Optimisation of Basis Sets and
Pseudopotentials
Sanliang Ling
University College London
4th CP2K Tutorial, 31st August – 4th September 2015, Zurich
Electronic structure methods in CP2K
GPW: Gaussian and plane waves method
-Goedecker-Teter-Hutter pseudopotentials
-Gaussian basis sets for valence electrons
GAPW: Gaussian and augmented plane
waves method
-all electron calculations
2
LCAO
LCAO: Linear Combination of Atomic Orbitals
MO coefficient atomic orbital
molecular orbital (MO)
(unknown) (basis function)
Jensen, Introduction to Computational Chemistry, Wiley (2007)
3
Gaussian type orbitals (GTOs)
normalisation exponent:
constant width of orbital
sum of lx, ly, lz determines type of orbital:
0 for s, 1 for p, 2 for d, 3 for f, etc
Jensen, Introduction to Computational Chemistry, Wiley (2007)
4
Contracted basis sets
contraction coefficient
(to be optimised)
Jensen, Introduction to Computational Chemistry, Wiley (2007)
5
Polarisation function
basis functions with higher angular momentum (than
the valence orbital)
first shell of polarisation functions are most important
p-function for H-Be, d-function for B-Ca, etc
adds additional flexibility to the basis set, provides
better descriptions to bonding
Jensen, Introduction to Computational Chemistry, Wiley (2007)
6
Diffuse function
basis function with small exponent
better representation of the “tail” of the wavefunction
important for loosely bound electrons (anions or
excited state) and molecules in the gas phase
Jensen, Introduction to Computational Chemistry, Wiley (2007)
7
All-electron basis set for GAPW calculations
Pople style basis sets (e.g. 6-31G*, etc)
Correlation consistent basis sets (aug-cc-pVDZ, etc)
and more
see $CP2K/cp2k/data, ‘ALL_BASIS_SETS’ and ‘EMSL_BASIS_SETS’
additional all-electron basis sets can be found from EMSL Basis Set
Exchange, see https://bse.pnl.gov/bse/portal
Potential needs to be defined in &KIND section for GAPW calculations, see
$CP2K/cp2k/data/POTENTIAL, choose “ALL” potential
8
Basis set for GPW calculations
MOLOPT basis sets: basis sets optimised from molecular
calculations, see ‘BASIS_MOLOPT’
‘DZVP-MOLOPT-SR-GTH’ for solids (‘SR’ denotes shorter
range, i.e. less and thus less diffuse primitives)
always check the basis set convergence (DZVP/TZVP/…)
do not use SZV for production run
more basis sets for GTH pseudos can be found in
‘BASIS_ZIJLSTRA’ and ‘GTH_BASIS_SETS’
all basis set files can be found in $CP2K/cp2k/data
9
Basis set construction
trade-off between computational cost and accuracy
route for systematic improvements
(SZV/DZVP/TZVP/TZV2P/…)
same basis set should perform in various chemical
environments, e.g. from isolated molecules to solids
lead to well conditioned overlap matrices (suitable for
linear scaling calculations)
condition number: ratio of the largest to smallest
eigenvalue of the overlap matrix
VandeVondele & Hutter, J. Chem. Phys., 127, 114105 (2007) 10
MOLOPT basis set
CP2K All-electron (Gaussian/NWCHEM)
SZV STO-3G
H-Rn
DZVP 6-31G*
limited TZVP 6-311G*
availability
TZV2P 6-311G(2df, 2pd)
SZV: single-zeta valence, i.e. one contracted function per orbital
DZVP: double-zeta valence, i.e. two contracted functions per orbital plus
one set of polarisation functions with l = lmax + 1
TZVP/TZV2P: triple-zeta valence, i.e. three contracted functions per
orbital plus one/two set of polarisation functions with l = lmax + 1
Matthias Krack, 1st CP2K Tutorial, Zurich, 2009 11
MOLOPT basis set format
element basis set name number of valence electrons in pseudo
H DZVP-MOLOPT-GTH DZVP-MOLOPT-GTH-q1
1 number of CGTO contraction coefficients
201721
11.478000339908 0.0249162432 -0.0125124214 0.0245109182
3.700758562763 0.0798254900 -0.0564490711 0.0581407941
1.446884268432 0.1288626753 0.0112426847 0.4447094985
0.716814589696 0.3794488946 -0.4185875483 0.6462079731
0.247918564176 0.3245524326 0.5903632167 0.8033850182
0.066918004004 0.0371481214 0.4387031330 0.8929712087
0.021708243634 -0.0011251955 -0.0596931713 0.1201013165
Gaussian exponents s-function p-function
principle quantum
number
2 0 1 7 2 1 number of
p-function
minimum angular maximum angular number of number of
momentum momentum Gaussian s-function
quantum number quantum number exponents 12
Basis set optimisation
number of Gaussian exponents (to be determined
before optimisation)
number of basis functions per angular momentum
choice of training molecules (transferability)
strategy of basis set optimisation, e.g. whether or not
to optimise different basis sets concurrently
weight of condition number in optimisation
13
Choice of training molecules
small molecules formed with different elements and
with different coordination environments
preferably with only two elements (including the
target element) in each molecule
a good source of small molecules (with optimised
geometries) can be found in the Supporting
Information of “Ahlrichs et al., Phys. Chem. Chem.
Phys., 7, 3297 (2005)”
14
Basis set optimisation: MOLOPT
contraction
exponents coefficients
objective basis sets training total condition
weight
function to be optimised molecules energy number
VandeVondele & Hutter, J. Chem. Phys., 127, 114105 (2007) 15
Basis optimisation with OPTIMIZE_BASIS
Choosing a reference (complete) basis
Performing accurate molecular calculations with ref. basis
Choosing a form of the basis to be fitted
Minimizing the objective function
Ω 𝛼𝑖 , 𝑐𝑗 = 𝐵 𝑀 ∆𝜌𝐵,𝑀 𝛼𝑖 , 𝑐𝑗 + 𝛾 ln κ𝐵,𝑀 𝛼𝑖 , 𝑐𝑗
(developed by Dr Florian Schiffmann) 16
Basis optimisation with OPTIMIZE_BASIS
• Reference (Complete) basis set
– check GTH-def2-QZVP and aug-GTH-def2-QZVP
included in $CP2K/cp2k/data/BASIS_ADMM
– generate uncontracted basis sets with the ATOM code
(see Marcella’s slides and examples in
$CP2K/cp2k/tests/ATOM)
• Molecular calculations
– use reference basis sets for all elements
– avoid homonuclear diatomic molecules
– use equilibrium geometry (i.e. GEO_OPT)
17
Generate uncontracted basis set with ATOM
&GLOBAL
PROJECT Na
PROGRAM_NAME ATOM
&END GLOBAL
&ATOM
ELEMENT Na
RUN_TYPE BASIS_OPTIMIZATION
ELECTRON_CONFIGURATION CORE 2s2 2p6 3s1
CORE 1s2
MAX_ANGULAR_MOMENTUM 1
&METHOD
METHOD_TYPE KOHN-SHAM
&XC
&XC_FUNCTIONAL PBE
&END XC_FUNCTIONAL
&END XC
&END METHOD
&OPTIMIZATION
EPS_SCF 1.e-8
&END OPTIMIZATION
&PP_BASIS
NUM_GTO 6 6
S_EXPONENTS 7.92602574 5.92602574 1.59655262 0.71279902 0.28969807 4.00675308
P_EXPONENTS 7.92602574 5.92602574 1.59655262 0.71279902 0.28969807 4.00675308
&END PP_BASIS
&POTENTIAL
PSEUDO_TYPE GTH
POTENTIAL_FILE_NAME POTENTIAL
POTENTIAL_NAME GTH-PBE-q9
&END POTENTIAL
&POWELL
ACCURACY 1.e-8
STEP_SIZE 1.0
&END POWELL
&END ATOM 18
Generate uncontracted basis set with ATOM
19
Generate uncontracted basis set with ATOM
Na CBS
8
2021111
23.51400109 1.0 1.0 1.0
2021111
11.54276369 1.0 1.0 1.0
2021111
4.98513381 1.0 1.0 1.0
2021111
2.06401264 1.0 1.0 1.0
2021111
0.83224580 1.0 1.0 1.0
2021111
0.31900235 1.0 1.0 1.0
2021111
0.06577574 1.0 1.0 1.0
2021111
0.02386738 1.0 1.0 1.0
20
GTH-def2-QZVP basis set
H GTH-def2-QZVP Go to https://bse.pnl.gov/bse/portal,
12
20071 select ‘H’ element and ‘Def2-QZVP’ basis
11.478000339908 0.024916243200 set, use ‘Gaussian 94’ format:
3.700758562763 0.079825490000
1.446884268432 0.128862675300 H 0
0.716814589696 0.379448894600
0.247918564176 0.324552432600
SZV S 4 1.00
190.6916900 0.70815167E-03
0.066918004004 0.037148121400 28.6055320 0.54678827E-02
0.021708243634 -0.001125195500 6.5095943 0.27966605E-01
1 0 0 1 1 1.8412455 0.10764538
6.50959430 1.00000000 S 1 1.00
1 0 0 1 1 0.59853725 1.0000000
1.84124550 1.00000000 S 1 1.00
1 0 0 1 1 0.21397624 1.0000000
0.59853725 1.00000000 S 1 1.00
1 0 0 1 1 0.80316286E-01 1.0000000
0.21397624 1.00000000 P 1 1.00
1 0 0 1 1 2.29200000 1.0000000
0.08031629 1.00000000 P 1 1.00
1 1 1 1 1 0.83800000 1.0000000
2.29200000 1.00000000 P 1 1.00
1 1 1 1 1 0.29200000 1.0000000
0.83800000 1.00000000 D 1 1.00
1 1 1 1 1 2.06200000 1.0000000
0.29200000 1.00000000 D 1 1.00
1 2 2 1 1 0.66200000 1.0000000
2.06200000 1.00000000 F 1 1.00
1 2 2 1 1 1.39700000 1.0000000
0.66200000 1.00000000
1 3 3 1 1
1.39700000 1.00000000 (use exponents between 0.05~20 only)
21
Input Structure: OPTIMIZE_BASIS
&GLOBAL
PROJECT optbas
PROGRAM_NAME OPTIMIZE_BASIS Ti FIT10
PRINT_LEVEL HIGH 10
&END GLOBAL 10011
&OPTIMIZE_BASIS 0.10001966 1.00000000
BASIS_TEMPLATE_FILE BASIS_SET_TEMPLATE 10011
BASIS_WORK_FILE WORK_BASIS_STRUCTURE 1.06186104 1.00000000 s-functions
BASIS_OUTPUT_FILE Ti_FIT10 10011
# USE_CONDITION_NUMBER Y 0.40963197 1.00000000
# CONDITION_WEIGHT 0.0005 10011
WRITE_FREQUENCY 10 4.39901876 1.00000000
&OPTIMIZATION 11111
MAX_FUN 50000 0.52985233 1.00000000
&END OPTIMIZATION 11111 p-functions
… 1.57394040 1.00000000
&TRAINING_FILES 11111
DIRECTORY ../ticl4 11.83843422 1.00000000
INPUT_FILE_NAME ticl4.inp 12211
&END TRAINING_FILES 0.25675246 1.00000000
… 12211 d-functions
&FIT_KIND Ti 1.02358115 1.00000000
BASIS_SET FIT10 12211
INITIAL_DEGREES_OF_FREEDOM EXPONENTS 4.21355677 1.00000000
&CONSTRAIN_EXPONENTS
BOUNDARIES 0.1 20 (Ti electron configuration: [Ne] 3s2 3p6 4s2 3d2)
USE_EXP -1 -1
&END CONSTRAIN_EXPONENTS
&END FIT_KIND
&END OPTIMIZE_BASIS (see $CP2K/cp2k/tests/QS/regtest-optbas) 22
Basis optimisation with OPTIMIZE_BASIS
23
Basis set superposition error
&GLOBAL
MOLOPT basis sets are PROJECT_NAME project
incomplete RUN_TYPE BSSE
&END GLOBAL
…
BSSE correction using the &FORCE_EVAL
…
Boys and Bernardi &BSSE
&FRAGMENT
counterpoise correction LIST 1..272
scheme &END FRAGMENT
&FRAGMENT
Boys & Bernardi, Mol. Phys., 19, 553 (1970)
LIST 273..368
&END FRAGMENT
&END BSSE
useful for binding energy …
SCF_GUESS ATOMIC
calculations, etc …
&KIND H_ghost
use larger basis sets to BASIS_SET DZVP-MOLOPT-SR-GTH
GHOST
reduce BSSE &END KIND
…
(see more examples in $CP2K/cp2k/tests/QS/regtest-gpw-3) 24
Pseudopotentials
Goedecker-Teter-Hutter (GTH) pseudopotentials
ionic charge error function
long-ranged term
Local part
short-ranged term
coefficients
rloc: range of Gaussian ionic charge distribution
Krack, Theor. Chem. Acc., 114, 145 (2005)
25
Pseudopotentials
Non-local part
coefficients
Gaussian-type projectors
normalisation spherical
constant radius
harmonics
Krack, Theor. Chem. Acc., 114, 145 (2005)
26
GTH pseudopotential format
Element Name Number of valence electrons
Ti GTH-PBE-q12 GTH-PBE
4 6 2 Number of valence electrons in each shell (s/p/d)
𝒓𝒍𝒐𝒄 0.38000000 2 8.71144218 -0.70028677
Number of non-
3 local projectors
𝒓𝒔 0.33777078 2 2.57526386 3.69297065
-4.76760461
𝒓𝒑 0.24253135 2 -4.63054123 8.87087502
-10.49616087
𝒓𝒅 0.24331694 1 -9.40665268
number of potential functions coefficients
27
GTH pseudopotential
LDA (PADE): H-Rn (including lanthanides)
PBE: H-Rn (excluding lanthanides)
PBEsol: H-Kr (plus a few selected)
BP: H-Kr (plus a few selected)
HCTH: a few selected elements
Non-linear core corrected (NLCC) pseudopotentials:
a few selected elements
All pseudopotentials can be found in $CP2K/cp2k/data, see ‘POTENTIAL’,
‘GTH_POTENTIALS’ and ‘NLCC_POTENTIALS’
Matthias Krack, 1st CP2K Tutorial, Zurich, 2009 28
Pseudopotentials optimisation
All-electron calculation using a chosen DFT functional
Choosing a form of the pseudopotential to be fitted
Minimising the differences between eigenvalues
and charges within an atomic sphere of the all-
electron atom and the pseudo atom
Quality check
Hutter et al., Phys. Rev. B, 58, 3641 (1998) 29
Pseudopotentials optimisation with ATOM
&OPTIMIZATION
Example: optimise pseudo- EPS_SCF 1.e-10
potential of O using PBE0 &END OPTIMIZATION
&PRINT
&GLOBAL &BASIS_SET
PROGRAM_NAME ATOM &END
&END GLOBAL &END
&ATOM &AE_BASIS
ELEMENT O BASIS_TYPE GEOMETRICAL_GTO
RUN_TYPE PSEUDOPOTENTIAL_OPTIMIZATION &END AE_BASIS
ELECTRON_CONFIGURATION [He] 2s2 2p4 &PP_BASIS
CORE [He] BASIS_TYPE GEOMETRICAL_GTO
MAX_ANGULAR_MOMENTUM 2 &END PP_BASIS
COULOMB_INTEGRALS ANALYTIC &POTENTIAL
EXCHANGE_INTEGRALS ANALYTIC PSEUDO_TYPE GTH
&METHOD POTENTIAL_FILE_NAME POTENTIAL
METHOD_TYPE KOHN-SHAM POTENTIAL_NAME GTH-PBE-q6
RELATIVISTIC DKH(2) &END POTENTIAL
&XC &POWELL
&XC_FUNCTIONAL PBE0 ACCURACY 1.e-10
&END XC_FUNCTIONAL STEP_SIZE 0.5
&END XC WEIGHT_PSIR0 0.1
&END METHOD &END POWELL
&END ATOM
(see more examples in $CP2K/cp2k/tests/ATOM/regtest-pseudo) 30
Pseudopotentials optimisation with ATOM
31
Pseudopotentials optimisation with the
standalone ATOM code
check ‘README_quick_GTH’ in $CP2K/potentials,
more details in Dr Matthias Krack’s slides
Matthias Krack, 1st CP2K Tutorial, Zurich, 2009
more options for quality check
32
A few final remarks …
read the original references on basis sets and
pseudopotentials before making your own basis sets and
pseudopotentials
do plenty of tests on optimised basis sets and
pseudopotentials and compare the results with reference
values before production run
All data files used by CP2K can be downloaded directly from
http://sourceforge.net/p/cp2k/code/HEAD/tree/trunk/cp2k/data/
33
Further Reading
Accuracy and Efficiency (by Dr Matthias Krack)
http://www.cecam.org/upload/talk/presentation_3002.pdf
Basis Sets and Pseudo-Potentials (by Dr Fawzi Mohamed)
http://www.cecam.org/upload/talk/presentation_2994.pdf
34
