composition analysis modules can calculate fragment ODI. Now, let see an example.
As vividly shown in the orbital isosurface maps of the D-pi-A.fchk given above, for the amino
group, the MO 53 fully localizes on the nitrogen atom, while the MO 56 delocalizes over the entire
group. Now we use fragment ODI to quantify this point. Boot up Multiwfn and input
24-26 // Index of the atoms in the amino group
Next, if you input 53, you will see
Orbital delocalization index of the fragment: 77.26
if inputting 62, you will see

Orbital delocalization index of the fragment: 33.41
Since fragment ODI of MO 62 is significantly smaller than that of MO 53, it is clear that
delocalization of MO 62 over amino group is much stronger than MO 53.
Note that you can also use the option "Print orbital delocalization index (ODI) for a batch of
orbitals" to calculate ODI and ODIfrag for a batch of orbitals. The ODI values will be printed
followed by ODIfrag values.
4.8.6 Calculate orbital composition contributed by AIM basins and

other type of basins
As mentioned in Section 3.10.7, Multiwfn is able to compute orbital composition based on

AIM partition via basin analysis module (main function 17), in other words, calculate orbital
composition contributed by AIM basins. In addition, due to the extremal flexiblity of the basin
analysis module, it is also possible to calculate orbital composition contributed by other kinds of
basins, such as ELF basins, electrostatic potential basins and Fukui function basins. In this section,
I will take CH3COCl as example to illustrate this point. You can use any kind of file as input file as
long as it contains GTF information, see Section 2.6 for more information.
If you are not familiar with basin analysis and find difficulty in understanding below examples, you are
suggested to read Section 3.20 to gain basic knowledge about basin analysis and check Section 4.17 to familiarize
yourself with the use of basin analysis module.
Calculate contributions of AIM basins to molecular orbitals

examples\CH3COCl.wfn // You can also use other formats, e.g. wfx/fch/molden/mwfn...
Note that .wfn and .wfx only contain occupied orbitals
1 // Use electron density to partition basins, namely yielding AIM basins
11 // Calculate orbital compositions contributed by various basins
476
Now you can directly input index of an orbital to calculate its composition. For example, we
input 5, you will see
Final data after normalization:
Basin: 1 Contribution: 0.731 %
Contributions from atoms:
1 (C ) Contribution: 6.860 %
2 (H ) Contribution: 0.731 %
5 (C ) Contribution: 14.708 %
6 (O ) Contribution: 4.527 %
7 (Cl) Contribution: 72.113 %
As you can see, Multiwfn first outputs contributions from various basins, and then, in order to
facilitate inspection, the contributions are outputted again according to the order of atoms
(commonly AIM basins and atoms have a one-to-one correspondence). If you use Hirshfeld method
to compute the composition of this orbital, you will find the result is similar; for example, Cl7 is
69.6% and C5 is 16.3%. Since generation of AIM basins is quite time-consuming for large system,
while AIM partition does not have special advantage, usually I recommend to use Hirshfeld or
Becke method to compute atomic contribution.
In the present interface, you can also use option -9 to define a set of atoms as fragment, so that
the fragment contribution can be outputted together. Besides, you can use option -4 to output all
atomic contributions in all orbitals to orbcomp.txt in current folder.
Calculate contributions of ELF basins to molecular orbitals

This time we will partition molecular space based on ELF, so that contributions from various
ELF basins to specific orbital can be obtained. Since each ELF basin usually corresponds to a local
region with featured electronic structure, this analysis may be useful in characterizing orbitals.
examples\CH3COCl.wfn
9 // Use ELF to partition basins
11 // Calculate orbital compositions contributed by various basins
10 // Study the 10th MO
Then you will see
477
In order to understand the chemical meaning of the data, we can select 0 to return to upper
level of menu, and then select option "0 Visualize attractors and basins". In the GUI window, select
basin 5 to visualize it, you will see the left figure shown below. If you use main function 0 to plot
isosurface map of the orbital 10, you will see the right figure shown below (isovalue = 0.09).
It can be seen that the orbital 10 partially exhibits bonding character between C5 and Cl7, while
basin 5 directly corresponds to bonding basin and can be symbolized as V(C5,Cl7), evidently this
basin should have notable contribution to orbital 10. From the orbital composition data shown above,
the contribution by basin 5 is indeed prominent (18.2 %), showing that the ELF basin contributions
evaluated in the aforementioned way is reasonable.
4.9 Bond order analysis
In this section I will illustrate how to use Multiwfn to perform different kinds of bond order
analyses to characterize chemical bonds.
478
4.9.1 Mayer bond order and fuzzy bond order analysis on acetamide
This instance exemplifies how to calculate Mayer bond order and fuzzy bond order for
acetamide. Related theories have been introduced in Sections 3.11.1 and 3.11.6, respectively. Finally,
I introduce a skill, namely labelling the calculated bond orders to molecular structure map by means
of GaussView, so that you can examine bond orders easier.
Calculation of Mayer bond order

We first calculate Mayer bond order. Note that calculating Mayer bond order requires basis
function information, thus currently .mwfn/.fch/.molden/.gms file must be used as input file.
9 // Bond order analysis
1 // Calculate Mayer bond order
Immediately you get below output:
Bond orders with absolute value >= 0.050000
# 1: 1(C ) 2(H ) 0.93802674
# 2: 1(C ) 3(H ) 0.93473972
# 3: 1(C ) 4(H ) 0.94494566
# 4: 1(C ) 5(C ) 0.96585484
# 5: 5(C ) 6(O ) 1.90392771
# 6: 5(C ) 7(N ) 1.11849509
# 7: 6(O ) 7(N ) 0.07620305
# 8: 7(N ) 8(H ) 0.83250273
# 9: 7(N ) 9(H ) 0.83869874
Total valences and free valences defined by Mayer:
Atom 1(C ) : 3.77555991 0.00000000
Atom 2(H ) : 0.93147308 0.00000000
Atom 3(H ) : 0.92778456 0.00000000
Atom 4(H ) : 0.93657474 0.00000000
Atom 5(C ) : 3.97788022 0.00000000
Atom 6(O ) : 2.05925868 0.00000000
Atom 7(N ) : 2.85041375 0.00000000
Atom 8(H ) : 0.86522064 0.00000000
Atom 9(H ) : 0.85875080 0.00000000
By default, only the bond order terms larger than specific criteria will be outputted, the criteria
can be adjusted in "bndordthres" in settings.ini. Mayer bond order often coincide with empirical
bond order well. In this example, bond order between C5 and O6 is 1.9, which is very close to ideal
value 2.0 (double bonds).
Total valence of an atom is the sum of Mayer bond orders that it formed. Free valence of a
atom measures its remained capacity to form new bonds by sharing electron pairs, for closed-shell
this quantity is always zero.
Then if you choose "y", entire bond order matrix will be outputted to bndmat.txt in current
folder.
479
Orbital occupancy-perturbed Mayer bond order analysis

Next, we want to try to find which orbitals have main contributions to Mayer bond order
between C5 and O6, calculating the so-called "Orbital occupancy-perturbed Mayer bond order" is
useful for realizing this goal. Hence, we select option 6 in bond order analysis module, and then
input 5,6. Below information will be outputted:
Mayer bond order before orbital occupancy-perturbation: 1.903928
Orbital Occ Energy Bond order Variance
1 2.00000 -19.10356 1.906089 0.002162
2 2.00000 -14.34920 1.903934 0.000006
3 2.00000 -10.28192 1.905514 0.001586
4 2.00000 -10.18447 1.903939 0.000011
5 2.00000 -1.03758 1.606912 -0.297016
6 2.00000 -0.90529 1.825661 -0.078267
7 2.00000 -0.73868 1.872011 -0.031917
8 2.00000 -0.58838 1.896333 -0.007594
9 2.00000 -0.54103 1.868127 -0.035801
10 2.00000 -0.46630 1.799351 -0.104576
11 2.00000 -0.44722 1.512533 -0.391394
12 2.00000 -0.40158 1.830827 -0.073101
13 2.00000 -0.39610 1.659075 -0.244853
14 2.00000 -0.36742 1.393122 -0.510805
15 2.00000 -0.26611 1.743383 -0.160545
16 2.00000 -0.24383 1.796157 -0.107770
Summing up occupancy perturbation from all orbitals: -2.03987
From the output we can know that, for example, if the two electrons are removed from orbital
15, then Mayer bond order between C5 and O6 will decreased from 1.903928 to 1.743383, we can
also say that the contribution from orbital 15 is 0.160545. The sum of contributions from all
occupied MOs is 2.03987, the reason that this value is not equal to 1.903928 is that Mayer bond
order is not a linear function of density matrix, we do not need to concern this.
Orbital 14 has the largest negative value of orbital occupancy-perturbed Mayer bond order,
therefore this orbital must be greatly beneficial to the bonding. This conclusion can be further
testified by visual inspection of the orbital isosurface, see the graph given at the end of Section 4.8.1.
As expected, this orbital shows strong character of π-bonding between C5 and O6.
Calculation of fuzzy bond order

Now we calculate fuzzy bond order. Unlike Mayer bond order, fuzzy bond order does not rely
on basis function, therefore you can also use such as .wfn/.wfx as input file. Calculating fuzzy bond
order is more time consuming than Mayer bond order, its advantage over Mayer bond order is that
the basis set sensitivity is greatly reduced, and using diffuse basis functions will never deteriorate
result.
In the bond order analysis menu, we select "7 Fuzzy bond order analysis", the result will be
printed, as shown below
Bond orders with absolute value >= 0.050000
# 1: 1(C ) 2(H ) 0.89666213
480
# 2: 1(C ) 3(H ) 0.89075146
# 3: 1(C ) 4(H ) 0.88888051
# 4: 1(C ) 5(C ) 1.08050669
# 5: 1(C ) 6(O ) 0.13600686
# 6: 1(C ) 7(N ) 0.11096284
# 7: 3(H ) 5(C ) 0.05357850
# 8: 5(C ) 6(O ) 2.00411901
# 9: 5(C ) 7(N ) 1.40339002
# 10: 6(O ) 7(N ) 0.24523041
# 11: 7(N ) 8(H ) 0.87233783
# 12: 7(N ) 9(H ) 0.88243512
Comparing the result with that of Mayer bond order, you will find the results of both types of
bond orders are very similar, in fact this is the common case. However, for highly polar bonds their
results may deviate with each other relatively evidently.
Skill: Labelling bond orders on molecular structure map by GaussView

If you have GaussView (version  6.0), you can use it to show the bond orders calculated by
Multiwfn on the molecular structure map to facilitate examining their values. Here I use Mayer bond
order of acetamide as instance to illustrate this point.
y // Export the bond order matrix as bndmat.txt in current folder
1000 // Hidden main function
13 // Convert the bndmat.txt in current folder to Gaussian .gjf file with bond order
information
Now we have gau.gjf in the current folder, which not only contains present molecular
coordinate, but also contains bond orders between the connected atoms (the connectivity is
automatically guessed based on current geometry, unless you employ a file containing connectivity
information as input file, such as .mol and .mol2, see Section 2.5 for detail).
Load the gau.gjf into GaussView, select "Results" - "Bond Properties", then after proper
adjustments, you can obtain below effect.
481
As can be seen, we also requested GaussView to use different colors to exhibit the bond orders. The
more green the color, the larger the bond order; the redder the color, the smaller the bond order.
4.9.2 Multi-center bond order analysis on Li6 cluster and phenanthrene
The electron structure character of complex systems, such as cluster or system containing wide
range electron delocalization is hard to be investigated by simple chemitry empirical rules, we have
to resort to wavefunction analysis methods. In this section, examples of applying multi-center bond
order to reveal multi-center interaction are given. If you are not familiar with multi-center bond
order, please check Section 3.11.2 to gain basic knowledge. Notice that multi-center bond order
analysis requires basis function information, therefore you have to use .mwfn/.fch/.molden/.gms file
as input file.
Part 1: Studying three-center bond in Li6 cluster

In the planar Li6 cluster, as shown in below map, there are two kinds of three-membered rings,
namely the three boundary ones and the central one. We will use multi-center bond order to study
which kind of three-membered ring is more stable.
examples\Li6.fch
2 // Multi-center bond order analysis
1,3,4 // Indices of the atoms in the boundary three-member ring
The output is
The multicenter bond order: 0.1247848038
The normalized multicenter bond order: 0.4997129069
Then we calculate three-center bond order of the central three-member ring, therefore we input
1,2,3, the result is
Since the number of atoms in both the rings is the same, you only need to compare their "The
multicenter bond order" values. The data are marked in the below graph. The pink texts denote
Mayer bond orders.
482
From the three-center bond order values, it is evident that the boundary three-member rings
are more stable (i.e. more strongly binded) than the central one, this conclusion is also somewhat
reflected by the Mayer bond orders. We can further demonstrate this conclusion by plotting LOL
graph in the cluster plane (see Section 4.4.2 on how to plot this kind of map)
It is clear that electrons tend to localize in the boundary three-membered rings to stabilize them,
the conclusion of this real space function analysis is in good agreement with the bond order analysis.
By checking Laplacian map, ELF map, electron density deformation and valence electron density
map, you can draw exactly the same conclusion.
The Li6.fch used in this example was produced at B3LYP/6-31G* level. Usually diffuse
functions should be employed for properly describing anionic systems, in this case you should
483
evaluate the multi-center bond order based on natural atomic orbitals (NAO) rather than based on
original basis functions as illustrated above, otherwise the result may be completely useless, see
Section 3.11.2 for detail (also see Part 3 of this section for example). Alternatively, you can remove
diffuse functions and perform single point task to generate the wavefunction used for multi-center
bond order analysis, however the basis set should be at least three-zeta quality, e.g. 6-311G(2d,p) or
def2-TZVP.
Part 2: Studying six-center conjugation in phenanthrene

The examples\phenanthrene.fch contains wavefunction of phenanthrene generated at
B3LYP/6-31G* level. The atomic numbering is shown below. In this instance, we will use multi-
center bond order to study which six-membered ring has stronger multi-center conjugation effect.

examples\phenanthrene.fch
2 // Multi-center bond order analysis
1,2,3,4,5,6 // Indices of the atoms in the boundary ring. Note that the inputting order must be
in consistency with atomic connectivity, namely inputting such as 1,3,5,6,4,2 will be meaningless
The output is

Note: It is worth to mention that in this case if you input the atomic indices in reversed order, namely 6,5,4,3,2,1,
the result will be different, namely 0.0591177129. However, since the difference between 0.05935 and 0.05911 is
marginal, we do not discrminate them. More information about influence of input order is mentioned Section 3.11.2
Next, we study the case of the central ring. We input 3,4,8,9,10,7, the result is
Clearly, the central ring has weaker electron conjugation character than the boundary one,
consequently we can also conclude that the boundary rings have stronger aromaticity. In Section
4.14.3, 4.15.2 and 4.100.13 we will further investigate the ring aromaticity by means of other
analysis methods.
Note that the data of "The normalized multicenter bond order" can be compared between rings
with different number of atoms. Since this value of boundary six-membered ring of phenanthrene
is 0.6245, while that of the boundary three-membered ring of Li6 cluster is 0.4997, we can infer that
the multi-center interaction in the former case may be more prominent.
Part 3: Calculate six-center bond order based on natural atomic orbitals (NAOs)
In Multiwfn, multi-center bond order can also be calculated based on natural atomic orbitals
(NAOs), as introduced in Section 3.11.2. The main advantage of using NAO as basis over the
common case is that reasonable result can still be obtained even diffuse functions are presented.
484
Here we calculate multi-center bond order in NAO basis for the phenanthrene, in this case
NBO output information with DMNAO keyword is required. The Gaussian input file involved in
this example is exampes\phenanthrene_DMNAO.gjf, the corresponding output file is
examples\phenanthrene_DMNAO.out. As you can see from the .gjf file, the NBO module embedded
in Gaussian is invoked and DMNAO keyword is passed into NBO module.
examples\phenanthrene_DMNAO.out
-2 // Multi-center bond order analysis in NAO basis
1,2,3,4,5,6 // Calculate six-center bond order for the boundary ring
The output is
As can be seen, the result is almost completely identical to the one we obtained earlier using option
2 in main function 9, it is expected since currently diffuse functions are not employed.
4.9.3 Calculate Laplacian bond order (LBO)
The Laplacian bond order (LBO) was proposed by me in J. Phys. Chem. A, 117, 3100 (2013),
see Section 3.11.7 for detail. LBO is very suitable for organic system and has close correlation with
bonding strength. Let us calculate LBO for C-C bond of ethane, ethene and acetylene.
examples\ethane.wfn // Optimized and produced at B3LYP/6-31G**
8 // Laplacian bond order
You will see the result:
The bond order >= 0.050000
# 1: 1(C ) 2(H ): 0.887111
# 2: 1(C ) 3(H ): 0.889492
# 3: 1(C ) 4(H ): 0.889492
# 4: 1(C ) 5(C ): 1.059879
# 5: 5(C ) 6(H ): 0.887111
# 6: 5(C ) 7(H ): 0.889492
# 7: 5(C ) 8(H ): 0.889492
As you can see, LBO is very close to formal bond order (1.0) for C-C and C-H. LBO only
reflects covalent bonding character, due to C-H is a weakly polar bond, the value is slightly smaller
than 1.0.
Then use examples\ethene.wfn to calculate LBO for ethene
# 1: 1(C ) 2(H ): 0.919443
# 2: 1(C ) 3(H ): 0.919443
# 3: 1(C ) 4(C ): 2.022583
# 4: 4(C ) 5(H ): 0.919443
# 5: 4(C ) 6(H ): 0.919443
Then calculate LBO for acetylene by using examples\C2H2.wfn
485
# 1: 1(C ) 2(H ): 0.958393
# 2: 1(C ) 3(C ): 2.767449
# 3: 3(C ) 4(H ): 0.958393
The LBO of the C-C bonds in the three systems are 1.060, 2.022 and 2.767, the ratio is
1:1.907:2.61. It is known that the ratio of the bond dissociation energy (BDE) of the three bonds is
1:1.85:2.61. Clearly, LBO has surprisingly good correlation with BDE, in other words, LBO exhibits
bonding strength fairly well (no other bond order definitions have so close relationship with BDE
in comparison with LBO)
Moreover, LBO predicts that the sequence of the C-H bonding strength in the three systems is
acetylene (0.958) > ethene (0.919) > ethane (0.889), this is completely in agreement with the
experimental BDE sequence! (Other bond order definitions, such as Mayer bond order, fail to
reproduce this sequence)
Finally, calculate LBO for O-H bond in water by using examples\H2O.fch, the result is 0.638.
This value is significantly smaller than the C-H bond order, reflecting that O-H bond is much more
polar than C-H bond.
4.9.4 Decomposition analysis of Wiberg bond order in NAO basis for

formaldehyde
This example briefly illustrates a unique feature of bond order analysis module of Multiwfn,
namely decomposing Wiberg bond order to atomic orbital pair and atomic shell pair contributions.
A very simple molecule formaldehyde will be used as example, of course you can extend the
analysis to much more complicated systems. Please read Section 3.11.8 first to understand basic
idea of this analysis method.
This analysis requires natural atomic orbital (NAO) information and density matrix in NAO
basis outputted by Weinhold's NBO program. For Gaussian user, you can run
examples\H2CO_DMNAO.gjf and use the corresponding output file (examples\H2CO_DMNAO.out)
as input file for this analysis. The orientation of the H2CO molecule in Cartesian system is shown
in below graph.
Boot up Multiwfn and input below command:

examples\H2CO_DMNAO.out
486
9 // Decompose Wiberg bond order in NAO basis

Then you can input two atom indices to obtain their Wiberg bond order calculated under NAO
basis, and meantime obtain major components (the threshold for printing components is controlled
by "bndordthres" parameter in settings.ini). For example, we input 1,4, below result is immediately
shown on screen:
Contribution from NAO pairs that larger than printing threshold:
Contri. NAO Center NAO type NAO Center NAO type
0.0823 2 1(C ) Val( 2S) S --- 21 4(O ) Val( 2S) S
0.1907 2 1(C ) Val( 2S) S --- 28 4(O ) Val( 2p) pz
0.9145 5 1(C ) Val( 2p) px --- 24 4(O ) Val( 2p) px
0.0658 7 1(C ) Val( 2p) py --- 26 4(O ) Val( 2p) py
0.2482 9 1(C ) Val( 2p) pz --- 21 4(O ) Val( 2S) S
0.3700 9 1(C ) Val( 2p) pz --- 28 4(O ) Val( 2p) pz
Contribution from NAO shell pairs that larger than printing threshold:
Contri. Shell Center Type Shell Center Type
0.0823 2 1(C ) 2S --- 2 4(O ) 2S
0.1907 2 1(C ) 2S --- 5 4(O ) 2p
0.2482 5 1(C ) 2p --- 2 4(O ) 2S
1.3504 5 1(C ) 2p --- 5 4(O ) 2p
Total Wiberg bond order: 1.9161
From above information, the detail of total Wiberg bond order of 1.9161 becomes quite clear.
According to the molecular graph shown earlier, the px type of NAO corresponds to the 2p atomic
orbital perpendicular to molecular plane, thus the px-px mixing results in  bond, its contribution to
the total bond order (0.9145) is close to unity, which is in line with chemical intuition. The 2s-2s
interaction only has weak contribution to the C=O bond, since the value 0.0823 is almost negligible;
the reason should be attributed to the fact that the orbital overlap is insufficient. In addition, the 2py-
2py interaction also plays insignificant role, the contribution is merely 0.0658. The interaction
between 2s(C)-2pz(O), 2pz(C)-2pz(O) and 2pz(C)-s(O) have remarkable contribution to total bond
order, which are 0.1907, 0.3700 and 0.2482, respectively, and the sum reaches as high as 0.8089.
The large contributions must mainly stem from good orbital overlapping.
In order to facilitate discussion, the program also outputs contribution to Wiberg bond order
from various atomic shell pairs. For example, as you can see from above information, interaction
between all 2p orbitals of carbon and all 2p orbitals of oxygen totally contributes 1.3504 of bond
order.
4.9.5 Study orbital contributions to Mulliken bond order for C-C bond
of CH3CONH2
The Mulliken bond order has been introduced in Section 3.11.4, it is also known as Mulliken
overlap population. This kind of bond order is not particularly useful, since it neither correlates well
with bonding strength nor closely related to bond multiplicity. However, a unique advantage is that
487
it can be exactly decomposed into orbital contributions, and positive and negative value correspond
to bonding and anti-bonding effect, respectively, this feature is useful for unraveling characteristics
of orbitals. In this section I will use CH3CONH2 as example to illustrate this point.
5 // Decompose Mulliken bond order between two atoms to orbital contributions
1,5 // Decompose C1-C5 bond
The result is
...[ignored]
Orbital 7 Occ: 2.000000 Energy: -0.738679 contributes 0.30341308
Orbital 8 Occ: 2.000000 Energy: -0.588379 contributes -0.03865545
Total Mulliken bond order: 0.66486037
It can be seen that many MOs have evident positive contributions, such as MO7 (0.303), and
MO10 (0.209); a few MOs have negative contributions, especially MO14 (-0.111). There are also
some MOs have almost vanished contributions, such as MO15 (0.005). Therefore, occupation of
MO7 and MO10 should enhance the strength of the C1-C5 bond, while occupation of MO14 must
be harmful for formation of the C1-C5 bond.
The value of the MO contributions to Mulliken bond order can also be understood in terms of
orbital isosurface map:
For MO7 and MO10, from above graph it can be seen that there is no nodal plane between C1
488
and C5, the isosurface substantially encloses the C1-C5 bonding region, thus MO7 and MO10 act
as bonding orbital for C1-C5 and have positive contribution to its Mulliken bond order. For MO14,
an evident nodal plane perpendicular to the C1-C5 bond can be clearly seen at the midpoint of C1-
C5, clearly MO14 behaves as an anti-bonding orbital for C1-C5 bond and thus should have negative
contribution to its Mulliken bond order. For MO15, the C1-C5 bonding region is not covered by the
orbital isosurface, this is why contribution from MO15 to C1-C5 Mulliken bond order is negligible.
Beware that in rare cases the contribution values to Mulliken bond order cannot be well
explained by isosurface map, showing deficiency of definition of Mulliken bond order. In this case
you may try to use the orbital occupancy-perturbed Mayer bond order instead (as illustrated in
Section 4.9.1), which is more robust.
It is noteworthy that decomposition of Mulliken and Mayer bond order can be carried out not
only based on molecular orbitals, but also based on localized molecular orbitals (LMOs), in the
latter case the discussion is usually more meaningful. To do so, in general you should use main
function 19 to yield LMOs and then carry out the decomposition analyses as usual.
4.9.6 Using intrinsic bond strength index (IBSI) to measure strength of

chemical bonds
The intrinsic bond strength index (IBSI) proposed in J. Phys. Chem. A, 124, 1850 (2020) has
certain ability in characterizing strength of covalent bonds, please check Section 3.11.9 for
introduction first. In this section I will illustrate its calculation.
As mentioned in Section 3.11.9, IBSI can be calculated in terms of either IGM or IGMH
formalism, they will be referred to as IBSIIGM and IBSIIGMH, respectively. To calculate the former,
the input file can only contain atom coordinate, while for latter, the input file must carry
wavefunction information. Here we calculate these two indices for acetylene, the .wfn file was
generated at B3LYP/6-31G** level, its geometry was optimized at the same level.
examples\C2H2.wfn
10 // Intrinsic bond strength index (IBSI)
1 // Start calculation. Since the current input file contains wavefunction information, by
default the IBSI to be calculated is IBSIIGMH
2 // Use high quality integration grid (using "ultrafine grid" will result in marginally better
numerical accuracy, while the cost will be correspondingly increased)
The result is
1(C ) 2(H ) Dist: 1.0657 Int(dg_pair): 0.26199 IBSI: 0.46062
1(C ) 3(C ) Dist: 1.2054 Int(dg_pair): 1.16889 IBSI: 1.60654
2(H ) 3(C ) Dist: 2.2711 Int(dg_pair): 0.09015 IBSI: 0.03490
2(H ) 4(H ) Dist: 3.3368 Int(dg_pair): 0.01304 IBSI: 0.00234
The "Dist" corresponds to distance between the two atoms, the Int(dg_pair) stands for the
∫ 𝛿𝑔pair d𝐫 term in the IBSI expression, the "IBSI" is the IBSIIGMH value.
489
Next, we calculate IBSIIGM. Choose option "2 Toggle type of IGM" to change the form of the
IGM to be calculated as IBSIIGM, then choose option 1 again and select "high quality" to carry out
the calculation, the result is
1(C ) 3(C ) Dist: 1.2054 Int(dg_pair): 1.11265 IBSI: 1.86652
2(H ) 3(C ) Dist: 2.2711 Int(dg_pair): 0.18091 IBSI: 0.08548
2(H ) 4(H ) Dist: 3.3368 Int(dg_pair): 0.01960 IBSI: 0.00429
Similarly, you can calculate ∫ 𝛿𝑔pair d𝐫 and IBSI for ethane and ethene, their .wfn files
generated at the same level as the C2H2.wfn have been provided as ethane.wfn and ethene.wfn in
"examples" folder. The calculated data of the C-C bond in the three systems are plotted with respect
to their bond dissociation energies (BDEs) in below map, in which the gIGM and gIGMH correspond
to the ∫ 𝛿𝑔pair d𝐫 calculated in terms of IGM and IGMH, respectively.
2.0
1.8 gIGM
IBSIIGM
1.6 gIGMH
IBSIIGMH
1.4
1.2
1.0
0.8
0.6
0.4
300 400 500 600 700 800 900 1000

Bond dissociation energy (kJ/mol)
As can be seen, the gIGM does not correlate well with bonding strengh, which is directly reflected
by BDE. The linear relationship between IBSIIGM and BDE is perfect, however the slope is large.
The gIGMH seems to have good positive correlation with BDE, it even outperforms the more
sophisticated IBSIIGMH, but currently it is unclear whether this relatively satisfactory relationship
can be transplanted to other types of chemical bonds or weak interactions.
Note that if the input file only contains geometry information, such as .pdb and .xyz, the default
IBSI to be calculated is IBSIIGM, and this is the only available choice.
Strictly speaking, the reference value of IBSI, namely the denominator in the IBSI expression,
should be calculated at the same level as current system, however in the present example we directly
used the built-in data. You can change it if you hope to obtain more rigorous result, the reference
value can also be evaluated using the present function, see Section 3.11.9 for detail.
490
4.9.11 Example of using AV1245 and AVmin indices to study

aromaticity
Please read Section 3.11.10 first to gain basic knowledge about the AV1245 and AVmin indices.
These indices can not only study aromaticity for small rings like multi-center bond order (MBCO),
but can also study arbitrarily large rings. In Section 4.9.11.1, AV1245 and AVmin will be employed
to distinguish aromaticity between the two kinds of six-membered rings in phenanthrene, then in
Section 4.9.11.2, porphyrin will be taken as instance to show the ability of these indices in
quantification of aromaticity of large rings.
4.9.11.1 Using AV1245 and AVmin to study local aromaticity of phenanthrene
In Section 4.9.2, MCBO has been used to study the difference in local aromaticity for the two
kinds of rings in phenanthrene, whose geometry and atomic numbering are shown below. In this
section, we will use AV1245 and AVmin to study it again.

11 // Calculate AV1245
1,2,3,4,5,6 // Calculate AV1245 for the boundary six-membered ring. Note that the order of
inputting should be in line with connectivity
The result is
4-center electron sharing index of 1 2 4 5: 0.01304820
AV1245 times 1000 for the selected atoms is 11.17127674
AVmin times 1000 for the selected atoms is 8.015375 ( 3 4 6 1)
Namely 1000*AV1245 and 1000*AVmin are 11.171 and 8.015, respectively. The AVmin value
corresponds to 4c-ESI of 3-4-6-1.
Next, we input 3,4,8,9,10,7 to calculate 1000*AV1245 and 1000*AVmin for the central ring,
the result are 5.012 and 3.996, respectively. Clearly, the boundary ring has stronger aromaticity than
the central ring since it has larger AV1245 and AVmin, this conclusion is in line with the MCBO
analysis in Section 4.9.2. In the original paper of AV1245, it is argued that AV1245 behaves as an
approximation of MCBO.
491
Calculating AV1245 and AVmin in natural atomic orbital (NAO) basis

In Multiwfn, AV1245 and AVmin can also be calculated based on natural atomic orbitals
(NAOs), as mentioned in Section 3.11.10. The main advantage of this way is that reasonable result
can still be obtained even diffuse functions are presented (while if you calculate AV1245 and AVmin
as what we have done earlier when diffuse functions are employed, the result will be quite
misleading)
Here we calculate AV1245 and AVmin in NAO basis for the phenanthrene, in this case NBO
output information with DMNAO keyword should be employed as input. The Gaussian input file
involved in this example is exampes\phenanthrene_DMNAO.gjf, the corresponding output file is
examples\phenanthrene_DMNAO.out. As you can see from the .gjf file, the NBO module embedded
in Gaussian is invoked and DMNAO keyword is passed into NBO module.
examples\phenanthrene_DMNAO.out
1,2,3,4,5,6 // Calculate AV1245 for the boundary ring
The output is
As can be seen, the result are almost identical to those we obtained earlier (11.171 and 8.015), this
is expected since currently diffuse functions are not employed.
4.9.11.2 Using AV1245 and AVmin to measure global aromaticity of porphyrin

In this example we use AV1245 and AVmin to quantify aromaticity corresponding to different
delocalization paths of porphyrin. The .fch file generated at B3LYP/6-31G* level can be
downloaded at http://sobereva.com/multiwfn/extrafiles/porphyrin.rar. The structure is shown below.
There are several possible global delocalization paths around the porphyrin, now we calculate
one of them. Although you can manually input atomic indices by tracing atom connectivity as
exhibited in above map, this process is very laborious, especially when the ring under study is large.
It is much better to use GaussView to visually select the atoms in the ring of interest and directly
extract their indices. To do so, we use GaussView to open the porphyrin.fch, then click the brush
492
icon , then hold down the left mouse button and let the cursor pass every atom in the ring,
then the atoms will be highlighted as yellow, as shown on the left side of the figure below. After
that, choose "Tools" - "Atom Selection", copy out the atom indices from the text box (see below) to
clipboard, namely 1,3-4,6-8,10-14,16-19,21-22,24.
Now boot up Multiwfn and input

porphyrin.fch
d // After entering this mode, you can input the atom indices in arbitrary order, because in
this case the actual atom sequence will be automatically guessed based on recognized connectivity
1,3-4,6-8,10-14,16-19,21-22,24 // Indices of the atoms in the selected ring
Now you can see below information:
Number of selected atoms: 18
Atomic sequence:
1 3 6 8 12 14 17 18 19 22 21 16
13 10 11 7 4 24
...[ignored]
As you can see, the atomic sequence has been properly recognized, it is fully in line with the actual
connectivity in the ring, therefore the result, 2.76, should be meaningful. The AVmin corresponds
to N3-C6-C12-C14, implying that this local region is the bottleneck of electron delocalization over
the whole path.
Similarly, we calculate AV1245 for other rings by inputting below commands
d
1-2,4-8,10-14,16-20,22-24
d
1,3-4,6,8-10,12-13,15-16,18-19,21-22,24
d
1-2,4-6,8-10,12-13,15-16,18-20,22-24
493
For all calculated rings, the selected atoms as well as results are summarized below, the blue
and green texts correspond to 1000*AV1245 and 1000*AVmin, respectively.
It unambiguous that the path passing through nitrogen of pyrroles but bypassing N-H group is the
most favorable delocalization channel, since its 1000*AV1245 and 1000*AVmin values (2.76 and
1.94) are both larger than other pathways. It is worth to mention that for the two paths shown at
bottom of the above map, although their AV1245 are unequal, their AVmin are exactly identical.
This observation suggests that despite average extents of electron delocalization on the two selected
paths are notably different, the bottlenecks are the same.
This system has also been studied in Section 4.4.9 via LOL-, the resulting graph is given
below, from which it can be clearly seen that the degree of electron delocalization along different
paths is significantly different, the top priority delocalized path is vividly revealed by red or orange
color. Obviously the most favorable delocalization path unveiled by AV1245 and AVmin is in good
agreement with that revealed by LOL-.
494
It is worth to note that the precondition of using the convenient "d" mode in inputting atomic
indices is that there is no atom in the ring connecting more than two other atoms in the ring. For
example, for the naphthalene shown below, you should not enter "d" mode and input 1-10 to
calculate the AV1245 and AVmin to study the global aromaticity around the whole system, because
the atoms 9 and 10 simultaneously connect three atoms, in this case the correct atomic sequence in
the ring cannot be automatically determined by Multiwfn.
4.10 Plot density-of-states (DOS) maps
In this section, I will illustrate how to use Multiwfn easily plot various kinds of density-of-
states (DOS) maps. The relevant theories and usage introductions of the DOS module can be found
in Section 3.12.
More in-depth discussions and DOS plotting examples can be found in my blog article "Plotting density-of-
states maps by Multiwfn to study electronic structure" (in Chinese, http://sobereva.com/482)
4.10.1 Plot total, partial and overlap DOS for N-phenylpyrrole
In this example, we will plot total, partial and overlap density-of-states (TDOS, PDOS and
OPDOS) for N-phenylpyrrole, whose structure is shown below. This example consists of six parts.
Please read Section 3.12.1 first if you are not familiar with DOS.
Because basis function information is required in plotting PDOS and OPDOS, we use .fch as
495
input file in these examples, using .mwfn, .molden or .gms file is also OK, but .wfn/.wfx cannot not
employed, since they do not contain information of basis function and virtual MOs. If you only need
to obtain TDOS, you can also simply use a plain text file recording MO energy levels or the Gaussian
output file with pop=full keyword as input file, see Section 3.12.1 for file format.
It is noteworthy that if you intend to plot PDOS and OPDOS based on the default Mulliken orbital composition
method, employing diffuse functions must be avoided, because they severely hurt the reliability of the orbital
compositions evaluated by Mulliken or SCPA method. However you can safely use diffuse functions if you let
Multiwfn calculate orbital composition via Hirshfeld or Becke method, but OPDOS cannot be plotted in this case.
The wavefunction of present system was generated at B3LYP/6-31G* level.
Hint: It is strongly suggested using .pdf or .svg format instead of the default .png format for saving DOS maps
in post-processing menu of DOS module, since in this case the DOS map can be scaled losslessly and the curves
look very smooth. The default graphic format can be set via "graphformat" in settings.ini, you can also change file
format via option "-1 Set format of saved image file" in the post-processing menu of DOS module.
Part 1: Plot total DOS (TDOS)

examples\N-phenylpyrrole.fch
10 // Plot various kind of DOS maps
0 // Plot map
Since currently no fragment is defined, only TDOS is plotted. The TDOS map pops up
immediately, see below
TDOS
9.00
8.00
7.00
Density-of-states
6.00
5.00
4.00
3.00
2.00
1.00
0.00
-0.80 -0.70 -0.60 -0.50 -0.40 -0.30 -0.20 -0.10 0.00 0.10 0.20
Energy (a.u.)
In the TDOS map, each discrete vertical line correspond to a molecular orbital (MO), the
dashed line highlights the position of HOMO. The curve is the TDOS simulated based on the
distribution of MO energy levels. In the negative part, the region around -0.40 a.u. has obviously
larger state density than other regions.
Clicking mouse right button on the graph to close it, and select option 0 to return to last menu.
As illustrated below, we can also change the energy unit and energy range, and the line height
can be used to indicate orbital degeneracy. Input below commands:
8 // Switch the unit from the default a.u. to eV
2 // Set energy range
-30,5,5 // Set lower and upper limits to -30 eV to 5 eV, the spacing between labels is 5 eV
9 // Using line height to show orbital degeneracy
496
0.05 // If energy difference between two orbitals is less than 0.05 eV, they will be regarded
as degenerate
0 // Plot TDOS map again
Now we have below map. The line height indicates degeneracy and corresponds to the axis at
right side. As can be seen, some orbitals have degeneracy of two.
0.360 10.0
TDOS
9.0
0.320
8.0
0.280
7.0
0.240
Density-of-states
6.0
Degeneracy
0.200
5.0
0.160
4.0
0.120
3.0
0.080 2.0
0.040 1.0
0.000 0.0
-30.00 -25.00 -20.00 -15.00 -10.00 -5.00 0.00 5.00
Energy (eV)
Part 2: Plot PDOS and OPDOS for fragments

Next, we will define the heavy atoms of pyrrole moiety as fragment 1 and that of the phenyl
moiety as fragment 2 to check their PDOS and OPDOS. In addition, we will define all hydrogens
as fragment 3.
-1 // Enter the interface for defining fragments. You can define up to 10 fragments. PDOS
will be plotted for all of them but OPDOS will only be drawn between fragment 1 and 2
1 // Define fragment 1
a 1-5 // Add carbons and nitrogen of pyrrole moiety (atoms 1~5) to the fragment
q // Save fragment 1
a 10-13,15,17 // Add phenyl moiety (atoms 10~13, 15 and 17) to the fragment
a 6-9,14,16,18-20 // Add all hydrogens to the fragment
0 // Return to last menu
2 // Set X-axis
-1.1,-0.1,0.1 // Set the range of X-axis to -1.2 ~ -0.1 a.u., so that all valence MOs can be
shown in the graph. The step between labels is set to 0.1 a.u..
0 // Draw TDOS+PDOS+OPDOS
497
The current graph is not very ideal. Close the graph, you can see many options used to
customize the graph, such as setting curve colors, setting legend texts. Try to play with them and if
you are confused you can consult Section 3.12.3. Here we select option 4 and input -2,9,1 to set
lower, upper limits and stepsize of left Y-axis (corresponding to TDOS and PDOS) to -2.0, 9.0 and
1.0, respectively. Select 14 and input the scale factor 0.2, then range of the right Y-axis
(corresponding to OPDOS) will be set to -0.4, 1.8 (because -2.0*0.2=-0.4 and 9.0*0.2=1.8).
Shrinking the range of the Y-axis at right side is equivalent to enhancing the amplitude of OPDOS
curve, which makes the variation of OPDOS in the map clearer. Then select option 1 to replot the
DOS map, you will see
9.00 1.80
TDOS
8.00 PDOS frag.1 1.58
PDOS frag.2
7.00 PDOS frag.3
OPDOS 1.36
6.00
1.14
5.00
Density-of-states
0.92
4.00
OPDOS
0.70
3.00
0.48
2.00
0.26
1.00
0.00 0.04
-1.00 -0.18
-2.00 -0.40
-1.10 -1.00 -0.90 -0.80 -0.70 -0.60 -0.50 -0.40 -0.30 -0.20 -0.10
Energy (a.u.)
The axis at left side corresponds to TDOS and PDOS, while the one at right side corresponds
to OPDOS. The red, blue and magenta curves and discrete lines represent PDOS of fragment 1, 2
and 3, respectively. It can be seen that in most valence MOs, the fragments 1 and 2 have comparable
amount of contribution. The fragment 3 (hydrogens) mainly contributes to the MOs between -0.60
~ -0.35 a.u. Green curve is the OPDOS between fragments 1 and 2, its positive part implies that the
MOs in corresponding energy range show bonding character between the two fragments (e.g. the
one at -0.8 a.u., which corresponds to MO14); there are also regions where OPDOS is negative, e.g.
the HOMO-1 (-0.213 a.u.) behave as antibonding orbitals between the two fragments.
Part 3: Plot PDOS of a specific atom orbital

Current molecule is in YZ plane, as an example, let us check the PDOS of px atomic orbital of
the nitrogen atom, it represents  electron on this site. Select 0 to return to last menu and then input
-1 // Define fragments
-2 // Fragment 2 is not needed, so we input corresponding negative value to unset it
-3 // Also unset fragment 3
1 // Redefine fragment 1
clean // Clean existing content of the fragment
all // Print out information of all basis functions
The information corresponding to nitrogen atom is extracted and shown below
Basis: 61 Shell: 25 Center: 5(N ) Type: S
498
Basis: 63 Shell: 27 Center: 5(N ) Type: X
Basis: 64 Shell: 27 Center: 5(N ) Type: Y
Basis: 65 Shell: 27 Center: 5(N ) Type: Z
Basis: 67 Shell: 29 Center: 5(N ) Type: X
Basis: 68 Shell: 29 Center: 5(N ) Type: Y
Basis: 69 Shell: 29 Center: 5(N ) Type: Z
Basis: 70 Shell: 30 Center: 5(N ) Type: XX
Basis: 71 Shell: 30 Center: 5(N ) Type: YY
Basis: 72 Shell: 30 Center: 5(N ) Type: ZZ
Basis: 73 Shell: 30 Center: 5(N ) Type: XY
Basis: 74 Shell: 30 Center: 5(N ) Type: XZ
Basis: 75 Shell: 30 Center: 5(N ) Type: YZ
Current system is calculated under 6-31G* basis set, according to the basis set definition, each
valence atomic orbital is represented by two basis functions of corresponding type. Therefore, what
we should do is to put basis functions 63 and 67 into the fragment, they collectively represent the
px orbital of nitrogen (For other kinds of basis set, you can consult Section 4.7.6 on how to identify
correspondence between basis functions and atomic orbitals). Input below commands
b 63,67 // Then you can input command all again, the basis functions added to present
fragment are marked by asterisks
q // Save fragment
0 // Return
0 // Plot TDOS and PDOS
Please analyze the resulting graph by yourself.
9.00
TDOS
8.00 PDOS frag.1
7.00
6.00
5.00
Density-of-states
4.00
3.00
2.00
1.00
0.00
-1.00
-2.00
-1.10 -1.00 -0.90 -0.80 -0.70 -0.60 -0.50 -0.40 -0.30 -0.20 -0.10
Energy (a.u.)
Part 4: Plot PDOS of all  molecular orbitals

This molecule is in YZ plane, assume that we only intend to study PDOS/OPDOS of π orbitals
of pyrrole and phenyl moieties and want to get rid of effect of all other MOs, although in the
499
fragment definition interface we can choose each PX basis function in turn, because there are too
many atoms, this process will take you a lot of time and thus is very boring. A much better way is
to use the condition selection command. Select 0 to return to last menu and then input
1 // Redefine fragment 1
clean // Clean existing content of the fragment
cond // Use conditions to select basis functions. You will be prompted to input three
conditions, the basis functions simultaneously satisfying the three conditions will be added to
current fragment
1-5 // The first condition is that the basis functions must belong to the heavy atoms in pyrrole
moiety (atoms 1~5)
a // The second condition is the index range of basis functions. Inputting a means any basis
function is OK (in other words, do not employ this condition)
X // The third condition is that the type of basis function should be PX
cond // Use conditions to select basis functions
10-13,15,17 // Atom index of the carbons in the phenyl moiety
a // No restriction on the index of basis functions
X // Basis function must be PX type
9.00 1.80
TDOS
8.00 PDOS frag.1 1.58
PDOS frag.2
7.00 OPDOS
1.36
6.00
1.14
5.00
Density-of-states
0.92
4.00
OPDOS
0.70
3.00
0.48
2.00
0.26
1.00
0.00 0.04
-1.00 -0.18
-2.00 -0.40
-1.10 -1.00 -0.90 -0.80 -0.70 -0.60 -0.50 -0.40 -0.30 -0.20 -0.10
Energy (a.u.)
This time the PDOS curves only cover high-energy regions, implying that most π MOs in
present system have higher energy than  MOs. Please use main function 0 of Multiwfn to visualize
corresponding MO isosurfaces.
Part 5: Plot PDOS for s, p, d atomic orbitals individually

Next, we plot PDOS for s, p, d atomic orbitals individually. Reboot Multiwfn and then input
500
l s // Add basis functions with angular moment of s to the fragment
q // Save fragment
l p // Add basis functions with angular moment of p to the fragment
q // Save fragment
l d // Add basis functions with angular moment of d to the fragment
q // Save fragment
Then close the graph and input
9 // Disable showing OPDOS curves
10 // Disable showing OPDOS lines
4 // Set range of Y axis
0,10,1 // Lower and upper limits are set to 0 and 10 with stepsize of 1.0
16 // Set legends
1 // Set legend of PDOS corresponding to fragment 1
s
p
d
0 // Exit the interface for setting legends
1 // Replot the map
Now you can see below map
501
10.00
TDOS
9.00 s
p
d
8.00
7.00
Density-of-states
6.00
5.00
4.00
3.00
2.00
1.00
0.00
-0.80 -0.70 -0.60 -0.50 -0.40 -0.30 -0.20 -0.10 0.00 0.10 0.20
Energy (a.u.)
From this map, it is clear that occupied frontier MOs are solely contributed by p orbitals.
If you want to plot PDOS for certain angular moment of orbitals for specifc atoms, it is also
very easy. For example, by inputting below commands in the fragment definition interface, a
fragment corresponding to all p orbitals of all the four carbons in the pyrrole moiety could be defined.
cond // Use conditions to select basis functions
1-4 // Atoms 1~4
a // No condition for basis function index
P // Basis function of P angular moment
Part 6: Plot PDOS based on orbital compositions derived by Hirshfeld method

Plotting PDOS requires orbital compositions. In above examples, the compositions were
evaluated using the default Mulliken method. This method is fast very, however, it is not quite robust
(especially for unoccupied orbitals), and the result is completely useless when diffuse functions are
employed. Here I also illustrate how to plot PDOS based on the orbital compositions derived by the
Hirshfeld method, which is more robust and fully compatible with diffuse functions. The
disadvantage is that Hirshfeld method is more expensive, and it can only evaluate contributions
from atoms, namely the fragments can only be defined as a set of atoms.
Here we repeat the example in "Part 2" but using compositions obtained by Hirshfeld method.
Boot up Multiwfn and input below commands:
10 // Plotting DOS
7 // Change the method for calculating orbital compositions
3 // Hirshfeld method. Then Multiwfn calculate orbitals compositions for all atoms in all
orbitals, for large system you need to wait for a while
1-5 // Set carbons and nitrogen of pyrrole moiety (atoms 1~5) as the fragment
10-13,15,17 // Set phenyl moiety (atoms 10~13, 15 and 17) as the fragment
502
6-9,14,16,18-20 // Set all hydrogens as the fragment
2 // Set X-axis
-1.1,-0.1,0.1
0 // Draw TDOS+PDOS
The resulting graph is almost identical to that plotted based on the compositions derived by the
default Mulliken method (however, the difference is often evident for the energy range composing
of unoccupied MOs, clearly the PDOS based on Hirshfeld is more reliable). Note that OPDOS
cannot be plotted when Hirshfeld method is employed to calculate orbital compositions.
4.10.2 Plot local DOS for 1,3-butadiene
If you do not know what is local DOS (LDOS), please check Section 3.12.4 first. Briefly
speaking, TDOS represents DOS curve for the whole system, PDOS describes DOS curve for an
atom (or fragment), while LDOS exhibits DOS curve for a point (i.e. space-resolved). In addition,
we can plot LDOS for a set of point constituting a line as color-filled map, the X-axis corresponds
to energy while the Y-axis shows position in the line. LDOS is useful when interpreting the data
from scanning tunneling microscope (STM), you can find relating experimental data in e.g. J. Phys.
Chem. Lett., 5, 3701 (2014).
In the current example, we plot LDOS for butadiene at selected points. First, we plot LDOS
for the point over 1.5 Bohr of terminal carbon of butadiene. Boot up Multiwfn and input following
commands:
0 // From output in command-line we can find the expected point should be 1.137 3.308 1.5
(1.5 Bohr above C1)
10 // DOS plotting module
10 // Draw local DOS for a point
1.137,3.308,1.5
Then you will see (you can compare it with TDOS map)
0.038
0.034
0.030
0.026
Density-of-states
0.023
0.019
0.015
0.011
0.008
0.004
0.000
-0.80 -0.70 -0.60 -0.50 -0.40 -0.30 -0.20 -0.10 0.00 0.10 0.20
Energy (a.u.)
503
Close the graph and select 0 to return to the last menu.
Next, we plot color-filled map along the line connecting the two points above 1.5 Bohr of the
two terminal carbons (C1 and C8), input below commands
11 // Draw local DOS along a line
1.137,3.308,1.5
-1.137,-3.308,1.5
200 // Evenly taking 200 points along line
Then close the graph that pops up and input
4 // Modify the ratio between Y and X axes
0.5 // The length of Y-axis will be half of X-axis
1 // Replot
Then you can see
The color in this graph represents density of states at different 3D spatial positions (Y-axis) and
different energies (X-axis). The pink arrows highlight the gaps at three different spatial positions.
If you still feel difficult in understanding meaning of the map, please check below figure, in
which some important information are explicitly labelled.
It is easy to understand, the lowermost horizontal line of the above graph (viz. the green dash
504
line at Y=0) corresponds to the LDOS curve map at the position of 1.5 Bohr above C1, which has
been plotted by us earlier.
4.10.3 Plot DOS map for unrestricted open-shell system: Na3O@Si12C12
In Section 4.10.1, we have plotted a closed-shell system, while in this section, I will illustrate
how to plot DOS for a typical open-shell system Na3O@Si12C12, which was studied in my work J.
Comput. Chem., 38, 1574 (2017) and is doublet. For open-shell cases calculated in unrestricted
formalism, there are two kinds of spins, they should be simultaneously taken into account, The .fchk
file can be downloaded here: http://sobereva.com/multiwfn/extrafiles/Na3O-Si12C12.rar, which
corresponds to UM06-2X/6-311G* wavefunction at optimized geometry.
First, we plot TDOS+PDOS map for alpha spin, the PDOS will correspond to the Na3O. Boot
up Multiwfn and input
Na3O-Si12C12.fchk
a 1,4,27,28 // These four atoms correspond to the Na3O moiety
q // Save fragment
0 // Return
0 // Plot TDOS+PDOS
You will see below graph. By default, for unrestricted wavefunction, only alpha MOs are taken
into account, therefore the below map is DOS map of alpha spin.
TDOS
27.81 PDOS frag.1
24.72
21.63
Density-of-states
18.54
15.45
12.36
9.27
6.18
3.09
0.00
-0.80 -0.70 -0.60 -0.50 -0.40 -0.30 -0.20 -0.10 0.00 0.10 0.20
Energy (a.u.)
Choose option 0 to return to last menu. If you want to plot the DOS map for beta spin, you
should select "6 Choose orbital spin" and then choose "2 Beta spin". If you do not want to distinguish
spin but want to take all MOs into account, you should choose "3 Both spins". Please choose beta
spin and replot the map via option 0 again, you will find the map is similar to that of alpha spin,
showing that in this system the spin polarization is not quite evident.
505
In order to make comparison between alpha and beta DOS maps intuitive, we can try to make
a figure, the upper part and lower part correspond to alpha spin and beta spin, respectively. Such a
map cannot be directly produced by Multiwfn, however it can be easily prepared via Multiwfn in
combination with third-part visualization software such as Origin, as shown below. The Origin
version I am using is 9.0
We first use Multiwfn to plot the alpha TDOS+PDOS map in aforementioned way, in the post-
processing menu, choose "3 Export curve and line data to plain text file in current folder". Rename
the DOS_curve.txt to alpha.txt. Return to DOS plotting interface, change to beta spin, plot the map
and then export the data set again, rename the DOS_curve.txt to beta.txt. The DOS_line.txt can be
deleted because we will not utilize it.
Boot up Origin, drag both alpha.txt and beta.txt into it to import them. Currently, in the
workbook corresponding to beta spin, B and C columns correspond to TDOS and PDOS curve data,
respectively. We choose "Set Column Values" option for column D (which is empty curremty), set
the content D to -Col(B); similarly we set column of E to -Col(C).
Next, we select proper option to plot line map. In the worksheet corresponding to alpha spin,
we add column A as X data, add columns B and C as two sets of Y data. In the worksheet
corresponding to beta spin, we add column A as X data, while add columns D and E as Y data. After
some adjustments, you will obtain below graph, which nicely exhibits DOS and PDOS for alpha
and beta spins, respectively.
30
 TDOS
 PDOS (Na3O)  HOMO
20
 TDOS
 PDOS (Na3O)
10
Density of states
-10
-20
 HOMO
-30
-0.8 -0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2
Energy (Hartree)
Note that in order to plot a horizontal line corresponding to DOS=0 and two vertical lines
highlighting HOMO level of alpha and beta spins, I also created the third worksheet and properly
filled the content. The value of the alpha and beta HOMOs can be directly found from the prompts
when you selecting option 0 in DOS module to draw the map, namely
Note: The vertical dash line corresponds to HOMO level at -0.171 a.u.
and
506
Note: The vertical dash line corresponds to HOMO level at -0.221 a.u.
The aforementioned alpha.txt, beta.txt as well as the Origin .opj file of the map have been
provided in "examples\DOS\" folder.
4.10.4 Plot photoelectron spectrum (PES) for Cr3Si12− cluster
In Section 3.12.4, the theory of PES and the interface for plotting PES have been introduced,
please read it if you have not. In present section, Cr3Si12- will be taken as example to illustrate how
to very easily plot PES, we will plot PES employing generalized Koopmans' theorem. This system
has been studied in J. Phys. Chem. A, 122, 9886 (2018) under PBE/6-311+G* level, it is worth to
note that the calculated first VIP of this system 2.56 eV.
Using the optimized structure of this system provided in supplemental material in the JPCA
paper, I carried out a single point task using the same level as the paper by Gaussian 16, the resulting
Cr3Si12-.fchk file can be downloaded here: http://sobereva.com/multiwfn/extrafiles/Cr3Si12-.rar.
Cr3Si12-.fchk
10 // DOS module
12 // Interface for plotting PES. You will find HOMO level has been shown on the screen,
namely -0.77 eV, which is the highest one among alpha HOMO and beta HOMO
3 // Set shift value to meet generalized Koopmans' theorem
1.79 // Should be 1st VIP + E(HOMO). For present case the value is -0.77+2.56=1.79 eV
4 // Set X-axis
1,4.5,0.5 // The energy span is 1.0~4.5 eV, with label step of 0.5 eV
9 // Set width of curve
10 // Make the curve thicker than default
1 // Plot the spectrum
The resulting spectrum is shown below. Note that the absolute value of Y-axis in fact is
meaningless, you can choose option "13 Toggle showing labels and ticks on Y-axis" once to switch
its status to "No" to remove the labels and ticks on the Y-axis.
507
The experimental spectrum provided in the JPCA paper is shown below
Clearly, our simulated spectrum is in very good agreement with the experimental one, showing
that our plotting procedure and methodology are completely reasonable.
4.10.5 Plot MO-PDOS map to reveal PDOS contributed by different

MOs for cyclo[18]carbon
Note: The first published paper that came up with the idea of MO-PDOS map is my work: Carbon, 165 461
(2020), see Fig. 2. Please cite this paper if MO-PDOS map is employed in your work.
This section illustrate how to plot and analyze MO-PDOS map. The so-called "MO-PDOS"
refers to a special kind of PDOS, which is used to reveal DOS contributed by different sets of MOs
(rather than by atoms or basis functions as the PDOS in common sense), the PDOS curves and
discrete lines corresponding to different sets of MOs are shown using different colors. If requsted,
the height of discrete lines can be used to reflect degeneracy of orbital levels.
We will plot MO-PDOS for cyclo[18]carbon, its structure optimized at B97XD/def2-TZVP
level is shown below, it is an exactly planar system with point group of D9h. The .fchk file of this
system corresponding to B97XD/def2-TZVP wavefunction at minimum point structure can be
download here: http://sobereva.com/multiwfn/extrafiles/C18.zip.
Occupied valence MOs of this system consist of three types, you can identify their indices by
viewing orbitals via main function 0:
(1)  MOs: 19-36
(2) in-plane  MOs: 37,39,40,45,46,49,50,53,54
508
(3) out-plane  MOs: 38,41,42,43,44,47,48,51,52

In the MO-PDOS map to be plotted, we will use different colors to respectively reveal the position
of energy levels of these orbitals as well as their contributions to total DOS.

C18.fchk
10 // Plot DOS
-2 // Enter the interface for defining MO fragments of MO-PDOS
1 // Define 1st fragment
19-36 //  MOs
2 // Define 2nd fragment
37,39,40,45,46,49,50,53,54 // in-plane  MOs
3 // Define 3rd fragment
38,41,42,43,44,47,48,51,52 // out-plane  MOs
0 // Return
0 // Plot DOS map
We immediately see below map
18.10
TDOS
PDOS frag.1
16.29
PDOS frag.2
PDOS frag.3
14.48
12.67
Density-of-states
10.86
9.05
7.24
5.43
3.62
1.81
0.00
-0.80 -0.70 -0.60 -0.50 -0.40 -0.30 -0.20 -0.10 0.00 0.10 0.20
Energy (a.u.)
In this figure, the red, blue and purple discrete lines indicate position of  MOs, in-plane 
MOs and out-plane  MOs, respectively. The  ones have evidently lower energies than the  ones,
while the two kinds of  MOs have similar energy distribution. From the broadened curves, we can
identify the respective contributions due to the three types of MOs, the sum of heights of the colored
curves just equals to the black curve, which portrays the total DOS. Since the defined fragments are
only composed of occupied MOs, the unoccupied region of the map is completely identical to usual
TDOS map.
We can further improve the setting of the MO-PDOS map. After closing the graph, we input
0 // Return to last menu from the post-processing menu
8 // Switch the energy unit to eV
2 // Set energy range and step
-28,1,3 // Set lower and upper limits of plotting region to -28~1 eV with step of 3 eV, so that
509
all occupied valence MOs and a few lowest-lying virtual MOs could be displayed in the map
9 // Enabling using height of discrete lines to indicate orbital degeneracy
[Press ENTER button directly] // Use default threshold to determine degeneracy
0 // Plot DOS map
Close the map, then in the post-processing menu we input
16 // Set the texts in the legends
1 // Set legend for PDOS 1
sigma MOs
in-plane pi MOs
out-plane pi MOs
0 // Return to post-processing menu
6 // Disable showing TDOS discrete lines
1 // Redraw the graph
You should see below map, which is quite satisfactory
10.0
TDOS
0.450 sigma MOs 9.0
in-plane pi MOs
out-plane pi MOs
0.400 8.0
0.350 7.0
Density-of-states
0.300 6.0
Degeneracy
0.250 5.0
0.200 4.0
0.150 3.0
0.100 2.0
0.050 1.0
0.000 0.0
-28.00 -25.00 -22.00 -19.00 -16.00 -13.00 -10.00 -7.00 -4.00 -1.00
Energy (eV)
From the height of the colored discrete lines, you can clearly find most occupied valence
orbitals are doubly degenerate.
Hint: Save and load status

If you want to save current status (plotting settings, fragment definition and orbital information)
of the above map to a file so that you can quickly recover the map at the next time, now you can
input 0 to exit post-processing menu, then input s, then input path of the file to save the status.
The status file corresponding to the above map is given as examples\DOS\C18_MO_PDOS.dat,
therefore if you want to directly replot the above map, after booting up Multiwfn and load the
C18.fchk, you should input
l // Load status file
examples\DOS\C18_MO_PDOS.dat
510
0 // Plot the map
4.11 Plot various kinds of spectra
Multiwfn has a very powerful spectrum plotting module. The basic principles, supported input
files and all options of this module have been detailedly introduced in Section 3.13, below sections
will briefly exemplify the usage of this module.
It is worth to note that there is also an article of introducing detailed steps on how to simulate
UV-Vis and ECD spectra using Multiwfn in combination with ORCA, see http://sobereva.com/485.
4.11.1 Plot infrared (IR) spectrum for NH3BF3
This example plots infrared (IR) spectrum for NH3BF3. Multiwfn can read in frequencies and
intensities from output file of Gaussian or ORCA vibration analysis task (“freq” keywords). Boot
examples\spectra\NH3BF3_freq.out // The output file of optimization and vibrational
analysis task of Gaussian at B3LYP/6-31G* level
11 // Plot spectrum
1 // The type of the spectrum is IR
0 // Show the spectrum right now
You will get below graph
6013.86 377.95
5392.43 338.90
4771.00 299.84
4149.56 260.79
IR intensities (km/mol)
3528.13 221.73
2906.70 182.68
2285.27 143.62
1663.83 104.57
1042.40 65.51
420.97 26.46
-200.46 -12.60
4000.0 3600.0 3200.0 2800.0 2400.0 2000.0 1600.0 1200.0 800.0 400.0 0.0
Wavenumber (cm^-1)
The left axis corresponds to curve (broadened data), the right axis corresponds to discrete lines
(original transition data). Plotting parameters such as full width at half maximum (FWHM),
broadening function, unit and range of the axes can be adjusted by corresponding options in the
interface. The graph and X-Y data set of discrete lines/curve can be exported by option 1 and 2,
respectively.
Note that after selecting option 0 to plot the spectrum, Multiwfn prints extrema information in
511
the console window

Extrema on the spectrum curve:
Maximum 1 X: 3578.5262 Value: 585.2897
Maximum 2 X: 3454.4848 Value: 110.4777
Maximum 3 X: 1695.2317 Value: 460.3222
Maximum 4 X: 1359.1197 Value: 1717.7809
Maximum 5 X: 1303.1010 Value: 5467.1462
...[ignored]
From this output you can obtain accurate position and height of absorption peaks. As will be
illustrated in Section 4.11.3, the maxima and minima can even be directly labelled on the spectrum.
It is well known that the frequencies produced under harmonic approximation deviate to
experimental vibrational frequencies systematically. In order to correct this problem, fundamental
frequency scale factor should be applied, this can be done easily in Multiwfn. We close the spectrum
and then choose "14 Set scale factor for vibrational frequencies ", then press ENTER button directly
to choose all vibrational modes, then press ENTER button directly again to employ the scale factor
fitted for B3LYP/6-31G* level, namely 0.9614, which can be found in Table 1 of J. Phys. Chem.,
100, 16502 (1996). After that, if you replot the spectrum, the resulting spectrum will correspond the
scaled one.
Some experimental IR spectra determine transmittance rather than absorption. To mimic this
kind of spectrum, you can select "4 Set left Y-axis" and then input e.g. 6000, 0, 400 to set lower
limit, upper limit and stepsize to corresponding values, respectively, and then input y to
automatically scale the right Y-axis. Since currently lower limit (0) is higher than upper limit (6000),
the Y-axis is inverted.
The procedure of plotting Raman, UV-Vis, electronic/vibrational circular dichroism
(ECD/VCD), ROA spectra is very similar to plotting IR spectrum, you only need to use proper input
file and select corresponding option after entering main function 11. If the quantum chemistry
program you used for spectrum calculation is not the one directly supported by Multiwfn, you can
manually extract data from corresponding output file and then write them into a plain text file
according to the format shown in Section 3.13.2, then the file can be used as input file of Multiwfn
for plotting spectrum.
4.11.2 Plot UV-Vis spectrum and contributions from individual

transitions for acetic acid
The spectrum plotting module of Multiwfn is quite flexible, not only the total spectrum but
also the contribution from individual transitions can be exported. This feature is particularly useful
when you want to identify nature of spectrum. In this section I will show how to realize this analysis,
UV-Vis spectrum of acetic acid is taken as example.
examples\spectra\acetic_acid_TDDFT.out // Calculated at TD-B3LYP/cc-pVDZ level by
Gaussian
11 // Plot spectrum
3 // The type of the spectrum is UV-Vis
512
15 // Output the spectrum including the contributions from certain individual transitions
0.01 // The criterion of selecting transitions is oscillator strength > 0.01
The curve of the UV-Vis spectrum together with the contributions from the transitions whose
absolute value of strength are larger than 0.01 have been outputted to spectrum_curve.txt in current
folder. The first two columns correspond to energies and molar absorption coefficients, the
correspondence between the other columns and transition modes are clearly indicated on screen:
Column# Transition#
3 2 //i.e. transition S0→S2
The discrete line data are outputted to spectrum_line.txt in current folder.

Now you can plot the data in the two files as curves in a single graph by your favourite program
(if you use Origin to plot, you can directly drag these two files into Origin window to import them).
In the two files, the first column should be taken as X-axis data, while the other columns should be
taken as Y-axis data. The spectrum plotted by Origin is shown below, if you are confused about the
procedure, you can consult the acetic_acid_TDDFT.opj provided in "examples\spectra" folder,
which is the .opj file of Origin 8.
10000
0.24
9000 Total
0.22
Molar absorption coefficient (L mol cm )
-1
S0→S2
8000 S0→S3 0.20
-1
S0→S5
7000 0.18
S0→S11
S0→S13 0.16
Oscillator strength
6000
0.14
5000 0.12
4000 0.10
3000 0.08
0.06
2000
0.04
1000 0.02
0 0.00
80 100 120 140 160 180 200
Wavelength (nm)
From the graph the underlying character of the total UV-Vis spectrum (black curve) is now
very clear. Although the S0→S3 transition (146.28nm) does not has very small oscillator strength
(0.036), no absorption peak directly corresponds to this transition, since its absorption curve (blue
curve) has been completely merged into the neighboring large absorption peak due to S0→S5
transition (cyan curve).
As mentioned in last section, after selecting option 0 to plot spectrum, Multiwfn directly prints
extrema of the spectrum curve. In current case the outputted data is
Maximum 1 X: 113.0588 Value: 5680.9864
Maximum 2 X: 122.5085 Value: 8239.4308
513
Maximum 3 X: 138.0728 Value: 4667.7944
Maximum 4 X: 159.7516 Value: 2123.3506
Maximum 5 X: 213.7324 Value: 48.5312
...[ignored]
Based on above outputs, we can calculate the contributions from different transitions to a given
peak. For example, we want to study the composition of the peak at 138.0728 nm. In
spectrum_curve.txt, move to the line corresponding to 138.07278 nm, you can find the total value
is 4667.79439, while the values in column 4 and 5 are 309.47267 and 4273.59757, respectively.
Therefore, the contribution from S0→S3 and S0→S5 can be respectively calculated as
309.47267/4667.79439100% = 6.63% and 4273.59757/4667.79439100% = 91.55%.
4.11.3 Plot electronic circular dichroism (ECD) spectrum for

asparagine
In this example we plot electronic circular dichroism (ECD) spectrum for asparagine. Boot up
Multiwfn and input
examples\spectra\Asn_TDDFT.out // Gaussian TDDFT task at PBE0/6-311G* level, 30
lowest excited states were calculated
11 // Plot spectrum
4 // ECD
2 // Read the rotatory strengths in velocity representation
0 // Show the spectrum
From the resulting spectrum, you will find the labels of X-axis and Y-axis are decimal. In order
to make the graph more beautiful, it is suggested to modify the scale so that label of each tick is
integer. Therefore, we close current graph and input below commands:
3 // Set X-axis
120,280,20 // Lower and upper limits, as well as spacing between ticks of X-axis
4 // Set left Y-axis
-90,100,20 // Lower and upper limits, as well as spacing between ticks of left Y-axis
y // Let program properly adjust right Y-axis to guarantee that zero point of left and right axes
are in the same horizontal line
Then you will see below graph
514
90.000 56.90
Delta molar absorption coefficient (L/mol/cm)
70.000 44.26
50.000 31.61
Rotatory strength (cgs)

30.000 18.97
10.000 6.32
-10.000 -6.32
-30.000 -18.97
-50.000 -31.61
-70.000 -44.26
-90.000 -56.90
120.0 140.0 160.0 180.0 200.0 220.0 240.0 260.0 280.0
Wavelength (nm)
Note that you can use exactly the same way as that illustrated in Section 4.11.2 to decompose
the total ECD spectrum to individual contribution from each transition.
Labelling minima and maxima labels on spectrum

One of the strengths of Multiwfn in plotting spectrum is that maxima, minima or both can be
directly labelled on the spectrum. To label wavelength of both maxima and minima, we input
16 // Enter the interface of setting status of showing labels of spectrum minima and maxima
1 // Change displaying status of labels
3 // Label both maxima and minima on the spectrum
0 // Return
4 // Set left Y-axis
-100,110,20 // Making range of left Y-axis slightly wider, because the labels will be shown
y // Correspondingly scale right Y-axis
0 // Plot spectrum again
515
155.3
100.0 63.2
80.0 50.6
60.0 37.9
40.0 25.3

193.0
234.4
20.0 139.2 12.6
(arb.)
0.0 0.0
144.5
132.4
210.9
-20.0 -12.6
-40.0 -25.3
-60.0 -37.9
-80.0 -50.6
169.1
-100.0 -63.2
120.0 140.0 160.0 180.0 200.0 220.0 240.0 260.0 280.0
Wavelength (nm)
You can also make Multiwfn label Y-axis value at the extrema on the map, now we do this, and
meantime we customize some plotting parameters. Input below commands
16 // Enter the interface of setting status of showing labels of spectrum minima and maxima
6 // Switch the content of the labels to Y-axis value
4 // Do not rotate the labels (this step is optional)
3 // Set decimal digits (this step is optional)
0 // No decimal digits, namely show data as integer
2 // Set label size
50 // Larger text size than default (30)
0 // Return
0 // Plot spectrum again
100.0 63.2
88
80.0 50.6
60.0 37.9
40.0 25.3
20.0 12.6
13 13
(arb.)
6
0.0 0.0
-7 -6
-20.0 -14 -12.6
-40.0 -25.3
-60.0 -37.9
-80.0 -50.6
-77
-100.0 -63.2
120.0 140.0 160.0 180.0 200.0 220.0 240.0 260.0 280.0
Wavelength (nm)
516
Hint: Save and load plotting settings

In order to replot the above map quickly in the future, I suggest saving plotting settings to a
file, namely input below commands:
s // Save plotting settings
Asn_ECD.dat // Save settings to Asn_ECD.dat in current folder
Next time, if you want to recover the above map, you simply need to input
examples\spectra\Asn_TDDFT.out
11 // Plot spectrum
4 // ECD
2 // Read the rotatory strengths in velocity representation
l // Load plotting settings
Asn_ECD.dat // Save settings to Asn_ECD.dat in current folder
Note that the Asn_ECD.dat corresponding to the above map has already been provided in
examples\spectra folder.
4.11.4 Plot conformational weighted UV-Vis and ECD spectra for

plumericin
For a flexible system with many thermally accessible conformation (or configurations), when
plotting its spectrum, it is crucial to take weighting average of various conformations into account,
otherwise the resulting spectrum is impossible to be compared well with experimental spectrum.
Fortunately, weighted spectrum can be very conveniently plotted by Multiwfn, I will show how to
do this in present section. Plotting conformationally weighted UV-Vis and ECD spectra of
plumericin are taken as instances.
Preparation
Before calculating conformationally weighted spectrum, we need to evaluate population of
these conformations, commonly Boltzmann's weight is used. We assume that plumericin has four
accessible conformations, and properly construct their initial geometries, then optimize them and
perform frequency analysis at B3LYP/6-31G* level with zero-point energy scale factor of 0.9806.
Based on optimized geometries, high-accuracy single point energies at M06-2X/def2-TZVP level
were calculated. Finally, we add the Gibbs thermal corrections produced by frequency analysis to
the single point energies to yield relatively accurate Gibbs energy of various conformations. After
that, according to Boltzmann's formula and relative Gibbs energy among these conformations, we
calculate their weights at 298.15 K. Then, using TD-PBE0/TZVP level we calculate the lowest 20
excited states for these conformations. The output file of the TDDFT tasks have been provided in
"examples\spectra\weighted" folder as a.out, b.out, c.out and d.out (the names are arbitrary, you can
also use other file names).
Now we write a plain text file named multiple.txt with below content (the name should not be
changed, otherwise Multiwfn will unable to properly recognize it):
examples\spectra\weighted\a.out 0.6046
examples\spectra\weighted\b.out 0.1950
517
examples\spectra\weighted\c.out 0.1686
examples\spectra\weightedt\d.out 0.0317
The first column records path of Multiwfn input file for various conformations, the second
column is the Boltzmann weights we calculated above. Below the four conformations will be called
as a, b, c and d, respectively.
PS: If you are using Linux system, and there are / symbols or space in the path, do not forget to add double
quotation marks at the two ends of the path, for example:
"/sob/weight/a.out" 0.8
"/sob/weight/b.out" 0.2
Plot conformationally weighted UV-Vis spectrum

examples\spectra\weighted\multiple.txt // The aforementioned file
11 // Plot spectrum
3 // UV-Vis
The resulting graph is shown below
26402.31 0.120
Weighted
1 ( 60.5%)
23674.07 2 ( 19.5%) 0.107
3 ( 16.9%)
4 ( 3.2%)
20945.83 0.095
18217.59 0.083
Oscillator strength
15489.36 0.070
12761.12 0.058
10032.88 0.046
7304.64 0.033
4576.40 0.021
1848.16 0.008
-880.08 -0.004
137.1 154.5 172.0 189.5 206.9 224.4 241.8 259.3 276.8 294.2 311.7
Wavelength (nm)
The thick red curve corresponds to conformationally weighted UV-Vis spectrum, while the
green, blue, purple and black curves correspond to UV-Vis spectrum of conformation a, b, c and d,
respectively. The weight of each conformation is also shown in the legend. From this graph the we
can very conveniently compare the character of weighted spectrum and spectra of individual
conformations.
The black discrete lines on the graph represent all transition data of the four conformations,
their heights have already been scaled by conformational weight. Hence, the thick red curve can be
regarded as broadened by all discrete lines shown on the graph.
Multiwfn provides another mode to plot spectrum of individual conformations. We close above
graph and choose "18 Toggle weighting spectrum of each system" once, then choose option 0 to
view the spectrum again, we will see
518
25486.27 0.120
Weighted
1 ( 60.5%)
22852.69 2 ( 19.5%) 0.107
3 ( 16.9%)
4 ( 3.2%)
20219.11 0.095
17585.53 0.083
Oscillator strength
14951.95 0.070
12318.37 0.058
9684.78 0.046
7051.20 0.033
4417.62 0.021
1784.04 0.008
-849.54 -0.004
137.1 154.5 172.0 189.5 206.9 224.4 241.8 259.3 276.8 294.2 311.7
Wavelength (nm)
The spectrum curve of each conformation shown on this graph has already been multiplied by
corresponding weight. Obviously, what this graph represents is contribution of each conformation
to the conformationally weighted spectrum. In other words, the height of thick red curve is simply
the sum of height of all other curves. It can be seen that conformation a (green line) has major
contribution to the conformationally weighted spectrum, their profiles are rather similar, this is
because a has as high as 60.5% population.
The discrete lines in above graph now have different colors, the color correspond to legend
shown at right-top side. For each conformation, since both discrete lines and curve currently have
identical color, we can say for example, green curve can be directly yielded by broadening the green
discrete lines.
Plot conformationally weighted ECD spectrum

Using the same procedure illustrated in the last section, we plot conformationally weighted
ECD spectrum and ECD spectrum for all the four conformations.
examples\spectra\weighted\multiple.txt
11 // Plot spectrum
4 // Plot ECD
2 // Read rotatory strengths in velocity representation
You will see
519
153.883 33.49
Weighted
1 ( 60.5%)
123.106 2 ( 19.5%) 26.79
3 ( 16.9%)
92.330 4 ( 3.2%) 20.09
61.553 13.40

30.777 6.70
0.000 0.00
-30.777 -6.70
-61.553 -13.40
-92.330 -20.09
-123.106 -26.79
-153.883 -33.49
137.1 154.5 172.0 189.5 206.9 224.4 241.8 259.3 276.8 294.2 311.7
Wavelength (nm)
From this graph you can see conformationally weighted ECD spectrum (thick red curve) as
well as ECD spectrum of individual conformations (other curves).
Then choose "18 Toggle weighting spectrum of each system" option once and plot spectrum
again, you will see
56.295 33.49
Weighted
1 ( 60.5%)
45.036 2 ( 19.5%) 26.79
3 ( 16.9%)
33.777 4 ( 3.2%) 20.09
22.518 13.40
11.259 6.70 Rotatory strength (cgs)
0.000 0.00
-11.259 -6.70
-22.518 -13.40
-33.777 -20.09
-45.036 -26.79
-56.295 -33.49
137.1 154.5 172.0 189.5 206.9 224.4 241.8 259.3 276.8 294.2 311.7
Wavelength (nm)
This graph decomposes the final weighted ECD spectrum to contribution of individual conformation.
Again, since conformation a (green) has very high population and thus dominates the final weighted
curve, most characters of these two curves are similar. However, influence from other conformations
cannot be simply ignored. From the graph it is easy to find that if conformation c (purple) is missing,
then there will not be an evident ECD peak at approximately 180 nm, since only ECD of c at this
wavelength has significant signal.
520
4.11.5 Plot Raman spectrum for 2-methyloxirane based on Raman

intensity
The procedure of plotting Raman spectrum is very similar with plotting IR spectrum, the only
additional step you would better do is to convert the Raman activities directly outputted by Raman
task of quantum chemistry codes to Raman intensities before plotting the spectrum, so that the
resulting spectrum can be comparable with the experimental one. The Raman intensities are
dependent of wavelength of incident light source and ambient temperature, while Raman activities
are not. This point has been emphasized in Section 3.13.1. In this section, I illustrate how to properly
plot Raman spectrum using (2S)-2-methyloxirane as example.
examples\spectra\2-methyloxirane_Raman.out // Output file of Raman task calculated at
B3LYP/6-31G* level by Gaussian09
11 // Plot spectrum
2 // Raman spectrum
14 // Apply frequency scale factor
[Press ENTER button] // Select all frequencies
[Press ENTER button] // Employ the fundamental scale factor 0.9614, which is suitable for
B3LYP/6-31G* level
19 // Convert Raman activities to intensities
15000 // Wavenumber (cm-1) of incident light. This value should be consistent with actual
experimental condition, the value we inputted here is arbitrarily chosen
298.15 // Assume that experimental temperature is 298.15K (You can also press ENTER
button directly, 298.15K will be used as default)
Then input 0 to plot spectrum, you will see below Raman spectrum
122.88 1244.70
110.18 1116.08
97.48 987.46
Raman scattering intensities
84.79 858.84
Relative Raman intensity
72.09 730.23
59.39 601.61
46.69 472.99
34.00 344.37
21.30 215.75
8.60 87.13
-4.10 -41.49
4000.0 3600.0 3200.0 2800.0 2400.0 2000.0 1600.0 1200.0 800.0 400.0 0.0
Wavenumber (cm^-1)
Multiwfn can also plot pre-resonance Raman spectrum. An example output file of pre-
resonance Raman task of Gaussian is examples\spectra\2-methyloxirane_Raman.out, the
521
corresponding input file (.gjf) is also given. This task calculates Raman activity at incident
wavelength of 150nm and 140nm, which are close to S0→S1 and S0→S2 TDDFT excitation
energies at the same calculation level (147.36nm and 138.81nm at TD-B3LYP/6-31G* level,
respectively). You can load this output file into Multiwfn and plot Raman spectrum as usual. The
only difference is that, before entering spectum plotting interface, Multiwfn asks you to choose the
incident frequency for which the Raman activities will be loaded. If you choose 2 or 3, the spectrum
you finally obtained will be pre-resonance Raman at corresponding frequency; if you choose " 1:
0.00000000", namely the static limit case, the resulting spectrum will be exactly identical to the one
we obtained earlier.
4.11.6 Simultanously plot multiple systems
In Multiwfn, it is very easy to plot spectrum for multiple systems simultaneously, these systems
may correspond to different conformations, different configurations, different molecules or different
calculation conditions. In this section two examples are provided.
Comparing spectra yielded by different theoretical methods and basis sets

In "examples\spectra\indigo" folder, you can find Gaussian output file of electronic excited
state task carried out at different levels. In this example, we plot them together so that their results
can be conveniently compared.
What we need to do first is to prepare a file named multiple.txt including path of various
systems with their legends, this file has been provided as examples\spectra\indigo\multiple.txt, its
content is:
examples\spectra\indigo\ZINDO.out ZINDO
examples\spectra\indigo\TD-PBE0.out TD-PBE0/6-31G*
examples\spectra\indigo\TD-PBE0_TZVP.out TD-PBE0/def-TZVP
Note that the legends must not simply be a digital, otherwise it will be interpreted as weight of
corresponding system (see Section 4.11.4). In addition, in Linux system, if the file path contains /
symbol, do not forget to add double quotation marks at the two ends of the path.
Boot up Multiwfn, load multiple.txt and then plot UV-Vis spectrum as usual, you will see below
graph
522
48324.6 0.80
ZINDO
TD-PBE0/6-31G*
43331.0 TD-PBE0/def-TZVP 0.72
(L mol - 1 cm - 1 )
38337.5 0.63
33343.9 0.55
Oscillator strength
28350.4 0.47
Molar absorption coefficient
23356.9 0.39
18363.3 0.30
13369.8 0.22
8376.3 0.14
3382.7 0.06
-1610.8 -0.03
171.0 211.0 251.1 291.1 331.1 371.2 411.2 451.3 491.3 531.3 571.4
Wavelength (nm)
From the graph it is clear that basis set only has small influence on the resulting spectrum,
while the spectrum profile of ZINDO differs from that of TD-PBE0 significantly. The curve of all
systems may be exported via option 2 as curveall.txt and then replotted via third-part program such
as Origin.
In "examples\spectra\indigo" folder you can also find ZINDO_30.out, it corresponds to a
ZINDO calculation with 30 excited states produced. You can also include it into multiple.txt.
Comparing spectra with and without spin-orbit coupling effect

ORCA program is able to take spin-orbit coupling (SOC) effect into account during TDDFT
calculation, here we plot and compare the UV-Vis spectrum with and without the SOC consideration.
Please download http://sobereva.com/multiwfn/extrafiles/SOC-TDDFT_ORCA.zip, it is an ORCA
output file of TDDFT task for Ir(ppy)3 coordinate, SOC treatment had been enabled in the
calculation via dosoc true keyword in the %tddft field. In the output information, there are excitation
energies and oscillator strengths before and after SOC correction.
Extract the .zip package, put the .out file in current folder, then create a multiple.txt with below
content
Ir_ppy3.out with SOC
Ir_ppy3.out without SOC

multiple.txt
11 // Plot spectrum
3 // Plot UV-Vis
y // For the first spectrum, let Multiwfn use the data with SOC consideration
n // For the second spectrum, let Multiwfn use the data without SOC consideration
Then you will enter the interface for setting up the spectrum. After slight adjustment of settings,
you will obtain below graph. Clearly, SOC effect has non-negligible influence on the spectrum for
present systems.
523
50000.0 0.347
with SOC
without SOC
45000.0 0.312
(L mol - 1 cm - 1 )
40000.0 0.277
35000.0 0.243
Oscillator strength
30000.0 0.208
25000.0 0.173
20000.0 0.139
15000.0 0.104
10000.0 0.069
5000.0 0.035
0.0 0.000
150.0 200.0 250.0 300.0 350.0 400.0 450.0 500.0 550.0
Wavelength (nm)
It is worth to note that the number of states with SOC is by far larger than that without SOC.
Because SOC effect splits each originally degenerate triplet state to three sublevels. For example,
assume that the TDDFT calculates 50 singlets and 50 triplets, then after taking SOC correction into
account, there will be 50+3*50=200 states. Since the number of states corresponding to SOC-
TDDFT is higher than that corresponding to regular TDDFT, when two (or more) set of data are
simulated as theoretical spectra, the first legend in multiple.txt must correspond to SOC-TDDFT
case, and when loading data for the first spectrum, you must choose y to let Multiwfn load SOC
corrected TDDFT data, as I illustrated above.
4.11.7 Plot Raman optical activity (ROA) spectrum for chiral molecule
S-methyloxirane
ROA is an important type of vibrational spectrum for chiral molecule, only chiral molecule has
ROA signal, see Section 3.21 for detail. In this example, I will illustrate how to plot ROA spectrum
for a typical chiral molecule S-methyloxirane based on output file of Gaussian freq=ROA task. The
Gaussian input and output files are S-methyloxirane_ROA.gjf and S-methyloxirane_ROA.out in
examples\spectra folder, respectively. As can be seen from the input file, this calculation takes three
incident light frequencies (500, 532 and 600 nm) into account. It is well-known that diffuse
functions are important for obtaining accurate ROA data, so aug-cc-pVDZ is used here.
examples\spectra\S-methyloxirane_ROA.out
11 // Plot spectrum
6 // ROA
2 // Three incident light frequencies are detected, here we select the 532nm case
2 // There are totally six kinds of data can be selected, here we select the commonly studied
"ROA SCP(180)", namely backscattered circular polarization ROA spectrum
14 // Scale frequencies by a scale factor
524
[Press ENTER button] // Select all frequencies

0.97 // Employ fundamental scale factor of 0.97, which is suitable for B3LYP/aug-cc-pVDZ
level
19 // Convert the ROA data outputted by Gaussian to "real" ROA intensities
532nm // Wavelength of incident light. This value should be consistent with actual
experimental condition
[Press ENTER button] // Assume that experimental temperature is 298.15K
3 // Adjust range of X axis of the spectrum
3200,200,400 // Lower limit, upper limit, stepsize
Now you can see below ROA spectrum:
2532.2 34333.5
2025.8 27466.8
1519.3 20600.1
1012.9 13733.4
506.4 6866.7
ROA intensity
0.0 0.0
-
-506.4 -6866.7
-1012.9 -13733.4
-1519.3 -20600.1
-2025.8 -27466.8
-2532.2 -34333.5
3200.0 2800.0 2400.0 2000.0 1600.0 1200.0 800.0 400.0
Wavenumber (cm - 1 )
If you intend to present this spectrum in paper, I suggest removing the labels in both the left
and right ordinates, since they are meaningless, only the shape of the curve is of chemical interest.
4.11.8 Skill: Plot spectrum for a batch of files via shell script
Note: There is an illustration video corresponding to this section: https://youtu.be/x6jp40DR24k.
In this section, I show how to plot spectrum for a batch of input files via shell script. Via this
way, all input files in current folder can be immediately converted to respective spectrum image file
by only one command!
Assume you are using Windows system, and you want to convert all Gaussian TDDFT .out file
in examples\spectra\indigo folder to UV-Vis spectrum, what you should do is:
• Copy the .out files to Multiwfn folder
• Copy examples\spectra\UV-Vis.txt and examples\spectra\batchspec.bat to Multiwfn folder
• Set "isilent" in settings.ini to 1 and save the file
• Double-clicking the batchspec.bat
Now the batch script invokes Multiwfn to process all.out files in current folder according to
525
the commands in the UV-Vis.txt. After a few seconds, you will find all spectrum image files have
been generated in current folder, the name is identical to the .out file.
If you want to plot IR spectrum for a batch of files in current folder, copy
examples\spectra\IR.txt to current folder and replace the "UV-Vis.txt" in the .bat script with "IR.txt",
then run the .bat file.
The content of UV-Vis.txt and IR.txt is very easy to understand if you already know how to run
Multiwfn in silent mode and batch mode. If you have not read Sections 5.2 and 5.3, after reading
them you will fully understand how the script works. Commonly, you should properly modify the
settings (range of axes) in the .txt file before employing it for producing spectrum for your systems.
Via the same way, you can also use Multiwfn to plot other kinds of spectra for a batch of input
files, you need to manually compile the .txt file containing proper commands.
In Linux environment, you can also use shell script to realizing the batch plotting. The
examples\spectra\batchspec.sh is a Bash script that have exactly identical function as the
batchspec.bat shown above.
4.11.9 Skill: Use spikes to indicate position of transition levels
In Multiwfn, it is possible to plot a set of spikes at the bottom of the simulated spectrum to
highlight position of specific transition levels. In Section 4.11.1 we have plotted IR spectrum for
NH3BF3, which has some featured vibration modes. This time we will use spikes with different
colors to highlight position of two kinds of modes on the map: (1) stretching vibration of B-N bond
(2) stretching vibration of N-H bonds. The index of these modes can be identified by inspecting
vibration animations in GaussView.
examples\spectra\NH3BF3_freq.out
11 // Plot spectrum
1 // The type of the spectrum is IR
23 // Set status of showing spikes to indicate transition levels
1 // Set the first set of spikes. We want to use black spikes to reveal all vibrations
a // Select all modes
5 // Black
2 // Set the second set of spikes
16-18 // Indices of stretching vibration mode of the three N-H bonds
1 // Red
3 // Set the third set of spikes
4 // Index of vibration mode of B-N bond stretching
2 // Green
0 // Return
4 // Modify Y-axis at left side
0,6000,600 // Set lower and upper limits as well as label spacing
y // Correspondingly scale Y-axis at right side
0 // Plot the graph
You will see below map
526
The red spikes in this graph clearly indicate that the N-H stretching modes have highest
frequencies, while the B-N stretching vibration mode (green spike) has frequency about 400 cm-1.
All other vibration modes are highlighted by black spikes.
It is worth to note that although the first set of spike is black and contains all vibration modes, the second and
third sets of spikes are plotted after it, therefore the N-H and B-N stretching vibration modes are in red and green
colors respectively rather than in black.
Properly using the spikes to indicate featured transitions can make the graph much more
informative. For example, when you plot UV-Vis map, you can use spikes in different colors to
distinguish different transition types (e.g. →* and n→*, or local excitation and charge transfer
excitation).
If some transitions are degenerate, you can enable Multiwfn to exhibit degeneracy in terms of
spike height. To do so, after defining spikes in option 23, choose its suboption "-3 Toggle
considering degenerate" and input a threshold for determining degeneracy. If energy span over two
or more transitions is less than the threshold, then they will be regarded as degenerate, only the
lowest lying one will be drawn as spike with height of degenerate degree, while other ones will be
invisible. For example, below is IR spectrum of cyclo[18]carbon, the green and blue spikes reveal
position of in-plane and out-plane vibration transitions, respectively. Most transitions have
degeneracy of two (full height), while a few are not degenerate and thus the spike height is only
half.
527
3500.0 250.0
(L mol - 1 cm - 1 )
3000.0
200.0
2500.0
IR intensities (km mol - 1 )

150.0
2000.0
1500.0
100.0
1000.0
50.0
500.0
0.0 0.0
2500.0 2200.0 1900.0 1600.0 1300.0 1000.0 700.0 400.0 100.0

Wavenumber (cm - 1 )
4.11.10 Examples of plotting NMR spectrum
Please read Section 3.13.5 first to gain basic knowledge about the module of plotting NMR
spectrum. In this section, a few examples will be given to show how to easily and flexibly plot NMR
spectrum in Multiwfn.
4.11.10.1 Plotting 1H and 13C NMR spectra for acetaldehyde
In this example we plot 1H and 13C NMR spectra for acetaldehyde, which is shown below.
examples\spectra\NMR\Acetaldehyde.out is output file of NMR task of Gaussian 09. The

geometry was optimized using B3LYP/def2-SVP level in vaccum, while the NMR task was
conducted at B97-2/def2-TZVP level under chloroform environment represented by SMD solvation
model. It was demonstrated that B97-2 is a good choice for theoretically evaluating NMR, at least
for 1H and 13C, see J. Chem. Theory Comput., 10, 572 (2014) for benchmark. The NMR output file
of tetramethylsilane (TMS) calculated at the same calculation level is
examples\spectra\NMR\TMS.out, as can be seen from line 355 and line 360, the isotropic magnetic
shielding value of C and H are 186.8707 and 31.5143 ppm, respectively, they will be taken as
reference values later.
We first plot 13C NMR spectrum. Boot up Multiwfn and input

examples\spectra\NMR\Acetaldehyde.out
11 // Plot various spectrum
7 // NMR
528
From option 6 in the interface, you can find the element currently considered is carbon. Now
if you directly select option 0, you will see 13C spectrum, however, the X-axis corresponds to
absolute shelding value. In order to make X-axis correspond to chemical shift, we should input
7 // Set how to determine chemical shifts
1 // Using reference shielding value to derive chemical shifts
186.8707 // Reference value of carbon in TMS (see above). Since this value is a built-in data,
in this step you can also directly input a to employ it
0 // Plot NMR spectrum
Now you can see
In the above map, the height of black spikes corresponds to the "Degeneracy" axis, while red
curves are broadended from the spikes, their values correspond to the "signal strength" axis. The
blue texts indicate the index of the atom corresponding to the peak.
You can also see following information on Multiwfn console window
Term: 1 Chemical shift: 29.657 ppm Atom: 1(C )
Term: 2 Chemical shift: 208.011 ppm Atom: 5(C )
In the NMR plotting interface, there are many options used to adjust various plotting settings,
such as range of X and Y axes, style of atom labels, color and thickness of spikes and curves, FWHM
parameter of broadening and so on, please play with them to improve the spectrum according to
your actual requirement.
Next, we plot 1H NMR spectrum. Input below commands

6 // Choose the element considered in plotting
H // Hydrogen
1 // Using reference shielding value to derive chemical shifts
a // As mentioned above, this input corresponds to using built-in reference data of TMS
evaluated at B97-2/def2-TZVP level under chloroform
It is important to notice that the shielding values of the three hydrogens in the methyl group
must be averaged, since methyl group rotates easily in actual environment and thus there is only one
NMR peak of hydrogens in this group. Thus we input
529
10 // Average shielding values of specific atoms

2-4 // H2, H3 and H4 are the hydrogens in the methyl group
Now you can see
As you can see, the plotting effect is fairly satisfactory. The information currently shown in
console window is:
Term: 1 Chemical shift: 2.070 ppm Atom: 2(H ) 3(H ) 4(H )
Term: 2 Chemical shift: 10.333 ppm Atom: 7(H )
4.11.10.2 Plotting NMR spectra for pyridine based on scaling method

As introduced in Section 3.13.5, there is another way of determining chemical shifts of 1H and
13C, namely scaling method. Via this method we do not need to calculate reference values, and good
chemical shifts could be obtained even using inexpensive calculation levels since the prefitted
scaling parameters eliminated most systematical errors. In this section we plot NMR spectrum for
pyridine based on the scaling method. examples\spectra\NMR\pyridine_scale.out is output file of
NMR task of Gaussian calculated at B3LYP/6-31G* level with chloroform environment represented
by SMD solvation model, while the geometry was optimized at B3LYP/6-31G* level in vaccum.
The error statistics of various levels given in http://cheshirenmr.info indicate that this level is one
of best levels of applying scaling method.
examples\spectra\NMR\pyridine_scale.out
7 // NMR
2 // Set slope and intercept to determine chemical shifts by scaling method
a // Use built-in slope and intercept parameters prefitted for B3LYP/6-31G* with
SMD(chloroform) level, namely slope of -0.9449 and intercept of 188.4418 for 13C NMR
0 // Plot NMR spectrum
Now you can see
530
Due to symmetry of pyridine, there are two peaks showing double degenerate character.
Similarly, you can plot 1H NMR spectrum via scaling method, namely input
H
2 // Set slope and intercept to determine chemical shift by scaling method
a
4.11.10.3 Plotting conformation weighted NMR spectrum for valine

In this section, I will illustrate how to plot conformation weighted NMR spectrum. Valine is an
essential amino acid, it has two conformers in aquous environment, as show below. The values in
the parenthese are my theoretically estimated conformation weights in water.
In this section we will simulate 1H NMR spectrum of valine in water and compare it with
experimental spectrum measured in D2O solvent. Note that since the three hydrogens in the
protonated amino group are fully substituted by deuterium in heavy water environment, we need to
eliminate its contribution from the spectrum. Also, we need to average shielding values of the
531
hydrogens in each of the two methyl groups.

The conf1.out and conf2.out in "examples\spectra\NMR\valine" folder are output files of NMR
task of Gaussian 16, the NMR calculations were conducted at B97-2/def2-TZVP level, the
geometries were optimized at B3LYP-D3(BJ)/6-311G** level, in both calculations the IEFPCM
model was employed for representing water environment.
We first create a text file named multiple.txt with following content (prefix may be added to
the file name, such as valine_multiple.txt):
examples\spectra\NMR\valine\conf1.out 0.825
examples\spectra\NMR\valine\conf2.out 0.175
As can be seen, we have specified two input files with corresponding conformation weights.
multiple.txt
7 // NMR
H
1 // Set reference shielding value to determine chemical shift
31.8294 // The TMS reference value that comes from examples\spectra\NMR\valine\TMS.out,
which was calculated via exactly the same way as current system
11-13 // Three methyl group hydrogens
14-16 // Three methyl group hydrogens
11 // Set strength of specific atoms
2,17,18 // The three hydrogens in the amino group
0 // Making them fully invisible in the spectrum
0 // Plot the NMR spectrum
Now you see the following spectrum
In order to improve the effect of the map, we close the graph and then input
532
3 // Set lower and upper limits of X-axis

4,0,0.5 // From 4.0 to 0.0 ppm with label spacing of 0.5 ppm
12 // Do not show spikes to make the spectrum clearer
18 // Other plotting settings
5 // Set X position of legends
1300 // Moving the position of the legends more left than default position
0 // Return
Now we select option 0 to replot, the current spectrum is already quite satisfactory
Experimental 1H NMR spectrum of valine in water is https://hmdb.ca/spectra/nmr_one_d/1582,

by comparing above map with it you can find our simulated NMR spectrum is reasonable and
captured all major features of the experimental one.
Note that if you only want to plot weighted spectrum or only plot the spectra for the two
conformers, you can choose corresponding suboptions in option 17.
4.11.10.4 Plotting multiple systems simultaneously

Multiwfn is able to easily plot NMR spectra of multiple systems on the same map, as
exemplified below, the prerequisite is that all systems must have the same number of atoms. In this
example, we will view the two conformers of valine as two individual systems.
Create a file named multple.txt (additional prefix may be added to the file name) with following
content, each line contains path of an input file and corresponding legend
examples\spectra\NMR\valine\conf1.out conformer 1
examples\spectra\NMR\valine\conf2.out conformer 2
Boot up Multiwfn, load the multiple.txt, and then run all commands recorded in the
examples\spectra\NMR\valine\drawmulti.txt file in turn, you will see below map. The meaning of
each command can be easily understood according to the prompt on screen.
533
4.12 Quantitative analysis of molecular surface
4.12.1 Electrostatic potential analysis on phenol molecular surface
Below I will introduce quantitative analysis of molecular surface by case study of phenol. The
theoretical basis has been documented in Section 3.15.1 and thus will not be repeated here. In this
section we only analyze electrostatic potential (ESP) on phenol vdW surface, in next section we will
then analyze average local ionization energy on the phenol surface.
examples\phenol_631Gxx.wfn // Phenol wavefunction produced at B3PW91/6-31G** level.
For most systems this level can give acceptable result of ESP analysis. For better accuracy, def-
TZVP is recommended, while more expensive def2-TZVP basis set is able to give ideal result
12 // Quantitative analysis of molecular surface
0 // Start the analysis under default settings. By default the mapped function is ESP
Now the calculation starts. Since computing ESP is time consuming, you need to wait for a
while. During the calculation some intermediate information are printed, most users do not need to
concern them. Below results will be printed on screen once the calculation has been finally finished:
Global surface minimum: -0.041203 a.u. at 1.455097 3.343708 -0.007902 Ang.
Global surface maximum: 0.085761 a.u. at -1.936645 3.093464 0.021360 Ang.
Number of surface minima: 3
# a.u. eV kcal/mol X/Y/Z coordinate(Angstrom)
1 -0.03046066 -0.828877 -19.112843 0.150202 -1.011077 -1.882004
2 -0.03045989 -0.828856 -19.112362 0.192185 -0.985412 1.877656
* 3 -0.04120321 -1.121196 -25.853368 1.455097 3.343708 -0.007902
534
Number of surface maxima: 5
1 0.02275520 0.619200 14.277975 -3.344441 -2.281045 0.047286
* 2 0.08576096 2.333674 53.811572 -1.936645 3.093464 0.021360
3 0.01935782 0.526753 12.146259 0.066223 -4.286661 0.040555
4 0.01980285 0.538863 12.425498 3.340574 -2.325727 0.021485
5 0.01583741 0.430958 9.937340 3.419218 1.225375 -0.019326
================= Summary of surface analysis =================
Volume: 835.71041 Bohr^3 ( 123.83953 Angstrom^3)
Overall surface area: 476.05682 Bohr^2 ( 133.30951 Angstrom^2)
Positive surface area: 231.09186 Bohr^2 ( 64.71232 Angstrom^2)
Negative surface area: 244.96497 Bohr^2 ( 68.59719 Angstrom^2)
Overall average value: -0.00020233 a.u. ( -0.12695332 kcal/mol)
Positive average value: 0.01877643 a.u. ( 11.78145591 kcal/mol)
Negative average value: -0.01810626 a.u. ( -11.36095315 kcal/mol)
Overall variance (sigma^2_tot): 0.00041488 a.u.^2 ( 163.34165024 (kcal/mol)^2)
Positive variance: 0.00031106 a.u.^2 ( 122.46642148 (kcal/mol)^2)
Negative variance: 0.00010382 a.u.^2 ( 40.87522876 (kcal/mol)^2)
Balance of charges (nu): 0.18762182
Product of sigma^2_tot and nu: 0.00007784 a.u.^2 ( 30.6464578 (kcal/mol)^2)
Internal charge separation (Pi): 0.01842642 a.u. ( 11.56183883 kcal/mol)
Molecular polarity index (MPI): 0.50154872 eV ( 11.56600 kcal/mol)
Nonpolar surface area (|ESP| <= 10 kcal/mol): 67.26 Angstrom^2 ( 50.45 %)
Polar surface area (|ESP| > 10 kcal/mol): 66.05 Angstrom^2 ( 49.55 %)
Above information contains various quantities relevant to ESP, see Section 3.15.1 for their
meanings and defintions. Now select option 0 in the post-processing interface to view molecular
structure and surface extrema (red and blue spheres correspond to maxima and minima,
respectively):
535
In side view:
Minimum 3 (-25.85 kcal/mol) is global minimum on the surface, its large negative value is
owing to the lone pair of oxygen. Maximum 2 (53.81 kcal/mol) is global maximum arising from the
positively charged H13, the ESP at this point is much larger than that at other maxima (where the
ESP ranges from 10 to 15 kcal/mol). This is because the presence of oxygen, which attracted a great
deal of electrons from H13. In complex, assume that only electrostatic interaction exists, monomers
always contact each other in maximally ESP complementary manner. So we can expected that in
phenol dimer, H13 and maximum 2 in a monomer, and O12 and minimum 3 in neighbour monomer
will be in a straight line (resulting hydrogen bonding), this is the exactly situation in actual geometry
of phenol dimer, see below graph. Notice that in the dimer, maximum 2 and minimum 3 shown
above have cancelled each other out.
536
Minimum 1 and 2 (both are -19.11 kcal/mol) are local minimum on the surface, mainly arise
from the abundant π electrons above and below the ring. It is well know that electrophiles always
prefer to attack the atom which has very low ESP around it, so C1 should be an ideal reaction site
for electrophilic reaction. This conclusion is partially consistent with the general knowledge that
hydroxyl is an ortho-para directing group. However, although the global minimum is closest to O12,
O12 is not the electrophilic reaction site; this contradiction reveals the inherent limitation of ESP
analysis method.
Note 1: Since the molecule has Cs symmetry, in principle, minimum 1 and 2 should have identical X and Y
coordinates. However, this cannot be exactly fulfilled in numerical process, because the points scattered on molecular
surface do not have the molecular symmetry, see Section 3.15.1 for detail. So X and Y coordinates of minimum 1
and 2 are slightly deviated to each other. If you want to refine the result, choose option 3 "Spacing of grid data for
generating molecular surface" and input a smaller value than default value. Smaller spacing of grid points yields
more accurate result, but bring higher computational burden.
Note 2: Due to limitation of the Multiwfn GUI, sometimes it is difficult to query the index of the ESP extrema
of interest, in this case using VMD instead is recommended, see Part 6 of this video https://youtu.be/QFpDf_GimA0.
Mutual peneration distance

The non-bonded radius defined in the framework of atoms in molecule (AIM) theory is the
shortest distance between a nucleus and ρ=0.001 a.u. isosurface. Let us calculate the non-bonded
radius for O12 and H13. Select option 10 in the post-processing interface and input 12, we can see
that the non-bonded radius of O12 is 1.701 Å. Select 10 and input 13, the non-bonded radius of H13
is 1.172 Å. In the phenol dimer, the H---O of the H-bond is 1.937 Å, therefore the so-called mutual
peneration distance is 1.701+1.172-1.937=0.936 Å. This is a nontrivial value, indicating the H-bond
is strong.
ESP statistical distribution on molecular surface

As the final part of ESP analysis, we examine the molecular surface area in each ESP range,
this is useful to quantitatively discuss ESP distribution on the whole molecular surface. We choose
option 9 in the post-processing interface, and then input:
all // All atoms are taken into the statistics (alternatively, if you input for example 2-4, then
only the local surfaces corresponding to atom 2, 3 and 4 will be taken into account, see Section
4.12.3 for illustration of the concept of local molecular surface)
-30,55 // The ESP range you are interested in. Since we have already known that the
minimum and the maximum ESP on the surface are -25.85 and 53.81 kcal/mol respectively, here
we input a slightly larger range to enclose them
15 // The number of intervals
3 // Both the inputted and outputted units are in kcal/mol
537
Then you will see surface area (in Å2) and corresponding percentage in the whole surface area
in each successive ESP range.
Begin End Center Area %
-30.0000 -24.3333 -27.1667 1.8192 1.3647
-24.3333 -18.6667 -21.5000 6.0284 4.5221
-18.6667 -13.0000 -15.8333 20.9732 15.7327
-13.0000 -7.3333 -10.1667 19.0390 14.2818
...
43.6667 49.3333 46.5000 1.2690 0.9519
49.3333 55.0000 52.1667 1.0457 0.7844
Sum: 133.3095 100.0000
By using these data you can use your favourite program to draw a histogram graph. For
example, we choose "center" column as X-axis and "Area" column as Y-axis to plot below graph
28
24
20
Surface area (Å )
2
16
12
0
-30 -20 -10 0 10 20 30 40 50
Electrostatic potential (kcal/mol)
From the graph it can be seen that there is a large portion of molecular surface having small
ESP value, namely from -20 to 20 kcal/mol. Among these areas, the negative part mainly
corresponds to the surface above and below the six-membered ring and shows the effect of the
abundant π-electron cloud; the positive part mainly arises from the positive charged C-H hydrogens;
the near-neutral part represents the border area between the negative and positive parts. There are
also small areas having remarkable positive and negative ESP value, corresponding to the regions
closed to the global ESP minimum and maximum, respectively.
ESP colored molecular surface

With the help of VMD program, one can plot very nice color-filled molecular surface map with
surface extrema for various real space functions based on the output of Multiwfn. Below is such a
plot for ESP, which was presented in my study of benzoapyrene diol epoxide, see Struct. Chem., 25,
1521 (2014). In which blue, white and red correspond to ESP varying from -30 to 35 kcal/mol, the
green and orange spheres correspond to ESP surface minima and maxima, respectively
538
If you would like to plot similar graphs, please download and follow this tutorial:
http://sobereva.com/multiwfn/res/plotESPsurf.pdf. However, there are many steps in this tutorial, if
you want to draw a map with similar or even better effect but in a much simpler way, see Section
4.A.13 or this video tutorial: https://youtu.be/QFpDf_GimA0. This section and video also illustrate
how to plot penetration map of van der Waals surface of different molecules, which is useful for
discussing intermolecular interactions.
Hint: Calculation speed of ESP of cubegen utility in Gaussian package is usually faster than Multiwfn. If you
have Gaussian installed on your machine and input file is .fch/fchk format, it is suggested to set "cubegenpath"
parameter in settings.ini file to actual path of cubegen, so that cubegen could be automatically invoked by Multiwfn
to evaluate ESP during the analysis. Please check Section 5.7 for detail.
If you are an ORCA user and meantime unable to access Gaussian, you can make use of "orca_vpot" utility in
ORCA package to try to reduce cost of ESP analysis on molecular surface, see
http://sobereva.com/wfnbbs/viewtopic.php?pid=937 for detail.
FAQ: Why some surface ESP minima (maxima) have positive (negative) value?
Some Multiwfn users asked me why they observed that some surface ESP minima (maxima)
have positive (negative) value. In fact this phenomenon is very common and it is never a problem
or bug. Mathematically, a minimum (maximum) refers to a point where its value is lower (higher)
than surrounding points, clearly it never implies that this point must have negative (positive) value.
If you are still confused, see below illustration
For neutral system, commonly surface ESP minima (maxima) having positive (negative) are
not chemically significant, you can simply ignore them in the discussion. If you want to remove
539
these insignificant minima (maxima), you can choose option 3 (4) in post-processing menu, and
then input d. Then you will find these unwanted extrema have disappeared.
It is also worth to note that for a cation (anion) system, commonly all surface extrema have
positive (negative) value, because the overall value of surface ESP extrema is always greatly
dominated by the net charge carried by the system.
Trick: Reuse data of mapped function generated during previous analysis

Here I introduce a trick. Maybe you have noticed that evaluation of ESP on vdW surface is
time-consuming, especially for large system with high-quality basis set. If you have performed ESP
analysis on a system, which will be analyzed again later, in fact you can export the ESP data to a
plain text file, so that next time when you analyze the same system you can directly make use of the
exported data. For other type of mapped functions, this trick also works.
Let us see an example. We first perform ESP analysis on vdW surface as usual, input below
commands:
12 // Quantitative molecular surface analysis
0 // Start the analysis
Once the calculation is finished, select option 7 to export the surface vertices with ESP values
to a plain text file named vtx.txt in current folder. After that select -1 to return to last menu.
Assume that we want to perform the analysis again. This time we can directly use the ESP data
recorded in the plain text file. Input below commands
5 // Loading mapped function values from external file during analysis
1 // Loading mapped function at all surface vertices from a plain text file
0 // Start the analysis
Once construction of molecular surface is complete, Multiwfn will prompt you to input the
path of the plain text file recording mapped function values at all surface vertices, at this point you
should simply input vtx.txt.
Since this time the mapped function values, namely ESP values, are not calculated but loaded
from vtx.txt directly, the analysis results immediately show up on the screen.
Trick: Perform ESP analysis on molecular surface solely based on cube files
Some quantum chemistry programs, such as ADF, Dmol3 and FHI-aims are unable to produce
a wavefunction file that supported by Multiwfn, however in this case it is still possible to perform
ESP analysis on molecular surface, as long as you can yield cube file for electron density and ESP
for your system by these codes. Once the cube files are obtained, you can input below command:
density.cub // Load cube file of electron density first
1 // Select the way to define surface
11 // Isosurface of the grid data in memory
0.001 // Use  = 0.001 a.u. to define the isosurface
2 // Select mapped function
1 // ESP
5
3 // The mapped function will be interpolated from an external cube file
ESP.cub // The cube file recording ESP
540
Note that the grid setting used for yielding density.cub and ESP.cub must be exactly the same,
and the grid spacing should not large, otherwise the analysis result will be inaccurate.
4.12.2 Average local ionization energy analysis (ALIE) on phenol

molecular surface
Below we will analyze average local ionization energy 𝐼 ̅ on phenol vdW surface. Boot up
Multiwfn and input
examples\phenol_631Gxx.wfn // Produced at B3PW91/6-31G** level
2 // Reselect mapped function
2 // Choose 𝐼 ̅ as mapped function
0 // Start the surface analysis.
Since calculation of 𝐼 ̅ is much simpler than ESP, the calculation is finished rapidly. Unlike
surface analysis for ESP, at this time only vdW volume, surface area, average and variance of 𝐼 ̅ on
vdW surface are outputted alongside extrema information.
Choose 0 to visualize extrema. In order to make the correspondence between extrema and
atoms clearer, we drag the "Ratio of atomic size" scale bar to 4.0, which corresponds to vdW surface,
and we disable showing of surface maxima, then we will see:
In side view
541
Low value of 𝐼 ̅ suggests that the electron at this position is not tightly bounded, the site with
lowest 𝐼 ̅ on vdW surface is usually recognized as the most vulnerable site to electrophilic attack or
to free radical attack. All sites that highly polarizable such as π electron and lone pair regions
commonly have corresponding surface minima of 𝐼 ̅ . In present instance, minimum 8 and 9
correspond to lone pair of O12, from the output on screen we can find both their 𝐼 ̅ values are
10.59eV. Minimum 4,5,11, and 3,7,10 correspond to π electrons, 𝐼 ̅ values of all of them are about
8.9eV and can be viewed as degenerate global minima. Worthnotingly, the minima above and below
the conjugated ring only present at ortho- and para-carbon. These observations perfectly explained
the effect of hydroxyl as an ortho-para directing group. Since 𝐼 ̅ at minimum 8 and 9 are obviously
larger than the 𝐼 ̅ at the minima around the carbon ring, oxygen should not be vulnerable site of
electrophilic reaction.
Plotting average local ionization energy colored molecular surface map

Note 1: There is a video illustration corresponding to this part, please have a look! https://youtu.be/-1sBa0lKhp8.
Note 2: Chinese version of this part is http://sobereva.com/514.
To provide a more complete viewpoint about distribution of 𝐼 ̅ on molecular surface, it is best

to plot molecular surface map colored according to 𝐼 .̅ This can be extremely easily done via script
and VMD program (http://www.ks.uiuc.edu/Research/vmd/). Below I show how to realize this
under Windows environment, the phenol is still taken as example. Please do below steps in turn:
• Copying the ALIE.vmd from "examples\scripts\" to VMD folder
• Copying the ALIE_extiso.bat and ALIE_extiso.txt from "examples\scripts\" folder to the
folder containing Multiwfn.exe
• Edit the ALIE_extiso.bat, change the default input file to examples\phenol_631Gxx.wfn,
change the VMD folder to actual VMD folder on your machine
• Double click the ALIE_extiso.bat icon to execute it. This script will invoke Multiwfn.exe in
current folder to conduct some calculations, then avglocion.cub, density.cub and surfanalysis.pdb
will appear in VMD folder
• Boot up VMD, input source ALIE.vmd in console window, you will immediately see below
graph (In order to gain better effect, I used Tachyon render to generate the image)
In above map, the displayed surface is  = 0.0005 a.u. isosurface. The reason why the
commonly used  = 0.001 a.u. isosurface is not adapted as the definition of the surface is because
if it is employed, then the 𝐼 ̅ distribution on the surface can hardly be distinguished. The cyan
spheres correspond to surface minimum of 𝐼 .̅ The color transition is Blue-White-Red, therefore the
542
blue color highlights the regions having relatively low 𝐼 ̅ value, where are favorable sites for
electrophilic attack.
By default, the color scale of 𝐼 ̅ is 0.32~0.36 a.u., if you find the color scale is not appropriate
for present system, you can input for example mol scaleminmax 0 1 0.31 0.38 in the VMD console
window to change the lower and upper limits to 0.31 and 0.38, respectively.
4.12.3 Atomic local molecular surface analysis for acrolein
It is well known that acrolein (see below) tends to undergo nucleophilic attrack at carbonyl
carbon and  carbon; in particular, the former is the primary site for hard nucleophilic reagent. The
so-called hard means the electron cloud of the nucleophilic reagent is difficult to be polarized; the
selectivity of reactive site for this case is usually dominated by ESP.
In this example, we will try to interpret the site-selectivity of acrolein by analyzing ESP on its
vdW surface. Note that average local ionization energy is only useful for studying electrophilic
attack, but completely useless for analyzing nucleophilic attrack.
examples\acrolein.wfn // Optimized and produced at B3LYP/6-31G** level
0 // Start the analysis for ESP
After the calculation is finished, choose 0 to visualize surface extrema:
As you can see, there is a surface minimum of ESP at the boundary of  carbon and it is very
close to  carbon. This observation indirectly reveals that nuclear charge of  carbon is more heavily
screened by electron cloud, and hence is less probable to be the site of nucleophilic attack. However,
quantitative analysis of ESP on the whole acrolein surface does not provide a direct and definitive
interpretation on the preference of reactive sites, because no surface maxima are found on carbonyl
and  carbons, hence we are unable to directly investigate the characteristic of carbonyl and  carbon.
543
In Multiwfn, the quantitative analysis can not only be applied to the whole molecular surface,
but is also applicable to local molecular surface to reveal characteristic of atom or fragment, see
Section 3.15.2.2 for introduction. Here we select option 11 in post-processing interface to calculate
and output properties of the local surfaces corresponding to each atom. Some of the results are
shown below
Note: Average and variance below are in kcal/mol and (kcal/mol)^2 respectively
Atom# All/Positive/Negative average All/Positive/Negative variance
1 -24.35251 NaN -24.35251 NaN NaN 72.72766
2 4.65672 5.74594 -1.45401 9.79575 9.01495 0.78080
3 1.30965 2.36405 -0.88391 2.70896 2.45972 0.24925
4 8.37174 10.35813 -6.00120 36.74187 17.22299 19.51888
5 7.07040 8.67973 -6.35468 49.99108 25.33899 24.65209
6 2.21578 3.02322 -0.73880 5.15593 4.95305 0.20288
7 15.34251 15.34251 NaN NaN 22.85377 NaN
8 14.68486 14.68486 NaN NaN 25.92230 NaN
As you can see, the average ESP values on the local surface of carbonyl carbon (atom 2), 
carbon and  carbon are 4.657, 1.310 and 2.216 kcal/mol, respectively. This result clearly explained
the site-selectivity; the carbonyl carbon is the most favorable site because the average ESP on its
local surface is the most positive, and hence nucleophilic reagents (especially the hard ones) tend to
be attracted to this site. In contrast, on the local surface of  carbon the average of ESP is the smallest
compared to the other two carbons, and thus  carbon has less capacity to attract nucleophilic
reagents.
Note that some of the outputted data are NaN (Not a Number), these are not bug but
understandable. For example, the average of positive part of ESP of atom 1 is NaN, this is because
oxygen has large electronegativity, and thus on the local surface of atom 1 the ESP is completely
negative, so the average of positive ESP is unable to be computed.
If you are confused about what is "local surface of atoms" or you want to visualize them, after
you choose option 11 you can select "y" to output the surface facets to locsurf.pdb file in current
folder. Each atom in this file corresponds to a surface facet, the B-factor value corresponds to its
attribution. By this file you can visualize how the whole molecular is partitioned, the method is:
boot up VMD program and drag the pdb file into the VMD main window, in "Graphics"-
"Representation" set the "Drawing method" as "Points", set the point size to 4, and set "Coloring
Method" as "Beta". In VMD main window select "Display"-"Orthographic" and deselect "Display"-
"Depth Cueing". Then load the molecular structure file of acrolein into VMD and render it as CPK
mode, you will see below graph
544
In the graph, each point represents a surface facet; different colors represent different local surface
regions, and each one corresponds to an atom.
Beware that the "imolsurparmode" parameter in settings.ini directly affect the result of local
surface analysis, currently we are using imolsurparmode=1.
4.12.4 Fukui function distribution on local molecular surface of phenol
I have exemplified how to study Fukui function by visualizing its isosurface (Section 4.5.4)
and by condensing it to atomic value via population analysis (Section 4.7.3). In this section, I will
introduce another way to study this function, namely examining its average value on local molecular
surface corresponding to each atom. The advantage of this scheme over simply visualizing
isosurface is that the relative magnitude of Fukui function can be obtained clearly and quantitatively.
Meanwhile, unlike condensed Fukui function, this scheme is free of ambiquity of choice of
population methods. I will still use phenol as example.
Boot up Multiwfn (referred to as Multiwfn A) and input following commands

examples\phenol.wfn
2 // Select the mapped real space function on the molecular surface
0 // The function value will be loaded from an external file
1 // Set the way to define the surface
1 // Use electron density isosurface as molecular surface
0.01 // Because Fukui function on default isosurface =0.001 is often too small, enlarging
the isovalue to 0.01 makes the discussion easier
0 // Start the surface analysis
Multiwfn will generate grid data of electron density and then generate the surface vertices.
After the coordinate of these vertices are automatically outputted to surfptpos.txt in current folder,
Multiwfn A pauses. Do not terminate Multiwfn A, we boot up another Multiwfn now (referred to as
545
Multiwfn B), and then input below commands in Multiwfn B

examples\phenol.wfn
5 // We use this module to generate Fukui function on the points recorded in surfptpos.txt
1
-,examples\phenol_N-1.wfn // Subtract a property of phenol_N-1.wfn from that of phenol.wfn
100 // Load the coordinate of the points to be calculated from an external file
surfptpos.txt
t.txt // Output the coordinate and calculated function values (Fukui function) of the points
(surface vertices) to this file
Now we terminate Multiwfn B, and return to Multiwfn A, then input
t.txt // Load the Fukui function values at the surface vertices from this file
11 // Output the quantitative data of Fukui function distributed on the local vdW surface
corresponding to each atom
You can find below content from the output
Atom# All/Positive/Negative average
1 1.88921E-03 1.88921E-03 NaN
2 8.30558E-04 8.30558E-04 NaN
3 1.20307E-03 1.20307E-03 NaN
4 1.28067E-03 1.28067E-03 NaN
5 1.06720E-03 1.06720E-03 NaN
6 9.43611E-04 9.43611E-04 NaN
... [ignored]
NaN means there is no negative value of Fukui function on the local molecular surfaces. From the
result it is clear that the average of the Fukui function on the local molecular surface corresponding
to ortho (C3 and C5) and para (C1) carbons are larger than that of meta carbons (C2 and C6), this
observation correctly reflects the fact that hydroxyl is an ortho-para- director.
4.12.5 Becke surface analysis on guanine-cytosine base pair
The concept of Hirshfeld and Becke surface analysis have been detailedly introduced in Section
3.15.5, please read them first. In this section I will exemplify how to perform Becke surface analysis
on guanine-cytosine (GC) base pair to analyze the weak interaction between the two monomers.
examples\GC.wfn // Generated at M06-2X/6-31+G** level, optimized at PM7 level
1 // Change the definition of surface
6 // Use Becke surface. You can also select 5 to use Hirshfeld surface
1-13 // The index range of the atoms you are interested in (cytosine in present case)
2 // Change mapped function
Multiwfn found numerous surface minima, which are meaningless in this case, and at the same
546
time three surface maxima are found

Number of surface maxima: 3
# Value X/Y/Z coordinate(Angstrom)
1 0.042331 -0.721650 2.605906 0.043754
2 0.049281 0.767524 0.959671 -0.013820
* 3 0.053205 1.628343 -1.269364 0.030387
You can choose 0 to visualize them, see below
Since the sequence of electron density at these maxima is 32>1, one can expect that the
sequence of H-bond strength is O24-H13  H25-N6  H29-O8. This conclusion is identical to the
analysis of AIM bond critical point analysis.
If you want to visualize the shape of Becke surface, simply choose option -3. If you want to
plot the Becke surface colored by mapped function value, you need to make use of VMD, as
described below. Select option 6 and input 100 to export all surface vertices to vtx.pdb in current
folder and in the meantime multiply the data by 100 (since electron density in the Becke surface is
small, while the B-factor in pdb format only has two decimal places). Each atom in vtx.pdb now
corresponds to a surface vertex, and its B-factor stands for electron density at corresponding position.
Drag examples\GC.pdb into main window of VMD program, this file contains geometry of
present system. Select "Graphics"-"Representation", change the drawing method to "Licorice" and
decrease bond radius to 0.2. Then drag the vtx.pdb into VMD main window, select "Graphics"-
"Representation", change the drawing method from "Lines" to "Points", set coloring method to
"Beta", enlarge the point size to 6, run command color scale method BWR in VMD console window.
Now you should see
Becke surface is represented by points, and the three red zones correspond to high electron
density region, which stem from H-bonds. This example demonstrated that Becke surface analysis
547
is useful to reveal the region where intermolecular interaction is evident.

Repeat above analysis but with dnorm as mapped function, and replot the Becke surface with the
same procedure. Then in the graphical representation window, select "Trajectory" tab, input -0.5
and 2.0 in the "Color scale data range", you will see below graph
As you can see, using dnorm as mapped function is also capable of revealing weak interactions;
the blue regions corresponds to close contact between the fragment you are interested in (cytosine)
and other fragments (guanine only). The advantage of using dnorm over electron density is that the
former does not rely on wavefunction information, so the input file you used can only contain
geometry information, such as .pdb and .xyz are completely acceptable in this case. However the
physical meaning and robustness of dnorm is not as good as electron density.
In fact, the present analysis can also be realized in terms of Hirshfeld surface analysis, the
computational cost is lower (especially when the number of atoms is high). After carefully reading
next section you will know how to do.
4.12.6 Hirshfeld surface analysis and fingerprint plot analysis on urea

crystal
In this section we perform Hirshfeld surface analysis for urea crystal. .cif is the most popular
crystal structure format, however currently Multiwfn does not directly support it. So I extracted a
cluster of ureas from its crystal structure. The file examples\Urea_crystal.pdb contains 11 ureas, the
central one will be defined as the fragment in our Hirshfeld surface analysis. If you do not how to
extract a cluster of molecules from molecular crystal, please follow this video tutorial:
https://youtu.be/fLXr6S-8-8g.
548
Notice that Hirshfeld surface analysis requires electron density of atoms in their free-states, in Multiwfn there
are two ways to provide them:
(1) Using built-in atomic densities, see Appendix 3 for detail. In this example we use this way because it is
more convenient than way 2 and works equally well
(2) Evaluating atomic densities based on atomic .wfn files, see Secion 3.7.3 for detail. What we need to do is
simply moving the "atomwfn" folder in "examples" folder to current folder. Then during generating Hirshfeld surface,
Multiwfn will automatically use the atomic wavefunction files in this folder.

examples\Urea_crystal.pdb
1 // Change surface type
5 // Use Hirshfeld surface
16,36,58,2,77,55,34,13 // The index of the atoms in the central urea
0 // Start calculation. Note that the default mapped function dnorm is used here
1 // Evaluating atomic densities based on built-in atomic densities
After the calculation is finished, select option 6 to export the surface vertices with dnorm value
to vtx.pdb, and then plot them in VMD by using the method shown in the last section, you will see
a graph very similar to the second figure in Section 3.15.5.
Next, we draw fingerprint plot. Input below commands

20 // Fingerprint plot analysis
0 // Start fingerprint analysis
3 // Set color scale of fingerprint plot
0,5 // Lower and upper limits
0 // Plot the fingerprint map on screen
Then you will see below graph
549
In this figure, X and Y axes correspond to di and de, respectively. It can be seen that there is a pair
of spikes at the bottom left of the plot (i.e. short di and de), this observation suggests that urea
behaves as both H-bond acceptor (the lower spike, di > de) and H-bond donor (the upper spike, di <
de). It is worth to note that the coloring method could be changed, if you want to use white
background color, you can choose option "6 Set color transition" and then select "3 Rainbow
started from white".
Fingerprint plot for local contact surface

In Multiwfn, the fingerprint plot can be drawn not only for overall Hirshfeld surface, but also
for local contact surface (see Section 3.15.5 for detail). Let us check the fingerprint plot of the local
contact surface between the four hydrogens in the central urea and all atoms in the peripheral ureas.
After closing the window of showing fingerprint plot, select -1 to return, then input 20 to enter
to the interface of fingerprint plot analysis again. Next, we need to define "inside atoms" set and
"outside atoms" set, only the points on the contact surface between the two sets will be taken into
account in the fingerprint plot analysis. Now, we choose option 1 to define the "inside atoms" set.
You will be asked to input two conditions in turn, their intersection will be employed as the set. We
first press ENTER button directly to use default atom range (i.e. all atoms in the central urea) and
then input H to select all hydrogens. As can be seen from screen, all the four hydrogens in the central
urea have been defined as the "inside atoms" set. Since the default "outside atoms" set is just all
atoms in peripheral ureas, we do not need to modify it.
Now, select option 0 to start the fingerprint plot analysis. Notice below prompt on screen:
The area of the local contact surface is 63.803 Angstrom^2
The area of the total contact surface is 93.438 Angstrom^2
The local surface occupies 68.28% of the total surface
This information indicates that the area occupied by the local contact surface defined by us is 63.8
Å2. Clearly, by properly utilizing this feature, you can easily obtain the area corresponding to any
specific contact between the central molecule and the peripheral molecules!
After that, choose option 0 to plot the fingerprint map, you will see
550
Since this time we only considered the four hydrogens in the central urea, which purely behave as
hydrogen donor, so only one spike at bottom left of the plot can be observed.
It is interesting to check the shape of the local contact surface. To this end, after closing the
fingerprint map we choose option 5 to export all points on the contact surface to finger.pdb in current
folder. Plot them in VMD by using the method shown in the Section 4.12.5, you will see
Clearly, this surface well portrays the contact between the hydrogens in the central urea and the
atoms in the peripheral ureas. The color corresponds to dnorm value, more blue imples more close
contact.
Next, we check the fingerprint plot between the hydrogens in the central urea and the oxygen
atom marked by yellow arrow in the above figure. Input below commands
-1 // Return
20 // Fingerplot analysis
1 // Set inside atoms set
[Press ENTER button directly]
551
H
2 // Set outside atoms set
76 // The index of the oxygen
[Press ENTER button directly] // Do not utilize element type as filter condition
0 // Start fingerprint analysis
From the outputted information on screen you can find the local contact surface produced this
time is 6.257 Å2, which corresponds to 6.7% of total contact surface area. Then we input
3 // Set color scale
0,5 // Lower and upper limits
0 // Show fingerprint map on screen
The fingerprint plot and the corresponding surface points are shown below
In the fingerprint plot you can see that the distribution scope of surface points is narrow, and
the spike is quite evident, suggesting strong H-bond character due to the contact of H and O.
Fingerprint is especially useful for comparison of intermolecular interactions in different
crystals, see CrystEngComm, 11, 19 (2009) for discussions.
Using VMD to plot color-mapped isosurface of Hirshfeld/Becke surface

Here I describe how to easily plot very pretty Hirshfeld (or Becke) surface mapped by electron
density with promolecular approximation, this map looks much better than those shown above. Urea
cluster is still taken as example. There is a video illustrating the whole process, it is highly suggested
to take a look at it: https://youtu.be/Vy9pLQt4aDs.
examples\Urea_crystal.pdb
1 // Change surface definition
5 // Use Hirshfeld surface
16,36,58,2,77,55,34,13 // The index of the atoms in the central urea
11 // Electron density under promolecular approxmation (namely the density of the system is
evaluated as superposition of bulit-in atomic densities in their free-states)
1 // Use built-in atomic densities to construct Hirshfeld surface
552
-2 // Export the grid data used to define Hirshfeld surface as surf.cub in current folder
13 // Calculate grid data of mapped function and export it to mapfunc.cub in current folder
Now you have surf.cub and mapfunc.cub in current folder, move them to the VMD folder. Then
copy the examples\scripts\hirsh_rho.vmd into the VMD folder. Boot up VMD, input source
hirsh_rho.vmd in VMD console window to run this script. For the present case it is better to also
input material change diffuse Translucent 0.8 in the console window to make the surface brighter.
Finally you can see below graph. Note that the plotting script sets color transition to Blue-White-
Red, which corresponds to electron density varying from 0.0 to 0.015 a.u. Clearly, from the graph
one can easily recognize the evident intermolecular interaction regions.
You can manually modify the .vmd plotting script to meet your expectation. Via basically the
same procedure, you can also plot dnorm mapped Hirshfeld (or Becke) surface, in this case you should
use examples\scripts\hirsh_dnorm.vmd script.
4.12.7 Predict density of molecular crystal of FOX-7
As introduced in Section 3.15.1, many condensed phase properties of a molecule can be

predicted based on the result of quantitative molecular surface analysis of electrostatic potential
(ESP). For example, in Mol. Phys., 107, 2095 (2009), Politzer et al. showed that crystal density of
molecules only containing C, H, N, O can be predicted as
M
 = +  ( tot
2
)+
Vm
where  = 0.9183,  = 0.0028 and  = 0.0443 when the wavefunction is generated at B3PW91/6-
31G** level and the unit of M/Vm and  tot

2
are g/cm3 and (kcal/mol)2, respectively.
In this section, I illustrate how to use above formula to predict density of molecular crystal of
FOX-7 (1,1-diamino-2,2-dinitroethene), which is an insensitive high explosive compound. More
illustrations of property prediction can be found in my blog article "Using Multiwfn to predict
crystal density, heat of vaporization, boiling point and solvation free energy" (in Chinese,
First, we optimize geometry of FOX-7 and yield wavefunction file at B3PW91/6-31G** level,
553
which is the level used by Politzer et al. in their Mol. Phys. paper. The resulting FOX-7.wfn has
provided as examples\FOX-7.wfn.
examples\FOX-7.wfn
0 // Start analysis for default real space function (ESP) on default surface (0.001 a.u.
isosurface of electron density)
After a while, you will find below output on screen
Volume: 942.48700 Bohr^3 ( 139.66220 Angstrom^3)
Estimated density according to mass and volume (M/V): 1.7606 g/cm^3
...[ignored]
Product of sigma^2_tot and miu: 0.00020164 a.u.^2 ( 79.40119 (kcal/mol)^2)
Internal charge separation (Pi): 0.03740373 a.u. ( 23.47121 kcal/mol)
From the output, we find that M/Vm=1.7606 g/cm3 and  tot

2
=79.40119 (kcal/mol)2, therefore
the density could be predicted as 0.9183*1.7606+0.0028*79.40119+0.0443=1.883 g/cm3. The

experimental density of FOX-7 crystal is 1.885 g/cm3, which can be found at corresponding wiki
page (https://en.wikipedia.org/wiki/FOX-7). Clearly, our prediction is extremely successful, the
error is merely -0.002 g/cm3! However, the suprisingly good result is fortuitous to a large extent,
since according to the test in the Mol. Phys. paper, the RMS error using above prediction formula is
0.047 g/cm3.
4.12.8 Quantitative analysis of orbital overlap distance function D(r)

on thioformic acid molecular surface
The content of this section was kindly contributed by Arshad Mehmood and slightly adapted by Tian Lu.
This example is a continuation of Section 4.5.7. Here I illustrate the quantitative analysis of
orbital overlap length function D(r) on molecular electron density isosurface of thioformic acid.
examples\ThioformicAcid.wfn // Thioformic acid optimized at B3LYP/6-311++G(2d,2p)
6 // Orbital overlap distance function D(r), which maximizes EDR(r;d) with respect to d
2 // Use default value of total number, start and increment of EDR exponents. Please consult
Section 4.5.7 for more information.
0 // Start analysis now!
Now the analysis starts. This step will take some time. Once calculation is finished, following
results will be printed on screen along with other information:
Global surface minimum: 2.789918 a.u. at 1.983402 -0.346198 1.757884 Ang
Global surface maximum: 3.541349 a.u. at -2.861073 -1.074395 -0.095237 Ang
The number of surface minima: 4
554
1 3.296958 -1.622302 2.055665 0.423383
2 3.218284 0.950838 2.841231 0.010991
* 3 2.789918 1.983402 -0.346198 1.757884
4 2.790103 2.078805 -0.344424 -1.731277
The number of surface maxima: 10
* 1 3.541349 -2.861073 -1.074395 -0.095237
2 3.401958 -0.539716 2.324441 0.014652
3 3.502485 0.041422 -2.239222 -0.308278
4 3.502696 0.089030 -2.257271 0.004859
5 3.494004 0.207205 -2.075594 -0.673828
6 3.496480 0.164501 -2.070255 0.707373
7 3.359381 0.542328 1.465019 -1.650518
8 3.358980 0.496994 1.456262 1.657761
9 3.311711 2.030429 1.904600 0.023877
10 2.918240 2.847314 -1.370280 0.021275
Now select 0 to view surface minima and maxima:
This graph shows molecular structure and surface extrema (red and blue spheres correspond to
surface maxima and minima, respectively). It can be seen that surface minima is present on oxygen
atom due to compact lone pair and surface maxima is located on sulfur atom due to its more diffuse
and weakly bound lone pair electrons.
4.12.9 Evaluate vdW surface area of the whole system as well as

individual fragment
Note: Chinese version of this Section is http://sobereva.com/487, which also contains more discussions.
After reading Section 4.12.1, you must have already known how to evaluate area of molecular
vdW surface. In this section, I will discuss more about this topic. Dopamine will be employed as
example, its properly optimized geometry is shown below
555
Evaluating vdW surface area of dopamine corresponding to condensed phase

According to the Bader's paper J. Am. Chem. Soc., 109, 7968 (1987), =0.001 and 0.002 a.u.
isosurfaces can be defined as vdW surface in gas and condensed phase, respectively. The volume of
the latter is smaller than the former, because in condensed phase the vdW surface penetration must
be evident due to intermolecular interaction. Here we will calculate area of vdW surface
corresponding to condensed phase for dopamine. Boot up Multiwfn and input
examples\dopamine.wfn // Generated using B3LYP/6-31G* level. Commonly the quality of
density at this level is absolutely adequate
1 // Isosurface of electron density
0.002 // Isovalue (a.u.)
6 // Start analysis without consideration of mapped function
You only need to pay attention to below line in the output:
That means, the area of the whole molecule is 181.6 Å2.
Evaluating surface area of amino group in dopamine

Next, I illustrate how to calculate surface area of a specific fragment, the amino group in the
dopamine is taken as example. In the post-processing menu, we input
12 // Output surface properties of specific fragment
3,19,20 // The indices of the atoms in the amino group
You will see
The contribution of the amino group to the whole vdW surface thus can be calculated as
27.9/181.6*100%=15.4%.
If you want to visualize the vdW surface that attributed to the amino group, we should input y
to let Multiwfn export locsurf.pdb in current folder. Then load this file into VMD visualization
program (http://www.ks.uiuc.edu/Research/vmd/), in the "Graphics" - "Representation" set
"Drawing method" as "Points", set "Coloring method" as "Beta", and then load the structure file of
present system (examples\dopamine.xyz) into VMD to also plot the molecular geometry in the map,
after slight adjustment you will see
556
In above figure, each point denotes a vertex comprising the electron density isosurface of 0.002
a.u., the blue area corresponds to the local region belonging to the amino group. Clearly, the partition
of the entire vdW surface is very reasonable, thus the outputted area of the amino group by Multiwfn
must be reliable and meaningful.
Evaluating vdW surface area without wavefunction information

Sometimes we are difficult to generate wavefunction file due to various reasons, in this case
we can still use Multiwfn to evaluate vdW surface area. In this case, the electron density we
employed should be promolecular density, which is the molecular electron density approximately
constructed by simply superimposing the electron density of each atom in its isolated state according
to the coordinates of the atoms in the molecule..
For example, we only have examples\dopamine.xyz in hand, you can boot up Multiwfn and
load this file, then input
2 // Isosurface of a specific real space function
1 // Promolecular electron density
0.002 // Isovalue (a.u.)
The calculated result is
Obviously the result is reasonable, the value 195.2 Å2 is in qualitative agreement with the 181.6
Å2 we previously calculated based on the B3LYP/6-31G* wavefunction.
If then we calculate area of the amino group moiety, the result will be 31.5 Å2, which is also
close to the 27.9 Å2 calculated based on the DFT density. In particular, the occupancy of this group
31.5/195.2*100%=16.1% is even nearly quantitatively consistent with the 15.4% we calculated
before.
4.12.10 Quantification of area of sigma-hole and pi-hole
Introduction
-hole and -hole correspond to local regions with evident positive electrostatic potential (ESP)
on van der Waals (vdW) surface due to depletion of -electron and -electron, respectively. The
557
region corresponding to these holes can behave as electron acceptor (local Lewis-acid) to form non-
covalent interactions that dominated by electrostatic attraction, such as halogen bond. If you are not
familiar with these two concepts, reading a review article J. Comput. Chem., 39, 464 (2017) is
recommended. In literatures, the -hole and -hole are commonly revealed via analysis of ESP
extrema on vdW surface, the ESP value at the corresponding extrema is often employed as a
quantitative measure of potential strength as electron acceptor.
In the present section, I will show it is also possible to use Multiwfn to calculate surface area
corresponding to selected -hole and -hole, and meantime based on the outputted file, the
corresponding local surface could be directly visualized in VMD. I suggest you reading part 2 of
Section 3.15.2.2, in which the algorithm used in this analysis is described. ClPO2 is taken as example
here, which contains -hole at the end of chlorine atom as well as -hole above and below the
phosphorus atom.
Quantitative analysis of ESP on vdW surface

First, we carry out regular quantitative analysis of ESP on vdW surface. Boot up Multiwfn and
input
examples\ClPO2.fch // Geometry and wavefunction were produced at PBE0/def2-TZVP
0 // Start analysis, the mapped function is default to ESP
As can be seen from the output, three ESP maxima on the vdW surface are found, their ESP
values and coordinates are shown below:
1 0.07562734 2.057924 47.456909 -1.949844 -0.043516 -0.320257
2 0.04258971 1.158925 26.725470 0.015842 -0.054525 3.395121
* 3 0.07568641 2.059532 47.493977 1.945305 0.044502 -0.234288
Now enter option 0 to check index and visually examine position of the surface ESP maxima,
see the left side of the figure below (all surface minima are hidden). If you plot ESP colored vdW
surface as well as surface extrema according to the method described in Section 4.A.13, you can
obtain right side of the below graph, in which red and blue correspond to positive and negative ESP,
respectively.
From above figure it can be seen that surface maxima 1 and 3 correspond to -hole at the two sides,
while maximum 1 corresponds to -hole.
Check surface region corresponding to positive ESP value

Since -hole and -hole correspond to evidently positive ESP value, it is naturally expected
558
that area of positive ESP region around a surface maximum is a direct measure of /-hole size.
Now assume that we want to measure area of -hole corresponding to maximum 3, in the post-
processing menu we should input below commands
14 // Calculate area of the region around a specific surface extreme
2 // Maximum
3 // Select maximum 3 (corresponding to one of -holes)
0 // Set criterion as 0 a.u.
Now we can find below output
Number of surface vertices in selected surface region: 4307
Area of selected surface region: 55.946 Angstrom^2
Average value of selected surface region: 0.02650 a.u.
Product of above two values: 1.48230 a.u.*Angstrom^2
The output indicates that there are 4307 surface vertices directly or indirectly connected to
maximum 3 with ESP values larger than 0 (i.e. positive ESP), the area of this local surface is 55.94
Å2 and average ESP is 0.0265 a.u. According to chemical intuition, the calculated area is obviously
too large compared to expected -hole area, what is the reason?
In current folder, you can find a file named selsurf.pqr, which contains coordinate all selected
surface vertices and its "Charge" column corresponds to ESP in a.u. Now we load this file into VMD
program. In addition, in the post-processing menu of Multiwfn, we choose option 5 to export a pdb
file containing molecular geometry, and then also load this file into VMD. In the "Graphics" -
"Representation" panel of VMD, we set "Drawing Method" of the molecule as "Licorice" with
"Bond Radius" of 0.2, then set "Drawing Method" of the surface vertices as "Point" with "Size" of
16, then set "Coloring Method" as "Charge". The current graph should look like below
In this graph, the more blue the point, the higher the ESP value. It is evident that our currently
selected local surface does not only correspond to a -hole, but corresponds to the entire positive
ESP surface region.
Calculate surface area corresponding to -hole

Clearly, if we want to only study a region corresponding to a -hole, the ESP criterion should
be set to a larger value than 0 but smaller than the ESP value at the -hole surface maximum (0.0756
a.u., see above). In order to find an appropriate criterion, in the "Graphics" - "Representation" panel,
we switch "Selected Molecule" to the entry corresponding to selsurf.pqr, then input charge > 0.04
in "Selected Atoms" textbox, now the graphical window becomes:
559
From the graph it can be seen that criterion of 0.04 a.u. is suitable for defining the local surface
corresponding to -hole of present system. Above map contains two blue local surfaces since there
is a -hole at each side of the phosphorus atom. To calculate area of each -hole, we input
2 // Maximum
3 // Select maximum 3 (corresponding to one of -holes)
0.04 // Set criterion as 0.04 a.u.
The result is
Number of surface vertices in selected surface region: 271
Area of selected surface region: 3.570 Angstrom^2
Average value of selected surface region: 0.05772 a.u.
Product of above two values: 0.20608 a.u.*Angstrom^2
The calculated 3.57 is a very reasonable area of a typical -hole. If you visualize the generated
Å2
selsurf.pqr by VMD to examine the selected local surface, you will find the region just corresponds
to one of the two -holes shown in above surface map. Evidently, the total area of the -holes in
current system should be 2*3.57=7.14 Å2.
Calculate surface area corresponding to -hole

Next, we use similar way to calculate area of the -hole at the end of the chlorine atom. In this
case we should not use 0.04 a.u. as criterion, because the ESP value at surface maximum of the -
hole is only 0.0425 a.u. In VMD, we can try different criterions by inputting charge > xxx until
finding the best one to represent the -hole. After a few attempts, 0.03 a.u. was found to be a
reasonable value, therefore we input below command in the post-processing menu
2 // Maximum
2 // Select maximum 2 (corresponding to the -hole)
0.03 // Set criterion value as 0.03 a.u.
The area is found to be 4.88 Å2, while average ESP value in this region is 0.03617 a.u., which is
evidently smaller than that of the -hole. If you plot the exported selsurf.pqr in VMD as points, and
set color scale as 0.0~0.05 (In the "Representation" panel, choose "Trajectory" tab, then set "Color
Scale Data Range"), you will see below map, indeed the selected surface region well exhibits
expected -hole character.
560
It is important to point out that the calculated area is directly dependent on the choice of the
criterion, while there is no unique way of determining the perfect criterion. In practical study, you
can try to define the criterion as e.g. 60% of ESP value at corresponding surface maximum, or
consider defining the criterion as a value lower than surface maximum by e.g. 10 kcal/mol.
Noticeably, the option 14 can not only measure area around a surface maximum, but can also
calculate area around a surface minimum. Thus you can try to use this feature to quantify the area
corresponding to various lone pairs.
4.12.11 Basin-like analysis of molecular surface for electrostatic

potential
Just as the whole 3-dimensions molecular space can be partitioned as basins based on e.g.
electron density and electron localization function so that character of local regions could be
discussed, it is also possible to employs analogous idea to partition the whole molecular surface as
individual local surface based on a specific mapped function, so that chemically interesting
information could be gained. In this example, we will decompose the whole vdW surface of ClPO2
to contributions that source from its surface ESP minima and maxima. Please read part 3 of Section
3.15.2.2 to gain basic knowledge about the algorithm employed in this analysis. The ClPO2 has
already been investigated by means of molecular surface analysis in Section 4.12.10, please read it
if you have not.
examples\ClPO2.fch // Geometry and wavefunction were produced at PBE0/def2-TZVP
0 // Start analysis, the mapped function is default to ESP
15 // Basin-like partition of surface and calculate areas
Then you can find below output on screen
Minimum 1 N_vert: 1596, 19.615 Angstrom^2 Avg. value: -0.023076 a.u.
Minimum 2 N_vert: 1613, 19.874 Angstrom^2 Avg. value: -0.022824 a.u.
Maximum 1 N_vert: 1312, 16.689 Angstrom^2 Avg. value: 0.028729 a.u.
Above output presents information of "surface basins" (i.e. local molecular surfaces) corresponding
to different surface extrema. The "N_vert" denotes the number of surface vertices belonging to the
561
surface basin, the area as well as average of the mapped function in the surface basin are also shown.
Multiwfn also exported a file named surfbasin.pdb in current folder, which contains all surface
vertices, their B-factors correspond to the index of the surface basin that the vertex attributed to
(positive and negative Beta values correspond indices of surface maxima and minima, respectively).
Index of a surface basin is identical to index of corresponding surface extreme, each surface basin
contains and only contains one surface extreme. Note that surface minimum with positive value and
surface maximum with negative value do not have accompanied surface basin, this is easy to
understand if you have correctly understand the algorithm described in Section 3.15.2.2.
In order to vividly examine the surface basins, you can load the surfbasin.pdb into VMD, then
set drawing method as "Points" while set coloring method as "Beta". Also, we choose corresponding
options in Multiwfn to export pdb file of molecular structure (option 5) and surface extreme (option
2) and then display them in VMD. Finally you can obtain below graph, calculated data are also
marked
In the current graph, red points around the minimum 1 collectively exhibit the region of surface
basin 1, while gray and iceblue points display surface basin corresponding to maxima 1 and 2,
respectively. Clearly, via the analysis we currently employed, we are able to make clear the intrinsic
contributions that stem from different extrema to the overall positive or negative surface region. For
example, the percentage contribution to the positive surface region due to the maximum 2, which
results from -hole of the chlorine atom, is 21.539/(16.689+21.539+16.040)100%=39.7%.
The sum of areas of all maxima (minima) is not exactly identical to the positive (negative) surface area outputted
in the "Summary of surface analysis" section, because there are some boundary surface facets, whose three vertices
do not have identical attribution. These facets are ignored during calculation of area and average of function value
of surface basins.
By the way, you can also make VMD to solely display specific surface basin. For example, by
inputting beta=-1 and beta=2 in "Selected Atoms" textbox of "Graphics" - "Representation" panel
of VMD and then set color as orange, you will respectively observe surface basin corresponding to
minimum 1 and maximum 2:
562
It is noteworthy that, due to the C2v molecular symmetry, the minima 1 and 2 should have
identical values, maxima 1 and 3 should also have identical values. The slight violation of the
equivalency, as shown in above computed data, is due to numerical aspect reasons. When you report
data of the surface basins corresponding to the -hole (maxima 1 and 3), it is reasonable to take the
average of them, that is the area in each side should be (16.040+16.689)/2=16.4 Å2.
4.12.12 Estimate kinetic diameter for small molcules

Note: Chinese version of this section with more discussions is http://sobereva.com/503.
The kinetic diameter is an important quantity in the study of gas separation. Most cited values
of kinetic diameter of small molecules are taken from Breck's book Zeolite Molecular Sieves;
Structure, Chemistry and Use, which was published in 1974. In J. Phys. Chem. A, 118, 1150 (2014),
the authors proposed a general way of calculating kinetic diameter purely based on isosurface of
electron density. As exemplified below (adapted from the J. Phys. Chem. A paper), the distance
enclosed by the two black arrows could be used to define the kinetic diameter
In the paper, it is found that the calculated values match best to the Breck's values if isovalue of
electron density is set to 0.0015 a.u. when PBE0/def2-TZVP is used in the wavefunction generation.
In this section, I will show how to use the quantitative molecular surface analysis module to
realize the above mentioned method to calculate kinetic diameter for a typical molecule, CO.
The .fch file yielded by optimization task at PBE0/def2-TZVP level has been provided as
examples\CO.fch.
Before doing the calculation, we should use main function 0 to check the orientation of the
molecule in the CO.fch, as shown below
563
Clearly, the molecular axis is exactly parallel to the Z-axis, therefore the kinetic diameter can be
calculated as difference between surface vertex with most positive X value and that with most
negative X value (the surface is defined as 0.0015 a.u. isosurface of electron density).
Now we conduct the calculation. Boot up Multiwfn and input
examples\CO.fch
0.0015 // Isovalue
After properly scrolling up, you can find below output:
Among all surface vertices:
Min-X: -1.7527 Max-X: 1.7528 Angstrom
Min-Y: -1.7527 Max-Y: 1.7528 Angstrom
Min-Z: -2.5093 Max-Z: 2.0951 Angstrom
That means the kinetic diameter can be calculated as 1.7528-(-1.7527)=3.505 Å. According to Table
2 of the J. Phys. Chem. A paper, the slope of fitting is 1.025, therefore the final estimated value
should be 3.505*1.025=3.593 Å, which is in qualitative agreement with the Breck's value (3.76 Å).
The CO is a very simple case, while for much more complicated molecule, you have to use
VMD (http://www.ks.uiuc.edu/Research/vmd/) to measure the distance between two proper surface
vertices to estimate the kinetic diameter. Again taking the CO as example, in the post-processing
menu, choose option 6 to export vtx.pdb in current folder, which records all surface vertices. Then
load this file into VMD, in the "Graphics" - "Representation", set "Drawing method" as "Points".
Then in the VMD main window, choose "Display" - "Orthographic". After that, activate the VMD
graphical window, press button 2 on your keyboard, then click two vertices at proper positions.
From below map, you can find the distance between the two vertices is 3.47 Å, which is very close
the value 3.505 Å given above.
564
Choosing proper surface vertices is not quite easy, please be very patient. If the vertices are
mis-selected, you can enter "Graphics" - "Labels", then delete the unwanted atoms labels and bond
labels.
4.12.13 Using local electron affinity to reveal electrophilic regions
We have studied average local ionization energy (IEL) in Section 4.12.2, please read it first if
you have not, since the present section can be viewed as an extension of that section. There is a
function closely related to IEL, namely local electron affinity (EAL), which was proposed in J. Mol.
Model., 9, 342 (2003) and defined as
−   i (r )  i
2
EAL (r ) = ivir
  (r )
2
i
ivir
where  denotes orbital energy,  is orbital wavefunction. EAL corresponds to user-defined function
27 in Multiwfn.
EAL approximately reveals electron affinity at a given point based on Koopmans'
approximation. It is expected that the more positive the EAL at a point, the stronger the
electrophilicity in this region. Clearly, this nature makes EAL have certain ability in revealing
favorable site of nucleophilic attack.
The best way of exhibit distribution of EAL should be mapping it to molecular surface via
different colors. In Section 4.12.2 I have illustrated how to plot IEL mapped molecular surface via
script of VMD program based on Multiwfn output files, below I illustrate how to plot this kind of
map for EAL via nearly the same way.
The examples\CH3Cl.fchk will be taken as example, it was generated at B3LYP/6-31G* level.
Note that EAL is meaningful only when diffuse functions are not employed. In addition, you must
use a file containing virtual orbitals as input file, such as .mwfn, .fch and .molden, because virtual
orbitals are involved in EAL calculation.
To plot the map, you should do following things (below procedure only works for Windows
platform, for Linux platform you should write similar script yourself)
• Copy LEA_isoext.bat and LEA_isoext.txt from "examples\scripts\local_EA" folder to current
folder. Edit the VMD path in the .bat file to actual VMD folder on your machine
• Copy LEA_isoext.vmd from "examples\scripts\local_EA" folder to VMD folder
565
• Edit the path of input file in LEA_isoext.bat as actual path, namely examples\CH3Cl.fchk
• Double click LEA_isoext.bat to run it. Then Multiwfn will be invoked to generate density.cub,
userfunc.cub (cube file of EAL) and surfanalysis.pdb (containing surface extrema of EAL on  = 0.01
a.u. isosurface), then they will be automatically moved to VMD folder
Boot up VMD and input source LEA_isoext.vmd in VMD console window to run this script,
then you will see below graph
This map shows EAL mapped  = 0.01 a.u. isosurface, the color scale is from -0.80 (blue) to
-0.30 (red) Hartree, cyan spheres correspond to maxima of EAL on this surface. As can be seen, the
regions around the hydrogens have highest EAL, the value at corresponding surface maxima is -0.31
Hartree. Presence of these regions may be attributed to the fact that the hydrogens have positive
charge and thus has higher tendency to attract additional electron. At the end of the Cl atom there is
also a region with relatively more positive EAL, the maximum at that region has EAL of -0.46 Hartree.
The appearance of this region should be closely related to the  hole of the Cl atom.
It is worth to note that the most appropriate color scale of EAL is usually very different from
system to system. If you find there is only one color over the entire isosurface, or the color in
different regions cannot be clearly distinguished, you should adjust lower and upper limits of color
scale. If you input for example mol scaleminmax 0 1 -1.0 -0.4 in VMD console window, then the
color scale will be changed to -1.0 ~ -0.4.
By the way, in order to fully understand how the script works, you are encouraged to manually
input the commands recorded in the LEA_isoext.txt one by one into Multiwfn window.
4.13 Process grid data
Main function 13 includes a bunch of subfunctions, by using them you can process the grid
data loaded from Gaussian-type cube file (.cub), DMol3 grid file (.grd), ParaView VTK Image Data
file (.vti), or the grid data directly generated by such as main function 5 of Multiwfn. In this section
I present several simple applications, please play with other subfunctions by yourself.
4.13.1 Extract data points in a plane
In this example we extract average XY-plane data between Z=28 and Z=32 Å to a plain text
566
file.
dens.cub // A cube file generated by Multiwfn or by some external programs, since cube file
is generally large, it is not provided in "example" folder. You can also use the grid data generated
internally by Multiwfn instead, that is use main function 5 to calculate grid data first and then choose
0 to return to main menu (the just generated grid data is present in memory)
13 // Process grid data
5 // Extract average plane data
28,32 // Range of Z (in Ångstrom)
Now the data points are exported to output.txt in current folder, including X,Y coordinates and
value. You can import this file to plotting software such as sigmaplot to draw plane graph.
Another example, we extract data point on the plane defined by atom 4,6,2.
dens.cub
8 // Output data in a plane by specifying three atom indices. This function is commonly used
to extract tilted plane, if the plane is parallel to XY, YZ or XZ, you should use function 1,2 or 3
instead respectively
0 // Use automatically determined tolerance distance. If vertical distance between any point
and the plane you defined is smaller than tolerance distance, then the point will be outputted.
1 // Project the data points in the plane you defined to XY plane, so that you can directly
import the outputted file to plotting software to draw plane graphs
Now the data value along with coordinates is exported to output.txt in current folder.
Notice that Multiwfn does not do interpolation during plane data extraction, hence if the quality
of grid data is not fine enough (namely spacing between points is large), then the extracted plane
data will be sparse (especially severe for the plane not parallel to XY, YZ or XZ plane).
4.13.2 Perform mathematical operation on grid data
Example 1
Assume that we have two cube files A.cub and B.cub, in this example we obtain their difference
cube file (viz. A.cub minus B.cub).
A.cub // Load the first cube file into memory
11 // Grid data calculation
4 // Subtract the grid data in memory by another grid data
B.cub // The cube file containing another grid data. Notice that this cube file must have
identical grid setting as the first cube file
Now the grid data in the memory has been updated, choose 0 to export it as a new cube file,
which is what we need.
Example 2
Assume that we have two cube files MO1.cub and MO2.cub, each of them records
567
wavefunction value of an orbital. In this example we will generate a cube file containing total
electron density deriving from these two orbitals. According to Born's probability interpretation,
square of an orbital wavefunction value is simply its density probability, therefore what we need is
the sum of square of the two grid data.
MO1.cub
10 // Perform A2+B2=C operation, where A is present grid data (MO1.cub), B is another cube
file (MO2.cub), C is the new grid data
MO2.cub // Load another cube file
After calculation, the grid data in memory has been updated to C.
0 // Output the updated grid data
totdes.cub // Filename of the new cube file, which contains total electron density of the two
orbitals
4.13.3 Scaling numerical range of grid data
The numerical range of ELF function is [0,1], in this example, we scale its numerical range to
[0,65535] (which is value range of unsigned 16bit integer). We first compute ELF grid data in
Multiwfn as described in Section 4.5.1, and then input
0 // Return to main menu from post-processing interface of grid data calculation
16 // Scale data range
0,1 // Original data range
0,65535 // The range after scaling. Please read Section 3.16.12 for the detail of scaling
algorithm.
Now the grid data has been scaled. You can choose function 0 to export the updated grid data
to Gaussian cube file, or extract plane data to plain text file by corresponding functions.
4.13.4 Screen isosurfaces in local regions
Sometimes we do not want all isosurfaces in the whole space are shown, because too many
isosurfaces will confuse our eyes. This section I will show how to screen the isosurfaces of not
interest
4.13.4.1 Screen isosurfaces inside or outside a region
This section I take electron density of phenol dimer as example. First we generate the grid data
as follows (you can also directly load a .cub/.grd file and then enter main function 13)
examples\phenoldimer.wfn
-1 // Visualize isosurface
568
As you can see, the isosurfaces appear on both phenol molecules.
Assume that we only want the isosurface around the right phenol will be shown, we need to set the
value of the grid points that close to the left phenol to a very small value, for example, zero.
Close the Multiwfn GUI window and input
13 // Set value of the grid points that far away from / close to some atoms
-0.7 // That means we will set the value of the grid points inside 0.7 times of vdW radius of
the atoms. If input 0.7, then the value of the grid points outside 0.7 times of the vdW radius will be
set
0 // Set the value to 0
2 // Defining mode. 2 means inputting atomic indices by hand (if choose 1, external file
containing atomic index list will be used to define fragment, see Section 3.16.9 for the format or the
next example)
1-13 // Range of atomic indices of the left phenol
Now the grid data has been updated, let us choose option -2 to visualize the isosurface of current
grid data. As you can see, the isosurface of the left phenol has disappeared.
4.13.4.2 Screen isosurfaces outside overlap region of two fragments

During analysis of inter-molecular interaction by NCI method (Section 4.20.1), what we want
to study is only the isosurfaces in inter-molecular regions. In order to screen isosurfaces in other
regions, we can set the value of grid points outside superposition region of scaled vdW regions of
two molecules to a very large value (at least larger than maximum value in current grid data). In this
section I give you a practical example.
The so-called "scaled vdW regions" is the superposition region of scaled vdW spheres of all
atoms in the fragment. While the "scaled vdW sphere" denotes the sphere corresponding to the
scaled vdW radii.
Below is a segment of dimeric protein plotted by VMD program (you will know how to draw
a similar picture after reading Section 4.20.2), red and blue representing backbone structure of the
two chains respectively. Isosurfaces of reduced density gradient exhibit weak interaction region.
However, these isosurfaces include both intermolecular and intramolecular parts, they are
569
interwinded and result in difficulty in visual study of weak interaction between the two chains.
In order to screen those intramolecular isosurfaces, we will use subfunction 14 in main function
13 of Multiwfn. First, we prepare two atom list files for the two chains (each chain corresponds to
a fragment). atmlist1.txt includes atom indices of chain 1, the head and tail of the file are:
159 <--- Total number of atoms in chain 1
1 <--- Atom index of the first atom in chain 1
2 <--- Atom index of the second atom in chain 1
...
159 <--- Atom index of the last atom in chain 1
Similarly, atmlist2.txt defines atom list for chain 2, its head and tail parts are:
159 <--- Chain 2 has 159 atoms too
160 <--- Atom index of the first atom in chain 2
161 <--- Atom index of the second atom in chain 2
...
318 <--- Atom index of the last atom in chain 2
Then boot up Multiwfn and input:

RDG.cub // The cube file of reduced density gradient corresponding to above graph
14 // Set value of the grid points outside overlap region of the scaled vdW regions of the two
fragments
1.8 // The value for scaling vdW radius. In your practical studies, you may need to try this
value many times to find a proper value
1000 // Set value of those grid points to 1000, this value is large enough
1 // Defining mode, 1 means using external file to define the fragment
570
atmlist1.txt // The name of the atom list file for chain 1

atmlist2.txt // The name of the atom list file for chain 2
Wait for a while, the grid data will be updated. Then choose function 0 to export it as cube file.
Using this new cube file to redraw above picture, we find all of intramolecular isosurfaces have
disappeared, the graph becomes very clear.
In fact, when the case is not complicated (as present example), preparing atomic list files are
not needed, you can choose defining mode as 2 and then directly input atomic indices (i.e. 1-159 for
chain 1 and 160-318 for chain 2).
4.13.5 Acquire barycenter of a molecular orbital
In this example, we will calculate barycenter of a molecular orbital. You can also obtain
barycenter of other real space functions by similar manner. The definition of barycenter is given in
Section 3.16.13. Below is the isosurface of the 10th MO of phenol.
571
Before calculate the barycenter of the MO, we need to obtain the grid data of the MO. We can
do this in Multiwfn, namely boot up Multiwfn and input following commands:
examples/phenol.wfn
4 // Choose orbital wavefunction
10 // The 10th orbital
2 // Medium quality grid. Finer quality of grid will give more accurate barycenter positions
0 // Return back to main menu
Now the grid data has been stored in memory, we will analyze it now
17 // Show statistic data
1 // Select all points
From the output, we can find that the X, Y, Z components of barycenter of the positive part of
the MO are (in Bohr) 2.629, -0.408, 0.000 respectively, while that of the negative part are -2.603,
-0.702, 0.000. The total barycenter is meaningless currently, since total integral value is zero for this
MO. However, total barycenter of absolute value of the MO is useful, especially for macromolecules,
from this we can understand where the MO is mainly located. In order to do this, we input:
13 // Get absolute value
17 // Show statistic data
1 // Select all points
We find X, Y, Z of total barycenter of the MO are -0.043, -0.558, 0.000 Bohr, respectively.
Since there is no negative region now, barycenter of negative part is shown as NaN (Not a Number).
4.13.6 Plot charge displacement curve
Multiwfn is able to calculate and plot integral curve for grid data, see the introduction in
Section 3.16.14. If the grid data is selected as electron density difference, then the integral curve is
commonly known as charge displacement curve (CDC), by which the charge transfer can be studied
visually and quantitatively, extremely suitable for linear systems. In this example, by means of CDC,
we will investigate the intermolecular charge transfer in polyyne (n=7) due to the externally applied
electric field of 0.03 a.u. along the molecular axis.
The polyyne.wfn and polyyne_field.wfn files in "example" folder correspond to the polyyne in
its isolated state and in the case that external electric field of 0.03 a.u. is applied, respectively.
B3LYP/6-31G* is used in the calculations, and the geometry optimized in isolated state is used for
both cases. In Gaussian program, the field can be activated via keyword field=z+300.
Before plotting the CDC, we must calculate the grid data of electron density difference between
these two files first. Boot up Multiwfn and input following commands:
examples\polyyne_field.wfn
1
-,examples\polyyne.wfn //Subtract the propery of polyyne.wfn from polyyne_field.wfn
572

We first visualize the isosurface of the electron density difference. After select -1 and set the
isovalue to 0.004, we will see the graph like below
The green and blue parts represent the regions where electron density is increased and decreased
after the external electric field is applied, respectively. It can be seen that although green and blue
parts interlace with each other, the total trend is that electron transferred to positive side of Z-axis
(namely toward the source of the electric field). Next we will plot CDC, which is able to characterize
the electron transfer in different regions quantitatively.
Click "Return" button in the GUI and then input
18 // Calculate and plot integral curve
Z // The curve will be plotted in Z direction
a // Select the entire range
In the menu, we first choose 2 to plot local integral curve of the grid data of the electron density
difference. You will see
From the graph we can examine the integral of electron density difference in the XY planes
corresponding to different Z coordinates. The Z coordinate and the value correspond to X and Y
axes of the graph. You can directly compare this curve with the isosurface graph shown above, the
peaks lower and higher than zero (dashed line) correspond to the blue and green isosurfaces.
Clicking right mouse button on the graph to close it, and then select option 1, the CDC will be
573
shown immediately
This graph is yielded by integrating the curve shown in the last graph along the molecular axis. In
its left part, although there are some fluctuations, the CDC gradually becomes more and more
negative and reaches minimum value of 1.4 in the midpoint of the X-axis (corresponding to the
center of the polyyne), that means due to the external electric field, the number of lost electrons in
the left part of polyyne is 1.4. In the right part of the graph, the CDC increases gradually from -1.4
and finally reaches zero, suggesting that 1.4 electrons are transferred to right part of the polyyne,
and due to the amount of increase and decrease of electron are cancelled with each other exactly in
the whole molecular space, there is no variation of the total number of electrons (in other words,
integral of the electron density difference in the whole molecular space is exactly zero).
4.13.7 Evaluation of electron density overlap
In this example we take methane dimer as example to evaluate where electron density of the
two monomers overlap with each other evidently. The structure of the methane dimer is
We first obtain grid data for an arbitrary real space function for the dimer. Boot up Multiwfn
and input:
examples\rho_overlap\dimer.pdb
100 // User define function, by default this function does not take any computational time
574
-10 // Set grid extension distance

2 // Decrease the distance to 2 Bohr to avoid waste of grid at system boundray
2 // Export grid data as userfunc.cub
Now we calculate wavefunction file by Gaussian for the two monomers, the input files are
monomer1.gjf and monomer2.gjf in examples\rho_overlap folder. Finally we obtained
monomer1.wfn and monomer2.wfn. Beware that the nosymm keyword must be used in Gaussian
calculation to avoid the automatic reorientation and translation.
Now we calculate electron density grid for monomer 1 over the whole space
monomer1.wfn
8
userfunc.cub // Use this cube file to define the grid, which correpsonds to the whole space
2 // Export grid data
Then rename the resulting density.cub to density1.cub. Repeat above steps for monomer2 to obtain
density2.cub.
Now we calculate grid data of min(rho(1),rho(2)), namely take the minimal value of the two
sets of electron density everywhere. Boot up Multiwfn and input:
density1.cub
11 // Mathematical operation on grid data
21 // Take min(rho(1),rho(2))
density2.cub
0 // Export resulting grid data
overlap.cub
we open the overlap.cub and density2.cub by text editor, copy atomic coordinates from the
latter to the former, and meantime change the number of atoms. Then the head of the overlap.cub
should look like below (the highlighted texts are modified parts)
Generated by Multiwfn
Totally 531846 grid points
10 -3.693194 -3.955866 -7.444301
63 0.119334 0.000000 0.000000
67 0.000000 0.119334 0.000000
126 0.000000 0.000000 0.119334
6 6.000000 0.000000 0.000000 3.371271
1 1.000000 0.000000 0.000000 5.444301
1 1.000000 0.000000 1.955867 2.677742
1 1.000000 -1.693195 -0.976988 2.677742
1 1.000000 1.693195 -0.976988 2.677742
6 6.000000 0.000000 0.000000 -3.371271
1 1.000000 1.693195 0.976988 -2.677742
1 1.000000 0.000000 0.000000 -5.444301
1 1.000000 -1.693195 0.976988 -2.677742
575
1 1.000000 0.000000 -1.955867 -2.677742
Boot up Multiwfn and load the overlap.cub, enter main function 0, change the isovalue to a
small one such as 0.0005, you will clearly see the density overlap region:
It is noteworthy that if then you enter main function 13, select subfunction 17 and then input
1, you will be able to obtain integral value of the density overlap function over the whole space,
namely the "Integral of all data". Clearly, the larger the integral, the higher extent the densities of
monomers overlap with each other.
By the way, if you feel the above mentioned steps are too lengthy, you can easily make use of
silent mode of Multiwfn to signficantly reduce the operation steps, see Section 5.2.
4.14 Adaptive natural density partitioning (AdNDP)

analysis
Theory basis of AdNDP analysis has been introduced in Section 3.17.1, please read it first.
Below I will show how to use AdNDP approach to study multi-centers orbitals of a frew practical
molecules. More detailed discussions about AdNDP analysis can be found in my blog article "Study
multi-center bonds by AdNDP approach as well as ELF/LOL and multi-center bond order" (in
Chinese, http://sobereva.com/138).
NOTICE: Using diffuse functions in AdNDP analysis is strongly deprecated, because they often cause
numerical problems (which sometimes leads to crash when Multiwfn loading input file) while never improve AdNDP
results!
4.14.1 Analyze Li5+ cluster
In Chem. Eur. J., 6, 2982 (2000), the authors showed that Li5+ cluster has two 4-centers 2-
electrons (4c-2e) bonds by examining ELF isosurfaces. In present example, we will use AdNDP
approach to study this cluster to verify their statement. We first optimize Li5+ cluster under
B3LYP/6-311G* level and then compile an input file of single point task for Gaussian. pop=nboread
keyword must be specified in route section, and $NBO AONAO DMNAO $END must be added to
the end of the input file. Run this file by Gaussian, and then convert check point file to .fch format.
The input file, output file and .fch file have been given in "examples\AdNDP" folder.
576
Boot up Multiwfn and input examples\AdNDP\Li5+.out, and then choose main function 14.
After Multiwfn loaded some necessary information, a menu appears. Since this cluster is small, we
can directly use exhaustive manner to search all possible 1c-2e, 2c-2e, 3c-2e, 4c-2e and 5c-2e
orbitals in turn. We first choose option 2 to search 1c-2e orbitals (namely lone pairs), however,
because occupation numbers of all tried 1c orbitals are lower than default threshold (which is a value
close to 2.0 and can be adjusted by option 4), the candidate orbitals list shown in front of the menu
is still empty. We then choose option 2 twice to search 2c-2e and 3c-2e orbitals in turn, we still
cannot find any orbital with high occupation numbers. Next we select option 2 again to search 4c-
2e orbitals, this time the candidate orbital list is no longer empty, there are two orbitals in it:
# 2 Occ: 1.9966 Atom: 1Li 2Li 3Li 4Li
# 1 Occ: 1.9966 Atom: 1Li 2Li 3Li 5Li
Due to their high occupation number, it clear that they are ideal 4c-2e orbitals, therefore we decide
to choose option 0 and input 2 to pick them out from candidate list and save as AdNDP orbitals.
Whereafter the list of AdNDP orbitals can be printed by option 5.
You may have noticed that the number of residual valence electrons (shown at the top of the
menu) has been updated to 0.020, which is already very close to zero, it is suggested it is
meaningless to continue to search 5c-2e orbitals because they would be impossible to be found.
Now you can choose option 7 to visualize the two 4c-2e AdNDP orbitals. In order to calculate
orbital wavefunction, Multiwfn needs to load basis set information from corresponding .fch file first.
Since Li5+.fch is in the same folder and has identical name as Li5+.out, the .fch file will be directly
loaded. When loading is finished, a GUI pops up, which is completely identical to the one of main
function 0. AdNDP orbitals can be plotted by selecting corresponding numbers in the right-bottom
list. The 0.05 isosurfaces of the two orbitals are shown below.
Grid data of AdNDP orbitals can be exported as Gaussian cube files by option 9, so that you
can plot them by some third-part visualization programs such as VMD and Molekel. You need to
input orbital index range, assume that we want to output the two 4c-2e AdNDP orbitals we just
found, we should input 1,2 and choose a proper grid setting, then they will be exported as
AdNDPorb0001.cub and AdNDPorb0002.cub in current folder.
By option 3, you can set the number of centers of multi-centers orbitals in the next exhaustive
search. So, assume that you have already known that there are two 4c-2e orbitals in present system,
you can directly choose option 3, input 4 and then select option 2 to start the exhaustive search of
4c-2e orbitals, the exhaustive searchs of 1c-2e, 2c-2e and 3c-2e orbitals will be skipped.
577
Evaluating AdNDP orbital energies

It is also possible to obtain energies of the AdNDP orbitals that have been picked out. To realize
this, you need to provide additional plain text file containing Fock matrix of present system in lower-
triangular sequence, so that orbital energies can be yielded after some transformations of this matrix.
The Fock matrix can also be loaded from .47 file. The most straightforward procedure is as follows:
Copy examples\AdNDP\Li5+.gjf as examples\AdNDP\Li5+_47.gjf, change content between
$NBO ... $END to archive file=C:\Li5+. Then after running this file, C:\Li5+.47 will be yielded,
which is input file of GENNBO program and contains the Fock matrix that we need. Then we choose
"16 Output energy of picked AdNDP orbital" in the AdNDP analysis interface, input the path of the
Li5+.47 (which has already been provided in "examples\AdNDP\" folder), Multiwfn will load it
and immediately print out AdNDP orbital energies, as shown below:
Energy of picked AdNDP orbitals:
Orbital: 1 Energy (a.u./eV): -0.325776 -8.8648
As expected, the two orbitals are degenerate in energy, since they have exactly equivalent shape.
4.14.2 Analyze B11− cluster
This time, we will try to reproduce the AdNDP analysis result of B11- cluster that given in
AdNDP original paper (Phys. Chem. Chem. Phys., 10, 5207 (2008)).
The files needed by this instance, namely B11-.out and B11-.fch can be found in
"examples\AdNDP" folder. The geometry was optimized under B3LYP/6-311+G*, while the
wavefunction was generated under HF/STO-3G level. You may wonder whether the result is
meaningful under such low level of basis set; actually, AdNDP analysis is rather insensitive to basis
set quality, even STO-3G is able to produce at least qualitative resonable result. In addition, using
larger basis set will bring additional cost at AdNDP analysis stage.
Boot up Multiwfn and input examples\AdNDP\B11-.out, then choose 14 to enter AdNDP
module. As usual, we select 2 to search 1c-2e orbitals first, but we find nothing (this is common
case). Then select 2 again to exhaustively search 2c-2e orbitals from the 11 atoms, Multiwfn will
totally try 11!/(11-2)!/2!=55 combinations, finally there are nine 2c orbitals present in candidate list
(sorted according to occupation number from large to small):
# 9 Occ: 1.9727 Atom: 6B 10B
# 8 Occ: 1.9727 Atom: 5B 11B
# 7 Occ: 1.9742 Atom: 7B 9B
# 6 Occ: 1.9742 Atom: 7B 8B
# 5 Occ: 1.9869 Atom: 2B 6B
# 4 Occ: 1.9869 Atom: 3B 5B
# 3 Occ: 1.9871 Atom: 9B 11B
# 2 Occ: 1.9871 Atom: 8B 10B
# 1 Occ: 1.9942 Atom: 2B 3B
All of them have occupation number close to 2.0, ostensibly one can directly pick all of them out as
AdNDP orbitals, however, it is not recommended to do so, because neighboring orbitals may share
the same densities. For example, the 1th and the 4th candidate orbitals share the some densities,
578
since both of them related to atom 3. In order to avoid overcount of electrons, firstly you should
pick out the first three orbitals by choosing option 0 and input 3, then the density of the first three
orbitals will be depleted from density matrix, after that wavefunction and occupation number of
remained candidate orbitals will be updated automatically. After that the candidate list becomes
# 6 Occ: 1.9538 Atom: 6B 10B
# 5 Occ: 1.9538 Atom: 5B 11B
# 4 Occ: 1.9556 Atom: 7B 8B
# 3 Occ: 1.9556 Atom: 7B 9B
# 2 Occ: 1.9750 Atom: 2B 6B
# 1 Occ: 1.9750 Atom: 3B 5B
Since some densities have been depleted, occupation number of remained six candidate orbitals
slightly decreased. Now, we pick out the first four candidate orbitals by choosing option 0 and input
4. Although both of 3th and 4th orbitals are related to atom 7, here we have to ignore the slight
overcount of electrons, otherwise their degeneration will be broken and thus the final AdNDP
pattern will not be consistent with molecular symmetry anymore (you can choose option 8 to
carefully inspect candidate orbitals before you decide to pick them out). Finally, we pick out the last
two orbitals (i.e. 6B-10B and 5B-11B). Currently the number of residual valence electrons is 16.307,
which reveals that it is probable to find several higher number of centers orbitals with nearly two
electrons occupied.
Now we select option 2 to start the search of 3c-2e orbitals, the current candidate orbital list is:
# 9 Occ: 1.7399 Atom: 1B 6B 10B
# 8 Occ: 1.7399 Atom: 4B 5B 11B
# 7 Occ: 1.7502 Atom: 1B 3B 4B
# 6 Occ: 1.7502 Atom: 1B 2B 4B
# 5 Occ: 1.8504 Atom: 1B 2B 6B
# 4 Occ: 1.8504 Atom: 3B 4B 5B
# 3 Occ: 1.8603 Atom: 1B 4B 7B
# 2 Occ: 1.8673 Atom: 4B 9B 11B
# 1 Occ: 1.8673 Atom: 1B 8B 10B
After we pick out two orbitals (1B-8B-10B and 4B-9B-11B), one orbital (1B-4B-7B) and two
orbitals (3B-4B-5B and 1B-2B-6B) in turn, the highest occupation number of remained candidate
orbitals is 1.41, which is obviously too low to be recognized as 3c-2e orbital, so they will not be
concerned. Currently the number of residual valence electrons is 7.03.
Then you can start to search higher number of centers orbitals, beware that this is never a trivial
task, and there is no absolute rule on how to reasonably pick out candidate orbitals, different picking
manners result in different AdNDP patterns. You may have to try many times before finally
obtaining an optimal AdNDP pattern. It is recommended to use option 11 to save present density
matrix and AdNDP orbital list into memory, so that you need not to worry about improper pick of
candidate orbitals, since saved state can be recovered anytime by choosing option 12.
Now choose option 2 to start the search of 4c-2e orbitals, the highest occupation is merely 1.71,
none of them could be picked out.
Select option 2 again to search 5c-2e orbitals, you will find many 5c candidate orbitals, the
first two have occupation numbers of 1.89, we pick out both of them.
Then choose option 2 to search 6c-2e orbitals, no good candidate can be found, the highest
579
occupation number is only 1.84. Then choose option 2 to search 7c-2e orbitals, we pick the highest
occupied one (1.90). Now the residual valence electron is only 1.34, which is much less than 2.0,
indicating that no additional 2e AdNDP orbital could be found, therefore now we can end the
AdNDP searching procedure. The amount of residual electron reflects the electrons that cannot be
fully represented by present AdNDP pattern (analogous to non-Lewis electron in the NBO
framework)
By choosing option 5, information of all AdNDP orbitals can be printed out:
# 1 Occ: 1.9942 Atom: 2B 3B
# 2 Occ: 1.9871 Atom: 8B 10B
# 3 Occ: 1.9871 Atom: 9B 11B
# 4 Occ: 1.9750 Atom: 2B 6B
# 5 Occ: 1.9750 Atom: 3B 5B
# 6 Occ: 1.9556 Atom: 7B 9B
# 7 Occ: 1.9556 Atom: 7B 8B
# 8 Occ: 1.9337 Atom: 6B 10B
# 9 Occ: 1.9337 Atom: 5B 11B
# 10 Occ: 1.8673 Atom: 4B 9B 11B
# 11 Occ: 1.8673 Atom: 1B 8B 10B
# 12 Occ: 1.8533 Atom: 1B 4B 7B
# 13 Occ: 1.8451 Atom: 3B 4B 5B
# 14 Occ: 1.8451 Atom: 1B 2B 6B
# 15 Occ: 1.8908 Atom: 1B 2B 6B 8B 10B
# 16 Occ: 1.8908 Atom: 3B 4B 5B 9B 11B
# 17 Occ: 1.9036 Atom: 2B 3B 5B 6B 7B 8B 9B
Total occupation number in above orbitals: 32.6607
Plotting a batch of AdNDP orbitals simultaneously by VMD script

Now you can use option 7 to visualize all picked AdNDP orbitals. However this time I show
how to plot AdNDP orbitals via VMD, which could simultaneously plot a batch of orbitals and the
graphical quality is good. VMD can be freely obtained at http://www.ks.uiuc.edu/Research/vmd/.
Choose option 9, select "Medium quality grid" and then input 1-17 to export all the 17 AdNDP
orbitals as cube files in current folder, the format of the file name is AdNDPorb[index].cub. Assume
that you have moved all of them to C:\ directory, you should edit examples\AdNDP\plotAdNDP.vmd
and change this line
set name "D:\\CM\\my_program\\Multiwfn\\AdNDPorb$idx.cub"
to
set name "C:\\AdNDPorb$idx.cub"
You also need to make sure that in the script, the values after "set istart" and "set iend" have been
set to 1 and 17, respectively, so that the cube files from AdNDPorb0001.cub to AdNDPorb0017.cub
will be loaded. The positive and negative phases of the orbital isosurfaces are determined by the
values after "set posclr" and "set negclr 0", the orbital isovalue is determined by "set isoval"
Now boot up VMD, copy all content in the plotAdNDP.vmd to VMD console window, all cube
files of AdNDP orbitals will be loaded into VMD. Now the VMD Main window looks like below
580
Each entry corresponds to an AdNDP orbital. Currently all the 17 orbitals are shown. If you double
click a "D" label, then corresponding orbital will be hidden in the graphical window. In order to
show the molecule structure, drag the examples\AdNDP\B11-.xyz into the VMD main window to
load it, then enter "Graphics" - "Representation" and change the drawing style as CPK.
If you make VMD only display all the nine 2c-2e and all the five 3c-2e orbitals, you will see
left and right parts of below graph, respectively
The two 5c-2e and one 7c-2e orbitals are shown below (In the graph the 7c-2e orbital looks
like 5c orbital, the main reason is that the isovalue in the plotting script is relatively high, i.e. 0.06).
581
4.14.3 Analyze phenanthrene
AdNDP analysis of phenanthrene (C14H10, see above) has been given in J. Org. Chem., 73,
9251 (2008), in this section we will repeat their result, you will learn how to use user-directed search.
Files used in this example can be found in examples\AdNDP folder with "phenanthrene" prefix.
First we load examples\AdNDP\phenanthrene.out and enter main function 14. In consistency
with the previous examples, we select option 2 twice to search 1c orbitals and then search 2c orbitals.
No 1c-2e orbitals can be found, while there are 31 candidate 2c orbitals present in the list. Ten of
them correspond to C-H σ-bonds and have no overlapping with each other, so we can pick them out
first, namely choosing option 0, input 8,15, then choose option 0 again and input 9,10. Next, we
successively pick out sixteen 2c candidate orbitals that corresponding to C-C σ-bonds (the most
careful input is 0 2 0 1 0 2 0 1 0 2 0 2 0 2 0 2 0 2, where space denotes pressing ENTER button
once).
Now there are only five orbitals remain:
# 5 Occ: 1.7182 Atom: 5C 6C
# 4 Occ: 1.7182 Atom: 14C 15C
# 3 Occ: 1.7192 Atom: 11C 13C
# 2 Occ: 1.7192 Atom: 1C 2C
# 1 Occ: 1.8033 Atom: 7C 10C
The first orbital with occupation number of 1.80 corresponds to the π-bond between C7 and C10,
we pick it out. The occupation numbers of the four remained orbitals are about 1.72, thus they are
not quite ideal 2c-2e bonds, we do not concern them at the moment.
Although we can use option 2 to exhaustively search 3c, 4c, 5c ... orbitals as usual, this may
be not a good idea for present system, since user-directed search is often more effective. We first
choose option 13 to check population of residual electrons on each atom, see below, this information
is usually helpful for guiding users to properly set up exhaustive search list. (Note: The exhaustive
search triggered by option 2 is only applied to the atoms in exhaustive search list, which contains
all atoms in present system by default)
1C : 1.0250 2C : 1.0370 3C : 1.0280 4C : 1.0414
5C : 1.0339 6C : 1.0262 7C : 0.1322 8C : 1.0414
9C : 1.0280 10C : 0.1322 11C : 1.0370 12H : 0.0117
13C : 1.0250 14C : 1.0262 15C : 1.0339 16H : 0.0121
17H : 0.0113 18H : 0.0117 19H : 0.0126 20H : 0.0111
21H : 0.0121 22H : 0.0113 23H : 0.0111 24H : 0.0126
From above data it is clear that hydrogens have almost vanished population, hence they can be
ignored during seach. Due to the same reason C7 and C10 can be ignored too. The other atoms,
whose occupation numbers are about 1.03, are the carbons composing the two 6-member rings in
582
both sides of the molecule. It can be expected that the two rings may be analogous to benzene ring
and hence representing local aromaticity in phenanthrene.
Based on this consideration, we select option 1 and input 1-6 to search AdNDP orbitals for the
fragment consisted of atoms 1, 2, 3, 4, 5, 6, we find
...ignored
# 5 Occ: 0.2157 Atom: 1C 2C 3C 4C 5C 6C
# 4 Occ: 0.2998 Atom: 1C 2C 3C 4C 5C 6C
# 3 Occ: 1.8214 Atom: 1C 2C 3C 4C 5C 6C
# 2 Occ: 1.9850 Atom: 1C 2C 3C 4C 5C 6C
# 1 Occ: 2.0000 Atom: 1C 2C 3C 4C 5C 6C
Evidently the first three are appropriate to be picked out as 6c-2e AdNDP orbitals, so we pick them
out now. Their 0.03 isosurfaces are shown below, which look very like π molecular orbitals of
benzene, implying that the boundary 6c ring has strong aromaticity as benzene.
Next, via the same way we search 6c-2e orbitals over another boundary ring, namely choose
option 1 again and input 8,9,11,13-15, after that we pick out three highest occupied orbitals.
Now the residual valence electrons is only 1.15, which is already very small, clearly the
AdNDP search should end here. Finally, totally 33 orbitals (27*2c-2e, 6*6c-2e) are found.
Note: When searching 6c-2e orbitals over the ring consisting of atoms 1~6, in fact there is another way to do
this (though more cumbersome), namely choose option -1 to enter the interface for defining exhaustive search list,
input clean to clean up the default content, then input a 1-6 to add ring atoms 1, 2, 3, 4, 5, 6 into the list, then input
x to save and exit. After that, select option 3 and input 6 to set the number of atoms in the next exhaustive search as
six, then choose option 2 to search 6-centers orbitals over the ring.
Evaluating AdNDP orbital energies

With similar procedure, we evaluate AdNDP orbital energies like Section 4.14.1. The NBO .47
file containing Fock matrix of current molecule has been provided as
examples\AdNDP\phenanthrene.47, which was yielded by examples\AdNDP\phenanthrene_47.gjf.
We choose option 16 and input the path of this file, Multiwfn immediately loads Fock matrix from
it and outputs the orbital energies:
...(ignored)
As you can see, the orbitals 28~33, which correspond to  orbitals, have energy much higher
583
than  orbitals. The three  orbitals (28~30) in the left six-membered ring are symmetric to the
counterpart orbitals (31~33) in the right six-membered ring. In each side, the two highest lying
orbitals (e.g. 32 and 33) are nearly degenerate and evidently higher than the lowest lying one (e.g.
31), this situation is very similar to occupied  orbitals of isolated benzene.
Evaluating composition of AdNDP orbitals

Sometimes composition of AdNDP orbitals is interesting. In the AdNDP module we can
directly choose option 15 to analyze orbital composition by NAO method, which as been introduced
in Section 3.10.4. NAO method is particularly suitable for analyzing AdNDP orbitals.
Choose option 15, then input index of a picked AdNDP orbital, for example 31, you will see
(by default only terms whose absolute contribution > 0.5 % are shown)
NAO# Center Label Type Composition
67 8(C ) px Val( 2p) 2.247%
76 9(C ) px Val( 2p) 1.929%
94 11(C ) px Val( 2p) 13.276%
105 13(C ) px Val( 2p) 32.846%
114 14(C ) px Val( 2p) 34.484%
123 15(C ) px Val( 2p) 15.204%
Condensed NAO terms to shells:
Atom: 8(C ) Shell: 39( 2p Val) 2.247%
Atom: 9(C ) Shell: 44( 2p Val) 1.929%
Atom: 11(C ) Shell: 54( 2p Val) 13.276%
Atom: 13(C ) Shell: 61( 2p Val) 32.846%
Atom: 14(C ) Shell: 66( 2p Val) 34.484%
Atom: 15(C ) Shell: 71( 2p Val) 15.204%
Condensed NAO terms to atoms:
Center Composition
8(C ) 2.251%
9(C ) 1.932%
11(C ) 13.276%
13(C ) 32.849%
14(C ) 34.487%
15(C ) 15.204%
As expected, this  type of 6c-2e orbital purely compose of px natural atomic orbitals, whose
axis is perpendicular to the plane of the phenanthrene. This orbital is delocalized over the ring, but
mostly distributed on atoms 13 and 14.
Note that there is another way of evaluating AdNDP orbital composition, namely exporting
AdNDP orbitals as AdNDP.molden in current folder by option 14, and then use this file as input file
of Multiwfn and perform orbital composition as usual (via e.g. Mulliken analysis, Hirshfeld analysis
and so on, see Section 4.8 for example). For present example, this .molden file contains 146 orbitals
because there are originally 146 natural atomic orbitals, however only the first 33 orbitals
correspond to AdNDP orbitals and thus meaningful.
584
4.14.4 Analyze Au20 cluster
In this section we perform AdNDP analysis for Au20 cluster, the needed files can be
downloaded from http://sobereva.com/multiwfn/extrafiles/Au20.rar.
Au20.out // Generated at B3PW91/Lanl2DZ level based on optimized geometry
14 // AdNDP analysis
2 // Search 1-center AdNDP orbitals. 100 candidates are found, whose occupation numbers
are very close to 2.0 and thus can be picked out
0 // Pick out orbitals
100 // Pick out all 100 candidate orbitals
2 // Perform exhaustive search of 2-centers orbitals. Nothing can be found
2 // Perform exhaustive search of 3-centers orbitals. Again nothing can be found
2 // Perform exhaustive search of 4-centers orbitals. Now you can see four candidates with
1.84 e and six candidates with 1.7589 e
4 // Pick out first four orbitals. The remaining orbitals now have occupancy of 1.6913, which,
although is not quite high, it is still worth to be picked out in current circumstance
6 // Pick out remaining six orbitals.
Plotting AdNDP orbitals

Next we plot all the ten picked 4c-orbitals by VMD using plotting script. We first export their
cube files, input below commands
9 // Export cube file of picked AdNDP orbitals
2 // Medium quality grid, which is adequate for producing smooth orbital isosurface for
present system
101-110 // The index range of the picked 4c-orbitals
After a while, we have ten cube files in current folder, the first one is AdNDPorb0101.cub, the
last one is AdNDPorb0110.cub. We intend to plot the orbitals 101~104 (occ=1.84) using red color
while 105~110 (occ=1.69) using orange color so that they can be clearly distinguished.
We edit examples\AdNDP\plotAdNDP.vmd, set "istart" and "iend" to 101 and 104, respectively.
Then modify the path after "set name" to make it correspond to the actual path containing the cube
files. In addition, add # symbol in the front of the two "mol modmaterial..." lines to comment them
out so that the isosurface will be drawn as opaque. After that we boot up VMD, copy all content
from the VMD script to the VMD console window, then you will find the first four 4c-orbitals have
been shown as isosurfaces in the VMD graphical window.
We also need to plot the orbitals 105~110 on the graph. We set "istart" and "iend" in the script
to 105 and 110, respectively. Set "posclr" to 3 to make isosurface color of positive phase to be orange
(note that these 4c-orbitals only have positive phase). Set "idinit" to 4, because the systems showing
orbital 101, 102, 103 and 104 have already been assigned to be ID=0, ID=1, ID=2 and ID=3 in
VMD, respectively, therefore the ID corresponding to the next orbital to be loaded must be 4. Then
copy the content of the script to VMD console window to run them, you will find the remaining six
4c-orbitals have also been shown.
Finally, choose "Graphics" - "Representations", click "Create Rep", change "Drawing method"
585
to "CPK". Now you should see below graph, each red and orange isosurface represents 4c-2e orbital
at each vertex and edge of the cluster, respectively.
4.15 Fuzzy atomic space analysis
Delocalization index, PDI, FLU and FLU-π were originally proposed for AIM atomic space, it
has been shown that, if they are calculated in fuzzy atomic space, the computational cost would be
significantly reduced, while the results are still reasonable. The introduction of basic concepts of
fuzzy atomic space has been given in Section 3.18.0. In this section, we will calculate these
quantities in the fuzzy atomic space defined by Becke. Multiwfn also supports Hirshfeld and
Hirshfeld-I methods to defined atomic space, however, the Becke's definition has the advantage that
the calculation is independent on reference atomic densities, so we do not need to concern the
preparation of .wfn files for atoms in their free-states, and therefore the calculation procedure is
somewhat simpler.
4.15.1 Study delocalization index of benzene
The definition of delocalization index (DI) has been detailedly introduced in Section 3.18.5. In
present instance we will calculate DI in Becke's fuzzy atomic space to study the extent of electron
delocalization between different atom pairs in benzene. Notice that the original DI is defined in
Bader's AIM atomic space, so the DIs calculated in this example may somewhat differ to the ones
in some literatures.
Boot up Multiwfn, and input following commands
examples\benzene.wfn // Generated under B3LYP/6-311G*
15 // Fuzzy atomic space analysis
4 // Calculate localization index (LI) and DI. Multiwfn first makes use of DFT numerical
quadrature scheme to calculate atomic overlap matrix (AOM) in each fuzzy atomic space, and then
convert AOM to DI and LI.
586
n // Do not output LI and DI to plain text file

Multiwfn automatically checks and output the error of AOM, for present calculation the error
is less than 0.001, which is completely negligible. If the error is too larger to be accepted, you can
set "iautointgrid" in settings.ini to 0, and set "radpot" and "sphpot" to a large value. When
"iautointgrid" is equal to 1, Multiwfn uses (40,230) grid points to calculate AOM, accuracy of which
directly affect the accuracy of LI, DI, as well as of PDI and FLU.
From the DI matrix outputted on the screen, we can see that DI between adjacent two carbon
atoms and adjacent C-H atoms are large (1.467 and 0.877, respectively), that means electron
delocalization between bonded atoms is strong, this is mainly due to the shared electron of σ bond.
In contrast, the DI between non-bonded carbon atoms is very small, about 0.1, nevertheless
evidently not zero, reflecting the high-degree delocalization nature of π electrons.
The DI calculated in fuzzy atomic space is essentially the fuzzy bond order proposed by Mayer.
According the DI data, we can say that the bond order between C-C bond and C-H bond in benzene
is 1.467 and 0.877 respectively, the former corresponds to single σ bond + "semi" π bond, while the
latter corresponds to typical single σ bond.
Since benzene is an exactly planar molecule, we can decompose DI to DI-σ and DI-π. Here we
calculate the latter. Input 0 to return to main menu, and then following commands
6 //Modify wavefunction
26 //Modify occupation number
0 // Select all orbitals
0 // Make occupation number of all orbitals to zero
17,20,21 // MO 17,20,21 correspond to π orbitals.
2 // Set their occupation numbers to 2 (closed-shell orbitals)
q // Return to upper level of menu
-1 //Return to main menu
Now recalculate DI as before, since the occupation numbers of all orbitals except π orbitals
have been set to zero, the result will be DI-π.
The DI-π between C1-C6, C1-C5 and C1-C4 are 0.438, 0.055 and 0.093 respectively, it is
obvious that π electron delocalization is larger for para-related than for meta-related carbon atoms.
4.15.2 Study aromaticity of phenanthrene by PDI, FLU, FLU-π and

PLR
PDI, FLU, FLU-π and PLR are very useful aromaticity indices, their definitions have been
introduced in Section 3.18.6, 3.18.7 and 3.18.9. In present instance, we will calculate them in
Becke's fuzzy atomic space to study aromaticity of different rings of phenanthrene.
587
We first calculate PDI. Boot up Multiwfn, and input following commands:

examples\phenanthrene.wfn // Optimized at B3LYP/6-31G* level
5 // Calculate PDI
Then Multiwfn starts to calculate atomic overlap matrix (AOM), this is a computationally
intensive work. After that AOM will be converted to delocalization index (DI), then DI matrix will
be outputted on screen. Finally, you will be prompted to input atom indices of the ring you are
interested in, the input order should be in consistency with atomic connectivity. We first calculate
PDI of the central ring, namely input 4,8,9,10,7,3, the result is
Delocalization index of 4(C ) -- 10(C ): 0.053069
PDI value is 0.047496
PDI value is just the average of the DIs between C4-C10, C8-C7 and C9-C3. Now we input
8,9,11,13,14,15 to calculate PDI of the boundary ring, the result is 0.0817. From this result it is
evident that the electron delocalization in the boundary rings is stronger than in the central ring, so
boundary rings possess larger aromaticity. Next, we use FLU and FLU-π to study the aromaticity,
and check if we can draw the same conclusion.
Input q to return to upper level of menu, and input 6 to calculate FLU, then input 4,8,9,10,7,3
and 8,9,11,13,14,15 in turn. The FLU of central ring and of boundary rings are 0.025311 and
0.007537 respectively, this result suggests that the boundary rings is more like to typical aromatic
system (benzene), and hence possesses larger aromaticity than the central ring. Note that since the
AOM has already been calculated during calculating PDI, so this time the calculation process of
AOM is automatically skipped.
Next, input q to return to upper level of menu, and input 7 to calculate FLU-π. First you need
to input the indices of π orbitals. By visually checking isosurface of all orbitals (or utilizing option
22 in main function 100), we know that 36, 40, 43, 44, 45, 46, and 47 are π orbitals, so here we input
36,40,43,44,45,46,47, then DI of π electrons will be outputted. After that you will be prompted to
input the atom indices in the ring, we input 4,8,9,10,7,3 and 8,9,11,13,14,15 in turn. The FLU-π of
central ring and boundary rings are 0.149288 and 0.035037, respectively. Obviously, FLU-π analysis
also confirms that boundary ring is more aromatic.
Finally, let us calculate para linear response index (PLR). PLR is based on linear response
kernel, which relies on virtual MOs information; however .wfn file only contains occupied MOs,
therefore we must use .mwfn/.fch/.molden/.gms file as input. Reboot Multiwfn and input following
commands
examples\phenanthrene.fch // Obtained at the same level as phenanthrene.wfn
15 // Fuzzy space analysis
588
10 // Calculate PLR
Multiwfn will calculate condensed linear response kernel (CLRK) matrix, after that input
4,8,9,10,7,3 and 8,9,11,13,14,15 in turn, the result is 0.247848 and 0.489434, respectively. Since the
former is much smaller than the latter, PLR also validates the conclusion that boundary rings have
larger aromaticity than central ring.
Note that both PDI and PLR can be separated as  and π parts to respectively investigate  and
π aromaticity, see Section 3.18.6 and 3.18.9 for detail.
4.15.3 Calculate fragment dipole moment to exhibit local polarity

Note: Chinese version of this topic is http://sobereva.com/558, which contains additional example and extended
discussion.
Dipole moment of a fragment may be defined as

DF =   Z AR A −  wA (r )  (r )r d r 
AF
 
where A is atomic index in the fragment F, ZA and RA are nuclear charge and position of atom A,
respectively. wA(r) is atomic weighting function of atom A.
As mentioned in Section 3.18.3, Multiwfn is able to calculate atomic and molecular
dipole/multipole moments; if you define an atom list, then the outputted molecular dipole and
multipole moments will correspond to the moments of the fragment. In this calculation, we use this
feature to calculate respective dipole moment of the two monomers in phenol dimer. We will use
Hirshfeld weighting function, since its calculation is easy and its physical meaning is relatively clear.
examples\phenoldimer.wfn // Wavefunction file of optimized phenol dimer
-1 // Select method for defining atomic space
3 // Hirshfeld based on built-in spherically averaged atomic densities
2 // Calculate atomic and molecular multipole moments
Then Multiwfn starts to calculate population number, dipole and multipole moments of every
atom, and finally prints the data for the whole system (the "molecular" in this context corresponds
to the entire current system):
***** Molecular dipole and multipole moments *****
Total number of electrons: 100.000331 Net charge: -0.000331
Molecular dipole moment (a.u.): 1.227306 -0.128087 0.650833
Molecular dipole moment (Debye): 3.119501 -0.325563 1.654252
Magnitude of molecular dipole moment (a.u.&Debye): 1.395088 3.545959
Molecular quadrupole moments (Standard Cartesian form):
XX= -56.973177 XY= 3.413050 XZ= 4.769915
YX= 3.413050 YY= -58.109370 YZ= 4.188508
ZX= 4.769915 ZY= 4.188508 ZZ= -57.714457
Molecular quadrupole moments (Traceless Cartesian form):
XX= 0.938737 XY= 5.119575 XZ= 7.154872
YX= 5.119575 YY= -0.765553 YZ= 6.282762
ZX= 7.154872 ZY= 6.282762 ZZ= -0.173184
589
Magnitude of the traceless quadrupole moment tensor: 0.999096
Molecular quadrupole moments (Spherical harmonic form):
Q_2,0 = -0.173184 Q_2,-1= 7.254708 Q_2,1= 8.261735
Q_2,-2= 5.911576 Q_2,2 = 0.983972
Magnitude: |Q_2|= 12.523257
Molecular octopole moments (Cartesian form):
XXX= 27.1548 YYY= -34.6763 ZZZ= 12.9803 XYY= 31.5970 XXY= -32.1231
XXZ= 11.3712 XZZ= 18.8632 YZZ= -6.4625 YYZ= 25.0005 XYZ= 21.8843
Molecular octopole moments (Spherical harmonic form):
Q_3,0 = -41.5772 Q_3,-1= 25.0762 Q_3,1 = 10.2272
Q_3,-2= 84.7574 Q_3,2 = -26.3932 Q_3,-3= -48.7725 Q_3,3 = -53.4710
Magnitude: |Q_3|= 124.8215
As can be seen, the dipole moment of the dimer is (1.227306 -0.128087 0.650833) a.u.
Next, we calculate dipole moment for the first phenol. We input
-5 // Define the atoms to be calculated
1-13 // Atom indices of the first phenol
You will see
Total number of electrons: 50.095624 Net charge: -0.095624
Molecular dipole moment (a.u.): 0.570356 -0.356257 0.284025
Molecular dipole moment (Debye): 1.449701 -0.905516 0.721919
Magnitude of molecular dipole moment (a.u.&Debye): 0.729997 1.855468
...[ignored]
showing that the dipole moment of the first phenol is (0.570356 -0.356257 0.284025) a.u., and the
phenol carries net charge of -0.096.
Then we input
-5 // Define the atoms to be calculated
14-26 // Atom indices of the second phenol
You will find the second phenol has dipole moment of (0.656950 0.228171 0.366808) a.u.
In summary, now we have three dipole moments:

• dimer: (1.227306 -0.128087 0.650833) a.u.
• 1st phenol: (0.570356 -0.356257 0.284025) a.u.
• 2nd phenol: (0.656950 0.228171 0.366808) a.u.
For easily visual inspection, we will plot the dipole moments as arrows in VMD visualization
program, which can be freely obtained at http://www.ks.uiuc.edu/Research/vmd/. The version I am
using is VMD 1.9.3. Since VMD is unable to load .wfn file, we need to convert the present system
to .xyz file. To do so, we return to main menu, enter main function 100 and choose subfunction 2,
then you will find corresponding option used to export .xyz file. We export the present system as
phenoldimer.xyz.
Boot up VMD and then load the phenoldimer.xyz into it. Copying all information from
examples\scripts\drawarrow.tcl script file to VMD console window to run it, a new custom
command "drawarrow" will be defined, which will be used to plot arrows. Then we input below
590
commands to VMD console:

draw color green
drawarrow all 1.227306 -0.128087 0.650833 2
draw color red
drawarrow "serial 1 to 13" 0.570356 -0.356257 0.284025 2
draw color yellow
drawarrow "serial 14 to 26" 0.656950 0.228171 0.366808 2
That means dipole moment of the whole system (selected by "all") will be plotted by green arrow,
that of the 1st and 2nd phenol molecules will be plotted by red and yellow arrows, respectively. The
"serial 1 to 13" and "serial 14 to 26" are selection syntax in VMD. The argument "2" at the end of
the commands makes the length of the arrows doubled, so that the dipole moments can be
represented by the arrows clearly.
After some adjustments of graphical effect (e.g. setting drawing method as "licorice" in
"Graphics" - "Representation" interface), you can see below map. The center of the red, yellow and
green arrows are placed at geometric centers of the two phenols and the dimer, respectively; the
arrow lengths correspond to the norm of the dipole moments (multiplied by 2).
Since the dipole moment vectors of the two monomers are nearly parallel with each other, their
vector sum, namely the dipole moment of the dimer, is significantly larger than the monomer dipole
moments.
Bear in mind that fragment dipole moment relies on the choice of origin if the net charge of
the fragment is not zero. For the present example, due to marginal charge transfer between the two
phenol molecules, the monomer dipole moment must be slightly dependent of origin. However this
is not an evident problem since the net charge of the monomer is quite small and the current origin
is appropriate (at nuclear charge center of the dimer).
4.16 Charge decomposition analysis and plotting orbital

interaction diagram
The theory of charge decomposition analysis (CDA), extended CDA and generalized CDA
(GCDA, J. Adv. Phys. Chem., 4, 111-124 (2015), DOI: 10.12677/JAPC.2015.44013), as well as
usage of CDA module have been detailedly introduced in Section 3.19. In this section I will present
591
several practical examples.

The CDA module supports .fch file, .mwfn file, Molden input file (.molden), GAMESS-US
output file (.gms) and Gaussian output file as input. In the next several sections only Gaussian .fch
files will be used to exemplify the usage of the CDA module.
4.16.1 Closed-shell interaction case: COBH3
In COBH3, CO makes use of its lone pair to form coordinate bond with BH3, which is an
electron-deficient system (Lewis acid). Therefore, electrons will transfer from CO to BH3 during
the formation of the complex. In this example we will employ CDA scheme to provide a deeper
insight on the electron transfer.
First we generate Gaussian output file for CO (fragment 1), BH3 (fragment 2) and COBH3
(complex). The .fch files and corresponding input files have been provided in
"examples\CDA\COBH3" folder. The calculations were performed at HF/6-31G* level. On how to
prepare the input files for CDA, see Section 3.19.2 for detail
Now boot up Multiwfn, and input following contents:
examples\CDA\COBH3\COBH3.fch // Gaussian .fch file of the complex
16 // Enter CDA module
2 // We define two fragments
examples\CDA\COBH3\CO.fch // Gaussian .fch file of fragment 1
examples\CDA\COBH3\BH3.fch // Gaussian .fch file of fragment 2
Immediately, below CDA result are outputted on screen
Orb. Occ. d b d - b r
1 2.000000 -0.000004 -0.000000 -0.000004 -0.000001
2 2.000000 0.001119 -0.000023 0.001141 0.000326
3 2.000000 -0.000002 -0.000471 0.000469 0.000313
4 2.000000 -0.013250 -0.000704 -0.012546 -0.005676
5 2.000000 0.041648 -0.003309 0.044957 0.232262
6 2.000000 0.037385 -0.020136 0.057521 0.212422
7 2.000000 -0.000543 0.000647 -0.001190 0.022166
8 2.000000 -0.000543 0.000647 -0.001190 0.022166
9 2.000000 0.171353 0.026952 0.144401 -0.741381
10 2.000000 -0.000569 0.043713 -0.044281 -0.038916
11 2.000000 -0.000569 0.043713 -0.044282 -0.038916
12 0.000000 0.000000 0.000000 0.000000 0.000000
13 0.000000 0.000000 0.000000 0.000000 0.000000
14 0.000000 0.000000 0.000000 0.000000 0.000000
15 0.000000 0.000000 0.000000 0.000000 0.000000
......
-------------------------------------------------------------------
Sum: 22.000000 0.236023 0.091027 0.144996 -0.335233
"Orb." denotes the indices of the orbitals of the complex; "occ." is corresponding occupation
number. "d(i)" stands for the amount of donated electrons from fragment 1 to 2 via corresponding
complex orbital, "b(i)" stands for the amount of electrons back donated from fragment 2 to 1 via
corresponding complex orbital. "r(i)" corresponds to the overlap population between the occupied
592
fragment orbitals (FOs) of the two fragments in corresponding complex orbital; its positive and
negative signs imply that in this complex orbital, the electrons of occupied FOs are accumulated to
and depleted (mainly due to Pauli repulsion) from the overlap region between the two fragments,
respectively. The sum of r(i), namely -0.335, reveals that repulsive effect dominates the overall
interaction between occupied FOs, which results in corresponding electrons moved away towards
nonoverlapping regions from overlap regions.
The difference between d(i) and b(i), to some extent can be viewed as the number of net
transferred electrons from fragment 1 to 2 due to formation of corresponding complex orbital. But
bear in mind, electron polarization effect is also included into this value.
From the data, it can be seen that the first three complex orbitals have almost zero b, d and r
values, this is because they are core orbitals of O, C and B, respectively, and hence it can be expected
that they are not involved in bond formation. The virtual complex orbitals have exactly zero b, d
and r terms, since their occupation numbers are exactly zero. Orbital 9 leads 0.171 electrons donate
from CO to BH3, which is the primary source of the donor-acceptor bonding, and causes as high as
0.741 electrons removed from overlap region between CO and BH3, which stabilized the complex
by diminishing electron repulsion. Orbital 5 and 6 have small contributions to electron donation,
meanwhile lead evident accumulation of electrons from respective occupied FOs to the overlap
region, which must be beneficial to the bonding between the two fragments. Orbital 10 and 11 are
π orbitals and degenerate in energy, they exhibit π-back donation characteristics.
Isosurfaces of orbital 9, 5 and 6 are shown below. As you can see, a node occurs in the overlap
region between CO and BH3 in orbital 9, while the isosurfaces of orbital 5 and 6 uniformly cover
the overlap region. These observations largely explained why r(9) is a large negative value, and r(5)
and r(6) are obvious positive values.
Notice that the definition of CDA used in Multiwfn is a generalized version proposed by me,
so that CDA is also applicable to post-HF and open-shell circumstances, see corresponding part of
Section 3.19.1 for detail. For the cases when original CDA is applicable (namely MO for FO, MO
or natural orbital for complex orbital), the d and b terms produced by generalized CDA are exactly
identical to the ones produced by original definition, while the r term is exactly twice of the one
produced by original definiton.
Note: The COBH3 example was also given in original paper of CDA, in which, althought the formulae of d, b
and r are correct, the data in their examples are incorrectly: all of the data should be divided by two. I also found that
all of the results of the CDA program written by Dapprich and Frenking (http://www.uni-marburg.de/fb15/ag-
593
frenking/cda), and the results of AOMix program (http://www.sg-chem.net) should be divided by two. So, if you
want to compare the CDA results calculated by Multiwfn with theirs, the d and b terms should be multiplied by two
first. But do not do this for r term, since the r term computed by Multiwfn has already been doubled with respect to
its original definition.
The amount of net electron transfer between the two fragments can be characterized by b-d
term, however it was argued that ECDA is a more reasonably method to calculate the amount of net
electron transfer, since the contribution of electron polarization effect (PL) is completely excluded.
ECDA result is outputted after CDA result:
========== Extended Charge decomposition analysis (ECDA) ==========
Contribution to all occupied complex orbital:
Occupied, virtual orbitals of fragment 1: 680.4194% 8.0593%
Contribution to all virtual complex orbital:
PL( 1) + CT( 1-> 2) = 0.3916 PL( 1) + CT( 2-> 1) = 0.1612
PL( 2) + CT( 1-> 2) = 0.4225 PL( 2) + CT( 2-> 1) = 0.1920
The net electrons obtained by frag. 2 = CT( 1-> 2) - CT( 2-> 1) = 0.2304
Commonly, you only need to pay attention to the last line. The data shows that the net number
of electrons transferred from fragment 1 to 2 is 0.2304.
In the menu that appears on screen, by using option 2, composition of FOs in a specific complex
orbital can be outputted. Here we select this option and input 9, the composition of complex orbital
9 are shown below
Occupation number of orbital 9 of the complex: 2.00000000
Orbital 7 of fragment 1, Occ: 2.00000 Contribution: 25.8560%
Only the FOs with contribution  1% to the complex orbital are shown (the threshold can be
altered by "compthresCDA" in settings.ini). As already mentioned, the electron transfer from CO to
BH3 is mainly due to the complex orbital 9, therefore from above data we can infer that the nature
of the CO→BH3 electron transfer can be largely interpreted as the mix between FO 7 of CO (an
occupied orbital) and FO 5 of BH3 (a virtual orbital). This viewpoint can also be manifested by
comparing the shape of the two FOs (see below) with the shape of complex orbital 9 given above.
594
Note: Sometimes a few contributions of FOs may be negative, this is a well-known drawback of Mulliken
analysis, which is the method employed in CDA module to calculate the complex orbital composition. Since the
magnitudes of the negative values are often small, you can simply ignore them.
We can also directly decompose d, b, r terms of a complex orbital to FO pair contributions, let
us do this for complex orbital 9. Now input 0 to return to the last menu, select option 6 and input 9,
then input a threshold e.g. 0.005, then all FO pairs whose contribution to any of d, b, r term larger
than 0.005 are printed:
FragA Orb(Occ.) FragB Orb(Occ.) d b d - b r
4( 2.0000) 2( 2.0000) 0.000000 0.000000 0.000000 -0.009969
7( 2.0000) 1( 2.0000) 0.000000 0.000000 0.000000 -0.005759
7( 2.0000) 2( 2.0000) 0.000000 0.000000 0.000000 -0.723845
7( 2.0000) 5( 0.0000) 0.176503 0.000000 0.176503 0.000000
7( 2.0000) 8( 0.0000) 0.006221 0.000000 0.006221 0.000000
7( 2.0000) 11( 0.0000) 0.005846 0.000000 0.005846 0.000000
7( 2.0000) 12( 0.0000) -0.023941 0.000000 -0.023941 0.000000
13( 0.0000) 2( 2.0000) 0.000000 0.021958 -0.021958 0.000000
From the output it is clear that the mix between FO 7 of CO and FO 5 of BH3 contributes most of
the d term of complex orbital 9.
Finally, we plot orbital interaction diagram. Input 0 to return to the last menu, and then select
5 to enter the menu for plotting orbital interaction diagram. Choose option 3 and input -30,10 to set
lower and upper energy limits of the plot to -30eV and 10eV, respectively. Then select option 1 to
plot the diagram under default settings, a graph will pop up on screen:
In above graph, occupied and virtual orbitals are represented as solid and dashed bars,
595
respectively. Orbital indices are labelled by blue texts. If two or more labels occur in the same bar,
e.g. 7 and 8, that means these orbitals are degenerate in energy. If composition of a FO in a complex
orbital is larger than 10%, then the corresponding two bars will be linked by red line, and the
composition will be labelled on the lines. By simply viewing the diagram we can directly understand
how the MOs of COBH3 are constructed by FOs of CO and BH3. For example, complex orbital 7
and 8 only link to FO 5 and 6 of CO in this diagram, hence we immediately know that these two
orbitals basically remain unperturbed during formation of the complex. In fact they are π orbitals of
CO, certainly they cannot participate to the σ type donor-acceptor interaction between CO and BH3.
There are many options used to adjust plotting parameters (such as label size, the rule for
drawing and linking bars, position of composition labels, energy range), please play with them, and
replot the graph to check their effects.
4.16.2 Open-shell interaction case: CH3NH2
In this example, I use CH3NH2 to illustrate how to perform CDA for the complex in which the
two fragments interact with each other covalently (open-shell interaction).
First we need to generate Gaussian output file for CH3 (fragment 1), NH2 (fragment 2) and
CH3NH2 (complex). For fragment 1 and 2, in present example we use unrestricted B3LYP method;
while for the complex, since this is a closed-shell system, we use restricted B3LYP method
(unrestricted B3LYP can also be used, the CDA result will be the same). The .fch files and
corresponding input files can be found in "examples\CDA\CH3NH2" folder, the geometry was pre-
optimized under B3LYP/6-31G** level.
Note that both CH3 and NH2 have 5 alpha and 4 beta electrons, while CH3NH2 has 9 alpha and
9 beta electrons. Evidently, the total numbers of alpha and beta electrons in the two fragments,
namely 5+5 and 4+4, does not match the ones of the complex. So, we must flip electron spin of one
fragment (either CH3 or NH2). In this example, we will flip electron spin of NH2, i.e. exchanging
all information of its alpha and beta electrons.
Boot up Multiwfn and input following contents:
examples\CDA\CH3NH2\CH3NH2.fch // Gaussian output file of the complex
16 // Enter CDA module
2 // We define two fragments
examples\CDA\CH3NH2\CH3.fch // Gaussian output file of fragment 1
examples\CDA\CH3NH2\NH2.fch // Gaussian output file of fragment 2
n // Do not flip electron spin of fragment 1
y // Flip electron spin of fragment 2, then NH2 will have 4 alpha and 5 beta electrons.
CDA and ECDA results will be calculated and printed on screen for alpha electrons and beta
electrons separately. As you can see, for alpha (beta) part, both d - b and CT(1->2) - CT(2->1) terms
are positive (negative), showing that alpha (beta) electrons are tranferred from CH3 to NH2 (from
NH2 to CH3). This is mainly because CH3 has more alpha electrons (5) than beta electrons (4), while
after flipping electron spin, NH2 has more beta electrons (5) than alpha electrons (4), hence when
they combine together to form CH3NH2, CH3 prefer to donate alpha electrons to NH2 and accept
beta electrons from NH2.
Result of total electrons, namely the sum of alpha and beta results are also outputted. Below is
596
total result of CDA and ECDA, respectively

d= 0.044335 b= 0.145181 d - b = -0.100847 r= -0.172318
CT( 1-> 2) - CT( 2-> 1) for all electrons: 0.1252
It is obvious that electron should transfer from CH3 to NH2, because nitrogen has evidently
larger electronegativity. The total ECDA result is in line with our expectation, it shows that the
amount of transferred electrons from CH3 to NH2 is 0.1249. However, the d - b term conflicts with
our expectation, the transfer direction is totally inverted. This example illustrates that d - b term is
not as reliable as ECDA to reveal total amount of net electron transfer for open-shell interaction,
although d and b terms are still quite useful for decomposing electron tranfer into orbital
contributions.
If you select to flip electron spin for CH3 rather than for NH2 when loading their Gaussian
output files, you will see the alpha and beta results of CDA and ECDA are exchanged, but the result
for total electrons remains unchanged.
Now select 2 and input 6, you will see both the composition of the 6th alpha orbital and the 6th
beta orbital of the complex are printed on screen. Though the complex is a closed-shell system and
thus the two orbitals are essentially the same, owing to the alpha and beta FOs in the two fragments
are not equivalent, the printed compositions have slight difference. Then input 0 to return.
Select option 5 to enter the menu for plotting orbital interaction diagram. By option 5 in this
menu, you can switch the spin of the orbitals that the diagram will be plotted for. We select it once
to switch the status to "Beta", then choose option 3 and input -30,10 to change the lower and upper
limits of the diagram to -30 and 10 eV, respectively. Then select option 1 to plot the orbital
interaction diagram. From the graph it is very clear that, beta orbital 3 and 4 of CH3NH2 are formed
by mixing beta FO 2 in CH3 and beta FO 2 in NH2. To illustrate this point more intuitively,
corresponding part is extracted from the whole diagram, and the orbital isosurfaces are attached on
it, see below
It can be seen that, beta complex orbital 3 shows bonding character (this is why r(3) is a positive
value, namely 0.103), which is constructed by slightly mixing beta FO 2 of CH3 into beta FO 2 of
NH2 with the same wavefunction phase. Beta complex orbital 4 is an anti-bonding orbital (this
explained why r(4) is a negative value, namely -0.056), formation of which is due to the mixture of
beta FO 2 of NH2 into beta FO 2 of CH3 in terms of different phase.
597
4.16.3 More than two fragments case: Pt(NH3)2Cl2
In this section I use cisplatin (see below) as instance to show how to perform CDA analysis on
the system consisting of more than two fragments.
We will define Pt2+ cation, (Cl2)2- anion and (NH3)2 as fragment 1, 2 and 3 respectively. This
definition of fragments is the best choice for present system. Note that the Pt2+ cation is in 3d8 low-
spin configuration, so that all fragments are in closed-shell status and thus we will not need to
analyze alpha and beta spins separately.
The related Gaussian input files and resulting .fch files have been provided in
examples\CDA\Pt(NH3)2Cl2 folder. Lanl2DZ pseudo-potential basis set was used for Pt, cc-pVDZ
was used for other atoms. B3LYP functional was chosen as the theoretical method. The geometry
of the complex has already been optimized. In the input file of Pt2+, "scf=xqc" keyword was used to
solve the unconvergence problem, but this is not needed in general.
Note 1: For certain types of basis sets, such as Pople series of basis sets (e.g. 6-31G*), by default Gaussian
employs Cartesian type basis functions rather than spherical-harmonic basis functions, which may results in
inconsistency problem of basis set between complex and fragment calculations. If you are not very familiar with
Gaussian, I highly recommend you always add 5d keyword in all Gaussian input files when mixed basis set is used
in the calculation of the complex.
Note 2: The sequence of the fragments is crucial. Because in the input file of complex, the atomic sequence is
Pt--Cl2--(NH3)2, we should not for example define Pt2+, (NH3)2 and (Cl2)2- as fragment 1, 2 and 3, respectively.
Now boot up Multiwfn and input:

examples\CDA\Pt(NH3)2Cl2\Pt(NH3)2Cl2.fch // Complex
16 // CDA module
3 // Define three fragments
examples\CDA\Pt(NH3)2Cl2\Pt.fch // Fragment 1
examples\CDA\Pt(NH3)2Cl2\Cl2.fch // Fragment 2
examples\CDA\Pt(NH3)2Cl2\(NH3)2.fch // Fragment 3
Then choose option 0 and input 1,2 to output CDA analysis result for fragment pair 1-2.
Similarly, we get the CDA result for fragment pair 1-3 and 2-3. We cannot obtain ECDA result for
present system because ECDA is only applicable to two-fragment cases. The total CDA results are
summarized below.
1-2 1-3 2-3
d 0.0017 0.0538 0.0538
b 0.5156 0.0071 0.0071
d-b -0.5139 -0.0467 0.0467
r 0.0581 -0.1415 -0.1415
The table shows that there are 0.5156 and 0.0071 net electrons transferred from (Cl2)2- and
(NH3)2 to Pt2+, respectively. Basically no electrons are donated from Pt2+ to its ligands. This result
is in line with our chemical intuition because according to classical theory, the electrons of
coordinate bonds are purely contributed from ligands. The charge transferation between (Cl2)2- and
(NH3)2 is trivial, mainly due to there is no direct chemical interaction between them. Note that the
r term between fragment 2 and 3 is a small but non-negligible negative value, exhibiting that there
598
is detectable electron repulsion effect between the ligands.

Deeper analyses on this system are left for you. You can discuss orbital contribution to charge
transferation based on the detail output of CDA analysis, use option 2 to study the composition of
complex orbital in fragment orbital basis, and use option 5 to plot orbital interaction diagram.
4.17 Basin analysis
Below I will show how to use basin analysis module of Multiwfn to perform basin analysis for
several molecules and for various real space functions. Related theory, numerical algorithms and
the usage of this module have been detailedly introduced in Section 3.20. If you are not familiar
with the concept of basin, please consult Section 3.20.1 first. While if you want to know more detail
about basin analysis, please concult Sections 3.20.2 and 3.20.3.
You should know that Multiwfn uses a grid-based method (in particular, near-grid method) to
locate attractors, generate and integrate basins; in other words, most tasks realized in the basin
analysis module rely on grid data. This is why in below sections "grid data" is frequently mentioned.
4.17.1 AIM basin analysis for HCN and Li6
In this example we will analyze basins of electron density (also known as AIM basins) for
HCN molecule, which is a very representative molecule. Only at final part of this section, I will also
show how to carry out basin analysis of electron density for Li6 atomic cluster, since this is a special
case, in which there are some "pseudoatoms".
After you carefully read this section, I believe you will understand most of basic operations of
the basin analysis module in Multiwfn.
Generate basins and locate attractors

examples\HCN.wfn
1 // The grid data to be calculated and thus analyzed is for electron density
2 // Medium quality grid. This is enough for most cases, if you want to obtain a better result,
you can choose "High quality grid", but much more computational time will be spent
Multiwfn now starts to calculate grid data for electron density, and then generates basins and
locates attractors based on the grid data; soon you will see attractor information, as shown below
(notice that the attractor sequence may be different to your actual case if parallel mode is enabled,
similarly hereinafter):
Attractor X,Y,Z coordinate (Angstrom) Value
1 -0.02645886 -0.02645886 -1.52058663 0.33959944
2 0.02645886 0.02645886 -0.51514986 49.48609717
3 -0.02645886 -0.02645886 0.64904009 76.55832377
From now on you can check the information of the located attractors anytime by option -3.
Visualization of basins and attractors
599
Now choose option 0 to visualize the basins and attractors. The purple labels indicate the index
of the located attractors. Since attractors of electron density are very close to nuclei, if you want to
see them, you should deselect "Show molecule" check box first. Here we select 1 from the basin list
at the right-bottom corner, the basin corresponding to attractor 1 will be immedately shown. By
default only interbasin part of the basin is shown, if you want to inspect the whole basin you should
select "Show basin interior" checkbox. After we select it, the GUI will look like below:
If you only want to visualize the region of the atomic basin in the vdW surface (defined as
=0.001 a.u. in the present context), you can choose "Set basin drawing method" - "rho>0.001
region only", then you can see:
Now click "RETURN" button to close the GUI.
You may have noticed below information in the command-line window

The number of unassigned grids: 0
The number of grids travelled to box boundary: 0
Commonly the number of these two types of grid should be zero, only in rare cases they are not
zero; in such cases, you can visualize them by respectively selecting "Unas" and "Boun" in the basin
list of the GUI. These grids do not belong to any basin, and generally they lack of physical meaning;
to understand when and why they occur please consult Section 3.20.2.
Integrating basins
Next, we calculate the integral of electron density (electron population number) in these basins.
600
Select function 2, then you will see many options. Each option with the serial  1 corresponds to a
real space function; if you select one of them, corresponding real space function will be integrated
in the generated basins. In present example we can select option 1, which corresponds to electron
density. However, since we have already calculated grid data for electron density, and the grid data
of electron density thus have been stored in memory, we can directly use it rather than recompute it
again to reduce computational time, so here we select option 0 to use "The values of the grid data
stored in memory". Since electron density at the grids is not needed to be re-evaluated, the integrals
are outputted immediately:
#Basin Integral(a.u.) Volume(a.u.^3)
1 0.7356142812 441.70000000
2 5.3511723358 566.84900000
3 7.9020047534 829.72600000
Sum of above values: 13.98879137
There should be totally 1+6+7=14 electrons in present system; unfortunately the sum of the
integrals of electron density is 13.98879, which evidently deviates from the ideal value!
Because the basin we are studying is AIM basin, the best choice to obtain the basin integral is
using function 7 rather than function 2. In function 7, mixed atomic-center and uniform grid is used,
while function 2 only employs uniform grid to integrate. We input:
7 // Integrate real space functions in AIM basins with mixed type of grids
1 // Integrate a specific function with atomic-center + uniform grids
1 // Select electron density as the integrand
The result is
#Basin Integral(a.u.) Vol(Bohr^3) Vol(rho>0.001)
1 0.7356461193 441.700 34.884
2 5.3552283604 566.849 102.832
3 7.9086490986 829.726 134.228
Sum of above integrals: 13.99952358
Sum of basin volumes (rho>0.001): 271.944 Bohr^3
As you can see, the sum of the integrals of electron density (13.9995) is almost exactly identical
to the expected value 14.0, obviously the result is much better than using function 2. The basin
volumes are also outputted. The terms "vol(Bohr^2)" do not have clear physical meaning, since they
are directly affected by the spatial range of grid setting. However, the terms "vol(rho>0.001)" are
useful, they exhibit the volume of the basin enclosed by the isosurface of electron density > 0.001
(Bader's vdW surface) and thus can be regarded as atomic sizes.
The atomic charge (AIM charge) of these atoms and their volumes are also outputted
1 (C ) Charge: 0.644589 Volume: 102.832 Bohr^3
2 (N ) Charge: -0.908918 Volume: 134.228 Bohr^3
3 (H ) Charge: 0.264329 Volume: 34.884 Bohr^3
Note that above AIM charges are inaccurate! To obtain more accurate integrals in AIM basins,
you should select 2 or 3 in function 7; compared to 1, they will refine the assignment of basin
boundary to improve the integration result, but additional computational cost must be afforded. Here
we try it, select option 2 in function 7, then input 1, the result is
1 (C ) Charge: 0.748979 Volume: 100.160 Bohr^3
2 (N ) Charge: -1.003917 Volume: 136.304 Bohr^3
601
3 (H ) Charge: 0.254938 Volume: 35.480 Bohr^3
Although the sum of the integral of electron density is still 13.9995, the charge of the atoms
are somewhat varied (more accurate than before).
To further improve the integration accuracy, when generating basins one should select a grid
setting better than "medium quality grid", e.g. "High quality grid" or even "Lunatic quality grid".
But bear in mind that for large systems, "high quality grid" (or a better one) may consumes very
large amount of computational time and memory.
Although as we have seen the integration accuracy of function 7 is much better than function
2, the former is only applicable to AIM basins, while the latter can be used for any type of basin
(e.g. ELF basin).
Note: If you used option 2 or 3 in function 7, during the bounary grid refinement process, the assignment of
basin boundary will be updated permanently, that means the result of later analysis (e.g. calculating LI/DI, electric
multipole moment) will become more accurate.
Hint: In summary, the general steps for obtaining reliable AIM charges after you entered basin
analysis module is
1 // Generate basin
2 // Medium quality grid. Select high quality grid if you wish to get more accurate result
2 // Integrate and meantime refine basin boundary
Calculate electric multipole moments in basins

Enter function 8 to calculate electric multipole moments in the AIM basins (you can also use
3 to do this, but the accuracy is much lower). Only the result for carbon is pasted below
Result of atom 1 (C )
Basin electric monopole moment: -5.250849
Basin electric dipole moment:
X= 0.000003 Y= 0.000003 Z= 1.095846 Magnitude= 1.095846
Basin electron contribution to molecular dipole moment:
X= 0.000003 Y= 0.000003 Z= 6.026238 Magnitude= 6.026238
Basin electric quadrupole moment (Cartesian form):
QXX= -0.759571 QXY= 0.000000 QXZ= -0.000002
QYX= 0.000000 QYY= -0.759571 QYZ= -0.000002
QZX= -0.000002 QZY= -0.000002 QZZ= 1.519142
The magnitude of electric quadrupole moment (Cartesian form): 1.519142
Electric quadrupole moments (Spherical harmonic form):
Q_2,0 = 1.519142 Q_2,-1= -0.000002 Q_2,1= -0.000002
Q_2,-2= 0.000000 Q_2,2 = 0.000000
Magnitude: |Q_2|= 1.519142
The formulae used to evaluate these terms are basically identical to the ones given in Section
3.18.3, the only differences are that in that formulae the nuclear positions should be replaced by
attractor positions, and the ranges of integration should be basins rather than fuzzy atomic spaces.
The electric monopole moment (-5.251) is just the negative value of electron population
number in the basin. Z-component of electric dipole moment of the basin is a positive value (1.096),
suggesting that in basin 2, most of electrons are distributed in the regions where Z-coordinate is
602
more negative than attractor 2. The ZZ-component of basin electric quadrupole moment is positive
(1.519), while the other diagonal components are negative, indicating that relative to attractor 2,
electron cloud in this basin contracts in Z-direction but elongates in other directions.
Calculate localization index and delocalization index in basins

Delocalization index (DI) is a quantitative measure of the number of electrons delocalized (or
say shared) between two regions, while localization index (LI) quantitatively measures how many
electron are localized in a region. For details about LI and DI please see Section 3.18.5. The only
difference relative to the statements in that section is that here the LI and DI will be calculated based
on basins rather than based on fuzzy atomic spaces.
Choose function 4, Multiwfn will start to calculate LI and DI. As you can see from screen, the
LI and DI matrix are outputted first based on basin indices, after that, Multiwfn identifies
correspondence between atom indices and basin indices, then print LI and DI matrix in atom indices,
as shown below
Detecting correspondence between basin and atom indices (criterion: <0.3 Bohr)
Basin 2 corresponds to atom 1 (C )
Basin 3 corresponds to atom 2 (N )
Basin 1 corresponds to atom 3 (H )
Note: In below output the indices are atom indices rather than basin indices
************** Total delocalization index matrix (atom index) **************
1 2 3
1 3.4930 2.5927 0.9004
2 2.5927 2.6652 0.0725
3 0.9004 0.0725 0.9729
Total localization index (atom index):
1: 3.496 2: 6.658 3: 0.259
Since present molecule is a close shell system, only total LI and DI are outputted, the LI and
DI for  and  electrons are not outputted separately. As you can see, between C1 and N2 atomic
basins, the DI is 2.593, exhibiting that in average there are 2.593 electrons shared by the two atoms.
To some extent DI can be regarded as covalent bond order, the DI value 2.593 is indeed comparable
with the formal bond order (3.0) between C and N in HCN. The diagonal terms are the sums of the
elements in corresponding row/column, for closed-shell cases they can be somewhat considered as
atomic valency. So we can say the nitrogen atom in HCN has atomic valency of 2.665.
The LI of H3 is only 0.259 which conspicuously deviates from the basin electron population
number. This observation reflects that in HCN, the electron in the AIM atomic space of hydrogen
can easily delocalize out.
Special case: Basin analysis when pseudoatoms are presented

Pseudoatom is also known as non-nuclear attractor (NNA) of electron density, it refers to
maximum of electron density that are not at nuclear position. There are various reasons that can
cause the NNAs, for example, existence of metal bond or quality of wavefunction is too poor. Here
I use Li6 cluster as example to illustrate how to deal with the case when NNAs are presented.
603
examples\Li6.fch
1 // For illustration purpose, here we only use low quality grid for saving time
0 // Visualize attractors and basins
Left part of below graph shows geometry of the cluster, the three green spheres indicate
position of the three NNAs; right part of the graph displays corresponding basin of one of NNAs.
As you can see, attractors 2, 4, 8 are NNAs.
If you plot color-filled map of valence electron density for the Li6 via the steps illustrated in
Section 4.6.2, you will immediately understand why there exists NNAs at center of the boundary Li
triangles. From below graph you can clearly see that at center of each boundary triangle there is
indeed a maximum of electron density, this observation also implies presence of three-center bonds
Next, we calculate population number of the AIM basins. Input below commands
604
2 // Integrate and meantime refine basin boundary
During integrating the basins, calculation will pause three times and meantime you will find
prompt like below on screen. This is because program does not know how to properly deal with the
three NNAs, namely attractors 2, 4, 8:
Warning: Unable to determine the attractor 2 belongs to which atom!
If this is a non-nuclear attractor, simply press ENTER button to continue. If y
ou used pseudopotential and this attractor corresponds to the cluster of all max
ima of its valence electron, then input the index of this atom (e.g. 9). Else yo
u should input q to return and regenerate basins with smaller grid spacing
Since we already know that attractors 2, 4, 8 are regular NNAs, according to the prompt, what
we should do is simply pressing ENTER button to continue the calculation. Finally, you will find
below output
Normalization factor of the integral of electron density is 0.999992
The atomic charges after normalization and atomic volumes:
2 (NNA) Charge: -1.249592 Volume: 273.680 Bohr^3
1 (Li) Charge: 0.757502 Volume: 69.792 Bohr^3
As can be seen, each NNA basin carries 1.249 electrons, therefore charge of the basin is -1.249.
When NNAs are presented, it is clearly unable to rigorously define AIM atomic charges, since sum
of all AIM atomic charges will be unequal to net charge of the whole system. This is one of severe
limitations of AIM atomic charge.
Drawing basins via VMD program

When showing basins, you can easily get much better graphical effect if you make use of VMD
program. See this video illustration: "Drawing AIM basins (atomic basins) in Multiwfn and VMD"
(https://youtu.be/9D5do80XcbI).
Briefly speaking, what you need to do is selecting option -5 in the basin analysis module, input
indices of the basins of interest, and then choose "3 Output all basin grids where electron density >
0.001 a.u." to export the basins as individual cube files. Then loading them into VMD and plot them
as isosurfaces with isovalue=0.5, you can get a graph like the following one (oxygen basin of
examples\acrolein.wfn):
605
As illustrated in the aforementioned video tutorial, you can even simultaneously draw multiple
basins and show critical points + bond paths in VMD. For example, below map represents atomic
basins of four non-hydrogen atoms of acrolein, the orange and purple spheres correspond to bond
critical points and nuclear critical points, respectively. The yellow thick lines are bond paths.
4.17.2 ELF basin analysis for acetylene
This time we analyze ELF basin for acetylene. Boot up Multiwfn and input
examples\C2H2.wfn
9 // ELF
0 // Visualize attractors and basins
In the GUI, as the graph shown below, you can see there are lots of closely placed attractors
encircling the C-C bond. In fact, they have very similar ELF values, and collectively represent the
ring-type ELF attractor. These attractors have been clustered together by Multiwfn automatically,
therefore all of them have identical attractor index, namely 2; in another word, attractor 2 is a
"degenerate" attractor, which has many member attractors. Correspondingly, basin 2 is composed
by the member basins. The interested user is recommended to visualize basin 2 in the GUI.
606
The presence of attractor 2 signifies the π electrons shared by the two carbons. According to
the well-known ELF symbolic method, basin 2 should be identified as V(C1,C3), which means this
basin is comprised by the valence electrons of C1 and C3.
Attractor 3 and 4 correspond to core-type ELF attractors, their basins should be identified as
C(C3) and C(C1), respectively, where the letter C out of parentheses stands for "Core". If you
deselect the "Show molecule" check box and select corresponding terms in the basin list then you
can visualize the basins. The graph shown below portrays C(C3).
Although basin 1 and 5 cover the whole hydrogen spaces, they should be identified as V(C3,H4)
and V(C1,H2), rather than C(H4) and C(H2), respectively. This is because hydrogen does not have
core electrons.
Now close the GUI window by clicking "RETURN" button, and then input
2 // Integrate a real space functions in the basins
Soon, we get the integrals, namely the average electron population number in each basin:
1 2.2159821115 768.39800000
2 5.3670485473 972.79700000
607
3 2.0949016050 0.83200000
4 2.0949016050 0.83200000
5 2.2159824751 768.55500000
Both C(C1) and C(C3) in average contain 2.095 electrons, which is in line with the fact that
carbon has two electrons in its core. Also, the average population number in V(C3,H4) and V(C1,H2)
are close to two, approximately reflecting that in average there is a pair of electron shared between
C and H.
According to classical chemical bond theory there are three electron pairs and hence six
electrons are shared by the two carbons, however in the V(C1,C3) basin the integral is only 5.37.
Although the deviation is relatively large, this is a normal situation. It is senseless to require that the
result of ELF basin analysis must be able to reproduce classical Lewis picture, and actually, ELF
analysis is more advanced and more close to real physical picture.
Next, enter function 3 to calculate electric multipole moments. We want the result for all basins
to be outputted, so we input -1. The result of C(C1) is pasted below
Basin 4
Basin electric monopole moment: -2.094902
Basin electric dipole moment:
X= -0.104745 Y= -0.104745 Z= 0.020728 Magnitude= 0.149575
Basin electron contribution to molecular dipole moment:
X= 0.000000 Y= 0.000000 Z= -2.388409 Magnitude= 2.388409
Basin electric quadrupole moment (Cartesian form):
QXX= -0.003034 QXY= -0.007856 QXZ= 0.001555
QYX= -0.007856 QYY= -0.003034 QYZ= 0.001555
QZX= 0.001555 QZY= 0.001555 QZZ= 0.006067
The magnitude of electric quadrupole moment (Cartesian form): 0.006067
Electric quadrupole moments (Spherical harmonic form):
Q_2,0 = 0.006067 Q_2,-1= 0.001795 Q_2,1= 0.001795
Q_2,-2= -0.009071 Q_2,2 = 0.000000
Magnitude: |Q_2|= 0.011205
First you should note that although in present case, due to the symmetry, the X and Y
components of the basin electric dipole moment should vanish, the actual values are not so close to
zero, rendering that the integration accuracy is not very high. This is why "high quality grid" is often
necessary when electric multipole moment analysis is involved. However, present result is still
useful on the qualitative level. The magnitude of electric quadrupole moment in the basin measures
how evidently does the electron distribution in the basin deviate from spherical symmetry. This
value for C(C1) is very small (0.0061), showing that the distribution of core electron of the carbon
basically remains unperturbed during the formation of the molecule.
Note: If the basins you are interested in are only valence basins, "medium quality grid" is enough for electric
multipole moment analysis, since valency density is not as high as core density, and hence does not need high
accuracy of integration.
Input 0 to return. Finally, let us choose option 4 to study LI and DI. The result is shown below
************ Total delocalization index matrix ************
1 2 3 4 5
608
1 1.31231709 1.03496561 0.16085935 0.02291204 0.09358009
2 1.03496561 2.71090548 0.32048710 0.32048709 1.03496568
3 0.16085935 0.32048710 0.51131492 0.00705643 0.02291205
4 0.02291204 0.32048709 0.00705643 0.51131492 0.16085937
5 0.09358009 1.03496568 0.02291205 0.16085937 1.31231719
Total localization index:
1: 1.560 2: 4.011 3: 1.834 4: 1.834 5: 1.560
Between C(C1) and C(C3), namely DI(3,4), the value is trivial, reflecting the general rule that
the electron delocalization between atomic cores is rather difficult. DI(1,3) and DI(2,3) are very
small values but not close to zero, representing that the electrons in C(C3) have a few probability to
exchange with the ones in V(C3,H4) and V(C1,C3), which are the only two basins adjacent to C(C3).
DI(2,1) and DI(2,5) are about 1.0, such a large value indicates that the electron delocalization
between C-C bonding region and C-H bonding region is easy. Though the average electron
population number in C(C1) and C(C3) are 2.095, their LI values are as high as 1.834, implying that
the core electrons of carbon highly prefer to stay in the core region rather than delocalize out. In
constrast, for V(C,H) and V(C,H), their LI values are less than their average electron population
numbers distinctly, revealing that the electrons in these basins do not express very strong
localization character.
In the ELF basin analyses, some researchers prefer to use the concept of variance (σ2) and
covariance (Cov) rather than LI and DI to discuss problems. Covariance of electron pair fluctuation
is simply the half of the negative value of DI, for example, Cov(2,5) = -DI(2,5)/2 = -1.035/2 = -
0.518. The variance of electronic fluctuation can be calculated as the half of the corresponding
diagonal term of the DI matrix outputted by Multiwfn, for instance, σ2(2) = DI(2,2)/2 = 2.710/2 =
1.355 (note that as mentioned earlier, the diagonal terms of the DI matrix outputted by Multiwfn are
the sums of the elements in corresponding row/column).
4.17.3 Basin analysis of electrostatic potential for H2O
In this example I will illustrate performing basin analysis for electrostatic potential (ESP), a
very simple molecule H2O is taken as instance. Although this kind of analysis is currently
uncommon in literatures, you will see this analysis is indeed useful; in particular, this analysis is
able to well reveal occurrence region of lone pairs.
It is noteworthy that unlike electron density and ELF which we have analyzed earlier, ESP has
positive part and negative part simultaneously. For such real space functions, Multiwfn will
automatically locate attractors (maxima) for positive part and locate "repulsors" (minima) for
negative part, but in Multiwfn all of them are collectively recorded as "attractors". You can easily
discriminate them by their colors in GUI.
Note: Calculation speed of ESP of cubegen utility in Gaussian package is faster than that of internal code of
Multiwfn. If you have Gaussian installed on your system and the input file is .fch/fchk, it is suggested to set
"cubegenpath" parameter in settings.ini to actual path of cubegen, then at the step of calculating ESP grid data in the
basin analysis, cubegen will be automatically invoked by Multiwfn to evaluate ESP, the time cost will hence be
greatly reduced. More information about this point can be found in Section 5.7.
Basic steps of performin ESP basin analysis

609
examples\H2O.fch // Optimized and produced at B3LYP/6-31G** level

1 // Select real space function used to partitioning basins
12 // ESP
Once the calculation of ESP grid data and the basin generation are finished, five attractors are
founded. Notice that this time the number of grids travelled to box boundary is not zero again (about
1954 grids), but it completely does not matter.
Choose option 0 to open GUI, you can see that attractors 1, 2 and 3 correspond to ESP maxima
due to nuclear charges. Occurrence of lone pair often makes ESP at corresponding region negative,
clearly attractors 4 and 5 exhibited this effect. Attractors 4 and 5 are colored by light blue because
they are lying in negative region, in fact they are not attractors but "repulsors" (minima) of ESP. By
clicking "4" in the basin list and checking "Show basin interior" box, you will see below graph,
which exhibits the corresponding basin region.
If you want to visualize which grids have travelled to box boundary during basin generation,
you can select "Boun" in the basin list, see below graph. Evidently, these grids lack of physical
meaning and hence can be simply ignored. They only present at the regions far from atoms.
Now click "RETURN" button to close the GUI.
Measuring geometry
Sometimes one needs to obtain geometry information between attractors and nuclei. As an
610
instance, we enter function -2, and input a1 c4, the distance between nucleus of atom 1 (namely
oxygen) and attractor 4 will be shown on the screen, the value is 2.272 Bohr. Next, input c4 a1 c5,
then the angle of "attractor 2 -- atom 1 -- attractor 3" will be outputted, the value is 86.01 degree,
which in some sense can be regarded as the angle between the two lone pairs.
Input q to exit the geometry measurement interface.
Clustering attractors
Assume that we want to cluster attractors 4 and 5 together as a degenerate attractor to make
they collectively represent the two lone pairs, we can input
-6 // Set parameter for attractor clustering or manually perform clustering
3 // Cluster specified attractors
4,5 // Attractors 4 and 5 will be clustered as a single one
0 // Return
Select option 0 to open GUI, as shown below, you can find that the index of all attractors have
changed, and the two attractors corresponding to the oxygen lone pairs now sharing the same index,
namely 4.
Integrating basins
Close the GUI by clicking "RETURN" button, choose option 2 and then select 1 to integrate
electron density in the ESP basins, the result is
1 0.8776693902 390.34100000
2 7.5388327160 26.52400000
3 0.8776693895 390.28300000
4 0.6822297997 654.32200000
Integral of the grids travelled to box boundary: 0.00000007
Since currently basin 4 is just the whole negative ESP region, the result shows that in average
there are 0.682 electrons in the negative ESP region. Indeed this value is not large (one may expects
that there should be about four electrons due to the two lone pairs), this is because ESP of most
regions of the molecular space is dominated by nuclear charges and hence positive.
Simultaneously showing ESP minimum points and isosurfaces

In J. Comput. Chem., 39, 488 (2017), the authors showed that by plotting ESP minimum points
and isosurfaces in the same map, the region of lone pairs can be very clearly exhibited. The isovalue
is chosen to be higher than global minimum value of ESP by 10 kcal/mol. Here we plot this kind
map.
Recall that when we just finished the basin generation, the value of each ESP minimum is
611
printed on screen (you can also choose option " -3 Show information of attractors" any time to show
them again):
1 -0.02645886 -0.76730701 -0.48343728 17.61230000
2 0.02645886 -0.02645886 0.09865769 79.16270000
3 0.02645886 0.76730701 -0.48343728 17.61230000
4 -0.82022474 0.02645886 0.99825901 -0.09222160
5 0.82022474 0.02645886 0.99825901 -0.09222160
Clearly, the global minimum value is -0.09222*627.51 = -57.9 kcal/mol, the isovalue should then
be set to -0.09222+10/627.51 = -0.07628 a.u.
Enter option 0, in the GUI window deselect "Attractor labels", then input
-2 // Visualize isosurface of the grid data in memory
In the GUI, input -0.07628 in "Isosurface value" box, deselect "Show both sign", choose
"Isosurface style" - "Use mesh", make sure that "Show atomic labels" has been activated, choose
"Other settings" - "Set atomic label type" - "Element symbol". Finally, click "Save picture" button
to save image file to current folder, you will see below effect (the value is manually labelled)
Perform basin analysis using external cube file

Multiwfn is able to perform basin analysis solely based on grid data, thus the real space
function to be analyzed could not be any one that formally supported by Multiwfn, such as the
Anisotropy of the Induced Current Density (Chem. Rev., 105, 3758 (2005)), which can be calculated
by AICD or GIMIC codes. In order to illustrate this important feature, now we redo some analyses
shown above but using the grid data directly generated by the cubegen utility in Gaussian.
Run below commands in console window of your system to produce density.cub and ESP.cub.
cubegen 0 fdensity H2O.fch density.cub 0 h
cubegen 0 potential H2O.fch ESP.cub 0 h
Then boot up Multiwfn and input

ESP.cub // This file contains ESP grid data. After loading it, the grid data will be stored in
memory
1 // Select real space function used to partitioning basins
2 // Generate the basins by using the grid data stored in memory
612
Now visualize located attractors in option 0:
Exit GUI, then input following commands

2 // Integrate real space functions in the basins
-1 // Use the grid data stored in external file as integrand
density.cub // This file contains electron density grid data
The result is very close to the one we obtained in Part 1 of present section. For example, the
electron population number in negative ESP region we get here is 0.339*2=0.678, while the
counterpart value we obtained earlier is 0.682.
At last, we select option 3 to calculate electric multipole moments for the basins. Because cube
file does not contain GTF (Gaussian type function) information, you will be prompted to input the
path of a file containing GTF information of present system, so that electric multipole moments can
be calculated. We input the path of the H2O.fch file, and then input -1, the electric multipole
moments of all basins will be immediately outputted on screen.
4.17.4 Basin analysis of electron density difference for H2O
In this example we analyze basins of electron density difference for H2O to quantitatively study
the electron density deformation during formation of the molecule.
Before doing the basin analysis, we need to generate grid data of electron density difference
first by main function 5, wavefunction file of all related elements must be available. Here we directly
use the set of atomic wavefunction files provided in Multiwfn package, namely copying "atomwfn"
subfolder in "example" folder to current folder, then during generating grid data of electron density
difference Multiwfn will automatically use them. There are several different ways to prepare atomic
wavefunction files, please recall Section 4.4.7 and consult Section 3.7.3.
After that, boot up Multiwfn and input:
examples\H2O.fch // Generated at B3LYP/6-31G** level
-2 // Obtain deformation property
3 // High quality grid. Because the variation of electron density difference is complicated,
using relatively high quality of grid is compulsory. Note that the "high quality grid" we selected
here only defines the total number of grids, and hence has different meaning to the one involved in
function 1 of basin analysis module
0 // After the calculation is finished, return to main menu
17 // Basin analysis module
613
2 // Generate the basins by using the grid data stored in memory (namely the grid data we
just calculated by main function 5)
Enter function 0 to visualize the result, you will see the left graph shown below. After deselect
"Show molecule", the graph will look like the right one
Positive (negative) part of electron density difference corresponds to the region where electron
density increases (decreases) after formation of the molecule. Light green spheres denote the
maxima of the positive part, while light blue ones denote the minima of the negative part.
If you feel difficult to imagine why the maxima and minima distribute like this, I suggest you
to plot plane maps for electron density difference. The left graph shown below is the electron density
difference map vertical to the molecular plane, while the right graph is the map in the molecular
plane.
By comparing the attractors with plane maps, it is evident that attractor 4 and 5 are the maxima
in the region where electron density is enhanced due to the formation of the O-H bonds. While the
presence of attractor 6 arises from the electron aggregation due to formation of the lone pairs.
Note 1: Attractor 6 is two-fold degenerate, namely as you can see, it corresponds to two attractors. This is
because the two attractors share the same value and they are very closely placed with each other.
Note 2: Attractor 8 does not have its counterpart in another side of symmetry plane. The reason of this problem
is that the grid quality employed is not high enough relative to the complicated characteristic of electron density
difference.
It is interesting to examine how many electrons are aggregated between C and H due to the
bond formation. There can be many ways to measure this quantity; the most reasonable one for
present case is to integrate the electron density difference in basin 4 or in basin 5. Let us do this now.
Choose function 2, and then select option 0 to take the grid data of electron density difference as
integrand. From the output we can find that the integral is 0.102 e.
614
If you would like to compare the attractors with the isosurface of electron density difference,
you can simply choose option -10 to return to main menu of Multiwfn, and then choose suboption
-2 in main function 13 to plot the isosurface of the grid data stored in memory, the attractors we
located will be shown together, as shown below (isovalue=0.05), where green and blue parts
correspond to positive and negative regions, respectively.
4.17.5 Study source function in AIM basins
Source function has been briefed in part 19 of Section 2.6. Commonly, bond critical point (BCP)
is taken as the reference point of source function when bonding problem is discussed. In this
example we calculate source function in AIM basins for ethane; in particular, based on source
function we will get the contribution from methyl group to the electron density at the BCP of its C-
H bond. Before calculating source function we should perform topology analysis first to find out
the position of the BCP.
examples\ethane.wfn // Optimized and produced at B3LYP/6-31G*
2 // Search nuclear critical points
3 // Search BCPs
0 // Visualize result, see below
615
Critical point 11 will be selected as the reference point of the source function. Of course,
selecting which BCP is completely arbitrary. Now close the GUI of topology analysis module
You'd better choose option 7 and then input 11 to check the electron density at this CP, because
theoretically the integral of source function in the whole space should equal to the electron density
at its reference point, therefore this value is important to examine if the integration of source
function is accurate enough. The electron density at CP11 is 0.276277.
From the information shown in the command-line window you can find the coordinate of CP11
is (0.0,-1.199262548,-1.909104063), copy it from the window to clipboard (if you do not know how
to do this please consult Section 5.4). Next, we will set CP11 as the reference point of the source
function. Although you can define reference point by "refxyz" parameters in settings.ini, there is a
trick can do the same thing, by which you needn't to close Multiwfn and then reboot it to make the
parameters take effect!
Input below commands
-10 // Return to main menu from topology analysis module
1000 // A hidden menu
1 // Set reference point
Paste the coordinate of CP11 to the window and then press ENTER button.
Enter GUI by choosing function 0, you will see
Now we integrate source function in the AIM basins. Input following commands
1 // Integrate a specific function with atomic-center + uniform grids
19 // Source function
The result is
Atom Basin Integral(a.u.) Vol(Bohr^3) Vol(rho>0.001)
1 (C ) 5 0.00361772 149.307 70.392
2 (H ) 8 0.00289355 468.732 50.058
3 (H ) 6 0.00280041 443.982 50.097
4 (H ) 7 0.00280276 418.343 50.095
5 (C ) 2 0.12224052 150.514 70.396
6 (H ) 1 0.12040142 475.383 50.054
616
7 (H ) 4 0.01051821 424.105 50.095
8 (H ) 3 0.01051742 450.072 50.097
Sum of above integrals: 0.27579202
Sum of basin volumes (rho>0.001): 441.284 Bohr^3
The sum of the integrals is very close to the electron density at CP13 (0.276277). The sum of
the integral in basin of 1, 2, 3 and 4 is 0.1204+0.1222+0.0105*2=0.2636, which represents the
integral in the space of methyl group and accounts for 0.2636/0.2758*100%=95.6% of the total
integral value, exhibiting that methyl group is the main source of the electron density of the BCP of
its C-H bonds.
4.17.6 Local region basin analysis for polyyne
Sometimes, the geometry of the system we studied is rather extended, for instance, polyyne
C14H2, which can be formally illustrated as
H1−C2≡C3−C4≡C5−C6≡C7−C8≡C9−C10≡C11−C12≡C13−C14≡C15−H16
If we are only interested in the electronic structure characteristic of local region in this system, by
properly setting up grid, basin analysis can be conducted only for the interesting region rather than
for the whole system to save computational time. As an example, in this section we will try to acquire
electron population number in the ELF basin of V(C7,C8) and V(C8,C9) with minimum
computational cost.
examples\polyyne.wfn // Optimized and produced under B3LYP/6-31G*
9 // ELF
8 // Set the grid by inputting center coordinate, grid spacing and box length
a8 // Take the position of atom 8 as box center
0.08 // Grid spacing (Bohr)
10,10,8 // Box length in X, Y and Z directions (Bohr). Note that current molecule
is aligned in Z-axis. Obvisouly, the larger the box, the longer the computational time
must be spent. While the box should not be too small, otherwise the basins of interest
may be truncated. Choosing appropriate box size highly relies on users' experience
After the calculation is finished, enter GUI by selecting option 0, you will see the
graph shown at the right side. Clearly, only several attractors near C8 are located. Basin
5 and basin 21 correspond to V(C7,C8) and V(C8,C9), respectively. Notice that although
attractor 1 and 6 are also located, due to their corresponding basins are not only large but
also close to box boundary, it can be expected basin 1 and 6 are severely truncated and
hence studying them are meaningless.
When you use above manner to study local region, you will always find there are
many grids travelled to box boundary. In present example, as shown in command-line
window, the number of such type of grids is 60668. You can visualize them by choosing
"Boun" in basin list of GUI, see the graph below
617
Now close GUI, select option 2 and then select 1, the result shows that the integral of electron
density in V(C7,C8) and V(C8,C9) are 2.78 and 5.02, respectively. Evidently, the bonding between
C8-C9 is much stronger than C7-C8, this is why the bond length of the former (1.236Ǻ) is shorter
than the latter (1.338Ǻ). Note that the electron population number in V(C,C) of acetylene is 5.37
(see Section 4.17.2), therefore it can be expected that C8-C9 is weaker than typical C-C triple bond,
mostly due to the electron global conjugation in polyyne.
The interested users can redo the basin analysis for the whole system with using the same grid
spacing (0.08 Bohr), the computational amount will be much larger than current example. For
V(C7,C8) and V(C8,C9), you will find the result does not differ from the one we obtained above
detectably.
4.17.7 Evaluate atomic contribution to population of ELF basins
In this section, I will use CH3NH2 as example to show how to obtain contribution of C and N
to population of V(C,N) ELF bond basin based on AIM partition of atomic space, this is useful to
examine bond polarity. You can also use the similar way to obtain atomic contribution to population
of any other kind of basins (e.g. LOL basin, ESP basin).
First, we need to generate a cube file named basin.cub, whose grid value corresponds to index
of ELF basins. Boot up Multiwfn and input
examples\CH3NH2.wfn
9 // ELF
Now enter option 0 to examine the basin index
618
Evidently basin 5 corresponds to V(N,C), which is the one we will study. Then close the GUI
and input
-5 // Export basin as cube file
a // Export basin.cub in current folder
Next, we generate AIM basins as usual, the grid setting must be exactly identical to basin.cub
1 // Regenerate basins
1 // Select real space function
9 // Use grid setting of another cube file, this is the safest way to ensure the grid data to be
generated has the same grid setting as basin.cub
basin.cub
0 // Check attractors
It is clear that the attractor index corresponding to N and C are 2 and 3, respectively. Then we
evaluate atomic contribution to population of the basins defined in basin.cub
9 // Then program loads basin.cub in current folder
2 // The index of the attractor corresponding to N
5 // The 5th ELF basin, i.e. V(N,C) basin
The result is 1.15866, namely N contributes 1.159 electrons to V(N,C) basin. Then input
3 // The index of the attractor corresponding to C
5 // The 5th ELF basin, i.e. V(N,C) basin
From the result we know that C contributes 0.463 electrons to the V(N,C) basin.
Since N contributes much more electrons than C to their ELF bond basin, it may thus be
concluded that C-N is a bond with significant polarity.
4.17.8 Calculating high ELF localization domain population and

volume (HELP, HELV)
The high ELF localization domain population and volume (HELP and HELV, respectively)
were proposed in ChemPhysChem, 14, 3714 (2013), it was shown that they are useful quantities in
619
characterizing lone pair electron and they have close relationship with molecular properties that
related to lone pairs, such as ionization potential and energy of frontier molecular orbitals, see the
ChemPhysChem paper for detail.The definitions of HELP and HELV are clearly illustrated in below
map, PH3 is taken as an instance.
From above map it can be seen that this method defines a region named high ELF localization
domain (HEL), which simultaneously satisfies three conditions:
(1) Electron density is larger than 0.001 a.u.
(2) Electron localization function (ELF) is larger than 0.5
(3) Every point belongs to the ELF basin corresponding to a lone pair
The volume and electron population of the HEL are denoted as HELV and HELP, respectively.
It is believed that this definition is better than (3) for representing lone pair area, because
insignificant region (i.e. outside vdW surface, which corresponds to  = 0.001 a.u.) and the region
without clear chemical meaning (i.e. ELF < 0.5) are excluded.
Now we use Multiwfn to calculate HELV and HELP for PH3. Boot up Multiwfn and input
examples\PH3.wfn // Generated at M06-2X/def2-TZVPP level, optimized at the same level
9 // ELF must be chosen to define basins if you intend to calculate HELP and HELV
Now we select option 0 to check attractor indices:
From above map, we can find attractor 5 corresponds to lone pair of the P atom.
620
Next, we input
10 // Calculate HELP and HELV
0 // Select basins and calculate their HELP and HELV
5 // The basin index corresponding the lone pair of the P atom
After a while, you will see:
Basin information: (constraints are not taken into account)
Population: 2.0719 Volume: 1561.1570 Bohr^3
High ELF localization domain population (HELP): 1.4858
High ELF localization domain volume (HELV): 80.5500 Bohr^3
As can be seen, HELP and HELV are 1.4858 and 80.55 Bohr3, respectively, which are very
close to the values in the original paper, namely 1.50 and 80 Bohr3. The marginal difference comes
from the fact that the calculation level we employed is not identical to the original paper, in addition,
the detail of numerical setting in our calculation must be somewhat different to the original paper.
From above output you can also find the population and volume of the ELF basin, which can
also be calculated via option 2 of basin analysis module.
4.18 Electron excitation analysis
Main function 18 of Multiwfn is very powerful, it is a collection of electron excitation analysis

methods and able to provide very deep insight into all aspects of electron transition characters. In
this section I will illustrate most of them using practical instances. Before following below examples,
please at least first read beginning of Section 3.21 to understand requirement on the input files.
Although I illustrate the analyses based on Gaussian output files, these functions are by no means
limited to Gaussian users!
4.18.1 Using hole-electron analysis to fully characterize electron

excitations
The hole-electron analysis module of Multiwfn is quite powerful, it is able to present very
comprehensive characterization for all kinds of electron excitations. If you are not familiar with
basic theories and ideas of hole-electron analysis, please read Section 3.21.1. The requirement on
input file for the hole-electron analysis has been described at the beginning of Section 3.21, please
carefully check it. Below I will employ two systems to illustrate the use of the hole-electron analysis,
the first one is a typical donor--acceptor system, the second one is a typical coordinate. The
Chinese version of this section is my blog article "Using Multiwfn to perform hole-electron analysis
to fully investigate electronic excitation character" (http://sobereva.com/434), in which there is an
additional example, namely studying Rydberg excitation of H2CO.
If hole-electron is employed in your work, please not only cite Multiwfn original paper but
also cite my this work: Carbon, 165, 461-467 (2020) DOI: 10.1016/j.carbon.2020.05.023, in which
hole-electron analysis is briefly described.
621
4.18.1.1 Example 1: NH2-biphenyl-NO2

The example in this section is quite long, please carefully and patiently read it. In this section
I will take the NH2-biphenyl-NO2 as example, its geometry is shown below.
In this system the biphenyl moiety behaves as a -linker, and it is well-known that the nitro
group and amino group act as electron acceptor and donor during electron excitation, respectively,
therefore it is expected that there must be some charge-transfer (CT) states corresponding to overall
electron displacement from the amino group side to the nitro group side.
Preparation
Here I assume that you are a Gaussian user (other quantum chemistry codes users can also
utilize the hole-electron analysis). We optimize the geometry at the B3LYP/6-31G* level, then carry
out TDDFT calculation using below settings (the input file has been provided as examples\excit\D-
pi-A.gjf), then five lowest singlet excited states will be evaluated. Note that IOp(9/40=4) must be
specified, the reason has been clearly mentioned at the beginning of Section 3.21. The CAM-B3LYP
is employed here because it is able to faithfully represent CT excitations.
%chk=D-pi-A.chk
# CAM-B3LYP/6-31g(d) TD(nstates=5) IOp(9/40=4)
After calculation, convert the .chk file to .fch/fchk file, the resulting .fchk file has been
provided as examples\excit\D-pi-A.fchk. The output file of this task has been provided as
examples\excit\D-pi-A.out.
Examining quantitative indices defined in the hole-electron analysis framework

Boot up Multiwfn and input below command
1 // Hole-electron analysis
examples\excit\D-pi-A.out // In fact you can also press ENTER button directly, because the
name of the .out file is identical to the .fchk file and they are in the same folder
1 // Study excitation between ground state (S0) and the first excited state (S1)
1 // Calculate distribution of hole, electron and so on as well as various indices
2 // Medium quality grid (this is suited for small and medium sized systems. For large systems,
you should use at least "high quality grid", or manually input a proper grid spacing)
Once calculation is finished, you will find below information on screen. The data (except for
the excitation energy) are calculated by grid-based integration. Clearly, for the same system, the
higher number of grids, the better accuracy of the data.
Integral of hole: 1.000343
Integral of electron: 0.999793
622
Integral of transition density: -0.000033
Transition dipole moment in X/Y/Z: 0.444063 -0.000186 -0.001753 a.u.
Sm index (integral of Sm function): 0.27369 a.u.
Sr index (integral of Sr function): 0.51896 a.u.
Centroid of hole in X/Y/Z: -4.531729 0.000425 0.003252 Angstrom
Centroid of electron in X/Y/Z: -4.010033 0.001760 0.002643 Angstrom
D_x: 0.522 D_y: 0.001 D_z: 0.001 D index: 0.522 Angstrom
Variation of dipole moment with respect to ground state:
X: -0.985930 Y: -0.002523 Z: 0.001150 Norm: 0.985934 a.u.
RMSD of hole in X/Y/Z: 1.443 1.160 0.437 Norm: 1.902 Angstrom
RMSD of electron in X/Y/Z: 1.596 0.974 0.634 Norm: 1.974 Angstrom
Difference between RMSD of hole and electron (delta sigma):
X: 0.153 Y: -0.186 Z: 0.197 Overall: 0.072 Angstrom
H_x: 1.520 H_y: 1.067 H_z: 0.536 H_CT: 1.520 H index: 1.938 Angstrom
t index: -0.998 Angstrom
Hole delocalization index (HDI): 22.68
Electron delocalization index (EDI): 17.11
Ghost-hunter index (def 1): -18.018 eV, 1st/2nd terms: 9.583 27.601 eV
Ghost-hunter index (def 2): -17.466 eV, 1st/2nd terms: 10.135 27.601 eV
Excitation energy of this state: 3.907 eV
In the output, the "Integral of hole" and "Integral of electron" are the integrals of hole and
electron over the whole space, respectively, they should be exactly 1.0 in theory. However, due to
the unavoidable numerical integration error, the calculated values have slight deviation from 1.0.
Since the deviation is extremely small, we can say that for the current excited state of the current
system, the grid setting we used is completely appropriate.
Note: If you find the integral of hole or electron deviates from 1.0 evidently, then the outputted indexes will be
unreliable. There are three possibilities: (1) You forgot to use IOp(9/40=3 or 4) (2) The grid quality is too low (3)
The extension distance is not large enough, therefore the spatial region of the grid points does not fully cover the
main distribution region of hole or electron (when Rydberg excitation is investigated, the default extension distance
should always be enlarged).
The rest terms in the above output in turn are: The integral of transition density over the whole
space (ideal value is 0), transition electric dipole moment, Sm and Sr indices, centroid coordinate of
hole and electron, D and D indices, X/Y/Z components and norm of variation of excited state dipole
moment with respect to the ground state one, RMSD () of hole and electron, Δ and Δ indices,
Hλ/HCT/H indices, t index, hole and electron delocalization index, Ghost-hunter index, excitation
energy (which is loaded from Gaussian output file directly).
Note that the "Ghost-hunter index (def 1)" is the definition of this quantity in its original paper, while the one
with "(def 2)" suffix is the form modified by me, which should be more reasonable. The two values after the "1st/2nd
terms" are the first term (dependent on configuration coefficients) and second term (namely -1/D) of the Ghost-
hunter index, respectively.
The transition dipole moment outputted above is obtained by integrating evenly distributed grids of transition
dipole moment density. It can also be directly read from the Gaussian output file, the X/Y/Z components are 0.4427,
-0.0005, -0.0012 a.u., which are very close to the ones output by Multiwfn, namely 0.444063, -0.000186, -0.001753.
This observation further reflects that the grid setting we employed is appropriate.
For the S0→S1 excitation under current study, from the above output, it can be seen that the D
index is merely 0.522 Å, which is obviously a very small value since it is even less than half length
of a typical C-C bond. The Sr index reaches 0.519 (the theoretical upper limit is 1.0), which is a
large value, implying that about half part of hole and electron has perfectly matched. So, by simply
623
examining Sr and D indices, we are already able to conclude that this excitation should be a typical
local excitation (LE). Then let us look at the t index, its total value is -0.998, which is much less
than 0, meaning that there is no significant separation of hole and electron distributions, futher
implying that this excitation should be attributed to LE type.
Visual study of various real space functions in the hole-electron framework

Now you should see the post-processing menu on the screen. The meaning of each option is
self-explanatory. Please read through each option carefully. Here we choose option 3, we will see
distribution of hole and electron at the same time:
In above figures, green represents the electron distribution, and blue represents the hole distribution,
isovalue has been set to 0.005. Both the hole and electron appear almost exclusively in the nitro
group, so there is no doubt that S0→S1 is a LE excitation, well verifying our conclusion based on
the D, Sr, and t indices. In addition, according to the above hole distribution map, the hole appears
to be composed of lone pair orbitals of oxygens since there is one lobe on each side of each oxygen.
Electron distribution has a nodal plane along the nitro group, therefore we can infer that the electron
distribution should be composed of * orbital. Now we can draw the conclusion that that S0→S1 is
a LE excitation with n→* feature.
Then close the graphical window and select option 8 to visualize Chole and Cele, which are
transformed from hole and electron distributions respectively to make their distribution behavior
smoother. The isosurface map is show below (In order to see clearly, transparent style is used).
As can be seen, the graph of Chole and Cele look obviously more intuitive, they are very sleek and do
not have any nodal character as hole and electron. Therefore using Chole/Cele map to replace the
hole/electron map is a good choice in many cases.
Next, let us take a look at the overlap function of hole and electron, namely the Sr function.
Close current graphical window, select option 4 in the post-processing menu, and then select option
2 to display the Sr function, you will see below map (isovalue is set to 0.005)
624
From the graph one can clearly find where the hole and electron overlap significantly. As can be
seen, around each oxygen there are four regions where the hole and electron are highly overlapped.
It is easy to understand why the Sr graph looks like this by comparing the hole and electron
isosurfaces shown earlier.
Then close the window and select option 7, charge density difference (CDD) between the
excited state and the ground state will be shown, see below. In this map, the isovalue is set to 0.005,
green and blue correspond to increase and decrease of the excited state density with respect to the
ground state density, respectively.
The CDD map and the map simultaneously showing hole and electron distributions (referred
to as "hole&electron map" later) are similar, but there are also differences. The key difference is that
in the CDD map, the hole and electron have been largely cancelled in their overlapping region; in
contrast, in the hole&electron map, the overlapping between hole and electron can be faithfully
exhibited. I think the hole&electron map is more useful than the CDD map to investigate the
intrinsic characteristics of electron excitation because it directly exhibits the pristine distribution of
hole and electron.
By the way, if we select option 18 in the post-processing menu, the program will start to
calculate the Coulomb attractive energy (also known as exciton binding energy) between the hole
and electron. The calculation is quite time-consuming even for medium sized system, so please wait
patiently. The final output is:
Coulomb attractive energy: 0.287031 a.u. ( 7.810524 eV )
Important note: By using VMD software, you can plot above mentioned functions in much
better quality with only a few steps. I strongly suggest you check the example given in Section
4.A.14 on how to do this.
Examining quantitative contributions to hole and electron

Next, I demonstrate how to evaluate contribution of MOs to hole and electron. After selecting
option 0 in the post-processing menu to return to the hole-electron analysis interface, we select
subfunction 2 and input an outputting threshold. Here we input 1, then MOs with contribution to
hole or electron higher than 1% will be shown on screen:
MO 52, Occ: 2.00000 Hole: 96.207 % Electron: 0.000 %
625
Sum of hole: 100.001 % Sum of electron: 100.001 %
It can be seen from the data that MO52 is absolutely dominant for hole, it contributes as high as
96.2%, while electron is mainly composed of MO57, with a contribution of 85.4%. This observation
implies that if one discusses electron excitation solely based MO52 and MO57, although in this case
the electron excitation can be qualitatively described, there are still non-negligible deviations. The
"Sum of hole" and "Sum of electron" shown above are the sum of the contributions of all orbitals to
hole and electron, respectively (including the terms not outputted), these two values in principle
should be exactly 100%, but currently there are 0.001% error. Such a small error can be completely
ignored, it comes from the fact that not all configuration coefficients are printed by Gaussian (only
configuration coefficients greater than 0.0001 are requested to be outputted during Gaussian
calculation via IOp(9/40=4))
Then we check contribution of atoms or fragments to hole and electron. Select subfunction 3
in hole-electron analysis interface, you will see
The number of non-hydrogen atoms: 16
Contribution of each non-hydrogen atom to hole and electron:
18(C ) Hole: -0.01 % Electron: 1.16 % Overlap: 0.00 % Diff.: 1.17 %
21(N ) Hole: 2.39 % Electron: 33.90 % Overlap: 9.00 % Diff.: 31.52 %
22(O ) Hole: 46.18 % Electron: 24.38 % Overlap: 33.55 % Diff.: -21.80 %
23(O ) Hole: 46.12 % Electron: 24.39 % Overlap: 33.54 % Diff.: -21.73 %
24(N ) Hole: 0.46 % Electron: 0.15 % Overlap: 0.26 % Diff.: -0.32 %
Since hydrogen atoms generally do not participate in electron excitation of chemical interest,
only the information of non-hydrogen atoms is outputted, including the atomic contributions to hole,
electron, hole-electron overlap, electron-hole difference (i.e. CDD). The indices of the atoms in the
nitro group are 21, 22, and 23, it can be seen from the data that the two oxygens of the nitro group
contribute most to the hole, the sum of their contributions is 246.192%. The spatial delocalization
of the electron is relatively stronger, the three atoms in the nitro group contribute a total of
224.4+33.983%, the rest part of electron is basically contributed by the atoms in the biphenyl
moiety.
Although the distribution characteristics of hole and electron can be examined by visualizing
isosurface map of hole and electron, the observed isosurfaces are obviously dependent on the choice
626
of isovalue. So, it is impossible to fully display the hole and electron distribution in all regions by
only one image. On the contrary, the quantitative atomic contributions given above are very
definitive.
The sum of "Diff." of 21N, 22O and 23O is about -12%, which shows that the integral value
of the density difference (CDD) in the nitro group is -0.12, revealing that the nitro moiety lost 0.12
electrons during electron excitation, and some of them transferred to the biphenyl moiety (if you
want to investigate charge transfer amount between specific fragments, it is recommended to use
the IFCT method, as illustrated in Section 4.18.8)
Showing atomic contributions to hole and electron in terms of heat map

We can also plot the atomic contributions as heat map, so that the major character can be
immediately and easily captured. Select "4 Set stepsize between labels in X axis" in current menu
and then input 1 to change the step of abscissa to 1, and then select "1 Plot hole/electron composition
as heat map", you will see below map immediately
In the figure, the numbers in the abscissa are indices of non-hydrogen atoms. This figure
describes the contribution of each non-hydrogen atom to hole, electron and their overlap by color
(e.g. 0.4 corresponds to 40%). For example, based on the list of contributions of non-hydrogen
atoms to hole/electron that shown earlier, we can know that the position 16 in the abscissa of the
figure actually corresponds to atom N24. In order to more conveniently find the correspondence
between the indices in the abscissa and actual atomic indices, you can open
corresponding .fch/.gjf/.out file by GaussView, Select "Edit" - "Atom List", then select "Edit" -
"Reorder" - "All atoms: Hydrogens Last", then you will see below graph in GaussView, in which
indices of all hydrogen atoms have been placed behind indices of non-hydrogen atoms. Clearly, in
this case the atomic indices in the molecular structure directly corresponds to the indices in the
abscissa of the heat map.
From indices in the above map, we find that the position of 13, 14 and 15 in the heat map
correspond to the two oxygens and one nitrogen atoms in the nitro group, the position 16
corresponds to the nitrogen in amino group, the rest are carbons in the biphenyl moiety. In the heat
map, the row corresponding to hole clearly revealed that the hole is almost solely contributed by the
two oxygens in the nitro group, since corresponding matrix elements are red (large value). Electron
is also mainly contributed by the nitro group, but other molecule regions also have non-negligible
627
contributions, this is why in the row corresponding to electron, blue color appears in the matrix
elements other than positions 13~15. The information conveyed by colors of the row corresponding
to overlap is that there is significant overlap between hole and electron on the oxygens of the nitro
groups, while the overlap in other areas of the systems is far from being so remarkable.
There are many options in the hole/electron composition analysis interface, they can be used
to adjust plotting effect of the heat map, save the heat map as a image file, switch whether to include
the hydrogen atoms into the heat map, and export the data to he_atm.txt in current folder so that you
can draw the heat map yourself in other programs such as Origin. These options will not be
explained one by one here, please try it yourself.
Investigating fragment contributions to hole and electron

The option -1 in the hole/electron composition analysis interface is important, it is used to load
fragment definition, then fragment contribution to hole and electron will be shown, and fragment-
based heat map could be drawn, which makes the discussion significantly more convenient. Here
we divide the system into four fragments in the following way.
Select option "-1 Load fragment definition", and then input

4 // Define four fragments
21-23 // The atomic indices of fragment 1 (the nitro group)
11-20 // The atomic indices of fragment 2 (the benzene neighbouring to nitro group)
1-10 // The atomic indices of fragment 3 (the benzene neighbouring to amino group)
24-26 // The atomic indices of fragment 4 (the amino group)
You will see below output immediately
Contribution of each fragment to hole and electron:
# 1 Hole: 94.68 % Electron: 82.67 % Overlap: 88.47 % Diff.: -12.01 %
# 2 Hole: 3.20 % Electron: 15.67 % Overlap: 7.09 % Diff.: 12.47 %
The data shows that 94.68% of hole is located on the nitro group, while 82.67% and 15.67%
of electron are located on the nitro group and neighbouring benzene ring, respectively. The degree
of overlap between hole and electron on the nitro group is about 90%. Since "Diff." of the nitro
group is -12.01%, and the excitation under current investigation is a single electron excitation,
therefore, it can be said that the electron on the nitro group is reduced by 0.1201 during the electron
excitation process, while the benzene ring neighbouring to the nitro group gained 0.1247 electrons.
Then, if you want to make representation of the above data more intuitive, we can select option 1 to
draw the fragment-based heat map. The abscissa at this time corresponds to fragment index, as
shown below:
628
It can be seen from the figure that the spatial distribution range of electron is larger than hole.
Collective comparsion of all electron excitations

At this point, various analyses in the hole-electron framework for the S0→S1 excitation of the
NH2-biphenyl-NO2 system have been completely completed. If you also want to analyze other
excited states, you should return to the menu of main function 18 by option 0, enter the hole-electron
analysis function again, and then select the corresponding excited state. Here we put together the D,
Sr, H, t index and hole-electron Coulomb attraction energy of all the five excited states calculated
in this system. The hole delocalization index (HDI) and electron delocalization index (EDI), which
have not been discussed earlier, are also given:
D(Å) Sr H(Å) t(Å) Ecoul(eV) HDI EDI
S0→S1 0.52 0.52 1.94 -1.00 7.81 22.7 17.1
S0→S2 3.48 0.65 3.15 0.56 4.71 7.2 9.5
S0→S3 0.57 0.55 1.70 -0.68 8.54 19.7 17.2
S0→S4 0.97 0.87 2.88 -1.55 5.56 7.0 7.0
S0→S5 0.54 0.87 2.93 -2.06 5.56 6.9 7.1
Below are hole&electron map, Chole&Cele map and Sr function map of all the five excitations.
The isovalues are set to 0.003. It can be seen that the Chole&Cele map can always display the main
distribution features of the hole&electron map in a clearer and more intuitive way. However, many
details are lost during the transformation; for example it is impossible to determine the specific type
of the electron excitation (such as n-*, -*) solely based on the Chole&Cele map.
Now we look at the indices together with the isosurface graphs. For the D index, only the
629
S0→S2 value has very large (3.48 Å), so it can obviously be considered as CT excitation. Indeed,
from above graph it can be seen that the distance between the center of the blue and green isosurfaces
(namely centroids of Chole and Cele) is large. While for other excitations, the centers of the blue and
green isosurfaces are close together, therefore they should be regarded as LE excitations.
Then we examine the Sr index. We find that the Sr indices of all excited states are relatively
large. In particular, the values of the S0→S4 and S0→S5 are rather large, up to 0.87, the main reason
is that these two excitations are highly localized -* type of excitation on the benzene ring. It is
worth to mention that although S0→S1 is also a highly localized excitation, its Sr (0.52) is even
smaller than the Sr (0.65) of S0→S2, which is a CT excitation. The reason why the Sr index of the
S0→S1 is not as large as expectation is not difficult to understand. As mentioned earlier, S0→S1
shows n→* feature, the main body of lone pair is on the NO2 plane, while * orbital has a nodal
plane on the NO2 plane, hence the overlap of hole and electron should be limited.
Next, look at the H index, which reflects the breadth of the average distribution of hole and
electron. It can be seen from the hole&electron map that both the hole and electron of S0→S1 and
S0→S3 are distributed in a local region, this is why their H indices are not large. Since the
distribution of hole and electron corresponding to the excitations from S0 to S2, S4 and S5 are
evidently wider than S0→S1, their H indices are evidently larger.
One can see that only the t index of S0→S2 is a slightly positive value, indicating that the
separation of hole and electron is obvious, so it is more reasonable to consider S0→S2 as a CT
excitation. The t indices corresponding to the excitations from S0 to other excited states are evident
negative values, suggesting that degree of separation of their hole and electron is very low.
By comparing the hole and electron isosurface maps, we can find that the HDI and EDI indices
indeed nicely quantified the uniformity of spatial distribution (i.e. degree of delocalization) of hole
and electron, respectively. It can be seen that both hole and electron of S0→S1 and S0→S3 are
highly localized, corresponding to large calculated value of HDI and EDI. In contrast, the hole and
electron distributions of S0→S2/S4/S5 are evidently more delocalized, this point is faithfully
revealed by their relatively small HDI and EDI values.
The hole-electron Coulomb attractive energy given in the table is closely related to the electron
excitation characteristics, and the most influential factor should be the D index. It is easy to
understand that the larger the D index is, the farther the distance between the main distribution
regions of hole and electron, and thus the weaker the Coulomb attractive energy. From the data, it
is indeed found that Coulomb attraction energy of S0→S2 (the only CT excitation) is the smallest
one of that of all the five electron excitations. While for the excitations of S0→S1 and S0→S3,
since their D indices are very small and according to the H index the spatial extent of their hole and
electron is very narrow, one can easily imagine that the corresponding Coulomb attraction should
be very strong. Indeed, as can be seen from the previous table, their hole-electron Coulomb
attraction energies are the most negative ones (-7.81 and -8.54 eV).
Combined with the above isosurface maps and quantitative data, we can unambiguously
identify the characteristics of the five excitations:
·S0→S1：LE excitation of n-* type on the nitro group
·S0→S2：CT excitation of -* type from amino group towards nitro group
·S0→S3：The same as S0→S1
·S0→S4：-* LE excitation occurring on the benzene ring attached to the nitro group
·S0→S5：-* LE excitation occurring on the benzene ring attached to the amino group
630
Here, the heat maps that exhibit contributions to the hole, electron and overlap from all
fragments for all the five excitations are given together. In order to facilitate parallel comparison,
the color scale for all excited states is uniformly set to 0.0~1.0.
From these heat maps, one can immediately make clear where the excited electrons come and where
they go by viewing color of the matrix elements. For example, from the graph of S0→S2, one can
easily recognize that the excited electron is mainly orignated from fragment 3 (the benzene attached
to amino group), most of them is transferred to fragment 1 (the nitro group), and a smaller part is
transferred to fragment 2 (the benzene attached to the nitro group). Another example, from the
S0→S4 map one can find that the excited electrons come from fragment 2, after excitation most of
them remain in fragment 2, but a few of them transferred to fragment 1, and the overlap between
hole and electron on fragment 2 is significantly higher than the other regions.
It is worth to note that the detailed characteristics of charge transfer between fragments during
electron excitation can be even better revealed by the IFCT method, see example in Section 4.18.8.
Plotting grid data in VMD

The grid data of hole, electron, Cele, Chole and so on can be exported to cube files by
corresponding options in the post-processing menu, and then you can render them in VMD program
to get better visualization effect. If you do not know how to do, please refer to the description of the
operations in VMD at the end of Section 4.18.3. Furthermore, one can first use VMD to plot
isosurface map for Cele and Chole, and then input such as below command to draw centroids of hole
and electron as purple and orange spheres to make the graph more informative
draw color purple
draw sphere {1.411500 -0.007015 -0.025494} radius 0.25 resolution 20
draw color orange
draw sphere {-2.069784 -0.000346 -0.000830} radius 0.25 resolution 20
For the S0→S2 excitation, the graph plotted in above way in VMD is shown below
631
In order to make understanding easier, an arrow is appended on the graph to highlight the CT
direction, the D index is also labelled together to make the graph more informative.
Tip: Obtaining a variety of indices for a range of excited states via script
By Linux shell script, a variety of indices for a range of excited states can be obtained at once.
For example, we want to obtain all indices for excitations of S0→S1,S2,S3, we should copy the the
input files D-pi-A.fchk and D-pi-A.out as well as the all_index.sh from the "examples\excit" folder
to a proper folder, then enter this folder in the Linux terminal, run chmod +x ./batch.sh to add
executable permission, then run ./all_index.sh to execute the script. Each excitation will be analyzed
in turn, the status will be shown on screen until the appearence of "Finished!". Then open the
resulting result.txt file in current folder, you will find
1 Sr index (integral of Sr function): 0.51896 a.u.
1 D_x: 0.522 D_y: 0.001 D_z: 0.001 D index: 0.522 Angstrom
1 RMSD of hole in X/Y/Z: 1.443 1.160 0.437 Norm: 1.902 Angstrom
[ignored...]
This summary of all indices. You can easily modify the script to meet your practical
requirement, if you do not familar with this point, please check Section 5.3.
4.18.1.2 Example 2: Ru(bpy3)2+ cation in water

Below we examine several excited states of the Ru(bpy3)2+ cation complex in water by hole-
electron analysis based on TDDFT output.
632
The corresponding Gaussian input file has been provided as examples\excit\Ru(bpy3)2+.gjf,

please calculate it by Gaussian. If you want to directly obtain the resulting output file and .fchk file,
you can download them at http://sobereva.com/multiwfn/extrafiles/Ru_bpy3_2+_TDDFT.zip. As
you can see from the .gjf file, the keywords used are B3LYP/genecp TD(nstates=50) scrf
IOp(9/40=3), where scrf requests Gaussian to employ IEFPCM solvation model to represent water
environment. Since this system is not small, and as many as 50 excited states are evaluated, in order
to avoid too high calculation time in hole-electron analysis and too large Gaussian output file,
IOp(9/40=3) is used instead of the IOp(9/40=4) employed in last section, the analysis accuracy at
this time is still completely sufficient.
We arbitrarily select three excited states to perform hole-electron analysis, the results are
D (Å) Sr H (Å) t (Å) hole (Ru%) ele (Ru%) MLCT(%)
S0→S24 0.30 0.71 2.73 -1.35 77.3 19.6 57.7
S0→S37 0.11 0.84 3.52 -2.10 16.9 8.8 8.1
S0→S40 0.13 0.71 2.00 -1.05 80.3 42.4 38.0
All D indices in this table are very small, while all Sr indices are fairly large. The main reason
is that the current molecule is a symmetric system, thus the CT transitions are multiple directional.
The MLCT(%) in the table denotes percent of metal-to-ligand charge transfer character, which can
be easily evaluated in terms of substracting the percentage of metal in hole (namely hole(Ru%)) by
that in electron (namely ele(Ru%)). Notice that, properly speaking, what we obtained is net MLCT
percentage, it has been somewhat cancelled with LMCT (ligand-to-metal charge transfer).
Below is hole&electron map of S0→S24 excitation with isovalue of 0.002. Since the hole and
electron distributions have a large overlap, for the sake of clarity, the isosurfaces of hole and electron
are given separately.
633
Combining the graph with the quantitative data in the table, it is clear that the S0→S24
excitation not only has metal-centered (MC) character, namely electrons of the metal are excited
into the metal's own empty orbital, but also have evident MLCT character. As shown in the
hole&electron isosurface map, the main body of both the hole and electron is on the metal, and it
can be seen that the isosurface of electron also has a large portion on the ligand. The calculated
percentage of MLCT feature is 77.3-19.6 = 57.7%, this value should be said to be very consistent
with the information conveyed by the hole-electron isosurface map. Note: The hole also has non-
negligible distribution on the ligand, which is 100%-77.3% = 22.7%. The reason why the hole
isosurface is invisible on the ligands is because its distribution is very diffuse in this region, the hole
on the ligands can only be clearly seen by decreasing the isovalue to a smaller value, such as 0.0005.
Then look at S0→S37 excitation. From the hole&electron isosurface map shown below we can
see that main body of both hole and electron is located at one of the ligands, therefore it is no doubt
that this is a LC (ligand-centered) excitation, corresponding to the case that electrons excited from
ligand to its own * orbitals. Because there is also hole and electron distribution on the metal,
therefore this excitation also shows some MLCT character, which is calculated to be 16.9% - 8.8%
= 8.1%.
Finally, look at the S0→S40 excitation, whose hole and electron isosurfaces are shown at right
part of above map. From the hole isosurface and the percent of hole on metal shown in the previous
table (16.9%), it is found that its hole distribution character is very similar with S0→S24, but the
amount of electron on the ligand is obviously not as large as S0→S24, only tiny part of it is
distributed on the four nitrogens directly coordinating with Ru, so the MLCT feature of S0→S40 is
634
conspicuously weaker than S0→S24. In contrast, its MC characteristic is definitely higher than
S0→S24.
By the way, for coordinate systems, if you want to obtain MLCT, LMCT, LLCT, MC, LC
separately, you should resort to the IFCT analysis illustrated in Section 4.18.8.
We already know that the electron in the excitations from S0 to S24 and S40 mainly sources
from Ru atoms, but how to unveil which are the atomic orbitals that electron excited from? Although
it can be more or less judged from the isosurface map of the hole, there is still some subjectivity. To
figure this out, we can calculate contributions of basis functions to hole and electron. For example,
after entering the hole-electron analysis function and selecting the 40th excited state, we select "4
Show basis function contribution to hole and electron" and then input an outputting threshold, such
as 2, then information of the basis functions contributing to hole or electron higher than 2% will be
printed:
Basis Type Atom Shell Hole Electron Overlap Diff.
22 D 0 1(Ru) 12 23.81 % 0.00 % 0.33 % -23.81 %
23 D+1 1(Ru) 12 8.90 % 13.63 % 11.01 % 4.72 %
24 D-1 1(Ru) 12 9.06 % 25.12 % 15.09 % 16.05 %
25 D+2 1(Ru) 12 13.96 % 6.94 % 9.84 % -7.02 %
26 D-2 1(Ru) 12 14.06 % 15.57 % 14.80 % 1.51 %
27 D 0 1(Ru) 13 3.66 % 0.00 % 0.06 % -3.66 %
30 D+2 1(Ru) 13 2.32 % -0.01 % 0.00 % -2.32 %
31 D-2 1(Ru) 13 2.33 % -0.03 % 0.00 % -2.35 %
Sum of above printed terms: 78.10 % 61.22 % 16.88 %
It can be seen from the data that the main contribution to the hole is the D basis function of the
Ru atom. The SDD pseudopotential basis set we currently used only describes 4s, 4p, 4d and 5s
electrons for Ru, so the excited electron of S0→S40 excitation must come from the 4d atomic
orbitals. The basis functions of Ru that have contributions to electron are also D type. Therefore
now we know that the MC component in the S0→S40 corresponds to d-d transition on Ru.
Notice that this is not the only way to determine which orbitals the excited electrons come from and move to,
for example you can also use Multiwfn to perform NTO analysis and examine the pattern of the NTO pair with
largest eigenvalue. However, if the electron excitation under study cannot be well described by any pair of NTOs,
then this approach will not work. In contrast, the hole-electron analysis does not have any limitation.
4.18.2 Illustration of transition density (matrix) and transition dipole

moment density (matrix) analysis
The transition density and transition dipole moment density are very important quantities
involved in electron excitation analysis, they can be studied in the form of real space functions by
hole-electron module, or studied in Hilbert space by plotting as colored matrix maps (also known
as heat maps). The examples in Sections 4.18.2.1, 4.18.2.2 and 4.18.2.3 will illustrate these analyses.
In addition, although the commonly studied transitions are those from ground state to excited states,
the these quantites between two excited states are also useful in some special studies, Section
4.18.2.4 will mention how to realize this.
More discussions about these topics can be found from my blog article: "Using Multiwfn to
plot transition density matrix and charge transfer matrix to investigate electron excitation
characteristics" (in Chinese, http://sobereva.com/436).
635
4.18.2.1 Analyzing transition density and transition dipole moment density

in real space
Theory of real space function form of transition density, namely T(r), has been introduced as
"Theory 4" in Section 3.21.1.1, isosurface map of T(r) is capable of revealing apparent coherence
region between hole and electron. While the real space function form of transition electric dipole
moment density, namely Tx(r), Ty(r) and Tz(r), are able to exhibit contribution of various regions to
transition electric dipole moment (Dx, Dy, Dz), this point has been introduced as "Theory 5" of
Section 3.21.1.1. In this section, N-phenylpyrrole will be taken as instance to illustrate this kind of
analysis, involved files are completely identical to those utilized in the example in Section 4.18.1.
examples\excit\N-phenylpyrrole.fch // The .fch file yielded by Gaussian TDDFT task
1 // hole-electron analysis module
examples\excit\N-phenylpyrrole.out // The output file of Gaussian TDDFT task with
IOp(9/40=4) keyword
1 // Analyze electron transition from ground state to the 1st excited state (S0→S1)
1 // Visualize and analyze hole, electron, transition density and so on
Now you can find below information in the output, they are Dx, Dy and Dz evaluated based on
integrating the grid data
Transition dipole moment in X/Y/Z: -0.000021 -0.000045 1.767332 a.u.
It is worth to note that these values are very close to those printed in Gaussian output file, as shown
below (line 773 of N-phenylpyrrole.out), indicating that the grid quality we currently employed is
good enough
state X Y Z Dip. S. Osc.
1 0.0000 0.0000 1.7813 3.1729 0.3935
...[ignored]
We choose option 5 in the post-processing menu to show isosurface of transition density, you
will then see left map in the image shown below, which exhibits the transition density in real space
representation. If the transition density is multiplied by negative of Z coordinate variable, then we
will obtain Z component of transition dipole moment density, namely Tz(r). To visualize it, we close
current GUI window and choose "6 Show isosurface of transition dipole moment density" and then
select "3: Z component", you will see right graph below
636
From the above T(r) map, we can see that hole and electron have strong coherence everywhere,
implying that distribution of both hole and electron covers the whole molecule (this can be easily
further confirmed by visualizing hole and electron distributions, as illustrated in Section 4.18.1).
From the right map shown above one can see that its positive (green) part is obviously larger than
negative (blue) part, recall that the integral of Tz(r) over the whole space is just the Z component of
transition electric dipole moment (Dz), this observation explains why Dz of S0→S1 is an large
positive value (1.7813 a.u.). If you have interesting, you can try to plot Tx(r) or Ty(r) maps to
interpret why the electron excitation under study has vanished Dx and Dy.
It is important to note that the S0→S1 of current system is relatively special, namely its
transition electric dipole moment vector (D) just points towards Z-axis. However, for most practical
cases, the D is not parallel to any of the three Cartesian axes, in this case we are not able to directly
study its source in terms of visualization of any of Tx(r) or Ty(r) or Tz(r). Fortunately, it is quite easy
to reorientate the molecule so that the D exactly points towards a selected Cartesian axis, thus
making the above analysis feasible. See Appendex 2 of Section 4.A.7 on how to realize this.
Next, we check isosurface of Tz(r) for S0→S4 excitation. We first return to menu of main
function 18, then repeat above steps, finally you will see
637
The green and blue isosurfaces occupy the same amount of space, indicating that positive and
negative contributions to Dz are exactly the same, this is why Dz of S0→S4 is zero. Note that the
Tz(r) almost solely distributes on the pyrrole region, this is because this electron excitation
corresponds to local excitation on pyrrole moiety (as shown in Section 4.18.1).
You may have felt that visual study of transition dipole moment density is interesting and useful;
indeed, via this way you can clearly identify contribution to transition dipole moment from different
molecular regions. The cube file of the T(r) as well as Tx(r), Ty(r) and Tz(r) can be exported via
post-processing menu, so that you can also plot them using other visualization software such as
VMD (Section 4.A.14), or use such as basin analysis module (Section 3.20) or domain analysis
module (Section 3.200.14) to further quantify their distributions, or plot them as plane map via main
function 4 (using interpolated function based on loaded grid data, i.e. user-defined function -1).
Visualizing transition magnetic dipole moment density

The transition dipole moment we discussed above is transition electric dipole moment. There
are also other kinds of transition dipole moments, such as transition velocity dipole moment and
transition magnetic dipole moment. Multiwfn is also capable of calculating transition magnetic
dipole moment and plotting the corresponding density isosurface map, related knowledges can be
found in "Theory 5" of Section 3.21.1. Since this quantity is less important than transition electric
dipole moment, I will not discuss it deeply, but only give a simple example. Using the .out and .fch
files of N-phenylpyrrole, we first enter hole-electron module and select the second excited state,
then input
-1 // By default, transition magnetic dipole moment density is not calculated by option 1 of
hole-electron module for saving time, we select this option now to make option 1 also calculate this
quantity
Once the calculation is finished, from screen you can find the transition magnetic dipole
moment evaluated based on the grid data:
Transition magnetic dipole moment in X/Y/Z: -0.503315 0.000119 -0.000177 a.u.
Then we select option 9 and select the component of transition magnetic dipole moment density
638
that you are interested in, after that you will see the corresponding isosurface map. Below map is X
component of transition magnetic dipole moment density plotted under isovalue of 0.005. The
relatively larger blue region compared to green region explains why the X component of transition
magnetic dipole moment is a negative value (-0.503 a.u.).
4.18.2.2 Plotting and analyzing transition density matrix (TDM)

Note: There is a video illustrating the procedure of using Multiwfn and Origin to plot the transition density
matrix, see https://youtu.be/JPlZk4Aa6bQ.
The transition density matrix (TDM) has been carefully introduced in Section 3.21.2, heat map
of TDM is particularly useful in understanding the nature of electron excitation, please read Section
3.21.2 first. In this section, we will analyze transition character of a linear system of donor--
acceptor type by means of the TDM heat map. The molecular structure is shown below.
The Gaussian output file as well as .fchk file can be found in "examples\excit\NH2_C8_NO2" folder,
the keyword is CAM-B3LYP/6-31G* IOp(9/40=4) TD(nstates=10).
It is worth to note that the TDM is closely related to the charge transfer matrix defined in hole-
electron analysis framework. Example of plotting heat map of charge transfer matrix is given in
Section 4.18.8.2. Usually heat maps of TDM and charge transfer matrix are very similar to each
other and convey basically the same information.
Atom transition density matrix

First, we plot heat map of "atom TDM", namely the index of the TDM corresponds to atom
index. Commonly, this kind of map is only suitable for studying chain-like systems such as present
639
molecule, otherwise the indices in the heat map will be difficult to be mapped to actual atoms in the
molecule. For systems with more complicated shape, typically "fragment TDM" should be plotted
instead, it will be described later.
Note that in "atom TDM", hydrogens are commonly ignored because they rarely contribute to
excitations of chemical interest. Therefore, before the excited state calculation of present molecule,
the indices of the hydrogens have been moved to the back of heavy atoms. As can be seen from the
molecular structure graph shown above, the range of heavy atoms is 1~12, while that of hydrogens
is 13~22.
Here we first study S0→S1 transition of this system. The below used fchk and out files are
yielded using CAM-B3LYP/6-31G* IOp(9/40=4) TD(nstates=10) keywords. Boot up Multiwfn and
input
examples\excit\NH2_C8_NO2\NH2_C8_NO2.fchk
2 // Plot heat map of transition matrix
examples\excit\NH2_C8_NO2\NH2_C8_NO2.out
1 // Study transition between ground state to the 1st excited state. Then Multiwfn will
calculate corresponding TDM
n // Do not diagonalize the newly generated TDM, because TDM in original form carries
more useful information
1 // As mentioned in Section 3.21.2, there are several ways that can contract the TDM
(represented in basis functions) to atom TDM. Here we use the way 1. Way 2 and way 3 can also be
used and can result in similar map, while way 4 is usually deprecated
1 // Plot heat map
Below map is immedately shown on screen, the pink line was manually added to highlight the
diagonal, the "hole" and "electron" texts were also manually labelled. By default, the lower limit of
color scale is 0, while upper limit is the largest matrix element.
Electron excitation can be regarded as hole→electron transition. As introduced in Section

3.21.2, the diagonal terms of TDM heat map can reflect that in which atoms the hole and electron
simultanouesly have large distribution. For off-diagonal elements, we should first examine X-axis
640
(corresponding to hole position) and then Y-axis (corresponding to electron position), we will be
able to recognize how electrons transfer among different sites. In the TDM map of present instance,
most elements in the diagonal are surrounded by green or red color, therefore this excitation must
be a global excitation, namely the excited electrons distribute over the whole system. The matrix
elements are not symmetric with respect to diagonal, it can be clearly seen that the upper left part
of the map is larger than the lower right part, in particular, the elements near the diagonal have
relatively large value. This observation reveals that electrons on non-hydrogen atoms are transferred
to atoms adjacent to them, more specifically, electrons on the atoms with smaller index tend to
transfer to atoms with larger index. Since the index of non-hydrogen atoms is ordered from the
amino group to the nitro group, hence it can be inferred that this S0→S1 excitation causes the
electrons to move integrally from the amino end to the nitro end.
If you feel difficult to understand above texts, you can compare the TDM map with below
hole&electron isosurface map (see Section 4.18.1 on how to plot it). You can find the TDM heat
map and the isosurface map convey similar information, and can confirm all of our conclusions
drawn based on the TDM map.
If you want to plot TDM heat map for other excited state, you can exit the heat map plotting
function, then re-enter this function and select the state to be studied. It is worth to note that if you
choose option "4 Toggle if taking hydrogens into account" once to switch its status to "Yes" and
then replot, you will see below map
The index range of hydrogens are 13~22, above map shows that hydrogens indeed do not evidently
participlate in electron excitation because their elements are very small (represented as purple color),
clearly it is meaningless to include hydrogens into S0→S1 TDM heat map.
641
Let us check another excitation, S0→S2. The heat map and hole&electron isosurface map are
given below.
The upper right corner of the heat map has a large value area, which corresponds to the nitro group
at the end of the system, therefore the hole and electron nust have large distribution at the same time
in this region. In addition, the value of the off-diagonal terms of rightmost column of the image is
not very small, so it can be considered that the nitro group transfers a certain amount of electrons to
the central region of the system, which is consistent with the phenomenon that can be seen in the
hole&electron isosurface map. This observation can also be described as there is a so-called
"coherence" between the nitro moiety and the intermediate region of the system in the S0→S2
excitation.
Fragment transition density matrix

Below we will plot TDM heat map based on fragment, namely the index of the map
corresponds to the index of self-defined fragments. The advantage of this kind of TDM map is that
the system to be studied is not necessarily linear, any shape of system (e.g. ring, star) can also be
easily investigated.
The system to be investigated next is shown below, the molecule is divided as five fragments,
which are represented as different colors. The Gaussian input, output and fchk files, as well as other
files involved in the following texts can be downloaded from http://sobereva.com/attach/436/file.rar.
642
We first study S0→S1 excitation. Boot up Multiwfn and input

tdmat.fchk
18 // Electron excitation
tdmat.out
1 // Study S0→S1 excitation
n // Do not diagonalize the newly generated TDM
1 // Use the way 1 to contract the TDM represented in basis functions to atom TDM
Now you can choose option 0 to plot atom TDM, however our present aim is to plot fragment
TDM. In order to do this, we should create a plain text file (which has already been provided as
tdmfrag.txt in the file.rar package), each line of it defines a fragment, the content of the file is
1-23
24-33
34-43
44-55
56-63
Note that you can also use such as 2,5-8,12-15,20 to define a batch of atoms with discontinuous
indices as a fragment.
Then input below commands
-1 // Load fragment definition
0 // Load from external file (you can also directly manually input atom indices)
tdmatfrag.txt // The file containing fragment definition
5 // Modify range of color scale
0,0.4 // Lower and upper limit
1 // Plot heat map
Now you can see below graph, whose index corresponds to fragment index, the hole&electron
isosurface map is also given together for comparison. The region marked with a blue frame is the
4th fragment (hexatriene).
643
According to the colors in the heat map, we know that electron and hole are mainly distributed
on the fragment 4, but they also simultaneously appear on fragments 1 and 5 to some extent, these
finding are consistent with the situation exhibited by the isosurface. Since no off-diagonal element
in the graph is quite large, present electron excitation does not cause a significant electron transfer
between various fragments. Roughly speaking, the main feature of this excitation is local excitation
on fragment 4.
Analogously, we plot fragment TDM between ground state and each of S2~S7 excited states,
the resulting maps are collectively shown below
The values of the off-diagonal elements in the maps of S0→S2~S5 are not significant with
respect to their diagonal elements, therefore interfragment electron transfer should not be obvious.
According to the diagonal terms, we can find that the transitions of S0→S2 and S0→S3 mainly
occurred in the fragment 1, while S0→S2 also marginally involves fragment 4. In general, both the
644
transitions can be regarded as local excitation. The S0→S4 is evidently a global excitation since all
diagonal terms are conspicuous. Main character of S0→S5 is local excitation on the fragment 3,
which corresponds to a benzene ring, but its neighboring fragments are also more or less involved.
S0→S6 and S0→S7 are somewhat mirrored with each other, from the figure it can be seen that
almost every fragment is involved during electron excitation, they either occupied by hole
distribution, electron distribution, or both. For S0→S6, we can speculate that the fragments 2, 3,
and 4 transferred certain amount of electrons to the fragment 1 because (1,2), (1,3) and (1,4)
elements are large, and meantime the fragments 3 and 5 also transferred some electrons to the
fragment 4.
Above figure and discussion obviously show that, when you want to discuss transition
character from ground state to a large number of excited states at the same time when writing an
article, it is very straightforward to provide a figure containing TDM heat maps of all excitations.
Skill: Plotting TDM heat map for a batch of excited states using shell script
If you want to study a batch of excited states in terms of TDM heat map, while you feel that
plotting map one by one is laborious, you can use Linux shell script to fully automate this process.
The script that generates the fragment TDM heat maps for specified range of excited states at
one time is examples\scripts\allTDM.sh. For example, if you put tdmat.fchk, tdmat.out, tdmfrag.txt,
and allTDM.sh used in above example into the Multiwfn directory and then enter this folder, run
chmod +x allTDM.sh to add executable permissions, and then run ./allTDM.sh, this script will
automatically call Multiwfn to generate 1.png, 2.png ... until 7.png in the current directory, they
correspond to TDM heat map of S0→S1, S0→S2 ... to S0→S7. The entire process can be completed
in a blink of an eye, clearly using the script is extremely convenient.
4.18.2.3 Plotting and analyzing transition dipole moment matrix

In fact, the heat map plotting function illustrated in last section is a general module, it can also
plot other kind of atom or fragment matrix. In Section 3.21.11, the concept of atom transition dipole
moment matrix (atom TDMM) is introduced. This matrix has three components X, Y and Z. For
example, the sum of all elements of the X component matrix just corresponds to the X component
of transition dipole moment. Therefore, by plotting TDMM as heat map, we are able to make clear
which atoms or fragments have conspicuous contributions to transition dipole moment.
Below, we still use the donor--acceptor employed in Section 4.18.2.2 as example to show how
to plot TDMM as heat map. Multiwfn is able to plot both transition electric dipole moment matrix
and transition magnetic dipole moment matrix, present section will limit to the former one.
Initially we need to generate a file containing atom TDMM. Boot up Multiwfn and input
11 // Decompose transition dipole moment as basis function and atom contributions
1 // The transition dipole moment to be studied is "electric"
y // Export atom TDMM
As shown in screen, matrices have been exported to .txt files in current folder with "AAtrdip"
prefix, in which the AAtrdipX.txt contains the X component of atom TDMM. Next we will plot heat
645
map based on this matrix.

Reboot Multiwfn and input
o // Load the file last time loaded
AAtrdipX.txt
Now you can find below information on screen
Sum of all elements (including hydrogens): -4.38875223
Maximum and minimum (including hydrogens): 0.65818572 -0.82543408
Sum of all elements (without hydrogens): -2.86353889
Maximum and minimum (without hydrogens): 0.65818572 -0.82543408
where -4.38875223 is just the X component of transition electric dipole moment of S0→S1
excitation, the value is identical to that can be found from Gaussian output file.
As also shown in above prompt, the minimum value of this matrix is a negative value -0.825,
however by default the lower limit of color scale of present function is 0, therefore we must change
the color scale, and it is better to make absolute value of lower and upper limits identical. You can
repeatedly try to find the value that makes the image best reflect the characteristics of the matrix. If
the range is too narrow, the parts that exceed the upper and lower limits of the color scale will be
displayed as white and black, respectively, which is not beautiful. If the range is too broad, the
difference of the matrix elements can hardly be distinguished by colors.
Now we input below commands in Multiwfn
5 // Modify range of color scale
-0.7,0.7 // Lower and upper limits
1 // Plot heat map
Immediately you can see below graph. In order to better understand the heat map, the isosurface
of transition dipole moment density of X component is also shown together (see Section 4.18.2.1
on how to plot it)
646
The bluer (redder) matrix element of this heat map contributes more negatively (positively) to the
transition dipole moment of X component. Since most part of the heat map are blue, the sum of all
matrix elements must be negative, explaining why the X component of the transition dipole moment
is a significant negative value (-4.388 a.u.). Because all matrix elements far from the diagonal are
very close to 0 (shown as green), hence the long-range coupling between atoms does not contribute
substantially to the transition dipole moment of X component. There are several areas of the figure
show very blue color, such as the regions near (2,2) and (9,9), showing that corresponding atoms
and neighboring ones have significant negative contributions, this point is also clearly reflected in
the isosurface map of transition dipole moment density. Some sites such as (1,2) are obviously
positive, which means that the coupling between the two atoms has a significant positive
contribution to the transition dipole moment of X component, this is also why in the isosurface map
there are green isosurfaces between atoms 1 and 2. The middle part of the heat map is basically
green, indicating that the value is very small; correspondingly, there is no isosurface in the middle
of the molecule on the isosurface map.
It can be seen that combining the transition dipole moment density and the transition dipole
moment matrix together is helpful for clarifying the intrinsic characteristics of the transition dipole
moment.
We can also plot TDMM based on fragment index, this is very easy and thus will not be further
illustrated. What you need to do is simplying define fragments in the heat map plotting function and
then plot the graph (please recall Section 4.18.2.3).
4.18.2.4 Investigating transition density and transition density matrix

between excited states
In last several sections, I have illustrated how to study transition character in terms of transition
647
(dipole moment) density as real space function and transition (dipole moment) density matrix
between ground state and excited states. In fact, these kinds of studies can also be used to analyze
transition between two excited states, such analyses may be useful in special applications, such as
transient absorption spectrum and two-photon process. In this section, I will show how to realize
these analyses.
Analysis of transition density in real space function between excited states

As shown in Section 4.18.2.1, Multiwfn is able to easily generate grid data of transition density
between ground state and a selected excited state. In fact transition density between two excited
states can also be generated. To do so, we should first generate transition density matrix (TDM)
between the two states, however, since all analyses related to real space function in Multiwfn are
based on orbitals, we then need to transform the TDM to corresponding natural orbitals. Finally, the
electron density evaluated based on these natural orbitals will directly correspond to transition
density.
Here S2→S3 transition of N-phenylpyrrole is taken as an example, below procedure will
generate cube file of corresponding transition density. Boot up Multiwfn and input
9 // Generate and export transition density matrix
2 // Generate transition density matrix between (TDM) two excited states
examples\excit\N-phenylpyrrole.out
2,3 // Assume that you want to analyze S2→S3 transition
[Press ENTER button directly to use default threshold]
y // Symmetrize the resulting TDM in usual way
y // Export present wavefunction information including the newly generated TDM to
TDM.fch in current folder
TDM.fch
16 // Generate natural orbitals based on the density matrix in .fch/.fchk file
SCF // We input this because the "Total SCF Density" field in the TDM.fch currently
correspond to S2→S3 TDM
y // Export new.molden, which contains natural orbitals corresponding to S2→S3 TDM, and
then let Multiwfn directly load it
2 // Export cube file
Now the generated density.cub in current records transition density between S2 and S3, you can also
choose option -1 to directly visualize the isosurface.
It is also possible to generate grid data of transition dipole moment density between excited
states. For example, if you set “iuserfunc” parameter in settings.ini to 22, which sets user-defined
function to −𝑥𝜌(𝐫), then when you use the new.molden generated previously as input file, the user-
defined function will correspond to X component of transition dipole moment density of S2→S3
648
transition. Clearly, what you should do next is calculating grid data of user-defined function.
Heat map of transition density matrix between excited states

Here we use the NH2-C8-NO2.fchk and NH2-C8-NO2.out employed in Section 4.18.2.2 as
example to illustrate how to plot heat map of transition density matrix between two arbitrarily
selected excited states, S1 and S2.
9 // Generate transition density matrix
2 // For two excited states
1,2 // The states are chosen as S1 and S2
[Press ENTER button to use default threshold]
n // Do not symmetrize the resulting TDM
n // Do not yield TDM.fch
Now we have tdmat.txt in current folder, which records TDM of S1→S2.
Reboot Multiwfn and then input
o // Load the file last time loaded
2 // Plot heat map for transition matrix
tdmat.txt // Matrix data will be loaded form this file
1 // Construct atom transition matrix in terms of way 1
1 // Plot heat map
The obtained graph is shown below, the isosurface map of density difference yielded by
substracting S1 density from S2 density is also given (see Section 4.18.13 on how to plot it).
649
Since the area corresponding to atoms 1~9 at the right side (X=11 and 12) of the heat map has large
value, we can speculate that a large amount of electrons is transferred from atoms 11 and 12 to atoms
1~9 during the S1→S2 excitation. Exactly same conclusion can also be gained from the isosurface
map of density difference. This example shows that the density matrix heat map is not only useful
and reliable for analyzing transitions from ground state to excited states, but also for transitions
between various excited states.
4.18.3 Analyze charge-transfer during electron excitation based on

electron density difference
In this instance we will analyze charge-transfer (CT) between the first singlet excited state and
ground state of the molecule shown below in ethanol solvent, which will be referred to as P2. The
related theory has been introduced in Section 3.21.3. The discussions in this example are somewhat
related to the ones involved in Section 4.18.1, however the methods employed in this section are
purely based on electron density difference.
Since the .wfn files corresponding to the excited state and ground state are large, they are not
provided. Instead, the input files of Gaussian for generating the two .wfn files are provided in
“examples\excit” folder (extP2.gjf and basP2.gjf). I assume that the corresponding .wfn files are
produced at "CT" subfolder in current folder. I would like to remind you once again, the geometries
in the wavefuncition files of the two states must be exactly identical, otherwise the result will be
meaningless! If you do not have Gaussian in hand, you can also directly download the extP2.wfn
and basP2.wfn from http://sobereva.com/multiwfn/extrafiles/extP2_basP2.zip.
First, we calculate grid data of electron density variation  during the excitation. Boot up
Multiwfn and input:
CT\extP2.wfn // Excited state wavefunction file
1 // Only one file will be dealt with
-,CT\basP2.wfn // Ground state wavefunction file. Corresponding density will be subtracted
from the excited one to generate 
2 // Medium quality grid. If the system is much larger than present one, more grid points is
required (e.g. using high quality grid)
Once the calculation is normally completed, you can choose option -1 to view the electron
density variation during the electron excitation (default isovalue is too large for visualizing density
difference, 0.005 is recommended for present case). Green and blue regions correspond to positive
and negative regions, respectively, they represent increase and decrease in electron density due to
the excitation.
650
However this density difference graph is not quite intuitive, since positive and negative parts
interwined together and there are many nodes. We will see that C+ and C- functions make the image
much clearer.
3 // Analyzing CT based on electron density difference grid data
The following information are displayed immediately. Note that if positive and negative parts
of qCT are obvious unequal, that means the grid setting used in generating  is too coarse, and you
need to calculate again with finer grid setting.
q_CT (positive and negative parts): 0.844 -0.844 a.u.
Barycenter of positive part in x,y,z (Angstrom): -2.659 -0.001 -0.000
Barycenter of negative part in x,y,z (Angstrom): 2.294 -0.009 -0.029
Distance of CT in x,y,z (Angstrom): 4.953 0.009 0.029 D index: 4.953
Dipole moment variation (a.u.) : 7.896 -0.014 -0.046 Norm: 7.896
Dipole moment variation (Debye): 20.070 -0.035 -0.117 Norm: 20.070
RMSD of positive part in x,y,z (Angstrom): 2.993 1.250 0.821 Total: 3.346
RMSD of negative part in x,y,z (Angstrom): 3.290 1.144 0.881 Total: 3.593
Difference between RMSD of positive and negative parts (Angstrom):
X: -0.297 Y: 0.106 Z: -0.060 delta_sigma index: -0.247
H_x: 3.141 H_y: 1.197 H_z: 0.851 H_CT: 3.141 H index: 3.469 Angstrom
t index: 1.811 Angstrom
Overlap integral between C+ and C- (i.e. S+- index): 0.742365
Above information are self explanatory, if you are confused, please consult Section 3.21.3. The
evident positive value of t index implies that the distribution of positive and negative of  has been
significantly separated due to strong CT. The large D index (4.95 Å) shows that the CT distance is
quite long. Clearly, S0→S1 transition of this system should be identified as typical CT excitation.
The excitation caused significant variation of dipole moment, as shown in the data, it is as high as
20.07 Debye. The distribution spatial distribution breadth of positive and negative parts of  are
similar, therefore the outputted  index, which measures difference of their RMSD, is merely -
0.247 Å.
By selecting option 1, isosurface of C+ (green) and C- (blue) functions can be shown up. The
isovalue of the graph shown below is 0.0015.
651
If the isovalue is increased to 0.0024, the positions of barycenters can be approximately located
(barycenters of C+ and C- exactly correspond to center of their isosurfaces).
From the graph it is evident that the direction of electron transfer is from the amino group side
(electron donor) to nitro group side (electron acceptor). However, the barycenters are not exactly
located at the two substituents, this observation suggests that the actual electron donor in this
electron excitation is not amino group but phenyl. This finding parallels to the fact that phenyl is a
weak electron donor.
Hint: If you would like to get better display effect of the C+ and C- isosurfaces, you can use VMD program
(freely available at http://www.ks.uiuc.edu/Research/vmd/) to display them, the procedure is: Boot up VMD first,
drag Cpos.cub into VMD main window, and then drag Cneg.cub into it. Select "Graphics"-"Representations", choose
the first term in "Selected Molecule", click "Create Rep" button to create a new representation (the existing
representation is used to show molecular structure), change the "drawing method" to "isosurface", set "Draw" to
"solid surface", change the isovalue to 0.0015, set "coloring method" to "ColorID" and choose "7 green". Now the
isosurface of Cpos has been properly displayed. Next, choose the second term in "Selected Molecule", use the similar
methods to set each options, but select "0 blue" in "ColorID", and use isovalue of -0.0015. Finally, the graph will
look like the one shown above. You can also set "Material" to "transparent" so that the overlap region of C+ and C-
can be clearly distinguished.
4.18.4 Calculate ∆r and Λ indices to characterize various electron

excitations for N-phenylpyrrole
In this section I will illustrate how to calculate the r index proposed in J. Chem. Theory
Comput., 9, 3118 (2013) and the  index proposed in J. Chem. Phys., 128, 044118 (2008) to
652
characterize electron excitations for N-phenylpyrrole. If you are not familiar with these two indices,
please check Section 3.21.4 and Section 3.21.14, respectively.
In my personal view of point, using the quantities such as D and Sr indices defined in hole-
electron framework to characterize electron excitation is already absolutely sufficient, as illustrated
in Section 4.18.1. Theoretically, the r and  indices may be regarded as approximations of D and
Sr, respectively. The only advantage of r and  is that in Multiwfn, they can be outputted for all
selected excited states at the same time and can be decomposed into orbital pair contributions. In
addition, calculation cost of r index is almost negligible.
The files used in this section is the N-phenylpyrrole.fch and N-phenylpyrrole.out in
"examples\excit" folder, they were yielded by Gaussian, the keywords are CAM-B3LYP/6-31+G(d)
TD(nstates=5) IOp(9/40=4). Since the optimized ground state geometry was used in the calculation,
therefore the analysis results can be regarded as corresponding to vertical absorption process.
Calculating r index
The r index is a quantitative indicator for measuring charge transfer (CT) length of electron
excitation, larger r index implies longer CT distance.
4 // Calculate r index
1-5 // Assume that we want to calculate r index for all the five calculated singlet excited
states
Immediately, the results are printed on screen:
Excited state 1: Delta_r = 1.499249 Bohr, 0.793368 Angstrom
The r indices imply that the excitations from ground state (S0) to the 3th, 4th and 5th excited
states possess strong CT character since they have large r, while the excitations of S0→S1 and
S0→S2 should be basically regarded as LE excitations because their r indices are not quite large
(the original paper of r suggests using 2.0 Å as criterion for distinguishing LE and CT excitations).
Bear in mind, definitive conclusion about the excitation character can only be finally drawn after
visualizing the hole and electron distributions using the hole-electron analysis module of Multiwfn.
In Multiwfn it is possible to decompose the r index as contributions of orbital pair transitions.
For example, we want to do this for the S0→S4 excitation, we should first enter the r index
calculation interface and then input
4 // Only calculate r index for a single excitation (S0→S4), in this case the result can be
decomposed
y // Print orbital pair contributions
0.01 // Only the orbital pairs having contribution larger than 0.01 Å will be printed
You will immediately see below information
Note: The configuration coefficients shown below have combined both excitation
and de-excitation parts
653
Sum of square of configuration coefficients: 0.497953
#Pair Orbitals Coefficient Contribution (Bohr and Angstrom)
378 37 41 0.5004500 3.7301590 1.9739153
379 37 43 0.4452400 1.8477898 0.9778083
381 37 47 -0.1067900 0.0929645 0.0491947
382 37 49 -0.0782100 0.0700060 0.0370456
383 37 53 -0.0639300 0.0285116 0.0150877
389 37 72 0.0436900 0.0215865 0.0114231
As you can see, MO37→MO41 transition has predominating contribution (1.97 Å) to the r
index of S0→S4 (3.11 Å), while the MO37→MO43 transition also has nonnegligible contribution
(0.97 Å).
Calculating  (lambda) index

The  index essentially measures overlapping degree of hole and electron of electron
excitations. Here we calculate it for all the five excitations for N-phenylpyrrole.
14 // Calculate  index
1-5 // Analyze all the five calculated singlet excited states
Immediately, the results are printed on screen:
Excited state 1: lambda = 0.684853
From above output, it can be found that the  indices are nearly inversely proportional to the
r indices, because the larger the hole-electron overlapping exent, usually the shorter the hole-
electron separation distance (but bear in mind, this relationship is not always true).
Then we decompose the  index for the fourth excitation. Inupt below commands
y // Do the  index analysis again
4 // The fourth excitation
y // Decompose analysis on  index
0.01 // Printing threshold
Then you will see all MO pairs having contribution to  index larger than 0.01:
Sum of square of configuration coefficients: 0.497953
#Pair Orbitals Coefficient Contribution
378 37 41 0.5004500 0.0865190
379 37 43 0.4452400 0.0915297
The data indicates that only occupied MO 37 has non-negligible overlap with unoccupied MOs;
specifically, only the overlap between MO37-MO41 and between MO37-MO43 is relatively
detectable.
654
4.18.6 Generate and analyze natural transition orbitals (NTOs) for

uracil
In this section I illustrate how to use Multiwfn to perform the very popular natural transition
orbital (NTO) analysis with uracil as example. Please first read Section 3.21.6 to acquire basic
knowledge of NTO. Although in this example the files outputted by Gaussian were used as input
file, in fact the files outputted by ORCA are also fully supported, see Section 3.21.1.2 for detailed
requirement about the input file.
Before showing how to perform NTO analysis, I would like to let you apprehend why NTO
analysis is meaningful. As an instance, we use Gaussian to perform TDDFT calculation at PBE0/6-
31G* level for singlet excited states of uracil, you will find below information
Excited State 3: Singlet-A" 6.0180 eV 206.02 nm f=0.0000 <S**2>=0.000
26 -> 30 0.54135
26 -> 31 -0.20634
28 -> 30 -0.15424
28 -> 31 0.36715
Clearly, in excitation of S0→S3, there is no dominant MO transition, the largest contribution of a

single MO pair is merely 0.541^2*2*100%=58.5%, therefore it is impossible to identify the nature
of this excitation by viewing only one MO pair. In such difficult cases, NTO analysis is often useful,
because after transforming MOs to NTOs, commonly you will be able to find only one pair of NTO
having eigenvalue very close to 1, transition between the two NTOs in this pair faithfully represents
the real character of the electron excitation.
The files needed by NTO analysis have mentioned at the beginning of Section 3.21. Briefly,
assume that you are a Gaussian user and you want to study electron excitation from ground state to
the lowest three singlet excited states for uracil at TD-PBE0/6-31G* level, what you need to do is
carrying out a normal TDDFT calculation with these keywords: # PBE1PBE/6-31G* TD
IOp(9/40=4), also you need to make Gaussian generate corresponding .fch file. The input file,
output file and.fch file have already been provided in "examples\excit\NTO" folder. The keyword
IOp(9/40=4) is very important, without it the NTO result will be evidently inaccurate, the meaning
of this IOp has been mentioned in Section 4.18.1.
Now we start to carry out NTO analysis. Boot up Multiwfn and input
examples\excit\NTO\uracil.fch
6 // Generate NTOs
examples\excit\NTO\uracil.out // Gaussian calculated three lowest excited states, you can
analyze any one of them
3 // Study transition from ground state (S0) to the 3rd excited state (S3)
Now Multiwfn loads transition information of S0→S3 from the Gaussian output file and
generate NTOs, the eigenvalues of occupied and virtual NTOs are shown below
The highest 10 eigenvalues of occupied NTOs:
0.000006 0.000007 0.000015 0.000016 0.000024
0.000063 0.000121 0.000582 0.134025 0.865529
The highest 10 eigenvalues of virtual NTOs:
655
0.865529 0.134025 0.000582 0.000121 0.000063
0.000024 0.000016 0.000015 0.000007 0.000006
It can be seen that, the eigenvalues of occupied and virtual NTOs have one-to-one correspondence,
the largest eigenvalue is 0.8655, that means that NTO pair contributes as high as 86.55% of the
S0→S3 transition. So, if we would like to characterize the nature of this transition, we can only
study the occupied NTO and virtual NTO in this NTO pair.
Now you can select if outputting .fch or .molden file containing the NTOs. We choose "2
Output NTO orbitals to .fch file" and input the path to output, such as C:\S3.fch. After the .fch has
been successfully generated, you can reboot Multiwfn and load the S3.fch, in main function 0 you
can visualize the NTOs, the orbital energies now correspond to NTO eigenvalue. To plot the two
NTOs in the NTO pair with 86.55% contribution, in the GUI of main function 0 you can select
"orbital info." - "Show up to LUMO+10" in the menu, in the text window you will find output like
below
Orb: 27 Ene(au/eV): 0.000582 0.0158 Occ: 2.000000 Type: A+B
We can see that the occupied NTO with index of 29 and the virtual NTO with index of 30 constitute
the NTO pair with eigenvalue of 0.8655, there we select corresponding index in the GUI to visualize
them, the isosurfaces are shown below
Undoubtedly, this S0→S3 excitation can be regarded as transition from lone pair of O12 to anti-
bonding  orbital of the uracil ring, at least we have 86.55% confidence to say that. From NTO
eigenvalues we notice that NTO28→NTO31 transition also has small contribution (13.40%) to the
excitation, please plot corresponding orbitals and discuss their characteristic.
The NTOs can also be subjected to quantitative analyses. For example, you can enter main
function 8 and use suitable options to analyze their orbital composition at quantitative level, or you
can use subfunction 11 of main function 100 to evaluate overlap extent and centroid distance
between selected two NTOs.
Note that NTO analysis has both advantage and disadvantage with respect to the hole-electron
analysis, this point has been mentioned at the end of Section 3.21.6. For many systems and
excitations, even if NTO transformation has been applied, there are still no NTO pair with
predominant contribution, in this case you have to use hole-electron analysis to facilitate discussion
of excitation characteristic. The drawback of hole-electron analysis is that it is unable to present
656
orbital phase information like NTO analysis, and it takes more computational cost.
Using script to carry out NTO analysis in batch

Sometimes we want to perform NTO analysis for a batch of excitations, for example, we want
to generate a batch of .fch files that respectively contain NTOs of S0→S1, S0→S2 and S0→S3
transitions, although you can manually do this in the interactive interface of Multiwfn, a more
efficient and clever way is employing shell script. If you are a Linux user, we can run below shell
script to yield S1.fch, S2.fch and S3.fch, which contain NTOs corresponding to S0→S1, S0→S3 and
S0→S3. This script is very easy to understand as long as if you have basic level of knowledge about
shell programming.
#!/bin/bash
cat << EOF > allNTO.txt
18
examples/NTO/uracil.out
EOF
for ((i=1;i<=3;i=i+1))
do
cat << EOF >> allNTO.txt
$i
S$i.fch
EOF
done
./Multiwfn examples/excit/NTO/uracil.fch < allNTO.txt
rm ./allNTO.txt
This script is also provided as examples\excit\NTO\allNTO.sh. If you do not make any modification
to the script, this script should be copied to the Multiwfn folder and run as ./allNTO.sh in Multiwfn
folder, then S1.fch, S2.fch and S3.fch will be yielded at the same folder. In practical studies, you
should properly modify the script according to actual situation, the range of the excitations to be
studied is determined by "i=1;i<=3".
4.18.8 Using IFCT method and heat map of charge transfer matrix to
study interfragment charge transfer during electron excitation
The interfragment charge transfer (IFCT) is a method derived based on hole-electron analysis
for quantitatively studying amount of charge transfer between different fragments. The situation of
charge transfer can also be very intuitively understood by means of heat map of charge transfer
matrix, which is a byproduct of IFCT analysis. Please carefully read Section 3.21.8 if you are not
familiar with these concepts.
In the next two sections, I will use two molecules to respectively illustrate how to perform
IFCT analysis and plot heat map of charge transfer matrix, these two kinds of analyses can provide
complementary perspectives.
657
4.18.8.1 IFCT analysis for 4-nitroaniline

In this example, I will illustrate how to use IFCT method to study CT between three fragments
(-NO2, -NH2 and the linker benzene) of 4-nitroaniline during its electron excitation.
The Gaussian input file of TDDFT task at PBE0/6-311G* level for 4-nitroaniline has been
provided as examples\excit\4-nitroaniline.gjf. Note that IOp(9/40=4) is used so that Gaussian can
print enough configuration coefficients. Run it by Gaussian to obtain .chk file and .out file, then
convert the .chk file to .fch via formchk. The .fch and .out files have been provided in
"examples\excit" folder.
8 // Calculate interfragment charge transfer in electron excitation via IFCT method
2 // We first analyze transition from ground state (S0) to the second excited state (S2)
11-13 // Atomic indices of amino group (fragment 1)
1-10 // Atomic indices of benzene group (fragment 2)
14-16 // Atomic indices of nitro group (fragment 3)
Then you will see
Contribution of each fragment to hole and electron:
1 Hole: 28.17 % Electron: 3.02 %
2 Hole: 65.89 % Electron: 29.52 %
3 Hole: 5.94 % Electron: 67.46 %
Construction of interfragment charger-transfer matrix has finished!
Variation of population number of fragment 3: 0.61519
658
1 -> 2: 0.08315 1 <- 2: 0.01987 Net 1 -> 2: 0.06328
1 -> 3: 0.19001 1 <- 3: 0.00179 Net 1 -> 3: 0.18821
2 -> 3: 0.44452 2 <- 3: 0.01754 Net 2 -> 3: 0.42697
The output is very easy to understand, if you are confused, please check corresponding
explanation in Section 3.21.8. For example, the data shows that during the S0→S2 excitation, nitro
group (fragment 3) donates 0.00179 electrons to amino group (fragment 1) and meantime accept
0.19001 electrons from amino group, therefore nitro group totally gains 0.18821 electron from
amino group. If the benzene fragment is also taken into account, the electron excitation totally
increases electron population of nitro group by 0.61519. The electron redistribution phenomenon in
the two terminal groups is not prominent, however, as the data shown (0.19453), the electron
redistribution within the benzene, which behaves as  linker, is remarkable.
For facilitating discussion, I summarized all IFCT analysis data in below table. The diagonal
terms correspond to amount of intrafragment electron redistribution.
Donor Amino group benzene nitro group
amino group 0.008 0.083 0.190
benzene 0.020 0.195 0.445
nitro group 0.002 0.018 0.040
As you can see from the table, there are three prominent interfragment CT terms (sorted according
to magnitude): benzene→nitro, amino→nitro, amino→benzene, all of them direct from amino
group towards nitro group.
For better and intuitively understanding above data, we plot hole-electron isosurface map using
the method described in Section 4.18.1:
In the graph, main distribution regions of hole and electron correspond to blue and green,
respectively. As you can see, hole mainly distributes on the amino group and benzene moiety, while
electron mainly locates at the benzene linker and nitro group, clearly the overall CT direction is
from amino group to nitro group, this is completely in line with our observation of quantitative
interfragment CT analysis.
Performing IFCT analysis based on Becke partition

Notice that by default Multiwfn employs Mulliken-like partition to evaluate atomic
contributions to hole and electron distributions, which are needed by the IFCT analysis. Since this
method is incompatible with diffuse functions, if basis set containing diffuse functions was used,
the calculated amount of CT transfer may be unreliable or completely meaningless. Fortunately, it
is also possible to employ Becke partition to evaluate interfragment CT quantities, in this case
659
diffuse functions could be safely used, but additional steps are needed, namely cube file of hole and
electron must be first generated using the hole-electron analysis module. Next, we will redo above
analysis in this way. Boot up Multiwfn and then input below commands
1 // Evaluate and analyze hole and electron
2 // Analyze S0→S2
10 // Export cube file for hole
1 // Total hole (i.e. hole distribution in complete form)
11 // Export cube file for electron
1 // Total electron (i.e. electron distribution in complete form)
Now hole.cub and electron.cub have been generated in current folder, so we can start IFCT
analysis. Input below commands
0 // Return
0 // Return to menu of main function 18
8 // IFCT analysis
Now the hole.cub and electron.cub in current folder have been detected, here we input y to
make Multiwfn carry out IFCT analysis based on them using Becke partition method to evaluate
their compositions. Then define fragments:
11-13 // Atomic indices of amino group (fragment 1)
1-10 // Atomic indices of benzene group (fragment 2)
14-16 // Atomic indices of nitro group (fragment 3)
Now the result is
1 -> 2: 0.08103 1 <- 2: 0.02383 Net 1 -> 2: 0.05720
1 -> 3: 0.18532 1 <- 3: 0.00285 Net 1 -> 3: 0.18248
2 -> 3: 0.43360 2 <- 3: 0.02264 Net 2 -> 3: 0.41097
The data is in good agreement with the result shown earlier, which was calculated based on
Mulliken-like partition.
Commonly I suggest using the default way to realize the IFCT analysis, since no additional
procedure is needed and the cost is relative low. However, if due to some special reasons diffuse
functions must be employed (e.g. anion systems, Rydberg excitation), you have to perform IFCT
analysis based on Becke partition.
More discussions and illustrations about the IFCT analysis can be found from my blog article
"Using the IFCT method in Multiwfn to evaluate amount of electron transfer between arbitrarily
defined two fragments during electron excitation" (in Chinese, http://sobereva.com/433).
4.18.8.2 Plotting heat map of charge transfer matrix to intuitively understand

nature of electron excitation
The charge transfer matrix (CTM) is closely related to transition density matrix (TDM) and
660
their heat maps often provide similar information for an electron excitation. The method of plotting
TDM has been illustrated in Section 4.18.2.2. In my opinion, the physical meaning of CTM is
somewhat more clear than TDM and can better reveal actual charge transfer character. In addition,
since CTM is derived in the theory framework of hole-electron analysis (see Section 3.21.1), the
CTM heat map can always well compare with distribution of hole and electron.
Here I still use the molecule studied in Section 4.18.2.2 as instance. Before plotting the heat
map of CTM, we should first generate CTM. Boot up Multiwfn and input
8 // IFCT analysis
-1 // Export atom-atom CTM to atmCTmat.txt in current folder
2 // Enter the function used for plotting heat map
atmCTmat.txt // Load matrix data from this file
1 // Show heat map
Now you can see below map, the purple line and texts are manually added.
This figure has similar features of the heat map of atom TDM given in Section 4.18.2.2, but there
are also differences that cannot be ignored. According to the IFCT point of view, each of the non-
diagonal elements of the current graph rigorously exhibits the amount of electron transferred
between atoms. Looking at the graph column by column, it can be visually seen that each atom on
the carbon chain transferred electrons to the atoms at its front and back ends, and the amount of
transfer to the nitro side is significantly more than to the amino side. For example, it can be seen
from the figure that in the fifth column, the value of the sixth element is larger than the fourth
element, so the amount of electron transfer of C5→C6 must be more than C5→C4.
Next, we also look into heat map of CTM of other excitation. The map of S0→S9 plotted in
661
the same way as above is given below, corresponding hole&electron isosurface map is also
appended. Because it was found that S0→S9 transition evidently involves some hydrogens,
therefore hydrogens are also taken into account in the map (by choosing "4 Toggle if taking
hydrogens into account" once).
It can be seen from the above heat map that, there is strong electron transfer from the region of
atoms 1~5 and 7~9 to the hydrogen atom with index of 13, this observation fully agrees with the
hole&electron isosurface map, namely there is a large green isosurface at the H13. In addition, from
the isosurface map we can see that atom 6 is basically only surrounded by green isosurface, that
means this atom does not transfer electrons to others while largely accepts electrons from others;
accordingly, the color of the row of Y=6 in the heat map is distinct, while the column corresponding
to X=6 is very dark.
From this example, we can find that the hole&electron isosurface map provides the most
intuitive visual effect, but if it is discussed together with the heat map of CTM, the charge transfer
can be understood more thoroughly from a quantitative point of view, it also avoids the possibility
that the arbitrariness of the choice of isovalue leads to an unreasonable judgment.
The CTM can also be plotted based on fragment. To do this, you simply need to load fragment
definition file or directly input fragment definition in the heat map plotting function, and then plot
the map again.
4.18.9 Generate transition density matrix and transform it to orbital

representation
Note: This section may be not interesting for most Multiwfn users, but valuable for experts
662
In Section 4.18.2, I have shown that in Multiwfn the transition density can be studied in terms
of real space function and colored matrix (heat map). Multiwfn can do even more for transition
density. As will illustrated in this section, Multiwfn is able to transform the generated transition
density matrix to orbital representation and export the orbitals as wavefunction file. This brings a
lot of conveniences; for example, when you analyze "electron density" based on this file, the actual
function to be studied will directly correspond to transition density. Note that these orbitals can be
regarded as natural orbitals of transition density matrix (TDM), but they are remarkably different to
the NTO (nature transition orbital), which has been introduced in Section 3.21.6.
Here will take the N-phenylpyrrole as example, whose transition density of S0→S1 has been
plotted as isosurface in Section 4.18.2. Our purpose in this section is to transform this transition
density as orbitals and export them as .wfx file so that then we can very easily study properties of
the transition density based on this file.
examples\excit\N-phenylpyrrole.fch // The .fch file yielded by Gaussian TDDFT task
9 // Generate and export TDM
1 // Generate TDM between ground state and excited state
examples\excit\N-phenylpyrrole.out // The output file of Gaussian TDDFT task with
IOp(9/40=4) keyword
1 // Analyze electron transition from ground state to the 1st excited state (S0→S1)
1 // Symmetrize the raw TDM. This is important, the natural orbitals cannot be properly
yielded later without symmetrization of the TDM
y // Export current wavefunction to TDM.fch in current folder, whose "Total SCF Density"
field records the just generated symmetrized TDM
Next, we transform the TDM to natural orbitals. Reboot Multiwfn and input
TDM.fch
16 // Generate natural orbitals based on the density matrix in .fch/.fchk file
SCF // The matrix to be transformed comes from the "Total SCF Density" field
y // Export the generated orbitals to new.molden and load it
Now we have new.molden in current folder, which contains natural orbitals transformed from
the S0→S1 TDM. The orbitals in memory now also correspond to these natural orbitals. Assume
that we also want to export them as .wfx file, we should input below commands
2 // Export various kinds of files
4 // Output current wavefunction as .wfx file
TDM.wfx // The path of the file to be generated
In the future, if you use the TDM.wfx as input file and calculate grid data of "electron density"
via main function 5, you will find the resulting isosurface map (after properly adjusting isovalue) is
exactly identical to the transition density T(r) graph shown in Section 4.18.2.1.
663
4.18.10 Gain molecular orbital pair contributions to transition dipole

moment
In order to gain a deeper insight into transition (electric) dipole moment, Multiwfn provides a
function used to decompose it to contributions from various MO pair transitions, see Section 3.21.10
for introduction. Here I present an example. The .fch and .out files involved in this example were
produced by TDDFT calculation of Gaussian.
10 // Decompose transition dipole moment as molecular orbital pair contributions
1 // Select the excitation from ground state (S0) to the first singlet excited state (S1)
Now below information about this excitation is shown on screen
Transition dipole moment in X/Y/Z: -0.000000 -0.000000 1.781438 a.u.
Norm of transition dipole moment: 1.781438 a.u.
Oscillator strength: 0.3935306
Then you can find several options on screen, they are self explanatory. We first choose option
1 and input for example 0.02, then all MO pairs having contribution larger than 0.02 are printed:
1213 35 -> 46 0.040230 0.000000 0.000000 0.037709
1214 35 -> 50 -0.047670 0.000000 -0.000000 0.040152
1239 36 -> 40 -0.101270 0.000000 -0.000000 -0.280355
1259 37 -> 40 -0.127550 0.000000 0.000000 -0.148489
1260 37 -> 52 0.069960 -0.000000 -0.000000 -0.122222
1262 37 -> 58 -0.036060 -0.000000 0.000000 -0.025065
1278 38 -> 39 0.672690 -0.000000 0.000000 2.796678
1280 38 -> 50 0.052570 -0.000000 0.000000 0.046724
2489 36 <- 40 -0.014330 0.000000 -0.000000 -0.039671
2506 37 <- 52 0.015140 -0.000000 -0.000000 -0.026450
2522 38 <- 39 -0.027240 0.000000 -0.000000 -0.113249
Sum of the 11 pairs shown above: -0.000000 -0.000000 2.165763
From the output, we can immediately find that transition of MO38→MO39 has dominating
contribution (2.796678 a.u.) to this S0→S1 excitation.
When there are too many MO pairs having nonnegligible contributions to transition dipole
moment and thus difficult to identify important MO transitions, you can let Multiwfn sort the MO
pairs according to their contributions to specific component of transition dipole moment. For
example, here we choose the option "4 Print orbital pairs according to absolute contribution to Z
component" and then input 5, then you will see the five MO pairs having largest contribution to Z
component of transition dipole moment:
1278 38 -> 39 0.672690 -0.000000 0.000000 2.796678
1239 36 -> 40 -0.101270 0.000000 -0.000000 -0.280355
1259 37 -> 40 -0.127550 0.000000 0.000000 -0.148489
664
1260 37 -> 52 0.069960 -0.000000 -0.000000 -0.122222
2522 38 <- 39 -0.027240 0.000000 -0.000000 -0.113249
By the way, oscillator strength (f) directly relates to norm of transition dipole moment, therefore
it can be expected that if the configuration coefficient corresponding to MO38→MO39 is set to zero,
namely ignoring its contribution, then f will be lowered evidently. As shown earlier, the original f of
S0→S1 is 0.39353. Let us quantitatively check how MO38→MO39 affects the f. To do this, we can
manually set configuration coefficient of this transition to zero and then re-examine the f value. We
0 // Return to menu of electron excitation analysis
-1 // Check, modify and export configuration coefficients of an excitation
1 // Choose the first excited state
1 // Set coefficient of a MO pair
38,39 // The MO indices of the MO pair
1 // The transition type is chosen as "Excitation", hence MO38→MO39 is selected (if
inputting 2, then what we selected will be MO38MO39)
0 // Set the configuration coefficient to zero
-3 // Export current excitation information to a plain text file
S1.txt // The path of the file to store excitation information of S0→S1
Now S1.txt has been generated in current folder, if you open it with text editor, you will find
the coefficient corresponding to MO38→MO39 is indeed zero.
Then reboot Multiwfn and input
o // Load the file used at the last time, namely examples\excit\N-phenylpyrrole.fch
10 // Decompose transition dipole moment as molecular orbital pair contributions
S1.txt
Now the printed f is only 0.1278, which is less than 1/3 of its original value (0.39353), showing
that MO38→MO39 has crucial influence on strength of S0→S1 excitation.
Since the coefficient of MO38→MO39 is as large as 0.6727, after setting it to zero, now the sum of the square
of remaining coefficients has been much less than 0.1, which is far from the ideal value of closed-shell case (0.5).
In my paper Carbon, 165, 461 (2020), I employed the function illustrated above to study the
nature of the extremely strong absorption of cyclo[18]carbon, you are suggested to look at Fig. 4
and relevant discussion. If this function is employed in your work, it is suggested to also cite this
paper.
4.18.11 Plot transition dipole moment vector contributed by molecular

fragments as arrows
In Section 4.18.2.1, I have shown how to plot transition dipole moment density in real space,
which is extremely useful for studying contribution of different regions in three-dimension space.
In fact, if using a special plotting script of VMD (http://www.ks.uiuc.edu/Research/vmd/) provided
below, transition dipole moments contributed by molecular fragments can be easily drawn as arrows,
which greatly facilitates discussion of composition of total transition dipole moment.
665
Here, azobenzene is taken as example. The input file of TDDFT task of Gaussian for
azobenzene is provided as examples\excit\Azobenzene.gjf. Note that IOp(9/40=4) is used and .chk
file is saved after calculation. Run it by Gaussian, and then convert azobenzene.chk to
azobenzene.fch. (If you do not have Gaussian in hand, you can also directly download the .out
and .fch files from http://sobereva.com/multiwfn/extrafiles/Azobenzene_exc.zip)
Boot up Multiwfn, load the azobenzene.fch, then input
11 // Decompose transition dipole moment as basis function and atom contributions
Azobenzene.out // The Gaussian output file obtained by running Azobenzene.gjf
2 // Assume that we want to study is electron excitation from ground state to the second
excited state
1 // The type of transition dipole moment to be decomposed is electric
n // Do not generate AAtrdip.txt currently
Now trdipcontri.txt is outputted to current folder, which contains transition dipole moment
contributed by each basis function and each atom. Move this file to VMD folder.
Return to main menu, then enter subfunction 2 of main function 100, export current molecular
geometry to azobenzene.pdb.
Copy examples\excit\loadip.tcl to VMD folder, this is a VMD script written by me, it can load
data from trdipcontri.txt. It also defines custom commands "dip" and "dipatm" used to draw
transition dipole moment contributed by specific molecular fragment as arrow.
Boot up VMD, drag the file azobenzene.pdb into VMD main window to load it, then run source
loaddip.tcl in VMD console window to execute the script. Assume that we want to divide the
molecule as three parts to separately investigate their contributions to transition dipole moment,
namely the first phenyl group (atoms 1~11), N2 part (atoms 12 and 13) and the second phenyl group
(atoms 14~24), we should run below commands in VMD console window
draw color red
dip "serial 1 to 11"
dip "serial 12 13"
dip "serial 14 to 24"
Now you will see three red arrows in the VMD graphical window. The length of cylindrical part of
the arrows correspond to magnitude of fragmental transition dipole moments, the center of the
arrows correspond to geometric center of the fragments. Note that when we use "dip" command, the
fragment geometry center and quantitative contribution to transition dipole moment by the selected
fragment are also shown in VMD console window.
In order to improve the graphical quality, we input color Display Background white in console
window to set white as background color, enter Graphics - Representation and set Drawing method
to CPK, and then choose Display - Orthographic in VMD main window. The final graph will look
like below.
666
As you can see, both the two phenyl groups have significant contribution to Y component of
total transition dipole moment (the red, green and blue of the axis shown at left-bottom part of the
graph correspond to X, Y and Z directions, respectively). For quantitative comparison purpose, total
transition dipole moment vector and its compositions are also listed below
Total： 0.1155 -2.8868 0.0
Phenyl group 1：0.14262 -1.43288 0.0
N2： -0.16948 -0.02138 0.0
Phenyl group 2：0.14262 -1.43288 0.0
If you also want to plot total transition dipole moment as green arrow on the graph, you can
input draw color green and then input dip all.
It is also possible to plot transition dipole moment contributed by each atom. To do that, we
input draw delete all to remove all existing arrows, and then input dipatm, you will immediately see
There is a very important point that should be paid attention to when using above method to
decompose transition dipole moment, namely contribution of a fragment is often dependent of
choice of origin, because transition charge of a fragment is often non-zero. For example, if we use
subfunction 6 of hole-electron analysis module to export atomic transition charges and then sum
667
them as fragment transition charges, you will find the value of the first phenyl group is 0.2116. Since
it is non-zero, it can be proved that if overall coordinate of the azobenzene is translated, the transition
dipole moment corresponding to this fragment must be varied; in other words, the result is not
definite. Therefore, one should carefully discuss fragmental transiton dipole moment in papers.
Another very important point is that since the transition dipole moment is decomposed via
Mulliken method, the analysis method shown above will be meaningless when diffuse functions are
presented in the electron excitation calculation.
4.18.13 Study electronic structure of a single excited state and

difference between two excited states
Most other subsections in Section 4.18 focus on exemplifying how to study electron transition
characters, however, sometimes we want to study character of a single excited state or difference
between two excited states in specific property. In Multiwfn, one can perform various kinds of
wavefunction analysis for an excited state as usual, however, the input file must contain
wavefunction of this excited state. For multi-configuration methods that can study excited state,
such as CIS and TDDFT, the excited state wavefunction must be recorded as natural orbitals (NOs),
because Multiwfn always load wavefunction in terms of orbitals.
The main purpose of this section is illustrating the function used to generate .molden file
containing NOs of an excited state, so that we can analyze wavefunction character of this state. I
strongly suggest you read Section 3.21.13 first, in which the details of generating NOs of excited
states are described.
NOTE: There are two types of CIS/TDHF/TDA-DFT/TDDFT excited state wavefunction (or
density matrix): (1) Unrelaxed density (2) Relaxed density. The difference has been detailed
described in Section 3.21.1.1. Briefly speaking, the former is not as real as the latter, but generating
the latter requires additional cost (much higher than simply evaluating excitation energy). Next, I
will first illustrate how to perform wavefunction analysis for an excited state and study difference
between two excited states based on unrelaxed density, while at final part of this section I will also
exemplify how to analyze excited state based on its relaxed density.
Example of wavefunction analysis of an excited state (based on unrelaxed density)

Here I take N-phenylpyrrole as example, assume that we want to examine Mayer bond orders
for the second singlet excited state. To do so, we first carry out a regular TDDFT calculation with
IOp(9/40=4) keyword, the examples\excit\N-phenylpyrrole.out is output file and examples\excit\N-
phenylpyrrole.fch is corresponding .fch file. The geometry was previously optimized for ground
state.
13 // Generate natural orbitals of specific excited states
2 // Choose the 2nd excited state
Now, NO_0002.molden has been generated in current folder, it records wavefunction of the
second excited state in terms of NOs.
668

NO_0002.molden
1 // Mayer bond order
From the output you will find the bond order of the N5-C10 bond, namely the bond linking
pyrrole and benzene moieties, is 0.796. If you repeat the calculation for examples\excit\N-
phenylpyrrole.fch, the result will correspond to ground state, and you will find the Mayer bond order
is 0.713. Clearly, the vertical excitation from S0 to S2 at minimum point of S0 weakens the strength
of N5-C10 detectably.
Plotting density difference between excited states

Next I illustrate how to plot density difference between various excited state (corresponding to
unrelaxed density). In fact this is very easy, you simply need to generate Multiwfn input files
containing NOs of the two excited states respectively, and then get their difference via the steps
illustrated in Sections 4.5.5 or 4.18.3.
I still take N-phenylpyrrole as example. We repeat aforementioned steps using the N-
phenylpyrrole.fch and N-phenylpyrrole.out to generate .molden files, when Multiwfn asks you to
input the index of excited states, we input 1-3, then NO_0001.molden, NO_0002.molden and
NO_0003.molden will be generated in current folder, clearly now we can study density difference
between 1-2, 1-3 and 2-3.
Assume that currently we want to visualize isosurface map of electron density difference
between the third and the first excited state, we reboot Multiwfn and input
NO_0003.molden
1 // One file will be dealt with the firstly loaded file
-,NO_0001.molden
After setting isovalue to 0.005, we will obtain below graph
669
The density difference map between other excited states can be obtained similarly.
Although you can also directly use your quantum chemistry program to generate wavefunction
file containing NOs for various excited states, the procedure is evidently much more cumbersome
than using Multiwfn, because as shown above, the advantage of Multiwfn is that it is able to
simultaneously generate .molden file containing NOs for a batch of excited states.
Calculate difference in fragment charge between excited states

Next, as an example, we will study difference of electron distribution at quantitative level by
comparing fragment charge of the pyrrole ring between excited states 3 and 1.
NO_0003.molden
1-9 // The pyrrole fragment, which is composed of atoms 1~9
11 // ADCH charge
1 // Use built-in atomic densities
You will find
Fragment charge: 0.598263
Namely the fragment charge of the pyrrole ring is 0.598 at the 3rd excited state. Repeat the
calculation for the NO_0001.molden, the charge of the pyrrole ring will be found to be 0.148. The
data shows that during the (hypothetical) transition from the 1st to the 3rd excited state, the pyrrole
fragment will lose 0.598-0.148=0.45 electron, this well explains why in the corresponding density
difference map there are obvious isosurfaces around the pyrrole ring and most of them are in blue
color. Do not forget that the current result still corresponds to unrelaxed excited state density.
Wavefunction analysis of an excited state (based on relaxed density)

At final part of this section, I show how to carry out wavefunction analysis for an excited state
based on its relaxed density. N-phenylpyrrole is still taken as example.
We prepare a Gaussian input file with below content. The full file as been provided as
examples\excit\ N-phenylpyrrole_relaxS2.gjf.
%chk=C:\N-phenylpyrrole_relaxS2.chk
# TD(nstates=5,root=2) cam-b3lyp/6-31+g(d) density
...[ignored]
Run this file by Gaussian, then the density matrix corresponding to relaxed density of the 2nd
excited state will be written into the N-phenylpyrrole_relaxS2.chk. Then use formchk utility to
convert it to N-phenylpyrrole_relaxS2.fch (which can also be directly downloaded from
http://sobereva.com/multiwfn/extrafiles/N-phenylpyrrole_relaxS2.zip).
We first need to transform the density matrix to NOs. Boot up Multiwfn and input
N-phenylpyrrole_relaxS2.fch
200 // Other functions, part 2
16 // Generate NOs based on the density matrix in .fch/.fchk
CI // The label of TDDFT density matrix in the file is “CI”
y // Export new.molden in current folder and then automatically load it, which contains the
newly generated NOs
Now the orbitals in memory have corresponded to the NOs generated based on the relaxed
density of the 2nd excited state, then we can do arbitrary wavefunction analysis, for example
670

1 // Mayer bond order
From the output you can find the bond order of the N5-C10 is 0.756, while as shown earlier,
this value corresponding to unrelaxed density is 0.796. The small difference implies that the analysis
result based on unrelaxed density is at least qualitatively correct and as useful as those derived based
on the accurate but expensive relaxed density.
It is also possible to calculate density difference based on relaxed density between two excited
states. You need to repeat above steps twice to respectively generate .molden file for two different
excited states, and then get density difference as usual based on the two .molden files.
For Gaussian users, in fact one can use such as “# PBE1PBE/6-31G* out=wfn TD(root=x)” keywords to export
NOs of excited state x to specific .wfn file, which can also be employed as input file for performing wavefunction
analysis of the excited state. However, do not forget that many functions in Multiwfn require basis function
information, which cannot be provided by .wfn file, thus in this case the kind of analyses can used is severely limited.
In addition, by solely using Gaussian it is also possible to yield and store the NOs to .fch file, as explicitly described
at the beginning of Chapter 4, however this procedure is relatively cumbersome. Notice that the NOs generated in
these ways correspond to relaxed excited state wavefunction. If you only need the NOs correpsonding to the
unrelaxed excited state wavefunction, simply adding “density=rhoci” keyword in route section.
4.19 Orbital localization analysis
Orbital localization is a very powerful and useful technique, it can transform canonical
molecular orbitals, which often show highly delocalized character, to localized form, which is
known as localized molecular orbital (LMO). The LMOs have close relationship with many classical
chemistry concepts such as chemical bond and lone pair, therefore they can be used to analyze and
unveil many problems of chemical interest. Before reading this section please read Section 3.22 first
to gain some basic knowledges. Some examples in other sections, such as the example in Sections
4.8.4 and 4.100.22, also utilized orbital localization function.
CAUTION: The default orbital localization method cannot be used when diffuse functions are
presented, otherwise the result may be misleading or completely meaningless! If diffuse functions
have to be employed, you should change to Foster-Boys orbital localization method.
4.19.1 Localizing molecular orbital of 1,3-butadiene by Pipek-Mezey

method
This section illustrates the use of orbital localization analysis of Multiwfn with trans-1,3-
butadiene as example.
Basic steps of performing LMO analysis

The input file is examples\butadiene.fch, which was yielded at B3LYP/6-31G** level. You can
first visualize its MOs via main function 0, you will found all MOs are highly delocalized, none of
them have direct correlation with classical concept of chemical bond theory. Now we perform orbital
localization to transform them to more chemically meaningful orbitals. Boot up Multiwfn and input
19 // Orbital localization. By default, Pipek-Mezey localization method based on Mulliken
population is used
671
2 // Perform localization for both occupied and unoccupied MOs

Since this system is small, convergence of localization iteration immediately achieved, and the
major character of the resulting orbitals are printed. After that Multiwfn automatically exports the
localized orbitals as new.fch in current folder and then load it. Now, the orbital coefficients in
memory have been completely updated to the localized orbitals. You can use different ways to
characterize them. We enter main function 0 to visualize these orbitals, you will find all orbitals
show strong localization character, in particular the occupied ones. Below are screenshot of three
localized orbital involving C6-C8, the first two are occupied, and they correspond to  bond and 
bond, respectively; the third one is unoccupied, it can be ambiguously identified as anti- bond
orbital.
Obtaining LMO energies

It is also possible to obtain energy for the localized MOs, you should provide Fock matrix
(strictly speaking, Kohn-Sham matrix in current case). To realize this for present system, we enter
main function 100 and choose subfunction 2, and then select option 10 to write current molecular
geometry as a Gaussian input file. We properly modify it to make Gaussian output NBO .47 file,
which contains Fock matrix, and we need to ensure that the calculation level is identical to the .fch
we used (i.e. B3LYP/6-31G**). The .gjf file after modification has been provided as
examples\butadiene_47.gjf. Run it by Gaussian, you will find the resulting BUTADIENE.47 at C:\
(this file has already been provided as examples\butadiene.47).
We have finished preparation. Now reboot Multiwfn and input
-4 // Allow Multiwfn to generate and print energy for localized orbitals
examples\butadiene.47
2 // Localize both occupied and unoccupied orbitals
After convergence finished, Multiwfn evaluates energy of the localized MOs and sort orbitals
according to their energies from low to high. The information of the highest occupied localized
orbitals and lowest unoccupied localized orbitals are shown below
14 Energy: -0.2750727 a.u. -7.4851 eV Type: A+B Occ: 2.0
15 Energy: -0.2750727 a.u. -7.4851 eV Type: A+B Occ: 2.0
16 Energy: 0.2185910 a.u. 5.9482 eV Type: A+B Occ: 0.0
17 Energy: 0.2185912 a.u. 5.9482 eV Type: A+B Occ: 0.0
Please plot these orbitals in main function 0 to try to identify their characters.
Showing center position of all LMOs

It is possible to show center position of all generated LMOs, so that the distribution of the
LMOs can be immediately understood. Here I illustrate how to realize this.
672

-8 // Switch status of “If calculating center position and dipole moment of LMOs” to “Yes”
-6 // Change localization method
10 // Foster-Boys
1 // Localize occupied orbitals (center position of unoccupied LMOs is generally
uninteresting therefore we do not localize unoccupied MOs)
n // Skip the dipole moment analysis
As you can see from the prompts, after orbital localization has been finished as usual and the
newly generated new.fch has been automatically loaded into Multiwfn, the program calculates center
position of each LMO. In order to make visualization of the centers easy, Multiwfn adds Bq atoms
(i.e. ghost atoms”) into the current system to represent the centers of the LMOs. The center
coordinates as well as correspondence between LMO indices and Bq indices are automatically
exported to current folder as LMOcen.txt. The content of this file of current instance is
LMO 1 corresponds to Bq 11, X,Y,Z: 1.1372 0.7759 -0.0000 Bohr
LMO 2 corresponds to Bq 12, X,Y,Z: -1.1372 -0.7759 -0.0000 Bohr
LMO 3 corresponds to Bq 13, X,Y,Z: 1.1372 3.3081 -0.0000 Bohr
...[ignored]
Then we return to main menu and enter main function 0 to visualize the LMO centers and
orbital isosurfaces. The plotting settings have been automatically set to a special status for best
visualizing LMOs purpose, currently you can see:
From this graph we can very clearly understand distribution of the generated LMOs. Each cyan
sphere is a ghost atom, representing center of a LMO. Under current setting the index of the ghost
atoms starts from 1, therefore the index shown in the graph directly corresponds to the LMO index.
For example, we want to simultaneously visualize the two LMOs at boundary C-C bond, since the
cyan spheres with labels 8 and 11 locate around the bond, we choose orbital 8 in the orbital list to
display it, then select “Show+Sel. isour#2” and choose orbital 11 to. After that, change isovalue to
0.13 and set the drawing style as transparent, you will see
673
Evidently, the two LMOs obtained via Foster-Boys method collectively represent the double-bond
character of the boundary C-C bond. The two LMOs are known as “banana” orbitals and do not
exhibit - separation character.
You can also try to visualize the center position of the LMOs generated via Pipek-Mezey
algorithm. Because this method represents each double-bond as a pair of separated  and  LMOs,
whose centers should be very close to each other, from below graph you can find the centers indeed
can hardly be distinguished:
It is worth to note that there are two ways to rapidly find the index of the LMO corresponding
to the bond of your interest. The first one is examining the orbital compositions printed during the
orbital localization, the second one is directly visualizing the LMO centers and find the index
showing above the cyan sphere at proper place, as I just illustrated.
4.19.2 Analyze variation of localized molecular orbitals for SN2

reaction
In this example, we study variation of localized molecular orbitals (LMOs) along with reaction
path to visualize variation of chemical bonds, a typical SN2 reaction is taken as instance.
This SN2 reaction involves formation of F-C bond and break of C-H bond, therefore in the
following analysis we will focus on corresponding two LMOs. The IRC of the SN2 reaction is shown
below, five points are taken into account, their .fch files have been provided in examples\SN2 folder.
674
First, we generate LMOs for transition state (TS) geometry. Boot up Multiwfn and input
examples\SN2\TS.fch
1 // Localize occupied orbitals
0 // Visualize orbitals
We can find there are two orbitals respectively corresponding to F-C and C-H bonds, we draw
them together for easier comparison
The green-blue isosurface and purple-yellow isosurfaces clearly portray the orbitals corresponding
to F-C and C-H bonds, respectively.
We draw the same kind plot for R.fch, TS-1.fch, TS+1.fch and P.fch, then put the graph together,
as shown below
From the graph, the variation of chemical bonds during the SN2 reaction is quite clear (R→TS-
1→TS→TS+1→P). In the reactant state, the green-blue isosurface corresponds to the lone pair of
fluorine atom, while the purple-yellow isosurface shows typical covalent bond character of C-H. As
the reaction proceeds, the two LMOs vary smoothly, the C-F covalent bond character becomes more
and more prominent, and in the final product state, the LMO represented by purple-yellow
isosurface has already corresponded to 1s orbital of H- anion.
4.19.3 Characterize Re-Re bond of [Re2Cl8]2- anion
It is widely accepted that Re-Re bond in [Re2Cl8]2- anion is a quadruple bond, with
configuration of (242). The  bond results from overlap of d z 2 − p z hybrid orbitals of the two
675
rhenium atoms, the two  bonds stem from overlap of their dxz and dyz orbitals, while the  bond is
formed by overlap of their dxy orbitals. Can this classic concept be validated via orbital localization
analysis?
The .fchk file of [Re2Cl8]2- anion produced under B3LYP with 6-31G* for Cl and SDD for Re
is provided as examples\Re2Cl82-.fchk. Load it into Multiwfn and carry out orbital localization for
occupied orbitals, from the output we can immediately identify the four LMOs corresponding to the
Re-Re bond:
Almost two-center LMOs: (Sum of two largest contributions > 80.0%)
57: 7(Cl) 73.8% 2(Re) 22.5% 58: 10(Cl) 73.8% 2(Re) 22.5%
59: 4(Cl) 73.8% 1(Re) 22.5% 60: 2(Re) 48.1% 1(Re) 48.1%
61: 8(Cl) 73.8% 2(Re) 22.5% 62: 3(Cl) 73.8% 1(Re) 22.5%
64: 1(Re) 45.9% 2(Re) 45.9% 66: 6(Cl) 73.8% 1(Re) 22.5%
67: 9(Cl) 73.8% 2(Re) 22.5% 69: 5(Cl) 73.8% 1(Re) 22.5%
83: 1(Re) 45.9% 2(Re) 45.9% 84: 2(Re) 43.6% 1(Re) 43.6%
Hint: If you want to more easily find the indices of the LMOs corresponding to the Re1-Re2 bond, the best way
is setting "iprintLMOorder" in settings.ini to 1 before booting up Multiwfn. After that, the compositions of LMOs
will be printed in the order of atoms and atom pairs.
Then we enter main function 0 to visualize them (under default isovalue of 0.05):
From the isosurface maps of the LMOs, it is clear that LMO60 corresponds to  bond, LMO64
and LMO83 correspond to  bond and LMO84 corresponds to  bond. This observation supports
the quadruple bond argument.
However, if we calculate Mayer bond order, we will find the situation is not so simple. The
Mayer bond order of Re-Re bond calculated by main function 9 is 2.94, why the value is
significantly lower than 4.0, which is expected?
To gain deeper insight, we perform "Orbital occupancy-perturbed Mayer bond order" analysis
for Re-Re bond using main function 9, the output is
Mayer bond order after orbital occupancy-perturbation:
Orbital Occ Energy Bond order Variance
...[Ignored]
60 2.00000 -0.14354 2.001481 -0.935449
...[Ignored]
64 2.00000 -0.12793 2.276388 -0.660542
...[Ignored]
83 2.00000 -0.01749 2.276388 -0.660542
84 2.00000 0.02809 2.466317 -0.470613
The result shows that the  LMO has contribution of 0.935, each  LMO contributes 0.661,
676
and the  LMO contributes 0.471. Although in principle Mayer bond order cannot be exactly
decomposed, these data is sufficient to help us to understand relative importance of each bonding
LMO. Clearly  bond is of most importance to the Re-Re bond, while the importance of the  bond
is relatively lowest.
Now a new problem arises, why the four LMOs have different contributions to Mayer bond
order of Re-Re bond? This may be answered by visualizing their isosurfaces with lower value of
isovalue. The graphs of the four LMOs with isovalue of 0.01 are illustrated below
From the graph we find that LMO60 basically only occurs around the two Re atoms, therefore
its contribution to Re-Re Mayer bond order should be close to 1.0. Both LMO64 and LMO83
slightly delocalize to four Cl atoms, therefore they do not purely show Re-Re bond character and
thus have lower contribution to Re-Re Mayer bond order. The LMO84 delocalizes to all the eight
Cl atoms, it is naturally expected that its contribution to the Re-Re Mayer bond order should be the
smallest.
By the way, if you have interesting, you can carry out orbital occupancy-perturbed Mayer bond
order analysis for Cr2, you will find the Mayer bond order is almost exactly 6.0, and all the six MOs
corresponding to Cr-Cr bond basically have contribution of 1.0, this is mainly because these orbitals
do not delocalize to any other atoms.
4.19.4 Study bond dipole moment based on two-center LMOs for

CH3NH2
Multiwfn supports a few methods for evaluating bond dipole moment, as mentioned in Section
4.A.11. In this section I use CH3NH2 as an example to show that based on two-center LMOs it is
possible to study bond dipole moment, which somewhat reflects bond polarity. Introduction of
related theory is given in Section 3.22.
examples\CDA\CH3NH2\CH3NH2.fch
-8 // Switch status of “If calculating center position and dipole moment of LMOs” to “Yes”
y // Perform dipole moment analysis for the LMOs
Now LMOdip.txt has been generated in current folder, in this file you can find below content,
which are calculated in a special method as shown in “Special topic 3” of Section 3.22:
Single-center orbital dipole moments (a.u.):
677
1 ( 5N ) X/Y/Z: -0.04465 0.01500 0.00001 Norm: 0.04710
2 ( 1C ) X/Y/Z: 0.00010 0.00014 0.00000 Norm: 0.00017
9 ( 5N ) X/Y/Z: -1.28448 0.41598 0.00001 Norm: 1.35016
Sum X/Y/Z: -1.32903 0.43112 0.00001 Norm: 1.39721
Two-center bond dipole moments (a.u.):
3 ( 5N - 6H ) X/Y/Z: -0.08951 -0.20801 0.30542 Norm: 0.38021
4 ( 1C - 4H ) X/Y/Z: -0.06753 0.08375 -0.00000 Norm: 0.10758
5 ( 5N - 7H ) X/Y/Z: -0.08950 -0.20802 -0.30543 Norm: 0.38022
6 ( 5N - 1C ) X/Y/Z: 0.17708 0.25312 -0.00000 Norm: 0.30891
7 ( 1C - 3H ) X/Y/Z: 0.08532 0.06534 -0.09566 Norm: 0.14387
8 ( 1C - 2H ) X/Y/Z: 0.08532 0.06534 0.09566 Norm: 0.14387
Sum X/Y/Z: 0.10119 0.05152 -0.00001 Norm: 0.11355
After entering main function 0, we can see below graph
By comparing the graph and content of LMOdip.txt, you can find LMO1 and LMO2, which
respectively correspond to core orbitals of N5 and C1, have negligible polarity (i.e. the “norm” are
basically zero), reflecting that the LMO centers are very close to the nuclear positions. The LMO9
corresponds to lone pair orbital of N5, its “norm” is as high as 1.35 a.u., showing that the LMO
center deviates from N5 nucleus significantly. LMOs 3 and 5 correspond to the N-H bonds, LMOs
4, 7 and 8 correspond to the C-H bonds, it is well known that polarity of C-H should be lower than
N-H, this point is well reflected by the difference in their “Norm” values. The “Norm” of LMO6,
which corresponds to the C-N bond, is 0.3089, well indicating the fact that C-N is a polar bond.
In addition, in current system the Y coordinates of N5 and C1 are -1.438 and 1.330 Bohr,
respectively. The Y component of the bond dipole moment of LMO6 is 0.253, which is an evident
positive value. This observation shows that the negative and positive charge centers are on the N5
and C1 sites, respectively, corresponding to the fact that nitrogen has larger electronegativity than
carbon.
678
4.20 Visual study of weak interactions
4.20.1 Studying weak interaction in 2-pyridoxine 2-aminopyridine by

NCI method
Please carefully read Section 3.23.1 first to understand theory and how to use Multiwfn to carry
out NCI analysis. Weak interaction character in 2-pyridoxine 2-aminopyridine system has already
been studied using AIM theory in Section 4.2.1, in this section we also perform NCI analysis for it,
and meantime I will show how to plot color-filled RDG map and AIM topology graph as a single
map.
examples\2-pyridoxine_2-aminopyridine.wfn
1 // NCI analysis
After a while, calculation of grid data is finished. You can then select -1 to visualize scatter
map, from which interactions in the system can be examined preliminarily.
Since there are spikes (points nearly approaching bottom) at very negative region of sign(2),
according to the description of NCI method given in Section 3.23.1, we immediately know that this
dimer system must contain evident attractive intermolecular interaction. There is also a spike at very
positive side, therefore steric effect should exist in present system.
Then select option 3 to export func1.cub and func2.cub, and use VMD to plot color-filled RDG
map using the method described in Section 3.23.1, we obtain below graph (to make the graph
brighter, I enabled additional lighting via "Display" - "Light 3")
679
The type of interactions in this system now is very clear. Steric effect exists within the aromatic ring
because the color of corresponding isosurfaces is red. The two hydrogen bonds NH-N and N-
HO should be strong, since corresponding RDG isosurfaces have blue color. Between two
hydrogens there is also a RDG isosurface, since its color is green, it should be regarded as van der
Waals interaction, which is very weak. This region corresponds to the spikes in the middle of the
scatter map.
Showing AIM information in colored RDG map

Below I illustrate how to plot AIM critical points (CPs) and bond paths on the color-filled RDG
map, the resulting graph will be more informative, since the trace of interactions can be vividly
shown, while this kind of information is not explicitly revealed by NCI analysis.
Return to main menu and input below commands to search CPs, generate paths and then export
them as CPs.pdb and paths.pdb in current folder, respectively.
2 // Search nuclear CPs
3 // Search bond CPs
8 // Generate bond path
-4 // Modify or export CPs
6 // Export CPs as CPs.pdb in current folder
0 // Return
-5 // Modify or print detail or export paths
6 // Export paths as paths.pdb in current folder
Then we close Multiwfn. Drag CPs.pdb and paths.pdb into VMD main window in turn to load
them, select "Graphics"-"Representation", change "Selected molecules" to the second term
(corresponding to CPs.pdb), change "Drawing Method" to "VDW" and set "Sphere Scale" from the
default 1.0 to the minimal value 0.1. Note that in the CPs.pdb file, C, N, O, F atoms correspond to
(3,-3), (3,-1), (3,+1), (3,+3), respectively. Here we only want to draw bond CPs (i.e. (3,-1) type of
CPs) on the graph with yellow color, therefore we input "nitrogen" in "Selected Atoms" text box
and press ENTER button, then change "Coloring Method" to "Color ID" and select "4 yellow" in
the drop-down box. Currently, the graph looks like below
680
Probably you feel that the spheres corresponding to CPs are too large, however we cannot
further decrease the "Sphere Scale" using graphical window due to limitation of VMD. To make the
spheres smaller, you have to use corresponding command in VMD console window. To find proper
command to do this, we select "File"-"Log Tcl Commands to Console", then change "Sphere Scale"
to other value (e.g. 0.2), you will immediately see corresponding text-line command in the VMD
console window, for present the command is mol modstyle 0 1 VDW 0.200000 12.000000, where
the argument 0.2 corresponds to size of the spheres. Therefore, to decrease the sphere size to e.g.
0.09, we should input mol modstyle 0 1 VDW 0.09 12.000000 in the console window, then in the
VMD graphical window you will see the spheres have already become smaller.
Next, we change the appearance of paths. In the "Graphics"-"Representation" panel, select the
third term in "Selected molecules" (corresponding to paths.pdb), change the drawing method to
"VDW", set coloring method as "Color ID" and select "3 orange", then use abovementioned skill to
set the sphere scale to 0.02. The final graph is shown below.
From this graph, not only the weak interaction regions are clearly revealed, but also the interaction
paths are vividly exhibited. Notice that there is no CP and path corresponding to the hydrogen-
hydrogen interaction, because in this region there is no position having vanishing electron density
gradient, and this is why in the scatter map the spike corresponding this H-H interaction does not
completely approach the bottom of the map. This observation reflects an advantage of NCI analysis
over AIM analysis, namely interaction can be revealed even if there is no corresponding bond CP.
The steps of showing the CPs and bond paths in VMD is somewhat lengthy, therefore I strongly
681
recommend to use VMD plotting scripts to automatically do all of above steps, please follow part 4
of this video tutorial: https://youtu.be/e4FpVc9ao48, you will find the process is extremely easy.
More information about this script can be found in Section 4.2.5.
4.20.2 Studying weak interaction in DNA by NCI method based on

promolecular density
Please read Sections 3.23.1 and 3.23.2 if you are not familiar with NCI analysis and the concept
of promolecular approximation. In this example, we will carry out NCI analysis for a DNA fragment
consisting of 10 base pairs. Since this system is fairly large, promolecular approximation is used for
approximately rapid construction of the moelcular electron density. This example is also illustrated
as part 3 of this video tutorial: https://youtu.be/e4FpVc9ao48.
Here we only study weak interaction character of a local DNA region, which is enclosed in the
transparent box:

examples\DNA.pdb
2 // NCI analysis based on promolecular density
7 // Use mode 7 for defining grid data
84,565 // Use midpoint of atom 84 and 565 as center of grid data. You can view molecule
682
structure in your favourite visualization tool to find two proper atoms used to define center
120,120,120 // Because the spatial scope of grid data is large, we need relatively large number
of grid points, otherwise the grid spacing will be too large, which results in bad quality of RDG
isosurfaces
9,9,9 // Set the extension distances in all directions to 9 Bohr
Hint: You can also use mode 10 to set up box interactively in a GUI window, the box size and position of box
center is more controllable
After the calculation of grid data is finished, choose option 3 to export sign(2) and RDG as
func1.cub and func2.cub respectively in current folder, and then copy them as well as
examples\RDGfill_pro.vmd to VMD installation folder. Boot up VMD and input source
RDGfill_pro.vmd in console window to draw color-filled RDG isosurface map, the resulting graph
after some adjustments is shown below. (For better visualization effect, open "graphics"-
"Representation" and change the drawing method of DNA to Licorice, change bond radius to 0.2.
Then enter “Display”-“Display settings...” to set “Cue Mode” to "Linear" and set “Cue Start/End”
to 2.25 and 3.75, respectively, so that distant atoms can be substantially screened). Finally, you will
get the graph below
It is clear that there are - stacking interaction between neighbouring base pairs (big flat
isosurfaces), and there are two strong hydrogen bonds among each base-pairs. The region pointed
by red arrow seems to be hydrogen bond because it connects hydrogen and oxygen, however since
the filled-color is green, we can conclude that it can only be regarded as vdW interaction.
The default isovalue 0.3 in RDGfill_pro.vmd is suitable for present case, but may be not
suitable for exhibiting weak interaction region of other systems, in that situations you need to adjust
it manually. You can either edit the .vmd file, or choose “Graphics”-“Representation” in VMD, then
select the representation with style “Isosurface” and reset isovalue by inputting expected value in
683
text box.
.
4.20.3 Visually studying weak interaction for water in bulk

environment by aNCI method
If you are unfamiliar with NCI and aNCI methods, please first read the introduction given in
Section 3.23.1, 3.23.2 and 3.23.3. The aNCI method illustrated in this section is a generalization of
NCI analysis method for fluctuation environment, e.g. molecular dynamics (MD) process.
In this example I will show how to use Multiwfn to visually study the weak interaction between
water molecules in the MD simulation of bulk water system. You can use any program to perform
the MD simulation, as long as you know how to convert the resulting trajectory from private format
to the general .xyz format, which can be recognized by Multiwfn and utilized in aNCI analysis.
Here I assume that you are a GROMACS 4.5 user. The detailed steps of the MD process are
given below (very different to GROMACS >= 5.0), all of the related files can be found in
examples\aNCI folder. If you do not want to perform the MD simulation by yourself, you can
directly download the wat.xyz, which will be utilized in aNCI analysis later:
http://sobereva.com/multiwfn/extrafiles/aNCI_wat_xyz.zip.
Generating MD trajectory by GROMACS

First, build a file named "emptybox.gro", which records a blank box, and the side length in
each direction is 2.5nm. Then run below command to fill the box with waters.
genbox -cp emptybox.gro -cs spc216.gro -o water.gro
Run below command and select "GROMOS96 53a6 force field" to obtain the top file of the
bulk water system. SPC/E water model is employed.
pdb2gmx -f water.gro -o water.gro -p water.top -water spce
Then carry out NPT MD by 100ps to equilibrate the bulk water at 298.15K, 1atm environment.
grompp -f pr.mdp -c water.gro -p water.top -o water-pr.tpr
mdrun -v -deffnm water-pr
Using VMD program to load water-pr.gro, select a water close to the center of the box. We
select the water with resid index of 101, which is highlighted in below graph. Note that the two
hydrogens in this water have index of 302 and 303, and the index of the oxygen is 301.
684
This water will be freezed in the following MD simulation. In order to do so, we generate index
file, namely inputting below commands
make_ndx -f water-pr.gro
ri 101
q
Run following command to do 1ns equilibrium MD simulation at 298.15K, the trajectory will
be saved every 1ps, and finally we will obtain 1000 frames. The water with resid index of 101 is
freezed via the keyword "freezegrps = r_101". Note that NVT ensemble instead of NPT is used,
because NPT process will scale the coordinate of the atoms, which somewhat destorys the effect of
freezing.
grompp -f md.mdp -c water-pr.gro -p water.top -o water-md.tpr -n index.ndx
mdrun -v -deffnm water-md
Load the water-pr.gro into VMD, then load water-md.xtc to the same ID, select "File"-"Save
Coordinate..." option and set the file type as xyz, then input all in the "Selected atoms" box, input 1
and 1000 in the "First" and "Last" window, respectively. Finally, click "Save" button to convert the
GROMACS trajectory to wat.xyz.
IMPORTANT NOTICE: The wat.xyz currently records atom names rather than element
names. For example, if you open this file via text editor, you will find each water contains OW,
HW1 and HW2, which are atom names. However, in standard .xyz file, only atom elements should
be recorded. Therefore, in general cases, you should manually replace all atom names in the .xyz
file generated by VMD with element names. Fortunately, this step can be skipped in present example
because there is no element in the periodic table named OW, HW1 and HW2, therefore, only the
first letter of atom names will be employed by Multiwfn to try to identify their elements, and they
can be properly recognized as oxygen and hydrogens, since after loading the .xyz file, you can find
prompt “Formula: H1022 O511” on the screen, which is what we expected. If you find there are
undesired elements in the the “formula”, that means you have to replace the corresponding atom
names in the .xyz file as their actual element names.
Generating grid data by Multiwfn

wat.xyz
685

3 // aNCI analysis
1,1000 // The range of the frames to be analyzed
7
301,301 // Using atom 301 (the oxygen of the freezed water) as the box center of the grid
data
80,80,80 // The number of grid points in each side
4.5,4.5,4.5 // Extend 4.5 Bohr in each side
Now Multiwfn starts to calculate electron density, its gradient and Hessian of each frame, then
their average quantities will be obtained, and finally Multiwfn calculates average RDG and average
sign(2). The whole process is time-consuming; at a common Intel 4-cores computer about half an
hour will be consumed. (Note that the electron density I referred here is produced by promolecular
approximation, which is constructed by simply superposing the density of the atoms in their free-
states)
After the calculation is finished, you can select option 1 to check the scatter plot between
average RDG (X-axis) and average sign(2) (Y-axis), see below, there is also option used to export
the corresponding data points to plain text file.
Select option 6 to export the grid data of average RDG and average sign(2) as avgRDG.cub
and avgsl2r.cub in current folder, respectively.
Since we wish to check the stability of weak interaction, we also select 7 to export thermal
fluctuation index to thermflu.cub in current folder. Note that this process requires recomputing
electron density of each frame, and thus is time-consuming.
Analysis
Copy avgRDG.cub, avgsl2r.cub, thermflu.cub as well as avgRDG.vmd and avgRDG_TFI.vmd
in "examples\aNCI" folder to the directory of VMD program.
Simply boot up VMD and input source avgRDG.vmd in its console window, the average RDG
isosurface will be shown with isovalue of 0.25, meantime the average sign(2) is mapped on the
isosurface by various colors. In order to make the graph clearer, one should screen unrelated atoms,
that is enter "Graphics" - "Representation"， then select the entry whose "style" is "CPK", and
686
input serial 301 302 303 in the "Selected Atoms" box and then press ENTER button. Now only the
water with resid index of 101 presents in the graph. After proper rotation and translation of view,
you will see
Unfortunately, around the water of interest, there are large amount of noisy isosurfaces, which
somewhat messed up the graph, thus it is better to shield them. This aim can be achieved by main
function 13 of Multiwfn, the steps are described below.
Then boot up Multiwfn and input
avgRDG.cub
13 // Set the value of the grid points far away from specific atoms
1.5 // If the distance between a grid point and any selected atoms is longer than 1.5 times of
vdW radius of corresponding atom, then the value of the grid point will be set as given value
100 // An arbitrarily large value (should be larger than the isovalue of the RDG isosurfaces)
2 // Inputting selected atoms by hand
301-303 // The index of the atoms are 301, 302 and 303
0 // Export the updated grid data to a new cube file
avgRDG.cub // The name of the new cube file
Copy the newly generated avgRDG.cub to the folder of VMD program to overwrite the old
one, then use the script avgRDG.vmd again to plot the graph, after some adjustments you will see
(for clarity, the view of two sides are shown at the same time)
687
The graph we obtained this time is very clear. The color scale is from -0.25 to 0.25,
corresponding to the color variation of Blue-Green-Red. More blue denotes electrostatic interaction
or H-bond effect in corresponding region is stronger, and more red suggests more intensive steric
effect. Green region implies low electron density, corresponding to vdW interaction. From the graph
one can see that there are two blue ellipses near the two hydrogens, rendering that in the MD process,
strong H-bonds are formed due to the O-H group. The slender green isosurface exhibits in which
direction this water prefers to interact with other waters by vdW interaction. There is a big lump of
isosurface above the oxygen, on which the red color appears in the middle part, while blue color
occurs at the two ends; the latter reflects that the two lone pairs of the oxygen act as H-bond
acceptors during the simulation, while the former reveals the repulsive interaction zone between
waters.
Next, we study the stability of the weak interactions. First disable present isosurfaces, and then
input the command source avgRDG_TFI.vmd in the console window, after some adjustments you
will see
The color scale is 0~1.5, still corresponding to the color transition of Blue-Green-Red. More
blue (red) means the thermal fluctuation index (TFI) is smaller (larger), and hence the weak
interaction in corresponding region is more stable (unstable). The graph shows that it is stable that
the water behaves as H-bond donor, while the stability of the water acting as H-bond acceptor is
slightly weaker; the vdW interaction region is totally red, rendering that vdW interaction is evidently
688
unstable compared to hydrogen bond.
4.20.4 Simultaneously revealing covalent and noncovalent interactions

in phenol dimer by IRI analysis
Before following this section please read Section 3.23.8 to gain basic knowledge about IRI
(Interaction Region Indicator) function. Here I use two examples to respectively show how to draw
sign(2) mapped IRI isosurfaces and IRI plane map to simultaneously reveal both covalent and
noncovalent interaction regions
Plotting sign(2) mapped IRI isosurface

Phenol dimer will be taken as instance. You will find almost all steps are identical to the NCI
analysis described in Section 3.23.1.
4 // IRI analysis
3 // High quality grid
Move func1.cub, func2.cub and plotting script examples\IRIfill.vmd to VMD folder. Then boot
up VMD and input source DORIfill.vmd in its console window to execute the script, you will
immediately see below graph.
The graphical effect is quite satisfactory. The weak interaction regions are exhibited as good
as NCI analysis, while the chemical bond regions are nicely portrayed by blue isosurfaces,
indicating strong attractive interaction. The color transition and color scale employed in the
IRIfill.vmd is the same as that of NCI analysis (i.e. RDGfill.vmd).
Note that the most suited isovalue of IRI may be different for different systems, the default
isovalue in the IRIfill.vmd script is 0.9. You can modify it in the .vmd script file, or manually adjust
it in "Graphics" - "Representations" panel of VMD.
Plotting plane map for IRI

Sometimes it is also useful to plot plane map for IRI to reveal interaction regions in a specific
plane. Below I will illustrate how to realize this for examples\GC.wfn, which is a base pair dimer.
Set "iuserfunc" in settings.ini to 99, so that user-defined function will correspond to IRI. Then
689
boot up Multiwfn and input

examples\GC.wfn
4 // Plot plane map
100 // User-defined function (now corresponds to IRI)
[Press ENTER button to use default number of grids]
1 // 1 Bohr
1 // XY plane, which is the plane all atoms are
0 // Z=0
1 // Rainbow
4 // Enable showing atom labels and reference point
16 // Pink
14 // Brown
-1 // Plot again
The red color areas (IRI  1.0) in this map clearly reveal the regions where chemically notable
interactions occur, including the chemical bond and steric regions within each monomer as well as
the weak interaction regions between the two bases.
690
4.20.5 Simultaneously revealing covalent and noncovalent interactions

in phenol dimer by DORI analysis
Frankly speaking, DORI (Density Overlap Regions Indicator) analysis is useless, since the IRI
analysis illustrated in the last section has similar capacity while works evidently better than DORI.
However I still use phenol dimer system to illustrate how to perform DORI analysis in Multiwfn.
Please read Section 3.23.4 first to understand basic knowledge about DORI.
4 // DORI analysis
Move func1.cub, func2.cub and plotting script examples\DORIfill.vmd to VMD folder. Then
boot up VMD and input source DORIfill.vmd in console window, you will immediately see below
graph.
The graphical effect of DORI map is obviously not as good as IRI map, especially the edge
region of the isosurfaces corresponding to weak interactions look quite ugly. Furthermore,
calculation cost of DORI is higher than IRI, therefore there is no any reason to use DORI.
The DORIfill.vmd employs the same color transition and color scale as RDGfill.vmd and
IRIvill.vmd. The most suited isovalue of DORI is different for different systems, the default isovalue
in the DORIfill.vmd script is 0.95, you can modify it if you find it is inappropriate.
4.20.6 Visualizing and analyzing van der Waals potential

Note: Chinese version of this topic is http://sobereva.com/551, which contains much more examples and
additional discussions.
The concept of van der Waals (vdW) potential proposed by me has been introduced in Section
3.23.7, in below examples I will illustrate how to visualize and analyze it. Do not forget to check
my this paper: ChemRxiv (2020) DOI: 10.26434/chemrxiv.12148572, in which idea,
implementation and application of the vdW potential analysis are carefully introduced.
691
4.20.6.1 Example 1: Helicene

In this example I will illustrate visualizing the vdW potential for helicene, whose structure is
shown below
To study vdW potential, you need to select a probe atom. For example, in this instance we want
to employ He atom as probe atom, therefore we change "ivdwprobe" parameter in settings.ini to 2.
examples\helicene.xyz
6 // Visualization of van der Waals potential
3 // High quality grid (the computational cost of vdW potential is extremely low, therefore
we use relatively good grid quality here)
As can be seen from the menu, now you can directly visualize vdW potential or its two
components, namely repulsion potential and dispersion potential. You can also export their grid data
to cube files. The unit used in this module is kcal/mol.
Now we choose option 3 to visualize isosurface map of the vdW potential, the resulting map
with isovalue of 0.6 (kcal/mol) is shown on the left side below. If you want to only visualize the
negative part, there is a trick: Set isovalue to -0.6 and unselect "Show both sign" check box, then
select "Isosurface style" - "Exchange positive and negative colors", the resulting map is shown on
the right side below ("Ratio of atomic size" has been changed to 4.0, the molecular representation
now corresponds to superposition of atomic vdW spheres).
692
In the above maps, the blue isosurfaces represent the regions where vdW potential is negative,
in these regions the dispersion attractive effect surpasses repulsion effect. It is expected that He
atom (or more generally, various small nonpolar molecules) tends to be attracted to the blue regions
due to dispersion interaction. All regions close to the nuclei are enclosed by green isosurface,
indicating that repulsive potential dominates the vdW potential in these places, this is the normal
case.
Can the vdW potential map be correlated with any practical observation? The answer is YES.
I carried out 2500 ps molecular dynamics simulation under 10 K for the complex consisted of a
helicene molecule and a He atom based on the GFN0-xTB theory in the Grimme's xtb code, the
trajectory frames (small spheres) and the isosurface of spatial distribution function (orange
isosurface) of the He atom are shown below
It can be seen that the majority of trajectory frames and the main distribution of spatial
distribution function are highly analogous to the blue regions in the vdW potential map showing
earlier, demonstrating that the vdW potential is indeed able to exhibit the favorable adsorption
regions if vdW interaction dominates the intermolecular interaction (of course, the precondition is
that both the adsorbate and the local region of adsorption sites are nearly nonpolar, otherwise
electrostatic interaction will largely control the adsorption behavior, in this case you should examine
electrostatic potential rather than vdW potential).
By the way, there is another way of visualizing grid data of vdW potential, namely plotting vdW potential
colored vdW surface map, the procedure in Windows system is: copying vdWpot.bat and vdWpot.txt from
"examples\scripts\vdWpot" folder to current folder, properly modifying the path of input file and VMD folder in
the .bat file, and then run this .bat file. After that, boot up VMD, copying all content from the vdWpot.vmd file in
"examples\scripts\vdwpot" folder to the VMD console window, you will see the vdW potential colored =0.001 a.u.
surface (note that the  is estimated using promolecular approximation). However, since the graphical effect of this
kind of map is not quite good, I prefer to study vdW potential in terms of isosurface map.
4.20.6.2 Example 2: Cyclo[18]carbon

The cyclo[18]carbon system was very extensively studied in my works Carbon, 165, 468
(2020), Carbon, 165, 461 (2020) and ChemRxiv (2019) DOI: 10.26434/chemrxiv.11320130. In this
example I will illustrate how to plot plane map of vdW potential for this system. The structure
optimized at B97XD/def2-TZVP level is given as examples\C18.xyz. As you can see, this system
is exactly planar and fully lying in the XY plane with Z=0.
693
We will plot color-filled map of vdW potential on the molecular plane. To do so, we need to
change user-defined function to vdW potential, namely setting "iuserfunc" parameter in settings.ini
to 92. We still use He element as probe atom like the last example, therefore "ivdwprobe" in
settings.ini should be set to 2.
examples\C18.xyz
4 // Plot plane map
[Press ENTER button directly to use recommended grid]
0 // Set extension distance
10 // 10 Bohr
1 // XY plane
0 // Z value
Click right mouse button on the graph to close it, then input
-0.8,0.8 // Note that unit is kcal/mol for vdW potential
12 // Dark green
14 // Brown
8 // Blue-White-Red
Now select option -1 to replot the map, you will see
In this map, red and blue colors represent positive and negative vdW potential, respectively.
694
From the map it can be seen that the vdW potential around the center of the cyclo[18]carbon is fairly
negative and thus this place has strongest ability in adsorbing nonpolar molecules. In the peripheral
region of this system the vdW potential is modestly negative, implying that nonpolar molecule can
only weakly interact with this system in this area due to dispersion attraction.
Please plot isosurface map of vdW potential as well as plane map of vdW potential in YZ plane,
so that you can better understand the overall distribution of vdW potential around this unusual
system. Via main function 3, you can also plot vdW potential between two given points, therefore
you can easily study variation of vdW potential starting from the ring center in the direction
perpendicular to the ring.
It is worth to note that it is very easy to obtain value of vdW potential at a given point, for
example, the center of this system, whose position is exact (0,0,0). You simply need to enter main
function 1, input 0,0,0, then choose either Bohr or Å as unit, then below information can be found
from screen:
User-defined real space function: -0.6979258635E+00
that is the vdW potential (with He as probe atom) is -0.70 kcal/mol.
Basin analysis on vdW potential

If you want to exactly obtain most negative value of vdW potential, you can make use of the
powerful basin analysis module, which is detailedly illustrated in Section 4.17. Now we use this
module to find the most negative value of vdW potential for the cyclo[18]carbon.
examples\C18.xyz
100 // User-defined real space function. Now it corresponds to vdW potential with He as
probe atom
Wait for a while until the calculation is complete, then you can choose option 10 to visualize
the located minima of the vdW potential
Only the minima around the ring center are of chemical interest. As can be seen, these minima
are automatically clustered together and share the same index (1), they can be viewed as degenerate
minima. The minima at peripheric region of the ring can be ignored because they basically resulted
from numerical noise.
Then input
695
-3 // Show information of attractors

y // Show attractors after sorting according to their values
Then you can see position and value of each minima before clustering. Finally you can see the
positions and values of the final minima (i.e. after clustering):
1 0.00000000 0.00000000 -0.01763924 -0.757315492E+00
21 0.00000000 -6.45596244 -0.52917725 -0.242350606E+00
36 0.00000000 -6.45596244 0.52917725 -0.242350606E+00
28 6.37658585 -1.08481336 -0.50271839 -0.242126321E+00
[...ignored]
Clearly, the minima of vdW potential (e.g. the points sharing attractor index of 1) in this system is
-0.757 kcal/mol. The printed coordinate "0.00000000 0.00000000 -0.01763924"
corresponds to average of all of its members.
If then you want to plot the minima via visualization softwares such as VMD, you can select
"-4 Export attractors as pdb/pqr/txt file", then select corresponding option to export attractors.pdb
in current folder, the meanings of atom index and residue index in this file are explicitly shown on
screen. Specifically, if you load this file into VMD, you can use "resid 1" as selection to plot the
global minima, since they share index of 1.
4.20.10 Visualize and quantify weak interactions by Independent

Gradient Model (IGM)
Please read Section 3.23.5 to gain basic knowledges about the Independent Gradient Model
(IGM) method proposed in Phys. Chem. Chem. Phys., 19, 17928 (2017). If you are not familiar with
NCI analysis, you should also first read Section 3.23.1, since many aspects of IGM analysis are
closely related to the NCI analysis. In this section I will illustrate the use of IGM analysis in
Multiwfn via several examples. More discussions and instances can be found in my blog article
"Investigating intermolecular weak interactions via Independent Gradient Model (IGM)" (in
Chinese, http://sobereva.com/407).
Notice that the IGM method illustrated in this section is based on promolecular approximation,
hence you can use any kind of input file containing atomic coordinate information, such as .xyz, .mol
and .pdb (see Section 2.5 for details). The VMD program used throughout this section is version
1.9.3, it is freely available at http://www.ks.uiuc.edu/Research/vmd/. Another form of IGM, namely
IGMH, is also supported in Multiwfn, see Section 4.20.11 for example.
4.20.10.1 Example 1: Guanine-cytosine (GC) base pair

The IGM framework includes many useful ideas and defined many useful concepts, a series of
analyses will be conducted in this example. A simple system guanine-cytosine (GC) base pair is
taken as instance here. Some analyses may be ignored in practical studies, and sequence of analyses
is completely arbitrary.
(1) Studying g function

We first study the distribution character of g function by plotting it as color-filled plane map.
696
examples\GC.pdb
4 // Plot plane map
22 // g
1 // XY plane
0 // Z=0
The graph shown on screen currently looks obscure, this is because the default color scale is
not suitable for present case, so we close the graph and input
1 // Set color scale
0,0.2 // Lower and upper limits
4 // Show atomic labels
1 // Red color
-2 // Set stepsize of axes
3,3,0.02 // Stepsize for X, Y and color bar
-1 // Plot again
You will see below graph
Above graph clearly reveals all interatomic interactions, and the magnitude of g is positively
relevant to interaction strength. As can be seen from the graph, all chemical bond regions have large
g value (the region with value higher than 0.2 is shown as white). The g function also outlines
three hydrogen bond regions among the base pair, where the g function has evidently smaller value
compared to the chemical bond regions.
697
The g can also be plotted as isosurface map. Return to main menu and input
22 // g
-1 // Show isosurface
The isosurface with isovalue of 0.15 and 0.02 are shown below (you can use higher quality of
grid or set the extension distance of grid data smaller to make the graph more smooth)
Since chemical bond regions have relatively large value of g, only chemical bonding
interactions are visible when isovalue is set to 0.15. Clearly, g may be used as a function to exhibit
chemical bonds like ELF and DORI functions, with additional advantage that only geometry
information is needed. The weak interaction regions can also be simultaneously visualized when
isovalue is decreased to a small value, e.g. 0.02.
(2) Studying ginter function between base pair

The ginter is a key function in the IGM analysis framework, it is designed to reveal interaction
regions between two (or even more) fragments defined by users. Here we plot this functions to study
the interactions between the two bases. Although as shown earlier, these interactions can also be
revealed by simply drawing g, the isosurfaces corresponding to intrafragment interactions severely
polluted the graph. Fortunately, the IGM analysis allows us to separate the g as ginter and gintra,
which solely reflect the contribution to g due to interfragment and intrafragment interactions,
respectively.
Return to main menu and input following commands
20 // Visual study of weak interactions
10 // IGM analysis
2 // Define two fragments
1-13 // Range of atoms in the first base
14-29 // Range of atoms in the second base
698

After calculation, you will see a post-processing menu. The meaning of each option has been
explained in Section 3.23.5. You can use option -1 to directly draw scatter map between g or ginter
or gintra versus sign(2). We choose option -1 and then select suboption 4, in the resulting map the
red and black points correspond to ginter and gintra, respectively.
If you are familiar with NCI method, you will naturally know how to discuss this graph, now
we try to identify character of peaks in the scatter graph. In the region where sign(2) is about -
0.04, you can find that the ginter has a remarkable peak (with height about 0.06), which implies
presence of hydrogen bonds. If ginter isosurface is set to an isovalue lower than about 0.06, the
corresponding isosurfaces should be visible in the graph. In the region where sign(2) is
approximately +0.02, there is also a small peak of ginter. Since positive sign(2) implies repulsive
interaction, the peak may reflect weak steric regions in the center of the two rings between the two
bases. In above scatter map, there is a very prominent peak of gintra around sign(2) = -0.3. Since
this peak corresponds to intrafragment interaction, and corresponding sign(2) is not only negative
but large, rendering attractive and strong interaction, the peak must result from chemical bond.
Using Multiwfn you can directly visualize isosurface of ginter and gintra. To do this, we close
the scatter map, select "4 Show isosurface of grid data", then choose corresponding options and
properly set isovalue, you will obtain below isosurface graphs
699
As can be seen, ginter and gintra indeed solely exhibit inter- and intra-fragment interactions,
respectively. This greatly facilitates separate discussion of the two kinds of interactions.
From the above ginter=0.02 isosurface map we are only able to visualize hydrogen bond
regions. To find steric region in the ring center between the two bases, we should further decrease
the isovalue of ginter to e.g. 0.008, as shown below. The ring-center steric regions are highlighted
by arrows.
(3) Drawing sign(2) mapped ginter isosurfaces

If sign(2) is mapped to ginter isosurfaces by different colors, then one can not only recognize
where weak interactions occur, but also immediately capture the character of the interactions.
Multiwfn itself is currently unable to plot color-filled isosurface map, we need to use VMD program
to do this, just like what we do in the NCI analysis.
Select "3 Output cube files to current folder" in IGM post-processing menu, then sign(2), g,
ginter and gintra will be exported to sl2r.cub, dg.cub, dg_inter.cub and dg_intra.cub in current folder,
respectively. Move sl2r.cub and dg_inter.cub as well as VMD plotting script
examples\IGM_inter.vmd into VMD folder. Boot up VMD, input source IGM_inter.vmd in the
console window, you will immediately see below graph
700
The default isovalue employed by the IGM_inter.vmd script is 0.01, you can manually change
isovalue by dragging the isovalue bar in "Graphics" - "Representation" panel. In the script the
default color scale of sign(2) ranges from -0.05 to 0.05, and the default color transition is Blue-
Green-Red. Therefore, the more blue the isosurface, the stronger the attractive interaction, while the
more red the isosurface, the larger the steric effect. Green zone in an isosurface implies that the
corresponding interaction is weak and may be regarded as van der Waals interaction.
Similarly, you can plot sign(2) mapped gintra isosurfaces. Just move sl2r.cub and
dg_intra.cub as well as the corresponding VMD plotting script examples\IGM_intra.vmd into VMD
folder, then input source IGM_intra.vmd in the console window to execute it.
(4) Decomposing interfragment interaction as atom and atom pair contributions

The atom pair g index is a quantitative indicator of contribution of an atomic pair to total
ginterbetween two fragments, while sum of all atom pair g indices of a given atom yields its atom
g index, which represents its importance for interfragment interaction. In Multiwfn the atom and
atom pair g indices can be easily evaluated. In the IGM post-processing menu, select option 6 and
then "2 High quality", Multiwfn will calculate these indices ("High quality" grid is absolutely fine
enough). The result will be exported to atmdg.txt in current folder, and then the program asks you
if also exporting atmdg.pdb in current, we choose y in present case.
Part of content of atmdg.txt is pasted here:
Atom dg index in fragment 1
Atom 6 : 0.417341
Atom 13 : 0.390307
Atom 8 : 0.342704
[ignored...]
Atom dg index in fragment 2
Atom 25 : 0.489922
Atom 24 : 0.350533
Atom 29 : 0.314113
[ignored...]
Atom pair dg index (zero terms are not shown)
13 24 : 0.192572
701
6 25 : 0.182227
8 29 : 0.169217
If you compare above data with the structure graph of present system shown below, you will
find the atom and atom pair g indices very meaningful and useful for discussing interfragment
interactions
The largest three atom g indices of fragment 1 are 25, 24 and 29, while that of fragment 2 are
25, 24 and 29, they are just the atoms closest to another fragment, undoubtedly they should have the
most important contributions to the interfragment interactions. H13-O24, N6-H25 and O8-H29 have
the largest atom pair g indices, reflecting that they are the most crucial interactions for formation
of the base pair.
Note that Multiwfn also exports IBSIW (intrinsic bond strength index for weak interactions)
to IBSIW.txt. Its definition has been described in Section 3.23.6. This index may have stronger
correlation with interatomic interaction strength than atom pair g index.
(5) Coloring molecular structure by atom g indices

Using VMD, it is also possible to map atom g indices on molecular structure, so that relative
importance of various atoms for interfragment interaction can be vividly exhibited. If needed, the
ginter isosurfaces can also be shown together. Now we plot such a map.
First, plot color-filled ginter isosurfaces using the IGM_inter.vmd script as mentioned earlier.
After that, we need to remove the default representation showing molecular structure, so we enter
"Graphics"-"Represenation", choose the first term (its current style is CPK), click "Delete Rep"
button. Then we drag the previously generated atmdg.pdb into VMD main window to load it. In this
file, the B-factor field records atom g indices; therefore, we should let VMD color the atoms
according to their B-factors. We enter "Graphics"-"Representation" again, set "Drawing method" to
CPK, and set "Coloring method" to Beta, then click "Trajectory" tab, set upper limit of color scale
to 9.0 and press ENTER button, now the system in graphical window should look like below
702
Since the color transition set by IGM_inter.vmd is Blue-Green-Red, the largest atom g index
in present case is 0.45 (as can be seen in atmdg.txt), while currently the range of color scale for
mapping atom g indices is set to 0.0~9.0, therefore in above map, the more green the atom, the
larger the atom g indices. The green atoms may be viewed as "hot atom" for interfragment
interactions. Contribution to interfragment interactions due to the blue atoms can be safely ignored,
since their g indices are very close to zero.
This example ends here, through this example I think you have already recognized basic steps
of IGM analysis. In next several examples I will illustrate more.
Skill: Plotting sign(2) colored IGM scatter map

In Part (2) of this example, I have shown how to directly use Multiwfn to plot scatter map
between sign(2) and various forms of g. As illustrated below, by using gnuplot program, the
scatter map can be colored with the same sign(2) color scale as VMD, so that you can easier
identify the correspondence between the peaks of the scatter map and the colored ginter isosurface
map. Very similar way has been employed to draw sign(2) colored RDG scatter map, as illustrated
in " Special skill 1" of Section 3.23.1.
Run below commands
examples\GC.pdb
10 // IGM analysis
1-13 // Range of atoms in the first base
14-29 // Range of atoms in the second base
2 // Output scatter points to output.txt
Then copy the exported output.txt and the plotting script examples\scripts\IGMscatter.gnu to
the file containing gnuplot executable file, then in this folder run command: gnuplot IGMscatter.gnu,
after that you will obtain IGMscatter.ps. If you open it by Acrobat or photoshop or Irfanview (with
ghostscript installed), you will see
703
In the current graph, the Y-axis corresponds to ginter. In the IGMscatter.gnu, the default color scale
is identical to the one adapted in IGMinter.vmd, namely -0.05~0.05.
If you want to use this script to plot gintra vs. sign(2), you should change "4:1:4" in the
IGMscatter.gnu to "4:2:4"; while if you want to plot g vs. sign(2), you should change it to"4:3:4".
If you find the range of X-axis is not appropriate, you can change the data behind "xtic" and
"xrange"; while if Y-axis is not appropriate, you can change "ytic" and "yrange".
4.20.10.2 Example 2: C60-coronene dimer

In this example, we will carry out IGM analysis for C60-coronene dimer, and finally plot below
map using VMD based on the data outputted by Multiwfn.
In above map, the major van der Waals interaction region (more specifically, the - stacking
region) is exhibited as green isosurface, the more red-colored atoms contribute to the interaction
more. If you think this graph is pretty and want to reproduce it, just follow below steps.
704
The pdb file of the dimer optimized at PM6-D3 level using Gaussian is provided as
examples\C60_coronene.pdb. Boot up Multiwfn and load it, then input below commands
10 // IGM analysis
1-60 // C60 is fragment 1
61-96 // Coronene is fragment 2
3 // Output cube files in current folder
6 // Evaluate atom and atom pair g indices
y // Export atmdg.pdb in current folder
Boot up VMD, enter below commands into VMD console window:
color scale method BWR
color Display Background white
axes location Off
display depthcue off
display rendermode GLSL
Then drag the atmdg.pdb into VMD main window, enter "Graphics"-"Representation", set
"Drawing Method" to "CPK", change bond radius from the default 0.3 to 0.8, set "Coloring Method"
to "Beta", set "Material" to "EdgyShiny". Then go to "Trajectory" tab, set lower and upper limit of
color scale to -5.0 and 5.0, respectively.
Next, we need to draw the ginter isosurface on the map. Drag dg_inter.cub into VMD main
window to load it, then enter "Graphics"-"Representation", change the default style from "lines" to
"Isosurface", set "Draw" to "Solid Surface", set "Show" to "Isosurface", then input 0.004 in the
"Isovalue" box and press ENTER button. Change "Coloring Method" to "ColorID" and select "7
green". Change the "Material" to "EdgyGlass".
Finally, we render the map. Select "File"-"Render", choose "Tachyon (internal, in-memory
rendering)" and click "Start Rendering" button, then you will obtain the graph shown at the
beginning of this section. The resulting graphical file is .tga format, you can use such as IrfanView
or photoshop to visualize it.
Note that in present example, color-filled effect is only applied to molecular structure, but not applied to
isosurface. This is because in VMD, setting of color transition is shared by all representations, that means we cannot
use Blue-White-Red color transition for molecular structure but use conventionally employed Blue-Green-Red color
transition for isosurface. Considering that in present system there is only one kind of interaction, namely van der
Waals interaction, and in conventional color-filled IGM map this region is basically colored by green, I decide
directly assign green color for the whole isosurface, so that we are able to freely set color transition mode for coloring
molecular structure.
4.20.10.3 Example 3: Oxazolidinone trimer

The IGM module of Multiwfn is extremely flexible, it can be applied to any number of
fragments. In this example I use oxazolidinone trimer to show this point. The geometry was taken
from J. Chem. Theory Comput., 11, 3065 (2015).
We first use ginter to reveal all interactions between the three monomers. Boot up Multiwfn
and input
examples\oxazolidinone_trimer.xyz
705
10 // IGM analysis
1-11 // Fragment 1: Monomer 1
Then we use aforementioned method to plot color-filled ginter isosurface map via
IGM_inter.vmd script, you will see below graph
From color of isosurfaces in the graph it is found that the 1-2 and 1-3 interactions correspond to
typical hydrogen bonding, while 2-3 interaction is significantly weaker and thus more appropriate
to be assigned as van der Waals interaction.
Assume that we only want to study interactions between 1-2 and 2-3, and meantime wish to
screen the ginter isosurface corresponding to 1-3 interaction, how to do that? The answer is: Only
define two fragments, making fragment 1 correspond to monomer 2, while making fragment 2
correspond to monomers 1 and 3. Now we do this, input below commands
10 // IGM analysis
1-11,23-33 // Fragment 2: Monomers 1 and 3
6 // Evaluate atom and atom pair g indices
y // Export atmdg.pdb in current folder
Then plot ginter isosurface again using the newly generated sl2r.cub and dg_inter.cub via
IGM_inter.vmd, and mean time color the structure according to atom g indices based on atmdg.pdb
file. You will finally obtain below graph.
706
Now the isosurface corresponding to monomer 1-3 interaction is invisible. Since the color
transition set by IGM_inter.vmd is Blue-Green-Red, and I did not manually adjust the automatically
determined color range for mapping atom g indices, therefore, in present graph, the site having
largest atom g index is rendered as red, it should be regarded as "hottest atom" for the interactions
under study. The green or cyan atoms have modest magnitude of atom g index, while contribution
of blue atoms to the interaction is completely negligible.
Finally, let us only highlight interaction between monomer 1 and 2 while completely ignore
monomer 3. Input below command
10 // IGM analysis
12-22 // Fragment 2: Monomers 2
Move the resulting dg_inter.cub and sl2r.cub to VMD folder and use the IGM_inter.vmd script
to draw corresponding color-filled isosurface map. It is better to make monomer 3 transparent, since
currently it is uninteresting. So we enter "Graphics"-"Representation", click the existing
representation with CPK style, input fragment 0 1 in the "Selected Atoms" box and press ENTER
button, now the monomer 3 is invisible. Next, click "Create Rep" button, input fragment 2 in the
"Selected Atoms" box and press ENTER button, then set the "Drawing method" as "Licorice" and
change the "Bond Radius" to 0.2, then set "Material" to "Ghost". Now you should see below graph,
in which only the interaction between monomer 1 and 2 is visible, while all interactions related to
monomer 3 are ignored.
PS: The concept of "fragment" in VMD is different to the "fragment" in IGM analysis of Multiwfn. In VMD,
when structure file is loaded into VMD, the bonding relationship is automatically determined, and then each unlinked
fragment is assigned to a unique fragment index. The index starts from 0.
707
As can be seen in this example, the partition of fragments is highly arbitrary. The union set of
all fragments is not necessarily equal to the whole system. When you intend to study intramolecular
interaction, a whole molecule can also be divided into multiple fragments to reveal interesting
interaction regions.
In this example only a simple system is taken as instance, however I think it is enough to
substantially exhibit the extreme flexibility and powerfulness of IGM analysis, Multiwfn and VMD
programs. The IGM method can also be easily applied to much more complicated systems; for
example, in my blog article http://sobereva.com/407 (in Chinese), I showed that IGM can clearly
reveal the interaction between two monomers in a tetramer consisted of four large flexible molecules.
Please play with the IGM analysis more!
It is worth to mention that g value at bond critical points (BCP) is positively correlated to
interaction strength (see Table 1 of IGM original paper), Multiwfn is also able to calculate it. First,
load a file containing wavefunction information into Multiwfn, then use main function 2 to carry
out topology analysis and locate BCPs, then using option 7 to examine properties of the BCPs, from
screen you can directly read g value. I do not explicitly present a corresponding analysis example
here, please try this kind of analysis yourself.
4.20.11 Using IGM based on Hirshfeld partition of molecular density

(IGMH) to study weak interaction
Please read Section 3.23.6 to gain basic knowledges about the IGMH method. The IGMH
proposed by me is more physically rigorous than the original IGM and its graphical effect is better,
at the expenses of higher computational cost. The use of the IGMH analysis function is exactly
identical to IGM, thus if you have carefully read Section 4.20.10, you will be able to realize IGMH
analysis smoothly. In this section, I use a dimer consisting of 2-pyridoxine and 2-aminopyridine as
an example, this system has also been studied via AIM analysis in Section 4.2.1.
examples\2-pyridoxine_2-aminopyridine.wfn // Since IGMH relies on wavefunction
information, therefore you should use such as .wfn, .fch, .mwfn, .molden, etc. as input file
708

11 // IGMH analysis
1-12 // Atom indices in fragment 1
13-25 // Atom indices in fragment 2
3 // Output cube files to current folder
Next, in order to plot sign(2) colored ginter isosurface map, we move the exported sl2r.cub
and dg_inter.cub from current folder to VMD folder, then copy the examples\IGM_inter.vmd to
VMD folder, then boot up VMD and run source IGM_inter.vmd in VMD console window to execute
the plotting script.
Since the blue around the center of the isosurface corresponding to the N-HN interaction is
darker than that corresponding to the N-HO interaction, it is expected that the H-bond interaction
of N-HN is stronger.
Hint: If you hope to make the radius of the isosurfaces in above map smaller, so that only the most important
interaction areas are shown, you can increase the isovalue from 0.01 to 0.015, namely entering "Graphics" -
"Representation", change the value in the "Isovalue" box to 0.015 and then press ENTER button. In this case, you
can also make the range of color scale narrower than default (-0.05~0.05) to make the color on the isosurfaces more
vivid, namely in the "Representation" interface click "Trajectory" tab, then input -0.045 and 0.045 respectively in
the two boxes and then press ENTER button.
You can also choose option 6 to calculate atom pair g indices, the result calculated with high
quality grid is shown below
12 13 : 0.072611
1 25 : 0.059073
2 13 : 0.032215
...[ignored]
From this quantitative data, we can further confirm the conclusion that the N-H12N13 is stronger
than the N-H25O1.
For comparison purpose, you can plot the same map via IGM module, the sign(2) colored
ginter isosurface map is shown below (the sign(2) was calculated using actual density rather than
promolecular density). As can be seen, the graphical effect of IGM is much poor than IGMH, since
the isosurfaces are too bulgy and thus relatively difficult to examine and compare, this problem
cannot be fully avoided even if you carefully adjust isovalue. Therefore, using IGMH instead of
IGM is always suggested if the system is not too large and thus additional computational cost is
709
affordable.
4.21 Energy decomposition analysis
Aside from the examples given in this Section, Multiwfn is also able to perform the so-called
simple energy decomposition, see Section 4.100.8 of example.
4.21.1 Examples of energy decomposition analysis based on forcefield

(EDA-FF)
In this section I will illustrate how to perform energy decomposition analysis between specific
fragments based on classical forcefield, this method will be referred to as EDA-FF. Please carefully
read Section 3.24.1 first to gain basic knowledges about the underlying ideas and implementations
of EDA-FF. If you have carefully read below examples, you should be able to easily apply this
method onto various kinds of systems.
More in-depth discussions can be found from my blog article "Using Multiwfn to perform
energy decomposition analysis based on forcefield" (in Chinese, http://sobereva.com/442).
4.21.1.1 Example 1: Water dimer

As first example, we perform the EDA-FF based on AMBER forcefield for a very simple
system, water dimer, whose most stable geometry is shown below
Relevant files
The related files have been provided in "examples\EDA\EDA_FF\waterdimer" folder, as
shown below:
• dimer.mol: The .mol file of water dimer containing its optimized geometry at B3LYP-
710
D3(BJ)/6-311G** level, which is very reliable for optimizing molecular clusters. Note that in this
file, the atomic sequence is O1 H2 H3 O4 H5 H6
• water.fchk: The .fchk file produced by optimization task of water monomer at B3LYP-
D3(BJ)/6-311G** level
• mollist.txt: Molecular list file corresponding to the water dimer
• water.txt: Molecular type file of water monomer
As you can see, the content of mollist.txt is simply

water.txt 2
corresponding to the fact that the water dimer has two water molecules, which are described by the
water.txt in current folder.
The content of water.txt is
OW -0.737121
HW 0.368560
HW 0.368560
indicating that the oxygen and hydrogens in the water have atom type of OW and HW, respectively.
The second column are atomic charges evaluated by Merz-Kollman (MK) method.
Detail of molecular type file

Here I describe how the molecular type file water.txt was constructed. The atom types can be
manually assigned according to practical chemical environment of the atoms and definition of atom
types in the forcefield original paper, but this process is troublesome if there are lots of atoms in a
molecule. Therefore, here I show how to use the popular GaussView to automatically assign the
atom types. Load the water.fchk into GaussView, click the icon with bold A letter to enter "atom list
editor", then click the icon with bold orange M letter to show atom types, then click the title of
"AMBER Type" column twice, the current status of the window should be
Then click "File" - "Export Data", save the file as water.txt. Next, via column mode of advanced
text editor such as Ultraedit, delete all columns except the "AMBER Type" column, and then delete
the first row in the file. Now, only atom types of all atom are presented in the water.txt. After that,
calculate MK (or CHELPG) charges based on the water.fchk using subfunction 13 (or 12) of main
function 7 (you can consult the example in Section 4.7.1), then copy the outputted charges from
screen (or from the exported .chg file) as the second column of the water.txt. At this point,
preparation of water.txt is finished.
Perform analysis
Now, we start to perform the EDA-FF analysis. Copy the water.txt to current folder, then boot
up Multiwfn and input
dimer.mol // The file containing dimer structure information (you can also use other formats
711
containing the geometry information as input file, such as the .fch file produced during optimization
task of the dimer)
21 // Energy decomposition analysis
1 // Energy decomposition analysis based on forcefield
3 // Load atom types and atomic charges
mollist.txt // The actual path of the molecular list file. At this point, the program read atom
types and charges from the water.txt and assign them to the two water molecules in the current
system
2 // Define fragments
2 // Two fragments will be defined
1-3 // The atomic indices of the fragment 1
4-6 // The atomic indices of the fragment 2
If you want to check if atom types and charges of all atoms in current system have been set up
properly, you can choose option 4, the output is
*** Fragment 1:
Atom: 1(O ) Charge: -0.737121 Type: OW
Atom: 2(H ) Charge: 0.368560 Type: HW
*** Fragment 2:
Atom: 4(O ) Charge: -0.737121 Type: OW
It is clear that the atom types and charges are all correctly assigned.
Now select option 1 to start the EDA-FF analysis, the result shows up immediately on screen:
Contribution of each atom in defined fragments to overall interfragment interac
tion energies:
Atom 1(O ) Elec: 12.53 Rep: 3.85 Disp: -2.21 Total: 14.17
Atom 2(H ) Elec: -6.24 Rep: 0.00 Disp: 0.00 Total: -6.24
Atom 3(H ) Elec: -16.87 Rep: 0.00 Disp: 0.00 Total: -16.87
Atom 4(O ) Elec: -23.52 Rep: 3.85 Disp: -2.21 Total: -21.88
Atom 5(H ) Elec: 6.47 Rep: 0.00 Disp: 0.00 Total: 6.47
Atom 6(H ) Elec: 6.47 Rep: 0.00 Disp: 0.00 Total: 6.47
Interaction energy components between all fragments:
Electrostatic Repulsive Dispersion Total
Frag 1 -- Frag 2: -21.15 7.71 -4.43 -17.87
The units in the output are all kJ/mol. The above information shows that the total interaction
energy between the two water molecules is -17.87 kJ/mol, which is close to the result -20.58 kJ/mol
obtained by highly accurate level CCSD(T)/CBS (see original paper of the S66 weak interaction
test set, J. Chem. Theory Comput., 7, 2427 (2011)). Although the given result has some error, at
least it is adequate for qualitative discussion purpose. The above data also indicates that electrostatic
interaction (-21.15 kJ/mol) has a decisive contribution to the binding energy between the two waters,
obviously the main essence of general hydrogen bond is dominated by electrostatic interaction.
Dispersion interaction also contributes to the binding, but the magnitude is relatively minor. The
712
exchange-repulsion effect (7.71 kJ/mol), to some extent, offsets the attractive interaction due to
electrostatic and dispersion effects.
In the original text of the S66 test set, the ratio of the dispersion interaction energy to the
electrostatic interaction energy of water dimer given by the very ideal DFT-SAPT method is 0.29,
which is qualitatively consistent with the value yielded by EDA-FF (4.43/21.15=0.21). Therefore,
with the very simple water dimer as instance, it can be seen that as long as the choice of forcefield
and atomic charges are suitable, the result of EDA-FF is generally reliable. For some systems, the
total interaction energy calculated by forcefield is not quite close to that evaluated by reliable
quantum chemistry method, but even so, in general the ratio between various physical components
provided by the EDA-FF is still meaningful. In my view of point, it is not a bad idea to
approximately estimate electrostatic interaction energy (Eele) via multiplying the total interaction
energy (Etot) obtained using quantum chemistry method by the ratio of the Eele and Etot that
evaluated by proper forcefield.
The above output also shows contribution of each atom to the total interaction between all the
defined fragments, so that you can easily recognize which atoms have a critical impact on the
interfragment interaction. The sum of all atomic contributions is equal to the total interaction energy
(if the system only has two atoms A and B, and each one is defined as a fragment, then the
contribution of atom A will be half of the interaction energy between A-B). From the data given
above, it can be seen that influence of each atom is not negligible. After all, the distance between
the atoms in the system is not far. The most important contribution to the attraction is the
electrostatic interaction of the O4 atom (-23.52 kJ/mol), this result is easy to understand since O4 is
the acceptor atom of H-bond. The H3, which directly acts with O4 to form the H-bond, also
contributes greatly to the binding (-16.87 kJ/mol) due to significant electrostatic effect. The data
shows that only oxygen atoms have nonvanishing repulsion and dispersion terms, this is because
the parameters of van der Waals potential of atom type HW is zero, hence HW atoms only behave
as point charges to exhibit electrostatic effect.
Interatomic interaction
If you choose option -3 once to switch its status from the default "No" to "Yes", then during
EDA-FF analysis via option 1, the program also outputs distance (Å), interaction energy (kJ/mol)
and its components of each atom pair to interatm.txt in current folder. The file content of present
example is
******* Between fragment 1 and fragment 2:
Atom_i Atom_j Dist(Ang) Electrostatic Repulsive Dispersion Total
1 4: 2.873 262.78 7.71 -4.43 266.05
1 5: 3.176 -118.86 0.00 0.00 -118.86
1 6: 3.176 -118.86 0.00 0.00 -118.86
2 4: 3.346 -112.79 0.00 0.00 -112.79
2 5: 3.762 50.16 0.00 0.00 50.16
2 6: 3.762 50.16 0.00 0.00 50.16
3 4: 1.916 -197.02 0.00 0.00 -197.02
3 5: 2.312 81.64 0.00 0.00 81.64
3 6: 2.312 81.64 0.00 0.00 81.64
From the above data, we can find that the interaction energy between each pair of atoms is very
large, which mainly comes from electrostatic interaction. For example, since the charges of the two
713
oxygen atoms O1 and O4 are large and have the same sign, the electrostatic mutual exclusion energy
is as high as 262.78 kJ/mol. The binding energy between the fragments appear to be orders of
magnitude far less than the above values, this is because when interaction energy between the
fragments is calculated, the electrostatic interactions of the atom pairs are largely positively and
negatively offset.
4.21.1.2 Example 2: Circumcoronene-Cytosine-Guanine trimer

In the L7 weak interaction test set given in J. Chem. Theory Comput., 9, 3364 (2013), a system
C3GC is a trimer consisted of circumcoronene (hereinafter abbreviated as C3), guanine (G) and
cytosine (C). The geometry has been optimized by the authors at TPSS-D/TZVP level, as shown
below. The GC base pair has formed triple H-bonds, and it is physically adsorbed on the C3 via -
 stacking interaction.
In this section, we will perform EDA-FF analysis on this system based on the AMBER force
field. The relevant files are provided in the "examples\EDA\EDA-FF\C3GC" directory.
Preparation works
Notice that Multiwfn can perform EDA-FF only when atomic indices in any molecule type are
contiguous. Otherwise, the atomic charges and types cannot be set for each atom in the system
through the molecular list file and molecule type files. The structure file given in the supplementary
material of the L7 test set is C3GC.xyz. This file cannot be directly used because the atomic indices
in each monomer is not contiguous. One of the simplest way to judge if the atomic indices are
contiguous is as follows: First load C3GC.xyz into Multiwfn, use subfunction 2 in main function
100 to convert it to C3GC.pdb (we do the conversion because GaussView does not support .xyz
format), then load this pdb file into GaussView, right click on arbitrary atom in a arbitrary molecule
(e.g. atom C5), choose "Select Fragments of Atom C5" (this option is available since GaussView 6).
At this point, all atoms in this molecule are selected as yellow color, then click "Tools" - "Atom
Selection". As can be seen in below screenshot, the indices displayed in the text box are 5-9, 13-17,
19, 25-29, clearly the atom indices are not contiguous and should be rectified.
714
The easiest way of making atom indices contiguous in each molecule is entering "Atom list
editor" of GaussView, then select "Edit" - "Reorder" - "All Atoms (Except the First) by Bonding",
after that the atom indices are reordered according to connectivity, and you will see the atom indices
in each monomer have become contiguous, as shown below. Now save this structure to C3GC.pdb
to replace the old one.
Then we copy each monomer from the trimer to individual GaussView window, save them to
respective .gjf files, change the keyword to "B3LYP/6-311G**" and use Gaussian to run them, then
calculate MK charges by Multiwfn based on the resulting .fch files. Also we make use of GaussView
to determine atom types for each monomer. Finally, combine the atom types and charges as single
file for each monomer, then we have the C.txt, G.txt and C3.txt, which have already been provided
in "examples\EDA\EDA-FF\C3GC" directory.
Finally, create a molecular list file mollist.txt (other name is also acceptable), the content is
actual paths of C.txt, G.txt and C3.txt as well as the number of corresponding molecule, notice that
the order of the file paths must be exactly in line with the molecule order in the geometry provided
in G3GC.pdb. Clearly, the content of mollist.txt should be (all the molecule type files are assumed
to be placed in C:\)
C:\C.txt 1
C:\G.txt 1
715
C:\C3.txt 1
Start analysis
All preparation works have completed, now we start the EDA-FF analysis. Boot up Multiwfn
and input
C3GC.pdb
1 // EDA-FF
3 // Load atom types and charges
mollist.txt // Input actual path of mollist.txt
2 // Define fragments
3 // Three fragments will be defined
1-13 // Atom indices in fragment 1, namely cytosine (C)
14-29 // Atom indices in fragment 2, namely guanine (G)
30-101 // Atom indices in fragment 3, namely C3
Select option 1 to carry out the EDA-FF calculation, the results are as follows (atomic
contribution part is ignored)
Electrostatic Repulsion Dispersion Total
Frag 1 -- Frag 2: -120.98 60.26 -45.54 -106.27
Frag 1 -- Frag 3: 1.88 44.86 -94.69 -47.95
Frag 2 -- Frag 3: 0.71 62.08 -132.62 -69.84
The data shows that the G-C binding is very strong, reached as high as -106.27 kJ/mol, mainly
because the electrostatic component is very large (-120.98 kJ/mol), which is the consequence of the
formation of the three pairs of H-bonds between the G and C. The total interaction energy between
C3 and G (-47.95 kJ/mol) as well as between C3 and C (-69.84 kJ/mol) are not small, mainly due
to the significant - stacking between them. Since the nature of - stacking is purely dispersion
effect, it can be seen that the dispersion interactions of C3-C and C3-G are very strong (-94.69 and
-132.62 kJ/mol, respectively), which are much higher than that between G-C (-45.54 kJ/mol). Since
C3 is essentially a finite graphene sheet, its interaction region with G and C is obviously nonpolar
(i.e. the atomic charges are very small), so the electrostatic component in C3-C and C3-G
interactions is negligible (merely 0.71 and 1.88 kJ/mol, respectively).
I also calculated binding energies between G-C, C3-G and C3-C using dimer models at B3LYP-
D3(BJ)/6-311+G** level, which is very robust for evaluating weak interactions, the results are
G-C (Frag 1 - Frag 2)：-143.97 kJ/mol
C3-C (Frag 1 - Frag 3)：-56.69 kJ/mol
C3-G (Frag 2 - Frag 3)：-76.27 kJ/mol
For C3-C and C3-G, it can be seen that the results calculated by AMBER forcefield and that
by quantum chemistry method are very close, but the quantitative difference for G-C is quite
conspicuous. This observation manifests the limited quantitative accuracy of forcefield when it is
applied to the weak interactions with great strength. However, in this case, if we only focus on the
ratio between various physical components, the result of EDA-FF is still useful and will not cause
evident misleading conclusion.
Coloring atoms according to contribution to interaction energy

By making use of .pqr file format, atomic properties can be easily visualized by coloring atoms
via different colors in VMD visualization program, this strategy is detailedly introduced in Section
716
4.A.10. Obviously, if atomic contributions to interaction energy between fragments are stored into
the .pqr files, then the importance and role of each atom will be able to be vividly shown in a graph.
Similar idea is also employed in the IGM analysis, as illustrated in Section 4.20.10.
Now we select "-4 Toggle if outputting atom contributions to .pqr files" option in the EDA-FF
interface to switch its status to "Yes", then select option 1 to carry out EDA-FF analysis. After the
calculation, atmint_tot.pqr, atmint_ele.pqr, atmint_rep.pqr, atmint_disp.pqr and atmint_vdW.pqr
appear in the current directory (some of them have been provided in "examples\EDA\EDA-
FF\C3GC\pqr"). The data of the atomic charge column in these .pqr files correspond to the
contribution of each atom to the total/electrostatic/repulsion/dispersion/vdW interaction energy
between the fragments, respectively, and the values are identical to those printed on screen. For
example, the value of the atomic charge column of the C1 atom in fragment 1 of the atmint_disp.pqr
corresponds to half of the dispersion interaction energy between C1 atom and all atoms of fragments
2 and 3.
We load the atmint_tot.pqr into VMD program, enter "Graphics" - "Representation", coloring
the atoms according to the "Charge" property, changing the lower and upper limits of the color scale
from the default ones to -50 and 50, respectively. Setting drawing method of the G and C parts as
CPK mode (using fragment 0 1 in the "Selected Atoms" box to choose them), and making C3 part
displayed in Licorice style (using fragment 2 to choose it). We also set the color scale method to
BWR (Blue-White-Red). Finally, the graph will look like below (if you are not familiar with VMD
and do not know how to realize these settings in VMD, please consult Section 4.A.10):
Since the current color scale used is Blue-White-Red, therefore the more blue the atom color
in the figure, the more negative the atomic contribution to the total binding energy between the
trimer (i.e. the more significant the attractive effect); while the more red the atom, the stronger the
repulsive effect it acts. Relatively white atoms only play trivial role on the trimer binding. As can
be seen from this figure, the colors of each H-bond acceptor atom and the hydrogen atom directly
acting with it are obviously blue, thus they contribute greatly to the stability of the G-C binding.
The color of all H-bond donor atoms is red, indicating that their existences are not conducive to the
binding, this is because the H-bond donor and acceptor atoms have large magnitude and same sign
717
of atomic charges, therefore there is a significant electrostatic mutual exclusion between them. From
above map, it can also be seen that all atoms of C3 as well as the atoms in the G and C that far away
from the hydrogen bonding region only have very light or purely white color, this observation does
not imply that they have nearly vanished contributions to the trimer binding, but indicates that their
contributions are relatively weak and thus difficult to be revealed under current color scale setting.
Assume that we want to vividly exhibit the dispersion interaction between C3 and GC base
pair, then we input below commands in the EDA-FF interface
2 // Redefine fragments
1-29 // Fragment 1, the GC base pair
30-101 // Fragment 2, the C3 part
1 // Start the EDA-FF calculation
Then below information is shown on screen, the data equals to the sum of C3-C and C3-G
interaction energies
Frag 1 -- Frag 2: 2.59 106.94 -227.31 -117.79
In the meantime, four new .pqr files are generated in current folder (they have been provided
in "examples\EDA\EDA-FF\C3GC\pqr2" folder). Load the atmint_disp.pqr among them into VMD,
coloring the atoms according to the way described above but using color scale of -10 to 10, then
select "Display" - "Orthographic" to modify the perspective, you will see
In the above figure, the more blue the atomic color, the greater it contributes to the dispersion
interaction between C3 and GC. It can be seen from the figure that each heavy atom in the GC part
contributes nearly equally to the dispersion interaction, mainly because they have almost the same
vertical distance to the C3 plane and the number of electrons carried by these atoms are not very
different. The hydrogen atoms in the GC pair contribute very little to the C3-GC dispersion
interaction, this is because the hydrogen atoms only have very few number of electrons. On the C3
part, the color of the carbons that directly contact with the GC pair is light blue, indicating their
notable contributions to the dispersion interaction. The color of the C3 atoms that far away from the
GC pair is white, reflecting that their influences on dispersion interaction are negligible (recall the
718
fact that dispersion attraction attenuates sharply with distance, it has 1/r6 asymptotic behavior).
Note: The heavy atoms in the GC pair in the above graph are very blue, while the atoms in the equivalent
position of C3 are not so blue, the reason is that: Because there are many atoms in C3, each heavy atom in the GC
pair can form dispersion interaction with a large range of C3 atoms, thus the sum of the terms is large. Since the
number of atoms in the GC pair is small, each atom of C3 can only interact with relatively few number of atoms in
the GC pair, so the sum of terms is not large. If you want to make atomic color of the C3 part more prominent, you
can set the color scale range of the representation corresponding to the C3 part to a smaller value than the -10~10
we previously used; for example, changing to -6 to 6 will yield satisfactory graph.
About determining binding energy of individual H-bond

Some readers may have thought that it would be great if the binding energy of each of the three
H-bonds between the G-C could be independently determined. There is no unique way to achieve
this goal, since this is equivalent to dividing the system into parts and must not be free of artifacts.
An seemingly easy way to realize this purpose is to directly define the donor and acceptor parts of
a H-bond as two fragments. For example, let us examine the H-bond of N10-H13...O14, we input.
2 // Redefine fragments
10,13 // Atomic indices of donor part of N10-H13...O14
14 // Atomic index of acceptor part of N10-H13...O14
1 // Perform EDA-FF analysis
The result is
Frag 1 -- Frag 2: 71.27 21.78 -9.40 83.65
Clearly the result is unreasonable, since the total binding energy was predicted to be a positive
value! The underlying reason is that, the electrostatic interaction is a kind of long-range effect (1/r
asymptotic behavior), therefore consideration of other atoms should not be simply ignored. I also
attempted to employ other ways to evaluate the N10-H13...O14 H-bond binding energy, although
some of them give seemingly acceptable result (for example, summing up atomic contribution of
N10, H12, H13 and O14 to total G-C binding energy), unfortunately the relative strength of the three
H-bonds cannot be faithfully explained. In my opinion, deriving individual H-bond interaction
energy is impossible for this system based on EDA-FF, the reason is that the three H-bonds are too
close together and thus the coupling is very strong, the polarization effect is obvious, and meantime
resonance-assisted effect is involved in these H-bonds, these factors make the total interaction
energy of the three H-bonds very difficult to be reasonably decomposed. However, if there are
several H-bonds and the sites are far away from each other, it should be possible to individually
evaluate the strength of each H-bond by estimation of interaction energy between the atoms in the
corresponding local region.
AFAIK, the best way of determining binding energy of individual H-bond is using atom-in-
molecules (AIM) analysis, see Section 4.2.1 for introduction and illustration.
Other aspects
It is worth to note that a possibly viable way aside from SAPT to evaluate dispersion interaction
energy is employing DFT-D3 dispersion correction using the zero-damping parameters fitted for
exchange-correlation functionals that completely failed to represent dispersion interaction. This
strategy has been utilized in the energy decomposition analysis example shown in Section 4.100.8.
The DFT-D3 dispersion correction for interfragment interaction energy using zero-damping BLYP
functional parameter is shown below
C -- G: -26.65 kJ/mol
719
C -- C3: -87.78 kJ/mol
G -- C3: -115.10 kJ/mol
It can be seen that the three values are close to the corresponding dispersion interaction energies
evaluated by AMBER forcefield (-45.54, -94.69, -132.62, respectively), rendering that using either
AMBER forcefield or DFT-D3 to estimate dispersion interaction energy is a reasonable approach.
Using UFF to conduct the EDA-FF analysis is often unsatisfactory. If we do this for present
system, the result is
Frag 1 -- Frag 2: -120.98 848.28 -80.36 646.94
Frag 1 -- Frag 3: 1.88 52.65 -107.41 -52.87
Frag 2 -- Frag 3: 0.71 68.89 -143.58 -73.99
It can be seen that the interaction predicted by UFF for C3-C and C3-G is normal, and the result
magnitude is close to that calculated by AMBER, but the interaction energy of G-C is extremely
positive, evidently this is completely unreasonable. From the data it is easy to find that the reason
is that the exchange-repulsion component is seriously overestimated. This is not an individual
phenomenon, but a common phenomenon of UFF, this is why I generally do not recommend
performing EDA-FF based on UFF (although optimizing the geometry using UFF prior to the EDA-
FF could alleviate this problem).
As mentioned in Section 3.24.1, for very large systems, using the very cheap EEM charges
(with parameters fitted for CHELPG charges at B3LYP/6-31G* level) instead of the rigorously
derived ESP fitting charges to perform the EDA-FF analysis may be a viable choice; however, my
test showed that this treatment can cause remarkable error in electrostatic interaction for present
system. The electrostatic interaction energy between G and C evaluated based on EEM charges is
merely -74.26 kJ/mol, which is much lower than that evaluated based on MK charges (-120.98
kJ/mol). Therefore, whenever possible, using ESP fitting charges (MK or CHELPG) in EDA-FF
analysis is strongly recommended, this is crucial for yielding electrostatic interaction energy with
satisfactory accuracy.
4.21.2 Shubin Liu's energy decomposition analysis for ethane rotation

barrier
The idea and usage of the Shubin Liu's energy decomposition (EDA-SBL) have been described
in Section 3.24.2. In this section we carry out EDA-SBL method to study the source of energy
different between ethane in optimized eclipsed and staggered conformations. Relevant input and
output files have been given in "examples\EDA\EDA_SBL" folder.
First, we optimize ethane in staggered conformation (D3 point group) and in eclipsed
conformation (D3h point group) at B3LYP/def-TZVP level using Gaussian, the latter in fact is a
transition state. Then using the geometries to create two Gaussian input files, namely
ethane_staggered.gjf and ethane_eclipsed.gjf, respectively. The two files correspond to single point
task at B3LYP/def2-TZVP level. As mentioned in Section 3.24.2, the route section contains
ExtraLinks=L608 keyword and at the end of the input file there is line -5 indicating that the currently
used functional is B3LYP.
720
Run the two .gjf files by Gaussian to yield ethane_staggered.out and ethane_eclipsed.out,
respectively, and then use formchk to convert the resulting .chk files to .fch files.
We first evaluate the energy terms defined by EDA-SBL method for staggered ethane. Boot up
Multiwfn and input
examples\EDA\EDA_SBL\ethane_staggered.fch
2 // Shubin Liu's energy decomposition
examples\EDA\EDA_SBL\ethane_staggered.out
Now Multiwfn loads relevant information from the Gaussian output file, and then evaluates
the steric term defined by EDA-SBL method. Finally, the EDA-SBL energy components are printed:
E_steric: 64.213411 Hartree
E_electrostatic: -146.114859 Hartree
E_quantum: 2.037465 Hartree
E_total: -79.863983 Hartree
The E_total is identical to the single point energy in the Gaussian output file.
We repeat the analysis for eclipsed ethane, then summarize the data into below table
Etotal Esteric Eelectrostatic Equantum
Eclipsed (a.u.) -79.85972 64.21925 -146.10780 2.02883
Staggered (a.u.) -79.86398 64.21341 -146.11486 2.03747
Diff. (kJ/mol) 11.2 15.3 18.5 -22.7
It can be seen that the eclipsed conformation has energy higher than the staggered one by 11.2
kJ/mol, which corresponds to the barrier of C-C single-bond rotation of the ethane. The data implies
that steric effect should be one of the major contributors of the barrier since Esteric is evidently
positive. In addition, the fairly large Eelectrostatic=18.5 kJ/mol suggests that the electrostatic
interaction is the dominating factor to determine the barrier height. In contrast, the variation of
Equantum, which reflects the change in energy purely due to quantum effect, significantly cancels the
steric and classical electrostatic terms and thus play an important role of reducing the barrier.
As you can seen on the screen, the EDA-SBL module also prints other intermediate quantities
comprising the Esteric, Eelectrostatic and Equantum, such as Pauli kinetic energy, so you can use them to
try to analyze the energy difference between the two conformations from more perspectives.
A thorough analysis using the EDA-SBL method for rotation barriers for a series of small
organic molecules is presented in J. Phys. Chem. A, 117, 962 (2013), interested users are suggested
to read it.
4.100 Other functions (Part 1)
4.100.4 Calculate kinetic energy and nuclear attraction potential

energy of phosgene by numerical integration
I suggest you read Section 3.100.4 first, the aim of this example is to show you the usefulness
721
and universality of numerical integration. In quantum chemistry program, kinetic energy and nuclear
attraction potential energy integration are evaluated analytically, analytical method has advantage
in both speed and accuracy. In Multiwfn you can evaluate them by general numerical integration
function. Local kinetic energy (or called “kinetic energy density”) is a built-in function, we first
integrate this function over the whole space to get total kinetic energy. Boot up Multiwfn and input:
examples\COCl2.wfn // HF/6-31G* wavefunction
4 // Integrate a function over the whole space
6 // Hamiltonian kinetic density K(r)
The result is 1031.1092, which is very close to the Gaussian outputted value 1031.1107.
Because I do not want the list of real space function becomes lengthy, so rarely used functions
such as local nuclear attraction potential energy are not chosen as built-in function, however you
can still easily make these functions available by hacking source code, please consult Appendix 2.
First, search “function userfunc” in function.f90, change the default content, namely
"userfunc=1.0D0", to "userfunc = -nucesp(x,y,z)*fdens(x,y,z)", then recompile Multiwfn, redo
above procedure but select function 100 as integrand, you will get nuclear attraction potential energy
-2839.1668, the value outputted by Gaussian is -2839.1629, evidently they are rather close. If you
want to obtain nuclear attraction potential energy contributed from a specific orbital, use subfunction
26 of main function 6 to set occupation number of other orbitals to zero, and then do the integration
as before.
To further illustrate the flexibility of numerical integration function in Multiwfn, assume that
you want to calculate expectation of r2 operator, you can modify the content of “userfunc” routine
to "userfunc = (x*x+y*y+z*z)*fdens(x,y,z)", then recompile Multiwfn and redo the integration, you
will get 444.6523, which is in excellent agreement with the Gaussian outputted value 444.6524
(behind “Electronic spatial extent (au): <R**2>”).
Tip: You may have already noticed that the two lines of the codes colored by blue above have presented in
"userfunc" routine, and they correspond to iuserfunc==12 and iuserfunc==3 respectively. So actually you needn't to
modify and recompile the source code of Multiwfn, by simply changing "iuserfunc" parameter in settings.ini from
the default value 0 to 12 and 3 respectively, the user-defined real space function will be equivalent to the two
functions. For more about the built-in real space functions in "userfunc" routine, see Section 2.7.
4.100.8 Perform simple energy decomposition by using combined

fragment wavefunctions
The principle of the simple energy decomposition has been introduced in Section 3.100.8,
please read it first. Here we try to use this method to analyze components of interaction energy
between NH3 and BH3 in the adduct NH3BH3 at HF/6-31G* level. The Gaussian input and output
files involved in this section can be found in "examples\EDA\EDA_gau" folder. The Gaussian used
here is Gaussian 09 E.01.
(1) Perform structure optimization task for NH3BH3, and then create a new input file with the
optimized coordinate, named NH3BH3.gjf.
(2) Duplicate NH3BH3.gjf as NH3.gjf and BH3.gjf, then remove BH3 fragment in NH3.gjf, and
remove NH3 fragment in BH3.gjf. Add pop=full nosymm keywords in both NH3.gjf and BH3.gjf.
(3) Run the input file of the two fragments by Gaussian respectively to generate NH3.out and
BH3.out.
722
(4) Boot up Multiwfn, input following commands

NH3.out //Fragment 1
8 // Generate Gaussian input file with initial guess combined from fragment wavefunctions
2 // There are two fragments in total
BH3.out //Fragment 2
Multiwfn generates new.gjf in current folder, this is the Gaussian input file for NH3BH3 with
fragment-combined wavefunction as initial guess.
(5) Make sure nosymm keyword is presented in the new.gjf. In order to output SCF energy in
each iteration, change "#" to "#P". pop=full keyword can be deleted.
(6) Run new.gjf by Gaussian to generate new.out.
From NH3.out, BH3.out and new.out, following information can be found
E(NH3) = -56.184295 a.u.
E(BH3) = -26.368199 a.u.
E(NH3BH3) = E_SCF_last = -82.611817 a.u.
E_SCF_1st = -82.535276 a.u.
PS: If you do not know where to find E_SCF_1st, search below information from new.out:
Cycle 1 Pass 1 IDiag 1:
E= -82.5352762791642
DIIS: error= 5.16D-02 at cycle 1 NSaved= 1.
The E_SCF_1st is the SCF energy at the first cycle of SCF process.
According to the equations shown in Section 3.100.8, we can calculate each energy term:
Etot = E(NH3BH3) - E(NH3) - E(BH3)= -0.059323 a.u.= -155.75 kJ/mol

Eorb = E_SCF_last - E_SCF_1st = -0.076541 a.u.= -200.96 kJ/mol
Eels+Eex = Etot - Eorb = 0.017218 a.u.= 45.20 kJ/mol
namely the total interaction energy between NH3 and BH3 is -155.75 kJ/mol, the orbital interaction
energy -200.96 kJ/mol stablized the adduct, while the sum of electrostatic and exchange-repulsion
energy, destablized the adduct by 45.20 kJ/mol.
In present system, compared to other energy components the dispersion interaction can be
safely neglected, so we did not discuss it. However if the system you studied is the complex bounded
by weak interaction, you must evaluate the dispersion component in total interaction energy, see the
discussion in Section 3.100.8 on how to do this.
4.100.12 Perform biorthogonalization analysis for orbitals of

unrestricted open-shell wavefunction
Chinese version of this example and related introduction is "Principle and application of biorthogonalization
method for unrestricted open-shell wavefunctions" http://sobereva.com/448.
The biorthogonalization algorithm between alpha and beta orbitals has been introduced in
Section 3.100.12, please read it first, this transformation makes interpretation of orbitals generated
by UHF or UKS calculation much easier, since only one set of orbitals will then need to be examined.
In this section I will illustrate the value of this analysis, examples\ethanol_triplet.fch is taken as
723
example.
This system has 14 occupied alpha and 12 occupied beta orbitals, first let us look at some of
them:
It can be seen that only alpha orbital 12 pairs well with beta orbital 12, while other alpha and
beta orbitals with the same index do not like with each other. Clearly, it is troublesome when we
discuss orbital characteristics of this wavefunction, because we must simultaneously inspect two
sets of orbitals.
Now we carry out the biorthogonalization for this wavefunction. Boot up Multiwfn and input
examples\ethanol_triplet.fch
12 // Perform biorthogonalization between alpha and beta orbitals
2 // Do biorthogonalization for all orbitals
The biorthogonalization for this wavefunction consists of three successive steps. For example,
the outputted information of the first step is
Doing biorthogonalization for alpha 1 to 14, Beta 1 to 12
Singular values of orbital overlap matrix:
1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 0.9999
0.9999 0.9998 0.9995 0.9992
As shown, this step performs biorthogonalization between alpha orbitals 1~14 and beta orbitals
1~12. The overlap integrals between the resulting alpha and beta orbitals with the same index are
all very close to 1.0, showing that the first 12 alpha orbitals have almost perfectly paired with the
first 12 beta orbitals.
Once all the three steps have been finished, we input n to skip the step of generation of orbital
energies, then biortho.txt and biortho.fch are generated in current folder.
The content of biortho.txt is shown below:
S = Singular value, E = Energy (in eV), O= Occupancy, A=Alpha, B=Beta
Orb: 1 S= 1.0000 O(A)= 1.0 O(B)= 1.0
...[ignored]
Orb: 11 S= 0.9995 O(A)= 1.0 O(B)= 1.0
Orb: 12 S= 0.9992 O(A)= 1.0 O(B)= 1.0
-----------------------------------------------
Orb: 13 S= 1.0000 O(A)= 1.0 O(B)= 0.0
724
Orb: 14 S= 1.0000 O(A)= 1.0 O(B)= 0.0
-----------------------------------------------
Orb: 15 S= 1.0000 O(A)= 0.0 O(B)= 0.0
Orb: 16 S= 1.0000 O(A)= 0.0 O(B)= 0.0
...[ignored]
The information is very easy to understand. Since singular value (S) of all orbitals shown above is
very close to 1.0, therefore the alpha orbitals match well with the beta orbitals having the same
index.
The exported biortho.fch contains wavefunction of the biorthogonalized orbitals, the “orbital
energies” information of these orbitals now correspond to the singular values. Next, if you input y,
this file will be immediately loaded, and then the orbitals in memory will correspond to the
biorthogonalized orbitals.
Here we input y to load the biortho.fch, then enter main function 0 to examine the newly
generated 11~14 alpha and 11~14 beta orbitals, you will see
From the isosurface map it can be seen that the current alpha and beta orbitals match with each
other perfectly. The concept "singly occupied molecular orbitals" (SOMO) was originally defined
for restricted open-shell wavefunction, but now it can also be ideally applied to current
wavefunction, the orbitals 13 and 14 now could be regarded as SOMO. It is well-known that spin
density of unrestricted open-shell wavefunction is determined by all occupied MOs, but after the
biorthogonalization, the spin density will only be contributed by the SOMOs. Obviously, the spin
density of current system directly corresponds to the sum of density of orbitals 13 and 14.
The HOMO and LUMO of singlet ethanol calculated at B3LYP/6-31G** level are shown
below. By comparing this graph and the last graph, one can find that the HOMO and LUMO look
very like the biorthogonalized alpha orbital 14 and 13, respectively. Hence it can be immediately
understood that the transition of ethanol between singlet and triplet states can be well represented
as HOMO→LUMO transition, because after such orbital transition both the HOMO and LUMO
will be occupied by one electron, this electronic structure just corresponds to that represented by the
biorthogonalized orbitals. Clearly, such valuable information cannot be obtained without the
biorthogonalization analysis, rendering importance of biorthogonalization in practical studies of
open-shell systems.
725
It worth to emphasize that the biorthogonalization does not alter any observable properties of
the system, such as total electronic energy, total density, spin density and so on. In addition, during
the biorthogonalization, the program does not update orbital energies (actually the orbital energies
should be changed), therefore after the biorthogonalization the orbital energies will be meaningless,
they are still the energies of the original molecular orbitals, and the order of the biorthogonalized
orbitals does not reflect order of their actual energies.
Biorthogonalization can also be applied to spin polarized singlet systems such as diradicals, an
example is given in my blog article "Principle and application of biorthogonalization method for
unrestricted open-shell wavefunctions" (in Chinese, http://sobereva.com/448).
Evaluating energy of biorthogonalized orbitals and ordering the orbitals

Next, I illustrate how to also make Multiwfn evaluate energies of the biorthogonalized orbitals
and order them according to their energies, the triplet ethanol is still taken as example. Run
examples\ethanol_triplet_47.gjf by Gaussian, then you will get ETHANOL_TRIPLET.47 in C:\
folder (which has been provided as examples\ETHANOL_TRIPLET.47). This file contains Fock/KS
matrix, which is needed during evaluation of orbital energies. Boot up Multiwfn and input
examples\ethanol_triplet.fch // Generated during calculation of ethanol_triplet_47.gjf
12 // Perform biorthogonalization between alpha and beta orbitals
2 // Do biorthogonalization for all orbitals
y // Evaluate energy of the biorthogonalized orbitals
examples\ETHANOL_TRIPLET.47 // Multiwfn will load Fock matrix from it and yield
energies of biorthogonalized orbitals
y // Ordering the biorthogonalized orbitals according to their energies
Now you can find the exported biortho.fch and biortho.txt in current folder. The content of the
later one is:
...[ignored]
Orb: 11 S= 0.9992 E(A)= -13.057 O(A)= 1.0 E(B)= -11.821 O(B)= 1.0
Orb: 12 S= 1.0000 E(A)= -11.918 O(A)= 1.0 E(B)= -11.866 O(B)= 1.0
-------------------------------------------------------------------------------
Orb: 13 S= 1.0000 E(A)= -11.439 O(A)= 1.0 E(B)= -5.806 O(B)= 0.0
Orb: 14 S= 1.0000 E(A)= -0.293 O(A)= 1.0 E(B)= 2.817 O(B)= 0.0
-------------------------------------------------------------------------------
Orb: 15 S= 0.9992 E(A)= 13.876 O(A)= 0.0 E(B)= 15.431 O(B)= 0.0
Orb: 16 S= 1.0000 E(A)= 15.057 O(A)= 0.0 E(B)= 15.305 O(B)= 0.0
...[ignored]
Because in present example we have evaluated orbital energies, therefore the energies are also
explicitly shown in the biortho.txt. It can be seen that the basically paired alpha and beta
726
biorthogonalized orbitals have detectably different energies, for example E(alpha 11)= -13.057eV
while E(beta 11)= -11.821 eV, this is because alpha and beta orbitals feel different effective
potentials (i.e. alpha and beta Fock opterators are different). In addition, one can find that the order
of orbital index has in line with order of orbital energy, because we have requested Multiwfn to
order the orbitals according to their energies (more specifically, the average energy of alpha and
beta orbitals with the same index). Note that the orbital energy information in the biortho.fch now
also corresponds to the actual energy of the biorthogonalized orbitals.
Then, you can input y to load the just generated biortho.fch so that we can then directly use
main function 0 to visualize the biorthogonalized orbitals. The two occupied alpha biorthogonalized
orbitals with highest indices, namely the two orbitals formally occupied by the two unpaired
electrons in present system, are shown below.
Their shapes are very close to the alpha MO 13 (-10.083 eV) and alpha MO 14 (-0.276 eV), and
meantime the energies are not quite different.
Note that as mentioned in Section 3.100.12, the orbitals in each batch are ordered individually,
therefore the relative order of orbitals in different batch will not alter due to the ordering. In addition,
the ordering is based on average energy of each alpha orbital and its beta counterpart, thus alpha-
beta orbital correspondence is kept during the ordering process.
4.100.13 Calculate HOMA and Bird aromaticity index for

phenanthrene
HOMA is the most prevalently used aromaticity index based on geometry equalization, see
Section 3.100.13 for detail. Here we use HOMA to study which ring of phenanthrene has stronger
aromaticity.
Since calculation of HOMA only requires molecular coordinate, you can simply use such
as .pdb and .xyz as input file. Of course, other files containing molecular coordinate, such as .wfn
and .fch files are acceptable too. The geometry in present instance is optimized under B3LYP/6-
31G* level.
727
examples/phenanthrene.pdb
13 // Calculate HOMA and Bird aromaticity index
0 // Start the calculation
You will see the default parameters are printed, they are taken from J. Chem. Inf. Comput. Sci.,
33, 70, you can also change these parameters by yourself via option 1 before the calculation.
Now we input the atom indices in the ring that we are interested in, we calculate HOMA for
the central ring first, so input 3,4,8,9,10,7, the input order must be consistent with atom connectivity.
You will immediately obtain the result shown below
Atom pair Contribution Bond length(Angstrom)
3(C ) -- 4(C ): -0.065698 1.427111
4(C ) -- 8(C ): -0.210455 1.458000
8(C ) -- 9(C ): -0.065698 1.427111
9(C ) -- 10(C ): -0.096016 1.435282
10(C ) -- 7(C ): -0.033673 1.360000
7(C ) -- 3(C ): -0.096016 1.435282
HOMA value is 0.432442
Obviously, C4-C8 deviates to ideal bond length 1.388 most significantly, giving rise to large
negative contribution to HOMA, in other words, significantly breaked aromaticity. The HOMA
value is calculated as 1 plus the contributions from all bonds in the ring.
Then input 8,15,14,13,11,9 to calculate HOMA for the boundary ring, the result is 0.855126.
Since this value is much closer to 1 than the one for central ring, HOMA suggests that the two
boundary rings have stronger aromaticity.
Bird index is another quantity used to measure degree of aromaticity, see Section 3.100.13 for
a brief description. Now choose option 2 to calculate Bird index for the two rings, you will find the
value for boundary ring is more close to 100 than the central ring. Likewise HOMA, Bird index also
indicates that the two boundary rings have stronger aromaticity.
4.100.14 Calculate LOLIPOP for phenanthrene
In this instance, we use LOLIPOP (Localized Orbital Locator Integrated Pi Over Plane) index
to evaluate which ring of phenanthrene has stronger π-stacking ability. For detail about LOLIPOP,
see Section 3.100.14 and the paper Chem. Commun., 48, 9239 (2012).

examples/phenanthrene.wfn
14 // Calculate LOLIPOP
728
1 // Specify π orbitals
36,40,43,44,45,46,47 //These orbitals are π orbitals. You can find out π orbitals by visualizing
orbital isosurfaces via main function 0 or by function 22 of main function 100
0 // Start the calculation
8,9,11,13,14,15 // The indices of the atoms in the ring you are interested in. This ring is
boundary ring
Wait for a while, from screen we can see that the LOLIPOP value is 8.233.
0 // Start the calculation again
7,3,4,8,9,10 // Central ring
The LOLIPOP value is 6.183
Since smaller LOLIPOP value corresponds to stronger π-stacking ability, we can expect that
the tendency of forming π-stacking structure for central ring is stronger than for boundary ring.
4.100.15 Calculate intermolecular orbital overlap integral of DB-TTF
Intermolecular orbital overlap integral is important in discussion of intermolecular charge

transfer, see Section 3.100.15 for introduction. In this example, we calculate HOMO-HOMO and
LUMO-LUMO overlap integrals between the two DB-TTF (Dibenzotetrathiafulvalene) monomers
in below dimer. The dimer structure was extracted from CSD (Cambridge Structural Database). The
wavefunction level we used is B3LYP/6-31G*, MO78 and MO79 correspond to HOMO and LUMO,
respectively.
As mentioned in Section 3.100.15, we must prepare wavefunction files containing basis

function information for the dimer and two monomers. For example, you can use Gaussian to
generate corresponding .fchk files, the Gaussian input files for this aim have been provided in
"examples\intermol" folder; if you do not want to run them by Gaussian by yourself, you can directly
download the resulting .fchk files from http://sobereva.com/multiwfn/extrafiles/intermol.rar.
It is worth to explain the three .gjf files. The coordinates in the two monomer input files DB-
TTF1.gjf and DB-TTF2.gjf were directly extracted from the dimer input file DB-TTFdimer.gjf, the
atoms of monomer 1 occur in this file prior to those of monomer 2. The keyword nosymm is used
in all the three files, it requests Gaussian do not automatically put the coordinates in the input files
to standard orientation, so that the coordinates in the monomer .fchk files are exactly identical to
those in the dimer .fchk file. Because the dimer .fchk file is only used by Multiwfn for generating
overlap matrix between basis functions while MOs of dimer are never be utilized, therefore
729
guess(save,only) is employed in the dimer input file to let Gaussian only generate and save initial
guessing orbitals, which do not need any cost. The pop=none is used together, it suppresses the
useless massive output of orbital expansion coefficients in output file.
Now, boot up Multiwfn and input below commands:
DB-TTFdimer.fchk // Wavefunction file of dimer
15 // Function for calculating intermolecular orbital overlap integral
DB-TTF1.fchk // Wavefunction file of monomer 1
DB-TTF2.fchk // Wavefunction file of monomer 2
After that, if we input i,j, then the intermolecular orbital overlap integral between MO i in
monomer 1 and MO j in monomer 2 will be printed. To obtain the integral between HOMO-HOMO,
we input 78,78, the result is -0.01411983; Then input 79,79, we will find the integral between
LUMO-LUMO is 0.01025897.
4.100.16 Example of easily calculating various quantities defined in the

conceptual density functional theory (CDFT)
4.100.16.1 Automatically calculate conceptual density functional theory

quantities for phenol
In this section, I will show how to very conveniently calculate almost all quantities defined in
the framework of conceptual density functional theory (CDFT). Phenol will be taken as example.
Since phenol is a neutral molecule, the N, N+1 and N-1 electrons states correspond to neutral,
anionic and cationic states, respectively.
Before following this example, please read Section 3.100.16 first, in which all quantities to be
computed and basic usage are described.
Preparation of needed wavefunction files

Phenol at neutral state should be optimized first, please do it by yourself. For present study, the
computation level for optimization is not necessarily high, even semi-empirical method such as
PM7 is even adequate. The examples\phenol.xyz contains geometry of phenol optimized at the
commonly used B3LYP/6-31G* level.
examples\phenol.xyz
16 // Calculate various quantities in conceptual density functional theory
1 // Generate .wfn files for N, N+1 and N-1 electronic states
[Press ENTER button directly] // The generated .gjf file will correspond to single point at
B3LYP/6-31G* level
[Press ENTER button directly] // Use default charge and spin multiplicity, namely 0 1 for N
state, -1 2 for N+1 state, and 1 2 for N-1 state
Now N.gjf, N+1.gjf and N-1.gjf have been generated in current folder, they are input files of
single point task used for generating N.wfn, N+1.wfn and N-1.wfn, respectively. Now you can
730
manually run them by Gaussian. If you do not change the generation path of the .wfn file in these
files, after running them they will be generated in current folder for Gaussian of Linux version,
while for Windows version they will be generated in Gaussian scratch folder.
You can also let Multiwfn directly invoke Gaussian to run the .gjf files. Assume that we have
set "gaupath" in settings.ini to actual path of Gaussian executable file, now you can input y in
Multiwfn window, then the .gjf files will be executed by Gaussian and the resulting .wfn files will
occur in current folder, after that the .gjf and .out files will be automatically deleted.
Multiwfn is also able to generate input files of ORCA program for generating the .wfn files. To do so, you
should select option -2 once, then choose option 1, then three ORCA input files will be generated. After manually
running them by ORCA, N.wfn, N-1.wfn, N+1.wfn will be available, and you should move them to Multiwfn folder.
Calculating global and atomic indices

Because N.wfn, N+1.wfn and N-1.wfn have already been provided in current folder, now we
can start calculation of CDFT quantities. We choose "2 Calculate various quantitative indices", then
Multiwfn starts to calculate Hirshfeld charges and extract information from the .wfn files, the results
are outputted to CDFT.txt in current folder. The content is shown below, they are completely self-
explanatory. The file has also been provided as examples\CDFT.txt.
Hirshfeld charges, condensed Fukui functions and condensed dual descriptors
Units used below are "e" (elementary charge)
Atom q(N) q(N+1) q(N-1) f- f+ f0 CDD
1(C ) -0.0587 -0.1185 0.0852 0.1439 0.0598 0.1018 -0.0841
2(C ) -0.0390 -0.1674 0.0268 0.0658 0.1284 0.0971 0.0626
3(C ) -0.0597 -0.1873 0.0319 0.0916 0.1276 0.1096 0.0360
...[ignored]
Condensed local electrophilicity/nucleophilicity index (e*eV)
Atom Electrophilicity Nucleophilicity
1(C ) 0.02576 0.45535
2(C ) 0.05533 0.20827
3(C ) 0.05496 0.28982
...[ignored]
Condensed local softnesses (Hartree*e) and relative electrophilicity/nucleophilicity
(dimensionless)
Atom s- s+ s0 s+/s- s-/s+
1(C ) 0.3761 0.1562 0.2661 0.4154 2.4076
2(C ) 0.1720 0.3356 0.2538 1.9510 0.5126
3(C ) 0.2393 0.3333 0.2863 1.3926 0.7181
...[ignored]
E(N): -307.464860 Hartree
E(N+1): -307.383614 Hartree
E(N-1): -307.163438 Hartree
E_HOMO(N): -0.218913 Hartree, -5.9569 eV
E_HOMO(N+1): 0.161297 Hartree, 4.3891 eV
E_HOMO(N-1): -0.464864 Hartree, -12.6496 eV
731
Vertical IP: 0.301421 Hartree, 8.2021 eV
Vertical EA: -0.081246 Hartree, -2.2108 eV
Mulliken electronegativity: 0.110088 Hartree, 2.9956 eV
Chemical potential: -0.110088 Hartree, -2.9956 eV
Hardness (=fundamental gap): 0.382667 Hartree, 10.4129 eV
Softness: 2.613235 Hartree^-1, 0.0960 eV^-1
Electrophilicity index: 0.015835 Hartree, 0.4309 eV
Nucleophilicity index: 0.116285 Hartree, 3.1643 eV
You can compare the condensed Fukui functions and dual descriptor shown above with those
manually calculated in Section 4.7.3, you can find the data are completely identical. Clearly, using
present module to compute the CDFT quantities is by far easier than manual calculation!
Calculating Fukui function and dual descriptor

Next, we study Fukui function (f) and dual descriptor (f), which are real space functions.
Choose option "3 Calculate grid data of Fukui function and dual descriptor", and then select "2
medium quality grid", then Multiwfn automatically calculates grid data of electron density for N,
N+1 and N-1 states. After that, you can choose corresponding option to visualize Fukui function or
dual descriptor, or export them as cube files in current folder. Various types of f and f plotted by
Multiwfn are collectively given in the following graph, all isovalues are set to 0.007:
From above map, you can find the f and f automatically generated by present module are
identical to those manually yielded in Section 4.5.3 by means of custom operation feature of main
function 5. Undoubtedly using present module is much more convenient!
As mentioned in Section 4.5.4, the dual descriptor can also be approximately evaluated based on spin density
of N-1 and N+1 states, however in present module, the dual descriptor as well as its condensed form are evaluated
in exact way.
It is worth to note that if we export the grid data as cube file by corresponding option, then we
can use the method described in Section 4.A.14 to very quickly and easily plot above functions as
isosurface map by VMD at state-of-the-art quality. For example, below is the f − function rendered
732
by VMD.
Present module can also be used to visualize local softness, local electrophilicity index and
local nucleophilicity index. For example, we will export cube file of local electrophilicity index,
recall that its definition is simply multiplying f + by global electrophilicity index. Therefore, we
select option "-1 Set the value multiplied to all grid data", input 0.4309 (which is the global
electrophilicity index in eV outputted by option 2, see CDFT.txt), then select option "5 Export grid
data of f+ as f+.cub in current folder". Now, the exported f+.cub corresponds to local electrophilicity
index, the unit is eV.
On the calculation of cubic

As described in Section 3.100.16, the electrophilicity index cubic is useful in studying weak
interaction, at least for halogen bonds. If you also want to calculate it, after enter the present module
you should select option -1 first to switch its status to "Yes", then simply follow the example shown
above (i.e. prepare .wfn files with the aid of option 1, then use option 2 to perform calculation), then
the CDFT.txt outputted by option 2 will contain cubic as well as its condensed value on every atom.
An intermediate quantity, namely the second vertical ionization potential, is also printed.
More specifically, for the present example, after entering the present module you should input
-1 // Toggle calculating cubic
1 // Generate .wfn files for N, N+1, N-1, N-2 electron states
[Press ENTER button directly] // Use B3LYP/6-31G* level
[Press ENTER button directly] // Use default charge and spin multiplicity, namely 0 1 for N
state, -1 2 for N+1 state, 1 2 for N-1 state, 2 1 for N-2 state
Then run the generated four .gjf files to generate N.wfn, N+1.wfn, N-1.wfn and N-2.wfn, then
choose option 2.
4.100.16.2 Illustration of studying orbital-weighted Fukui function and

orbital-weighted dual descriptor
Note: Chinese version of this section with extended discussions and more examples is http://sobereva.com/533.
Please read Section 3.200.16.3 first, in which orbital-weighted Fukui function (𝑓𝑤+ and 𝑓𝑤− )
and dual descriptor ∆𝑓𝑤 are introduced, also some important notes are given. In this section we
will use these functions to study several systems.
733
Part 1: C60
In this part we will use the orbital-weighted functions to reveal reactive sites of C60, which has
high point group symmetry and its frontier molecular orbitals are highly degenerate. Molecules like
this are unable to be reasonably studied via the Fukui function and dual descriptor in standard form.
The .fch file of this system generated at B3LYP/6-31G* level can be downloaded at
http://sobereva.com/multiwfn/extrafiles/C60.zip, which is input file of the present analysis.
C60.fch
16 // Calculate various quantities in conceptual density functional theory
In the current menu, you can use option 4 to set the  parameter used in the subsequent orbital-
weighted calculations, in this example we keep the default value (0.1 Hartree) unchanged, it should
be properly changed only when you find the result is not satisfactory.
We first visualize isosurface of the orbital-weighted functions. Input below commands
7 // Calculate grid data of OW Fukui function and OW dual descriptor
2 // Medium quality
Then you can use corresponding option to visualize isosurfaces of 𝑓𝑤+ , 𝑓𝑤− , 𝑓𝑤0 and ∆𝑓𝑤 ,
which are collectively shown below. Note that isovalue should be changed to proper value,
otherwise isosurfaces may be even invisible. Isosurface of 0.0003 a.u. is used to plot below maps.
In above map, green and blue isosurfaces represent positive and negative parts, respectively.
As you can see, distribution of all orbital-weighted functions are in line with molecular point group
symmetry, this is a unique advantage of the orbital-weighted form. In contrast, if you plot density
of HOMO (corresponding to frozen orbital form of f −) or calculate and plot f − via finite difference
(N − N-1), you will find their distributions are counterintuitive (inconsistent with molecular
symmetry) and thus useless in revealing reactive sites. As introduced in Section 4.5.4, a region with
large positive f − or with prominent negative f corresponds to the site having remarkable
nucleophilicity, or equivalently, vulnerable to electrophilic attack. From above map we can find that
the sites with highest tendency of undergoing electrophilic attack is the region above [6,6] type of
C-C bond (i.e. the bond shared by two adjacent six-membered rings). This observation is fully in
line with experimental finding (see original paper of ∆𝑓𝑤 , namely J. Phys. Chem. A, 123, 10556
(2019), for extensive discussion), and this conclusion can also be further confirmed according to
distribution of minima of average local ionization energy over vdW surface, see Section 4.12.2 on
how to perform this kind of analysis.
734
The orbital-weighted form of Fukui function and dual descriptor are contributed by multiple
orbitals. If you want to check weights to better understand how the orbital-weighted functions work,
in the present module you can choose "5 Print current orbital weights used in orbital-weighted (OW)
calculation", then you will see
10 Highest weights in orbital-weighted f+
Orbital 181 (LUMO ) Weight: 12.47 % E_diff: 1.752 eV
Orbital 182 (LUMO+1) Weight: 12.47 % E_diff: 1.752 eV
10 Highest weights in orbital-weighted f-
Orbital 180 (HOMO ) Weight: 9.06 % E_diff: -1.752 eV
Orbital 179 (HOMO-1) Weight: 9.06 % E_diff: -1.752 eV
The "E_diff" is the difference between orbital energy and the chemical potential approximately
evaluated as average of E(HOMO) and E(LUMO). Evidently, the more the orbital energy close to
the chemical potential, the higher weight the orbital will have.
You can also calculate condensed 𝑓𝑤+ , 𝑓𝑤− , 𝑓𝑤0 and ∆𝑓𝑤 , so that you can easily study their
values at each atomic site quantitatively. To do so, in the present module you should choose "6
Calculate condensed OW Fukui function and OW dual descriptor". However, these condensed
quantities are meaningless for the C60 we studied above, since all atoms are spatially equivalent.
It is worth to note in passing that extrema of 𝑓𝑤+ , 𝑓𝑤− , 𝑓𝑤0 and ∆𝑓𝑤 on molecular surface can
be exactly located via quantitative molecular surface analysis module, so that one can quantitatively
compare their values in different regions. Many examples of using this module has been given in
Section 4.12. Below we will examine extrema of ∆𝑓𝑤 on  = 0.01 a.u. isosurface. First we set
"iuserfunc" parameter in settings.ini to 98, since as mentioned in Section 2.7, ∆𝑓𝑤 corresponds to
the 98th user-defined function. Then boot up Multiwfn and input
C60.fch
735
0.01 // Use  = 0.01 a.u. isosurface to define the surface

-1 // User-defined real space function, which now corresponds to ∆𝑓𝑤
3 // Spacing of grid points for generating molecular surface
0.25 // Use slightly larger grid spacing than default to reduce cost. This setting is already fine
enough for present investigation
0 // Start analysis
After the calculation is complete, select option 0 in post-processing menu to visualize extrema.
In order to make the graph clear, we change the "Ratio of atomic size" in GUI to 3.0 to enlarge
atomic spheres. The graph in the GUI window is shown below. Value of a few minima (blue spheres)
of ∆𝑓𝑤 are labelled, the values can be found from text window. The red spheres correspond to
maxima.
As can be seen, the ∆𝑓𝑤 above [6,6] bond is remarkably more negative compared to other areas,
thus these positions have highest reactivity for electrophilic reaction. Although there are also surface
minima above center of each five-membered ring, the value is slightly positive, therefore the five-
membered ring does not have evident tendency to participate in electrophilic reaction.
Part 2: Cyclo[18]carbon
The cyclo[18]carbon was very systematically studied in my work Carbon, 165, 468 (2020),
Carbon, 165, 461 (2020) and ChemRxiv (2019) DOI: 10.26434/chemrxiv.11320130. This system
has high symmetry (D9h) and thus very suitable to be studied via the orbital-weighted functions.
The .fchk of this system can be directly downloaded via
http://sobereva.com/multiwfn/extrafiles/C18.zip, which was generated at B97XD/def2-TZVP
level. The isosurface of ∆𝑓𝑤 =0.0008 a.u. of this system is shown below.
736
The cyclo[18]carbon contains two kinds of C-C bond, a short one and a long one, they occur
alternatively. Two short bonds are highlighted by red arrows. From above map it can be clearly seen
that the short and long C-C bonds are vulnerable to electrophilic and nucleophilic attacks,
respectively, since the former are enclosed by negative isosurfaces while the latter are enclosed by
positive isosurfaces.
In Multiwfn, it is also possible to plot ∆𝑓𝑤 as plane map. As an example, we will plot ∆𝑓𝑤 as
color-filled map on the molecular plane of the cyclo[18]carbon. We first set "iuserfunc" parameter
in settings.ini to 98, then boot up Multiwfn and input
C18.fchk
4 // Plot plane map
100 // User-defined real space function, which now corresponds to ∆𝑓𝑤
1 // XY plane
0 // Z=0
We close the map that pops up, then adjust some settings in the post-processing menu and
replot, after that you will see below map. The blue contour line corresponds to vdW surface.
737
From above map we can find that the inner edge of the ring is more reactive than the outer edge,
since inner edge has larger magnitude of ∆𝑓𝑤 .
Part 3: CH3Cl
Finally, we employ ∆𝑓𝑤 to study reactivity of CH3Cl. The input file is examples\CH3Cl.fchk.
We calculate grid data of ∆𝑓𝑤 using the same way as above examples, however, in order to get
better graphical effect, this time we do not visualize isosurface in Multiwfn directly, but export grid
data of ∆𝑓𝑤 as OW_DD.cub via corresponding option, then use the method described in Section
4.A.14 to easily render it as isosurface map via VMD program. The isosurface of 0.008 a.u. is shown
below.
As can be seen, the toroidal negative region appears around the Cl atom, implying that this
region shows Lewis-base character. At the two ends of C-Cl bond the ∆𝑓𝑤 is notably positive,
showing that the end of carbon site is vulnerable to nucleophilic attack (e.g. SN2 reaction), while
the end of Cl site behaves as Lewis-acid, which is in line with the fact that there is a -hole region.
By the way, you can also study extrema of ∆𝑓𝑤 on molecular surface like the example of C60
to make discussion of ∆𝑓𝑤 at quantitative level.
738
4.100.18 Yoshizawa's electron transmission route analysis for

phenanthrene
In this example, we will use Yoshizawa's formula (Acc. Chem. Res., 45, 1612 (2012)) to analyze
favourable electron transimission routes for phenanthrene. Related theory, requirement of input file
and program options have been introduced in Section 3.100.18. The numbering scheme of the
carbons is shown below.

examples\phenanthrene_NAOMO.out //The Gaussian output file containing "NAOMO"
information
18 // Yoshizawa's electron transport route analysis
2 // Select YZ plane, which is the molecular plane
Then program will detect which atom has expected pz atomic orbitals, and load their expansion
coefficients in all MOs.
Now we select 1, and input 2,11 to check the the transmission probability between 2 and 11.
From the output we can know that the transmission probability is 0.855879. The contributions from
each MO are also shown. From the output we also know that the probability will be 2.144432 if
only HOMO and LUMO are considered. Although Yoshizawa's paper said that in common one only
need to taken HOMO and LUMO into account, it seems that this approximation is not true in
quantitative level. The distance route 5.674656 Å is the distance between atom 2 and 11.
Next, we examine which transport routes are the most favourable. Suppose that this time we
only want to consider HOMO and LUMO, so we choose option -1 and input 47,48. Then choose
option 2, you will see
Note: The routes whose absolute value < 0.010000 will not be shown
Note: The routes whose distance < 0.0000 or > 9999.0000 Angstrom will not be shown
Atom 7 -- Atom 10 Value and distance: 3.743670 1.359562
Atom 7 -- Atom 11 Value and distance: -2.831680 3.720990
...
739
The routes are ranked by transmission probability (absolute value). If we ignore the atoms
linking three carbons and the duplicated routes due to symmetry, the most favourable four routes
are 7-10, 6-10, 7-11, 6-14. This conclusion is completely in line with below graph in Yoshizawa's
paper!
Note that Yoshizawa stated that connection 9-10 (corresponding to 7-10 in our numbering
scheme) is predicted to be the best route from the rule, but it is too close to construct a metal-
molecule-metal junction for connection, so the route was not marked on the his graph.
Now we check the feasible routes started from atom 2. Choose option 3, and input 2, you will
see
To atom 10 Value and distance (Angstrom): -2.831680 3.720990
To atom 11 Value and distance (Angstrom): 2.144432 5.674656
...
Evidently, 2-10 is the most favourable route.
4.100.19 ELF analysis on the promolecular wavefunction combined

from fragment wavefunctions
Before reading this section please read Section 3.100.19 first to gain basic knowledge.
It is usually interesting to analyze the characteristic of promolecular wavefunction, which
corresponds to the state without any electron transfer and polarization due to the interaction between
the fragments constituting the system. Commonly, we can use "custom operation" in main function
3, 4 and 5 to realize this purpose, see Section 3.7 for introduction of custom operation and some
illustrative applications in Sections 4.5.4 and 4.5.5. For example, we want to study electron density
distribution of promolecular state of a complex AB, by using custom operation, we can very
conveniently instruct Multiwfn to calculate electron density of A and that of B respectively, and then
sum them up as electron density of the promolecule state. However, this process is not applicable to
non-linear real space functions such as ELF; that is to say, ELF of promolecular state of AB is not
equal to the sum of ELF of A and ELF of B, the result is completely meaningless. For such cases,
we should combine fragment wavefunctions first as promolecular wavefunction, and then calculate
ELF for it to obtain the ELF distribution in promolecular state.
Below I will show how to use Multiwfn to produce promolecular wavefunction for COBH3
based on fragment wavefunctions of CO and BH3, and then discuss the corresponding ELF character.
The .wfn files used below and the corresponding Gaussian .gjf files can be found in
"examples\genpromol" folder.
Boot up Multiwfn and then input
740
examples\genpromol\COBH3\CO.wfn // The path of wavefunction file of fragment 1

19 // Generate promolecular .wfn file from fragment wavefunctions
2 // Two fragments in total
examples\genpromol\COBH3\BH3.wfn // The path of wavefunction file of fragment 2
Now the promolecular wavefunction file of COBH3 has been outputted to promol.wfn in
current folder.
Let us plot ELF for this promolecular wavefunction. Reboot Multiwfn and input
promol.wfn
4 // Draw plane map
9 // ELF
2 // XZ plane
0 // Y=0
Interestingly, even in the promolecular state, from the resultant graph it looks as if the carbon
and boron have been bonded to each other. In order to make clear how the relaxation of electron
distribution affects the ELF character of COBH3, we decide to draw difference map of ELF between
the actual state and promolecular state.
examples\genpromol\COBH3\COBH3.wfn // Wavefunction file of actual state of COBH3
4 // Draw plane map
1 // Deal with only one file
-,promol.wfn // Subtracting property of COBH3.wfn by that of promol.wfn
9 // ELF
2 // XZ plane
0 // Y=0
1
-0.2,0.4 // Set the color scale from -0.2 to 0.4, since as you can see from the command-line
window, in this plane the data range is from -0.248 to 0.436
741
From this difference map of ELF, it is very clear that after electron relaxation, the electron
localization character between the C-B bond enhanced evidently.
Next I will show how to create promolecular wavefunction when open-shell fragments are
involved. CH3NH2 is taken as example, the two fragments are CH3 and NH2 free-radicals. Boot up
Multiwfn and input
examples\genpromol\CH3NH2\CH3.wfn
19 // Generate promolecular .wfn file from fragment wavefunctions
2 // There are totally two fragments (including the loaded one)
examples\genpromol\CH3NH2\NH2.wfn // Wavefunction file of the second fragment
n // Do not flip spin of orbitals of CH3
y // Flip spin of orbitals of NH2. If you do not understand why the spin should be flipped, you
can consult the corresponding CDA example in Section 4.16.2.
Now you have promolecular wavefunction of CH3NH2 in current folder. Try to plot ELF for it
and compare the result with actual state (i.e. examples\genpromol\CH3NH2\CH3NH2.wfn)
4.100.21 Calculate molecular diameter and length/width/height
As described in Section 3.100.21, subfunction of 21 of main function 100 can be used to

calculate a variety of that solely based on molecular geometry. This function can also calculate
molecular diameter and length/width/height, in this section I will use examples/alpha-
742
cyclodextrin.pdb an instance to illustrate this point. Please first read Section 3.100.21 to understand
how this function work.
examples/alpha-cyclodextrin.pdb
21 // Calculate quantities that purely based on geometry
size // Calculate molecular diameter and length/width/height
You will see below information on screen
Farthest distance: 44(H ) --- 123(H ): 14.027 Angstrom
vdW radius of 44(H ): 1.200 Angstrom
vdW radius of 123(H ): 1.200 Angstrom
Diameter of the system: 16.427 Angstrom
Radius of the system: 8.213 Angstrom
Length of the three sides: 15.341 14.714 9.511 Angstrom
The diameter 16.427 Å printed on screen is calculated as 14.027+21.2. The radius 8.213 Å is
simply the half of the diameter. The length/width/height of the molecule, namely the three values
after "Length of the three sides" are calculated as follows: Multiwfn first automatically rotates the
molecule so that its three principal axes just parallel to the three Cartesian axes, and meantime
translates the molecule to put its geometry center to original point. Then according to the position
of boundary atoms (i.e. the atoms having maximum/minimum value of X/Y/Z coordinate) and
atomic Bondi van der Waals radii, the length/width/height of the molecule can be derived
straightfoward.
If you want to visually depict the length/width/height, you can then choose option 1, you will
see below graph (the "Ratio of atomic size" has been set to 4.0, in this situation radius of atomic
sphere just equals to atomic van der Waals radii). As you can see, the blue box tightly encloses the
molecular van der Waals surface, the length of its three sides are simply the length/width/height of
the molecule, i.e. 15.341, 14.714 and 9.511 Å, respectively.
743
You can also select option 2 to export the molecule in rotated and translated coordinate to
new.pdb in current folder. This file contains "CRYST1" field, which records cell lengths. You can
directly load this file into VMD (http://www.ks.uiuc.edu/Research/vmd/) visualization program, and
then input pbc box command in VMD console window to show box. After slight adjustment of
plotting effect and manually editing the graph, you will obtain below graph, which clearly illustrates
the molecular geometry character.
It is also possible to visualize molecular principal axes in VMD. Simply copy all content of
examples\principal_axes.tcl into VMD console window, you will see below graph, in which the red,
blue and green axes respectively represent the three molecular principal axes.
More discussions about this topic are given in my blog article "Using Multiwfn to calculate the
length, width and height of molecules" (in Chinese, http://sobereva.com/426).
4.100.22 Analyze π electron character of non-planar system:

cycloheptatriene
More application examples of this module in studying  electron structure is given in my this paper: Theor.
Chem. Acc., 139, 25 (2020), you are highly encouraged to read it. If this module is involved in your work, please not
only cite Multiwfn original paper but also cite this paper, thanks!
The purpose of this example is illustrating how to study  electron structure of non-planar
systems based on detected  type of localized molecular orbitals (LMOs), and show how to evaluate
 composition of molecular orbitals. We will use a simple non-planar molecule cycloheptatriene as
example. If you have not read Section 3.100.22, please read it first, in which details and algorithm
744
are described.
Note that the way of detecting  orbitals for non-planar systems is very different to that for
exactly planar systems. I have already given example for the latter case. In Section 4.5.3, I have
mentioned how to make Multiwfn automatically detect  molecular orbitals for an exactly planar
system and then separately study  and  electron structure by means of ELF- and ELF-. In
Section 4.4.9, I illustrated how to study  electron delocalization path for a planar system porphyrin
by plotting LOL- map.
Care must be taken that diffuse function should not be employed, otherwise the analyses
described in this section may result in unreliable conclusion.
Detecting  type of LMOs

For non-planar systems, to separately study  and  electrons, the molecular orbitals must be
firstly transformed to LMOs. If you are not familiar with LMOs, see Section 3.22.
examples\cycloheptatriene.fch
-1 // Current orbitals are in localized form
0 // Detect  LMOs under default settings and then set their occupation numbers
There are three  LMOs identified:
Expected pi orbitals, occupation numbers and orbital energies (eV):
23 2.000000 -8.361389
24 2.000000 -6.762986
25 2.000000 -5.798201
Total number of pi orbitals: 3
Total number of electrons in pi orbitals: 6.000000
Assume that we will study  electron character, we choose option 2 to set occupation number
of all other orbitals to zero.
Studying  electron structure based on  type of LMOs

Because the occupation number of orbitals other than the  ones have been set to zero, and
meantime density matrix has been automatically updated, now we can carry out any kind of analysis,
the result will only contributed by  electrons. For example, we plot electron density isosurface map
(isovalue=0.03 a.u.) and LOL isosurface map (isovalue=0.5) using main function 5 in usual way,
they are shown below as left and right graphs, respectively.
745
From both the graphs, it is easy to recognize that C4-C6, C1-C2 and C3-C5 must be stronger than
C1-C4 and C2-C3, since  electrons are delocalized much more substantially among the
corresponding two atoms in the formers
Based on the  type of LMOs, we can also carry out other type of analyses, for example, Mayer
bond order calculation. We calculate Mayer bond order as illustrated in 4.9.1, the result is
# 1: 1(C ) 2(C ) 0.70534437
# 2: 1(C ) 4(C ) 0.18472895
# 3: 1(C ) 5(C ) 0.10215925
# 4: 2(C ) 3(C ) 0.18472484
# 5: 2(C ) 6(C ) 0.10215398
# 6: 3(C ) 5(C ) 0.80861164
# 7: 4(C ) 6(C ) 0.80860020
# 8: 5(C ) 6(C ) 0.07556895
The order of the  bond order is C4-C6 > C1-C2 > C1-C4, this result is in full agreement with our
expectation via visualzing the ELF- map.
The function illustrated in this section is quite powerful and can also be applied to fairly large
systems. For example, below left graph is LOL- isosurface map (isovalue=0.55) of helicene plotted
using above procedure. Moreover, using option "5 Set constraint of atom range", one can define
constraint of the region for identifying  LMOs. For example, below right graph shows the electron
density of helicene contributed by five  LMOs located at the central two six-membered rings.
Multi-center bond order calculation can also be normally carried out under the LMO
746
representation, therefore you can easily identify strength of  conjugation of different rings in above
system via this analysis.
Evaluating  composition of occupied MOs

Based on detected  LMOs, we can evaluate  composition of any orbital of present system.
Let us check  composition of occupied MO of the cycloheptatriene. The new.fch in current folder
was automatically exported when we perform earlier orbital localization, all occupied LMOs are
recorded in this file.
Boot up and input below commands:
new.fch // Load it to retrieve occupied LMOs
-1 // Detect  orbitals and then evaluate  composition for orbitals in another file
examples\cycloheptatriene.fch // This file contains MOs
[Press ENTER button directly to use printing threshold of 50%]
Now all occupied MOs with  composition higher than 50% have been shown:
Orbital 23 (Occ= 2.00000) pi composition: 65.003%
After this analysis, the orbitals in memory have been replaced with the ones recorded in
examples\cycloheptatriene.fch. Therefore now you can return to main menu and then enter main
function 0 to plot isosurfaces of the  MOs, as shown below
You can see that the evaluated  compositions are very reasonable. For example, in MO23, you
can see that there are evident  characters at C6-C7 and C5-C7 bonding regions, in addition, there
are isosurfaces around H10, H11 and H14, these observations explain why  composition of MO23
is not quite close to 100%. In constrast,  character of orbital isosurface of MO25 is not so evident,
therefore MO25 can be viewed as a quasi- orbital.
Evaluating  composition of unoccupied MOs

It is also possible to evaluate  compositions for unoccupied MOs. In this case, unoccupied
LMOs are also needed. Now we do this kind of analysis. We return to main menu and then input
2 // Localize both occupied and unoccupied orbitals
747
3 // Switch the LMOs in consideration to "all localized orbitals"
2 // Change the default density threshold for identifying  orbitals. Because current geometry
is highly distorted, more loose density threshold must be adapted, otherwise you will find no
unoccupied  LMO could be recognized
0.05 // A value much larger (more looser) than the default threshold
-1 // Detect  orbitals and then evaluate  composition for orbitals in another file
Now six  LMOs have been identified, as shown below. You can visualize them if you have
interesting
Expected pi orbitals, occupation numbers and orbital energies (eV):
23 2.000000 -8.361389
24 2.000000 -6.762986
25 2.000000 -5.798201
26 0.000000 -0.739482
27 0.000000 0.770830
110 0.000000 74.313859
Then we input
examples\cycloheptatriene.fch // The file containing MOs
[Press ENTER button directly to use printing threshold of 50%]
As can be seen, there are three unoccupied MOs having  composition larger than 50%, while
 compositions of occupied MOs are identical to those we obtained earlier. Now we enter main
function 0 to plot isosurface map for MO27, which is one of unoccupied MOs and has 
compositions of 72.5%. From below map, it is clear that major character of this orbital is indeed .
More discussions and illustrations about analyzing  electron character can be found in my
blog article "Separate investigation of  electronic structure in Multiwfn" (in Chinese,
748
4.200.4 Study iso-chemical shielding surface (ICSS) and magnetic

shielding distribution for benzene
Iso-chemical shielding surface (ICSS) denotes isosurface of magnetic shielding value, which
presents intuitive picture on aromaticity. If you are familiar with NICS, you can also simply view
ICSS as the isosurface of NICS with inverted sign. Please see Section 3.200.4 for more information.
In this example we will study benzene. Since this is a planar system, we will plot ICSSZZ instead of
ICSS, namely only the component of magnetic shielding tensor perpendicular to molecular planar
will be taken into account. ICSSZZ must be more physically meaningful than ICSS, just like NICSZZ
is a better aromaticity index than NICS (as demonstrated in Org. Lett., 8, 863 (2006)). Meanwhile I
will also show how to plot magnetic shielding values along a line and in a plane.
AFAIK, ICSSZZ was first proposed by me in this manual. If ICSSZZ is involved in your work,
please cite my study work involving ICSSZZ: Carbon, 165, 468 (2020).
You should first prepare a Gaussian input file of standard single point task for present system,
which will be taken as template input file later. This file has already been provided as
examples\ICSS\benzene.gjf, in which the geometry has already been optimized at a reasonable level.
examples\ICSS\benzene.gjf // Note that molecular plane is in XY plane
4 // Generate grid data of ICSS or related quantities
1 // Low quality grid, magnetic shielding tensor at 130910 points will be calculated by
Gaussian later. Using "medium quality grid" could result in smoother maps, but the calculation will
be much more expensive
n // Do not skip the step of generating Gaussian input file, because this is the first time we
carry out analysis and thus currently we do not have Gaussian input/output files in hand
Now Multiwfn generates a lot of Gaussian input files of NMR task in currenet folder based on
the template file. The files are named as NICS0001.gjf, NICS0002.gjf ... NICS0017.gjf. Run these
files by Gaussian, the NICS0001.gjf must be run as the first one. The output files can be directly
downloaded from here: http://sobereva.com/multiwfn/extrafiles/benzene_ICSS.rar.
Note: If these files cannot be runned by your Gaussian normally, please check the tail of the output file, there
are two common reasons:
(1) The %mem is too small to finish the task, you need to set %mem in the template .gjf file to a large value
and retry.
(2) The "NICSnptlim" in settings.ini is too large, you should properly reduce it and try again. The reason is that
in Gaussian there is a limit on the number of Bq atoms, and it is somewhat dependant of the version of Gaussian and
your computer. For G09 D.01 and E.01, you should add "guess=huckel" keyword, otherwise due to memory
allocation bug, the NICSnptlim has to be reduced to a very small value to make Gaussian run normally (in this case
the overall computational cost will be quite high). If error occurs in Link401 when "guess=huckel" is used, then try
to use "guess=core" instead. For G16, the guess keyword is not needed.
Hint: You can make use of the script "examples\runall.sh" (for Linux) or "examples\runall.bat" (for Windows),
which invokes Gaussian to run all .gjf files in current folder to yield output files with the same name but with .out
suffix.
Assume that the output files (NICS0001.out, NICS0002.out...) have been placed in
749
"C:\benzene" folder, in Multiwfn you should input C:\benzene\NICS. We want to study ICSSZZ first,
therefore we choose "5: ZZ component", then Multiwfn loads all Gaussian output files and convert
magnetic shielding tensors to grid data of ICSSZZ. After that you will see a new menu, you can
directly visualize isosurface of the grid data by option 1, export it as cube file by option 2 or reselect
the form of ICSS by option -1. The isosurface of ICSSZZ=2.0 ppm is shown below
As you can see, the green isosurface (positive Z-component shielding value), completely
covers the region above and below the benzene ring, suggesting that due to the induced ring current
originated from the globally delocalized π-electrons, the Z-direction external magnetic field is
largely shielded in these regions, this observation implies the strong aromaticity of benzene. From
below scheme we can understand the ICSSZZ more deeply; in the cylindrical region perpendicular
to and through the benzene, the direction of induced magnetic field (purple arrows) is exactly
opposite to external magnetic field (B0), this is why in this region Z-component of magnetic
shielding value is large.
You can also see, blue isosurface (negative Z-component shielding value) presents in the outlier
region of benzene, exhibiting de-shielding effect. This is mostly because the induced magnetic field
is parallel to B0 and thus enhances B0 in this region.
If you properly rotate viewpoint, you will clearly find the C-H bond is also completely covered
by the green surface. The reason is that the σ-electrons involved in the C-H bonding form
conspicuous local induced ring current, so the external magnetic field is also strongly shielded
around the C-H bond.
If you want to export current grid data as .cub file so that you can visualize it via third-part
softwares such as VMD, you can close the GUI window and select 2 to export the grid data to
ICSSZZ.cub in current folder.
Directly study ICSS based on existing Gaussian output files

Assume that you have already obtained the Gaussian output files for the ICSS purpose, and
you want to directly study ICSS, you should input following commands after booting up Multiwfn:
examples\ICSS\benzene.gjf
750

4 // Generate grid data of ICSS or related quantities
1 // Low quality grid
y // Skip generating Gaussian input files and thus directly load Gaussian output files
C:\benzene\NICS
5 // Study ICSSZZ
As you can see, the grid setting we adopted this time is exactly identical to that we used to
generate the Gaussian input/output files, this point is extremely important. If the grid setting is not
the same, the file loading must be failed.
Plot plane map of magnetic shielding value

Next I will show how to plot magnetic shielding value in a plane. Since we already have grid
data of ICSSZZ in hand, magnetic shielding value at any point in a line/plane can be easily obtained
by means of linear interpolation technique based on the grid data.
We first plot color-filled map for ICSSZZ in the YZ plane with X=0. This plane is normal to
benzene and crosses C4-H10 and C1-H7. Set "iuserfunc" in settings.ini to -1, and then boot up a
new Multiwfn instance and input below commands
ICSSZZ.cub
4 // Plot plane map
100 //User-defined function, which now corresponds to the function interpolated by the grid
data of ICSSZZ.cub
[Press Enter button]
0 // Set extension distance of the plot
8 // 8 Bohr
3 // YZ plane
0 // X=0
Now the graph pops up, close it and then input
4 // Show atom labels
11 // Crimson text
1 // Change lower and upper limit
-5,45
-1 // Replot the map
751
From the graph one can find that although Z-component of magnetic shielding in the center of
benzene is an evident positive value, the magnitude is by far less than in the region above and below
the ring plane. The reason is clear, that is benzene only has π-aromaticity, while its σ-electrons are
not globally delocalized to form σ-aromaticity, so the shielding effect is relatively weak in the plane
due to lack of contribution of σ-ring current.
Curve map of magnetic shielding value

Next, we plot curve map to study the variation of magnetic shielding in the line perpendicular
to ring plane and starting from ring center. Choose -5 to return to main menu and input
3 // Plot curve map
2 // Input coordinate of two points to define a line
0,0,0,0,0,12 // The line starts from ring center (0,0,0) and ends at 12 Bohr above the ring
plane
It can be seen that the maximum of Z-component of magnetic shielding value occurs about 1.8
Bohr above the ring plane.
752
Probably you have noticed that in some regions the curve map and plane map are not smooth
enough, this is because the grid of ICSSZZ.cub is relative coarse (the grid spacing is large). If we
use finer grid when calculating ICSSZZ, the graphs will be improved. Also beware that since the
extension distance used in the calculation of grid data of ICSSZZ is only 12 Bohr, when we plot
curve or plane map based on the interpolated data of ICSSZZ, the spatial range involved in the map
should not be too large. For example, we cannot plot the curve map from (0,0,0) to (0,0,20).
ICSS is very useful for discussing aromaticity and anti-aromaticity, many instances can be
found in the original paper of ICSS (J. Chem. Soc. Perkin Trans. 2, 2001, 1893), and in some
applicative papers, such as J. Phys. Chem. A, 116, 5674 (2012) as well as my research work on
cyclo[18]carbon, namely Carbon, 165, 468 (2020).
I strongly recommend the users do some more practices about plotting and analyzing
ICSS/ICSSZZ, I provided some ideal exercise systems in "examples\ICSS" folder, including azulene,
cyclobutadiene, cycloheptatriene, porphyrin, propane and pyracylene; among them cyclobutadiene
is the most simple one. Below is the ICSS=0.5 isosurface of cyclobutadiene showing in two styles;
from the graph it is clear that this system shows strong anti-aromaticity character, the 4n π-electrons
cause evident de-shielding effect in the cylindrical region perpendicular to and through the ring, this
situation is in complete contrast to benzene.
I wrote a very detailed post to discuss ICSS, in which all systems in "examples\ICSS" folder
are involved, see "Using Multiwfn to study aromaticity by drawing iso-chemical shielding surfaces"
(in Chinese, http://sobereva.com/216).
4.200.5 Plot radial distribution function of electron density
Multiwfn is capable of plotting radial distribution function (RDF) for any real space function,
see Section 3.200.5 for detail. This function is particularly useful for studying electronic structure
character of sphere-like system.
This section consists of two parts. In part 1, we will plot RDF for electron density of fullerene
(C60); while in part 2, I will show how to plot RDF of electron density for a Rydberg orbital to
characterize it quantatively.
Part 1: RDF of electron density for fullerene
753
Since .wfn file of fullerene at B3LYP/6-31G* level is large, I only provide the corresponding
Gaussian input file for you (C60.gjf in "example" folder), please properly modify and run it by
Gaussian to produce C60.wfn.
C60.wfn
5 // Plot RDF for a real space function
3
1,6 // Set the lower and upper limit of RDF to 1.0 and 6.0 Å, respectively
0 // Calculate RDF and its integration curve
After the calculation is finished, select option 0, below RDF map will be shown on the screen,
the X-axis corresponds to radial distance
As you can see, the peak of RDF is about 3.5 Å, this is because the distance between nucleus
of carbons and the sphere center is 3.545 Å. It is known that electron density has maximum at
nuclear position for any atom except for hydrogen.
If you carefully examine the RDF curve, you will find that the curve on the right side of the
peak is slightly higher than that on the left sight. The reason is that the amount of  electrons at
outer side of fullerene is richer than inner side. You can also draw and analyze ELF isosurface map
to confirm this point.
You can also choose option 2 to plot integration curve of RDF, as shown below
754
Probably you have noticed that after the calculation there is a prompt on screen:
Integrating the RDF in the specified range is 344.9321931513
which means integrating the RDF from r=1.0 to r=6.0 Å yields 344.932, it also corresponds to the
value of integration curve at r=6.0. This value deviates evidently from our expectation, namely the
number of electrons in current system (360). One reason is that the default number of integration
points is not large enough (500 and 2030 for radial and angular parts, respectively. You can manually
increase them), while another reason is more important, that is the size of current system is too large,
using single-center integration method is too difficult to yield very accurate result, at least for
integrating electron density.
Part 2: Using RDF of electron density to quantitatively characterize Rydberg orbital for
acetone
Rydberg orbitals denote the spatially very diffuse MOs, their orbital shapes are akin to atomic
orbitals, since electrons in Rydberg orbitals can be regarded as weakly bounded by a small cation
core, which behaves as an atomic nucleus. In order to faithfully represent the diffuse character of
Rydberg orbitals, basis set with substantial diffuse functions must be employed, e.g. aug-cc-pVTZ.
The example file we used below is formaldehyde calculated at B3LYP/aug-cc-pVTZ level by
Gaussian. We will first visualize Rydberg orbitals as isosurfaces, and then calculate RDF of electron
density corresponding to these orbitals to characterize them quantitatively.
Boot up Multiwfn, load examples\H2CO_aVTZ.fch, then enter main function 0. Since Rydberg
orbitals are very diffuse, in order to avoid truncating their isosurfaces when viewing them, you
should first select "Other settings"-"Set extension distance" and input a large value, here we input
12. After that, change isovalue to a much smaller value than default, e.g. 0.01. Then we arbitrarily
select some virtual MOs to examine their features. You will find a lot of virtual MOs show very
diffuse character, for example, MO10 and MO11 are shown below:
755
The main distribution region of both of them are far from the molecule. MO10 is almost spherically
symmetric, thus is can be denoted as s type of Rydberg orbital. While MO11 has two phases, they
equally distribute at the two sides, the overall shape is very close to atomic p orbitals, therefore
MO11 can be identified as p type of Rydberg orbital.
How to quantitatively demonstrate that the main distribution region of these Rydberg orbitals
is far from the molecular center? One of the best way is plotting RDF of electron density
corresponding to these orbitals. Here we plot this kind of RDF map for MO11. We close the GUI of
main function 0, and then input below commands:
26 // Modify orbital occupation number
0 // Select occupation number of all orbitals to zero
11 // Select orbital 11
2 // Set occupation number of orbital 11 to 2.0 (assume it is doubly occupied)
q // Return
200
5 // Plot RDF
3 // Set lower and upper limit of radial plotting
0,10 // From 0 to 10 Å
4 // Set angular number of integration points. The default value is unnecessarily high for
present purpose, therefore we set it to a smaller value to reduce computational time
302 // 302 angular points
0 // Calculate RDF for electron density (which is the default real space function)
1 // Plot the RDF map
756
From the map it can be seen that the global maximum peak is at about 4.6Å, indicating that the
major distribution region of this orbital wavefunction is very far from origin (in present .fch file,
Cartesian origin corresponds to molecular center), and thus MO11 can be unambiguously identified
as a Rydberg orbital.
Please also plot such a RDF map for a regular valence virtual MO, e.g. MO9, which is *
orbital. Where is its peak position?
4.200.6 Studying correspondence between orbitals in different

wavefunction files
In this section, two examples are given to illustrate how to use the function introduced in
Section 3.200.6, this function aims to reveal relationship between orbitals in different wavefunction
files.
4.200.6.1 Revealing relationship between HF and MP2 orbitals of CH3NH2
In this section, we study the correspondence between HF/6-31+G* MOs and MP2/6-31+G*
natural orbitals (NO) for CH3NH2.
After booting up Multiwfn we input
C:\CH3NH2_MP2.wfn // MP2/6-31+G* wavefunction file, there are 48 NOs
6 // Analyze correspondence between orbitals in two wavefunctions
[Press ENTER button directly to choose all orbitals]
C:\CH3NH2_HF.wfn // HF/6-31+G* wavefunction file, there are 9 MOs
[Press ENTER button directly to choose all orbitals]
Then you will see
1: 2( 94.63%) 1( 5.36%) 4( 0.01%) 3( 0.00%) 6( 0.00%)
2: 1( 94.64%) 2( 5.36%) 3( 0.00%) 9( 0.00%) 6( 0.00%)
757
3: 3( 85.21%) 4( 9.87%) 7( 3.24%) 9( 0.88%) 6( 0.79%)
4: 4( 86.88%) 3( 10.66%) 6( 1.94%) 9( 0.48%) 7( 0.02%)
5: 6( 54.88%) 7( 41.11%) 9( 3.20%) 4( 0.45%) 3( 0.34%)
6: 8( 57.50%) 5( 42.47%) 7( 0.00%) 6( 0.00%) 4( 0.00%)
7: 5( 57.50%) 8( 42.48%) 7( 0.00%) 4( 0.00%) 6( 0.00%)
8: 9( 86.62%) 6( 10.30%) 7( 1.63%) 4( 0.89%) 3( 0.50%)
9: 7( 53.98%) 6( 32.07%) 9( 8.76%) 3( 3.28%) 4( 1.89%)
10: 9( 0.00%) 7( 0.00%) 6( 0.00%) 4( 0.00%) 2( 0.00%)
11: 5( 0.00%) 8( 0.00%) 3( 0.00%) 7( 0.00%) 4( 0.00%)
... (ignored)
47: 6( 0.01%) 7( 0.00%) 9( 0.00%) 3( 0.00%) 4( 0.00%)
48: 9( 0.04%) 7( 0.00%) 6( 0.00%) 4( 0.00%) 3( 0.00%)
The first column denotes the index of the MP2 NOs, the largest five contributions from the
Hartree-Fock MOs to them are shown at right side. As can be seen, the first (second) NO is nearly
equivalent to the second (first) MO. While the 9th NO cannot be solely represented by any MO, it
mainly arises from the severe mix of the 7th (53.98%) and 6th MOs (32.07%), the 9th MO also has
non-neglectable contribution (8.76%).
If you want to get all coefficients and compositions of an orbital of present wavefunction (e.g.
the 5th NO), then simply input 5, you will see
1 Contribution: 0.000 % Coefficient: 0.001509
2 Contribution: 0.000 % Coefficient: -0.000042
Total: 99.980 %
this output shows that NO 5 = 0.7408 MO 6 − 0.6412 MO 7 − 0.1788 MO 9  The

99.98% at the last line indicates that the 5th NO can be perfectly represented by linear combination
of these nine MOs.
4.200.6.2 Study contribution of lone pair of nitrogen to MOs of dopamine
Sometimes it is useful to investigate a problem like "How large is contribution of lone pair of
an heteroatom to a molecular orbital?", using orbital localization technique in combination with
subfunction 6 of main function 200 can provide a clear answer to this problem. In this section, we
will study contribution of lone pair of nitrogen in amino group to occupied MOs for a Donor--
acceptor type of molecule. The wavefunction file is examples\excit\D-pi-A.fchk, the structure and
atomic numbering are shown below
758
We should first generate localized molecular orbitals (LMOs), because commonly a lone pair
can be represented by a LMO.
19 // Orbital localization analysis
The LMOs are automatically exported to new.fch in current folder. From the outputted LMO
composition on the screen, we can find there are several LMOs closely related to the N24, namely
the nitrogen in amino group. Below are relevant lines of orbital composition output.
4: 24(N ) 99.3% 5: 18(C ) 98.9% 6: 6(C ) 98.6%
...
19: 24(N ) 60.1% 25(H ) 32.9% 20: 21(N ) 57.5% 18(C ) 33.7%
27: 2(C ) 45.6% 3(C ) 44.5% 28: 24(N ) 53.6% 6(C ) 37.8%
33: 23(O ) 65.8% 21(N ) 27.5% 34: 24(N ) 59.9% 26(H ) 33.2%
43: 5(C ) 46.2% 4(C ) 44.9% 44: 24(N ) 77.7% 6(C ) 9.9%
To confirm which one corresponds to lone pair of N24, we enter main function 0 and check
isosurface of the highlighted LMOs one by one, we find LMO 44 can be regarded as lone pair orbital
of N24, the isosurface map with isovalue of 0.1 is shown below
Now we can check contribution of this LMO to various MOs. Reboot Multiwfn and input
6 // Analyze correspondence between orbitals in two wavefunctions
1,56 // We want to check all occupied MOs (indices range is 1~56)
new.fch // The file containing LMOs
44,44 // Only the 44th orbital in the new.fch will be taken into account
From the output, we can find some MOs have large composition of the LMO 44, relevant lines
are shown below (Since only LMO 44 is taken into account, other LMOs have exactly zero
contribution)
44: 44( 10.04%) 43( 0.00%) 42( 0.00%) 41( 0.00%) 40( 0.00%)
* 45: 44( 8.73%) 43( 0.00%) 42( 0.00%) 41( 0.00%) 40( 0.00%)
* 49: 44( 33.29%) 43( 0.00%) 42( 0.00%) 41( 0.00%) 40( 0.00%)
759
* 53: 44( 14.28%) 43( 0.00%) 42( 0.00%) 41( 0.00%) 40( 0.00%)
* 56: 44( 19.74%) 43( 0.00%) 42( 0.00%) 41( 0.00%) 40( 0.00%)
We return to main menu, enter main function 0 and visualize MO 49 to check whether LMO
44 (lone pair of N24) really has large contribution (33.29%) to it. The isosurface map with isovalue
of 0.05 is shown below
As can be seen, indeed MO 44 is largely contributed by lone pair of N24, since a large part of its
isosurface covers the lone pair region of N24, the computed composition 33.29% should be a
reasonable value.
4.200.7 Parse output file of “polar” task of Gaussian to obtain

(hyper)polarizability and calculate related quantities
As introduced in Section 3.200.7, Multiwfn has ability to parse the abstract output information
of “polar” task of Gaussian and reorganize the data to a much more readable format, and meantime
many useful quantites are outputted together. In this section an example is given.
In "examples\polar" folder, NH3_polar_static.out and NH3_polar_dynamic.out are output files
of static and frequency-dependent (hyper)polarizability calculations for NH3 via a common
exchange-correlation functional PBE0. The corresponding Gaussian input files are also provided in
the same folder. In the calculation, a large basis set with abundant diffuse functions (aug-cc-pVTZ)
was used, notice that diffuse functions play a vital role in yielding reasonable (hyper)polarizability
and thus they are absolutely indispensable.
Studying static polarizability and first hyperpolarizability

First, we use Multiwfn to parse data from the NH3_polar_static.out. Boot up Multiwfn and
input
examples\polar\NH3_polar_static.out
200 // Main function 200
7 // Parse (hyper)polarizability task of Gaussian
1 // Start parsing. We select option 1 because third-order derivative of current level (normal
DFT functional) is supported by Gaussian
Now you can see below information on screen:
Dipole moment:
X,Y,Z= 0.000000 0.000000 -0.596558 Norm= 0.596558
Static polarizability:
XX= 13.692700
760
XY= 0.000000
YY= 13.693300
XZ= 0.000000
YZ= -0.000334
ZZ= 15.323600
Isotropic average polarizability: 14.236533
Isotropic average polarizability volume: 2.109636 Angstrom^3
Polarizability anisotropy (definition 1): 1.630600
Eigenvalues of polarizability tensor: 13.69270 13.69330 15.32360
Note: It is well known that the sign of hyperpolarizability of Gaussian 09/16 s
hould be inverted, the outputs shown below have already been corrected
Static first hyperpolarizability:
XXX= 0.000000
XXY= 12.903200
XYY= 0.000000
YYY= -12.900200
XXZ= 9.617100
XYZ= 0.000000
YYZ= 9.564710
XZZ= 0.000000
YZZ= 0.016285
ZZZ= 26.613200
Beta_X= 0.00000 Beta_Y= 0.01929 Beta_Z= 45.79501
Magnitude of first hyperpolarizability: 45.795014
Projection of beta on dipole moment: -45.795010
Beta || : -27.477006
Beta ||(z) : 27.477006
Beta _|_(z) : 9.159002
The output is easy to understand, all of the outputted quantities have been detailedly explained in
Section 3.200.7.
Studying frequency-dependent polarizability and hyperpolarizability

Next, we extract frequency-dependent (hyper)polarizability from the NH3_polar_dynamic.out.
examples\polar\NH3_polar_dynamic.out
-1 // Let Multiwfn load frequency-dependent (hyper)polarizability
1 // Start parsing
Multiwfn detected there are three set of data:
761
1 w= 0.000000 ( Static )
2 w= 0.065600 ( 695.04nm )
3 w= 0.071900 ( 634.14nm )
If we input 2, then the (hyper)polarizability data corresponding to 0.0656 a.u. incident light will be
loaded and parsed.
After outputting information about dipole moment and polarizability, Multiwfn asks you to
choose the type of hyperpolarizability to be outputted:
1: Beta(-w;w,0) 2: Beta(-2w;w,w)
Note: Option 2 is meaningless if "DCSHG" keyword was not used
You can select either one according to your requirement. Since as can be seen in
NH3_polar_dynamic.gjf, the keyword we used is polar=DCSHG, both the two types are available.
Now if you choose option 2 to parse (-2w;w,w), not only first hyperpolarizability will be
shown, but also quantities related to Hyper-Rayleigh scattering (HRS) will be printed, as shown
below (see Section 3.200.7 for detail)
<beta_ZZZ^2>: 8.12843560E+02
<beta_XZZ^2>: 1.45318790E+02
Hyper-Rayleigh scattering (beta_HRS): 30.954
Depolarization ratio (DR): 5.594
|<beta J=1>|: 59.743
|<beta J=3>|: 41.624
Nonlinear anisotropy parameter (rho): 0.697
Dipolar contribution to beta, phi_beta(J=1): 0.589
Octupolar contribution to beta, phi_beta(J=3): 0.411
< (beta_ZXZ+beta_ZZX)^2 - 2*betaZZZ*betaZXX >: 2.04387968E+02
Next, if you want to study evolution of scattering intensity with respect to polarization angle
of incident light, you should input y and then input an initial angle, for example, -180. After that
HRS_angle.txt will be generated in current folder, which contains scattering intensity corresponding
to polarization angle varying from -180 to 179 with stepsize of 1. If you use such as Origin to
plot the data as "Polar theta(X) r(Y)" map, you will obtain below map, in which the radial distance
of the red curve at different angles corresponds to calculated HRS intensity. The corresponding
Origin .opj file has been provided as examples\polar\HRS_angle.opj
762
90
1000 120 60
800
600 150 30
HRS intensity (a.u.)

400
200
0 -180 0
-150 -30
-120 -60
-90
Studying second hyperpolarizability

Gaussian is also capable of calculating static and dynamic second hyperpolarizability (). An
example input file is examples\polar\NH3_gamma.gjf, the output file is also given in the same folder.
In this task polar=gamma keyword was used, and two external frequencies are specified (532 nm
and 680 nm). Below I illustrate using Multiwfn to parse data of (-2w;w,w,0) at 532 nm.
examples\polar\NH3_gamma.out
-1 // Let Multiwfn load frequency-dependent (hyper)polarizability
7 // This option is specific for parsing polarizability and second hyperpolarizability
3 // 532 nm
Dynamic polarizability and relevant quantities at 532 nm now have been shown on screen.
Then choose 2 to further parse gamma(-2w;w,w,0), then you will see
XXXX= 3.966470E+03
YXXX= 0.000000E+00
ZXXX= 8.493370E-05
XYXX=XXYX= 0.000000E+00
[...ignored]
ZZZZ= 3.204030E+04
Magnitude of gamma: 1.067181E+04
X component of gamma: 2.134993E+03
Y component of gamma: 2.134897E+03
Z component of gamma: 1.023580E+04
Average of gamma (definition 1), gamma ||: 1.450569E+04
Average of gamma (definition 2): 1.461464E+04
gamma _|_: 4.653651E+03
As can be seen, all components of  tensor have been very clearly shown, and some quantities
763
defined based on  are also shown, they are very useful in the study of . Note that above data are
given in input orientation.
Similarly, you can use Multiwfn to parse gamma(-w;w,0,0). If you want to parse
gamma(0;0,0,0), do not choose option -1 before parsing.
4.200.8 Studying polarizability and hyperpolarizability based on sum-

over-states (SOS) method
4.200.8.1 Calculate polarizability and hyperpolarizability for NH3

In this example I will show how to use Multiwfn to calculate polarizability and
hyperpolarizability based on sum-over-states (SOS) method for HF molecule. Please make sure that
you have read Section 3.200.8.1.
As introduced in Section 3.200.8.1, SOS calculation requires information of a large amount of
electronic states, including electric dipole moments, excitation energies, and transition dipole
moments between these states. Commonly these information can be obtained by ZINDO, CIS,
TDHF and TDDFT calculations. In this example we use the very popular CIS method. According
to my experiences, the more expensive methods TDHF and TDDFT, which can produce more
accurate excitation energies, do not necessarily give rise to better (hyper)polarizability than CIS
when used in combination with SOS technique.
Preparation
In this example we use Gaussian program to carry out the CIS calculation. However, Gaussian
itself cannot output enough information for SOS calculation. Though for CIS (and ZINDO) there is
a keyword alltransitiondensities, which makes Gaussian output transition density moment between
each pair of excited states, however electric dipole moment of all states still cannot be obtained in
a single run. Fortunately, we can use the very powerful electron excitation analysis module of
Multiwfn to generate all information needed by SOS based on the output file of CIS/TDHF/TDDFT
task of Gaussian or ORCA program.
First, run examples\NH3_SOS.gjf by Gaussian to produce output file NH3_SOS.out, and use
formchk utility to convert the checkpoint file to NH3_SOS.fch. If you do not have Gaussian in hand,
you can directly download them at http://sobereva.com/multiwfn/extrafiles/NH3_SOS.rar.
Since the SOS results converge often slow with respect to the number of excited states taken into account, we
produce as high as 150 excited states in this example to substantially avoid truncation error. Of course, employing
higher number of excited states needs more computational time in both of the CIS calculation and the subsequent
SOS calculation in Multiwfn. In most practical studies, 100 states is generally large enough, and even 70 is often
enough to provide usable results. Calculation of hyperpolarizability, especially the high-order ones, has very
stringent requirements on the quality of basis set, abundant diffuse functions are absolutely indispensable. In this
example we employ def2-TZVPPD (J. Chem. Phys., 133, 134105), which is a high-quality basis set optimized for
calculation of molecular response properties. Since this is not a built-in basis set in current version of Gaussian, it
was picked from EMSL website (https://bse.pnl.gov/bse/portal). The keyword IOp(9/40=5) is important, because in
the ZINDO/CIS/TDHF/TDDFT task by default Gaussian only outputs the transition coefficients whose absolute
values are larger than 0.1, while IOp(9/40=5) lowers the criterion to 0.00001, so that much more coefficients can be
outputted, and thereby we can obtain accurate transition dipole moments by Multiwfn at next step. Worthnotingly,
you can also use TDHF or TDDFT instead of CIS, for example you can write #P TD(nstates=150) CAM-B3LYP/gen
IOp(9/40=5).
C:\NH3_SOS.fch
18 // Electron excitation analysis module
5 // Calculate transition dipole moments and dipole moment for all excited states
764
C:\NH3_SOS.out
3 // Generate SOS.txt
The file SOS.txt generated in current folder contains all information needed by SOS
(hyper)polarizability calculation. This file can be directly used by SOS module of Multiwfn.
SOS.txt
8 // Study (hyper)polarizability by sum-over-states (SOS) method
Note that all units used in the SOS module are atomic units.
Calculation of polarizability (alpha)

Now, select 1 and input 0 to calculate static polarizability (0;0), the result is
1 2 3
1 14.610687 -0.000000 0.000000
2 -0.000000 14.610687 0.000000
3 0.000000 0.000000 14.552482
Isotropic average polarizability: 14.591285
Isotropic average polarizability volume: 2.162205 Angstrom^3
Eigenvalues: 14.552482 14.610687 14.610687
As can be seen, not only the polarizability tensor, but also some related quantities are outputted.
Their definitions can be found in Section 3.100.20. The isotropic average polarizability we obtained
here is 14.59, which is in perfect agreement with the experimentally determined value 14.56! (Mol.
Phys., 33, 1155)
Then we calculate dynamic polarizability (-;) and assume the frequency of external field
to be 0.0719 a.u. Select option 1 again, input 0.0719, from the output we can see the dynamic
isotropic polarizability at =0.0719 is 14.86, which is slightly larger than the static counterpart.
Calculation of first hyperpolarizability (beta)

Next, we calculate first hyperpolarizability and consider the static case (0;0,0). Select option
2 and input 0,0. Only the  component along the dipole moment direction, namely ||, is what we
are particularly interested in, since only this quantity can be determined experimentally. From the
output we find ||(0;0,0) is -38.98. The corresponding experimental value is not available, however
this value is close to the highly accurate value calculated by CCSD(T) method (-34.3, see J. Chem.
Phys., 98, 3022 (1993)).
Subsequently, we calculate (-2;,) at =0.0656 a.u. Select option 2 again and input 0.0656,
0.0656 to set the frequency of both two external fields as 0.0656 a.u. This time the || value is -49.69,
which is again in excellent agreement with the experimentally determined value -48.91.2 (see A.
Hernández-Laguna et al. (eds.), Quantum Systems in Chemistry and Physics, Vol. 1, p111)
Note: Although in this example the agreement between our SOS/CIS calculations and the reference values is
suprisingly good, this is not always hold for other systems. The SOS/CIS method sometimes severely overestimates
 value!
Calculation of second hyperpolarizability (gamma)

Finally, we calculate second hyperpolarizability . Select 3 and input 0,0,0 to assume static
electric fields. Since calculation of  is evidently more time-consuming than , Multiwfn does not
765
automatically use all states but allow you to set the number of states to be taken into account. Larger
value in principle gives rise to better result, but of course more time will be consumed in the
calculation. Here, we input 150 to use all states. After a while, result is shown on screen, the average
of  is 928.74. Beware that this value may be inaccurate (reference value is not available, so I am
not sure if this is a good result), one of the main reasons is that the basis set we used in the electron
excitation calculation is not large enough. Accurate calculation of  usually requests a basis set like
d-aug-cc-pVTZ (or a even better one), which has an additional shell of diffuse functions compared
to the commonly used aug-cc-pVTZ.
Multiwfn is also capable of calculating third hyperpolarizability (-;1,2,3,4) where
=1+2+3+4, but we do not do this in present example, because this quantity is fairly
unimportant, and the calculation is terribly expensive when the number of states in consideration is
large; moreoever, a sky-high quality of basis set must be employed in the calculation...
Variation of (hyper)polarizability with respect to number of considered states

A very important point in SOS calculation is that the number of states used must be high enough;
in other words, if n states are involved in your SOS studies, the variation of the (hyper)polarizability
with respect to the number of states have to be converged before n, otherwise n must be enlarged.
By using Multiwfn we can readily examine if the convergence condition is satisfied. Here we check
the convergence of static . Select option 6 and input 0,0, after a while, the result will be exported
to beta_n.txt and beta_n_comp.txt in current folder, the meaning of each column is clearly indicated
on screen. Then you can use your favourite tool to plot the data in the beta_n.txt. If you are an Origin
user, you can directly drag this file into Origin window and plot the data as curve maps. Below
graphs show the variation of static || as well as static isotropic average polarizability <> with
respect to number of considered states.
16
14 -20
12
-30
10
-40
< >
8
||
6 -50
4 -60
2
-70
0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Number of states Number of states
It is clear that both of the two quantities have basically converged at n=100. Since we employed 150
states in our calculations, the error due to the truncation of states can be safely ignored. From the
graph one can also see that if the number of states is truncated at 70, the results are still qualitatively
correct.
You can also analogously use options 5 and 7 to study the convergence behavior of  and ,
respectively.
Variation of (hyper)polarizability with respect to frequency of external fields

In the SOS module of Multiwfn, one can also easily study the variation of dynamic
(hyper)polarizability with respect to the frequency of external fields. For example, we investigate
the variation of (-1;1,0) as 1 varies from 0 to 0.5 a.u. with stepsize of 0.01. Write a plain text
file, each row corresponds to a pair of 1, 2 (2 is fixed at zero in this example), for example
766
0.00 0
0.01 0
0.02 0
...
0.5 0
Tips: For convenience, you can utilize Microsoft Excel program to generate frequency list, and save the table
as .txt file (you can select such as "Text (tab delimited)", but do not choose "Unicode text").
Then choose option 16, input the path of the plain text file, the  will be calculated at each pair of
frequencies, the result will be outputted to beta_w.txt and beta_w_comp.txt in current folder, the
meaning of each column of the data is clearly indicated on screen. The data in these files can be
directly plotted as curve map by Origin, for example the magnitude of (-1;1,0) is shown below:
110000
100000
90000
80000
Magnitude of  (−1;1,0)
70000
60000
50000
40000
30000
20000
10000
0
0.0 0.1 0.2 0.3 0.4 0.5
Frequency of external field 1
Simiarly, you can also use options 15 and 17 to study the variation of  and  with respect to
frequency of external fields, respectively.
Scanning both 1 and 2 of (-(1+2);1,2)

By subfunction 19 of the SOS module, you can also scan both 1 and 2 of (-(1+2);1,2).
For example, we input below commands in the SOS module
19
-0.6,0.6,100 // Lower limit, upper limit and number of steps of 1 (in a.u.)
-0.6,0.6,100 // Lower limit, upper limit and number of steps of 2 (in a.u.)
After a while, beta_w.txt and beta_w_comp.txt are generated in current folder, the meaning of
each column of the files is clearly mentioned on screen. To visually study how total  varies with
respect to change in 1 and 2, you can plot relief map with first, second and 7th columns as X, Y
and Z data, respectively. The map shown below was plotted by Sigmaplot:
767
This kind map exhibits one photon resonance (-1;1,0) or (-2;0,2), sum frequency
generation and difference frequency generation character of present system. Occurrence of a large
peak at the position where one of 1 and 2 is zero implies significant Electro-optics Pockels effect
at the corresponding frequency, a large peak at a frequency 1=2 indicates strong second harmonic
generation (SHG) effect at that frequency, a large peak at frequency 1= -2 shows remarkable
optical rectification effect at that frequency. See ACS Appl. Nano Mater., 2, 1648 (2019) for example
of discussion of this kind of map.
4.200.8.2 Perform two- and three-level model analysis for first

hyperpolarizability of NH2-biphenyl-NO2
In literatures about computational study of hyperpolarizability, two-level model are frequently

adapted to shed light on the major factors that influence the first hyperpolarizability, this analysis
also provides clear insight into the difference in  between analogous systems. The three-level
model implemented in Multiwfn is a natural extension of the two-level model to include two excited
states into account. Please read Section 3.200.8.2 first to gain basic knowledge about the two- and
three-level models. In this section I will illustrate basic step of performing these analyses
It is important to emphasize that the two- or three-level analysis is useful only when these two
conditions are satisfied:
• Only one component of  (namely XXX or YYY or ZZZ) plays dominant role. This
component should be subjected to the analysis
• One excited state (for two-level model) or two excited states (for three-level model) has much
larger contribution to  than all other excited states
Most donor--acceptor type of systems well satisfy the above two conditions. In this section
we take a typical donor--acceptor system NH2-biphenyl-NO2 as example. The D-pi-A.fchk and D-
pi-A.out in "examples\excit" folder were produced by Gaussian TDDFT task for this system at
CAM-B3LYP/6-31g(d) level, five lowest-lying singlet excited states were yielded. It is worth to
768
note that without sufficient diffuse functions, (hyper)polarizability cannot be estimated at

quantitative accuracy level; however, our present aim is simply using the two- and three-level
models to very roughly discuss the factors influencing the , so using these two files to conduct the
analysis is acceptable.

18 // Electron excitation analysis module
5 // Calculate transition dipole moments and dipole moment for all excited states
examples\excit\D-pi-A.out
3 // Generate SOS.txt, which contains all information needed by the two- or three-level
analysis
Now we are ready to perform the two/three-level analysis. Before doing this, we need to
confirm which component of  should be studied. The molecular geometry of D-pi-A.fchk is shown
below (displayed by main function 0). As can be see, the direction of donor--acceptor path is fully
parallel to X-axis, hence it is expected that only XXX of current system is prominent.

SOS.txt
8 // Study (hyper)polarizability by sum-over-states (SOS) method
20 // Two or three-level model analysis of 
Now program prints some information about the excited states, which are helpful for
identifying the excited state to be analyzed.
Excitation energy (E,eV) and transition dipole moment (X,Y,Z,total) between gro
und state to excited states (a.u.)
State 1 E= 3.9069 X= 0.44441 Y= -0.00044 Z= -0.00182 Tot= 0.44442
State 2 E= 4.0624 X= 2.52778 Y= -0.00309 Z= -0.00733 Tot= 2.52779
State 3 E= 4.4166 X= -0.00263 Y= -0.00167 Z= -0.02343 Tot= 0.02364
State 4 E= 4.7912 X= 0.00069 Y= 0.34055 Z= -0.01274 Tot= 0.34079
State 5 E= 4.8872 X= -0.00391 Y= 0.19205 Z= -0.17054 Tot= 0.25687
Because excited state 2 has much larger magnitude of "Tot" (norm of transition dipole moment)
than others, and meantime the excitation energy of excited state 2 is almost the lowest, thus this
769
state can be unambiguously identified as the so-called crucial state.

Next, in the Multiwfn window we input
1 // Choose X direction. In other words, XXX will be subjected to the analysis
2 // Select excited state 2 for two-level model analysis
The outputted information is shown below
Excited state 2
Excitation energy 4.0624 eV
Transition dipole moment component: 2.5278 a.u.
Oscillator strength component: 0.635943 a.u.
Variation of dipole moment component: 6.5629 a.u.
beta evaluated by the two-level model: 11289.099582 a.u.
All terms involved in the two-level model have been, see Section 3.200.8.2 for the expression
of the two-level model. If you have a similar system, for example the two benzene rings in current
system are replaced with three benzene rings, you can use the above outputted quantities to discuss
how various factors influence the difference in XXX.
We can also carry out three-level model analysis. In the Multiwfn window, we input
20 // Perform two or three-level model analysis of  again
1 // Investigate XXX again
1,2 // Choose excited states 1 and 2 for the three-level model analysis
Below are outputted information
Excited state 1
Variation of dipole moment component: -1.0098 a.u.
Excited state 2
Variation of dipole moment component: 6.5629 a.u.
Transition dipole moment between states 1 to 2: 1.475397 a.u.
Contribution of excited state 1: -58.048845 a.u.
Contribution of excited state 2: 11289.099582 a.u.
Contribution of cross term: 927.898744 a.u.
beta evaluated by the three-level model: 11251.731412 a.u.
It can be seen that all terms involved in the three-level model analysis have been given. We
also find that the contribution of excited state 1 to XXX is quite small, and the cross term due to
coupling between the two states is also negligible, implying that employing two-level model
analysis is already completely adequate for present system, including more states into the analysis
does not provide new insight.
770
The reason why the contribution solely due to excited state 1 is fairly low is quite easy to
understand. According to the two-level model, contribution to a given  component it is positively
proportional to variation of dipole moment and square of transition dipole moment in that
component. Compared to the excited state 2, both the two quantities of excited state 1 are
conspicuously smaller and thus excited state 1 can be safely ignored.
Hint: In some cases, the output file of electron excitation task may contains large number of states, while for
performing two- or three-level model analysis we always only need the first few low-lying excited states. In order
to significantly reduce the cost of generating the SOS.txt in this situation, you can properly set "maxloadexc"
parameter in settings.ini. For example, the output file contains as many as 100 excited states, however it is anticipated
that the state of interest should be no higher than the 5th excited state, hence you can set "maxloadexc" to 5, then
only the first 5 excited states will be recognized, loaded and subjected to calculation.
4.200.12 Calculate energy index (EI) and bond polarity index (BPI)
In this section I will illustrate how to calculate EI and BPI indices, which were defined in J.
Phys. Chem., 94, 5602 (1990). If you are not familiar with these two quantities, please read Section
3.200.12 first. The geometry and wavefunction involved in this example were produced at HF/6-
31G* level, which is the one used in above mentioned paper.
We will calculate BPI for C-N bond of CH3NH2, before this we first need to calculate reference
EI value for C and N atoms, which correspond to EI of C in ethane and N in H2N-NH2, respectively.
examples\EI_BPI\ethane.fch
12 // Calculate energy index (EI) or bond polarity index (BPI)
1 // C1 atom
You will see the EI value for C in reference molecule ethane is -0.667639 a.u.
examples\EI_BPI\N2H4.fch
12 // Calculate EI or BPI
1 // N1 atom
You can see the EI value for N in reference molecule H2N-NH2 is -0.718126 a.u.
Next we calculate EI for C and N in CH3NH2. Reboot Multiwfn and input
examples\EI_BPI\CH3NH2.fch
1 // C1, the result is -0.693374 a.u.
5 // N5, the result is -0.698092 a.u.
The BPICN in CH3NH2.is computed as
BPI CN = (EI C − EI Cref ) − (EI N − EI ref
N )
= −0.693374 + 0.667639 + 0.698092 − 0.718126

= −0.046
As a comparison, use examples\EI_BPI\F2.fch to calculate reference value for F, the result
should be -0.992542 a.u., and use examples\EI_BPI\CH3F.fch to calculate EI for C and F in CH3F
molecule, the result should be -0.750302 and -0.885961, respectively. Then compute the BPICF value
771
as -0.750302+0.667639+0.885961-0.992542= -0.189. Since BPICF in CH3F is evidently more

negative than BPICN in CH3NH2, it can be concluded that C-F bond in CH3F is more polar than C-
N bond in CH3NH2.
Via EI index we can also evaluate the so-called group electronegativity, which is often more
useful than atomic electronegativity. Here we calculate electronegativity for -CH3 group, which is
simply the negative of EIC for CH3 radical. Boot up Multiwfn and input
examples\EI_BPI\CH3.fch // Optimized and produced at UHF/6-31G*
1 // Carbon atom
The result is -0.630656 a.u., corresponding to electronegativity of CH3 group of 0.631. Then
we use examples\EI_BPI\F.fch to calculate group electronegativity for -F, the result is 0.957. It is
clear that -F group has much higher electronegativity, and thus has stronger capacity to attract
electrons than -CH3 group due to its lower average energy per valence electron.
4.200.13 Study orbital contributions to density difference

In this section, I will use a few examples to illustrate how to derive contributions of various
kinds of orbitals to density difference () of different types, and show which chemically valuable
information could be obtained by means of these analyses. Please read Section 3.200.13 first to gain
basic knowledge about the function employed in this section.
4.200.13.1 Contribution of MOs to Fukui function f − of phenol
The basic concept of Fukui function has been detailedly introduced in Section 4.5.4. The Fukui
function f − is frequently employed in literatures to identify favourable sites of electrophilic attack,
it is a special instance of  and is defined as 𝜌𝑁 (𝐫) − 𝜌𝑁−1 (𝐫), where N is the number of electrons
in original state. It is interesting to study relative contribution of various MOs to the f −, so that we
can better recognize its character from orbital point of view. The underlying idea is that if densities
of MOs are regarded as basis functions that can be used to linearly expand f −, then the optimal
expansion coefficients will be able to be used to measure contribution of each MO to the f −.
Evidently, if there is no orbital relaxation effect, the f − must be exactly identical to probability
density of HOMO of N-state. Of course this assumption is not true, because in practice all MOs of
N-state must somewhat undergo deformation when the number of electrons changes from N to N-1.
Below we will take phenol as example to illustrate how to calculate contribution of all occupied
MOs to f −. The files carrying orbital wavefunctions of N and N-1 states have been provided as
phenol.wfn and phenol_N-1.wfn in "examples" folder, respectively. Before calculation of the
contribution, we first need to generate cube file of f −. To do so, boot up Multiwfn and input
examples\phenol.wfn // Wavefunction file of N-state
-,examples\phenol_N-1.wfn
772

2 // Export the grid data as density.cub in current folder
Now, we can directly enter the function used to derive orbital contributions to . Input below
commands
13 // Evaluate orbital contributions to density difference or other grid data
density.cub // The file containing grid data of f −
0 // Choose orbital range and start analysis
o // Only consider orbitals with non-zero occupation in the fitting. For present case all
occupied MOs are chosen (note that the wavefunction file we loaded is in .wfn format, in fact it
only contains occupied orbitals, thus unoccupied orbitals cannot be chosen even if you want)
Soon the contribution values are listed in ascending order:
Orbital 20 Value: -0.108
[ignored]
Orbital 23 Value: 0.111
Sum of all values: 1.000
Fitting error (definition 1): 0.8911
Because by default the sum of all contributions is constrained to 1.0, the "Sum of all values" is
exactly 1.0. The default constraint makes sense for present case, since N and N-1 just differ by 1
electron. The "Fitting error" displays the fitting quality. The smaller the fitting error, the better the
fitting quality. However, the best way of examining the fitting quality is visually comparing the
isosurface of the actual  and the fitted . From above data, it can be clearly seen that MO25
(HOMO) dominates the f −, while MO24 (HOMO-1) and MO23 (HOMO-2) also play nonnegligible
roles.
In the post-processing menu, we can choose options 2 and 3 to visualize isosurface of provided
grid data (i.e. the one loaded from density.cub) and that of fitted grid data, respectively. The
difference between the two set of grid data can be visualized via option 4. The corresponding graphs
are given below, the isovalue is set to 0.005. For comparison purpose, the three MOs having largest
contributions are also given below, the isovalue is 0.07, and the contribution values are marked in
the parenthese.
773
From above map, it can be seen that our fitted grid data is very close to the provided one, at
least the fitted one is able to reasonably reproduce the basic character in  region. The main
difference between the provided and the fitted grid data is the  region, as clearly reflected in the
"diff" subgraph. This observation implies that lose of an electron causes severe relaxation of 
electrons, which cannot be fully represented based on the occupied MOs of N-state.
The MO23, MO24 and MO25 shown in above map are all  orbitals, this fact explains why
the  part of the f − cannot be faithfully represented in the fitted grid data. Notice that a few MOs
have negative contributions, the one with most negative value is MO20 (-0.108). If you check this
orbital, you will find it is a  orbital. However, since its shape is not quite simiar to the "diff" grid
data, its existence does not improve the fitting quality or lower the difference between the provided
and fitted grid data.
From above map we can also find the profile of isosurface of MO25 is very similar to the f −,
this is why MO25 has dominant contribution and also why f − can often be reasonably approximated
by HOMO distribution. It is obvious that if there is no orbital relaxation effect, contribution of
MO25 should be exactly 1.0 while all other MOs should have contribution of 0.0.
It is worth to stress that the range of the orbital considered in the fitting affects the resulting contributions. For
example, you only want to calculate contribution of MOs 10~15, you will find their contributions are different when
MOs 1~15 are chosen and when MOs 10~15 are chosen.
4.200.13.2 Contribution of NBO orbitals to Fukui function f − of 1,3-butadiene

This time we calculate contribution of various highly occupied NBOs to f − of 1,3-butadiene.
The .fch file of N and N-1 states, as well as NBO plot files .31 and .37 generated for N state have
been provided in "examples\orb_densdiff\butadiene" folder. The geometry has been optimized for
N-state.
Note that the NBO orbitals recorded in the .37 file can be divided as two categories, the Lewis
type and Rydberg type. The Lewis NBOs have occupation number of approximately 2.0 while the
Rydberg ones are almost unoccupied. Since only the Lewis NBOs significantly contribute to
electron density of N-state, we will only take these orbitals into account.
First, we generate cube file of f − type of Fukui function. Boot up Multiwfn and input
774
examples\orb_densdiff\butadiene\butadiene.fch
-,examples\orb_densdiff\butadiene\butadiene_N-1.fch
1 // Low quality grid (since the current system is very small, low quality grid is adequate)
Now reboot Multiwfn and input
examples\orb_densdiff\butadiene\BUTADIENE.31
37 // Load the BUTADIENE.37 in the same folder, which records NBO orbitals
Now if you enter main function 6 and choose option 3 to examine orbital information, you will
find the first 15 orbitals correspond to Lewis type of NBOs due to their high occupation numbers.
Next, we input below commands in the main menu
density.cub // The file containing grid data of f −
1-15 // Range of Lewis NBOs
[ignored]
The data clearly shows that NBO4 and NBO9 equally and fully dominate the f −, while
participation of other orbitals can be safely ignored.
The isosurface (isovalue=0.01) of the provided f − and the fitted one, as well as isosurface
(isovalue=0.1) of NBO4 and NBO9 are collectively shown below
775
Since from the graph it is obvious that NBO4 and NBO9 correspond to  bond of C1-C4 and
C6-C8, respectively, and the fitted function is qualitatively consistent with the provided f −, we can
conclude that the f − is mainly composed of the  electrons over the C1-C4 and C6-C8 bonds.
4.200.13.3 Contribution of NBO orbitals to density difference between S0

and S1 states of H2CO
Finally, we study contribution of NBOs to  between S1 and S0 states of H2CO.
All relevant files used in this section have been provided in "examples\orb_densdiff\H2CO"
folder, including the NBO plot files generated for ground state, wavefunction file of ground state
(S0.fch) and wavefunction of the first excited state (S1.wfn). The Gaussian input files used for
generating these files are also provided.
We first generate the  between S1 and S0 states. Boot up Multiwfn and input
examples\orb_densdiff\H2CO\S1.wfn
-,examples\orb_densdiff\H2CO\S0.fch
-1 // Visualize the isosurface
The isosurface at isovalue=0.03 is shown below.
Now we calculate contribution of NBO orbitals to S0→S1 to characterize the nature of S0→S1
transition. Reboot Multiwfn and input below commands
examples\orb_densdiff\H2CO\H2CO.31
37 // Load the H2CO.37 in the same folder, which records NBO orbitals
density.cub // The file containing grid data of S0→S1
1 // Set constraint on the sum of contributions
2 // Set the constraint to a specific value
0 // Since electron excitation does not alter the number of electrons, the sum of contributions
is set to be constrained to zero
[Press ENTER button to consider all orbitals] // Note that during electron excitation, a portion
776
of electrons is excited to empty orbitals, therefore only taking Lewis NBOs into account is evidently
inadequate, so all orbitals should be taken into account in the present context
[ignored]
It can be seen that the NBO having dominant positive value is NBO32. The NBO having the
most negative contribution is NBO8, its magnitude of negative contribution is by far greater than
any other NBO. From the "Natural Bond Orbitals (Summary)" field printed by NBO module in the
Gaussian output file examples\orb_densdiff\H2CO\S0.out, it can be seen that the NBO8 is identified
as " LP ( 2) O 3", namely lone pair of O3.
The isosurface of fitted  (isovalue=0.03) as well as isosurface of NBO8 and NBO32 (both
isovalues are set to 0.17) are shown below.
By comparing the "fitted" graph with the S0→S1 graph shown earlier, it is apparent that the
fitted  is almost completely identical to the rigorously calculated one, hence the fitting quality in
the current instance is fairly high and the contribution values must be very reliable and meaningful.
From the orbital isosurface maps it is obvious that the NBO8 indeed corresponds to lone pair of O3,
while NBO32 corresponds to anti- orbital of C1-O3, therefore the S0→S1 excitation could be
unambiguously identified as n(O3)→*(C1-O3) type of transition.
You can also calculate contribution of various NAOs to the S0→S1 excitation, the only
difference from the above example is that after loading the H2CO.31, you should input 33 to load
the H2CO.33, which records NAO orbitals. You will find the S0→S1 excitation mainly involves
transitions from highly occupied p type NAOs to those nearly unoccupied.
777
4.200.14 Application of domain analysis
The domain analysis refers to quantitative analysis for the region enclosed by isosurface of
given real space function, see Section 3.200.14 for details. To illustrate the powerfulness and flexible
of the domain analysis module, two practical applications of the domain analysis are given below.
4.200.14.1 Integrate real space functions within reduced density gradient
(RDG) isosurface to study weak interaction quantitatively
Before reading this, please read Section 3.23.1 to understand how to use reduced density
gradient (RDG) to reveal weak interaction regions. In this section, I show the possibility of
characterizing weak interaction by integrating domains enclosed by RDG isosurfaces.
System 1: Phenol dimer

First, we use phenol dimer as example. Boot up Multiwfn and input
examples\phenoldimer.wfn
14 // Integrate real space functions within isosurfaces of a real space function
Here we want to study RDG domains defined as regions enclosed by isosurface of RDG=0.5;
in other words, these domains composed of grid points where RDG<0.5. Therefore, we select option
2 and choose "13 Reduced density gradient", and then select option 3 and input criterion, namely
<0.5 (In fact, RDG<0.5 is the default setting and you do not need to manually do these steps). Next,
input below commands:
1 // Start calculation grid data and generate domains
-10 // Adjust extension distance
0 // Set extension distance to zero to avoid wasting of grid points at boundary area, where
RDG isosurfaces commonly do not occur
2 // Medium quality grid (i.e. grid spacing=0.1 Bohr), generally this is accurate enough
Now Multiwfn starts calculation of grid data for the selected real space function (i.e. RDG),
and then identifies individual RDG domains according to the criterion of RDG<0.5. Finally, four
domains are found, the number of grid points constituting the domains are shown as the last column:
Domain: 1 Grids: 208
To visualize them, select "3 Visualize domains". In the GUI, you can select domain index at
right-bottom list. The 2nd and 4th domains are shown below:
If you have read Section 3.23.1, you must know these domains correspond to H-bond and van der
778
waals (vdW) interactions between the two phenols, respectively. We can study properties of these
domains in terms of integrating specific real space functions within corresponding regions. We
select option 1 and input 2 (domain index), then input 1 to select electron density as the function to
be integrated, the result is:
Integration result: 0.6932973049E-02 a.u.
Volume: 0.290000 Bohr^3 ( 0.042974 Angstrom^3 )
Average: 0.2390680362E-01
Maximum: 0.2766231164E-01 Minimum: 0.1994457517E-01
Similarly, we do this for domain 4:

Integration result: 0.1181667175E-01 a.u.
Average: 0.4550123895E-02
Maximum: 0.6217203920E-02 Minimum: 0.3102088666E-02
From the output we know the number of electrons involved in the domains corresponding to
H-bond and vdW interactions are 0.006933 and 0.011817, respectively. They can be interpreted as
overlapping electrons and are closely related to strength of same type of interactions (see discussion
in Section 5.2 of DORI original paper J. Chem. Theory Comput., 10, 3745 (2014)). However, since
these two domains correspond to different type of weak interactions, the magnitude of overlapping
electrons is not positively correlated to their strengths, namely we are unable thus to say that the
vdW interaction between the two phenols is stronger than the intermolecular H-bond. The "Volume"
in the output denotes volume of the domain, we can find that vdW interaction involves much wider
spatial region than H-bond. "Average" correspond to average value of real space function in the
domain, from this quantity one can easily infer that the strength of interaction per contact region of
H-bond must be significantly higher than that of vdW interaction, since as shown above, their
average values are 0.0239 and 0.0045, respectively, the former is much larger than the latter.
System 2: 2-pyridoxine 2-aminopyridine

Intermolecular H-bonds of 2-pyridoxine 2-aminopyridine (PP) has been investigated in Section
4.2.1 by means of AIM analysis, while this time we will analyze them by means of qint index. This
index was proposed in J. Phys. Chem. A, 115, 12983 (2011) for judging interaction of H-bond at
various intermolcular distance, please check Section 3.200.14 for its definition. Commonly, the
more negative of the qint index, the more stable the interaction.
qint index is defined based on integrating domains enclosed by RDG=0.6 isosurfaces, therefore
we need to first calculate RDG grid data and generate corresponding domains. Note that it is not
always appropriate to set extension distance of RDG grid data to zero. For present system, if you
calculate RDG grid data by main function 5 with extension distance of zero, you will see some RDG
isosurfaces are truncated by box boundary, as shown below and highlighted by red arrows. In this
case domain integrating module of Multiwfn does not work.
779
Therefore, when we calculate RDG grid data for this case, extension distance should be set
somewhat larger than zero, 3 Bohr is safe enough for avoiding unexpected truncation. Extension
distance should also never be set to a too large value, otherwise the number of grid points to be
calculated will be very high and thus very time-consuming.
examples\2-pyridoxine_2-aminopyridine.wfn
14 // Integrate real space functions within isosurfaces of a real space function
3 // Change the default criterion of defining domain
<0.6
1 // Start calculation of grid data
-10 // Change extension distance
3 // 3.0 Bohr of extension distance
Now visualize resulting domains. Domains 2 and 4 are shown below, clearly they correspond
to H-bond of N23-H25O1 and N2-H12N13, respectively.
Now select "5 Calculate q_bind index for a domain" and input 2, the resulting qint and related
details of the domain are shown below
q_att: 0.00490957 a.u.
q_rep: 0.00005590 a.u.
q_bind: -0.00485367 a.u.
Volume (lambda2<0): 0.599000 Bohr^3
Volume (lambda2>0): 0.010000 Bohr^3
Volume (Total): 0.609000 Bohr^3
Similarly, we obtain results for domain 4
780
q_att: 0.00805204 a.u.
q_rep: 0.00037346 a.u.
q_bind: -0.00767858 a.u.
Volume (lambda2<0): 0.745000 Bohr^3
Volume (lambda2>0): 0.052000 Bohr^3
Volume (Total): 0.797000 Bohr^3
Since the qbind corresponding to N2-H12N13 (-0.007678) is much more negative than that
of N23-H25O1 (-0.004854), the former should be stronger than the latter.
Since both the interactions are H-bond, it is also possible to simply compare the number of
electrons contained in the domains to estimate their relative strength. We choose "2 Perform
integration for all domains" and then select electron density to print out integral of electron density
in all domains:
Domain Integral (a.u.) Volume (Bohr^3) Average
1 0.1212147834E-03 0.111000 0.1092025076E-02
2 0.1647945811E-01 0.609000 0.2705986553E-01
3 0.1714432107E-02 0.319000 0.5374395319E-02
4 0.2618188584E-01 0.797000 0.3285054685E-01
5 0.8932587581E-02 0.412000 0.2168103782E-01
6 0.7764521413E-02 0.394000 0.1970690714E-01
Integration result of all domains: 0.6119409984E-01 a.u.
Volume of all domains: 2.642000 Bohr^3 0.391504 Angstrom^3
Not only the integration value of domain 2 (0.01648) is evidently smaller than domain 4
(0.02618), but also the average value of domain 2 (0.02706) is smaller than domain 4 (0.03285),
therefore we have strong evidence to say N2-H12N13 is stronger than N23-H25O1.
With similar steps illustrated in this section, you can also integrate other real space functions
such as potential energy density and spin density in the domains enclosed by isosurfaces of other
real space functions, e.g. DORI and ELF.
Note that the accuracy of integration in the domains is directly determined by grid setting,
higher quality of grid leads to better accuracy. For example, when visualizing domains, if you found
a domain only consists of very few grid points, and its profile is very coarse, then integration
accuracy of this domain must be very low.
4.200.14.2 Visualize molecular cavity and calculate its volume by domain

analysis module
This is an application instance of domain analysis module, I will briefly illustrate how to use
Multiwfn to visualize molecular cavity and calculate cavity volume. More information about this
topic can be found in my blog article http://sobereva.com/408 (In Chinese).
The idea of using domain analysis to study molecular cavity is very simple: We first calculate
promolecular density, then we define the regions having electron density lower than a threshold e.g.
0.0001 as molecular cavity. More than one such regions may exist, the domain analysis module
automatically assigns them with different domain indices. After that, by visualizing the domains, it
should be easy to find the domain corresponding to the molecular cavity. The choice of threshold is
somewhat arbitrary, commonly 0.001~0.0001 a.u. is appropriate.
781
-cyclodextrin is used as an example in this section. Before studying the cavity using domain
analysis module, it is suggested to first visualize promolecular density under various isovalues. Boot
up Multiwfn and input:
examples\alpha-cyclodextrin.pdb
1 // Promolecular density
-10 // Set extension distance
0 // Zero extension distance, namely let the box just enclose the molecule
-1 // Show isosurface map
Click "Show data range" in the GUI window to show the box of grid data as blue frame, and
set isovalue to 0.01 and 0.001 respectively, you will see
It is easy to understand, if we use threshold of 0.001 a.u., then the domain corresponding to the
cavity in the center of the molecule cannot be defined, because the internal region and external
region are connected via the three channels pointed by red arrows. While in the case of 0.0001 a.u.,
the molecular cavity is clearly identificable and thus we could use domain analysis module with this
threshold to study the cavity.
Return to main menu, and then input below commands
14 // Domain analysis
2 // Choose the real space function to be calculated and used for partitioning domains
1 // Promolecular density
3 // Define the rule of determining domains
<0.0001 // Regions with electron density less than 0.0001 will be defined as domains
1 // Calculate grid data and assign domains
-10 // Change extension distance
0 // No extension distance
After calculation is finished, you will see below information from screen. There are totally six
domains found, the number of grids in each domain is also shown
Domain: 1 Grids: 4
782
Then we can visualize each domain by option 3. After examining each domain, we find that
the domain with index of 5 is the one corresponding to molecular cavity, as shown below:
Since this domain well represents shape of actual molecular cavity, its volume is a good indicator
of cavity size.
Close the GUI, select option 1 to carry out integration within domain, and input 4 to choose to
integrate the domain corresponding to the cavity. Since currently we are only interested in the
volume of the domain, we choose "100 User-defined real space function" as integrand (By default,
this function is 1.0 everywhere and thus does not take any computational cost), then below
information are printed on screen
Integration result: 0.3172320000E+03 a.u.
Average: 0.1000000000E+01
Maximum: 0.1000000000E+01 Minimum: 0.1000000000E+01
Position statistics for coordinates of domain points (Angstrom):
X minimum: -2.7823 X maximum: 3.1445 Span: 5.9268
Y minimum: -2.6823 Y maximum: 2.6095 Span: 5.2918
Z minimum: -3.1770 Z maximum: 3.9140 Span: 7.0910
From the output we find the volume of this domain, namely the size of the cavity, is 47.0 Å3. The
length of the cavity is 7.09 Å (i.e. the span distance between the domain point with maximum Z and
that with minimum Z coordinate).
783
Furthermore, we can export selected domain as domain.cub in current folder by using option
10, so that the domain can be portrayed as isosurface in third-part visualization tools, such as VMD.
In the resulting domain.cub, the grids within the selected domain have value of 1, while grids in
other regions (as well as boundary grids) have value of 0, therefore isosurface of domain can be
rendered with isovalue between 0 and 1 (commonly 0.5 is used). We do this for the 4th domain and
plot is as isosurface in VMD, below graph will be obtained
Alternatively, we can use option 11 to export boundary grids of selected domain to domain.pdb,
in which each particle corresponds to a boundary grid. You can load this file into VMD and render
the particles as spheres. Then if you want to measure domain, you can click keyboard button 2, then
click two spheres in the graphical window, the linking line and distance between the two selected
spheres will be shown, as illustrated below ("Display" - "Orthographic" was selected for easier
inspection)
784
4.200.18 Studying bond length/order alternation (BLA/BOA) as well as

alteration of bond angle and dihedral for specific paths
Note: Chinese version of this section is http://sobereva.com/501, which contains more discussion.
The basic knowledge about definition and calculation of bond length alternation (BLA) and
bond order alternation (BOA) have been introduced in Section 3.200.18. Multiwfn can not only
calculate BLA and BOA, but can also calculate variation of bond length, bond order, bond angle
and dihedral along a given path. This function is quite useful in studying characteristics of
conjugated chains. In this section two examples will be given.
In this function, any file format that carries geometry information can be used as input file,
such as .xyz, .pdb and .mol. However, if you also want to study BOA, the input file must contain
basis function information, such as .mwfn, .fch, .molden or .gms.
4.200.18.1 BLA and BOA of thiophene oligomer

In this section, we will calculate BLA and BOA for a thiophene oligomer with 5 repeat units,
the .fchk file generated at PBE0/6-31G* level can be downloaded here:
http://sobereva.com/multiwfn/extrafiles/TP5.zip. The geometry was optimized at B3LYP/6-31G*
level.
Before calculation, you need to first determine indices of the atoms in the conjugated chain of
interest. The easiest way of doing this is using GaussView. Now we use GaussView (version  6.0)
to open the TP5.fchk, choose button in the "Builder" panel, then hold down the left mouse
button and let the cursor pass through each atom in the conjugated chain to select them as yellow.
After that, the image in the GaussView window should like this:
Then enter "Tools" - "Atom Selection", you will find the atom indices of the selected chain is
10,12,14,16-17,19,21,23-24,26,28,30-31,33,35. Also, as can be seen from the above figure, the
index of the atom at the beginning side and ending side is 1 and 35, respectively.
Note: Of course, it is not absolutely necessary to use GaussView for the present function. However, without
GaussView, you have to manually record indices of all atoms in the chain by means of visual inspection, obviously
this process is fairly cumbersome!

TP5.fchk
18 // Calculate bond length/order alternation (BLA/BOA) and study variation of bond
characteristics with respect to bond index
10,12,14,16-17,19,21,23-24,26,28,30-31,33,35 // The indices of the atoms in the chain
785
1,35 // Index of the atom at the beginning side and ending side
Then Multiwfn automatically identifies the atom sequence of the chain based on your inputted
information. As can be seen from screen, the identified sequence is
Sequence of the atoms in the chain from the beginning side to the ending side
1 2 3 4 9 10 12 14 16
17 19 21 23 24 26 28 30 31
33 35
If you compare the sequence with the graph of molecular structure, you will find the sequence is
completely correct, therefore the subsequent data should be meaningful.
Next, the bond index, indices of the two atoms composing the bond, bond length and bond
order are given:
Bond Atom1 Atom2 Length (Angstrom) Mayer bond order
1 1 2 1.3678 1.6303
2 2 3 1.4232 1.2924
3 3 4 1.3795 1.5445
4 4 9 1.4468 1.0985
5 9 10 1.3799 1.5273
[ignored...]
Finally, some statistical data as well as BLA and BOA are shown:
The number of even bonds: 9
The number of odd bonds: 10
Average length of even bonds: 1.4300 Angstrom
Average length of odd bonds: 1.3779 Angstrom
Bond length alternation (BLA): 0.0521 Angstrom
Average bond order of even bonds: 1.2109
Average bond order of odd bonds: 1.5465
Bond order alternation (BOA): -0.3356
As mentioned on the screen, the bond data have also been exported to bondalter.txt in current
folder, you can plot "bond length vs. bond index" and "bond order vs. bond index" curve maps. The
below map was plotted by Origin, the corresponding .opj file as been provided as bondalter.opj in
"examples" folder.
786
1.48 1.7
1.46 1.6
1.44 1.5
Mayer bond order

Bond length (Å)
1.4
1.42
1.3
1.40
1.2
1.38
1.1
1.36
1.0
0 2 4 6 8 10 12 14 16 18 20
Bond index
Multiwfn also asks you if outputting variation of bond angle and dihedral along the atom
sequence, we input n because they are not what we are currently interested in.
4.200.18.2 Study variation of bond lengths, bond angles and dihedrals in the
ring of cyclo[18]carbon
The below map is a frame of ab-initio molecular dynamic trajectory of the cyclo[18]carbon at
200 K. This simulation was conducted in my research article ChemRxiv (2019) DOI:
10.26434/chemrxiv.11320130. In this section we will study variation of bond lengths, bond angles
and dihedrals along the ring.

examples\C18_MD.xyz // A frame extracted from molecular dynamics trajectory
18
1-18 // The atom indices in the ring
1,1 // The path under study is a closed path, i.e. a ring, in this case the two inputted atom
indices must be the same. Index of any atom in the ring could be inputted, it will be regarded as the
beginning atom
The outputted bond length variation is shown below
Bond Atom1 Atom2 Length (Angstrom)
1 1 2 1.208
787
2 2 3 1.375
3 3 4 1.209
[...ignored]
17 17 18 1.201
18 18 1 1.384
Next, we input y to let Multiwfn output variation of bond angles and dihedrals along the defined
path, you will see:
Note The unit of printed values is degree
Atoms: 1 2 3 Angle: 164.841
Atoms: 2 3 4 Angle: 154.976
Atoms: 3 4 5 Angle: 155.200
[ignored]
Atoms: 17 18 1 Angle: 162.906
Atoms: 18 1 2 Angle: 158.520
Atoms: 1 2 3 4 Dihedral: 28.69, deviation to planar: 28.69
[...ignored]
From the above output we can easily examine how bond angles and dihedrals vary along the
18-membered ring. Note that the value range of dihedral (D) is 0~180, the "deviation to planar" is
identical to D if the D is within 0~90, while it corresponds to 180−D if the D is within 90~180.
If you copy the data from screen (see Section 5.4 if you do not know how to do this) and plot
them via external software such as Origin, you can obtain below maps, which very clearly exhibit
geometric characteristics of the ring:
Since bond angles and dihedrals fluctuate evidently along the ring, we can conclude that the ring
undergoes prominent geometry deformation during the molecular dynamics simulation.
788
4.200.20 Using bond order density and natural adaptive orbital to study
chemical bonds
Note: Chinese version of this topic is "Using bond order density (BOD) and natural adaptive orbital (NAdO) to
graphically study chemical bonds" (http://sobereva.com/535), which contains much more examples and fuller
discussions.
The theory of bond order density (BOD) and natural adaptive orbital (NAdO) has been
detailedly introduced in Section 3.200.20, please carefully read it first. In this section I will present
two examples to show how to use the BOD and NAdO to study chemical bonds.
4.200.20.1 Plot bond order density for N2 molecule
The delocalization index (DI) represents average number of electron pairs shared by two atoms
and can be regarded as a definition of (covalent) bond order. By plotting BOD, we can better
understand the nature of its value. The integral of the BOD defined for two atoms over the whole
space exactly corresponds to the DI between the two atoms, therefore BOD is able to exhibit local
contributions everywhere to DI.
N2 molecule is taken as example in the present section, we will plot its BOD as color-filled
map in the molecular plane. In Multiwfn, delocalization index (DI) can be calculated based on fuzzy
partition via fuzzy atomic space analysis module (main function 15) or based on atom-in-molecules
(AIM) partition via basin analysis module (main function 17); correspondingly, both the two
modules can export atomic overlap matrix (AOM), which is needed by BOD analysis. In the present
example, we use the former (the latter works equally well but more expensive).
examples\N2.fch // Optimized and generated at B3LYP/def-TZVP level. You can also use
other files (e.g. .molden and .mwfn) as long as the file contains basis function information
3 // Calculate and output atomic overlap matrix to AOM.txt in current folder
20 // Bond order density (BOD) and natural adaptive orbital (NAdO) analyses
1 // Use atomic overlap matrix (AOM) for the analysis
[Press ENTER button] // Load the AOM.txt in current folder
1,2 // Indices of the two atoms to be analyzed
Then NAdOs are generated and you can find below information
Generating natural adaptive orbitals (NAdOs)...
Eigenvalues of NAdOs: (sum= 3.11681 )
1.00000 0.99999 0.99999 0.05728 0.05728 0.00113 0.00113
The values are eigenvalues of the NAdO orbitals, the sum (3.11681) just corresponds to the DI
between the two atoms. At the same time, NAdOs.mwfn is generated in current folder, in which the
first 7 orbitals are the NAdO orbitals, their occupation numbers correspond to the NAdO
eigenvalues. All other orbitals in this file are identical to the virtual MOs in the N2.fch.
Next, we input y to let Multiwfn load the newly generated NAdOs.mwfn. From now on, electron
density function directly corresponds to the BOD function. Now we plot the BOD as color-filled
map. Input below commands
789

4 // Plot plane map
1 // Electron density (which corresponds to BOD in the present context)
[Press ENTER button] // Use recommended grid setting
0 // Set extension distance
2 // 2 Bohr
3 // YZ plane
0 // Z=0
Now the BOD map pops up. After some adjustments on plotting settings, you can see below
map
It can be seen that the distribution of BOD is reasonable, its main body is distributed in the
bonding area, showing that the electrons in this region has major contribution to the DI value and
play crucial role in the covalent bond.
4.200.20.2 Study BOD and NAdO orbitals for C-C bonds in butadiene
In this section, we will study BOD and NAdO orbitals for two kinds of C-C bonds in 1,3-
butadiene, whose geometry is shown below. This time the analysis will be performed based on the
AOM generated by AIM partition (using fuzzy partition like the last example is also reasonable).
790

examples\butadiene.fch // Generated at B3LYP/6-31G** level
17 // Basin analysis module
1 // Use electron density to define basins (i.e. AIM basins)
6 // Output orbital overlap matrix in atoms to AOM.txt in current folder
4,6 // Indices of the two carbons at the center of the system
Now you can see
Eigenvalues of NAdOs: (sum= 1.11455 )
0.90121 0.24533 0.07666 0.00358 0.00062 0.00006 0.00001
0.00000 -0.00000 -0.00003 -0.00004 -0.00316 -0.00966 -0.02645
-0.07359
That means the DI of the central C-C bond is 1.114. Then input y to load the newly generated
NAdOs.mwfn file. After that, we use main function 5 to calculate grid data of electron density, which
now corresponds to BOD, and then plot it as isosurface. The resulting map with isovalue of 0.05 is
shown below.
It is obvious that most electrons that have significant contribution to the C4-C6 bond are
concentrated around this bond, this is what we expected.
Next, enter main function 0 to visualize various NAdO orbitals. Note that if you select "Orbital
info." - "Show all" in the GUI menu, you can see below orbital information
Orb: 1 Ene(au/eV): 0.000000 0.0000 Occ: 0.901215 Type:A+B (? )
[...ignored]
Orb: 14 Ene(au/eV): 0.000000 0.0000 Occ:-0.026445 Type:A+B (? )
Orb: 15 Ene(au/eV): 0.000000 0.0000 Occ:-0.073586 Type:A+B (? )
Orb: 16 Ene(au/eV): -0.023511 -0.6398 Occ: 0.000000 Type:A+B (? )
791
[...ignored]
The first 15 orbitals are NAdOs, their "Occ" values correspond to NAdO eigenvalues, which
can be regarded as contribution of the NAdOs to DI. All orbitals with index larger than 15 are the
virtual MOs in the examples\butadiene.fch, they are not of our interest. From the above list, it can
be seen that only the first two NAdO orbitals have prominent contribution to DI, their isosurface
maps are shown below, the eigenvalues are also labelled.
It can be seen that the first NAdO, which looks like a  type of localized orbital (see example
in Section 4.19.1 for more information about this point), has key contribution (0.901) to the DI value
(1.114); this is fully understandable, since the major ingredient of the C4-C6 bond must be 
interaction. The second NAdO also has nonnegligible contribution (0.245). Since it shows typical 
orbital character, we can infer that weak  interaction exists in the C4-C6 bond.
Now we turn our attention to the boundary C-C bonds, namely C1-C4 (or C6-C8). Reboot
Multiwfn and input below commands (you can first manually backup the previous NAdOs.mwfn to
avoid overwriting)
1,4 // Indices of the two carbons at the boundary of the system
y // Load the newly generated NAdOs.mwfn
After that, use main function 0 to visualize the only two orbitals having significant
contributions to DI, as shown below (isovalue=0.05)
The  type of NAdO orbital of C1-C4 has comparable eigenvalue to that of C4-C6, indicating
that both two kinds of C-C bonds has similar strength of  interaction. In contrast, the  type of
NAdO orbital of C1-C4 has much higher contribution to DI than that of C4-C6, well reflecting the
792
fact that the boundary C-C bonds have much stronger  interaction than the central one.
At the end of this section, I would like to emphasize a quite noteworthy point of the
BOD/NAdO method, namely you can directly specify the bond to be studied, as already fully
illustrated above. For example, for the dopamine shown below, if you want to study BOD or
corresponding NAdOs for a bond, e.g. C5-C7, you simply need to input 5,7 when Multiwfn asks
you to input atom indices, clearly this feature makes bonding analysis quite convenient! Below map
shows BOD=0.03 isosurface for C5-C7.
Also it is noteworthy that the BOD/NAdO analysis can also be performed to visualize DI
between two basins based on the BOM.txt exported by basin analysis module. For example, you can
use basin analysis module to generate electron localization function (ELF) basins and then export
basin overlap matrix (BOM) to BOM.txt by corresponding option. After that, in the BOD/NAdO
analysis function, select to load the BOM.txt, and then input index of two basins of interest, the
NAdOs for the two local regions will be generated. Via this manner, you can visually study e.g.
electron sharing between two lone pair regions or between a lone pair and a covalent bonding region.
Clearly, the BOD/NAdO analysis module is extremely flexible!
4.300.1 Visualizing free regions and calculating free volume for a

porous system
Note: Chinese version of this topic is http://sobereva.com/539, which contains more discussions.
Please check Section 3.300.1 for basic information and algorithm used in the function
illustrated in this section. Below I will exemplify how to use Multiwfn to view free regions (i.e.
pores or cavities) in a frame produced by molecular dynamics simulation under periodic boundary
condition. The method illustrated in this section can also be used for examining free regions in
experimentally determined molecular crystal.
Below we take examples\coal.pdb as instance, which is a frame of molecular dynamics
simulation conducted by LAMMPS program for coal:
793
Note that if you open this file by text editor, you can find the following line:
CRYST1 31.064 31.100 31.093 90.00 90.00 90.00 P 1 1
Meaning that box length in X, Y, Z is 31.064, 31.100, 31.093 Å, respectively.

examples\coal.pdb
1 // Viewing free regions and calculating free volume in a box
1 // Set grid and start calculation
[Press ENTER button directly] // Use default grid spacing (0.3 Å), which is fine enough
Then you can find following information from screen
Box lengths are directly loaded from "CRYST1" field from input file
Box length in X, Y, Z: 31.064 31.100 31.093 Angstrom
Origin in X,Y,Z is 0.000 0.000 0.000 Angstrom
Grid spacing in X,Y,Z is 0.302 0.302 0.302 Angstrom
The number of points in X,Y,Z is 104 104 104 Total: 1124864
As you can see, since "CRYST1" field is present in the inputted .pdb file, the box lengths are
automatically read and utilized in the calculation (When this information is not present, Multiwfn
will ask you to manually input box lengths).
Once the calculation of grid data is complete, you will find information about size of free
regions:
Volume of entire box: 30038.649 Angstrom^3
Free volume: 14787.821 Angstrom^3, corresponding to 49.23 % of whole space
This output shows that about half of the whole box is not lying within the vdW surface of the system.
In the newly appeared menu, select option 3 to visualize isosurface of smoothed grid dat, then
after selecting "Show molecule" and "Show data range" check boxes in the GUI, you will see
molecular structure along with the isosurfaces representing free regions:
794
It can be seen from the above map that the actual free regions are very vividly and clearly
revealed by the isosurfaces. The default isovalue is 0.5. If you increase it, the isosurfaces will shrink
and only significant pores will be visible. In contrast, if you decrease isovalue, existing isosurfaces
will inflate and more insignificant pores will appear in the graph.
In the post-processing menu you can also export the grid data as .cub file, so that you can
render it via third-part visualization tools, such as VMD and ChimeraX. You can also find option
used to visualize isosurface of primitive grid data, however, as you will see, this isosurface can
hardly used to exhibit pore character in the present system, rendering importance of the smoothing
algorithm employed in Multiwfn, see Section 3.300.1 for detail of the grid data calculation algorithm.
It is worth to note that as shown in Section 4.200.14.2, domain analysis module of Multiwfn is
also able to visualize cavity and calculate cavity volume, however this module is only suitable for
investigating internal cavity of a single molecule, it cannot be used to study all pores in a large box
like the present section.
4.300.2 Example of fitting radial atomic density as STOs or GTFs
Multiwfn is able to fit spherically averaged electron density of an isolated atom as multiple
Slater type orbitals (STOs) or Gaussian type functions (GTFs), please read Section 3.300.2 to gain
basic knowledges. In this section, I will illustrate how to use this function to realize the fitting.
4.300.2.1 Crudely fitting radial density of silicon as several STOs
In this section we will fit radial density of silicon atom as linear combination of a few STOs.
Since the number of fitting functions is small, the fitting procedure is rapid and evaluation of fitted
density is quite expensive, however, the fitting quality is not expected to be very high.
examples\atomwfn\Si.wfn // Generated at ROHF/6-31G* level
2 // Fitting atomic radial density as multiple STOs or GTFs
795
3 // Check or set initial guess of coefficients and exponents of fitting functions

2 // Set initial guess as "crude fitting by a few STOs with variable exponents". Then from
screen you can find four STOs will be employed in the fitting, their initial status are
Coefficient Exponent
STO 1: 1.000000E+03 2.700000E+01
STO 2: 3.000000E+02 9.000000E+00
STO 3: 2.000000E+01 3.000000E+00
STO 4: 1.000000E+00 1.000000E+00
The initial parametes look reasonable. Then input below commands

0 // Return to upper level of menu
1 // Start fitting
By default, 4000 evenly distributed points with spacing of 0.001 Å are used for fitting, clearly
they cover radial range of r = 0~4 Å. If you have carefully read Section 3.300.2, you will find the
outputted information during the fitting is quite easy to understand. The second half of the output is
shown below
Integral of fitted density calculated using 100 points: 13.87020162
Fitted coefficients are scaled by 1.00935807
Fitted parameters (a.u.) after scaling:
Coefficient Exponent
STO 1: 5.473079E+00 2.030994E+00
STO 2: -3.644214E+02 3.564418E+00
STO 3: 4.085449E+02 3.727240E+00
STO 4: 2.166949E+03 3.290713E+01
RMSE of fitting error at all points: 15.688409 a.u.^2
Pearson correlation coefficient r: 0.997997 r^2: 0.995998
As you can see, the integral of the originally fitted density over the whole space is 13.8702,
therefore the coefficients of the fitting functions are scaled by 14/13.8702=1.009358, where 14 is
the actual number of electrons of silicon. In the current fitting, both coefficients and exponents of
the four STOs are optimized, the final parameters are printed under "Fitted parameters (a.u.) after
scaling" title. The RMSE is a quantity useful in quantitatively measuring fitting quality. The r2
coefficient between fitted density and actual density is as high as 0.995, implying that the fitting is
reasonable; however, it is highly suggested also employing other ways to further examine the fitting
quality and confirm the fitting reliability, so that the fitted parameters can be safely used in practical
studies to estimate density.
In the newly appeared menu you can see many options, whose meanings are either self-
explanatory or have been described in Section 3.300.2.2. To quantitatively check fitting quality at
the 4000 fitting points, we select option 1, then you will see
Radial distance (Angstrom), actual density (a.u.), difference between fitted an
d actual density (a.u.) as well as relative difference
# 1 r: 0.00100 rho: 1632.43876137 Diff: 453.01616705 ( 27.75 %)
# 2 r: 0.00200 rho: 1580.36293050 Diff: 381.88236744 ( 24.16 %)
# 3 r: 0.00300 rho: 1507.61802448 Diff: 338.82345452 ( 22.47 %)
796
# 4 r: 0.00400 rho: 1427.66455905 Diff: 309.93260095 ( 21.71 %)
[ignored...]
# 3997 r: 3.99700 rho: 0.00000506 Diff: -0.00000387 ( -76.49 %)
# 3998 r: 3.99800 rho: 0.00000504 Diff: -0.00000386 ( -76.48 %)
# 3999 r: 3.99900 rho: 0.00000502 Diff: -0.00000384 ( -76.48 %)
# 4000 r: 4.00000 rho: 0.00000500 Diff: -0.00000383 ( -76.48 %)
The "Diff" is difference between fitted density and actual density, the value in the parenthese is
relative error. As can be seen, the error in the region very close to nucleus is not small, however it
does not matter since this region is not of chemical interest, usually chemists mainly focus on
electron density in valence region. The relative error in the region very far from nucleus is
significant (as large as 76%), it does not matter too since the density in this region is quite low and
thus has negligible influence on most quantities involved in common studies.
Next, we choose option 3 to visually inspect curves of fitted density and actual density using
logarithmic scaling, then you will see
The fitting is evidently successful, since the fitted density curve is close to actual density curve in
the region within 1.8 Å. Athough the two curves diverge detectably from each other in longer range,
the density in that region is quite low (much less than 0.01 a.u.) and thus this is not a big problem.
It is worth to mention that Bondi vdW radius of silicon is 2.1 Å.
You can also select option 4 to plot comparison map between fitted density and actual density
using linear scaling, you will further find that the fitted density indeed nicely reproduces the actual
density.
If you select option 5, then fitdens.txt will be outputted in current folder. This file contains
fitted density (the second column) at points evenly distributed from 0 to 10 Å, the grid spacing is
half of the fitting points, namely 0.001/2=0.0005 Å. From the data in this file you can find that the
fitted density varies smoothly and monotonically, no negative value can be found, further implying
that the fitting is successful and the fitted parameters are reliable.
Finally, we select option 6 to examine integral of fitted density over the whole space via
different number of points of Gaussian quadrature, you will see:
797
Number of integration points: 40 Integral: 13.99999927
[ignored...]
As can be seen, in all cases the integral is almost exactly identical to the actual number of electrons
(14), further demonstrating that our fitting is successful.
Since our fitted density has passed quality check in many ways, we can finally conclude that
the fitted parameters of the four STOs can be safely and reliably employed in future researches.
4.300.2.2 Accurately fitting radial density of bromine as many GTFs

In order to reliably and exactly fit radial density, commonly dozens of GTFs with fixed
exponents should be employed. In this example we will fit radial density of bromine atom in this
way. This kind of fitting is suitable for any element in the periodic table. In contrast, the method
illustrated in the last section does not work well for elements after the third row in periodic table,
the fitting quality is often unsatisfactory even if the exponents of the STOs are allowed to be variable,
the reason is that the curve of actual density of very heavy elements is much more complicated than
that of light elements because they have many atomic orbital shells.
examples\atomwfn\Br.wfn // Generated at ROHF/6-31G* level
2 // Fitting atomic radial density as STOs or GTFs
3 // Check or set initial guess of coefficients and exponents of fitting functions
3 // Set initial guess as "fine fitting by 30 GTFs with fixed exponents"
As can be seen from screen, 30 GTFs will be employed in the fitting, all coefficients are initially
set to 1.0, while their exponents span large range, the lowest one is 0.05, while the largest one is
2.68E7, the ratio between two neighbouring GTFs is 2.0. The GTFs with small, medium and large
exponents are mainly used to represent tail region, valence region and the region very close to
nucleus, respectively.
Then input 0 to return to upper level of menu and then choose option 1 to start fitting. During
fitting, you can find 8 redundant fitting functions are automatically eliminated to avoid numerical
problems, one of them has very small exponent (0.05), while others have very large exponents:
Delete redundant function (coeff= 1.00000E+00 exp= 3.35544E+06), refitting...
Negative fitted density is found at r= 0.0000 Angstrom
Delete redundant function (coeff=-8.53319E+11 exp= 1.67772E+06), refitting...
798
Delete redundant function (coeff=-3.59820E-04 exp= 5.00000E-02), refitting...
Totally 8 redundant fitting functions have been eliminated
Then you can find error statistics:

RMSE of fitting error at all points: 2.124191 a.u.^2
Pearson correlation coefficient r: 0.999997 r^2: 0.999994
From this data we can find the fitting quality is almost perfect! The r2 is almost exactly 1.0!
Please use the same way as illustrated in the last section to examine fitting quality, you will
find current fitting is also completely successful. For example, after choosing option 3 we can see
the following map, which exhibits that the fitting within r=2.5 Å is perfect. Although the difference
is detectable in longer range, the absolute difference is smaller than 4E-5 a.u. (you can check this
point using option 1), which is fully negligible.
Clearly, the fitting procedure illustrated in this section is quite ideal when you want to reach
very high fitting quality.
This fitting module in Multiwfn is quite flexible, there are many options used to control fitting
strategy, see Section 3.300.2 for more information.
4.300.3 Example of using unit sphere representation to visually study

(hyper)polarizability
Note: Chinese version of this section is http://sobereva.com/547, which contains more discussions.
Please read Section 3.300.3 if you are not familiar with unit sphere and vector representation
analysis of (hyper)polarizability. In this section, I will take CH3NHCHO and cyclo[18]carbon as
instances to show how to use Multiwfn in conjunction with VMD visualization software to realize
these kinds of analysis and to demonstrate the usefulness of these methods.
799
4.300.3.1 First-order hyperpolarizability of CH3NHCHO

In this section, we analyze SHG (second-harmonic generation) type of dynamic first
hyperpolarizability () at 1030 nm for CH3NHCHO, one of our purposes is to reproduce the Fig.
2(e) in the original paper of the unit sphere representation method (J. Comput. Chem., 32, 1128
(2011)). We will use the same calculation level as the authors, namely B3LYP/6-311+G** for
geometry optimization and HF/6-311++G** for hyperpolarizability calculation. The authors
employed GAMESS-US program but we will employ Gaussian to calculate .
First, use Gaussian to run examples\polar\CH3NHCHO\polar.gjf, its content is shown below

#P HF/6-311++g(d,p) polar=DCSHG CPHF=rdfreq
[blank line]
B3LYP/6-311++G** opted
[blank line]
0 1
[coordinate optimized at B3LYP/6-311++G** level]
[blank line]
450nm 1030nm
In this input file, polar=DCSHG requests Gaussian to calculate hyperpolarizability of SHG type,
namely (-2;,). Two frequencies of incident light are loaded from the end of input file, as
requested by CPHF=rdfreq keyword. Note that #P must be employed, otherwise Multiwfn will be
unable to parse (hyper)polarizability from the output file.
Now we use Multiwfn to parse the output file and export  as .txt file. Boot up Multiwfn and
input below commands
examples\polar\CH3NHCHO\polar.out
200 // Other function (Part 2)
7 // Parse "polar" task of Gaussian. PS: If you are not familiar with this function, please check
Section 3.200.7 for introduction and 4.200.7 for example
-1 // Request Multiwfn to parse dynamic (hyper)polarizability
-4 // Request Multiwfn to export parsed (hyper)polarizability as .txt file
1 // Start parsing (hyper)polarizability
2 // As shown on screen, the second option corresponds to 1030 nm case
2 // Load SHG form of 
n // Do not perform analysis related to hyper-Rayleigh scattering
Now polarizability tensor () and SHG form of  corresponding to 1030 nm have been
exported to alpha.txt and beta.txt in current folder, respectively.
As mentioned in Section 3.300.3, to realize unit sphere representation, generally you should
let Multiwfn load a file containing atom coordinate, so that Multiwfn can determine proper radius
of the sphere. Here we let Multiwfn directly load atom coordinate from Gaussian output file. To do
so, we change "iloadGaugeom" in settings.ini to 2, that means requesting Multiwfn to load atom
coordinate in standard orientation from the loaded Gaussian output file. Then boot up Multiwfn and
input
examples\polar\CH3NHCHO\polar.out
3 // Visualize (hyper)polarizability via unit sphere and vector representations
800
Now you can find many options used to adjust plotting parameters, such as radius and length
of the arrows on sphere, number of arrows and so on, currently we use default setting. We select
option 2, from prompt on screen you can find Multiwfn automatically loads  tensor from beta.txt
in current folder because it exists, and then export beta.tcl in current folder, which corresponds to
VMD plotting script of unit sphere representation. You can also find beta_vec.tcl has been exported
in current folder, which is VMD plotting script of vector representation.
Since we also want to display molecular structure in VMD, we need to generate a file
containing atom information that can be recognized by VMD, therefore we input
0 // Exit current function
2 // Generate new file
1 // Export current geometry as .pdb file
CH3NHCHO.pdb
Now we have CH3NHCHO.pdb in current folder, and we can close Multiwfn program.
Moving the beta.tcl and beta_vec.tcl from current folder to VMD installation folder, then boot
up VMD and input source beta.tcl and source beta_vec.tcl in VMD console window to run these
two plotting scripts in turn. Next, drag CH3NHCHO.pdb to "VMD Main" window to load it, then
enter "Graphics" - "Representation" and change "Drawing Method" to "CPK". Now you can see
below figure in VMD graphical window, two side views are given:
You can find this map is almost exactly identical to Fig. 2(e) in J. Comput. Chem., 32, 1128 (2011),
the marginal differences come from numerical aspects and the fact that definition of B3LYP in
Gaussian is slightly different to that in GAMESS-US. In this map, the arrows on sphere correspond
to scaled eff vectors, the arrow direction is in line with eff, and the length equals to norm of eff
multiplied by scale factor. The arrows are colored according to its length, the shortest (longest)
arrow is colored as blue and red, respectively, the white arrows have medium length. The starting
points of the arrows are evenly distributed on a sphere surface, the sphere radius can be controlled
by corresponding option in Multiwfn.
What can we learn from above maps? In order to elucidate this point, three featured zones are
labelled. The pink arrows are normal vectors at corresponding point of the sphere surface.
801
• Region 1: If two external electric fields are applied to the molecule along the direction
indicated by the pink arrow, their combination effect will result in occurrence of induced dipole
moment in the same direction, as exhibited by the small arrows on the sphere.
• Region 2: Similar to region 1, but the induced dipole moment occurs in the inverted direction
as applied electric field.
• Region 3: If two external electric fields are applied from top to bottom, as illustrated by the
pink arrow, their combination effect will lead to induced dipole moment pointing to the right. This
seemingly weird phenomenon reflects anisotropic response character of this molecule. Clearly,
without employing the unit sphere representation method, this point can hardly been noticed.
Since the  in the present example is SHG type corresponding to 1030 nm, the external electric
fields mentioned above in fact vary with frequency of 1030 nm, they source from 1030 nm light
radiating in the direction perpendicular to them.
The large green arrow in above map is referred to as vector representation, it corresponds to
scaled (x, y, z) vector and exhibits primary character of the . It can be seen that its direction is
basically identical to the vector sum of all arrows on the sphere. Undoubtedly, this vector
representation is concise and useful in representing , however, anisotropy character of  is fully
ignored.
4.300.3.2 Polarizability and second-order hyperpolarizability of

cyclo[18]carbon
The cyclo[18]carbon is an intriguing system having unusual electronic structure, its various
characteristics, including (hyper)polarizability, have been very comprehensively explored in my
work Carbon, 165, 461 (2020) and Carbon, 165, 468 (2020). In this section, we employ unit sphere
representation to visually investigate its polarizability and second-order hyperpolarizability ().
The Gaussian input file for calculating (hyper)polarizability, including , has been provided as
examples\polar\C18\gamma.gjf, which utilizes LPol-ds basis set, the basis set file can be obtained
from http://sobereva.com/345. In this example, we only study static  and , therefore 0.0 is
specified after molecular coordinate as frequency of incident light. The geometry has been
optimized at B97XD/def2-TZVP level.
We first extract  and  from Gaussian output file and write it as .txt file. Boot up Multiwfn
and input below commands
802
examples\polar\C18\gamma.out // Output file of aforementioned input file

7 // Parse "polar" task of Gaussian
-4 // Request Multiwfn to export parsed (hyper)polarizability as .txt file
7 // Start parsing  and 
Now we have alpha.txt and gamma.txt in current folder. Then we input
0 // Exit current function
3 // Visualize (hyper)polarizability via unit sphere and vector representations
-3 // Change scale factor of length of the arrows on sphere surface
0.005 // This value is smaller than default, since  of cyclo[18]carbon is fairly large. If default
value is used, you will find the arrows are too long
1 // Perform unit sphere representation analysis of . Since alpha.txt already exists in current
folder,  tensor is automatically loaded from it
Now we have alpha.tcl file in current folder, which is the VMD plotting script of unit sphere
representation of . Move it to VMD folder, boot up VMD and input source alpha.tcl in VMD
console window to run it.
In order to show molecule structure, again we need to export current molecule structure by
Multiwfn as .pdb file, the steps have been described in the last section. After loading it into VMD
and making its drawing method as CPK, we will see
The arrows on this figure reflect magnitude and direction of induced dipole moment when an
external electric field of the same strength is applied to different directions from the center of the
molecule. It can be clearly seen from the figure that the polarizability of the cyclo[18]carbon ring
in the direction parallel (perpendicular) to the ring plane is large (small). This observation is easy to
understand, as indicated in my cyclo[18]carbon research paper, this system has 36 highly
delocalized electrons (18 in-plane and 18 out-of-plane ones), therefore when electric field is applied
parallelly to the ring, these electrons will be significantly polarized, resulting in large induced dipole
moment; in contrast, the electrons in this system is not so easily be polarized in the direction
perpendicular to the ring. Note that in my cyclo[18]carbon paper, the value of the  components
parallel to and perpendicular to the ring are reported to be 392 and 98 a.u., respectively, evidently
803
the figure shown above is fully in line with these quantitative values.
Similarly, we use unit sphere representation to visually study . In the Multiwfn window we
input
-3 // Change scale factor of length of the arrows on sphere
1E-5 // This value is significantly smaller than default one, since magnitude of  is quite large
1 // Perform unit sphere representation analysis of . Since gamma.txt already exists in
current folder,  tensor is automatically loaded from it
Move the newly generated gamma.tcl from current folder to VMD folder and input source
gamma.tcl in VMD console window to run it.
Notice that the  tensor parsed by subfunction 7 of main function 200 corresponds to input
orientation (in contrast, the parsed  and  correspond to standard orientation), therefore, the
molecular structure file loaded into VMD must also correspond to input orientation, otherwise the
unit sphere representation map will be misleading. In order to yield the .pdb file corresponding to
input orientation, we change "iloadGaugeom" in settings.ini to 1, then reboot Multiwfn and input
examples\polar\C18\gamma.out // Geometry in input orientation will be loaded from this file
2 // Generate new file
1 // Export current geometry as .pdb file
C18.pdb
Load the C18.pdb into VMD and show it in CPK style, you will see below figure
The character of this map is similar to that of  map. From the figure it can be seen that
combination effect of three electric fields applied parallelly to the ring can induce relatively strong
dipole moment variation in the same direction, while in the direction perpendicular to the ring this
phenomenon is much weaker.
4.300.4 Example of simulating scanning tunneling microscope (STM)

image
Note: Chinese version of this topic is http://sobereva.com/549, in which extended discussion is given and
cyclo[18]carbon is taken as example.
Please check Section 3.300.4 if you are not familiar with theory of STM or simulation of STM
804
in Multiwfn. In the next two sections, we will respectively simulate STM of constant height and
constant current modes for phenanthrene. The simulation will be based on the wavefunction in
examples\phenanthrene.fch, which was generated at B3LYP/6-31G* level.
It is important to note that in order to simulate STM in Multiwfn, the molecule must be parallel
to XY plane (though the molecule is not necessarily planar), however, in the phenanthrene.fch the
molecule is parallel to YZ plane. Therefore, the molecule must be rotated prior to STM simulation.
Of course, we can first reorientate the molecule and then conduct a single point task via quantum
chemistry to generate wavefunction file, but a better way is using Multiwfn to directly rotate the
wavefunction and geometry, namely inputting below commands in Multiwfn
6 // Check & modify wavefunction
33 // Rotate wavefunction, namely X→Y, Y→Z, Z→X
0 // Rotate all orbitals
y // Also rotate molecule structure. Then the molecule will be on XZ plane
33 // Rotate wavefunction again
0 // Rotate all orbitals
y // Also rotate molecule structure
Now the phenanthrene has exactly been on XY plane of Z=0 Å (you can check this point via
main function 0). Then we enter main function 100, choose subfunction 2 and then select
corresponding option to export the present wavefunction to a new .fch file. In the next sections, this
new .fch file will be referred to as mol.fch.
4.300.4.1 Simulating constant height STM image for phenanthrene
Here we simulate STM image of constant height mode for phenanthrene. Boot up Multiwfn
and input
mol.fch
4 // Simulating STM image
From the message on screen it can be seen that the Fermi level (EF) has been set to average of
HOMO energy and LUMO energy, the bias voltage (V) has been automatically set to the difference
between HOMO energy and EF, in this case only HOMO can contribute to the STM image. In order
to obtain expected STM image, it is crucial to properly define the V. In the case of negative V,
electrons flow from sample to STM tip, and the more negative the V, the more MOs may contribute
to the STM image. Also, note that the distance between the atoms in the sample and the tip
significantly affects STM image. From the information on option 7 you can find the default Z
coordinate of the plane to be plotted is 0.7 Å. Since all atoms in the mol.fch have Z coordinate of 0
Å, the distance between the nuclei and the tip is 0.7 − 0.0 = 0.7 Å. In this example, we will plot
STM image with V= -5.0 V at Z=1.2 Å.
Now input below command
2 // Set bias voltage
-5 // Bias voltage of -5.0 V
7 // Set Z coordinate
1.2 // Z=1.2 Å
0 // Calculate tunneling current on the plane
Now you can find below information on screen
805
Lower limit of MO energy considered in the calculation: -8.362 eV
Upper limit of MO energy considered in the calculation: -3.362 eV
The MOs taken into account in the current STM simulation:
MO 44 Occ= 2.000 Energy= -7.6491 eV Type: Alpha&Beta
Totally 4 MOs are taken into account
Grid spacings in X and Y are 0.118428 0.082980 Bohr
Calculating, please wait...
Maximal value (LDOS) is 0.019891 a.u.
It can be seen that there are 4 occupied MOs whose energy is lying between EF+eV (-8.362 eV)
and EF (-3.362 eV), therefore the plane data of the STM image corresponds to sum of their
probability densities in the plane multiplied by their occupation numbers (2.0 in present case since
they are all close-shell MOs), this data is also known as local density-of-states (LDOS). As shown
in the prompt, the largest value of LDOS in the calculated plane is 0.01989 a.u.
Now you are in the STM plotting menu, you can find many options used to adjust plotting
effect, they are all self-explanatory, please play with them. We directly choose option 0 to plot the
image under default setting, you will see
In this map, the brighter the white, the larger the LDOS and thus the stronger the tunneling
current (I), since the Tersoff-Hamann model shows that I is positively proportional to LDOS. It can
be seen that I signal is more prominent over the two boundary six-membered rings than the central
one.
4.300.4.2 Simulating constant current STM image for phenanthrene
In this section we again plot STM image for phenanthrene but using constant current mode.
Boot up Multiwfn an input
mol.fch
806
4 // Simulating STM image

1 // Switch the mode of STM image to constant current
2 // Set bias voltage
-5 // Again we use bias voltage of -5.0 V
In the constant current mode, LDOS is calculated for every evenly distributed point in a 3D
region, whose X, Y and Z range can be set by options 5, 6 and 7, respsectively, usually the default
setting is appropriate. We directly choose option 0 to start the calculation, from the information on
screen you can find the maximal value of LDOS in the calculated region is 0.048 a.u.
In the post-processing menu you can find several options, we first use option 1 to visualize
isosurface map of tunneling current, which corresponds to LDOS in the present context. The
isosurface corresponding to LDOS=0.015 a.u. is shown below. Note that although the choice of
isovalue is arbitrary, it should be between 0 and the maximum value (0.048 a.u. in this example)
In above map, the isosurfaces of tunneling current normally occur over the carbons, implying that
the default calculated region is appropriate for the present case. The blue box can be shown by
clicking "Show data range" check box, it displays the region where LDOS was calculated.
Next, we plot plane map. Select option "3 Calculate and visualize constant current STM image"
and then input the expected value of tunneling current, we input 0.01 in this example. After that,
Multiwfn starts to calculate the Z value where tunneling current (LDOS) is approximately equal to
0.01 a.u., evidently Z is different at different (x,y) positions. Then from screen you can find
Minimal Z is 0.700000 Angstrom
Maximal Z is 1.206432 Angstrom
They are proper lower and upper limits of the color scale of STM image, respectively.
Now you are in the interface of plotting STM image of constant current mode, we input below
commands
2 // Choose map type
2 // Color-filled map with contour lines
7 // Set stepsize in X,Y,Z axes
1.5,1.5,0.05
-3 // Change other plotting settings
2 // Set number of digits after the decimal point for the labels
1 // Set X axis
1 // Set Y axis
2 // Set Z axis
807
0 // Return
0 // Plot the STM image
Then you can see this map on screen:
In above map, the value corresponds to Z distance of STM tip. It can be seen that at constant
current (LDOS) of 0.01 a.u., the Z position of STM tip is relatively high over the two boundary six-
membered rings. The characteristics of this map is very similar to the STM image of constant height
mode.
If you want to further investigate STM plane map with other constant current value, you can
exit the plotting interface, then enter the option "3 Calculate and visualize constant current STM
image" again and input the expected current value.
4.300.5 Calculate electric dipole moment and multipole moments
As described in Section 3.300.5, Multiwfn is able to analytically calculate electric dipole

moment, quadrupole and octopole moments. In this section we calculate these quantities for a simple
molecule, uracil.
examples\uracil.wfn
5 // Calculate electric dipole moment and multipole moments
The calculation is quite fast, you will immediately see below information, which are very easy
to understand if you have read Section 3.300.5. As clearly indicated on screen, the unit is a.u. unless
otherwise specified.
Dipole moment (a.u.): 0.473309 1.823683 0.000736
Dipole moment (Debye): 1.203031 4.635340 0.001872
Magnitude of dipole moment: 1.884103 a.u. 4.788911 Debye
Quadrupole moments (Standard Cartesian form):
XX= -43.878975 XY= 1.812455 XZ= -0.000544
YX= 1.812455 YY= -28.230039 YZ= -0.004831
808
ZX= -0.000544 ZY= -0.004831 ZZ= -34.463441
Quadrupole moments (Traceless Cartesian form):
XX= -12.532235 XY= 2.718682 XZ= -0.000816
YX= 2.718682 YY= 10.941169 YZ= -0.007246
ZX= -0.000816 ZY= -0.007246 ZZ= 1.591066
Magnitude of the traceless quadrupole moment tensor: 13.645453
Quadrupole moments (Spherical harmonic form):
Q_2,0 = 1.591066 Q_2,-1= -0.008367 Q_2,1= -0.000942
Q_2,-2= 3.139264 Q_2,2 = -13.552376
Magnitude: |Q_2|= 14.001908
Octopole moments (Cartesian form):
XXX= 14.0946 YYY= 8.6123 ZZZ= 0.0027 XYY= 6.6643 XXY= 46.4098
XXZ= -0.0099 XZZ= 2.7809 YZZ= -7.6843 YYZ= 0.0159 XYZ= -0.0033
Octopole moments (Spherical harmonic form):
Q_3,0 = -0.0062 Q_3,-1= -52.5167 Q_3,1 = -5.9005
Q_3,-2= -0.0129 Q_3,2 = -0.0500 Q_3,-3= 103.2618 Q_3,3 = -4.6630
Magnitude: |Q_3|= 116.0929
Note that you can also use subfunction 2 of main function 15 to calculate above quantities,
however it calculates numerically based on multicenter grids, the cost is significantly higher while
the numerical accuracy is slightly lower. However, it has a unique advantage, namely it can calculate
dipole and multipole moments for specific fragments in current system, see Section 4.15.3 for
example.
4.A Special topics and advanced tutorials
The contents in this section involve more than one main functions of Multiwfn, or contain
special usages and skills.
4.A.1 Study variation of electronic structure along IRC path

If you can read Chinese, please read http://sobereva.com/200, which essentially covers all content of this section.
In this tutorial, I will briefly show you how to use Multiwfn to study variation of electronic
structure along the IRC path of Diels-Alder adduction. We will study the variation of Mayer bond
order, and will animate the deformation of ELF isosurface. With the similar fashion you can also
easily investigate variation of other properties, such as atomic charges, electron density, aromaticity
and so on.
809
Gaussian 09 was used throughout this tutorial. Unless otherwise specified, all calculations will
be performed under Windows 7 64bit system. In this tutorial the files marked by crimson can be
found in "examples\IRC" or "examples" folder.
Before starting this tutorial, you should setup running environment for Gaussian first,
otherwise Gaussian cannot be properly invoked in Windows environment. The setup method is:
Enter “control panel”-“System properties”-“Advanced”, click “Environment variables” button, then
click “New” button in “User variables” frame, input GAUSS_EXEDIR as variable name, input the
install directory of Gaussian as variable value (e.g. D:\study\g09w\, assuming that g09.exe is in this
folder). After that modify "PATH" environment variable to add the install directory of Gaussian into
it.
1 Perform IRC calculation

Run DA_IRC.gjf by Gaussian to produce DA_IRC.out. We will find this IRC path actually
contains 18 and 13 points in the two directions, respectively. B3LYP/6-31+G* is used in this
calculation.
2 Generate wavefunction file for each point of IRC

Write an input file of single point task of Gaussian (DA_SP.gjf), which will be used as
"template" later. The geometry in fact can be arbitrarily filled.
DO NOT write anything here (e.g. %chk)
#p B3LYP/6-31G* nosymm
DA adduction
0 1
C -0.26156800 1.56679300 0.69509600
C -0.26156800 1.56679300 -0.69509600
C 0.50031400 -0.43279300 -1.43864500
C -0.26156800 -1.32826500 -0.70392000
C -0.26156800 -1.32826500 0.70392000
H -1.19341600 1.44689700 -1.23752300
H -1.19341600 1.44689700 1.23752300
H 0.52507600 2.08588200 1.23621300
H 0.38154400 -0.37781200 2.51847500
H 0.38154400 -0.37781200 -2.51847500
H 1.46467600 -0.09643600 -1.07409700
H -1.04094400 -1.89294000 -1.21418700
H -1.04094400 -1.89294000 1.21418700
H 1.46467600 -0.09643600 1.07409700
H 0.52507600 2.08588200 -1.23621300
C 0.50031400 -0.43279300 1.43864500
 blank line
 blank line
Notice that the basis set we used here (6-31G*) is different to the one used in IRC task (6-
31+G*), because Mayer bond order does not work well when diffuse functions are presented. By
the way, ignoring diffuse functions will not lead to detectable change of ELF isosurface. Also note
810
that the "nosymm" keyword is specified, because if we do not do this Gaussian will automatically
translate and rotate the molecule to put it to standard orientation, which may leads to discontinuity
problem in the animation of ELF (You will see molecule suddenly jumps in certain frames of the
animation).
IRCsplit.exe is a tool used to produce .wfn/.chk file for each point of IRC and SCAN tasks of
Gaussian, IRCsplit.f90 is the corresponding source code, by which you can compile Linux version
of IRCsplit. Boot up IRCsplit.exe by double click its icon and then input
DA_IRC.out //The file of the output file of the IRC task
DA_SP.gjf //The template file used to generate single point input files
2 //Only yield .chk files
C:\DA_IRCchk\DA //The path and prefix of the finally generated .chk files
18,13 // The program detected that in DA_IRC.out there are 18 and 13 points in the two
directions of IRC, respectively. Here we extract all of them, together with the TS point
Now you can find DA_SP0001.gjf, DA_SP0002.gjf ... DA_SP0032.gjf in current folder. Please
manually check one of them to verify the reasonableness of these input files. Note that
DA_SP0014.gjf corresponds to the TS geometry.
Build a new folder "C:\DA_IRCchk" and copy the .gjf files as well as the script runall.bat into
it. Double clicking the icon of runall.bat, which will invoke Gaussian 09 to run all of the .gjf files.
Now you have DA0001.chk, DA0002.chk ... DA0032.chk in "C:\DA_IRCchk" folder. Copy
chk2fch.bat to this folder and run it, then the formchk utility in Gaussian package will be
automatically invoked to convert all .chk files to .fch files.
3 Calculate Mayer bond orders for all IRC points

Mayer bond order of C1-C16 is the one we are particularly interested in, whose formation is
the key process of the DA adduction. Since by default Multiwfn only outputs Mayer bond orders
with value > 0.05, while C1-C16 must be very weak at the initial stage of DA adduction, we need
to set "bndordthres" parameter in the settings.ini file in Multiwfn folder to 0.0, so that all of the
bond orders larger than 0.0 can be outputted.
Write a plain text file (MBObatch.txt) and put it into Multiwfn folder, the content is
9 // Enter bond order analysis module
Note: If you are confused why this file is written in such manner, please read Section 5.2 to study how to run
Multiwfn in silent mode.
Then write a plain text file with .bat suffix (MBObatchrun.bat) and put it into Multiwfn folder,
the content should be
for /f %%i in ('dir C:\DA_IRCchk\*.fch /b') do Multiwfn C:\DA_IRCchk\%%i < MBObatch.txt >
C:\DA_IRCchk\%%~ni.txt
batchrun.bat in fact is a Windows batch script. Double clicking its icon to run it, the .fch files
in "C:\DA_IRCchk\" folder will be sequentially loaded into Multiwfn, and the calculated Mayer
bond orders will be exported to .txt files in the same folder.
4 Plot Mayer bond order

Now what we should do next is to extract the bond order of C1-C16 from the DA0001.txt,
DA0002.txt ... DA0032.txt. The most convenient way is to utilize "grep" command in Linux. So we
copy all of these .txt files to a folder in Linux system, then in this folder we run
grep "1(C ) 16(C )" * > out.txt
811
The out.txt file now contains C1-C16 bond order of all points in the IRC:
DA0001.txt:# 9: 1(C ) 16(C ) 0.05929972
DA0002.txt:# 7: 1(C ) 16(C ) 0.06877306
DA0003.txt:# 7: 1(C ) 16(C ) 0.07926829
DA0004.txt:# 7: 1(C ) 16(C ) 0.09089774
DA0005.txt:# 7: 1(C ) 16(C ) 0.10380144
DA0006.txt:# 7: 1(C ) 16(C ) 0.11815120
DA0007.txt:# 7: 1(C ) 16(C ) 0.13417828
DA0008.txt:# 7: 1(C ) 16(C ) 0.15218555
...
The last column is the values of Mayer bond order of C1-C16, you can plot them by your
favorite program now, you will see
1.1
1.0
0.9 C1-C16
0.8
0.7
Mayer bond order
0.6
0.5
0.4
0.3 TS
0.2
0.1
0.0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32
IRC point
Clearly, C1-C16 become stronger and stronger as the reaction proceeds, its Mayer bond order
gradually increases to 1.0 (typical single bond).
With the same method, we also calculate the Mayer bond order of C1-C2 and C4-C5, namely
run below commands
grep "1(C ) 2(C )" * > out2.txt
grep "4(C ) 5(C )" * > out3.txt
Plot the data in out.txt, out2.txt and out3.txt together, you will see
812
TS
2.0
1.8
1.6
1.4
Mayer bond order
1.2
1.0
0.8 C1-C16
C1-C2
0.6
C4-C5
0.4
0.2
0.0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32
IRC point
This graph vividly shows that the C1-C2 smoothly becomes to a single bond from a double
bond during the DA adduction, and the reaction increases the double-bond character of C4-C5
significantly.
5 Make animation of ELF isosurface

Next we make animation to study how the ELF isosurface varies during the DA adduction.
Create a plain text file ELFbatch.txt in Multiwfn folder with below content
9 // ELF
2 // Export the grid data to ELF.cub in current folder
Create a script file named ELFbatchrun.bat, whose content is

for /f %%i in ('dir C:\DA_IRCchk\*.fch /b') do (
Multiwfn C:\DA_IRCchk\%%i < ELFbatch.txt
rename ELF.cub %%~ni.cub
Run ELFbatchrun.bat, Multiwfn will sequentially load the .fch files in "C:\DA_IRCchk" and
export the corresponding ELF grid data to DA0001.cub, DA0002.cub ... DA0032.cub in current
folder.
We use VMD 1.9.1 program (freely available at http://www.ks.uiuc.edu/Research/vmd/) to
render isosurface for these cube files. Move all of the cube files to VMD folder, and create a plain
text file named isoall.tcl in the VMD folder, the content is
set isoval 0.88
axes location Off
color Display Background white
for {set i 1} {$i<=32} {incr i} {
set name DA[format %04d $i]
813
puts "Processing $name.cub..."
mol default style CPK
mol new $name.cub
#translate by -0.100000 0.20000 0.000000
#scale to 0.30
rotate y by 50
rotate z by 90
rotate x by -30
rotate y by -20
mol addrep top
mol modstyle 1 top Isosurface $isoval 0 0 0 1 1
mol modcolor 1 top ColorID 3
render snapshot $name.bmp
mol delete top
This file essentially is a VMD script, in which the command set isoval 0.88 means the isosurface of
0.88 will be plotted, the default view point is adjusted by scale, rotate and translate commands. for
{set i 1} {$i<=32} {incr i} means the file from DA0001.cub to DA0032.cub will be processed.
Now boot up VMD, and input the command source isoall.tcl in its command line window, then
you will have DA0001.bmp, DA0002.bmp ... DA0032.bmp.
There are numerous programs that can convert single-frame graphic files to animation, such as
Atani, Premiere, Vegas, Ulead Video Studio, Videomach, etc. Here we use ImageMagick tool in
Linux to do this, and we choose gif as the animation format, since gif animation can be directly
embedded into webpages.
Copy all of the .bmp files to Linux system, and run below command in the corresponding folder:
convert -delay 12 -colors 100 -monitor *.bmp ELF_IRC.gif
in which -delay controls the time inverval between each frame in the animation, and -colors
determines the number of colors used, the larger the value, the more smoothly the color changes,
but the larger the animation file. You can run convert --help to study more arguments of this tool.
If the the resultant ELF_IRC.gif cannot be properly displayed on your system, use your
webpage explorer or advanced image explorers (e.g. IrfanView) to open it. The deformation of ELF
isosurface in this animation very intuitively exhibits how the new bonds are formed and how the
characteristic of existing bonds changed.
4.A.2 Calculation of spin population
As there are many ways to calculate atomic charges (see Section 3.9 for introduction and 4.7
for examples), there are various ways to calculate spin population. Spin population is defined as the
population number of alpha electrons minusing that of beta electrons. Spin population is a key
quantity for characterizing electronic structure of open-shell systems, i.e. radicals and
antiferromagnetic systems. From spin population we can clearly know where the spin electrons are
mainly distributed. Moreoever, we can discuss contribution from different regions (atomic orbitals,
atoms or fragments) to the total magnetic dipole moment m due to electron spin. If spin population
of a region is x, then its contribution to m will be xμB, where the Bohr magneton μB=eћ/(2me) (e:
814
electron charge, me: mass of electron) represents the magnetic moment produced by a single electron.
Note that in chemical systems the movement of electron in orbitals and nuclear spins also have
contributions to m, but the magnitude is evidently weaker and thus can often be neglected.
In Multiwfn, the spin population defined in many different ways can be calculated by three
modules, they are briefly discussed below.
(1) Population analysis module (main function 7). In this module, if you select Mulliken or
Löwdin population analysis, alpha, beta and spin population of each basis function, shell and angular
moment orbitals will be outputted. If you select modified Mulliken population analysis (e.g. SCPA),
only the alpha/beta/spin population of each atom will be shown. If you first define a fragment via
option -1 in main function 7, then spin population of the fragment will be printed together. Do not
use these methods when diffuse basis functions are presented in your basis set, otherwise the result
may or may not be reliable.
(2) Fuzzy atomic spaces analysis module (main function 15). After you entered this module,
select option 1 and choose electron spin density, the spin population of each atom will be shown.
They are calculated by integrating electron spin density in fuzzy space of each atom. If you want to
obtain spin population of a fragment, you should first use option -5 to define the atoms to be
calculated.
By default the fuzzy atomic space defined by Becke is employed, so the result can be called
Becke spin population. If before calculation you selected option -1 to switch to Hirshfeld or
Hirshfeld-I fuzzy atomic space, then the result will correspond to Hirshfeld or Hirshfeld-I spin
population. All the Becke, Hirshfeld and Hirshfeld-I methods are reliable in all cases. For more
detail you can consult Section 3.18.
(3) Basin analysis module (main function 17). In this module, you can use AIM method to
calculate spin population. Please consult Section 4.17.1 on how to perform integration of real space
function in AIM atomic basins. If electron spin density is chosen to be the integrand, then the result
will correspond to AIM spin population. In general I do not suggest using this method, because the
computational cost is evidently higher than using population analysis module and fuzzy atomic
spaces analysis module.
Overall, if you only need to calculate atomic spin population, using fuzzy atomic spaces
analysis module is recommended, while SCPA is also a good choice when no basis function is
employed. However, if more detailed information are requested, such as spin population in different
angular moment orbitals, please use or Löwdin Mulliken population analysis.
4.A.3 Overview of methods for studying aromaticity
Aromaticity is a fundamental concept in organic chemistry and wavefunction analysis realm.

Previously I wrote a post to thoroughly discuss the methods for studying aromaticity, see "The
methods for measuring aromaticity and their calculations in Multiwfn" (in Chinese,
http://sobereva.com/176). Multiwfn supports a very large number of methods for investigating
aromaticity, they are summarized in below table and will be briefly introduced in turn. There are
also many other methods, such as induced ring current, ARCS, magnetic susceptibility exaltation,
aromatic stabilization energy (ASE), CiLC; they will not be mentioned since they are not directly
related to the capacities of Multiwfn.
Method Principle Year Pop. Reliab. Univ. Ref. Anti. / Cost Value
815
1 Molecular orbital Hückel 1951 ++ 0 0 N Y Y 0 +

2 AdNDP Hückel 2008 + 0 + N Y Y 0 +
3 NICS Magnet. 1996 +++ ++ ++ N Y Y + +++
4 ICSS Magnet. 2001 0 ++ ++ N Y Y +++ +
5 HOMA Geom. 1972 + + 0 Y Y N −− +
6 Bird Geom. 1985 −− 0 − Y ? N −− −
7 Multi-center BO Delocal. 1990 + +++ ++ N N Y 0 +++
8 ELF-σ/π Delocal. 2004 + 0 − N Y Y 0 +
9 PDI Delocal. 2003 0 + 0 N N Y 0 +
10 ATI Delocal. 2005 −− + 0 N N Y − 0
11 PLR Delocal. 2012 −− + 0 N N Y 0 0
12 ΔDI Delocal. 2003 −− − − N N Y 0 −−
13 FLU, FLU-π Delocal. 2005 0 + 0 Y/N N Y 0 +
14 RCP properties  1997 − 0 0 N N ? − 0
15 Shannon aromat.  2010 N Y N −
16 EL index  2012 −− − − Y Y N − −−
17 AV1245/AVmin Delocal. 2017 − + ++ N N Y −− ++
In the table, "+++", "++", "+", "0", "−" and "− −" correspond to very high, high, relatively high,
normal, relatively low and low, respectively. "Y" and "N" stand for "Yes" and "No", respectively.
The meaning of each column are given below.
Principle: The principle behind the method. "Hückel" = Hückel rule; "Magnet." = Magnetic
properties; "Geom." = Molecular geometry; "Delocal." = Electron delocalization character; "" =
Electron density distribution.
Year: The year that the method was first time proposed.
Pop.: Popularity in recent years.
Reliab.: Reliability, measuring if the method is able to faithfully reveal aromaticity.
Univ.: Universality. A method with high universality must be able to be applied to a wide variety of
kinds of systems and situations, such as the rings containing heteroatoms and transition metals, non-
equilibrium geometry (e.g. transition state of Diels-Alder adduction), excited state, etc.
Ref.: If the method relies on reference systems. A universal method must avoid this feature.
Anti: If the method is also able to measure anti-aromaticity.
σ/π: If the method can be used to separately discuss σ and π aromaticity.
Cost: The computational cost to apply the method.
Value: The overall value. This is the most important descriptor.
Next, the methods presented in above table will be briefed sequentially, and how to realize
them in Multiwfn will also be mentioned.
1. Molecular orbital (MO): The famous Hückel 4n+2 and 4n rule for determining aromaticity
character was first explicltly presented in J. Am. Chem. Soc., 73, 876 (1951). For a molecule, if there
are totally 4n+2 electrons in π () MOs, and this set of MOs share the similar delocalization pattern,
then the ring involved in these MOs will show π () aromaticity. If there are 4n electrons, then the
ring should possess anti-aromaticity. Note that for Möbius type of molecule, the 4n+2 and 4n rule
are inverted.
In order to use the Hückel rule to determine aromaticity, one should first pick out proper MOs
816
by visualizing MO isosurfaces, you can use main function 0 for this purpose. If the system is exactly
planar, you can directly make use the function introduced in Section 3.100.22 to find out the indices
of all π MOs.
2. AdNDP (Adaptive natural density partitioning): The MO method shown above
commonly is only applicable to the molecule containing only one ring. When there are multiple
rings, such as phenanthrene, the MOs are useless, since MOs in general delocalize over the whole
molecule and thus cannot be used to study local aromaticity of different rings. The AdNDP method,
which was proposed in Phys. Chem. Chem. Phys., 10, 5207 (2008), is able to overcome this difficulty.
AdNDP has been carefully introduced in Section 3.17, and many examples are given in Section 4.14.
3. NICS (Nucleus-independent chemical shift): NICS uses the negative value of magnetic
shielding value at ring center to measure its aromaticity. This is the most popular aromaticity index
nowadays, it was originally proposed in J. Am. Chem. Soc., 118, 6317 (1996) and reviewed in Chem.
Rev., 105, 3842-3888. There are also a few variants, among them the best one is NICS(1)ZZ, see Org.
Lett, 8, 863 (2006). For non-planar system, it is often difficult to calculate NICS(1)ZZ, in this case
you will find the function introduced in Section 3.100.24 quite useful.
4. ICSS (Iso-chemical shielding surface): The original paper of ICSS is J. Chem. Soc., Perkin
Trans., 2, 1893 (2001). This method analyzes aromaticity by visualizing isosurface of magnetic
shielding value around the molecule. See Section 3.200.4 for introduction and Section 4.200.4 for
example. The main drawback of this method is that calculating grid data of magnetic shielding
values is fairly time-consuming.
5. HOMA (Harmonic oscillator measure of aromaticity): HOMA measures aromaticity
based on bond lengths in the ring of interest. See Section 3.100.13 for introduction and Section
4.100.13 for example.
6. Bird index: The same as above.
7. Multi-center bond order (MCBO): MCBO is an indicator of electron delocalization ability
over a ring and is the aromaticity index I most strongly recommended. Larger MCBO value
corresponds to stronger aromaticity. See Section 3.11.2 for introduction. Some applications of
MCBO in aromaticity studies can be found in J. Phys. Org. Chem., 26, 473 (2013), Phys. Chem.
Chem. Phys., 2, 3381 (2000) and J. Phys. Chem. A, 109, 6606 (2005). It is straightforward to discuss
π and σ aromaticities separately by MCBO, that is before calculating MCBO value, first set
occupation number of all σ and π MOs to zero respectively by subfunction 22 of main function 100.
Note that the definition of MCBO in many literatures differ with that in Multiwfn by a constant
coefficient. The calculation cost of MCBO for six-membered ring can be ignored, however the cost
increases exponentially with the number of atoms in the ring.
8. ELF-σ/π: The ELF calculated solely based on π orbitals and all other orbitals are referred to
as ELF-π and ELF-σ, respectively. It was argued that the value of bifurcation point of ELF-π (ELF-
σ) is an indicator of π (σ) aromaticity, some applications can be found in J. Chem. Phys., 120, 1670
(2004), J. Chem. Theory Comput., 1, 83 (2005) and Chem. Rev., 105, 3911 (2005). An example of
calculating ELF-σ/π is given in Section 4.5.3. In addition, Section 4.4.9 presented an example of
studying LOL-π (which is very similar to ELF-π) by plotting plane map. I do not think ELF-σ/π is
a very ideal method for measuring aromaticity, mostly because this method often suffers from
ambiguity (you will recognize this point if you have used this method to study many practical
systems). Also note that the bifurcation values of ELF-σ/π in a lot of literatures are incorrect; if you
try, you will find it is impossible to reproduce their results at all. (So do not always believe literatures
817
but believe in yourself!)

9. PDI (Para-delocalization index): This aromaticity index is only applicable to six-
membered rings. PDI was first proposed in Chem. Eur. J., 9, 400 (2003) and reviewed in Chem.
Rev., 105, 3911 (2005). Please check Section 3.18.6 for introduction of PDI and Section 4.15.2 for
example of using PDI.
10. ATI (Average two-center indices): ATI was first proposed in J. Phys. Org. Chem., 18, 706
(2005). In fact ATI does not contain any new idea, it simply replaces the delocalization indices
involved in PDI formula with corresponding Mayer bond orders, and according to the discussions
in J. Phys. Chem. A, 109, 9904 (2005), there is no essential difference between Mayer bond order
and delocalization index in physical nature. If you would like to use ATI, you can directly calculate
Mayer bond order by Multiwfn and then manually calculate ATI according to its formula.
11. PLR (Para linear response index): As ATI, PLR is also very akin to PDI. The only
difference between PLR and PDI is that the delocalization indices in PDI are replaced with
corresponding condensed linear response kernels. Original paper of PLR is Phys. Chem. Chem.
Phys., 14, 3960 (2012). You can check Section 3.18.9 for introduction of PLR and Section 4.15.2
for example of using PLR.
12. DI: This method was proposed in Chem. Eur. J., 9, 400 (2003) for measuring aromaticity
of 5-membered systems. Consider below case
The DI is simply defined as the difference of delocalization index (DI) between the formal C=C
bond and the C-C bond. The DI can be either calculated by fuzzy atomic space analysis module or
by basin analysis module (though the definition of atomic spaces are different in these two module,
the results are similar in common). In fact, you can also use Mayer bond order instead of DI. I do
not believe DI is reliable, since aromaticity is an overall property of a system, while the
delocalization over C-X bond is completely ignored in DI.
13. FLU and FLU-π (Aromatic fluctuation index): They were proposed in J. Chem. Phys.,
122, 014109 (2005). See Section 3.18.7 for introduction and Section 4.15.2 for example.
14. RCP properties: In Can. J. Chem., 75, 1174 (1997) it was shown that the density and the
curvature of density perpendicular to the ring plane at ring critical point (RCP) closely relate to
aromaticity of the ring. The larger the density, or the more negative the curvature, the larger the
aromaticity. You can use topology analysis module of Multiwfn to apply this method. Detail
introduction can be found in Section 3.14.6, an example is given in Section 4.2.1.
15. Shannon aromaticity: This method was proposed in Phys. Chem. Chem. Phys., 12, 4742
(2010), which measures aromaticity based on electron density at bond critical points (BCP) in the
ring. See Section 3.14.6 for introduction and the example given in Section 4.2.1.
16. EL index: The idea of EL index is quite similar to HOMA, the most prominent difference
is that the bond lengths in HOMA formula are replaced with electron density ellipticity at BCPs in
the ring. For more detail see the original paper Struct. Chem., 23, 1173 (2012). Electron density
ellipticity at BCPs can be directly calculated by topology analysis module of Multiwfn. Since the
ellipticity at BCP is usually unclear for strongly polar bonds, EL index may be unreliable for the
ring containing heteroatoms. In addition, EL index shares the same drawback of HOMA, that is
818
reference system is need. If reference system cannot be obtained, such as the case of metal clusters,
this method does not work.
17. Aromaticity indices defined based on information-theoretic quantities: It was
demonstrated in ACS Omega, 3, 18370 (2018) that the average of information-theoretic quantities
of the atoms constituting a ring is closely related to aromaticity. This method is supported as
subfunction 12 of main function 15, see Section 3.18.11 for details.
18. AV1245 and AVmin: AV1245 can be viewed as an approximation of MCBO. The MCBO
commonly cannot be applied to rings containing more than 12 atoms, while AV1245 can be easily
employed to study aromaticity of quite large rings. AVmin is closely related to AV1245, it is able to
reveal bottleneck of electron delocalization and thus aromaticity of a selected path. See Section
3.11.10 for introduction and Section 4.9.11 for example.
4.A.4 Overview of methods for predicting reactive sites
There are numerous methods able to predict reactive site of electrophilic, nuelcophilic and
radical reactions, and almost all of them are supported by Multiwfn. In this section, I will summary
and briefly introduce the methods available in Multiwfn. The interested reader is highly
recommended to take a look at Acta Phys.-Chim. Sinica, 30, 628 (2014), in which various methods
for predicting electrophilic sites are carefully introduced and thoroughly compared. You may also
find the slideshow "Predicting reactive sites" in "Related resources and posts" Section of Multiwfn
website useful.
1 Electrostatic potential (ESP). If you are not familiar with ESP, please consult corresponding
introduction in Section 2.6. Since electrophile (nucleophile) locally carries positive (negative)
charge, and thus tends to be attracted to the region where ESP is negative (positive), the position
and value of minima (maxima) of ESP on molecular vdW surface is often used to reveal favorable
site of electrophilic (nucleophilic) attack. ESP analysis can be realized via quantitative molecular
surface analysis module, see Section 4.12 for detailed introduction and Section 4.12.1 for example.
There are also alternative ways to study ESP; as illustrated in Section 4.12.3, the average of ESP on
local vdW surface corresponding to each atom is also very useful, and this approach is more reliable
and robust than analyzing ESP extrema on vdW surface. For planar system, one can also calculate
and compare the ESP value above 1.6Å (approximately equal to vdW radius of carbon) of molecular
plane from different atoms to examine their reactivities; to do this, you need to use main function 1,
which directly outputs various real space function values at given points.
However, as shown in my paper Acta Phys.-Chim. Sin., 30, 628 (2014), ESP is usually not a
reliable property for predicting reactive sites.
2 Average local ionization energy (ALIE) and local electron affinity (LEA). If you are not
familiar with ALIE, please read corresponding introduction in Section 2.6. ALIE can be studied in
ways analogous to ESP. The most common way to predict reactive sites in terms of ALIE is
analyzing minima of ALIE on vdW surface, see Section 4.12.2 for example. Also, you can study
average of ALIE on local vdW surface or evaluate ALIE above 1.6Å of molecular plane for planar
system.
ALIE analysis is applicable to electrophilic and radical attacks, but it is useless for nucleophilic
attack. However, the local electron affinity (LEA) defined in similar way may be useful for this
purpose, see J. Mol. Model., 9, 342 (2003). LEA is supported in Multiwfn as user-defined function
819
27, see corresponding description in Section 2.7 for detail. The best way of analyzing LEA should
be plotting LEA mapped molecular surface map, as explicitly illustrated in Section 4.12.13.
3 Atomic charges. It is easy to understand that favorable electrophilic and nucleophilic
reactive sites should carry negative and positive atomic charges respectively, so that they can attract
electrophile and nucleophile to attack them. Multiwfn supports a lot of methods to calculate atomic
charges, see Section 3.9 for introduction and Section 4.7 for some instances. Among the available
atomic charges, the best one for predicting reactive site purpose may be Hirshfeld, interested readers
are suggested to consult J. Phys. Chem. A, 118, 3698 (2014) and especially Theor. Chem. Acc., 138,
124 (2019), the latter very nicely demonstrated reliability and value of Hirshfeld charge in predicting
both electrophile and nucleophile reactive sites. Do not use Mulliken charges, which may be the
worst one, though it is the most popular charge model.
4 Frontier molecular orbital (FMO) theory. Atom with larger contribution to HOMO
(LUMO) is more likely to be the preferential site of electrophilic (nucleophilic) attack. Multiwfn
supports many kinds of methods to calculate molecular orbital composition, see Section 3.10 for
introduction and Section 4.8 for examples. Commonly I suggest using Becke or Hirshfeld method.
Mulliken method works equally well if no diffuse functions are presented. NAO method is also a
good choice, but not suitable for analyzing virtual MOs. Besides, you can also directly visualize the
isosurface of MOs by main function 0 to discuss their compositions.
5 Fukui function and condensed Fukui function. The Fukui function proposed in J. Am.
Chem. Soc., 106, 4049 (1984) by Parr is the most prevalently used method for predicting reactive
sites nowadays. Please consult Section 4.5.4 for introduction and illustration. Fukui function is a
real space function, which is commonly studied by means of visualization of isosurface. In order to
faciliate quantitative comparison between difference sites, one can calculate condensed Fukui
function based on atomic charges, please consult Section 4.7.3. In addition, as illustrated in Section
4.12.4, distribution of Fukui function can also be characterized by means of local quantitative
molecular surface analysis technique. Furthermore, Multiwfn is able to evaluate contribution of
various kinds of orbitals (MO, NBO, NAO, etc.) to Fukui function to characterize it in terms of
orbital perspective, see Section 4.200.13.1 for example and Section 3.200.13 for introduction of the
algorithm.
6 Dual descriptor and condensed dual descriptor. As demonstrated in Acta Phys.-Chim.
Sinica, 30, 628 (2014), the dual descriptor proposed in J. Phys. Chem. A, 109, 205 (2005) may be
the most robust method for predicting reactive sites, at least for electrophilic reaction. Like Fukui
function, dual descriptor also has condensed version. Dual descriptor and the condensed version are
introduced in Section 4.5.4 and 4.7.3, respectively.
Note that the easiest way of calculating Fukui function, dual descriptor as well as their
condensed version is using subfunction 16 of main function 100, as introduced in Section 3.100.16
and illustrated in Section 4.100.16.1. An additional advantage is that many other important
quantities defined in the framework of conceptual density functional theory can be obtained together
without any additional cost, including Mulliken electronegativity, hardness, electrophilicity and
nucleophilicity indexes, softness, condensed local softness, relative electrophilicity and
nucleophilicity and so on, which are also quite useful for studying reactivity problems.
7 Orbital-weighted Fukui function and orbital-weighted dual descriptor: They are special
form of Fukui function and dual scriptor, the unique advantage of this orbital-weighted form is able
to reasonably deal with systems with degenerate or nearly degenerate frontier molecular orbitals,
820
such as C60, coronene and cyclo[18]carbon, usually these kinds of system have high point group
symmetry. See Section 3.100.16.3 for introduction and 4.100.16.2 for illustrative application.
8 Orbital overlap distance function. Analysis of this function may be useful for revealing
favorable reactive site, see Section 4.12.8 for example.
4.A.5 Overview of methods for studying weak interactions
There are a lots of ways to characterize weak interactions, and most of them are supported by
Multiwfn, here I give you a brief summary. If you can read Chinese, I suggest reading my blog
article "An overview of the weak interaction analysis methods supported by Multiwfn"
(http://sobereva.com/252), in which this topic is discussed more deeply and extensively.
(1) AIM topology analysis is a very popular method for studying both strong and weak
interactions. Its use in weak interaction analysis is partially illustrated in Section 4.2.1.
(2) NCI analysis proposed in 2010 may be viewed as an visualization extension of AIM
analysis, this method rapidly became quite popular after it was proposed. The example of using NCI
analysis is given in Sections 3.23.1, 4.20.1 and 4.20.2. NCI analysis is also able to be employed to
study weak interaction in dynamic environment such as molecular dynamic simulation, see the
introduction in Section 3.23.2 and the accompanied example in Section 4.20.3. Integrating domain
of NCI is a useful way to discuss weak interactions quantitatively, examples are provided in Section
4.200.14.
IRI and DORI analysis is closely related to the NCI method. Advantage of IRI and DORI is
that all kinds of interactions can be simultaneously visualized, including both chemical bonds and
weak interactions, as illustrated in Sections 4.20.4 and 4.20.5. Relatively speaking, the IRI is
evidently preferred over DORI, since the graphical effect of IRI is better and computational cost is
lower.
(3) IGM analysis. A key advantage of IGM analysis compared to NCI analysis is that this
method is able to visually study intrafragment and interfragment interaction regions separately by
properly defining fragments. Contributions by atoms and atom pairs can be quantified as atom g
index and atom pair g index defined in the IGM framework, respectively, and atoms can be colored
according to the atom g indices to vividly exhibit the role played by various atoms. Two forms of
IGM are supported, namely original IGM and the IGMH proposed by me, see Sections 3.23.5 and
3.23.6 for introduction as well as Sections 4.20.10 and 4.20.11 for examples. Graphical effect of
IGMH is much better than IGM, but the computational cost is evidently higher.
(4) Electrostatic potential (ESP) analysis. ESP has been introduced in Section 2.6, this is a
extremely important real space function for studying electrostatic dominated weak interactions.
There are many different ways to carry out ESP analysis:
·Visually studying ESP color-mapped molecular vdW surface, this analysis can be used to
quickly figure out potential electrostatic interaction sites and qualitatively study interaction strength.
See the end of Section 4.12.1 and J. Mol. Model., 13, 291 (2007) for example.
·Studying ESP minima and maxima on molecular vdW surface. This can done by quantitative
molecular surface analysis module, see Section 4.12.1 for example and Section 3.15 for more details.
The value of these ESP extrema on vdW surface strongly correlate with electrostatic interaction
energies, and you can find many papers have used this method, for example Phys. Chem. Chem.
Phys., 15, 14377 (2013), J. Mol. Model., 13, 305 (2007), Int. J. Quantum. Chem., 107, 3046 (2007),
821
Phys. Chem. Chem. Phys., 12, 7748 (2010), J. Mol. Model., 14, 659 (2008), J. Mol. Model., 18, 541
(2012), J. Mol. Model., 15, 723 (2009), Chapter 6 of book Practical Aspects of Computational
Chemistry (2009).
·Studying area and averaged ESP value corresponding to characteristic region, such as -
hole, -hole and lone pair, see Section 4.12.10 for example.
·Superposition analysis of ESP contour map. This method was proposed by Tian Lu in J. Mol.
Model., 19, 5387 (2013), it is quite vivid, easy-to-use and powerful. It was demonstrated that
stability of complex configurations can be fairly well predicted by this method. Section 4.4.4
showed how to plot ESP contour map.
·In J. Phys. Chem. A, 118, 1697 (2014), the authors showed that by making use of ESP at
nuclear positions the electrostatic dominated intermolecular interaction energies can be very
accurately predicted. See Section 4.1.2 for introduction and example.
(5) van der Waals (vdW) potential. The vdW potential has same importance as ESP, especially
for the case that the interaction is dominated by vdW interaction rather than electrostatic interaction.
Multiwfn is able to easily perform vdW potential analysis in various forms. See Section 3.23.7 for
introduction and Section 4.20.6 for example. An in-depth introduction and discussion of vdW
potential can be found in my research paper ChemRxiv (2020) DOI: 10.26434/chemrxiv.12148572.
(6) Atomic charge analysis. Atomic charge is a very simple and intuitive model for describing
charge distribution and can be used to analyze the strength of electrostatic interaction between
different sites. The functions for calculating atomic charges are introduced in Section 3.9, and some
practical examples are given in Section 3.7.
(7) Hirshfeld and Becke surface analysis. This kind of analysis is extremelly useful for
revealing weak interaction in molecular crystals, but can also be applied to molecular clusters, see
examples in Sections 4.12.5 and 4.12.6 as well as theory introduction in Section 3.15.5.
(8) Bond order and delocalization index (DI) analysis. Commonly weak interactions are
dominated by electrostatic and/or vdW interactions, so bond order and DI analysis, which mainly
reflect covalent character are often not useful in these cases. However, for "strong" weak interactions,
such as low-barrier hydrogen bonds (LBHB) and charge-assisted halogen bonds, covalent
contribution may be not negligible, and thus bond order and DI analysis can be applied. Bond order
calculations are illustrated in Section 4.9. In Multiwfn, DI can be calculated based on fuzzy atomic
space or AIM basin, the former is equivalent to fuzzy bond order, while the latter can be evaluated
in basin analysis module, see example in Section 4.17.1.
(9) ELF analysis. In Theor. Chem. Acc., 104, 13 (2000), Fuster and Silvi defined CVB index
based on ELF to distinguish strength of H-bonds. J. Phys. Chem. A, 115, 10078 (2011) employed
this method to study a large amount of resonance-assisted hydrogen bonds and find this index is in
good correlation with other H-bond strength indices. CVB index can be easily calculated in
Multiwfn, see Section 3.200.1 for detail. There are also other papers using ELF to study H-bonds,
e.g. Chem. Rev., 111, 2597 (2011).
(10) Charge variation analysis. Weak interactions often accompanied by charge transfer and
polarization, therefore studying how the electrons are transfered between or within molecules, as
well as how the electron density is polarized due to the presence another molecule are important.
There are many available ways to investigate these points:
·Plotting difference map of electron density between complex and monomers. This is the
most straightforward and intuitive way to study variation of electron density. The procedure is
822
illustrated in Section 4.5.5.

·Plotting charge displacement curve. After generating grid data of density difference, in order
to quantitatively study the charge variation in a direction, you can plot charge displacement curve,
see Section 3.16.14 for introduction and Section 4.13.6 for example.
·Variation of atomic charges of monomers in their isolated states and in complex state can
quantitatively and clearly show how the electrons are transfered between different atoms/fragments
due to the interaction.
·After generating grid data of electron density difference between complex and monomers,
you can use basin analysis module to integrate basin of density difference to study amount of
electron variation in various characteristic regions (e.g. the region corresponding to σ-hole). You
can consult the example in Section 4.17.4.
·Charge decomposition analysis (CDA). CDA is used to reveal underlying details of charge
transfer, the amount of donation and back-donation of electrons between two fragments due to
various complex MOs can be studied. In addition, the CDA module of Multiwfn can tell you how
the fragment MOs are mixed and hence yield complex MOs. CDA is commonly applied to strong
interaction, but it may be also useful for exploring weak interactions. The theory of CDA is
introduced in Section 3.19, practical examples are given in Section 4.16.
·Multiwfn has a function dedicated to analyze charge transfer in electron excitation based on
electron density difference, many important quantities characterizing the transfer can be obtained,
see Section 3.21.3 for introduction and 4.18.3 for example. Based on the grid data of electron density
difference between complex and monomers, this function may be also useful for studying charge
transfer due to weak interaction.
(11) Mutual penetration distance of vdW surfaces. For the same kind of weak interaction,
generally the larger the penetration of vdW surface, the stronger the interaction strength. For a
noncovalently interacting atom pair AB, the difference between the distance of A-B and the sum of
their non-bonded radii is termed as mutual penetration distance. The non-bonded atomic radius is
the closest distance between a nucleus and the molecular vdW surface, and can be obtained by
option 10 in post-processing interface of quantitative molecular surface analysis module of
Multiwfn.
(12) Energy decomposition analysis is a kind of important approachs for characterizing the
nature of weak interactions, physical components of total interaction energy can be separately
obtained. Multiwfn is capable of performing simple energy decomposition analysis in combination
with Gaussian to provide deeper insight into weak interactions, see Section 4.100.8 for example.
Multiwfn can also perform energy decomposition analysis based on molecular forcefield, this
function is very useful, flexible can be used to evaluate/decompose the weak interaction energy for
very large systems, see Section 3.24.1 for introduction and Section 4.21.1 for example.
(13) LOLIPOP index is useful for measuring π-π stacking ability, see the introduction in
Section 3.100.14 and the example in Section 4.100.14.
(14) Source function analysis is defined in the framework of AIM theory. Gatti et al. suggested
using source function to study both strong and weak interactions. Introduction of source function
can be found in Section 2.6, and tutorial of performing source function analysis is given in Section
4.17.5. A thorough review is Struct. & Bond., 147, 193 (2010), in which H-bond analysis is involved.
(15) Atomic multipole moment analysis. The definition of atomic multipole moment can be
found in Section 3.18.3. Atomic multipole moment measures the anisotropy distribution of electron
823
density around an atom, which has important impact on interatomic electrostatic interactions. See
Section 7.4.3 of the Bader's book Atoms in molecules-A quantum theory for illustrative examples.
In Multiwfn, atomic multipole moment can be calculated by both fuzzy space analysis module and
basin analysis module, for the latter case see Section 4.17.1 for example.
(16) Orbital overlap. For weak interactions involving orbital interaction, you can use Multiwfn
to study orbital overlap, which is closely related to orbital interaction strength. The example Section
in 4.100.15 illustrated how to calculate intermolecular orbital overlap integral. Section 4.0.2
exemplified how to visualize overlap degree of two NBO orbitals, high (low) overlap degree
commonly implies large (small) second-order perturbation energy E(2) between the two NBOs.
(17) As demonstrated in J. Mol. Model., 19, 2035 (2013), interaction energy of halogen-bond
complexes is well correlated with the properies of (3,-1) critcial point of Laplacian of electron
density at σ-hole location. The topology analysis of Laplacian of electron density can be
conveniently realized in main function 2. Section 4.2.2 showed how to perform topology analysis
for LOL, however you can use the same method to analyze Laplacian of electron density.
(18) The cubic electrophilicity index defined in conceptual density functional theory
framework has close relationship with strength of weak interaction energy. In J. Phys. Chem. A, 124,
2090 (2020) it is shown that condensed form of cubic at halogen atom in halogen bond dimers has
nice linear relationship with binding energy, therefore this quantity may be useful in predicting
strength of interaction and revealing interaction nature in some cases. This quantity can be
calculated in a fully automatical way via subfunction 16 of main function 100, see Section 3.100.16
for detail.
There are also other possibly ways to study weak interactions, but they are not directly relevant
to Multiwfn. These methods include: NBO E(2) and NBO deletion analyses, rehybridization
analysis (specific for H-bond, based on natural population analysis), variation of bond length and
vibrational frequency, Mayer energy decomposition analysis (Phys. Chem. Chem. Phys., 8, 4630
(2006)), magnetically induced current (Phys. Chem. Chem. Phys., 13, 20500 (2011)), interacting
quantum atoms (IQA, see J. Phys. Chem. A, 117, 8969 (2013) for example), SAPT analysis
(supported by PSI4, Molpro etc. see WIREs Comput. Mol. Sci., 2, 254 (2012)), ETS-NOCV
(supported by ORCA, see J. Chem. Theory Comput., 5, 962 (2009)).
4.A.6 Calculate odd electron density and local electron correlation

function
Part 1: Odd electron density

Some papers calculate the so-called odd electron density (OED), its was originally defined in
Chem. Phys. Lett., 372, 508 (2003), and employed in many papers, e.g. Theor. Chem. Acc., 130, 711
(2011) and J. Phys. Chem. C, 116, 19729 (2012). Here I introduce how to plot it by using Multiwfn
in combination with .wfn file produced by Gaussian.
Spatial natural orbitals are yielded by diagonalizing total density matrix and have occupation
number within 0 and 2.0. The OED of the kth natural orbital is defined as
 kodd (r) = min( 2 − nk , nk ) k (r)
824
where  k (r ) and nk are its electron density and occupation number, respectively. Thus for nk1,
the prefactor corresponds to occupation number, while for nk1, the prefactor corresponds to the
complement to achieve a closed shell. In other words, the min(2-nk, nk) term can be regarded as the
probability of the electron to be unpaired in this orbital.
The total OED is defined as the summation of OED for all natural orbitals:
 odd (r ) =   kodd (r )
k
Now we calculate OED for a closed-shell system OC-BH3 at CCSD level (while at HF/DFT
level, this quantity is obviously zero everywhere). The Gaussian input file can be found in
examples\COBH3_CCSD.gjf, note that density out=wfn keywords are used. The resulting file
examples\COBH3_CCSD.wfn contains all CCSD natural orbitals.
We first calculate total OED. Boot up Multiwfn and input
examples\COBH3_CCSD.wfn
26 // Modify occupation number
odd // Take min(2-nk, nk) as occupation number for all orbitals
q // Return
-1
Then we plot isosurface map of electron density as usual
Then we set isovalue to 0.005, you will see the total OED, as shown below. Although the use
of OED in above mentioned papers aims to study diradical character, in my personal viewpoint, this
function is also useful to reveal where electron correlation is significant. As you can see in the below
graph, electron correlation is very strong in the multiple-bonds region of CO.
Next, we plot OED for natural orbital 11. What we need to do first is cleaning the occupation
number for all other orbitals. Close the GUI and input
825
26 // Modify orbital occupation numbers
1-10 // Select orbitals 1 to 10
0 // Set occupation number to 0
12-57 // Select orbitals 12 to 57
0 // Set occupation number to 0
Then return to main menu and plot electron density as usual. That's all.
Below we plot total OED for a typical diradical system C4H8 at M06-2X level. In this case,
unrestricted open-shell calculation is needed and guess=mix keyword should be used to achieve
symmetry-broken state. In addition, pop=no out=wfn must be specified so that spatial natural orbital
will be generated by mixing alpha and beta density matrix, diagonalization and then be outputted
to .wfn file. The natural orbitals obtained by such a UHF/UDFT calculation are sometimes referred
to as unrestricted natural orbitals (UNO). The Gaussian input file for producing the .wfn file is
examples\C4H8-UNO.gjf, and the resulting .wfn file is examples\C4H8-UNO.wfn. Please use this
file to plot total OED like above example, the isosurface map with isovalue of 0.02 should look like
below (its character is quite similar with spin density map)
Part 2: Local electron correlation functions

The local electron correlation functions proposed in J. Chem. Theory Comput., 13, 2705 (2017)
are very useful real space functions to reveal total, dynamic and nondynamic electron correlation in
various molecular regions. The definition of these function can be found in entry 87, 88 and 89 of
Section 2.7. As you can see, their forms are closely related to OED. Now let us plot local total
electron correlation functions IT for OC-BH3.
Set "iuserfunc" in settings.ini to 87, then boot up Multiwfn and input
examples\COBH3_CCSD.wfn
100 // User-defined function, now it corresponds to IT
Set isovalue to 0.013, then you will see
826
This graph is rather similar with the isosurface map of OED. Since IT is a real space function
specific for revealing electron correlation, our observation clearly demonstrates that OED is also
capable of exhibit electron correlation.
Local electron correlation function has another two forms, namely ID and IND, which aim for
representing dynamic and nondynamic electron correlation, respectively. They can be plotted with
the same way as shown above, but you need to set "iuserfunc" to 88 and 89, respectively.
Plotting individual contribution to local electron correlation functions is also possible, you can
screen uninteresting natural orbitals by setting their occupation numbers to zero via suboption 26 of
main function 6.
Using subfunction 4 of main function 100 you can integrate local electron correlation functions
over the whole space, the result indicates the magnitude of electron correlation for the whole system.
For example, we return to main menu and input
4 // Integrate a real space function over the whole space
The result, which is referred to as total electron correlation index, is 1.576. Repeat this
calculation for dynamic and nondynamic electron correlation functions, you will find the resulting
indices are 1.267 and 0.309, respectively. Evidently, dynamic correlation governs the total electron
correlation effect for present system.
In fact, the total, dynamic and nondynamic electron correlation indices can also be calculated
by subfunction 15 of main function 200, which is much faster and more convenient, see Section
3.200.15 for introduction. Now we have a try. Return to main menu and input 200 then 15, you will
immediately see
Nondynamic correlation index: 0.30880061
Dynamic correlation index: 1.26746042
Total correlation index: 1.57626103
4.A.7 Study (hyper)polarizability density

Chinese version of this section is my blog article " Using Multiwfn to calculate (hyper)polarizability density"
http://sobereva.com/305.
By means of custom operation of main functions 4 and 5, (hyper)polarizability density can be

easily plotted by Multiwfn as plane map and isosurface map. This is quite useful for discussing
nature of (hyper)polarizability of a given molecule. If this feature is used in your work, citing my
paper J. Comput. Chem., 38, 1574 (2017) is recommended, in which the (hyper)polarizability
827
density analysis is involved. Other illustrative applications can be found in e.g. Phys. Chem. Chem.
Phys., 19, 23951 (2017).
Part 1: Basic theory of (hyper)polarizability density

There is a well-known Taylor expansion relationship for dipole moment
E
μ( F ) = − = μ0 + αF + (1 / 2)βF 2 + (1 / 6) γF3 + 
F
E 2E  3E 4E
μ0 = − α=− 2 β=− 3 γ=− 4
F F = 0 F F = 0 F F = 0 F F = 0
where F is external electric field vector, E is system total energy,  and 0 are current electric dipole
moment and permanent dipole moment, respectively. ,  and  are polarizability, the first and
second hyperpolarizability tensors, respectively.
Similarly, Taylor expansion with respect to F can be applied to electron density
 (r, F) =  ( 0) (r ) + ρ(1) (r )F + (1 / 2)ρ( 2 ) (r )F2 + (1 / 6)ρ( 3) (r )F3 + 
 (r )  2  (r )  3  (r )
ρ(1) (r ) = ρ ( 2 ) (r ) = ρ ( 3 ) (r ) =
F F = 0 F2 F = 0 F3 F = 0

Since μ(F) = −  (r, F)rdr , by comparing above equations, we find
μ0 =  − ρ( 0) (r )rdr α =  − ρ(1) (r )rdr
β =  − ρ( 2 ) (r )rdr γ =  − ρ( 3) (r )rdr
where (1) is known as polarizability density, while (2) and (3) are known as the first and second
hyperpolarizability densities, respectively. Using (hyper)polarizability densities, we can easily
investigate contribution of various regions to total molecular (hyper)polarizabilities. In current
example we will only focus on (3). Of course, by analogous treatment we can also easily study (1)
and (2).
The second hyperpolarizability density (3) is a third-order tensor function, it can be explicitly
represented as
 3  (r )
 (r ) =
( 3)
Fi Fj Fk

ijk
F =0
It is impossible to discuss all of its components, since there are as many as 3*3*3=27
components. Assume that the ZZZZ is the most crucial component of , we can simply study ZZZ(r):
 3  (r )
 zzzz =  − z zzz
( 3)
(r )dr  zzz
( 3)
(r ) =
Fz3 F =0
z
Clearly, the − z zzz (r ) is simply the contribution of point r to the ZZZZ. If it is plotted as
( 3)
isosurface map or plane map, the source of ZZZZ can be vividly revealed.
The easiest way of obtaining the  zzz

( 3)
is using finite difference method (see my article
http://sobereva.com/305 on how to derive it)
828
 (2 F z ) − 2 ( F z ) + 2 ( − F z ) −  ( −2 F z )
 zzz
( 3)
=
2( F z )3
where Fz is strength of the external electric field applied along Z axis. The functions such as (Fz)
and (-Fz) denote the electron density distribution yielded when Fz is applied along positive and
negative directions of Z-axis, respectively. The Fz in this case corresponds to finite difference step
size, it should not be too large or too small, otherwise numerical error will be significant. According
my experience, 0.003 a.u. is a good choice of Fz.
Part 2: Preparation work

Below we will study the second hyperpolarizability density for a typical small molecule H2CO.
The orientation of the molecule must be clarified, in current case the C=O bond is parallel to the Z-
axis, as shown below
Before performing hyperpolarizability density analysis, we first carry out a regular static 
calculation using Gaussian, so that we can then judge whether our hyperpolarizability density
analysis is reasonable. The Gaussian input and output files have been provided as polar=gamma.gjf
and polar=gamma.out in "examples\hyperpolar" folder. As can be seen at the end of the output file,
the ZZZZ is 0.145571D+04 a.u. (1455.71 a.u.). The theoretical method we used here is Hartree-Fock,
evidently this is not a good choice for practical studies, but it is acceptable for illustrating purpose.
Next, we need to obtain files containing wavefunctions yielded with -2Fz, -Fz, +Fz, +2Fz
external electric fields, respectively. The corresponding Gaussian input files have been provided in
"examples\hyperpolar" folder, the generated .wfx files are also provided. Note that for Gaussian
users, using .wfx and .fch formats are recommended for present study, while using .wfn format is
highly deprecated, since recording accuracy of wavefunction in this format is relatively poor.
NOTE 1: In general, when applying external electric field, you should add nosymm keyword in Gaussian input
file so that Gaussian will not automatically put the system in standard orientation. However, since I found the
standard orientation of H2CO just meet our expectation, namely C=O bond is parallel to Z-axis, therefore nosymm
keyword is ignored.
NOTE 2: The direction of external electric field applied via field keyword in Gaussian is contrary to common
convention. If 0.003 a.u. field is applied along positive direction of Z-axis, the keyword should be field=Z-30.
Part 3: Draw isosurface map of (3)ZZZ and -z(3)ZZZ

Please first familiarize yourself with custom operation feature of Multiwfn by reading such as
Section 4.5.4. Here we will calculate (3)ZZZ according to the finite difference equation shown earlier
by using custom operation in main function 5, so that we can obtain its grid data and directly it as
isosurface map.
829

examples\hyperpolar\H2CO+2FZ.wfx // The first term in above mentioned equation, namely
(2FZ)
5 // Five files will be operated on the firstly loaded file
-,examples\hyperpolar\H2CO+FZ.wfx // The -2(FZ) term in the equation. The coefficient
of 2 is taken into account by employing subtraction operation twice
-,examples\hyperpolar\H2CO+FZ.wfx
+,examples\hyperpolar\H2CO-FZ.wfx // The 2(-FZ) term
+,examples\hyperpolar\H2CO-FZ.wfx
-,examples\hyperpolar\H2CO-2FZ.wfx // The -(-2FZ) term
2 // Medium quality grid. This is adequate for small system
After calculation, we input
6 // Divide the grid data by a value
0.000000054 // The denominator of the equation, namely 2(Fz)3=20.0033
Now you will see the isosurface map of (3)ZZZ:
However, from this map it is still somewhat difficult to discuss contribution of various regions
to the ZZZZ, because the -z prefactor has not been taken into account. To obtain -z(3)ZZZ, we input
below commands
11 // Performing mathematical operation on grid data
20 // Multiply the grid data by a coordinate variable
Z // Multiply Z coordinate
830
11 // Performing mathematical operation on grid data

5 // Multiply the grid data by a constant
-1 // Multiplying -1. The grid data in memory now has corresponded to -z(3)ZZZ
Now you will see
In the figure, the regions enclosed by green (blue) isosurfaces have positive (negative)
contribution to the γZZZZ. It can be seen that in the molecular valence region, the contribution is
basically negative, however at the two ends of the molecule the contribution is remarkably positive,
and the magnitude obviously exceeds the negative region, this is why the current system has an
evidently positive value of γZZZZ (the value calculated by Gaussian is 1455.71 a.u.). For a large
system, through such a picture, you can immediately explain why γ is big or small, or why it is
positive or negative. Undoubtedly this kind of analysis is extremely useful for study underlying
characteristics of (hyper)polarizability.
It is noteworthy that one can also investigate how basis set and theoretical method affect the
result of (hyper)polarizability by plotting difference between grid data of (hyper)polarizabilities
generated by two different levels as isosurface map, namely first calculating the two sets of grid
data, and then use mathematical operation in main function 13 to obtain grid data of their difference.
One can verify if the γZZZZ evaluated by integrating the grid data of -z(3)ZZZ is close to the γZZZZ
directly evaluated by Gaussian. We close the GUI of main function 13, enter its option 17 and then
input 1, the statistical data of the grid data over the whole space will be printed, from it you can find
below line:
Integral of all data: 1429.5132718689
This is the γZZZZ obtained by integrating the cubic grids of the -z(3)ZZZ, it is quite close to the
1455.71 a.u. outputted by Gaussian, showing that our calculation steps are correct and the result is
meaningful. (If we use high quality grid and increase the extension distance of the grid data from
the default value to 10 Bohr, then the result will be 1459.43 a.u., which is closer to the γZZZZ
outputted by Gaussian)
831
Part 4: Plot plane map of (3)ZZZ and -z(3)ZZZ

Using custom operation feature in main function 4 of Multiwfn, the (3)ZZZ and -z(3)ZZZ can
also easily plotted as various kinds of plane map. We first plot contour line map for (3)ZZZ, input
below commands:
examples\hyperpolar\H2CO+2FZ.wfx
4 // Plot plane map
5 // Five files will be operated on the firstly loaded file
-,examples\hyperpolar\H2CO-2FZ.wfx
Press ENTER button to employ default grid setting
10 // Increase to 10 Bohr
3 // YZ plane
0 // X=0
Close the graph that pops up and then input
-7 // Multiply the data by a value
18518518.5185185 // Namely the inverse of the denominator 0.000000054
-1 // Replot
Then you will see below map, which corresponds to a slice of the (3)ZZZ isosurface map shown
earlier. The solid and dashed lines correspond to positive and negative regions, respectively.
832
Next, we plot plane map for -z(3)ZZZ. Quit Multiwfn, set "iuserfunc" in settings.ini to 23, we
do this because in this case the user-defined function will correspond to the -z function, if we repeat
above steps but replace the function 1 (electron density) with function 100 (user-defined function),
then what we plotted will just correspond to -z(3)ZZZ. The resulting map is shown below. You can
try to further improve the graph quality by using the lots of options provided in Multiwfn.
Part 5: Rapidly export grid data of (3)ZZZ and -z(3)ZZZ

In main function 13, you can find corresponding option used to export grid data in memory as
cube file. In addition, as described in Section 5.2, Multiwfn can be run in silent mode. Therefore, it
is possible to rapidly export cube file of -z(3)ZZZ and (3)ZZZ. The input stream file for this purpose
has been provided as examples\hyperpolar\hyperdens.txt, please open it to check its content.
Assume that all relevant .wfx files have been placed in proper folder, then after running Multiwfn <
hyperdens.txt, you will find density.cub and -Zrho3zzz.cub in current folder, they correspond to cube
file of (3)ZZZ and -z(3)ZZZ, respectively.
Part 6: Obtain atomic contribution to (hyper)polarizability

Thanks to the flexibility of Multiwfn, one can easily obtain atomic contribution to
(hyper)polarizability by integrating corresponding (hyper)polarizability density in each atomic
space based on its grid data. Commonly, employing fuzzy atomic space is recommended, because
the cost is very low. Here we will calculate atomic contributions to ZZZZ by integrating grid data of
-z(3)ZZZ.
Set "iuserfunc" in settings.ini to -1, so that the user-defined function will correspond to the
interpolated function based on the loaded grid data. Then boot up Multiwfn, load the cube file
corresponding to -z(3)ZZZ into it, then input below commands:
1 // Integrate a real space function in every atomic space
The result is
833
Atomic space Value % of sum % of sum abs
1(C ) 80.42330427 5.643648 5.643648
2(H ) 428.30264522 30.055833 30.055833
3(H ) 428.19625953 30.048367 30.048367
4(O ) 488.10117815 34.252152 34.252152
Summing up above values: 1425.02338717
The values under "Value" label are atomic contributions to ZZZZ, and the values under the "%
of sum" are percentage contributions. From the data it is clear that the two hydrogens and the oxygen
have major contribution to ZZZZ. The sum of all values is 1425.02 a.u., which is also quite close to
the ZZZZ value (1455.71 a.u.) outputted by Gaussian polar=gamma task.
Appendix 1: Equation for polarizability density and first hyperpolarizability density

Below expressions are given so that you can also easily evaluate polarizability density and the
first hyperpolarizability density by following above mentioned steps.
Polarizability density:
 ( F z ) −  (− F z )
 (1)
z =
2F z
 zz =  −  z(1) (r ) z d r
First hyperpolarizability density:
 ( F z ) − 2  (0) +  ( − F z )
 zz(2) =
( F z )2
 zzz =  −  zz(2) (r ) z d r
Appendix 2: Making molecular dipole moment parallel to a Cartesian axis

Above I introduced how to evaluate (hyper)polarizability density for a specific Cartesian
component (Z), e.g. (2)ZZ and (3)ZZZ. However, what we are actually interested in is often the
component along the direction of molecular dipole moment. To study the nature of this quantity in
terms of (hyper)polarizability density, we need to reorientate the molecule so that its dipole moment
is exactly parallel to a Cartesian axis, this could be easily done via VMD program
(http://www.ks.uiuc.edu/Research/vmd/).
For example, if after calculation the dipole moment vector is found to be (-1.8916 0.7861 0.0),
you should convert the current molecular geometry to a format that can be recognized by VMD,
such as .pdb and .xyz. Then load the file into VMD, run below commands in VMD console window:
set sel [atomselect top all]
$sel move [transvecinv "-1.8916 0.7861 0"]
Then the molecule will be reorientated and the dipole moment vector will just point to the positive
direction of X-axis. If then you also run $sel move [transaxis z 90], the dipole moment will point
towards Z-direction. After that, choose "File" - "Save coordinate" to save current geometry to a .xyz
file, finally you can use geometry in this file to carry out single point calculation with external field
to yield needed wavefunction. Notice that if you are a Gaussian user, do not forget to properly use
"nosymm" keyword to avoid automatic reorientation during the single point calculation.
834
4.A.8 Analyze wavefunction higher than CCSD level
In the most commonly used program Gaussian, the highest level of wavefunction is CCSD.
Although CCSD wavefunction is absolutely sufficient for almost all cases, due to some special
reasons, one may want to study wavefunction produced at higher level. Below I describe how to
make Multiwfn able to analyze CCSD(T) wavefunction generated by PSI4 program
(http://www.psicode.org), and arbitrary order of coupled cluster and CI wavefunction (including
Full CI) yielded by Kallay's MRCC program (http://www.mrcc.hu).
(1) PSI4
The version of PSI4 I currently use is 1.2.1. Below is an example of input file, which calculate
hydrogen fluoride at CCSD(T)/cc-pVTZ level, and produce HF_CCSDpT.fchk in current folder.
molecule HF {
H 0.0 0.0 -0.831975
F 0.0 0.0 0.092442
set basis cc-pVTZ
grad, wfn = gradient('CCSD(T)', return_wfn=True)
fchk_writer = psi4.FCHKWriter(wfn)
fchk_writer.write('HF_CCSDpT.fchk')
The resulting HF_CCSDpT.fchk records Hartree-Fock MOs and CCSD(T) density matrix. If
you directly feed this file into Multiwfn, because Multiwfn never utilizes density matrix but only
load orbitals from the file, the result of following analyses will correspond to Hartree-Fock level. In
order to make Multiwfn analyze CCSD(T) wavefunction, you should do below steps:
1. Boot up Multiwfn and load HF_CCSDpT.fchk as usual
2. Enter main function 200 and select subfunction 16. This function is used to transform density
matrix in the .fch/.fchk file into natural orbitals, see Section 3.200.16 for more detail.
3. Input CCSD, then “Total CCSD Density” field in the .fchk file will be loaded, and you will
immediately see occupation numbers of natural orbitals (NOs) yielded by diagonalization of
CCSD(T) density matrix.
4. Input y. Then new.molden is generated in current folder, which records NOs at CCSD(T)
level. This file is automatically loaded into Multiwfn, therefore the orbitals in memory now
correspond to NOs of CCSD(T) wavefunction, and thus all following analyses will correspond to
CCSD(T) wavefunction.
Note that if this is an open-shell system, you can choose the type of NOs that to be generated,
including, spatial NOs, alpha/beta NOs and spin NOs. See Section 3.200.16 for more detail.
(2) MRCC
The version of MRCC I currently use is Sept 25, 2017. Below is an example of input file, which
calculate hydrogen fluoride at CCSDT/cc-pVTZ level.
basis=cc-pvtz
calc=CCSDT
mem=2500MB
dens=1
835
geom=xyz
H 0.0 0.0 -0.831975
F 0.0 0.0 0.092442
After run it by MRCC, you will find a file named MOLDEN in current folder, which is a
Molden input file and records Hartree-Fock MOs. In current folder you can also find a file named
CCDENSITIES, which records the 2nd and 1st reduced density matrix (2RDM and 1RDM). In
order to make Multiwfn analyze CCSDT wavefunction, we must convert the 1RDM into natural
orbitals and save them to .molden file.
Change the name of MOLDEN to e.g. mrcc.molden, then boot up Multiwfn and load it. Enter
main function 1000 and select subfunction 97, input path of the CCDENSITIES file. Then input the
number of frozen core orbitals. By default, MRCC freezes core MOs in electron correlation
calculation. Current system has two core electrons (in the output file it can be seen that " Number
of core electrons: 2"), and this is a closed-shell system, each occupied MO has two electrons,
therefore only one core MO is frozen, so we input 1. Soon, generation of natural orbitals by
diagonalizing CCSDT density matrix is finished and occupation numbers are printed on screen. We
use subfunction 2 of main function 100 to export a .molden file. then reboot Multiwfn and load it,
now both the basis function information and GTF information in memory correspond to CCSDT
wavefunction, therefore all the following analyses will be carried out at CCSDT level.
The procedure of analyzing CI wavefunction produced by MRCC is exactly the same as those
shown above. Below is an example input file of calculating elongated LiH at FCI/aug-cc-pVDZ
level without frozen core treatment.
basis=aug-cc-pvdz
calc=fci
mem=2500MB
dens=1
core=0
geom=xyz
H 0.0 0.0 0.0
Li 0.0 0.0 3.0
4.A.9 Calculate TrEsp (transition charge from electrostatic potential)

charges and analyze excitonic coupling
1. Theory about TrEsp

General form of electrostatic potential (ESP) of a molecule, say A, can be written as
ZI  (r ' )
aaA ' (r ) =  a , a '  −  aa' dr'
I | r − RI | | r − r' |
836
where ZI and RI are nuclear charge and coordinate of atom I, respectively.  is Kronecker function.
a,a is transition density between state a and a'.
The ESP we commonly studied is ESP of a single state, i.e. a=a'. When a and a' correspond to
different states, the potential may be referred to as "transition electrostatic potential", which
measures ESP exerted by excition corresponding to a-a' transition.
It is known as exact ESP of a single state can often be well approximately represented as the
potential evaluated based ESP fitting charges (e.g. CHELPG and MK charges, see Sections 3.9.10
and 3.9.11),
ZI  (r ' ) qaI
aA (r ) =  − a dr'  
I | r − RI | | r − r' | I | r − RI |
I
where qa is ESP fitting charge of atom I at electronic state a.
In light of this, J. Phys. Chem. B, 110, 17268 (2006) puts forward the concept of TrEsp
(transition charge from electrostatic potential), and shows that exact transition ESP can be well
approximated as below
 aa' (r' ) I
qaa
 (r ) = − 
A
dr'   '
| r − r' | I | r − RI |
aa'
I
where qaa' is TrEsp of atom I derived from transition density of a-a'.
The way of calculation of TrEsp charges is almost exact the same as evaluation of common
ESP fitting charges, the only differences are that the nuclear contribution should be ignored, and
density of a single state should be replaced with transition density between two states.
2. Example of calculating TrEsp charges

Now I use a simple molecule 4-nitroaniline to illustrate how to calculate TrEsp charges for its
S0-S2 transition. Here I assume you are a Gaussian user (If you prefer to use ORCA program, the
steps will be slightly lengthy, but can be significantly simplified via writting shell script. Please
check #2 of this post for detail: http://sobereva.com/wfnbbs/viewtopic.php?pid=389).
First, run the Gaussian input file examples\4-Nitroaniline_TrESP.gjf, the keywords

PBE1PBE/6-31g(d) TD density=transition=2 out=wfn mean transition density between ground
state (S0) to S2 will be generated at TD-PBE0/6-31G(d) level, and then it will be automatically
diagonalized to yield corresponding natural orbitals, which are finally saved to specified .wfn file.
If you are confused or do not have Gaussian in hand, you can directly download related files from
http://sobereva.com/multiwfn/extrafiles/TrEsp.zip
S0S2.wfn // The .wfn file generated in above process
837
12 // CHELPG fitting method (you can also use MK or RESP method instead)
5 // Choose form of ESP
3 // Transition electronic (i.e. the ESP specific for evaluating TrEsp)
Calculation of ESP for even medium-sized systems is time-consuming, you need to wait
patiently. Finally, the TrEsp charges are shown on the screen:
Center X Y Z Charge
1C -0.006197 0.042893 2.284588 0.213411
2C -0.006197 -2.564075 2.283674 -0.167947
3C -0.004195 -3.920061 0.000000 0.184157
...[ignored]
14N 0.001111 4.086240 0.000000 -0.064630
15O 0.003902 5.154581 -2.036539 -0.094910
16O 0.003902 5.154581 2.036539 -0.094899
RMSE: 0.000970 RRMSE: 0.044908
The sum of charges is exactly zero, which is what we expected, becuase electronic transition process
does not alter total number of electrons.
Beware that, the above outputted TrEsp charges must then be manually divided by √2! This
is because the natural orbitals in the exported .wfn file were generated based on symmetrized form
of transition density matrix (TDM), however the symmetrization was done via a strange way by
Gaussian, namely TDMi,j=(TDMi,j+TDMj,i)/√2 rather than TDMi,j=(TDMi,j+TDMj,i)/2 as expected,
therefore this problem should be manually fixed via dividing the resulting charges by √2.
In fact, in Multiwfn the transition charge can also be calculated by Mulliken method via hole-
electron analysis module, see Section 3.21.1.3, and the computational cost is almost negligible.
However, Mulliken transition charges must not be as good as TrEsp charges for approximately
representing transition electrostatic potential and analyzing intermolecular excitonic coupling
purposes.
Skill 1: Accelerating calculation of TrEsp by making use of cubegen utility

Making use of cubegen utility in Gaussian package can decrease cost of ESP relevant analyses
in Multiwfn and is thus highly recommended, please read Section 5.7 for detail. The cubegen can
also be used to reduce computational cost of TrEsp charges, as will be shown below.
Since cubegen calculate ESP based on density matrix information in .fch/fchk file, we must
first generate TDM and store it into a .fch file, the function mentioned in Section 3.21.9 can do this.
We first use PBE1PBE/6-31g(d) TD IOp(9/40=4) keywords to carry out electron excitation
calculation and meantime keep the corresponding .fch, the corresponding files for 4-Nitroaniline
are 4-Nitroaniline_IOp.gjf, 4-Nitroaniline_IOp.out and 4-Nitroaniline.fchk in the aforementioned
TrEsp.zip package.
4-Nitroaniline.fchk
9 // Generate and export TDM
1 // Generate TDM between ground state and excited state
838
4-Nitroaniline_IOp.out
2 // Generate TDM between S0 and S2
y // Symmetrize the resulting TDM in usual way, namely TDMi,j=(TDMi,j+TDMj,i)/2
y // Export TDM.fch, whose density matrix field corresponds to the just generated TDM
Please make sure that "cubegenpath" parameter in settings.ini has been set to actual path of
cubegen utility in Gaussian folder, then reboot Multiwfn and input
TDM.fch
12 // CHELPG fitting method
3 // ESP specific for evaluating TrEsp
Since cubegen calculates ESP quite fast, immediately the TrESP charges are shown on screen.
You do not need to manually divide the result by √2, because the TDM generated by Multiwfn has
been symmetrized in correct way.
It is worth to note that if you want to verify whether the fitted TrEsp charges are reasonable,
you can compare the electric dipole moment computed via these charges and the transition electric
dipole moment printed by Gaussian (or other quantum chemistry codes). As it is well known that
ESP fitting charges are able to well reproduce electric dipole moment, commonly the TrEsp charges
are also able to well reproduce actual electric transition dipole moment.
After calculation of TrEsp, we choose y to let Multiwfn export the charges to TDM.chg file in
current folder. Then boot up and load this file, you will find below information on screen
Component of electric dipole moment:
X= -0.011607 a.u. ( -0.029501 Debye )
Y= -1.781254 a.u. ( -4.527495 Debye )
Z= 0.000011 a.u. ( 0.000029 Debye )
In 4-Nitroaniline_IOp.out you can find below information

Ground to excited state transition electric dipole moments (Au):
state X Y Z Dip. S. Osc.
1 -0.0000 -0.0000 0.0001 0.0000 0.0000
2 -0.0165 -1.7911 0.0000 3.2083 0.3408
3 0.0210 0.0188 0.0000 0.0008 0.0001
Since the electric dipole moment evaluated based on our TrEsp charges is very close to the
exact transition electric dipole moment, it is clear that our TrEsp charges must be reasonable.
Skill 2: Imposing customized charge constraint during TrEsp fitting process

The restrained electrostatic potential (RESP) module of Multiwfn has been detailedly
introduced in Section 3.9.16. As you can see, this module is more general and more powerful than
the MK or CHELPG module, because you can arbitrarily impose customized constraints on the
resulting ESP fitting charges, for example, you can request some atoms must have exactly the same
charge, or request sum of a batch of charges must equal to a predefined value.
Here, I present an example to illustrate how to use the RESP module to calculate TrEsp based
on MK fitting grid with additional constraint that atomic charge of all hydrogens must be zero.
Firstly, write a plain text file (e.g. chgcons.txt) with below content:
7 0
839
8 0
9 0
10 0
This file will be used in the RESP module. The 7~10 are atom indices of the hydrogens, the 0 means
their charges will be constraint to zero during fitting.
S0S2.wfn // The .wfn file we previously used
18 // RESP module
3 // Transition electronic
6 // Set charge constraint in one-stage fitting
1 // Load charge constraint setting from external plain text file
chgcons.txt // The file containing charge constraint
2 // Start one-stage ESP fitting calculation with customized constraint. The default fitting grid
is MK (you can also change to CHELPG by option 3)
The result is
Center Charge
1(C ) 0.158613
...
7(H ) 0.000000
8(H ) -0.000000
9(H ) 0.000000
10(H ) -0.000000
11(N ) 0.178112
...
Clearly, our charge constraints have been in effect, and all other atoms still have reasonable
TrEsp charges. You can learn more about the RESP module by reading corresponding example in
Section 4.7.7. It is worth to note that when you select “Transition electric” in option 3, the default
atom equivalence constraint is automatically removed and the restraint strength in one-stage fitting
is automatically set to zero, since these treatments are not useful in current case.
3. Evaluating excitonic coupling energy based on TrEsp

General form of intermolecular Coulomb interaction energy can be expressed as
ZI ZJ  (r )  b,b (r' )
A B
V A, B
=  a ,a ' b,b '  +   a ,a drdr'
IA J B | R I − R J | | r − r' |
aa',bb'
Z I  bB,b (r ) Z  A (r )
−  a ,a '   dr −  b,b '   J a ,a dr
IA | RI − r | J B | RJ − r |
This quantity may have different physical meaning. For example
V00A,,00
B
: Coulomb interaction energy between A and B in their ground states
V00A,,11
B
: Coulomb interaction energy between A in ground state and B in the first excited state
840
V01A,,10B = V10A,,01
B
: Excitation energy transfer couplings between transition of the two molecules
Calculation of the integrals in VaaA,',Bbb' is difficult, there is a method known as transition density
cube (TDC), which calculates the integrals by numerical integration based on evenly distributed
grids, its cost is extremely high for large system. Fortunately, it was shown that by using TrEsp
charges calculated for two molecules, their excitonic coupling energy VaaA,',Bbb' can be readily
evaluated using below formula at commonly satisfactory accuracy:
qaI ,a ' qbJ,b'

VaaA,',Bbb'  
IA J B | R I − R J |
In Multiwfn, you can easily calculate excitonic coupling energy based on TrEsp charges of two
molecules. The steps are briefly outlined below:
(1) Optimize dimer structure
(2) Extract coordinate of each monomer and write it into Gaussian input file, properly change
keywords and perform electron excitation calculation to yield .wfn file containing natural orbitals
derived from transition density. Note that nosymm keyword must be used to avoid Gaussian
automatically translating and rotating the overall monomer coordinate.
(3) Generate TrEsp charges for each monomer using respective .wfn file generated at last step,
then export TrEsp charges as .chg file.
(4) Manually combine content of the two monomer .chg file as a single .chg file. The monomer
coordinate in this file should be consistent with optimized dimer coordinate.
(5) Load the dimer .chg file into Multiwfn, enter main function 7 and select option -2, then
input atom list of the two monomers in turn, the excitonic coupling energy will be printed
immedately.
4.A.10 Intuitively exhibiting atomic properties by coloring atoms
There is a very useful way of intuitively exhibiting atomic properties calculated by Multiwfn,
namely coloring atoms in VMD program (http://www.ks.uiuc.edu/Research/vmd/), here I illustrate
how to do that via two examples. More detailed discussions and examples can be found in my blog
article "Using Multiwfn+VMD to exhibit atomic charges, spin populations, charge tansfer and
condensed Fukui function via coloring atoms" (http://sobereva.com/425).
(1) Coloring atoms according to atomic charges

First, I illustrate how to use this manner to vividly represent atomic charges of polyyne. This
system was also involved in Section 4.13.6.
The first step is calculating atomic charges. Boot up Multiwfn and input below commands:
examples\polyyne.wfn
11 // ADCH charge (this type of charge is generally recommended)
1
y // Export atomic coordindates and atomic charges to polyyne.chg in current folder
841
Now reboot Multiwfn, then input

polyyne.chg
2 // Export new file
1 // The format of the new file is .pqr
polyyne.pqr
Now we have polyyne.pqr in current folder. The .pqr format is very similar to the popular .pdb
format, the major difference is that in the .pqr format the last two columns are specific for recording
atomic charges and atomic radii, respectively. In current file, the atomic charges correspond to the
polyyne ADCH charges, while the atomic radii correspond to Bondi van der Waals radii.
The .pqr file can be recognized by VMD. We boot up VMD, then drag the polyyne.pqr into
VMD main window to load it. After that, we modify plotting settings:
(a) Use white background: Inputting color Display Background white in VMD console
window
(b) Modifying drawing style and coloring setting: Enter "Graphics" - "Representation", set the
drawing method as "CPK", set the coloring method as "Charge". Then choose "Trajectory" tab, input
-0.4 and 0.4 in the "Color Scale Data Range" text boxes and press ENTER button.
(c) Changing color transition style: Enter "Graphics" - "Colors", choose "Color Scale" tab,
change the default RWB to BWR (Blue-White-Red)
(e) Choose "Display" - "Orthographic" to use orthographic perspective.
Now you will see below graph in VMD OpenGL window. The atoms at both ends are hydrogen,
all the other atoms are carbon.
In above graph, the red and blue colors reflect that the atom has positive and negative charge,
respectively. The deeper red (blue) the more positive (negative) the charge. As can be seen, since
carbon has larger electronegativity than hydrogen, the two hydrogens have evident positive charge
and the carbons bonded to them have evident negative charge. The white color indicates that the
charge of the carbons in the middle region of the molecule is close to zero.
The polyyne is a highly conjugated system, it is expected that external field could significantly
polarize its charge distribution. To study this problem, we use the same procedure to plot the map
based on examples\polyyne_field.wfn, which was generated under 0.03 a.u. external electric field
along molecular axis. The resulting graph is shown below, the direction of the external electric field
is from the right side to the left side.
It can be seen that the atomic charge distribution is no longer symmetric. Since the source of
the field is at right side, large amounts of electrons transferred from left to right, as a result, the net
charge of the carbon atoms at left side become positive, while the ones at right side become negative.
(2) Coloring atoms according to atomic contribution to molecular orbitals

The atomic coloring method is not only able to be employed to exhibit atomic charges, but can
also be used to exhibit other atomic properties. As an example, I illustrate how to represent atomic
842
contributions to molecular orbitals by coloring atoms, examples\N-phenylpyrrole.fch is taken as

example molecule.
First, we calculate orbital composition of an orbital. Boot up Multiwfn and input
8 // Orbital composition
3 // SCPA method
36 // Select MO 36 as example
Then we copy all atomic contributions from the Multiwfn window to a text file using the
method described in Section 5.4.
0 // Return
2 // Export new file
2 // The format of the new file is xyz, because .xyz is very similar to .chg
N-phenylpyrrole.chg // Name of the new file
Now manually modify the N-phenylpyrrole.chg by your favorable text editor (Ultraedit is
recommended), delete the first two lines, and copy the orbital composition to the last column using
column mode, then save the file. Finally, the content of the N-phenylpyrrole.chg should be
C -0.00000000 1.12162908 1.82507914 0.396870
C -0.00000000 0.71310006 3.13457424 0.099898
C -0.00000000 -0.71310006 3.13457424 0.099898
...[ignored]
H 0.00000000 2.14896316 -3.05414223 0.000000
H 0.00000000 -0.00000000 -4.31975533 0.000000
We use Multiwfn to load this .chg file and convert it to N-phenylpyrrole.pqr, then use VMD to
visualize it using exactly the same procedure described in last example. However, this time the lower
and upper limit of color scale should be set to -50 and 50, respectively. The resulting graph is shown
as left part of below map; as a comparison, the corresponding isosurface map of MO36 is shown at
right part.
843
The more red the atom, the greater its contribution to the orbital. As can be seen, the atomic
coloring introduced in this section well reflects actual orbital distribution. For very large molecules,
the isosurface map may become quite complicated, while the atomic coloring map should be much
clearer.
Of course, the atomic coloring method is also applicable to other kinds of atomic properties
calculated by Multiwfn, such as condensed Fukui function, atomic spin population, atomic transition
charge, source function of atoms, integral of electron energy in atomic space, variation of atomic
charge during electronic transition or intermolecular interaction. More examples can be found in my
blog article http://sobereva.com/425 (in Chinese).
4.A.11 Overview of methods for studying chemical bonds

PS: Chinese version of this Section is http://sobereva.com/471, which also contains extended discussion.
In this section, I present an overview of all methods that may be used to study chemical bonds.
You will find Multiwfn is indispensably useful in characterzing and unraveling nature of the bonds.
Most analyses can be applied to both ground state and excited states (see Section 4.18.13 for more
information about this point)
1 AIM (Atoms-in-molecules) analysis

In the framework of AIM, the bond critical point (BCP) is the most representative point of a
bond, hence character of chemical bonds can be characterized by various properties at
corresponding BCPs, for example:
·Electron density and potential energy at BCP, namely (BCP) and V(BCP), are often used
to discuss bonding strength. For the same kind of bond, they are usually positively and negatively
correlated to bonding strength, respectively.
·Laplacian of electron density at BCP, namely 2(BCP), is often used to judge whether or
not a bond mainly shows covalent character. Negative and positive values imply that the major
nature of the bond is covalent and non-covalent, respectively. But notice that this criterion is often
wrong (e.g. CO has positive 2(BCP) but it is evidently a polar covalent bond)
·In Angew. Chem. Int. Ed. Engl., 23, 627 (1984) it was argued that negative and positive
values of energy density at BCP, i.e. H(BCP), implying the bond has covalent and non-covalent
nature, respectively. But this criterion is not always true; for example, the Ca-O in CaO is typical
ionic bond, but its H(BCP) is negative.
·The V(BCP)/G(BCP) was proposed in J. Chem. Phys., 117, 5529 (2002), where G(BCP)
denotes Lagrangian kinetic energy density at BCP. It was argued that <0, >1 but <2, >2 of this
quantity respectively imply that the bonding mainly belongs to close-shell interaction, intermediate
(mixed) interaction and covalent interaction.
·The eta index was proposed in J. Phys. Chem. A, 114, 552 (2010) and further studied in
Angew. Chem. Int. Ed., 53, 2766 (2014), it is defined as |1(r)|/3(r), where 1 and 3 are the smallest
and largest eigenvalues of Hessian matrix of electron density, respectively. It was argued that if eta
index at BCP is smaller than 1, then the bonding should be closed-shell interaction; while if it is
larger than 1, the interaction should has covalent nature, and the more positive the value, the stronger
the covalent character. However I found this argument is not always true, for example this quantity
844
of both Ni-C and C-O bonds in Ni(CO)4 is less than 1, but undoubtedly they should be attributed to
polar covalent bonds.
·Bond degree (BD) was proposed in J. Chem. Phys., 117, 5529 (2002) and defined as
H(BCP)/(BCP). The physical meaning of BD is energy density of unit electron at BCP. For
covalent interaction (usually H(BCP)<0), the more negative the BD, the stronger the bonding; while
for non-covalent interaction (usually H(BCP)>0), the more positive the BD, the weaker the
interaction.
·Bond ellipticity was proposed in J. Am. Chem. Soc., 105, 5061 (1983) and defined as (r)=
[1(r)/2(r)]-1. At BCP the 1 and 2 must be negative and exhibit the curvature of electron density
perpendicular to the bond. The larger the value deviates to 0, the stronger the tendency that the
electron density has unsymmetric distribution in the plane perpendicular to the bond at the BCP.
·The source function with BCP as reference point has been employed to study chemical bonds,
see Struct. & Bond., 147, 193 (2010) for comprehensive review and Section 4.17.5 for analysis
example.
There are some other research papers utilized BCP properties to discuss chemical bonds, for
example J. Am. Chem. Soc., 120, 13429 (1998) and J. Comput. Chem., 39, 1697 (2018). It is worth
to note that properties at certain critical points can also be used to estimate metallicity of crystals,
see J. Am. Chem. Soc., 124, 14721 (2002), Chem. Phys. Lett., 471, 174 (2009) and J. Phys.: Condens.
Matter, 14, 10251 (2002).
The bond path is also a very important concept in the AIM framework, it rigorously reveals the
main interaction path connecting various atoms. It is important to understand the fact that chemical
bond must be accompanied by a bond path and BCP, while present of bond path and BCP does not
imply existence of chemcal bond.
AIM topology analysis has been systematically introduced in Section 3.14 and illustrated in
Section 4.12.1, all above mentioned quantities can be easily and rapidly evaluated by Multiwfn. It
is noteworthy that in Multiwfn many real space functions at critical point (or specific point) can be
decomposed as contributions from various orbitals (usually MOs or LMOs), see Section 4.2.4 for
example; moreover, any real space function can be plotted along bond paths, see Section 4.2.3 for
example. These useful features can often provide much deeper insight into the bonding.
The above mentioned real space functions can also be plotted as curve map, plane map or
isosurface map so that one can visually study their distribution, see Section 4.3, 4.4 and 4.5 for
practical example, respectively. The contour line map and isosurface map of 2(r) is particularly
useful and frequently employed.
2 Bond order and delocalization index analysis

Bond order is a very useful and straightforward way of characterizing chemical bonds.
Multiwfn supports a lot of bond order definitions, please check Section 3.11 for detailed introduction.
Different bond orders have different characters and physical meanings. For example, Laplacian
bond order (LBO) measures covalent component of a bond and usually has good relationship with
bond dissociation energy (BDE), while Mayer bond order essentially reflects the number of
electrons shared by two interacting atoms. The bond analysis module of Multiwfn is also able to do
more things than just calculating the value of bond order. For example, Multiwfn can decompose
some bond orders as contributions from various orbitals, the Wiberg bond order can be decomposed
as contributions from atomic orbital pairs. Many detailed analysis examples of bond order are given
in Section 4.8.
845
It is worth to mention that the delocalization index (DI) is essentially equivalent to Mayer bond
order and fuzzy bond order in physical nature. The difference arises from how the atomic spaces are
defined. For nonpolar bonds, the DI is usually very close to Mayer and Fuzzy bond orders, but they
quantitatively differ for polar bonds. For DI, the AIM atomic basins are employed as atomic spaces.
The DI can be calculated via basin analysis module, see Section 3.18.5 for detailed introduction of
DI and Section 4.17.1 for DI analysis example. Commonly I do not suggest employing DI, because
its cost is by far higher than evaluation of Mayer and Fuzzy bond orders.
For the same kind of chemical bond, e.g. C-O bond in different transition metal coordinates,
the Mayer bond order and LBO are positively correlated with bonding strengh. For different kinds
of chemical bonds, Mayer bond order should not be used to compare bonding strength, for example,
the bond in N2 and that in P2 has evidently different BDEs, but they have basically the same Mayer
bond order because both of them are typical triple bonds. In contrast, the LBO is able to faithfully
reflect that the bond in P2 is much weaker than N2. Please check original paper of LBO (J. Phys.
Chem. A, 117, 3100 (2013)) for more discussions, comparisons and examples.
Plotting variation of bond order versus reaction coordinate is an absolutely very good idea to
shed light on the underlying change of electronic structure in a chemical reaction, see Section 4.A.1
on how to easily realize this.
3 Bond order density and natural adaptive orbital analysis

The concept of bond order density (BOD) and natural adaptive orbital (NAdO) has been
introduced in detail in Section 3.200.20, they are fairly useful if you want to graphically discuss
bond order of a given covalent bond. The BOD is a real space function representing everywhere
contribution to bond order (strictly speaking, delocalization index in the present context), while
NAdO unveils nature of bond order in terms of orbitals. See Section 4.200.20 for application
examples, you will find this method particularly useful in many situations.
4 Orbital localization analysis

Molecular orbitals (MOs) commonly are unable to be used to study bonding character because
they are highly delocalized and do not directly correspond to chemical bonds. The orbital
localization is a very powerful technique, it can transform the MOs to localized molecular orbitals
(LMOs), which are highly localized and have very close relationship with bonding. Via LMOs,
numerous useful information about chemical bonds can be extracted, such as bond polarity, bond
multiplicity, bond type, the atomic orbitals that participate in the bonding and so on. Please check
Section 3.22 for detailed introduction of LMOs and follow the LMO analysis examples in Section
4.19.
5 AdNDP analysis
The purpose of the adaptive natural density partitioning (AdNDP) method is somewhat similar
to orbital localization method, the advantage of AdNDP is that it is also able to derive orbitals with
semi-delocalization character from the complicated multi-electron wavefunction. If the AdNDP
analysis has been properly performed, then the resulting orbitals will faithfully reveal all multi-
center bonds in current systems. The drawback of AdNDP analysis is that user must manually pick
out orbitals from candidate list, this process is slightly cumbersome and requires the user has
adequate chemical intuition. When there is no multi-center bond, using orbital localization is much
preferred over AdNDP because it is fully automatic, fast and free of subjectivity; while if you suspect
that present system may have evident mulit-center bonds and you want to study them, commonly
846
AdNDP is the only choice. The AdNDP method is detailedly introduced in Section 3.17, related
examples are provided in Section 4.14.
6 Analysis of ELF and relevant real space functions

ELF is a very important real space function, it is able to reveal localization and delocalization
of electrons in chemical systems. Brief introduction of ELF can be found in Section 2.6. In Multiwfn
ELF can be analyzed in many different ways, as shown below
·Visualizing study. In Multiwfn, the ELF can be drawn as curve map by main function 3,
drawn as plane map by main function 4 and plotted as isosurface map via main function 5, see
Sections 4.3, 4.4 and 4.5 for examples, respectively. From the graphs one can easily identify which
region contains evident covalent interaction (i.e. evident share of electrons) and judge the nature of
a given chemical bond. In addition, the bond multiplicity can be inferred from the shape of ELF
isosurface around the bond. Multiwfn is also capable of studying ELF- and ELF-, so that 
interaction and  interaction can be studied separately, see Sections 4.5.3 and 4.100.22 for example.
Note that there are a lot of real functions having analogous distribution feature as ELF, though
their underlying ideas may be not very similar to ELF. Multiwfn supports most of them and they
can also be plotted in the exactly the same way as ELF. These real space functions include LOL,
SCI, SEDD, RoSE, PS-FID and DORI. The LOL is introduced in Section 2.6 and sometimes
preferred over ELF; introduction of other real space functions can be found in Section 2.7. Section
4.20.5 presented an example of analyzing the DORI function, the special feature of DORI compared
to ELF is that DORI is able to reveal all kinds of interactions, including both covalent and non-
covalent ones.
The negative part between two atoms in 2 map is able to reveal the region where electrons
concentrate due to formation of covalent bond, this point is similar with ELF. In J. Phys. Chem.,
100, 15398 (1996), Bader believes that 2 and ELF are homeomorphic and their similarities and
differences are able to provide complementary information in understanding chemical bonds.
However, notice that for bonds involving very heavy atom, 2 map often fails to reveal covalent
character. For example, 2 in the interacting region of Re-Re bond is entirely positive.
By using Multiwfn and shell script as well as third-part software, anime of ELF or other
functions during a chemical process (often represented as trajectory resulting from intrinsic reaction
coordinate or rigid scan tasks) can be easily generated, such an anime is able to very vividly exhibit
variation of characters of chemical bonds, see Section 4.A.1 on how to make the anime.
·Basin analysis of ELF (or similar functions): This kind of analysis can be carried out via
basin analysis module (main function 17), see Section 4.17.2 for example. All ELF basins
collectively make up the whole space, each ELF basin corresponds to a region with featured
electronic structure. For example, the ELF basins may correspond to covalent bond, lone pair, core
region, etc. By analyzing character of bond basins, one can acquire many information about the
bonds, such as average number of electrons that occurs in the bonding region, degree of electron
localization in the bonding region, dipole moment of the bonding region. Contribution of each atom
to the electron population in the bonding region can also be obtained, as illustrated in Section 4.17.7.
·Topology analysis of ELF (or similar functions): This kind of analysis allows one to obtain
accurate position of ELF maximum (also known as ELF attractor) and (3,-1) type of ELF critical
point (also known as ELF bifurcation point), the former displays the most representative point of a
ELF basin, while the value of the latter somewhat reflects the extent of electron sharing (degree of
delocalization) between two ELF basins. Topology analysis of ELF can be realized via main
847
function 2, see Section 4.2.2 for example. For practical studies based on topology analysis of ELF
and LOL, see Nature, 371, 683 (1994) and J. Comput. Chem., 30, 1093 (2009). Tracing variation of
ELF attractors is particularly useful for understanding the change in electronic structure and bonding
character during a chemical process, illustrative examples of such analysis are RSC Adv., 5, 62248
(2015), Chem. Phys., 501, 128 (2018) and Comput. Theor. Chem., 1154, 17 (2019).
Note that basin analysis is also able to give positions of ELF attractors, the procedure is even more simple than
using topology analysis module, however the accuracy of the positions given by basin analysis module is not as good
as topology analysis module, since basin analysis is carried out based on even-distributed grids.
7 Analysis of valence electron density

As clearly illustrated in my article Acta Phys. -Chim. Sin., 34, 503 (2018), visualizing electron
density of valence electrons is a very useful way of revealing electron structure and studying
character of chemical bonds, please carefully read this paper. Moreover, basin analysis can be
applied to valence electron density to unveil more information of chemical interest. See Section
4.6.2 for example on how to carry out this kind of analysis.
8 Electron density difference analysis

Formation of chemical bond always leads to significant electron reorganization (polarization
and charge transfer), in particular, formation of covalent bond must be accompanied with the
phenomenon that electrons concentrate to the bonding region. Plotting electron density difference
(EDD) map is one of the best ways to reveal this point, EDD can be very easily plotted as curve
map, plane map and isosurface map in Multiwfn via mainfunctions 3, 4 and 5, respectively. EDD
can be defined in different ways, if you want to study the bond formed between two fragments, you
should study the EDD between the whole system and the two fragments, see Section 4.5.5 for
example; if you want to study the reorganization of electron density due to forming bonds between
various atoms in the system, you should study deformation density, which is defined as the
difference between the electron density of the whole system and all atom in their isolated states, see
Do not forget that Multiwfn also provides advanced technique for analyzing EDD, for example,
basin analysis can be applied to EDD, see Section 4.17.4 for example. Also one can plot charge
displacement curve to better quantitatively study electron reorganization along specific direction,
see Section 4.13.6 for example.
It is worth to note that plotting difference map of ELF between whole system and its fragments
is also valuable, see illustration in Section 4.4.8.
9 Analysis of g function and IBSI index

The real space function g is defined in the framework of IGM theory, see Section 3.23.5 for
introduction. g is capable of revealing all kinds of interactions, including both chemical bonding
and weak interaction, as well as both covalent and non-covalent ones. Moreover, the magnitude of
g in the bonding region is often positively correlated with bonding strength, therefore one can
easily examine bonding strength in different regions by inspecting colors in color-filled map or by
properly adjusting isovalue in isosurface map. In addition, the isosurface of g can be mapped by
sign(2) function via various colors, this makes the isosurface map informative. Please check
Sections 4.20.10 and 4.20.11 for IGM examples; although the examples focus on studying weak
interactions, the same procedure can also be migrated to chemical bond analysis.
The intrinsic bond strength index (IBSI) is defined based on integral of g over the whole space.
In J. Phys. Chem. A, 124, 1850 (2020) is was shown that it has the ability to measure bonding
848
strength and distinguish type of bonds to a certain extent, see Section 3.11.9 for introduction and
10 Quantifying amount of charge transfer due to bonding

Formation of chemical bond between two different fragments must result in detectable charge
transfer (CT) between the two fragments. The amount of CT can be obtained as difference between
the fragment charge in actual system and the net charge of the fragment in its isolated state. The
fragment charge is defined as sum of charges of the atoms in the fragment. In the population analysis
module of Multiwfn, if a fragment has been defined, the fragment charge will be directly outputted
when calculating atomic charges, see Section 4.7.1 for example.
11 Charge decomposition analysis (CDA)

Using fragment charge we can easily discuss the total amount of CT, however, in order to
examine details of charge transfer due to bonding, the CDA must be employed. CDA is able to
explicitly show electron donation and back-donation between each pair of user-defined fragments
at resolution of orbital interactions, also it provides clear information about how the MOs of the
entire system is composed of MOs of individual fragments. See Section 3.19 for introduction and
Section 4.16 for example.
12 Density-of-states (DOS) analysis

The partial DOS (PDOS) curve map is useful for intuitively exhibiting bonding and anti-
bonding due to interaction between user-defined fragments (may be defined as a batch of atoms,
shells or atomic orbitals) at various energy range, see Section 3.12 for introduction and Section
4.10.1 for example.
13 Energy decomposition analysis

Energy decomposition analysis is used to decompose the bond energy to different physical
components to provide deeper insight into the bonding nature. The “simple energy decomposition”
supported by Multiwfn can be applied to chemical bonds, please check Section 4.100.8 for example.
You need Gaussian to use this function.
14 Studying bond polarity

It is often interesting to study polarity of a bond, there are several possible ways, as shown
below. The results often differ significantly, since the concept itself cannot be uniquely defined.
·Calculating respective contribution from the two bonding atoms to the LMO corresponding
to the bond (A and B), then the ionicity of the bond can be evaluated as |A-B|. Clearly the larger
this value, the higher the bond polarity. To obtain A and B, you should first perform orbital
localization, then find the LMO corresponding to the bond in main function 0, and finally use main
function 8 to evaluate composition of the LMO via proper method.
·Firstly evaluating respective contribution from the two bonding atoms to the population
number of the ELF basin corresponding to the bond, as illustrated in Section 4.17.7, then take the
difference of the two contribution values to estimate the bond polarity.
·Calculating bond polarity index. See Section 3.200.12 for introduction and 4.200.12 for
example.
· It is worth to note that Laplacian bond order only reflects covalent component of a bond,
while Mayer bond order may be regarded as total bond order. Therefore, in certain cases, the
difference between Laplacian and Mayer bond orders may be used to reveal bond polarity.
849
15 Studying bond dipole moment

There are three possible ways in Multiwfn:
·Calculate bond dipole moment based on two-center localized molecular orbitals, please
check introduction in Section 3.22 and example in Section 4.19.4.
·Perform ELF basin analysis and check dipole moment of the basin corresponding to the bond
of interest. See Section 4.17.2 for illustration. At the same time, quadrupole moment of the bond
basin can also be obtained.
·Calculating bond dipole moment in Hilbert space. See Section 3.200.2 for introduction
16 PAEM (potential acting on one electron in a molecule) analysis

The PAEM refers to the total potential acting on an electron at a point. By analyzing PAEM at
proper positions between two atoms, the nature (covalent or non-covalent) can be determined. See
Section 4.3.3 for illustration.
4.A.12 Overview of methods for analyzing electron excitation
In this section, I present a systematic overview of all methods supported by Multiwfn that can
be used for analyzing electron excitation problems.
Note: Chinese version of this section corresponds to my blog article http://sobereva.com/437.
1 Hole-electron analysis
All kinds of excitations can be essentially described as "hole-to-electron" transition, that is,
"hole" is the region where the excited electron leaves, and "electron" is the region where the excited
electron eventually goes. Hole-electron analysis corresponds to subfunction 1 of main function 18,
see Section 3.21.1 for introduction and 4.18.1 for illustration. This analysis is very powerful and
universal and is an almost indispensable analysis method for all kinds of electron excitation
problems. Specifically, the hole-electron analysis has below capacities:
·Displaying isosurfaces of hole and electron. From this picture, one can intuitively understand
how electrons are excited
·Transforming the hole and electron distributions to a form described by Gaussian function,
making them significantly easier to examine visually
·Calculating quantitative indices that measure characteristics of electron excitation, including
the Sr index, which measures the degree of overlap of hole and electron; the D index, which
measures the distance between hole and electron centroids; the  index, which measures the breadth
of hole and electron distributions; the t-index, which measures degree of separation of hole and
electron, and so on.
·Plotting density difference map, which corresponds to subtracting hole from electron
·Calculating contribution of basis functions, atomic orbitals, atoms, molecular fragments and
molecular orbitals to hole and electron, so that the nature of hole and electron can be thoroughly
analyzed. Moreover, amount of hole and electron as well as hole-electron overlapping extent on
various atoms and fragments can be directly displayed as heat map (color-filled matrix map), which
is very convenient for visual horizontal comparison.
·Calculating Coulomb attraction between hole and electron, which is a common definition of
exciton binding energy.
2 Nature transition orbital (NTO) analysis
850
When doing electron excitation calculations, it is often found that many orbital transitions have
negligible contribution to electron excitation, this phenomenon makes viewing orbitals to discuss
electron excitation characteristics difficult, and in this case it is necessary to examine multiple
orbitals simultaneously. After transforming the molecular orbitals to NTOs using subfunction 6 of
main function 18, for most cases the electron excitation can be solely described by only one pair of
NTO transition, thus making the discussion much simpler. See Section 3.21.6 for introduction of
NTO analysis and Section 4.18.6 for practical example.
3 Λ index and Δr index

The Λ index proposed in 2008 may be the earliest index to quantitatively examine characteristic
of electron excitations, its intrinsic physical meaning is a measure of degree of overlap between
electron and hole. The Δr proposed in 2013 is another index for characterizing electron excitation
based on the idea of Λ index. The Δr essentially measures the centroid distance between electron
and hole. Λ and Δr are described in detail in Sections 3.21.14 and 3.21.4, and they can be computed
via subfunctions 14 and 4 of main function 18, respectively.
In fact, with the Sr and D indices defined in the hole-electron analysis framework, it is no longer
necessary to use the Δr and Λ indices, since Sr and D are in principle more significant in physical
meaning. However, since Multiwfn is able to calculate Δr and Λ for a large number of selected
excited states simultaneously, if you simply want to roughly examine electron excitation
characteristics for a batch of excited states at once, employing Δr and Λ is still a good choice.
4 IFCT analysis
The full name of IFCT is "interfragment charge transfer", which is a method proposed by me
to estimate amount of electron transfer between atoms or fragments in the process of electron
excitation. The calculation cost is extremely low. This method has been detailed described in Section
3.21.8 and illustrated in Section 4.18.8. Although using difference between fragment charge of
excited state and that of ground state can also study variation of electron population during electron
excitation, one cannot understand details of charge transfer at "who transferred to whom" level,
therefore IFCT analysis has important and irreplaceable practical value for investigating problems
of electron excitation. In particular, when studying transition metal coordinates, exact amount of
MC, LC, LLCT, MLCT and LMCT can be separately evaluated by the IFCT analysis.
5 Analysis based on density difference between excited state and ground state
Density difference analysis is a prevalently used and widely accepted method for studying the
difference in charge distribution between two electron states of a system. Multiwfn supports a
variety of analysis methods based on the density difference between excited state and ground state,
as shown below:
·Plotting density difference map
First of all, Multiwfn can easily calculate density difference between excited state and ground
state and plot it as curve map, plane map and isosurface map through main functions 3, 4, 5, see
examples in Section 4.3, 4.4 and 4.5, respectively. Moreover, not only the density difference
between excited state and ground state can be drawn, but also the density difference between two
excited states can be easily drawn via Multiwfn, see illustration in Section 4.18.13.
·Smoothing density difference and calculating statistical data of density difference
The original density difference map between excited state and ground state is not easy to
examine, because its positive and negative regions are interlaced and appear to be messy. After
851
calculating the density difference grid data, one can use subfunction 3 of main function 18 to
transform it to replace the positive and negative parts of the density difference with very smooth
Gaussian functions, then the image will become much more intuitive and easier to analyze. At the
same time, the program outputs various statistical data about the density difference, such as centroid
coordinates of positive and negative parts, charge transfer distance, degree of separation between
positive and negative parts. See introduction in Section 3.21.3 and example in Section 4.18.3.
·Local integral curve and charge displacement curve
If the system under studying is linear or an interface system (such as a dye molecule attached
to TiO2 surface), the local integral curve and charge displacement curve can be plotted along the
direction of the molecular chain or perpendicular to the interface. The local integral curve shows
the integral value of the density difference on each section perpendicular to the chosen direction,
while the charge displacement curve shows the integral of density difference from the beginning
side to the current position. These two kinds of maps are useful to quantitatively study electron
transfer feature along a certain direction. It is easy to draw these two kinds of graphs in Multiwfn,
please check Section 3.16.14 of the manual for introduction and Section 4.13.6 for example.
·Basin integration for density difference
Multiwfn is able to perform basin integration for density difference, so that one can study
variation of number of electron in some featured local regions, see Section 4.17.4 for example.
6 Analyzing difference between excited state and ground state in electron population or
atom/fragment charges
Main function 7 is used to perform population analysis or atomic charge calculation, and if a
fragment is defined by subfunction -1 before evaluating atomic charges, fragment charge will also
be given in the output. See corresponding examples in Section 4.7. After calculating the fragment
charges of excited state and ground state separately, the difference between them can be used to
understand how many electrons were lost or gained at different fragments during the electron
excitation, and thus the influence of electron excitation on the charge distribution can be investigated
at quantitative level.
Although IFCT analysis is able to realize the same purpose, the advantage of using
atomic/fragment charge to discuss this problem is that there is a large room of choice of the method
for evaluating atomic charges, and the charge distribution of excited state can correspond to relaxed
density.
7 Draw transition density isosurface map, plotting heat map of transition density matrix
The transition density matrix (TDM) is very useful for unveiling the underlying nature of an
electron excitation. TDM has two forms:
(1) The three-dimensional real space form, which can be expressed by drawing isosurface map.
Large value at a point corresponds to large overlap of hole and electron at this place, see Section
3.21.1.1 for detailed introduction and Section 4.18.2.1 for analysis example.
(2) The matrix form in the common sense. This form of TDM can be exhibited as heat map
(namely color-filled matrix map), which may be atom-based or fragment-based. Its diagonal
elements vividly show which atoms or fragments are simultaneously occupied by hole and electron,
while the non-diagonal elements directly reflect the direction and extent of electron transfer between
corresponding atoms or fragments. See Section 3.21.2 for introduction of the TDM heat map and
Section 4.18.2.2 for analysis example.
852
8 Analyzing heat map of charge transfer matrix

If each atom is defined as a fragment in the aforementioned IFCT analysis, the amount of
charge transfer between various atoms and the amount of charge redistribution within each atom
will constitute a matrix, which is referred to as "atom-atom charge transfer matrix" by me and may
be further contracted to fragment-fragment charge transfer matrix. Both of the matrices can be
drawn as heat maps by subfunction 2 of main function 18, see Section 3.21.2 and 3.21.8 for
introduction as well as Section 4.18.8 for practical example. The information carried by the charge
transfer matrix heat map is very similar to the TDM heat map, and the way of analysis is exactly the
same, but the charge transfer matrix is more strictly defined and has clearer physical meaning.
Moreoever, the charge transfer matrix is completely in agreement with the hole and electron
distributions given by the hole-electron analysis module, therefore I believe the charge transfer
matrix map analysis is a better method than the popular TDM heat map analysis.
9 Analyses on transition dipole moment

For absorption process, the larger the oscillator strength of an electron excitation, the stronger
the corresponding absorption peak. The transition probability between two excited states is mainly
determined by oscillator strength, which is proportional to square of corresponding transition
electric dipole moment. Therefore, it is very meaningful to conduct an in-depth analysis on intrinsic
factors affecting the transition electric dipole moment. Multiwfn provides a number of functions for
decomposing transition dipole moment (including both the electric one and magnetic one), as
described below.
·Drawing transition dipole moment density
The transition dipole moment density is a function that measures the contribution of a point in
the three-dimensional space to the transition dipole moment, and its integral over the whole space
is exactly equal to the transition dipole moment. Obviously, if the transition dipole moment density
is plotted as isosurface map or plane map, contribution of each region to the transition dipole
moment can be vividly exhibited. See Section 3.21.1.1 for introduction and Section 4.18.2.1 for
example.
·Drawing heat map of transition dipole moment matrix
Subfunction 2 of main function 18 can draw heat map of transition dipole moment matrix,
either atom-based or fragment-based. The sum of all matrix elements is exactly the transition dipole
moment of the system, so the diagonal elements in the map show the contribution of atoms or
fragments to the transition dipole moment solely by themselves, while the non-diagonal elements
reflect the atom-atom or fragment-fragment coupling contribution to transition dipole moment.
Clearly, the internal structure of transition dipole moment can be clearly understood via this kind of
heat map. See Section 4.18.2.3 for analysis example.
· Decomposing transition dipole moment to basis function contributions and atom
contributions
Subfunction 11 of main function 18 can decompose transition dipole moment into contribution
of each atom and each basis function, see Section 3.21.11 for details. Moreover, based on the data
outputted by Multiwfn, via a VMD script one can draw arrows to exhibit contribution vector of
custom fragments to transition dipole moment, so that the contribution of each part of the system to
the transition dipole moment can be intuitively understood, see Section 4.18.11 for example.
·Decomposing transition dipole moment into contributions of orbital transitions
Subfunction 10 of main function 18 can decompose transition dipole moment into contribution
853
of each orbital transition, and at the same time the program outputs the oscillator strength evaluated
based on the current electron excitation information. Therefore, when many orbitals have significant
participation in electron excitation, this function can be used to immediately identify which orbital
transitions have critical influence on the oscillator strength, so that further discussion can be made.
In addition, one can set configuration coefficients of some orbital transitions to zero in subfunction
-1 of main function 18, and then enter this function again to check influence on the oscillator strength
due to ignoring those orbital transitions. See corresponding introduction in Section 3.21.10 and
example in Section 4.18.10.
·Calculating transition dipole moments between excited states and dipole moment of each
excited state
Transition dipole moments between excited states are important for some studies. For example,
they are needed by the sum-over-states (SOS) method, which can be used to calculate
(hyper)polarizability (see Section 3.200.8); in addition, simulating transient absorption spectrum
needs oscillator strength (f) between excited states, while evaluation of f requires transition dipole
moment between corresponding two excited states. In Multiwfn, subfunction 5 of main function 18
can evaluate transition dipole moments between excited states, and dipole moments of each excited
state can also be directly outputted. See Section 3.21.5 for details about this function.
10 Analyzing excited state wavefunction

Multiwfn is extremely powerful on electronic structure analysis, the analyses can not only be
applied to ground state, but can also be applied to excited state, as long as the input file contains
excited state wavefunction. Note that if the excited states were calculated by
CIS/TDHF/TDDFT/TDA-DFT methods, the input file must record natural orbitals (NOs) of
corresponding excited state. By using Multiwfn, the NOs can be generated based on the excited
state density matrix in .fch file, see Section 3.200.16 for detail; the NOs can also be generated based
on configuration coefficients, as shown in Section 3.21.13.
After loading excited state wavefunction into Multiwfn, one can carry out a variety of
electronic structure analyses. For example, main function 9 can be used to calculate various kinds
of bond orders for excited state, main function 7 can perform population analysis and calculate
atomic charges for excited state, main functions 3,4,5 are able to plot more than one hundred of real
space functions for excited state, AIM analysis can be applied to excited state by main functions 2
and 17, weak interaction of excited state can be visually studied via main function 20, excited state
aromaticity can be investigated via a bunch of methods in Multiwfn (see Section 4.A.3). By
comparing analysis result of excited state and ground state, the impact on electronic structure caused
by electron excitation can be fully shed light on.
11 Orbital composition analysis

Multiwfn has a very powerful orbital composition analysis module (main function 8), which
supports all orbital composition analysis methods. Via this function, one can study the MOs or NTOs
that mainly involved in electron excitation to make clear the role that played by various atomic
orbitals, atoms and fragments.
12 Examining overlapping extent and centroid distance between orbitals

Subfunction 11 of main function 100 is used to calculate overlapping extent and centroid
distance between two selected orbitals. Evidently, this function is useful for studying electron
excitation. For example, using this function to analyze the MO pair or NTO pair that dominates the
854
electron excitation, one can investigate charge displacement and separation degree during the
electron excitation.
13 Evluating atomic transition charges

The atomic charge we commonly say is for a single electronic state, it is essentially determined
by density matrix of this state. It is also possible to calculate charge for each atom using transition
density matrix between two states, these charges are known as atomic transition charges. Just as the
method of calculating atomic charges is not unique, there are many different methods for calculating
the atomic transition charges. Multiwfn can calculate Mulliken atomic transition charges, see
corresponding description in Section 3.21.12. Multiwfn can also calculate atomic transition charges
via electrostatic potential fitting method, J. Phys. Chem. B, 110, 17268 (2006) and some other
literatures called this kind of charges as TrEsp (transition charge from electrostatic potential). Basic
theory and calculation example of TrEsp can be found in Section 4.A.9 of the manual. The main use
of atomic transition charges is quickly calculating electrostatic potential corresponding to transition
density, thereby examining the exciton coupling between molecules, this point is also described in
detail in Section 4.A.9.
14 Investigating contributions of orbital transitions to electron excitation

Computing contribution of an orbital transition to electron excitation is rather simple, see
beginning of Section 3.21 for introduction. For facilitating analysis, when you enter subfunction -1
of main function 18, contribution of ten orbital transitions that have largest contributions to the
selected electron excitation will be directly listed, see Section 3.21.0 for more information.
15 Identifying ghost states

Asymptotic behavior of exchange potential of pure or hybrid DFT functionals with low HF
exchange composition is obviously incorrect. When TDDFT with such exchange-correlation
functional is used to calculate excited states of large conjugate systems, a batch of artificial charge
transfer excited states with low energy tend to occur. The ghost states have no any physical meaning,
their existences not only wastes computation time, but may also cause beginners to mistake a ghost
state as an emission state. The ghost-hunter index proposed in J. Comput. Chem., 38, 2151 (2017)
can be used to diagnose whether an excited state produced by TDDFT calculation is a ghost state.
This index is automatically outputted after performing hole-electron analysis analysis, see Section
3.21.7 for detailed introduction and 4.18.1 for example. If a ghost state is found, the researchers can
avoid these states in their discussions, or try to eliminate these states by using a DFT functional with
higher HF exchange composition or long-range corrected functionals.
16 Evaluating contribution of NBO orbitals to electron transition

As sufficiently exemplified in Section 4.200.13.3, contribution of NBO orbitals to electron
transition can be obtained by fitting NBO orbital densities to density difference between two
electronic states. Since NBO orbitals often have clear feature and chemical meaning, this method is
able to provide deeper insight into the nature of electron transitions. The same module can also be
used to study contribution of any other kind of orbitals (e.g. LMO) to electron transitions, see
Section 3.200.13 for theory and algorithm introduction.
Others
Subfunction 15 of main function 18 is able to quickly print all major molecular orbital
transitions in every excited state, this is useful if you want to examine basic characteristics of each
855
electron excitation in terms of molecular orbitals. See Section 3.21.15 for introduction of this
function.
Do not forget Multiwfn has main function 11, which can plot UV-Vis and ECD spectrum based
on the oscillator/rotatory strengths and excitation energies outputted by quantum chemistry codes.
This module is by far more powerful and flexible than any other plotting tools and able to provide
detailed information about the spectrum. Please check Section 3.13 for introduction and Section
4.11 for abundant examples.
It is also worth mentioning the biorthogonalization method, which may be also useful in
studying the nature of triplet excited state, that is this method can usually describe the triplet excited
state calculated by UKS or UHF method in terms of orbital transition model, so that discussion the
nature of the excitation could be simplified. See Section 3.100.12 for introduction and 4.100.12 for
example.
Finally, note that only for the aforementioned entry 5 (density difference analysis), entry 6
(atom/fragment charge analysis) and entry 10 (excited state wavefunction analysis), arbitrary
electron excitation calculation method could be used as long as they can produce excited state
wavefunction. For example, for the density difference analysis, the difference can be made between
the electron density of the lowest triplet excited state and singlet state calculated by KS-DFT, the
difference can also be made between excited state density produced by EOM-CCSD and the ground
state density yielded by CCSD. While for other kinds of analyses, such as hole-electron analysis,
IFCT analysis, only CIS, TDHF, TDDFT and TDA-DFT can be employed for calculating excited
states.
4.A.13 Plot electrostatic potential colored van der Waals surface map
and penetration graph of van der Waals surfaces
Note 1: I strongly suggest looking at this video tutorial https://youtu.be/QFpDf_GimA0, which clearly and
sufficiently illustrates most content in this section.
Note 2: Average local ionization energy (ALIE) on molecular surface can also be plotted via VMD script, see
Note 3: Chinese version of this tutorial is http://sobereva.com/443, it contains more discussion and examples
than this section.
1 Foreword
In the tutorial "Plotting electrostatic potential colored molecular surface map with ESP surface
extrema via Multiwfn and VMD" (http://sobereva.com/multiwfn/res/plotESPsurf.pdf), I detailedly
described how to plot electrostatic potential (ESP) colored molecular van der Waals (vdW) surface,
this kind of map is very important and frequently occurs in literatures. However there are a large
number of steps in the tutorial. In order to make plotting this kind of map as easy as possible, here
I introduce a script-based method to draw similar graph, and meantime I will introduce how to plot
penetration graph of vdW surfaces, which is very useful for discussing intermolecular interactions.
However, I still suggest you also read the tutorial "plotESPsurf.pdf" after reading the content in
present section, so that you can understand more details and are able to manually improve the effect
of the obtained graph.
2 Preparation
VMD program is needed in the present plotting, it can be freely downloaded from
http://www.ks.uiuc.edu/Research/vmd/, the version I used here is 1.9.3. Here I assume that you are
856
using Windows system, however the method described below is also suitable for Linux system, see
Part 9 of this section.
All files utilized below have been given in "examples\drawESP" folder, they are briefly
introduced here:
• .bat files: Batch process files of Windows system. They are used to invoke Multiwfn to
calculate needed data for plotting graphs in VMD. The content of the files is very easy to understand
and can be easily modified. If you do not know how to run Multiwfn in silent mode, please check
Section 5.2
• .txt files: Input stream files of Multiwfn that involved in the .bat files.
• .vmd files: VMD plotting scripts.
Before plotting, you should do below things:

(1) Move all .bat and .txt files to the folder containing Multiwfn executable file
(2) Modify the VMD path in the .bat files to actual path of VMD in your machine
(3) Copy all .vmd files to VMD folder
(4) Add below content to the end of the vmd.rc file in VMD folder:
proc iso {} {source ESPiso.vmd}
proc iso2 {} {source ESPiso2.vmd}
proc pt {} {source ESPpt.vmd}
proc pt2 {} {source ESPpt2.vmd}
proc ext {} {source ESPext.vmd}
Hint: When .fch/fchk file is used as input file, the computational cost of ESP may be evidently
reduced by allowing Multiwfn to invoke cubegen utility in Gaussian package, see Section 5.7 for
detail.
3 Plotting ESP colored vdW surface for single molecule

Here we take acetamide as example. Move the CH3CONH2.fch in "examples" folder to the
folder containing Multiwfn executable file, modify the file name to 1.fch. Double click the
ESPpt.bat, Multiwfn will be invoked to carry out quantitative molecular surface analysis (main
function 12) for the 1.fch, once the calculation has done, the exported mol1.pdb and vtx1.pdb will
be automatically moved to the VMD folder. Then boot up VMD and input command pt in VMD
console window, then the ESPpt.vmd will be activated to load the mol1.pdb and vtx1.pdb to draw
below map:
The default lower and upper limits of color scale is -50 and 50 kcal/mol, respectively, and the
857
default color transition is BWR (Blue-White-Red), therefore in above map the white area
corresponds to the region having almost vanished ESP value, while the red and blue points have
evidently positive and negative ESP, respectively. You can manually change the default setting by
modifying the ESPpt.vmd, the settings can also be changed in VMD GUI interface, see the
plotESPsurf.pdf tutorial for detail.
In above map, the ESP colored vdW surface is represented in terms of surface vertices, the
graph can also be drawn in an alternative way, namely mapping ESP on electron density isosurface,
we do this now. Double click the ESPiso.bat, then Multiwfn will be invoked to calculate and export
cube file of electron density and ESP, the resulting density1.cub and ESP1.cub will be automatically
moved to the VMD folder. Then boot up VMD and input command iso in VMD console window,
then the ESPiso.vmd will be activated to load the two cube files to draw below map. Notice that in
order to gain slightly better effect, I used the bulit-in Tachyon render to obtain below graph, namely
selecting "File" - "Render", change to "Tachyon (internal, in-memory rendering)" and click "Start
Rendering" button (The resulting file is in .tga format, you need to use advanced image viewer to
view it, such as IrfanView, which is freely available at https://www.irfanview.com).
This graph looks more pretty than the former one. Unfortunately, calculation of ESP grid data
is much more time-consuming than calculation of ESP on surface vertices, since the number of
surface vertices is by far less than the number of grid points.
4 Show ESP extrema on molecular surface simultaneously

It is possible to append the ESP surface extrema on the graph. To do this, double click
ESPext.bat, it will do all things that ESPpt.bat do, but it also outputs surfanalysis.pdb and moves it
to the VMD folder. This file records all surface extrema. Then boot up VMD and input command pt
or iso to draw corresponding map first, and then input ext, then ESPext.vmd will be activated to load
the surfanalysis.pdb and render the surface extrema as small spheres. The combination of pt+ext
and iso+ext are shown at left and right sides of below graph, respectively. Note that in order to make
ESP extrema at backside visible, I have changed the material of electron density isosurface to
"Transparent" (namely enter "Graphics" - "Representation", switch to "density1.cub", change
"Material" to "Transparent". If you want to make this as default setting, modify the ESPiso.vmd and
change the "$id EdgyGlass" to " id Transparent")
858
In above graph, the orange and cyan spheres correspond to the positions of maxima and minima of
ESP on the vdW surface. You can also manually mark ESP values on the extrema by image editor,
see the plotESPsurf.pdf tutorial on how to do this.
As I have mentioned in Section 4.12.1, even for neutral system, there may be some surface
minima (maxima) with positive (negative) value, which are often chemically insignificant and can
be ignored. If you do not want to plot them on the graph, you can replace the content of ESPext.txt
by that of examples\drawESP\ESPext_noinsig.txt. The additional four lines in this file with respect
to ESPext.txt is used to remove these insignificant extrema.
5 Plotting ESP colored penetration map of vdW surface of monomers

Here I use water tetrameter to illustrate how to plot this kind of map. The files used in this
instance are provided in "examples\water_tetramer\fch" folder. The Gaussian input files of the four
water molecules are 1/2/3/4.gjf, respectively, their coordinates were directly extracted from the
optimized tetramer coordinate, which can be found in complex.gjf. Run these .gjf files by Gaussian,
you will obtain the 1/2/3/4.fch. Notice that nosymm keyword has been employed, otherwise the
Cartesian coordinates of the monomers will be no longer consistent with those in complex, because
without this keyword Gaussian will automatically put the systems to standard orientation.
Copy the 1/2/3/4.fch files to the folder containing Multiwfn executable file, run ESPpt.bat, then
Multiwfn will be invoked to calculate the four .fch files in turn, the resulting mol1/2/3/4.pdb and
vtx1/2/3/4.pdb will be automatically moved to the VMD folder. Then boot up VMD and input pt2
to activate the ESPpt2.vmd script, you will immediately see left part of below graph. If you run
ESPiso.bat and then input iso2 in VMD instead, then the ESPiso2.vmd will be activated to draw the
right part of below graph based on the exported density1/2/3/4.cub and ESP1/2/3/4.cub.
859
From above map, the inter-penetration between the vdW surfaces of the four monomers due to
formation of hydrogen bonds can be clearly seen. In addition, the mapped colors show that the
tetramer was formed in ESP positive-negative complementary way, revealing the electrostatic nature
of the hydrogen bonds.
As an exercise, please try to plot the ESP colored vdW surface penetration map of Guanine-
Cytosine dimer via above two ways, the .fch file of the two monomers can be downloaded at
http://sobereva.com/multiwfn/extrafiles/GC_fch.rar. Notice that before plotting, you should
manually delete the .pdb and .cub files generated for previous systems in VMD folder.
6 Hint: On the adjustment of material

For some systems, the ESP colored map plotted via iso command is not quite ideal. For
example, the below map looks messy
In this case, you can enter "Graphics" - "Materials", set "EdgyGlass", which is the material
currently used for representing the surface. Then adjust its each setting, especially the "Opacity". If
we change the setting to below case, you will find the difference of ESP on the vdW surface now
can be distinguished more clearly.
860
7 Hint: Plotting ESP mapped vdW surface for very huge systems
Sometimes we need to plot ESP mapped surface for systems consisting of several hundreds of
atoms, in this case even single point calculation using DFT with 6-31G* is very expensive or
computationally infeasible. To plot the map for this case, below is my suggested steps:
• Perform single point task or optimization task via Grimme's xtb code
(https://www.chemie.uni-bonn.de/pctc/mulliken-center/software/xtb/xtb). The xtb conducts all
calculations based on GFN-xTB theory, which can be regarded as a semi-empirical version of DFT.
--molden argument should be employed to make xtb output Molden input file (molden.input). Since
xtb is extremely fast, even for a system composed of several hundreds of atoms, single point task
can be finished within no more than 1 minutes on a personal computer.
• Load the molden.input into Multiwfn, then use corresponding option in subfunction 2 of main
function 100 to export .fch file (e.g. xtb.fch).
• Make sure that "cubegenpath" in settings.ini has been properly set. Ensure that you have
defined GAUSS_MEMDEF environment variable, see Section 5.7 for detail.
• load the xtb.fch into Multiwfn, using main function 5 to calculate grid data and export cube
file for electron density and ESP in turn, the exported density.cub and totesp.cub should be manually
renamed to density1.cub and ESP1.cub, respectively. Notice that "High quality grid" should be
employed since the system is large. The calculation of ESP is relatively time-consuming, for
example, using a common Intel 4 cores CPU, the time cost for a system with 336 atoms takes half
an hour.
• Move the density1.cub and ESP1.cub, as well as the aforementioned ESPiso.vmd in
examples\drawESP\ to VMD folder.
• Boot up VMD and input source ESPiso.vmd in VMD console window.
Now you can see the ESP colored vdW surface map. I suggest you also properly adjust material
setting, as mentioned earlier. Below is a system containing 336 atoms.
861
Even if you only have a personal computer with 4 cores, you can obtain above map starting
from a structure file with total time cost of no more than 1 hour. While if you have a server with
dozens of cores, the graph can be obtained within 10 minutes.
It is worth to note although the xtb code is extremely fast, the ESP quality corresponding to the
wavefunction generated by xtb is usually satisfactory. According to my test, no evident difference
can be found between the ESP colored molecular surface graph generated based on the xtb
wavefunction and the high-quality B3LYP/def2-TZVP wavefunction.
More information about this topic can be found in my article "Rapidly plotting electrostatic
potential colored van der Waals surface for huge system" (in Chinese, http://sobereva.com/481).
8 Other worthnoting points

Color bar of ESP can be appended on the map, see illustration near the end of this video:
https://youtu.be/QFpDf_GimA0.
I strongly suggest readers to check the content of the .bat, .txt and .vmd files to make clear how
they work. As you can see, the ESPpt.bat and ESPiso.bat can at most deal with four .fch file
(1/2/3/4.fch), you can also extend them to more molecules. In the ESPpt2.vmd and ESPiso2.vmd,
the variable "nsystem" was set to 4, namely at most mol4.pdb&vtx4.pdb and
density4.cub&ESP4.cub will be loaded and plotted, clearly if you want to use the plotting scripts to
simultaneously plot more monomers, the "nsystem" should be enlarged.
It is noteworthy that the value 0.15 in the ESPpt.txt is the grid spacing in quantitative molecular
surface analysis; if you increase it, the surface vertices will become more sparse, and calculation
cost will be reduced. The default commands in the ESPiso.txt correspond to using medium quality
grid for electron density while use low quality grid for ESP (for saving computational cost purpose),
this combination is suitable for small and medium sized systems, but for large systems, you may
need to modify the file so that high and medium quality grids are employed for electron density and
ESP, respectively, otherwise the resulting isosurface may be unsmooth and the mapped color may
be blurred.
For better graphical effect, users are suggested to manually change lower and upper limits of
color scale, so that the variation of ESP on vdW surface can be represented by color as clear as
possible. For charged systems, the default color scale should always be changed, otherwise the vdW
surface will be monochromatic. For these systems, you should load the input file, enter main
function 12, select option 1 to carry out quantitative molecular surface analysis for ESP, copy the
862
global minimal and maximal values of ESP to the text box for setting color scale, as shown below,
and then press ENTER button to make the setting take effect.
Note: If the map was plotted by ESPext.bat, you should take the ESP values in a.u. and then set them as the
color scale. However, if the map was plotted by ESPpt.bat, you should open the vtx1.pdb in VMD folder by text
editor, the first few lines clearly indicates the unit used in this file, you should read ESP values in that unit from the
Multiwfn console window and the set the color scale.
9 On the plotting ESP colored vdW map under Linux

The method described above can also be used in Linux (and may be also MacOS) environment.
In "examples\drawESP" folder, you can find ESPiso.sh, ESPpt.sh and ESPext.sh, they are
corresponding Linux scripts of the .bat file mentioned above.
For example, you want to use ESPiso.sh to draw ESP colored vdW surface for cosplay.fchk,
what you need to do is
• Install Multiwfn by strictly following Section 2.1.2. Install VMD in usual way
• Copy ESPiso.sh, ESPiso.txt and ESPiso.vmd from "examples\drawESP" to a working
directory
• Copy cosplay.fchk to the working directory
• Edit the ESPiso.sh, change the 1.fchk to cosplay.fchk
• Enter the working directory, run chmod +x ./ESPiso.sh, and then run ./ESPiso.sh. (Then you
may find error prompts such as "File not found" and "No such file or directory". They are harmless,
simply ignore them)
• Input vmd to boot up VMD, and then input source ESPiso.vmd in the VMD console window
to plot the map
4.A.14 Very easily rendering cube files as state-of-the-art isosurface

map via VMD script
Note: Chinese version of this tutorial is http://sobereva.com/483, it contains more discussions and examples
than this section.
Introduction
Although in most cases, the isosurface map directly drawn by Multiwfn is already satisfactory,
better effect can be gained if you use VMD to render the isosurface. VMD can be freely obtained
via http://www.ks.uiuc.edu/Research/vmd/. In fact, in Section 4.5.5 I already exemplified how to
863
draw isosurface map based on cube files produced by Multiwfn, however the procedure is somewhat
lengthy, and the effect cannot reach state-of-the-art level. In this section, I will show it is possible
to use VMD script to plot very high-quality isosurface map via very few steps. The methods in this
section only works for Windows platform, but you may also be able to find a way to make the
method work under Linux.
The VMD script is examples\scripts\showcub.vmd. Before using it, you should move it to VMD
folder and add a line source showcub.vmd into the vmd.rc file in VMD folder, so that the four
customized commands in this file will be available after booting up VMD. The commands are
described below.
➢ cub and cubiso: Used for showing single cube file. Example of usage:
cub DD : Draw DD.cub in current folder as isosurface map, both positive and negative parts
are shown as green and blue with default isovalue of 0.05 and -0.05, respectively.
cubiso 0.02 : Changing isovalue of both positive and negative parts to 0.02.
cub DD 0.02 : Equivalent to using cub DD and then cubiso 0.02.
➢ cub2 and cub2iso: Used for showing two cube file simultaneously. Example of usage:
cub2 f+ f- : Draw f+.cub and f-.cub in current folder as green and blue isosurfaces, respectively.
Note that only positive part of the cubes will be shown.
cub2iso 0.02 : Changing isovalue of both the two isosurfaces to 0.02.
cub2 f+ f- 0.02 : Equivalent to using cub2 f+ f- and then cub2iso 0.02.
After showing the isosurfaces in VMD graphical window using above commands, you can use
batch file VMDrender_full.bat or VMDrender_noshadow.bat in examples\scripts folder to invoke
Tachyon render to get better effect, as will be illustrated later. The difference between the two files
is that the shadow effect is enabled in the former while disabled in the latter.
Next I give two practical examples. Before following them, please copy the aforementioned
two .bat files and showcub.vmd to VMD folder, and then properly set up vmd.rc. The VMD version
I am using is 1.9.3.
Example 1: Spin density map of C4H8 singlet diradical

examples\C4H8.wfn // .wfn file of C4H8 singlet diradical
5 // Spin density
2 // Export the grid data to spindensity.cub in current folder
Now, move the spindensity.cub to VMD folder, boot up VMD and input orb spindensity 0.01
in VMD console window, you will see isosurface map of the cube file with isovalue of 0.01 has
been shown in the graphical window.
In order to get better effect, in VMD we choose "File" - "Render" - "Tachyon", then click "Start
Rendering", you will find vmdscene.dat has appeared in VMD folder. Now double click
VMDrender_full.bat, which will use vmdscene.dat as input file of Tachyon render to yield an image
file named full.bmp in current folder. The resulting graph is shown below, the quality is clearly very
good!
864
Example 2: Hole-electron map of NH2-biphenyl-NO2

The hole-electron analysis illustrated in Section 4.18.1 is extremely useful in understanding
the nature of electron excitation. Although Multiwfn can directly plot hole and electron distributions
simultaneously in the built-in GUI window, much better effect can be obtained by means of VMD.
I still use the NH2-biphenyl-NO2 analyzed in Section 4.18.1 as example, we will plot isosurface
of both hole and electron of S0→S2 transition via VMD. To do so, boot up Multiwfn and input
1 // Hole-electron analysis
examples\excit\D-pi-A.out
2 // Study excitation between ground state (S0) and the second excited state (S2)
1 // Calculate distribution of hole, electron and so on as well as various indices
After the calculation is finished, choose options 10 and 11 in turn to export grid data of hole
and electron to hole.cub and electron.cub in current folder, respectively. Then move them to VMD
folder, boot up VMD and input cub2 electron hole. You will find no isosurface has been shown, this
is because the default isovalue (0.05) is not suitable for grid data. We use cub2 command to test
different isovalues, finally we find after inputting cub2 0.005 the graphical effect is satisfactory, that
is the isosurfaces can substantially represent distribution character of both hole and electron. The
current graph shown in the VMD graphical window is shown below, green and blue correspond to
electron and hole, respectively.
Current effect is already nice, however, the overlapping character between hole and electron
can not be clearly recognized. In order to improve the effect, we enter "Graphics" - "Representation",
set "Material" to "EdgyGlass", then choose "electron.cub" in "Selected Molecules", also set its
"Material" to "EdgyGlass". Next, choose "File" - "Render" - "Tachyon", then click "Start
Rendering". If this time we directly use VMDrender_full.bat to render the graph, you will find the
graph is too dark. In order to get best effect, we open VMDrender_full.bat using text editor, change
865
the "-trans_raster3d" to "-trans_vmd", and then add an argument "-shadow_filter_off". Finally we

execute this .bat file to yield full.bmp, which is shown below, the effect is perfect! (Note that I used
Photoshop to increase the brightness of the graph by 20)
4.A.15 Calculating information-theoretic quantities and some relevant

quantities
Prof. Shubin Liu proposed many information-theoretic quantities and applied them to a wide
variety of chemical problems, and got many valuable findings. Multiwfn is able to calculate all
information-theoretic quantities. There is a document "Calculating information-theoretic quantities
and some relevant quantities by Multiwfn" in "Resources" page of Multiwfn website dedicated to
describe how to use Multiwfn to calculate these quantities, please check it.
866
5 Skills
5 Skills
5.1 Make Multiwfn support more quantum chemistry

programs
Although currently Multiwfn is able to directly accept Molden input file (.molden) as input file,
only the file generated by a few programs is formally supported (see the related description in
Section 2.5). If the Molden input file is generated by other programs then the analysis result may be
incorrect. For these cases, you can use Molden2aim program written by Wenli Zou
(https://github.com/zorkzou/Molden2AIM) to produce standardized Molden input file.
Using Molden2aim is easy. First move the Molden input file (e.g. ltwd.molden) into the
directory where molden2aim.exe is placed, then properly modify its setting file m2a.ini, then boot
up Molden2aim and follow its prompts to input commands, finally you will get ltwd_new.molden
(standardized Molden input file) and may be other files (.wfn, .wfx, etc.).
Since Molden2aim v2.0.5, orbital spin-types are explicitly written at the end of the converted
wfn file ($MOSPIN $END field), this information is automatically loaded by Multiwfn.
Molden2aim can output GTFs with angular moment up to g. Although g GTFs is not originally
defined in wfn format, these g GTFs can also be correctly identified and loaded into Multiwfn.
The "charge" field of the wfn file outputted by Molden2aim is the serial number of the element
in periodic table rather than the effective nuclear charges, even if effective core potential (ECP) is
used. This treatment is inconsistent with the wfn file outtputted by Gaussian, in which "charge" is
effective core charges when ECP is used (e.g. the "charge" of Au under Lanl2DZ is 19.0). So, if
ECP is used and you want to calculate electrostatic potential, do not forget to modify "charge" field
in the wfn file outputted by Molden2aim to effective nuclear charges.
5.2 Running Multiwfn in silent mode
Multiwfn aims for ease of use, so it is designed as an interactive program. Nevertheless

Multiwfn can also run in silent mode, by which you do not need to press any button during Multiwfn
running. Here I introduce how to do that.
For Windows users

For example, you want to obtain the graph in example 4.4.1 silently, you need to write an
inputstream file first, the content is (red texts are comments):
4  Main function 4
1  Real space function 1
1  Color-filled map
 Space line, corresponding to press ENTER button once (use default grid setting)
867
5 Skills
2  XZ plane
0  Y=0
0  Option 0: save graph to current directory
I suppose that the inputstream file is named as 4.4.1.txt, I have already provided this file in
“examples” directory. Now change “isilent” parameter in settings.ini from 0 to 1 (or add "-silent"
argument in command line), this makes Multiwfn forbid displaying any graph or GUI automatically
during running, otherwise you have to close the window by clicking mouse button. Then enter
command-line environment of Windows (click “Start”-“run” and type “cmd”) and run:
Multiwfn HCN.wfn < 4.4.1.txt > medinfo.txt
Here I assume that Multiwfn.exe, 4.4.1.txt and HCN.wfn are in current directory. After a few seconds,
you will find that image file has appeared in current directory. From medinfo.txt you can find all
intermediate information outputted by Multiwfn.
What does the content in the inputstream file mean? The answer is: the texts in each line of the
inputstream file are just what you need to input in interactive mode. It is very easy to write a new
inputstream file by following the prompts on screen in interactive mode. The symbols “<” and “>”
are redirection operators, they respectively tell Multiwfn that the content in 4.4.1.txt is inputstream,
while the outputstream should be stored to medinfo.txt. This redirection mechanism is not provided
by Multiwfn but provided by operation system. Notice that the name of input file is not given in
inputstream file, because it appears as an argument.
You may have noted that when the task is finished, some errors like below appear:
forrtl: severe (24): end-of-file during read, unit -4, file CONIN$
Image PC Routine Line Source
Multiwfn.exe 00588F1A Unknown Unknown Unknown
Multiwfn.exe 00586438 Unknown Unknown Unknown
Multiwfn.exe 00530B3A Unknown Unknown Unknown
......
Actually they are not errors, and hence you can safely ignore them. However, if you really want to
exit Multiwfn gracefully to avoid printing the "errors", you should properly write the inputstream
file so that command q is inputted in the main function menu at the final step.
Another example, assume that you want to save detail compositions of orbitals 1 to 3 of
COCl2.fch, just create an inputstream file orbana_1_3.in with below content:
8  Orbital composition analysis
1  Mulliken method
1  Orbital 1
2  Orbital 2
3  Orbital 3
Then run the command: Multiwfn COCl2.fch < orbana_1_3.in > orbana_1_3.txt.
Notice that if you are using Powershell in Windows environment, since the "<" redirection
operator is unsupported, you should use "Get-Content" command and pipline feature instead. For
example, above command should be written as (assume that Multiwfn.exe is in current folder): Get-
Content orbana_1_3.in | ./Multiwfn.exe COCl2.fch > orbana_1_3.txt
For Linux / MacOS users

If you are a Linux or Mac OS user, you can not only run Multiwfn silently as introduced above,
868
5 Skills
but also make use of "echo" command to avoid explicitly writting an inputstream file. The last
example can be equivalently realized by running this command:
echo -e "8\n1\n1\n2\n3" | Multiwfn COCl2.fch > orbana_1_3.txt
Each \n means pressing ENTER button once.

If you prefer to use shell script, you can also add below lines into your shell script file:
Multiwfn COCl2.fch > orbana_1_3.txt << EOF
EOF
5.3 Running Multiwfn in batch mode
If you are familiar with writing shell scripts and you have carefully read last section, you must
have already know how to use Multiwfn to process a batch of files, it is indeed very easy. I will
briefly introduce this point in this section.
For Windows users

• Example 1
Assume that you want to generate .cub files of ELF for these input files: ultravox.wfn,
chinaski.fch, strawberry_egg.wfn, you can create a plain text file named batchrun.bat (the suffix
must be .bat, while file name is arbitrary) with below content:
Multiwfn ultravox.wfn < genELFcub.txt > null
move ELF.cub ultravox.cub
Multiwfn chinaski.fch < genELFcub.txt > null
move ELF.cub chinaski.cub
Multiwfn strawberry_egg.wfn < genELFcub.txt > null
move ELF.cub strawberry_egg.cub
del null
where genELFcub.txt is inputstream file for generating ELF cube file, it is a plain text file having
below content is
5  Main function 5, calculate grid data
9  Real space function 9, namely ELF
2  Option 2: Medium quality grid
2  Option 2: Exporting cube file in current directory
Put the all of aforementioned files to a folder containing Multiwfn.exe, then double clicking the
icon "batchrun.bat" or input command batchrun in command line window, the task will start and
the three ELF cube files will be generated in current folder sequentially.
• Example 2
Shell script is very useful and powerful, it can automate a lot of repetitive work. As a simple
example, you want to generate ELF .cub file for all .wfn files in current folder, and hope that the
869
5 Skills
name of the resulting .cub files are [Input_file_name]_ELF.cub, then you can write a .bat file with
below content
for /f %%i in ('dir *.wfn /b') do (
Multiwfn %%i < genELFcub.txt > null
rename ELF.cub %%~ni_ELF.cub
Run the .bat file, the .cub files will be yielded in turn. Assume that one of input files is
yoshiko.wfn, the corresponding resulting .cub file will be yoshiko_ELF.cub.
For Linux users

Similarly, you can run Multiwfn under Linux environment in batch mode and write script to
significantly make your research easier. To realize the aforementioned example 1 under Linux, you
can create a file runthree.sh with below content (assume that you have properly installed Multiwfn
as mentioned in Section 2.1.2 and thus you can directly invoke Multiwfn by Multiwfn command)
Multiwfn ultravox.wfn < genELFcub.txt > null
mv ELF.cub ultravox.cub
Multiwfn chinaski.fch < genELFcub.txt > null
mv ELF.cub chinaski.cub
Multiwfn strawberry_egg.wfn < genELFcub.txt > null
mv ELF.cub strawberry_egg.cub
rm null
Put the runthree.sh and all input files in current folder, run this command: chmod
+x ./runthree.sh;./runthree.sh, then the calculation will start. (The chmod +x command is used to
add executable permission, it may be not necessary for certain cases)
To realize the aforementioned example 2 under Linux, you should create a shell script file with
below content and then run it
#!/bin/bash
for inf in *.wfn
do
echo Running ${inf} ...
Multiwfn ${inf} < genELFcub.txt > /dev/null
mv ELF.cub ${inf//.wfn}_ELF.cub
done
The shell environment of Linux platform is much more powerful than Windows. The
examples\scripts\gjf2xyz.sh is a Bash shell script that can convert all .gjf files in current folder
to .xyz file with same name, you can try to run it and check the script to understand how it works.
If you are unable to fully understand the content, just Google shell scripting.
Via sed command in Linux you can easily modify content of settings.ini in your script. For
example, to replace "iuserfunc=0" by "iuserfunc=30", you can input below command
sed -i 's/iuserfunc=../iuserfunc= 30/g' settings.ini
Substantially and flexibly employing shell scripts can automate much more kinds of analyses
than that shown above. For example, in Section 4.18.6 I illustrated that with a simple script, natural
transition orbitals (NTOs) of all selected excited states can be yielded and exported to various files
in one run.
870
5 Skills
5.4 Copy outputs from command-line window to clipboard
Sometimes the outputs of Multiwfn on command-line window are needed to be stored

permanently or transported to third-part softwares via plain text file. Here I describe how to copy
these outputs to Windows clipboard, assuming that you want to copy Hessian matrix of electron
density.
After you select "Mark", drag left mouse button from point A to point B
Then press ENTER button, the information highlighted by white rectangle will be stored to
clipboard, you can paste them to anywhere, such as plain text file.
For Mac OS or Linux system running in graphical environment, you can also copy the output
of Multiwfn from console to plain text file by similar manner.
871
5 Skills
5.5 Make command-line window capable to record more

outputs
Occasionally you may find command-line window cannot record entire outputs of Multiwfn.
For example, you select option 6 in wavefunction modification module to get density matrix for a
relative big wavefunction, however only the last part of the matrix can be found in the command-
line window. The solution of the problem is to enlarge buffer size of the window, please follow
below steps.
Boot up Multiwfn, click title of the window by right mouse button, click "Properties", select
"Layout" page, you will find the default buffer size of the window is 300 (see the screenshot below),
that means only up to 300 lines can be recorded in the window, which is obviously too small. Change
the value to a larger value, for example 9999, and then click OK button. After that you will find the
window capable to record much more outputs (If the complete output still cannot be recorded,
enlarge buffer size again).
The buffer size setting is saved permanently in system, you needn't to set this value again at
next time you boot up Multiwfn.
For Linux and Mac OS, you can also find similar option used to set buffer size of terminal.
5.6 Rapidly load a file into Multiwfn
Probably sometimes you feel inputting the path of input file is cumbersome, especially when
the path is very long. Below I provide you some tricks, which make this step much more easier.
If you want to rapidly load a file into Multiwfn without inputting its path, you can boot up
872
5 Skills
Multiwfn and then directly drag the icon of the file into the Multiwfn command-line window.
In Windows platform, an even more simple method is directly dragging the file onto the icon
of Multiwfn.exe, then the file will be automatically loaded into Multiwfn. Notice that in this situation,
the "current folder" is the position of the input file.
If directly inputting letter o, the file last time loaded will be loaded again, whose path is
recorded as "lastfile" variable in settings.ini file.
Assume that the file you last time loaded is C:\sob\lover\K-ON\Mio.wfn, and this time you
want to load C:\sob\lover\K-ON\Azusa.wfn, you can simply input ?azusa.wfn, namely the path of
the folder last time involved can be replace with a question mark.
5.7 Make use of cubegen utility in Gaussian package to

reduce computational time of electrostatic potential analyses
Calculation speed of ESP by internal code of Multiwfn is not as fast as the cubegen utility in
Gaussian package, especially for very large system when only very limited number of CPU cores is
available. In this circumstance, you can make use of cubegen to calculate the ESP data instead of
Multiwfn internal code to reduce the overall time-consuming.
The way of making use of cubegen in analyses about ESP is rather easy: Set "cubegenpath"
parameter in settings.ini file to actual path of cubegen executable file (For example,
"D:\study\G16W\cubegen.exe" in Windows platform or "/sob/g16/cubegen" in Linux platform).
Then if you use .fch/fchk/chk file as input file of Multiwfn, cubegen will be automatically invoked
by Multiwfn at the right time to calculate ESP data.
Availability
Below cases and functions currently support this feature:
·Plotting curve map of ESP (main function 3)
·Plotting plane map of ESP (main function 4)
·All functions that request grid data of ESP (e.g. using main function 5 to evaluate grid data
of ESP, using main function 17 to perform basin analysis for ESP, using subfunction 14 of main
function 200 to perform domain analysis for ESP)
·Calculating ESP fitting atomic charges such as CHELPG, MK and RESP (via corresponding
subfunctions in main function 7)
·Calculating TrEsp charges (see Section 4.A.9 on how to do this)
·Quantitative molecular surface analysis with ESP as mapped function (main function 12)
Although many other functions in Multiwfn also need ESP information, they do not support
utilizing cubegen because only very few number of points are needed to be calculated.
Even if you are not a Gaussian user, as long as the quantum chemistry program you used can
produce .mwfn or .molden file, or you are using GAMESS-US/Firefly, you can also benefits from
cubegen, because by using subfunction 2 of main function 100, Multiwfn can convert the
loaded .mwfn/.molden/.gms file to .fch file. Then, if you use the .fch file as input file, cubegen can
be invoked during electrostatic potential calculation. The examples/scripts/gbw2fch.sh is a Bash
script that convert all .gbw files to .fch file by invoking orca_2mkl and Multiwfn commands in turn,
873
5 Skills
if you are an ORCA user you will find it is quite useful.
Notes
·The wavefunction invoked by cubegen used for evaluating ESP comes from density matrix
in .fch/fchk file. This file may have more than one density matrices, by default the SCF density
matrix is used. The type of density matrix employed could be chosen by "cubegendenstype"
parameter in settings.ini. For example, if a .fch file was produced via "# MP2/cc-pVTZ density"
keyword, then the file will contain both SCF density matrix and MP2 density matrix. If you want to
calculate ESP at MP2 level by means of cubegen, you should change the value of "cubegendenstype"
from the default "SCF" to "MP2".
·When you are dealing with very large .fch/fchk file, running cubegen via Multiwfn may be
failed. Commonly, this is because the memory to be allocated by cubegen exceeded default upper
limit. To solve this problem, you can run for example export GAUSS_MEMDEF=4GB in you
console before performing your analysis, in this case the limit of memory will be increased to 4GB,
which is absolutely large enough.
·This feature is fully compatible with the cubegen in Gaussian 09 and 16 packages. For
cubegen in Gaussian 16, the number of CPU cores assigned for cubegen is identical to the "nthreads"
parameter in settings.ini file. However, since parallel implementation of cubegen in Gaussian 09
seems buggy, only one CPU core will be employed for it. Multiwfn judges whether it is G09 or G16
cubegen by checking if there is "g16" or "G16" string in the "cubegenpath".
·Even though you have modified current wavefunction via some functions of Multiwfn, e.g.
using subfunction 26 of main function 6 to manually change orbital occupation numbers, the ESP
calculated by cubegen still corresponds to the original wavefunction, since the density matrix in the
input file was not modified. If you indeed want to calculate ESP based on modified wavefunction
via cubegen, you should first use subfunction 2 of main function 100 to export .fch file, the SCF
density matrix in this file will correspond to the density matrix constructed by current wavefunction,
and thus then you can use this .fch file as input file to carry out ESP analyses.
Why Multiwfn crashes when invoking cubegen?

If Multiwfn crashes when (or after) invoking cubegen, there are three common reasons:
(1) The .fch/fchk file is problematic or inappropriate, for example:
• The file does not contain wavefunction information (e.g. the one produced by ONIOM task
with forcefield as low method)
• The file contains wavefunction but does not contain density matrix of expected type. For
example, you require cubegen to deal with MP2 density matrix, but current file only contains SCF
density matrix
• The file is corrupted. For example, incomplete file transfer, formchk does not work normally
(2) The wavefunction is quite large but you did not properly define the "GAUSS_MEMDEF".
(3) The cubegen is too old to compatible with the fch/fchk file, you need to upgrade the
Gaussian.
The outputted information of cubegen is dumped to nouseout file in current folder. If you are
confused why the invoked cubegen cannot run normally, you can try to manually examine this file.
874
5 Skills
5.8 Some tips on achieving ideal drawing quality
Many functions of Multiwfn involve plotting graph, below are some suggestions for getting
ideal drawing quality:
• Do not take screenshot, but use option like "save picture" or "export graphical file" to
generate image file. This is particularly important for Linux system, because the quality of the graph
shown on the screen is often poor. The effect of image file is independent of platform.
• If the graph to be plotted only consists of texts, straight lines and curves, it is strongly
suggested to change "graphformat" in settings.ini to a vector format, such as .pdf and .svg (see
Section 2.8 for detail), because in this case the texts and lines look very smooth, the graph can be
losslessly scaled, and meantime the file size is very small. Since vector format is not directly
supported by many applications programs, you can then manually convert the image file from vector
format to pixel format.
• When the graph contains gradient colors, such as isosurface map, color-filled map and so on,
the default .png format is the best choice. In order to make the axes, texts and outlines of objects as
smooth as possible, I suggest save the image file with large size (see Section 2.8 on how to change
size), then use advanced image editor such as Photoshop, or Irfanview (my favourite image viewer)
to shrink the graph to the size you actually need, in this process the resample will be done
automatically, anti-aliasing effect can thus be equivalently realized. By the way, before choosing the
option used to generate image file, I suggest properly zooming to make the system fill the entire
drawing area as much as possible.
• Please always pay attention to all options in the menu, Multiwfn have many options to control
plotting parameters, graph quality under default setting may be significantly improved by properly
using corresponding options. In addition, in settings.ini file there are many parameters controlling
plotting effect, I suggest you briefly check comment of each parameter in this file to find the ones
you often need to change.
• For plotting isosurface map, use relatively high quality of grid (i.e. relatively small grid
spacing, or equivalently, relatively high number of grid points) to generate grid data. If lighting
effect is not satisfactory, in the menu bar of the GUI for showing isosurface, use "Set lighting" in
"Other settings" to adjust lightings. In addition, using VMD to plot this kind of map can get much
better rendering effect, see Section 4.A.14 for related introduction and example.
• Use VMD plotting script whenever possible. For example, as illustrated in Section 4.2.5,
using VMD to plot paths and critical points can yield much better effect than directly plotting them
in the GUI window of topology module of Multiwfn; as also introduced in Section 4.A.14, via VMD
plotting script very ideal isosurface map of grid data can be obtained via merely a few steps
875
Appendix
Appendix
1 Setting up running environment for Gaussian in

Windows
Some functions of Multiwfn can directly invoke Gaussian (the precondition is that "gaupath"
in settings.ini has been set to actual path of Gaussian executable file). In order to make Windows
version of Gaussian run normally in this case, you must define "GAUSS_EXEDIR" environment
variable, otherwise error “No executable for file l1.exe” will occur and Gaussian running will fail,
because Gaussian does not know where to find the l1.exe executable file. Below are steps for setting
up this environment variable.
(1) For Windows XP users: Enter “Control panel”-“System properties”-“Advanced”
(2) For Windows 7 users: Enter “Control panel”-“System”-“Advanced system setting”-
“Advanced”
(3) For Windows 10 users: Click right mouse button on the start button, enter “Control panel”-
“System”-“Advanced system setting”-“Advanced”
After that, click “Environment variables” button, then click “New” button (in “User variables”
frame), input GAUSS_EXEDIR as variable name, input installation directory of Gaussian as variable
value (e.g. D:\study\g09w\, assuming that g09.exe is in this folder).
In addition, it is very important to note that when Multiwfn invoke Gaussian in Windows
environment, Gaussian will search for Default.Rou in current folder rather than in Gaussian scratch
path. Therefore, if there are important settings in Default.Rou, such as the default number of cores
to be utilized, you should copy this file to current folder to make the settings take effect during
calculations.
2 The routines for evaluating real space functions
Below are the routines in function.f90 file. You can make use of them to construct new real
space function by yourself. For more details please check the comments in the code of corresponding
routine.
Calculate real space functions

function calcfuncall: A warpper for calculating any supported real space function at a given
point
function userfunc: User-defined real space function
function linintp3d: Function value obtained by trilinear interpolation from grid data in memory
function splineintp3D: Function value obtained by cubic B-spline interpolation from grid data
in memory
function fmo: Orbital wavefunction value
876
Appendix
function fdens: Electron density

function fspindens: Spin or Alpha or Beta electron density
function fgrad: Gradient (x,y,z components or its norm) of density, or reduced density gradient
(RDG)
function flapl: Laplacian of electron density (xx or yy or zz part or total)
function Lagkin: Lagrangian kinetic G(r), or its components
function Hamkin: Hamiltonian kinetic K(r), or its components
function calcprodens: Promolecular density
function signlambda2rho: sign[2(r)](r)
subroutine signlambda2rho_RDG: Calculate sign[2(r)](r) and RDG at the same time
function signlambda2rho_prodens: sign[2(r)](r) with promolecular approximation
function RDGprodens: RDG with promolecular approximation
subroutine signlambda2rho_RDG_prodens: Calculate sign[2(r)](r) and RDG at the same
time with promolecular approximation
subroutine IGMprodens: Calculate promolecular density gradient of usual type or Independent
Gradient Model (IGM) type
function ELF_LOL: ELF or LOL or SCI (strong covalent interaction index)
function avglocion: Average local ionization energy
function loceleaff: Local electron affinity
function edr: Electron delocalization range EDR(r;d)
function edrdmax: Orbital overlap distance function D(r)
function delta_g_IGM: g(r) defined in IGM method
function linrespkernel: Approximate form of DFT linear response kernel for closed-shell
function pairfunc: Exchange-correlation density, correlation hole and correlation factor, on-top
pair density
function srcfunc: Source function
function infoentro: Shannon information entropy function or Shannon entropy density
function totesp: Total ESP
function nucesp: ESP from nuclear or atomic charges
function eleesp: ESP from electrons
function totespskip: ESP without contribution of nuclues defined by iskipnuc parameter
subroutine planeesp: Calculate ESP in a plane
subroutine espcub: Calculate grid data of ESP from electrons
function twoorbnorm: Product of norm of two orbitals
function beckewei: Generate Becke weighting function
function densellip: Ellipticity of electron density and  index
function xLSDA: Integrand of LSDA exchange functional
function xBecke88: Integrand of Becke88 exchange functional
function cLYP: Integrand of LYP corelation functional
function DFTxcfunc: Various kinds of integrand of DFT exchange-correlation functionals
function DFTxcpot: Various kinds of DFT exchange-correlation potentials
function weizsacker: Integrand of Weizsacker functional (steric energy)
function KED: Various kinds of kinetic energy density of electrons
function stericpot: Steric potential, which negative value is one-electron potential
877
Appendix
function stericcharge: Steric charge

function stericforce: Magnitude of steric force
funciton paulipot: Pauli potential
function pauliforce: Magnitude of Pauli force
function paulicharge: Pauli charge
function Fisherinfo: Fisher information density
function calcatmdens: Promolecular density calculated based on Lagrange interpolation of
built-in atomic radial density
function IRIfunc: Interaction region indicator (IRI)
function PAEM: Potential acting on one electron in a molecule
function SEDD: Single exponential decay detector (SEDD)
function DORI: Density Overlap Regions Indicator (DORI)
function localcorr: Local electron correlation function
function elemomdens: Electron linear momentum density
function magmomdens: magnetic dipole moment density
function energydens_grdn: Gradient norm of energy density
function energydens_lapl: Laplacian of energy density
function vdwpotfunc: Van der Waals potential and its two components
function orbwei_Fukui: Orbital-weighted Fukui function and dual descriptor
Calculate derivatives of real space functions

subroutine gencalchessmat: A general routine used to calculate value, gradient and Hessian
matrix at a given point for some real space functions
subroutine orbderv: Calculate wavefunction value of a range of orbitals and their derivatives
at given point, up to third-order
subroutine EDFrho: Calculate contribution from EDFs to density and corresponding
derivatives (up to third-order)
subroutine calchessmat_dens: Calculate electron density, its gradient and Hessian matrix
subroutine calchessmat_prodens: Calculate electron density, its gradient and Hessian matrix
with promolecular approximation based on built-in atomic densities
subroutine gendensgradab: Generate electron density and gradient norm for alpha and beta
electrons at the same time
subroutine calchessmat_lapl: Calculate Laplacian of electron density, its gradient and Hessian
matrix (Hessian is currently not available)
subroutine calchessmat_ELF_LOL: Calculate ELF/LOL, its gradient and Hessian matrix
(Hessian is currently not available)
subroutine calchessmat_mo: Calculate gradient and Hessian matrix for an orbital wavefunction
subroutine stericderv: Calculate the first-order derivative of steric potential
3 Detail of built-in atomic densities
Some analyses, such as Hirshfeld/ADCH population analyses and Hirshfeld orbital

composition analysis request atomic densities. Though, as shown in Section 3.7.3, the atomic
densities can be evaluated based on atomic .wfn files, the process is slightly complicated, namely
878
Appendix
the element .wfn files needed must be prepared and sphericalized first. In order to simplify these
analysis tasks, a set of built-in atomic densities (available from H to Lr) was provided in Multiwfn
and one can directly choose to use it.
These built-in atomic densities were evaluated at highly accurate computational level at atom
ground-states, and have been sphericalized (density distribution of many atoms in their ground states
are not spherically symmetric). Main group elements with index <=18 were calculated at B3LYP/cc-
pVQZ level, those >18 were calculated at B3LYP/ANO-RCC level (except for Ca, UGBS is used
since the ANO-RCC of Ca in EMSL website is wrong). Transition metals were calculated at
HF/UGBS level. Lanthanides and Actinides were calculated at B3LYP/SARC-DKH level (except
for U and Np, for which ROHF was used instead of B3LYP because DFT cannot reproduce their
correct ground state configurations). For all elements heavier than Ar, DKH2 method was employed
to take scalar relativistic effect into account. Unless otherwise specified, open-shell systems was
treated by unrestricted open-shell formalism.
The atomic densities are recorded as radial points in "atmraddens.f90", the second kind of
Gauss-Chebyshev method was used to generate the position of radial points, and the point
distribution is identical for all elements. The atomic density at any point is evaluated in terms of
Lagrange interpolation method based on the points.
If you want to replace built-in atomic density of an element with that calculated by yourself,
after booting up Multiwfn and loading corresponding atom wavefunction file, select subfunction 10
in main function 100 (a hidden option), then Multiwfn will calculate radial electron density and
output the result to sphavgval.txt in current folder. You can directly copy the Fortran codes in this
file into corrsponding field of atmraddens.f90.
4 Details about supplying inner-core electron density for

the wavefunctions involving pseudo-potential
In Section 2.5, the characteristic and the meaning of electron density function (EDF) have been
introduced. EDF information is used to represent the inner-core density replaced by pseudo-
potential, so that for the wavefunctions involving pseudo-potential, the result of the wavefunction
analyses that purely based on electron density can be almost exactly identical to full-electron
wavefunction.
When the input file you used contains GTF information, and meantime some atoms utilized
pseudo-potential, Multiwfn automatically finds proper GTF information from built-in EDF library
for these atoms. The built-in EDF library was developed by Wenli Zou and coworkers, it was
originally released as a part of Molden2aim program (https://github.com/zorkzou/Molden2AIM).
The quality of this EDF library is quite good, and is better than the EDF field contained in the .wfx
file produced by Gaussian program. This EDF library covers the entire periodic table, up to index
of 120. For most elements, it contains both EDF information for large and small core pseudo-
potentials. Some descriptions about this library posted by its developer can be found in
http://bbs.keinsci.com/thread-5354-1-1.html as well as J. Comput. Chem., 39, 1697 (2018). If you
do not want Multiwfn to read EDF information from this library automatically, set "isupplyEDF" in
settings.ini to 0.
As mentioned in Section 2.5, the .wfx file produced by Gaussian directly carries EDF field
879
Appendix
when pseudo-potential is used. When this kind of file is used as input, Multiwfn by default reads
EDF information from EDF field of this .wfx file rather than from the built-in EDF library. If you
do not want Multiwfn read EDF information from this file but from the built-in EDF library, you
can change "readEDF" in settings.ini from 1 to 0.
Worhnotingly, Multiwfn also allows reading EDF information from atomic .wfx file produced
by Gaussian rather than from the built-in EDF library of Multiwfn (This feature is rarely useful,
since as mentioned above, the quality of the built-in EDF library in Multiwfn is better than the EDF
library embedded in Gaussian). An example is provided as below:
examples\Pt(NH3)2Cl2.wfn is a file corresponding to Pt(NH3)2Cl2, Lanl2 pseudo-potential
accompanied by Lanl2DZ basis set is used for Pt and Cl, while 6-31G* is used for other atoms.
examples\Pt_lanl2.wfx and examples\Cl_lanl2.wfx are the Pt and Cl atomic .wfx files produced by
Gaussian 09, in which Lanl2 is also used, therefore their EDF fields represent the inner-core electron
density of Pt and Cl replaced by Lanl2.
In order to borrow EDF information from the atomic .wfx files in the analysis of
Pt(NH3)2Cl2.wfn, first we need to set "isupplyEDF" parameter in settings.ini to 1. Then boot up
Multiwfn and input following commands
examples\Pt(NH3)2Cl2.wfn
Pt // Load EDF information for element Pt
examples\Pt_lanl2.wfx // Take EDF information of Pt from this file
Cl // Load EDF information for element Cl
examples\Cl_lanl2.wfx // Take EDF information of Cl from this file
q // We have finished, exit
Now we can perform wavefunction analysis as usual. But it is better to first carry out some
tests to check if inner-core electron density has been properly represented, for example, we integrate
electron density over the whole space
100 // 100 Other functions (Part1)
4 // Integrate a function over the whole space
The result is 132.00, which is the expected total number of electron of Pt(NH3)2Cl2. Assume
that we did not load the EDF information, then the result will be 52.00, which is just the number of
valence electrons of Pt(NH3)2Cl2.
Note that you can also directly input atomic indices instead of element name, for example,
inputting 4,8-10,11 means selecting atoms 4,8,9,10,11 in present system. Of course, the atoms you
selected each time must corresponds to the same element and the same pseudo-potential.
It is users' responsibility to prepare the atomic .wfx files. Since there are too many kinds of
pseudo-potentials and elements, evidently I am uncapable of providing all of them for you.
5 Check sanity of wavefunction
The input files of Multiwfn generated by various quantum chemistry codes are not always
standard. For example, .molden file generated by many programs are problematic in content or
format. After you feeding them to Multiwfn and you want to make sure if the wavefunction has been
correctly loaded, there are two useful ways to check sanity:
880
Appendix
(1) Enter main function 100 and select subfunction 4, then choose electron density. If the
integral of electron density over the whole space is very close to the actual number of electrons,
then the wavefunction is ready for use.
(2) Enter main function 1000 (a hidden function) and select subfunction 100. This function will
check satisfaction of normalization condition of all orbitals, and show you the maximum deviation
to unity and integer. If both the two maximum deviations are evidently larger than zero, then there
must be some severe problems in the input file; if any one of them is very close to zero, the input
file should be fully compatible with Multiwfn.
6 Special functions
There are some special functions in Multiwfn, they are mainly used for debugging, special
purposes, and some of them are requested by some Multiwfn users. Here a few of them are
mentioned.
1 Adding Bq atoms at specific positons

Sometimes we want to highlight special positions on 3D map, for example, reference point,
centroid position of a real space function, the position used to plot local DOS and so on. To realize
this, you can enter main function 1000, select subfunction 12, and manually input the X, Y, Z
coordinate of the Bq atoms (ghost atoms) to be added. You can add arbitrary number of Bq atoms.
Once all Bq atoms have been added, input q to return. Then in the GUI showing 3D molecular
structure you will see the Bq atoms, which are displayed as cyan spheres. It is worth to note that in
the main function 0, you can choose showing or not showing labels of Bq atoms by selecting
corresponding term in "Other settings" - "Set atomic label type".
2 Calculate nuclear attractive energy between a fragment and an orbital

You can use this function by entering subfunction 90 of main function 1000 (hidden in main
menu). This function calculates attractive energy between all nuclei in a user-defined fragment and
an orbital, namely:
ZA ZA
EMO i −frag =  i  |r − R
Afrag |
i   |  i (r ) |2  |r − R
Afrag |
dr
A A
During the calculation, the atomic contributions to the result are outputted in turn, for example,
below output
Processing center 2(H ) / 3
Accumulated value: -8.3269642826 Current center: -0.0629848533
means
ZA
2H
EMO i −frag =  w2 H (r ) |  i (r ) |
2
 |r − R
Afrag |
dr = −0.06298
A
where w2H(r) is atomic weighting function of 2H atom defined by Becke's partition.

The integral is evaluated by Becke's multi-center integration method, therefore the "radpot"
and "sphpot" in settings.ini affect integral accuracy. Commonly the default value is accurate enough.
Note that, as shown in above formula, the occupation number of orbital is not taken into account in
881
Appendix
the calculation.
3 Output Becke's integration points

The Becke's multi-center numerical integration algorithm proposed in J. Chem. Phys., 88, 2547
(1988) has been employed by almost all popular quantum chemistry programs to integrate
exchange-correlation functionals, it is also employed by Multiwfn in many wavefunction analysis
functions, for example, calculating Hirshfeld charge and ADCH charge, fuzzy atomic space analysis
(main function 15) and so on.
Via subfunction 93 of main function 1000 (a hidden function), you can export integration points
of the Becke's algorithm to intpt.txt in current folder. The meaning of each column will be explicitly
shown on screen after exporting. The number of integration points is determined by "radpot" and
"sphpot" in settings.ini. The position and weight of angular grid are determined by Lebedev method,
while those of radial grid are determined by second kind Gauss-Chebyshev method.
After entering this function, Multiwfn asks you if also outputting Becke's integration weights.
If you input y, the weights will be calculated and outputted together.
882

Multiwfn 3.7

Uploaded by

Copyright:

Available Formats

Multiwfn 3.7

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Multiwfn 3.7

Uploaded by

Copyright:

Available Formats

What are the recommendations for users of the software?

What are the recommendations for users of the software?

What are the different input file requirements for different functions of the software?

What are the different input file requirements for different functions of the software?

Multiwfn

− A Multifunctional Wavefunction Analyzer −

Beijing Kein Research Center for Natural Sciences (www.keinsci.com)

This content in the manual is organized in the following sequence:

!!!!!!!!!! ALL USERS MUST READ !!!!!!!!!! ..........................................i

Linux and Mac OS USERS MUST READ ............................................ ii

2 General information ............................................................................9

4 Tutorials and Examples ..................................................................343

Input files supported by Multiwfn

Special points of Multiwfn

Major functions of Multiwfn

MO-PDOS. Up to 10 fragments can be very flexibly and conveniently defined. Local

transition density, transition electric/magnetic dipole moment and charge density

Real space functions supported by Multiwfn

density, integrand of Rényi entropy, steric energy/potential/charge, Pauli potential/force/charge,

Things that Multiwfn can do

atom/fragment/orbital contribution, centroid position, displacement and overlap, exciton binding

Tian Lu, Feiwu Chen, Multiwfn: A Multifunctional Wavefunction Analyzer, J. Comput.

2.1.1 Windows version

2.1.2 Linux version

• Uncompress the Multiwfn binary package

• Run below command to add executable permission to Multiwfn executable file:

2.1.3 Mac OS version

2.2 Using Multiwfn

2.3 Files of Multiwfn

2.4 Parallel implementation

2.5 Input files and wavefunction types

Wavefunction types supported by Multiwfn include restricted/unrestricted single-determinant

Contained information types

Molden input file (.molden or .molden.input or molden.inp) : Currently, a wide variety of

GTF and basis function informations.

H 0.000000 0.758953 -0.477232 0.150977

H 0.000000 -0.758953 -0.477232 0.150977

REMARK ESP charges

ATOM 1 O O 1 1 0.000 0.123 0.000 -0.680698 2.9000

ATOM 2 H O 1 1 0.757 -0.490 0.000 0.340338 2.6000

ATOM 3 H O 1 1 -0.757 -0.490 0.000 0.340361 2.6000

.vti, .dx and DMol3 grid file (.grd):

2.6 Real space functions

where i is occupation number of orbital i,  is orbital wavefunction,  is basis function. C is

2 Gradient norm of electron density

3 Laplacian of electron density

4 Value of orbital wavefunction

5 Electron spin density

 s (r) =   (r) −   (r)

6 Hamiltonian kinetic energy density K(r)

alternative definitions. One of commonly used definition is

7 Lagrangian kinetic energy density G(r)

K(r) and G(r) are directly related by Laplacian of electron density

2  (r) / 4 = G(r) − K (r)

8 Electrostatic potential from nuclear / atomic charges

9 Electron localization function (ELF)

For closed-shell system, since  =   = (1 / 2) , D and D0 terms can be simplified as

density  W (r ) = (1 / 8)  /  , therefore the D(r ) =  S (r ) −  W (r ) reveals the excess

10 Localized orbital locator (LOL)