Theoretical

(Soft) Condensed Matter Physics

M.Sc. Physics and M.Sc. Applied Physics

PHYSIKALISCHES KOLLOQUIUM
PHYSIKALISCHES
Theoretical KOLLOQUIUM
ANTRITTSVORLESUNG
Condensed Matter Physics
AM 12. NOVEMBER 2018 UM 17 UHR C.T.
A
IM
NTRITTSVORLESUNG
Lecturer:
GROßENJoachim
HÖRSAALDzubiella
12. NOVEMBER
AM Tutors: 2018
Sebastien Groh and
UM 17 U
Sven HR C.T.
Pattloch
IM GROßEN HÖRSAAL

Advanced Physics Course, Institute of Physics, Freiburg, SoSe 2024

COMPUTATIONAL PHYSICS OF SOFT AND FLUID MATERIALS

PROF. DR. JOACHIM DZUBIELLA

Prof. Joachim
PHYSIKALISCHES INSTITUT Dzubiella
, UNIVERSITÄT FREIBURG
Applied Theoretical Physics - Computational Physics
Macromolecules and complex fluids serve as integral components in the development of modern
soft and 'smart' functional materials, e.g., for applications in medicine, liquid phase catalysis, or
Cenergy
OMPstorage
UTAdevices.
TIONFrom ALa P HYSphysics
statistical OF SOthese
ICS perspective ANDareFtypically
FT liquids LUIDmod- MATERIALS
Institute of Physics - University of Freiburg
elled as classical, interacting many-body systems of varying complexity. While the structural and
thermodynamic properties of a ‘simple’ liquid can still be studied by analytical (statistical me-

PROF.Spring
DR. J2024
OACHIM DZUBIELLA
chanics) theory to a wide extent, applied fluid materials display much higher complexity and one
- Version 0.2
typically resorts to particle-resolved (molecular) computer simulations for an accurate treatment.
Here, the challenge is to set-up and analyse these 'computer experiments' in a meaningful way

PHYSIKALISCHES INSTITUT, UNIVERSITÄT FREIBURG

and, if possible, to reduce and interpret the large amount of generated data by simple and trans-
parent models. In this way, computer simulations establish an important bridge between theoret-
ical physics and applied material science, providing useful structure-property-function relation-
ships for the rational design of new materials. In this talk I will introduce and discuss these ideas
Macromolecules and complex
by means of a few illustrative fluids
examples serve
of soft and as
fluidintegral
systemscomponents in the development
of varying complexity, i.e., start- of modern
ing from
soft and simple
'smart'liquids like hard-sphere
functional materials,colloids to fully
e.g., for atomistic models
applications of multi-component
in medicine, liquid phase catalysis, or
electrolytes or polymers relevant for modern applications .
energy storage devices. From a statistical physics perspective these liquids are typically mod-
elled as classical, interacting many-body systems of varying complexity. While the structural and
thermodynamic properties of a ‘simple’ liquid can still be studied by analytical (statistical me-
chanics) theory to a wide extent, applied fluid materials display much higher complexity and one
 1

Version 0.2: March 2024 - April 2024.

Some chapters merged from the Computational Physics script and renamed. A general intro-
duction and the Intro-talk as appendix were added. Some parts were restructured, followed by
proofreading and little text extensions by SG, Sven Pattloch, and JD.
Version 0.1: December 2022 - May 2023.
Note: scripted by M.Sc. Applied Physics student Olga Korobeinicheva. Title page image credit:
Wikimedia Commons.

This script is based on lectures given by Prof. Joachim Dzubiella at the Institute of Physics
at the University of Freiburg in Master courses for (soft) condensed matter theory starting
from summer semester 2022. It is dynamic and constantly evolving, hopefully ever improving
through the process.
Contents

1 Introduction 8

2 Bonds and Interactions 11

2.1 Chemical and physical bonding in solids and liquids . . . . . . . . . . . . . . . . 11
2.1.1 Covalent bonds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.1.2 Ionic bonds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.3 Metallic bonds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.1.4 Physical bonds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Interactions and potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.1 Interatomic pair potentials . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3 Structure of solids 23
3.1 Crystal lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.1.1 Common crystal structures . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1.2 Facets and Miller indices . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 Reciprocal lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.1 Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.2 Bragg interpretation of the Laue condition . . . . . . . . . . . . . . . . . 31
3.2.3 Brillouin zones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.3 Crystal structure factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4 Elastic and thermal properties of simple crystals 35

4.1 Phonons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.1.1 1D atomic chain case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.1.2 Long wavelength limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.1.3 Extension to 2 atoms per primitive basis . . . . . . . . . . . . . . . . . . . 39
4.2 Thermal properties of phonons . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2.1 Einstein model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2.2 Debye model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.3 Thermal properties of simple electronic systems . . . . . . . . . . . . . . . . . . . 43
4.3.1 Thermal properties in metals . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.3.2 Energy bands, states, and gaps in brief . . . . . . . . . . . . . . . . . . . . 46

5 Recap of thermodynamics and statistical mechanics 50

5.1 Basic thermodynamic quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.2 Macro- vs. microstates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.3 Ergodicity and averages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

6 Structure-property relationships in simple liquids 56

6.1 From fluctuations to material properties . . . . . . . . . . . . . . . . . . . . . . . 56
6.1.1 Heat capacity and compressibility . . . . . . . . . . . . . . . . . . . . . . 56
6.1.2 Time correlation function . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

2
Contents 3

6.1.3 Einstein diffusion law (random walk) . . . . . . . . . . . . . . . . . . . . . 58

6.2 From the structure to material properties in the canonical ensemble . . . . . . . 59
6.2.1 The one-body density profile ρ(~r) . . . . . . . . . . . . . . . . . . . . . . . 59
6.2.2 The radial distribution function, g(r) . . . . . . . . . . . . . . . . . . . . . 61
6.2.3 From g(r) to macroscopic observables . . . . . . . . . . . . . . . . . . . . 63
6.2.4 Relation of g(r) to scattering . . . . . . . . . . . . . . . . . . . . . . . . . 65

7 Classical DFT (Density Functional Theory) 67

7.1 Mathematical background, functional calculus . . . . . . . . . . . . . . . . . . . . 67
7.2 Functionals in statistical mechanics: fluids in external fields . . . . . . . . . . . . 69
7.3 DFT (classical) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
7.3.1 DFT: basic theorems and variational principle . . . . . . . . . . . . . . . 72
7.3.2 Typical examples (approximations) for Fexc [ρ] and applications . . . . . . 74
7.3.3 λ-Integration and mean-field ("Vlasov") approximation . . . . . . . . . . . 77
7.3.4 More example applications . . . . . . . . . . . . . . . . . . . . . . . . . . 79

8 Electrostatics of liquid electrolytes and the Poisson-Boltzmann Equation 80

8.1 Electrostatics of molecular systems . . . . . . . . . . . . . . . . . . . . . . . . . . 80
8.2 Applications of PB in various geometries. . . . . . . . . . . . . . . . . . . . . . . 85
8.2.1 Spherical case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
8.2.2 Planar case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
8.2.3 Cylindrical case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

9 Coarse-graining of multi-component systems: effective interactions 93

9.1 Potential of mean force: effective interaction . . . . . . . . . . . . . . . . . . . . . 94
9.1.1 Two-body approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
9.1.2 Internal degree of freedom . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
9.2 Typical coarse-graining procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
9.3 Effective macromolecular interactions: examples . . . . . . . . . . . . . . . . . . 98
9.3.1 Depletion Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
9.3.2 DLVO interaction (Hamaker and Debye-Hückel) . . . . . . . . . . . . . . 101

10 Polyatomic molecules and polymers 105

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
10.1.1 Molecules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
10.1.2 Polymers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
10.1.3 Structures and architectures . . . . . . . . . . . . . . . . . . . . . . . . . . 106
10.1.4 Example of specific polymers . . . . . . . . . . . . . . . . . . . . . . . . . 107
10.1.5 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
10.2 Chain models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
10.2.1 Freely jointed chain (FJC) model . . . . . . . . . . . . . . . . . . . . . . . 110
10.2.2 Freely rotating chain (FRC) model . . . . . . . . . . . . . . . . . . . . . . 111
10.2.3 Hindered rotation model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
10.2.4 The Gaussian (or Rouse) chain . . . . . . . . . . . . . . . . . . . . . . . . 113
10.2.5 Form factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
10.2.6 Real chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

11 Free energy calculations and phase diagrams 116

11.1 Free energy calculations and entropy . . . . . . . . . . . . . . . . . . . . . . . . . 116
11.1.1 Thermodynamic Integration . . . . . . . . . . . . . . . . . . . . . . . . . . 116
11.1.2 Particle insertion method . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
11.1.3 Free energy perturbation approach . . . . . . . . . . . . . . . . . . . . . . 119
Contents 4

11.1.4 Reaction coordinate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

11.1.5 Importance sampling & Umbrella sampling . . . . . . . . . . . . . . . . . 121
11.2 From free energies to phase diagrams . . . . . . . . . . . . . . . . . . . . . . . . . 122

12 Appendix: Introduction Talk 126

List of Figures

1.1 Length scales in condensed matter physics. . . . . . . . . . . . . . . . . . . . . . 8

1.2 Length- and time-scales in condensed matter physics. . . . . . . . . . . . . . . . . 9
1.3 Macromolecules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.1 Bond types illustration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.2 Simple covalent bond illustration. . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 N a+ Cl− ionic bond illustration. . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 NaCl and CsCl structures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.5 Potential energy between two ions. . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.6 Illustration to calculation of the Madelung constant in 2D. . . . . . . . . . . . . . 15
2.7 Schematic illustration of H-bond in water molecule. . . . . . . . . . . . . . . . . . 16
2.8 Schematic illustration of London dispersion forces. . . . . . . . . . . . . . . . . . 17
2.9 Scheme of pair, V (2) , and the three body potential, V (3) . . . . . . . . . . . . . . 18
2.10 The complexity of a system can vary in shapes, polydispersity, and components
(x-axis) and the form of external potentials, possibly driving the system out of
equilibrium. Real systems are typically found on the top right corner of such a
diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.11 Typical atomic pair potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.12 Phase (liquid/solid) as a function of the packing fraction. . . . . . . . . . . . . . 22

3.1 Schematic of how atoms are arranged in crystalline, polycrystalline, and amor-
phous matter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 Schematic illustration of London dispersion forces. . . . . . . . . . . . . . . . . . 24
3.3 Hexagonal lattice in 2D. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.4 Schematic illustration of the conventional cell. . . . . . . . . . . . . . . . . . . . . 25
3.5 Cubic cells: simple, bcc, fcc. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.6 Illustration to the Miller indices. . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.7 Planes with different Miller indices in cubic crystals. . . . . . . . . . . . . . . . . 27
3.8 Schematic small-angle X-ray scattering setup. . . . . . . . . . . . . . . . . . . . . 29
3.9 Schematic small-angle X-ray scattering setup. . . . . . . . . . . . . . . . . . . . . 29
3.10 The Ewald sphere of the reciprocal lattice illustrating the Laue condition. . . . . 31
3.11 Schematic drawing to the geometry of the Bragg’s law. . . . . . . . . . . . . . . . 31
3.12 The Bragg interpretation of the Laue condition. . . . . . . . . . . . . . . . . . . . 32
3.13 Illustration to the derivation of the Bragg equation. . . . . . . . . . . . . . . . . 32
3.14 Construction of the. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4.1 1D atomic chain. Schematic illustration for longitudinal wave. . . . . . . . . . . . 36

4.2 1D atomic chain. Schematic illustration for transverse wave . . . . . . . . . . . . 36
4.3 First Brillouin zone of cubic lattice. . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.4 Optical and acoustical branches of the dispersion relation for a diatomic linear
lattice. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

5
List of Figures 6

4.5 Illustration to the Einstein’s model of heat capacity in solids. . . . . . . . . . . . 41

4.6 Illustration to the consideration of mode’s counting in 3D space. . . . . . . . . . 41
4.7 Illustration to the Debye’s model of heat capacity in solids. . . . . . . . . . . . . 43
4.8 Schematic model of a crystal of sodium metal. . . . . . . . . . . . . . . . . . . . . 44
4.9 Illustration for 1D problem of particle in a box. . . . . . . . . . . . . . . . . . . . 44
4.10 Fermi-Dirac distribution function at different temperatures. . . . . . . . . . . . . 45
4.11 Illustration to the periodic potential. . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.12 Square-well periodic potential (KPM) . . . . . . . . . . . . . . . . . . . . . . . . 47
4.13 Plot of energy ε vs. wavenumber k for the Kronig-Penney potential. . . . . . . . 48
4.14 Schematic illustration of energy gaps for insulator, semiconductor and metal. . . 49

5.1 Bulk modulus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

6.1 Random walk and diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

6.2 Forces of pair potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.3 Density distribution from averaging . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.4 Shell histogram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.5 (a) Pair distribution function, g(r), and (b) corresponding coordination number. 62
6.6 Applied external pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.7 scattering at macromolecular liquids . . . . . . . . . . . . . . . . . . . . . . . . . 65

7.1 Schematic plot of ρ(x) for LDA approximation. . . . . . . . . . . . . . . . . . . . 74

7.2 "Test particle" trick illustration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.3 Schematic plot for the Lennard-Jones potential. . . . . . . . . . . . . . . . . . . . 78

8.1 Dissolution process of sodium chloride crystal in water . . . . . . . . . . . . . . . 81

8.2 Schematic drawing of a micelle structure . . . . . . . . . . . . . . . . . . . . . . . 82
8.3 Schematic drawing of a charged colloid. . . . . . . . . . . . . . . . . . . . . . . . 82
8.4 Schematic drawing of a charged rod. . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.5 Schematic plot ρ(z) in planar case. . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.6 Schematic drawing of planar, spherical and cylindrical cases. . . . . . . . . . . . . 86
8.7 Schematic plot of ρ(z) in planar geometry for only counterions case. . . . . . . . 88
8.8 Schematic plot of ρ± (z) in planar geometry for added electrolyte case. . . . . . . 89
8.9 Double layer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
8.10 Two planar charged surfaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

9.1 Examples for two-component mixtures . . . . . . . . . . . . . . . . . . . . . . . . 93

9.2 Binary system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
9.3 Coarse graining a one component system, with internal degree of freedom . . . . 96
9.4 Typical coarse-graining procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
9.5 Veff through simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
9.6 Binary system with colloids and polymers . . . . . . . . . . . . . . . . . . . . . . 99
9.7 Binary system with colloids and polymers . . . . . . . . . . . . . . . . . . . . . . 99
9.8 Overlap volume of two colloids . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
9.9 Total effective pair interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
9.10 Van der Waals interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
9.11 Interaction between i and k across medium j . . . . . . . . . . . . . . . . . . . . 102
9.12 Hamaker interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
9.13 Charged colloids and salt in a water solution . . . . . . . . . . . . . . . . . . . . 103
9.14 Sketch of the DLVO potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

10.1 Structures and architectures of polymers . . . . . . . . . . . . . . . . . . . . . . . 106

10.2 Dendrimer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
List of Figures 7

10.3 Polypeptide backbone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

10.4 Definition of bond parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
10.5 Torsion angle vs. energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
10.6 Contour length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
10.7 Center-of-mass of polymer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
10.8 Freely jointed chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
10.9 Freely rotating chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
10.10Energy landscape and torsion angle . . . . . . . . . . . . . . . . . . . . . . . . . . 113
10.11The Gaussian chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

11.1 Bad sampling potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

11.2 Solvation of a LJ-sphere in water . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
11.3 Particle insertion method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
11.4 Insertion of multiple particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
11.5 Sampling unlikely states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
11.6 Umbrella sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
11.7 Free energy vs. density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
11.8 Phase diagrams with binodal and spinodal curves . . . . . . . . . . . . . . . . . . 125
Chapter 1

Introduction

Condensed matter physics is the field of physics that deals with the macroscopic and micro-
scopic physical properties of matter, especially the solid and liquid phases which arise from
electromagnetic forces between atoms. At high temperatures, a many-body system (molecules,
atoms) is in a dilute gas form. When temperature decreases, energetic attractions typically
dominate entropy and the particles condense into liquid and solid phases, cf. Fig. 1.1, which
feature distinct structures, dynamics, and macroscopic behavior. à Condensed (solid and liquid) phases
Interactions
Bonded / nonbonded / electrostatic atomic interactions
à Structure of molecules, solids and liquids

Simple phase diagram

Figure 1.1: Left: Simple condensed matter systems can be found in solid and liquid phases,
when a dilute gas condenses at relatively low temperatures. Right: Phase diagram of a simple
https://en.wikipedia.org/wiki/Phase_diagram
many-body system. The colored solid lines are coexistence lines between phases. Complex
systems, however, can have many more sub- and intermediate-phases (e.g., liquid crystals).

Condensed matter physicists seek to understand the behavior of these phases by experiments to
measure various material properties, and theoretically by applying the physical laws of quantum
mechanics, electromagnetism, statistical mechanics, and other theories to develop mathematical
models. Condensed matter physics is arguably the most active field of contemporary physics; it
overlaps with chemistry, materials science, engineering and nanotechnology, and relates closely
to atomic physics and biophysics.
In condensed matter physics, one can broadly distinguish between solid state physics and soft
condensed matter physics. Together, these fields span a wide range of length- and time-scales,
see Fig. 1.2. Solid state physics deals mostly with the very dense and crystalline (ordered) states
of matter and focuses on dense periodic lattices of atoms, molecules, as well as the electrons in
it. Electrons themselves can behave like a dilute gas, correlated liquids, or are highly localized
with distinct states. Electronic theory is a significant part of solid state physics.
In this lecture script we focus more on the structural arrangements of atoms and molecules in
the solid and liquid states, not so much on electrons. Large molecules, so called macromolecules,
or colloids and polymers, are especially the focus in soft condensed matter, see also Fig. 1.3.

8
Chapter 1. Introduction 9

Figure 1.2: Length- and time-scales in condensed matter physics, from electrons to materials.
For modeling, different methods, ranging from quantum mechanics to classical dissipative up
to continuum mechanics have to be employed.

Because macromolecules are much larger (& 10 − 100 nm) than atoms (' 0.1 nm), cf. Fig. 1.2,
the materials they constitute are very easily deformable under weak external mechanical and
energetic fields, hence, the term ‘soft’.
Soft matter is hence a subfield of condensed matter comprising a variety of physical systems that
are easily deformed or structurally altered by thermal or mechanical stress of the magnitude of
thermal fluctuations. They include liquids, colloids, polymers, foams, gels, granular materials,
liquid crystals, and a number of biological materials. Soft matter structures, such as liquids,
feature short-range order but no long-range order. Interestingly, glasses and amorphous solids
have liquid-like structures and are studied more in the soft matter community than in the solid
state community.
Soft materials share an important common feature in that predominant physical behaviors occur
at an energy scale comparable with room temperature thermal energy. At these temperatures,
quantum aspects are generally unimportant. Applications of soft condensed matter include
but are not limited to, foams and adhesives, detergents and cosmetics, paints, food additives,
lubricants and fuel additives, rubber in tires, etc., as well as biological materials, such as blood,
muscle, milk, yogurt, gelatine, etc. Also liquid crystals (as in LCDs, OLEDs, bacteria, etc.)
are an important subfield of soft matter. Polymers play a big role in soft matter applications
and polymer physics constitutes also an important subfield of soft matter.
The major part of this lecture focuses on general theoretical concepts to describe classical (non-
electron, non-quantum) many-body systems, in particular, the liquid state of simple systems in
soft condensed matter physics. The theoretical methods are thus based on thermodynamics and
advanced classical statistical physics, such as liquids state theory, classical density functional
theory, and polymer physics.
An Introduction powerpoint talk to this lecture script including more background is also pro-
vided in the Appendix.
Chapter 1. Introduction 10

Molecular ‘fluids’

Molecular systems: water, salt, (conjugated) organic

molecules, peptides, proteins, etc. (~ 0.1-10 nm)

+ Cl- Na+
-

~0.3
~ 2.5 nm
nanometer

(p-hexaphenyl, C36H26) ~ 5 nm
Polymeric systems

Long chain molecules (often in solution);

polyelectrolytes, polymer brushes & gels, etc. (~10-100 nm)

(thermosensitive)
core-shell hydrogels

hydrophilic hydrophobic
swollen collapsed
polymer
(polyethyleneglycole) switchable at critical temperature,
brushes
Colloidal fluids
size and properties tuneable

Macromolecules (~10 nm -1 µm) dispersed in a solvent

Examples: milk, ink, paint, blood, bacteria, big proteins,

model colloids, etc.

Figure 1.3: Examples of molecules and macromolecules constituting liquids and solids in soft
condensed matter physics.
Chapter 2

Bonds and Interactions

2.1 Chemical and physical bonding in solids and liquids

Atoms can form molecules by covalent bonding. Then, in the solid state, a large number of
atoms or molecules can also form strong chemical or physical bonds, leading to the creation
of a densely packed and usually well-organized aggregate. Ordered structures in solids can
be categorized into two-dimensional (2D) or three-dimensional (3D) periodic arrangements,
typically called crystals. These arrangements determine the long-range order and symmetry
exhibited by different types of solids in condensed matter. Weaker atomic or molecules bonds
or high temperatures lead to other important phases, such as the liquid phase, which features
only short-range order. At high temperature a liquid evaporates into a gas, where single atoms
or molecules are usually isolated.
Strong chemical or physical bonding in molecules and/or solids involves various types of inter-
actions that contribute to the stability and structure of the material. We typically distinguish

1. Covalent bond (directional bonding).


2. Ionic bond. 





3. Metallic bond. non-directional bonding



4. Physical bonding (hydrogen bond, van der Waals);



they are much weaker than types 1)-3).

Hence, a chemical bond is the association of atoms or ions to form molecules, crystals, and other
structures. The bond may result from the electrostatic force between oppositely charged ions
as in ionic bonds or through the sharing of electrons as in (directional) covalent bonds, or some
combination of these effects. Chemical bonds are described has having different strengths: there
are "strong bonds" or "primary bonds" such as covalent, ionic and metallic bonds, see above, and
"weak bonds" or "secondary bonds" such as dipole?dipole interactions, the London dispersion
force, and hydrogen bonding. The latter can also be termed as "physical bonds".
The classification of types of bonding in solids is primarily determined by the extent of overlap
between electronic wave functions. This overlap can be better understood if you recall the
concept of atomic orbitals, which describes the probability distribution of finding electrons
around an atom. For further reference, see Table 1.1 in [1].
In particular, the "valence band" refers to the outermost orbital containing electrons in a solid,
which defines the outermost electronic configuration of the material. The behavior of these
valence electrons significantly influences the properties and characteristics of the solid.

11
Chapter 2. Bonds and Interactions 12

Figure 2.1: The principal types of crystalline binding. In (a) neutral atoms with closed electron
shells are bound together weakly by the van der Waals forces associated with fluctuations in
the charge distributions. In (b) electrons are transferred from the alkali atoms to the halogen
atoms, and the resulting ions are held together by attractive electrostatic forces between the
positive and negative ions. In (c) the valence electrons are taken away from each alkali atom
to form a communal electron sea in which the positive ions are dispersed. In (d) the neutral
atoms are bound together by the overlapping parts of their electron distributions. From p.48,
[2].

2.1.1 Covalent bonds

Consider, for example, the di-atom H+
2 . We can write its Hamiltonian the following way:

~2 Ze2 Z 0 e2 ZZ 0 e2
H= − ∆ − − + ,
| 2m
{z } 4πε0 rA 4πε0 rB 4πε0 R
| {z }
kinetic operator Coulomb energies

where Z, Z 0 – valencies, ε0 – vacuum permittivity, e – elementary charge.

Next step would be deriving ψ from approximated solution of Schrödinger equation: Hψ = Eψ
(see derivation on p.5 in [1]).
Chapter 2. Bonds and Interactions 13

Figure 2.2: The simplest of a covalent bond (the H2+ molecule). a) Definition of the symbols,
b) bonding and antibonding combinations of atomic orbitals. The bonding combination leads
to an accumulation of charge between the nuclei which in turn gives rise to a reduction in the
Coulomb energy. c) The splitting of the atomic energy level into the bonding and antibonding
states. The greatest gain in energy is achieved when the bonding state is fully occupied – i.e.,
contains two electrons – and the antibonding state is empty (covalent bonding). From [1], p.5.

We obtain two possible solutions: "bonding" and "anti-bonding". Bonding case minimizes the
energy."Bond length" is defined as the distance between 2 bonded atoms − (for example,
C-C in diamond: 1.54 Å = 0.154 nm). Typically values lay in range of 1-2 Å.
This splitting also creates "band structure" in solids. Covalent bonding is directional, because
overlapping wave functions are anisotropic (see Fig. 1.3 in [1]). Typical binding energies of
covalent bonds (1 eV ≈ 100 molkJ
):

• C (diamond): 7.30 eV per atom (712 kJ/mol);

• Si: 4.64 eV per atom (448 kJ/mol);

• Ge: 3.87 eV per atom (374 kJ/mol).

Though C, Si, Ge prefer tetrahedral bonds (see Fig. 1.4 in [1]). Compare energy values to mean
thermal energy kB T (e.g. 1 kB T ≈ 2.5 kJ/mol at T = 300 K).

2.1.2 Ionic bonds

We define the ionization energy I as the energy we need to remove e− from a neutral atom. We
can also the electron affinity A which is the energy gain for adding e− to an atom (or energy
cost to remove e− from anion – negative atom). Essentially ionic bonding is the combination
of atoms with low I and high A. For example, consider Na+ + e− + Cl −→ Na+ + Cl− . In this
case, INa = 5.14 eV, ACl = −3.71 eV. (We count this negatively because it is an energy gain).
Then the net energy cost to move e− from Na to Cl is, I + A = 1.43 eV.
Chapter 2. Bonds and Interactions 14

Figure 2.3: Na+ Cl− ionic bond illustration.

This has to be compared to the electrostatic interaction. The electrostatic attraction at r0

from the Coulomb law is about ' −4.51 eV. That means we have the total gain of −3.08 eV =
(−4.51 + 1.43) eV which is in total achieved by ionization and the formation of an ionic pair.
General note: ionic crystals are always mixture of elements. Also their crystal structures are
determined by minimization of total Coulomb energy and optimal use of space for different types
of ions ("size" or "excluded volume" effect, also we will discuss later: hard sphere example, fcc
structure).

Figure 2.4: The two structures typical for ionic bonding in solids: a) NaCl structure; b) CsCl
structure. From [1].

Let us write the potential energy of atom i as: Ui ' Vij with pair potential:
X

e2 Cn
Vij ' ± + n,
4πε0 rij rij
| {z }
Coulomb

where typically n = 12, C12 > 0; in general, small contributions come also from dispersion (van
der Waals) forces, n = 6, C6 < 0, see later in the chapter.
Chapter 2. Bonds and Interactions 15

Figure 2.5: Potential energy between two ions.

We can write the total energy as the following sum:

e2 X ±1 Cn X 1
 
Utot = ·Ui = N − +
X
,
i
4πε0 r i6=j Pij rn i6=j Pijn
X ±1
with A = – Madelung constant (it depends on the crystal structure) and rij = rPij ,
Pij
i6=j
where r – nearest neighbour distance. Non-trivial, calculation of Madelung constant is difficult
(due to its conditionally converging sum).

Figure 2.6: Illustration to calculation of the Madelung constant in 2D. Also see exercise sheet.

Typical binding energies in crystal:


• NaCl (A = 1.748): 7.95 eV per ion-pair; 





• NaI: 7.10 eV per ion-pair; so-called "Madelung energies"



• KrBr: 6.92 eV per ion-pair.



Chapter 2. Bonds and Interactions 16

In ionic crystals electrons cannot move freely, hence they are typically non-conducting. Though
there are cases where ionic crystals can exhibit conductivity, for example at high temperatures,
they can undergo a process called thermal ionization, where some electrons gain enough energy
to break free from their bound states and become mobile. This can result in increased con-
ductivity or even partial or complete melting of the crystal. Also present defects or impurities
can create additional energy levels within the band structure, allowing for the movement of
electrons.
Qualitative measure of ionicity is given by "electronegativity scale" (by Pauling, Millikan, also
see Table 1.2 in [1]):

X = 0.184(I + A).
In ionic bonds atom with highest electronegativity X is the "anion" (-), which draws electrons
and "cation" (+) is the atom that loses electron. Generally, covalent and ionic bonds have very
different distribution of valence charges (see Fig. 1.8 in [1]).

2.1.3 Metallic bonds

Metals have extended wave functions (see Fig. 1.9 in [1]), which creates certain features of the
metallic bonds. Metallic bonds are more similar to covalent bonds but non-directional. Positive
atom cores are strongly "screened" by many neighbouring electron clouds. Overlapping valence
bands form a delocalized "conduction band" of an electron gas (or liquid). Metals also have
high conductivity (we will talk about this more later in Chapter 4.3).

2.1.4 Physical bonds

Hydrogen bond ("H-bond")
Hydrogen bonding is a type of intermolecular bonding that occurs when a hydrogen atom,
covalently bonded to a highly electronegative atom interacts with another electronegative atom
in a different molecule. Typical example – water molecule H2 O.

Figure 2.7: Schematic illustration of H-bond in water molecule. High electronegativity of O

drives electron to the center. Here the symbol δ + means a partial positive charge, and δ − means
a partial negative charge, so the O-H bonds in water molecules are polar. Image source.

Strong local electrostatic interaction (essentially Coulomb interaction) is created and also it
leads to local structure in liquids. Hydrogen bonding is also extremely important in biological
Chapter 2. Bonds and Interactions 17

molecules, where the bonds hold together strands of DNA or proteins in water. However, H-
bonding is not always easy to characterize because it depends on electronegativity X of atoms,
which can vary. Besides, H-bond is context-dependent (e.g., different in vacuum and in solvent).
Typical magnitude of energies are ∼ 0.1 eV ∼ 10 kJ/mol.

Van der Waals bond

The van der Waals (vdW) force refers to a type of intermolecular force that arises due to tem-
porary fluctuations in electron density within molecules. Specifically, the fluctuation-induced
force is a component of the vdW force that occurs when temporary imbalances in electron
distribution induce a temporary dipole in one molecule, which then creates a dipole in a neigh-
boring molecule. These dipoles result in an attractive force between the molecules, known as
the fluctuation-induced or London dispersion force.

Figure 2.8: Schematic illustration of London dispersion forces. At the infinite distance there is
a homogeneous electron distribution (left), but once the molecules are brought closer together
electrons fluctuate in a correlated way and dipole-dipole attraction occurs (right).

Using quantum mechanical perturbation theory and multiple expansions leads to an interaction
of approximative form (between 2 atoms A and B):

3 IA IB αA αB C6
V (r) ≈ − = 6,
2 IA + IB r 6 r
where I – ionization energy, α – polarizabilities (Pi = αi Ei ). (In your exercise sheet you will
also see the classical derivation for fluctuating harmonic oscillators).
Van der Waals forces have several features we mention here, for further reading consider p.53,
Chapter 3 in [2]. The vdW forces are ubiquitous: they are always acting, also between atoms
with closed electron shells (e.g. noble gas). They can also lead to molecular crystals (or low-T
noble gas crystals). VdW force leads to Hamaker interaction between macromolecules and also
motivates the generic Lennard-Jones atom-atom potential.

2.2 Interactions and potentials

Now to describe generally a system of N particles we would consider their positions and mo-
menta and how they change in time. These are typically trajectories in phase space {~rN (t), p~N (t)}.
Generally, the trajectories derive from a Hamiltonian H of typical form

H = Hkin + Hext + Hint . (2.1)

The kinetic energy Hkin is defined as

N
p~i 2
Hkin ≡ K̂ = (2.2)
X

i=1
2mi
Chapter 2. Bonds and Interactions 18

with p~i and mi being the momentum and mass of the particle i, respectively, and the symbol
ˆ indicates an instantaneous value (not an average). Hext describes the action of an external
potential Vext (e.g., an electrostatic field)
N
Hext = Vext (~
ri ) (2.3)
X

i=1

acting on particle i at position r~i .

In general the force field can be defined by the negative gradient of the potential energy from
particle-particle interactions:
F~i = −∇i Hint . (2.4)
The instantaneous potential energy from particle-particle interactions can be approximated as
N N
1X (2) 1 X (3)
Hint ' V (~ ri , r~j ) + V (~ ri , r~j , r~k ) + higher order contributions (2.5)
2 i6=j ij 6 i6=j6=k ijk

with V (2) being the pair potential and V (3) the three body contribution (see Fig. 2.9). The
latter expresses the change in the interaction of two particles, if a third one approaches. It
can often be neglected but its magnitude is system-dependent and should be scrutinized. It
is for example non-negligible in highly polarizable systems. The pair potential approximation
is applied almost in every study, as already implicitly assumed above when we discussed, e.g.,
Lennard-Jones. Usually one simply denotes the pair potential with V (r).

Figure 2.9: Pair potential, V (2) , between particle i and j, and three body potential, V (3) ,
represented by an angular potential.

We can now define a ‘simple’ system: Definition of a "simple" system:

• consists of only pair potentials (no triplet or higher order)

• only distant dependent potentials (no anisotropic (angle-dependent) potentials)

V (r~1 , r~2 ) ≡ V (|r~1 − r~2 |) ≡ V (r)

• no bonded potentials (such as molecules, polymers etc.)

Definition of a "complex" system: everything else!

We will mostly discuss simple systems, in particular, when we talk about ‘theory’, but of
course in simulations we want to address more complex systems, including, e.g., metals with
many-body interactions or bonded molecules, such as complex solvents or polymers. A nice
diagram of complexity of an interacting many-particle system is shown in Fig. 2.10.
 Complexity of a molecular system
Chapter 2. Bonds and Interactions 19

H. Löwen, J. Phys. Condens. Matter 13,

R415 (2001)

‚Applied‘ and biological

systems (intrinsic
nonequilibrium, driven,
reaction, etc.)

polymers,
… water, salt, …
proteins,
Energy Hamiltonian
X
THEO4 Hint = V (|ri rj |)
i6=j
“simple”
Figure 2.10: The complexity of a system can vary in shapes, polydispersity, and components
(x-axis) and the form of external potentials, possibly driving the system out of equilibrium.
Real systems are typically found on the top right corner of such a diagram.

Let us now look at some more examples for typical pair potentials.

2.2.1 Interatomic pair potentials

With ‘inter’-atomic potentials we mean interactions between atoms or particles which are not
part of the same molecule. A very nice summary of interactions can be found in the book by
Israelachvilli [3].
Except for the case of ideal gas where particles are not interacting with each other, i.e. V (r) = 0,
all atoms have a short-ranged repulsion due to the Pauli exclusion principle (Born-Mayer
1932):
Vsr (r) = Ae−αr (2.6)
Also all atoms have a dispersion (van der Waals) attraction due to the inherent fluctuating
electronic polarization:
c6
VvdW (r) = − c6 ≡ material constant (2.7)
r6
To model a complete atomic pair potential, both contributions have to be taken into account.
The combination of both effects can be seen in Figure 2.11.
Combining both contributions we get the "Buckingham" pair potential for atoms (1938):
c6
VB (r) = Ae−αr −
r6
6 !
 rmin

−2α(r−rmin )
= 6e − αrmin (2.8)
αrmin − 6 r

or, a slightly modified but more popular version, the Lennard-Jones pair potential (∼1930)
Chapter 2. Bonds and Interactions 20

which we introduced already above:

 12  6 !
σ σ
VLJ (r) = 4 −
r r
12 6 !
rmin rmin
 
= −2 (2.9)
r r
1
with the identities rmin = σ2 6 and VLJ (σ) = 0 and VLJ (rmin ) = −. The approximation of the
short-ranged part as r−12 is simply taken for convenience as this way it just becomes also a
power-law, like the van-der-Waals-potential ∝ r−6 .

In shorter terms with A = 4σ 12 and C = 4σ 6 we can write

A C
VLJ (r) = 12
− 6 (2.10)
r r

Figure 2.11: Schematic Lennard-Jones atomic pair potential with minimum at rmin .

The Lennard-Jones force is thus given by the negative gradient of VLJ (r):
!
σ 12 σ6
FLJ (r) = −∇r VLJ (r) = 4 12 13 − 6 7 (2.11)
r r

The generalization of the 12-6 Lennard-Jones potential is the n-m Mie potential:
1  m  n 
 mm σ σ
 
m−n
Vn-m (r) = − (2.12)
m−n nn r r
1
with the identities rmin =  m m−n and V
n-m (σ) = 0. The prefactor is defined auch that  still


n
describes the minimum energy of the potential. Due to the arbitrary exponents it is a very
versatile potential form to fit, e.g., coarse-grained macromolecular interactions.

The Morse-potential is an intramolecular potential which can be described as the following:

 
VMorse (r) =  e−2α(r−rmin ) − 2e−α(r−rmin ) (2.13)

It is a good model for metals as the potential is more "bonding"-like.

The Weeks, Chandler, Anderson (WCA) pair-potential, is an intermolecular potential to

describe interactions for all particles on all scales, i.e. colloid (≈ 10 − 100 nm macromolecules
Chapter 2. Bonds and Interactions 21

dispersed in a solvent, blood, cells bacteria, ...). The WCA potential is well suited to describe
short ranged, steric "excluded volume" repulsion. It is like a step repulsion which goes to zero
at r = σ21/6 . The functional form of the WCA is given by:

VLJ (r) +  if r ≤ σ21/6

(
VWCA (r) = (2.14)
0 otherwise

If we consider charged, Ionic materials, see above, we have to take into account Coulomb
interactions:
VC λB
(r) = Z1 Z2 (2.15)
kB T r
2
where Zi indicates the ion valency and λB = 4π0ekB T is the so-called Bjerrum length. The
Bjerrum length indicates the separation at which the electrostatic interaction between two
elementary charges is comparable in magnitude to the thermal energy scale, kB T .
For charge-induced polarization the potential is given as:

Z 2α 1
Vpol (r) = − (2.16)
2(4π0 )2 r4
where α is polarization and Z is the ion valency.
The Born-Mayer potential is a way to describe ions by combining the short-ranged potential
with the Coulomb potential:
B
Vij (r) = Ae−2rij + Zi Zj (2.17)
rij
This works well for ionic crystals as for example i = +, − in a NaCl-crystal (sodium chloride).

Ions in water
Alternatively one could use the Lennard-Jones with the Coulomb potential for ions in water:

4
 12  6 !
Vij σ σ λB
= − + Zi Zj (2.18)
kB T kB T r r rij

which is a very popular version in literature, in particular for molecular simulations in bio-
physics.

Finally, we note also on a model potential: the hard-sphere/disk (HS) potential is a conve-
nient potential to model particles in the statistical mechanical theory of fluids and solids for all
scales. Within this potential, particles are defined simply as impenetrable spheres/disks that
cannot overlap in space. The functional form of such a model is given by:

∞ if 0 ≤ r ≤ σ
(
VHS (r) = (2.19)
0 otherwise

Note there is no energy in the system in equilibrium (no overlapping states)! Hence, the whole
system behavior is only driven by entropy. It is such a good reference model for ‘excluded-
volume’ interactions. The packing fraction defines the fraction of space filled by particles
brought together. Assuming hard-sphere with diameter σ, the packing fraction is defined as:

π 3N
φ= σ = vHS ρ with 0 < φ < 1 (2.20)
6 V
where vHS = 43 πr3 and ρ being the HS volume and the density, respectively. As sketched
in Fig. 2.12, a first order transition between liquid/gas and solid occurs for 0.45 < φ < 0.58,
Chapter 2. Bonds and Interactions 22

while φ = 0.74 corresponds to a closed-packed structure such the one obtained for a fcc crystal
structure. The reason for the phase transition is that the free volume per sphere can be increased
(leading to a maximization in entropy) by assuming an ordered crystal state. Hence, in a hard
system order is driven by entropy!

Figure 2.12: Phase (liquid/solid) as a function of the packing fraction.

Chapter 3

Structure of solids

Figure 3.1: Schematic of how atoms are arranged in crystalline, polycrystalline, and amorphous
matter. Image source.

Minimization of the total free energy of bonded materials determines their crystal structure.
The periodic arrangement in 3D (2D as well) is essentially what we call the crystal structure. It
is characterized by the long range order (apart from defects, tilted domains, etc.). In contrast
to the "amorphous state" (where atoms are "frozen" or "glassy"), exhibiting no long-range order,
but local (also "short-range order") order exists, like in interacting liquids. Materials in practice
are often "composites" (mixtures), e.g., alloys (bronze: copper and tin).

3.1 Crystal lattice

Lattice is an infinite set of points defined by integer sums of a set of linearly independent
primitive lattice vectors. Let us define the lattice vector in 2D:

~rn = n1~a + n2~b,

with n1 , n2 being integers. ~rn describes the lattice points. We can also define the cell area:
|~a ×~b|.

23
Chapter 3. Structure of solids 24

Figure 3.2: Schematic illustration of a unit cell in 2D.

If one unit cell contains one atom (i.e., 1 lattice point), as shown in the Fig. 3.2, then it is
called a "primitive cell" – the smallest possible unit cell. Crystal lattices can exhibit certain
symmetries (see Fig. 7, p.8 in [2]):

• γ = 90◦ – rectangular lattice;

• γ = 90◦ and in addition a = b – square lattice;

• γ = 120◦ and a = b – hexagonal lattice. The hexagonal structure is the closest-packed 2D

structure of disks or repulsive systems (91% occupied by disk area).

Figure 3.3: Hexagonal lattice in 2D. a = b and γ = 120◦ .

If a = b and γ arbitrary, the lattice is better described as a non-primitive "centered rectangular

lattice" with a 6= b and γ = 90◦ , containing 2 atoms per unit cell (see Fig. 3.4 below). The
points A and A0 form the so-called "basis": a group of atoms reflecting identical environments
at the lattice points. This type of a unit cell is also called "conventional cell", it reflects the
symmetry of the lattice better.
Chapter 3. Structure of solids 25

Figure 3.4: Schematic illustration of the conventional cell.

In general: crystal structure = lattice + basis. We can expand our considerations to the 3D
case. There the lattice vector is:

~rn = n1~a + n2~b + n3~c,

with angles: α, β, γ. The volume of a primitive cell in 3D is defined as:

Vc = |~a · (~b ×~c)|.

(Also see Table 2.1, p.23 in [1] for classification of basis vector systems and Fig. 2.3, p.24).
In 3D, in addition to rectangular-centered, we can have base-, face- and body-centered cells.
In 3D, symmetries allow for 14 possible basic lattices, the "Bravais lattices" to describe all
structures, however they do not necessarily describe primitive cells.
Why there are, e.g., 3 different cubic cells? Structures may have the same symmetry (cubic,
for example), but a different base, i.e., have different primitive cells.

Figure 3.5: Simple cubic (primitive), body-centered cubic (bcc), face-centered cubic (fcc) cells.

3.1.1 Common crystal structures

For more details see Chapter 1 in [2], specifically for 3D lattice types: Table 1-2 and Fig. 8-12
on p.9-11. The coordination number of a lattice is the number of nearest neighbors any point
Chapter 3. Structure of solids 26

of the lattice has. The packing fraction φcp is the maximum proportion of the available volume
that can be filled with hard spheres. We summarize some properties:

1) face-centered cubic (f cc) (see Fig. 3.5)

• highest possible packing for spheres (φcp ∼ 74 %)

• coordination number = 12
• ABC stacking of 2D hexagonal planes

2) hexagonal close packing (hcp)

• ABAB stacking
• coordination number = 12

3) body-centered cubic (bcc) (see Fig. 3.5)

• coordination number = 8

4) diamond structure (see Fig 2.12, p.35 in [1])

• 2 interpenetrating f cc structures (basis - 2 identical atoms)

• coordination number = 4
• tetrahedral
√
3a
• nearest-neighbour distance , where a – edge length of cubic cell; φcp ∼ 34 % only
4
(quite loosely packed)

5) Zinc Blende structure (ZnS)

• like diamond structure, but basis consists of 2 different atoms

• coordination number = 4

6) Ionic structures

• such as NaCl and CsCl

• derived from f cc and bcc structures, respectively (NaCl: 2 shifted f cc for Na+ and
Cl− ; CsCl: replace central atom in bcc by other ion)
• coordination number (NaCl) = 6
• coordination number (CsCl) = 8

3.1.2 Facets and Miller indices

Finite pieces of crystals have facets: if we slice a crystal along a plane, we obtain a 2D surface
of a certain lattice structure. We can use an index scheme. First, we find the intercepts on the
axes in terms of the lattice constants ~a1 , ~a2 , ~a3 . Second, we take the reciprocals (inverted) of
those numbers and reduce to 3 smallest possible integers with the same ratio. The resulting
set of indices, written as (hkl), are called "Miller indices" and is the index characterizing the
1 1
plane. See Fig. 3.6, there 2, 2, 1 → , , 1 → 1, 1, 2; this set (h, k, l) is called the "index of the
2 2
plane". A negative intercept is denoted with a bar above the values, e.g., (11̄2).
Chapter 3. Structure of solids 27

Figure 3.6: Illustration to the Miller indices definition. Intercept of plane with lattice axis at
2~a1 , 2~a2 and ~a3 , defining plane as (112).

Figure 3.7: Planes with different Miller indices in cubic crystals. Image source.

The vector perpendicular to the plane ~n for the specific case of a cubic system is given by Miller
indices, ~n = [hkl], but this is not true generally. Generally, it is that ~n ∼ ~v1 × ~v2 where ~vi are
the plane-spanning vectors.
Chapter 3. Structure of solids 28

3.2 Reciprocal lattice

How do we determine and measure structure of condensed matter? For large macromolecules
(e.g. colloids) measurements are in real space (microscopy, etc.). For small atoms/molecules
we use reciprocal space methods: diffraction by electrons (10 eV - 1 keV), neutrons (10 meV
- 1 eV), phonons (1 keV - 100 keV) (see Fig. 1. p.24 in [2], Fig. 3.11, p.68 in [1]). Different
wavelengths can probe structure on different length scales.
To model diffraction and related reciprocal aspects, it is important to introduce the concept of
the reciprocal lattice. First, we define the reciprocal vectors:

~b1 = 2π ~a2 ×~a3 ~a3 ×~a1 ~a1 ×~a2

; ~b2 = 2π ; ~b3 = 2π .
~a1 · (~a2 ×~a3 ) ~a1 · (~a2 ×~a3 ) ~a1 · (~a2 ×~a3 )
| {z }
|...| = Vc – cell volume

There exists an orthogonality relation between the vectors ~bi and ~aj in the form of:

~bi ·~aj = 2πδij ,

1, if i = j,
(
where δij = is the Kronecker delta. The set of vectors {~bi } forms a new lattice,
0, if i 6= j
described by:
~ = ν1~b1 + ν2~b2 + ν3~b3 ,
G
~ = h~b1 + k~b2 + l~b3 is the
where ν1 , ν2 , ν3 correspond to Miller indices hkl, in the sense that G
vector perpendicular to the (hkl) planes.
Consider now the electron density of the crystal n(~r). The crystal is invariant under lattice
translation with vector T~ = n1~a1 + n2~a2 + n3~a3 , hence:

n(~r + T~ ) = n(~r),
which can be written as the Fourier decomposition:
~
n(~r) =
X
n~k eik·~r ,
{~k}

1
Z
~
with ~nk = dV n(~r)eik·~r , representing the Fourier transform of n(~r) in a single unit
Vcell unit cell
cell. {~k} is the set of vectors that leave n(~r) invariant, i.e.:
~ ~ ~ !
n(~r + T~ ) = n~k ei(k·~r) ei(k·T ) = 1.
X

{~k}

Let us insert the reciprocal lattice for {~k}:

 
~ ~
 
eiG·T = exp i(ν1~b1 + ν2~b2 + ν3~b3 )(n1~a1 + n2~a2 + n3~a3 ) = exp2πi (ν1 n1 + ν2 n2 + ν3 n3 ).
 
| {z }
∈ Z(integer)

The reciprocal lattice fullfills the condition of:

~ ~
ei(k·T ) = 1 as e2πin = 1. (3.1)
Therefore, G
~ is an appropriate representation of the lattice symmetry in the reciprocal space.
Chapter 3. Structure of solids 29

3.2.1 Scattering

Figure 3.8: Schematic small-angle X-ray scattering setup.

Let us define the scattering amplitude as:

Z
~ ~
F (θ) = dV n(~r)e−i(k−k0 )·~r , (3.2)

with F (k) ∼ F (θ) and intensity I(θ) = |F (θ)|2 . We make use of the Born approximation (more
details we will discuss later). For now, we note the following properties of F (θ):

• it is proportional to n(~r);
~ ~ ~
• it is proportional to the "phase factor": ei(k0 −k)·~r = e−i∆k·~r .

Figure 3.9: Assume the wavevectors of the incoming and outgoing beams are ~k0 and ~k. The
difference in phase angle for the incident wave is equal to ~k0 ·~r = |k||r| sin φ and for the diffracted
wave it is ~k ·~r. Then for the total difference θ in phase angle: (~k0 − ~k) ·~r. Adapted from Fig. 6,
p.30 in [2].

• F (θ) depends on particular scattering process, for example in case of Rayleigh scattering
(photon-atom, with large photon wavelength, depends on energy and mass of scatterer).

Let us assume elastic scattering: |~k0 | = |~k|. We can insert the following expression of Fourier
~
decomposition using reciprocal lattice symmetry into the Eq. (3.2): nG~ eiG~r = n(~r). Then:
X

G
Chapter 3. Structure of solids 30

XZ  
F (θ) = ~ − ∆~k) · ~r ,
dV nG~ exp i(G (3.3)
G
 
~ − ∆~k) · ~r = 1 for G
where the exponent exp i(G ~ = ∆G
~ or ∼ 0 else .
Now we arrive to the "Laue condition" or "Laue equations":

if ∆~k = G
~ → F (θ) has maximal amplitude. (3.4)
Hence, the diffraction pattern provides a mapping of the reciprocal lattice, which is a mathe-
matical representation of the arrangement of points in the crystal lattice!
This mapping allows us to analyze and understand the structure of the crystal based on the ob-
served diffraction pattern. It is often said that "scattering is an experimental Fourier transform"
because scattering experiments reveal information about the spatial distribution of scattering
centers in the sample. By measuring the scattering intensity at different angles or positions,
we can effectively transform the spatial information into reciprocal space, analogous to how a
Fourier transform converts a function from the time or spatial domain to the frequency domain.
∆~k = G
~ can also sometimes be expressed, using the (hkl) – the Miller indices:

~a1 · ∆~k = 2πh; ~a2 · ∆~k = 2πk; ~a3 · ∆~k = 2πl,

Rewriting the Eq. (3.4) as: ~k = G
~ + ~k0 or k 2 = (G + k0 )2 , since k 2 = k 2 , we can obtain a different
0
form of Laue condition (note: G ~ and −G ~ are invariant):

2~k0 G ~ 2 = 0 or 2~k0 G
~ +G ~ =G
~2 (3.5)
Thus we arrive to two important relations (see Chapter 3.4, p.61 in [1] for proofs):

1) G ~ hkl = h~b1 + k~b2 + l~b3 , are vectors perpendicular to the (hkl)

~ with components hkl, also G
planes;

2) using Eq. (3.1) we can define the nearest distance between the Miller planes:

2π
dhkl = , (3.6)
~ hkl |
|G
~ hkl | = n · 2π for all planes n = 0, 1, .., ∞.
i.e., in general |G d

The Laue condition can be conveniently pictured by the Ewald sphere, as shown below in
Fig. 3.10. Diffraction will only take place for reciprocal lattice points on the sphere surface
with: ∆~k = G
~ and |~k0 | = |~k| which is the radius of the Ewald sphere.
Chapter 3. Structure of solids 31

Figure 3.10: The Ewald sphere of the reciprocal lattice illustrating the Laue condition ~k − ~k0 = G
~
(see Eq. (3.4)). Adapted from Fig. 3.4, p.60 in [1].

3.2.2 Bragg interpretation of the Laue condition

2π
With dhkl = (Eq. (3.6)) and the Laue condition, Eq. (3.5): 2~k0 G~ =G
~ 2 we get:
~ hkl |
|G
2~k0 G ~ 2hkl ⇐⇒ 2|k0 | cos φ = 2πn ,
~ hkl = G
dhkl

Figure 3.11: Schematic drawing to the geometry of the Bragg’s law. Here cos φ = sin θ, which
we use in the derivation.

2πn  2π  2πn
2k sin θ = ⇐⇒ 2 sin θ = .
dhkl λ dhkl
Simplifying the expression above we arrive at the Bragg’s law (1913):

2dhkl sin θ = nλ . (3.7)

See Fig. 3.12 in for Bragg’s interpretation in real and reciprocal space.
Chapter 3. Structure of solids 32

Figure 3.12: The Bragg interpretation of the scattering condition. Since the vector G ~ hkl lies
perpendicular to the lattice planes (hkl) in real space, the scattering appears to be a mirror
reflection from these planes. From Fig. 3.6, p.62 in [1].

Intuitive real space picture: positive interference only if 2d sin θ ∼ λ (= wavelength of incoming
wave).

Figure 3.13: Illustration to the derivation of the Bragg equation 2d sin θ = nλ; here d is the
spacing of parallel atomic planes and 2πn is the difference in phase between reflections from
successive planes. From Fig. 2, p.24 in [2].

3.2.3 Brillouin zones

The Laue condition of the form Eq. (3.5) is also encoded in the concept of the Brillouin zone.
Our starting point is the "Voronoi-decomposition". Let us consider a lattice, we connect all
points, draw planes through the middle of each connecting line perpendicular to it. The smallest
remaining volume enclosed by these planes in real space is the "Wigner-Seitz cell" (it is a
Chapter 3. Structure of solids 33

primitive cell). The Wigner-Seitz primitive cell in the reciprocal lattice is called Brillouin
zone (see Fig. 3.14 below).

Figure 3.14: Construction of the Wigner-Seitz cell. From Fig. 10, p.35 in [2].

~ = G2 , then it follows ~k Ḡ = G (Voronoi condition).

   2
Consider Eq. (3.5), 2~k G
2 2
The Brillouin zone hence contains all vectors ~k, which can be Bragg-reflected by the crystal
(more explanations and illustrations see Fig. 9a, p.34 in [2]). Historically, Brillouin zones were,
dψ
however, introduced to understand periodic electronic wavefunctions in crystals (e.g. =0
dr
at the Wigner cell’s boundaries).

3.3 Crystal structure factor

From the above considerations, we can write the scattering amplitude as:
Z
~
FG~ = N dV n(~r)e−iG·~r = N · SG~ ,
cell
for N cells. Then the we define: SG~ as the crystal "structure factor" (not to confuse with
S(~k) later). It is useful to split the information in the electron density distribution n(~r) defined
by the atoms at positions ~rj in the unit cell (S atoms in cell (basis)) as:
S
n(~r) = nj (~r − ~rj ) .
X

j=1
| {z }
density due
to atom j

For the structure factor then we have:

XZ ~ r ~ rj
Z
~
−iG·~ −iG·~
SG~ = dV nj (~r − ~rj )e = dV nj (~r − ~rj )e−iG(~r−~rj ) , (3.8)
X
e
j j

where the first term of the sum is property of lattice and the second term is technically the
Fourier transform of j-th atom’s electron distribution, the so called "atomic form factor" fj . In
practice it is important to account for different atom types in scattering (e.g., electrons versus
neutrons versus photons). The values of fj can be tabulated for single atoms or have to be
calculated/measured (with approximative deconvolutions of the signal). Important to note is
that measured is not the scattering amplitude, but the intensity proportional to amplitude
squared:
Chapter 3. Structure of solids 34

N X
N
~
~ ∼ |F ~ |2 ∼ e−iG(~rj −~rk ) fj fk . (3.9)
X
I(G) G
j k

1 X X −iG(~ ~ rj −~
For identical atoms: I(G) ~ with S(G)
~ ∼ N f 2 S(G) ~ = e rk )
. See examples (bcc,
N j k
fcc) on p.39-40 in [2]. Later in liquids, we will see, that I(~k) ∼ S(~k). (S(~k) is a more general
term of the structure factor, and also applies to liquids).
Chapter 4

Elastic and thermal properties of

simple crystals

4.1 Phonons
At temperatures above absolute zero (T = 0 K), the atoms comprising a crystal lattice are not
static but exhibit dynamic motion. This atomic motion gives rise to a range of fascinating
phenomena, which have significant implications in condensed matter physics.

• One prominent consequence of atomic motion in a crystal is the propagation of sound

waves. As the atoms move, they transmit mechanical disturbances through the crystal
lattice, resulting in sound propagation.

• Furthermore, the motion of atoms in a crystal lattice leads to lattice vibrations, also
known as phonons. Phonons represent quantized vibrational modes of the crystal lattice,
resulting from the collective motion and interactions of atoms within the material. These
lattice vibrations play a fundamental role in shaping the material’s physical properties,
such as thermal conductivity, electrical conductivity, and optical behavior.

• In addition to the lattice vibrations, other couplings come into play within the crystal
lattice. These include interactions between the atoms and electrons, as well as external
fields that influence the behavior of the lattice.

• Resulting lattice vibrations have significant implications for the material’s heat capacity,
heat conductivity, and thermodynamic properties. By studying the behavior of these vi-
brations, we can better understand the mechanisms governing the transfer and dissipation
of heat in solids.

Let us explore the determination of the frequency of elastic sound waves in terms of wave
vectors and elastic constants. The mathematical solution for finding the frequency becomes
particularly straightforward when considering specific propagation direction, such as the [100],
[110], and [111] planes. In these cases, entire planes of atoms actively participate in the motion
induced by the elastic sound wave.

4.1.1 1D atomic chain case

Consider small displacements in a cubic crystal (see Fig. 4.1 and Fig. 4.2). We assume that
the elastic response of the crystal is a linear function of the forces. Then, we can use simple
classical linear treatment and adding simplifying assumption of looking only at the nearest
neighbor interactions.

35
Chapter 4. Elastic and thermal properties of simple crystals 36

Consider a plane s, that is subjected to a force. Neighboring planes are in a equilibrium distance
a. Then, the force on s is

Fs = C(us+1 − us ) + C(us−1 − us ), (4.1)

where C is a force constant and the terms in a mutually cancel due to symmetry. This expression
is linear in the displacements and is of the form of Hooke’s law.

Figure 4.1: 1D atomic chain. Schematic illustration for longitudinal wave. Oscillation happens
in the direction of propagation. (Dashed lines) Planes of atoms when in equilibrium. (Solid
lines) Planes of atoms when displaced as for a longitudinal wave. The coordinate u measures
the displacement of the planes. From p.90, [2].

Figure 4.2: 1D atomic chain. Schematic illustration for transverse wave. Oscillation is perpen-
dicular to the direction of propagation. Red dashed line shows planes of atoms as displaced
during passage of a transverse wave. From p.90, [2].

Fs can also be interpreted as force on one atom in plane (what gives us an effective 1D model).
Chapter 4. Elastic and thermal properties of simple crystals 37

The constant C is derived from harmonic approximation of binding potential and represents the
force constant between nearest-neighbor planes; it will also differ for longitudinal and transverse
waves. According to the Eq. (4.1) we can write the equation of motion of an atom in plane s
as:

d2 us (t)
M = C(us+1 (t) + us−1 (t) − 2us (t)),
dt2
where M is atomic mass. (We see in continuous limit, that M ü = C ü, is a wave-equation like
in electrodynamics. Also consider referring to the decent Wikipedia article: Phonons).
We look for time-dependent wave solutions of form: us (t) = ûs eiωt for all s. Then:

−M ω 2 ûs = C(ûs+1 ) + ûs−1 − 2ûs ).

For a lattice spacing a, the space solutions are of the form: us±1 = u0 e−ika(s±1) (or us = u0 eikas ),
then:

−ω 2 M u0 eiska = Cu0 (ei(s+1)ka + ei(s−1)ka − 2eiska )

As we cancel the term u0 eiska on both sides:

ω 2 M = −C(eika + e−ika − 2),

with 2 cos ka = eika + e−ika , we finally find:

2C
ω2 = (1 − cos ka) (4.2)
M
This is a "dispersion relation", a function ω(k). It describes the dispersion of sound in a simple
cubic solid (for long wavelengths). (Also recall for light: ω = ck). We call these elastic modes in
crystals "phonons". The Phonon is actually a quantized quasi-particle, because at low T in a
full quantum mechanical treatment the modes of vibration are quantized. (Recall QM: energy
of an elastic wave ε = (n + 12 )~ω, like a harmonic oscillator).
Eq. (4.2) can be also written as following using the trigonometry formula sin2 x = 12 (1 − cos 2x):
 4C  
1
  4C 1/2 
1

ω =
2
sin 2
ka or ω= | sin ka |.
M 2 M 2

Figure 4.3: Dispersion relation plot of ω versus k. The region of k  1/a λ  a corresponds to
the continuum approximation; here ω is directly proportional to k. From [2], p.92.
Chapter 4. Elastic and thermal properties of simple crystals 38

d(ω 2 )
ω(k) has maxima at kmax = ± πa as derived from ω 2 = ( 2C
M (1−cos ka)) and dk ±π/a = M
2Ca
sin ka =
±π/a
0. The ratio of displacements of successive planes is:

us+1 u0 exp(i(s + 1)ka)

= = eika .
us u0 exp(iska)
From this follow the properties of the first Brillouin zone:

• the range [−π; +π] for ka covers all independent values of the exponential

• we do not need to consider values of |ka| > π, as they would describe the same motion as
k-values inside the first Brillouin zone (see Fig. 5, p.93 in [2])
π
• at the boundaries kmax = ± , the solution is not a travelling wave, but a standing wave
a
instead, as us = u0 exp{±isπ} = u0 (−1)s , meaning that atoms move in exactly opposite phases
(us = ±1 if s is even or odd → the wave does not move neither right nor left)

The transmission velocity of a wave packet is defined by the group velocity, via:

dω
vg = or ~vg = ∇~k ω(~k).
dk
The group velocity is a property of wave packets, which are localized disturbances or "packets"
of energy that travel through a medium. This concept is also applicable to sound waves, in that
case, these packets represent compressions and rarefactions of air molecules. From definition
then the speed of sound refers to the rate at which sound energy travels through a medium.
However, in a dispersive medium the speed of sound becomes a function of sound frequency
due to the dispersion relation. Each frequency component within a sound wave propagates at
its own speed, known as the phase velocity ("speed of amplitudes") and defined as:
ω
vp = .
k
Overall, phase velocity helps to characterize the behavior of individual wave components, while
group velocity focuses on the propagation of the entire wave packet. Only if ω ∼ k, group and
phase velocity are the same.
 Ca2 1/2 1
For dispersion relation Eq. (4.2), we find vg = cos ka, thus vg (± πa ) = 0, as expected
M 2
for a standing wave. (For more information and animations, see Group velocity and Phase
velocity Wiki pages).

4.1.2 Long wavelength limit

For ka  1 (long wavelength λ = 2π
k → ∞), we expand cos ka = 1 − 12 (ka)2 and find from the
dispersion relation (Eq. (4.2)):
C 
ω2 = k 2 a2 .
M
This results means that the frequency is directly proportional to the wavevector, which makes
group and phase velocities
q independent of frequency (also known from continuum theory of
q
a3
elastic waves (vp = M
C
a∼ E
ρ from c ∼ Ea and M ∼ ρ), where E is the elastic modulus and
ρ the atom mass density).
Chapter 4. Elastic and thermal properties of simple crystals 39

4.1.3 Extension to 2 atoms per primitive basis

In crystals with 2 or more atoms per primitive basis, the phonon dispersion relation displays
more complex features. This is a good model for ionic 2-compound crystals. In a given prop-
agation direction, the dispersion relation ω(k) for each polarization mode now exhibits two
branches: the acoustical branch and the optical branch. We refer also to Chapter 4, p.95 in [2]
and Chapter 4.3, p.87 in [1].

Figure 4.4: Optical and acoustical branches of the dispersion relation for a diatomic linear
lattice, showing the limiting frequencies at k = 0 and k = kmax = πa with the lattice constant a.
From p.95, [2].

The name "optical" branch arises from the fact that oppositely charged lattice points will create
dipole moments which can be excited by an electric field of a light wave, hence atoms of basis
vibrate against each other (see Fig. 10, p.98 in [2]). Generally, acoustic branches: ω → 0 for
k → 0, atoms vibrate in the same direction; optical branches: ω is finite for k → 0; almost no
dispersion. Also note the frequency gap, seen in Fig. 4.4. It implies that within this range of
frequencies, there are no modes or solutions that can propagate or exist → not all frequencies
lead to waves (typical in polyatomic lattices).

4.2 Thermal properties of phonons

The thermodynamic (thermal) properties of a solid are closely related to the phonon spectrum
(energy of harmonic oscillators).
 ∂U 
Recall the isochoric heat capacity CV = , where U – internal energy, V – volume, T –
∂T V
temperature. Then the internal energy of lattice due to phonons may be written as the sum
of the energies over all phonon modes. We use the now the quantum treatment to be accurate
also at low temperatures. Here, phonons are characterized by discrete frequencies and their
occupation states. The energy can thus be written:

Ulat = (4.3)
XX
hnk,λ i~ωk,λ ,
k λ
where k – wave vector (here simply for 1D), λ – polarization (transversal or longitudinal), nk,λ
– thermal occupancy of states, ~ωk,λ – energy of oscillating mode of frequency ω. The thermal
occupancy of bosons (since phonons have no spin) is given by the Bose-Einstein distribution:
Chapter 4. Elastic and thermal properties of simple crystals 40

1
hni = , (4.4)
exp{β~ω} − 1
1
where β = and kB – Boltzmann’s constant. Remember, for large T , the expression turns
kB T
into hni ∼ e−β~ω (the classical Boltzmann distribution).
From Eq. (4.3) and Eq. (4.4) it follows that the total energy of oscillators with frequencies ωk,λ
is:

~ωk,λ
Ulat =
XX
.
k λ
exp{β~ωk,λ } − 1
XZ
It is convenient to replace the summation over wavevector k by an integral as: dk ·
XX
→
k λ λ
XZ
c→ dωDλ (ω). Suppose that the crystal has Dλ (ω)dω modes of a given polarization λ in
λ
the frequency range ω to (ω + dω), then

XZ ~ω
ulat = dωDλ (ω) , (4.5)
λ
exp{β~ω} − 1

dn(ω)
where Dλ (ω) is the "density of modes" or "density of states" (DoS), i.e. D(ω) = , where
dω
n(ω) is the total number of modes of frequency ω and related to the dispersion relation ω(k)
or k(ω). Let x = β~ω, then the heat capacity contribution of the lattice is:

∂Ulat x2 e x
CV,lat = = kB dωDλ (ω) x
X
.
∂T λ
(e − 1)2
Central problem: what is D(ω)? The approach to the answer has historically 2 starting points
which we summarize in the following.

4.2.1 Einstein model

In 1907 Albert Einstein assumed that every atom is in a harmonic well created by the interaction
with its neighbors. Further, he assumed that every atom is in an identical harmonic well and has
an oscillation frequency ωE (known as the “Einstein” frequency). The assumption is essentially
a classical one (‘equipartitioning’) that every mode has the same average frequency (i.e., same
average energy). Within this model we have thus: lattice basis s = 1; every mode has one mean
frequency ωE ; N modes per polarization direction. The expression for the density of states we
can then write as:

D(ω) = N δ(ω − ωE ).
where δ(..) is the Dirac distribution. Using Eq. (4.5) we obtain the heat capacity as:

eβ~ωE
CV = 3N kB (β~ωE )2 .
(eβ~ωE − 1)2
Chapter 4. Elastic and thermal properties of simple crystals 41

Figure 4.5: Heat capacity of an Einstein solid as a function of temperature. Experimental value
of 3N kB is recovered at high temperatures (Dulong–Petit law).

Looking at the limit cases for temperature: at high temperatures T → ∞, CV = 3N kB – we

recover the correct Dulong–Petit law (experiment, 1819); while at low temperatures T → 0,
CV = e−β~ωE , however, experimentally we know that the heat capacity is proportional to ∼ T 3
(⇒ Einstein’s Ansatz is not good for T → 0, which makes sense because of his simplified classical
assumption of an average equipartitioning, instead of using a spectrum of energy distributions).

4.2.2 Debye model

The Debye ansatz considers the energy spectrum better. Here we ask, how many modes can be
found in a 3D cube of volume V = L3 ? How many solutions are there for periodic waves?

Figure 4.6: Illustration to the consideration of mode’s counting in 3D space.

We impose periodic boundary conditions on N 3 primitive cells (N cells = N atoms) within a

cube of side L, determining k based on the given condition:

eikα α = eikα (α+L) for α = x, y, z,

Then with sα = 0, 1, 2, 3... we get:
2π
kα = ± sα ,
L
Chapter 4. Elastic and thermal properties of simple crystals 42
q 2π
• k = |~k| = s2x + s2y + s2z ;
L
 2π 3
• one allowed value of ~k per volume in k-space.
L

Considering volume of a sphere of radius k, V = 34 πk 3 we can write the total number of modes
with wavevectors < k = |~k| (per polarization direction) as:
 L 3  4 
n(ω) = πk 3 .
2π 3

dn(ω) V k 2 dk
D(ω) = = . (4.6)
dω 2π 2 dω
The expression for density of states in Eq. (4.6) shows direct relation to dispersion relation
dk
through the term .
dω
In the Debye’s approximation, as in classic elastic continuum (or photons), we assume ω ≈ v · k,
linear dispersion relation with v – constant velocity of sound for each polarization type. (Recall
dk 1
phonon acoustic branch and = ).
dω v
Then the density of states becomes:

V k2 V ω2
D(ω) = = .
2π 2 v 2π 2 v 3
This motivates Debye’s model in the form:

V
D(ω) = ω 2 θ(ωD − ω),
2π 2v3
| {z }
∗

where θ(ωD − ω) is a step-function. Hence, ωD is a cut-off frequency, the so-called "Debye

frequency". (Note, the term with ∗ is analogous to the density of states in Planck’s radiation
law.) The Debye cut-off frequency imposed correctly that in a discrete crystal no very high
frequencies are possible (recall ωmax in dispersion relation). The value of the cut-off frequency
can be derived from the normalization condition:
Z ωD
V
N= dωD(ω) = ω3 ,
0 3 · 2π 2 v 3 D
then we can write:
 6π 2 v 3 N 1/3 1
ωD = ∼ inverse size of unit sell.
∼ ρ1/3 ∼ (4.7)
V a
(See dispersion relation acoustic branch ωmax ∼ v πa ). Then for the density of states we obtain:

3N 2
D(ω) = 3 ω θ(ωD − ω).
ωD
To find the expression for thermal energy we use Eq. (4.5) and the assumption of all 3 polar-
ization directions being isotropic:

3N ~ω 3
Z ωD
ulat = 3 · 3 dω ,
ωD 0 eβ~ω − 1
with substitution terms x = β~ω (x is dimensionless; x = 1 and kB T = ~ω) and xD = β~ωD , it
follows:
Chapter 4. Elastic and thermal properties of simple crystals 43

9N (kB T )4 x3
Z xD
ulat = dx .
(~ωD )3 0 ex − 1
Defining the Debye temperature in terms of ωD and rewriting the expression for the energy:

~ωD
θ= = xD T ; (4.8)
kB
 T  3 Z xD x3
ulat = 9N kB T dx . (4.9)
θ 0 ex − 1
Then lattice contribution to the heat capacity is then found through differentiating the Eq. (4.9)
with respect to temperature T as follows:
3 Z xD
∂ulat T x4 ex

CV,lat = = 9N kB T dx (4.10)
∂T θ 0 (ex − 1)2 .

Figure 4.7: Heat capacity of a solid, according to the Debye approximation. The horizontal
scale is the temperature normalized to the Debye temperature θ. The region of the T 3 law is
below 0.1θ. The asymptotic value at high values of T /θ is 24.943 J mol−1 K−1 . From Fig. 7,
p.113, [2].

The limits of this model are briefly:

 T 3 x 3
1. High T : x → 0 =⇒ CV = 9N kB D
= 3N kB .
θ 3
12π 4  T 3  T 3
2. Low T : x → ∞ =⇒ CV ' N kB ∼ ∼ T 3.
5 θ θ
Debye’s model works well for simple solids in the full temperature range (see experimental plots
Fig. 8 and 9, p.113-114 in [2]). For more information also consider Chapter 5 in [2] or [1].

4.3 Thermal properties of simple electronic systems

4.3.1 Thermal properties in metals
So far we have only considered the atoms in crystal as simple entities with an effective inter-
action; now let us consider electrons separately. We start with a very simple model of the free
Chapter 4. Elastic and thermal properties of simple crystals 44

electron gas which is good for ideal metals. Assumptions for metals are: positive N+ atoms
(small and negligible, i.e., serve as neutralizing background) and N = N− electrons (valence
electrons become conduction electrons).

Figure 4.8: Schematic model of a crystal of a sodium metal. The atomic cores are Na+ ions:
they are immersed in a sea of conduction electrons. In an alkali metal the atomic cores occupy
a relatively small part (∼15 %) of the total volume of the crystal, but in a noble metal (Cu, Ag,
Au) the atomic cores are relatively larger and may be in contact with each other (will become
relevant in the next section.

Recall the single electron in a 1D container:

Figure 4.9: Illustration for 1D problem of particle in a box. Dashed lines represent the energy
levels and solid lines the respective wavefunctions of a free electron confined in a box of length
L.

~ d2 ψn
In this case the Schrödinger equation takes form of Hψn = − = εn ψn , with ψn (0) =
2m dx2
2π
ψn (L) = 0 → ψn = A sin x. Then for the energy spectrum we obtain:
λn
~  nπ 2
εn = . (4.11)
2m L
N
Next, we fill the container with N electrons up to the "Fermi level" nF = (by fixing an
2
appropriate chemical potential µ). 2 comes from the fact that 2 electrons (fermions) can
1

occupy the same state. We define the Fermi energy via:

Chapter 4. Elastic and thermal properties of simple crystals 45

~  nF π 2 ~ Nπ 
εF = = .
2m L 2m 2L
Typical Fermi-energies in metals are on the order of 10 eV. Recall your basic course on ther-
modynamics and statistical physics: increasing temperature leads to the Fermi-Dirac (FD)
distribution, then the probability of occupation of state with energy ε is as following:
1
f (ε) = (4.12)
exp{(ε − µ)/kB T } + 1.

Figure 4.10: Fermi-Dirac distribution function at different temperatures: T0 = 0 K and T2 > T1 .

At the absolute zero temperature (T0 ), the probability of an electron to have an energy below
the Fermi energy EF (which is the same as µ at T = 0 K) is equal to 1, while the probability
to have higher energy is zero.

Extending the case to 3D with electron density ρ = N/V , we get:

~  3π 2 N 2/3
εF = ∼ ρ2/3 .
2m V
Then the density of states (DoS) is:

dN (ε) V  2m 3/2 √ 3 ε1/2

D(ε) = = 2 ε = N 3/2 .
dε 2π ~ 2 ε
F

In a more general way we can write: D(εF ) = 3N

2εF . With this result, the total energy of free
electron gas in 3D then becomes:
Z ∞
U= dεεD(ε).
0

dx
Z ∞ n−1 
 x
This expression can be also expressed through Fermi integrals ∼ (see Chapter 6
0 ex + 1
in [2]). Limits:
3
1. T → ∞ : U = N kB T and P V = N kB T – classical ideal gas.
2
π 2 kB
2T2 2
2. T → 0 : U = U0 + · N and P = ρεF 6= 0, where U0 = const. – kinetic energy at T = 0.
4 εF 5
Z εF
εF
The energy the takes form of U0 = dεεD(ε), with Fermi temperature: TF = and f (ε)
0 k B
is a step-function. For the final expression of electronic heat capacity we obtain:
Chapter 4. Elastic and thermal properties of simple crystals 46

CV 1 ∂U 2T
π 2 kB
= ' . (4.13)
N N ∂T 2 TF
Considering the contribution of both lattice and electronic heat capacity gives us a very different
CV
case than for classical ideal gas at T → 0: electrons contribute ∼ T , while phonons ∼ T 3 . The
N
electronic contribution can be neglected in many cases. However, we skipped some derivation
steps here for simplicity, refer to Chapter 6 in [2] and Chapter 6.4 in [1] for a more detailed
description.

4.3.2 Energy bands, states, and gaps in brief

While the free electron gas model of electrons in solid can give us insights into the origins of
thermal properties, conductivity and some properties of simple metals, like alkali metals, it fails
to explain the nature of distinction between metals, semiconductors and insulators as well as
their optical and electronic features. Within the so-called nearly free electron gas we will see
how the electronic states in solids form so-called bands and band-gaps. Firstly, we introduce
the periodic potential of the atomic cores (refer to schematic drawing in Fig. 4.8 in the previous
section).

Figure 4.11: Potential energy of a conduction electron in the field of the ion cores of a linear
lattice. The positions of the ion cores are indicated by the points with separation a (lattice
constant). From [2], p.166.

Within 1D lattice of unit a: V (x) = VG eiGx , where G – reciprocal lattice vectors, with
X

G
~ = (h 2π , 0, 0) = (G, 0, 0) and h = ..., −2, −1, 1, 2, .... V (x) is real valued, hence also the following
G a
form, which will be used later:

V (x) = VG e| iGx +{ze−iGx} .

X

G>0 2 cos Gx
Let us write the equation for wave function:
 1 
p̂2 + V̂ (x) ψ(x) = εψ(x). (4.14)
2m
2πn
We want to express ψ(x) in most general Fourier series. For that we use k ∼ , with L –
L
system size. Then:

ψ(x) = ck eikx =
X X
ψk .
k k
Combining expressions for kinetic energy and potential energy we can write the total energy as
follows:
Chapter 4. Elastic and thermal properties of simple crystals 47

p̂2 1  d 2 ~ 2 X 2 ikx
ψ(x) = −i~ ψ(x) = k ck e ,
2m 2m dx 2m k

VG eiGx ψ(x) =
X XX
VG ck ei(G+k)x ,
G G k

~ 2 k2
ck eikx + VG ck ei(G+k)x = ε
X XX X
ck eikx .
k
2m G k k

~ 2 k2
Set λk = and use ck . Then for all k we arrive at:
X
2m k

(λk − ε)ck + VG ck−G = 0. (4.15)

X

Eq. (4.15) is called "central equation" and it represents the algebraic form of wave equation
for coefficients ck in reciprocal
X space and couples ck to ck−G , ck−G0 , ck−G00 ... X
Then the wave
function must obey ψk (x) = ck−G ei(k−G)x
with energy εk and with uk (x) = ck−G e−iGx .
G G
Considering these, we can rewrite the wave function as a plane wave with coefficient uk (x) in
the form of:

ψk (x) = uk (x)eikx . (4.16)

Note that uk (x) is a Fourier transform over G, i.e. invariant under crystal lattice translation.
Physicist F. Bloch proved the important theorem that the solutions of the Schrödinger equation
for a periodic potential must be of a special form like the one in Eq. (4.16). Generally the "Bloch
theorem" states: "The eigenfunctions of the wave equation for a periodic potential use the
product of a plane wave times u~k (~r), having the periodicity of the crystal lattice.
The central equation (Eq. (4.15)) is not generally exactly solvable, but lets us consider the
Kronig-Penney model (KPM):

Figure 4.12: Square-well periodic potential as introduced by Kronig and Penney. From [2],
p.168.

Schrödinger equation can be solved conventionally without Bloch’s theorem making a spatial
separation Ansatz (see p.169 in [2]). It simplifies in the δ-limit of KPM: V0 → ∞ and b → 0.
Then:

V (x) = 2 VG cos Gx = Aa
X X
δ(x − sa),
G>0 s
Chapter 4. Elastic and thermal properties of simple crystals 48

with constant A and s ∈ Z = 0, ..., a1 , i.e., crystal length is unity (s = 1). Note that UG =
Z 1 XZ 1
dxV (x) cos Gx = Aa dxδ(x − sa) cos Gx + Aa a · s = A, meaning that
cos Gsa = A |{z}
X
0 s 0 s 1
all VG are equal for the δ-function potential.
Rewriting the central equation from Eq. (4.15):

( λk −εk )ck + A ck−Gh = 0,

X
|{z}
h
~ 2 k2
2m

2πh
with Gh = and ck−Gh – Miller (or "Bloch") indices in 1D. We want to solve Eq. (4.15)
X
a h
for εk . Let us define fk = ck−Gh , which has a periodic property fk = fk−Gh . Using this then:
X

h
− 2mA
~2
· fk
ck−Gh =  2 ,
2mεk
k − 2πh
a − ~ 2

~2 Xh 2πh 2 2mεk i−1

=− k− − 2 .
2mA h
a ~
1
Math handbook tells us, that cot(x) = , applying this and using trigonometric rela-
X

n nπ + x
tions we can obtain the equation:

mAa2
= cos ka, (4.17)
Ka
s~ 2 sinKa + cos Ka
√
2mεk
with K = . Equation above has solutions only for certain K (or εk ) values as presented
~
in the 4.13, (see also Fig. 5, p.170 in [2]).

Figure 4.13: Plot of energy ε vs. wavenumber k for the Kronig-Penney potential. Notice the
energy gaps at ka = π, 2π, 3π.... From [2].

In the plot εk or ε(k) "energy gaps" are occurring. They are called "electronic band gaps",
where no electronic states exist. Generally, they are computed by numerical methods using, e.g.
Chapter 4. Elastic and thermal properties of simple crystals 49

density functional theory (DFT), (see more in Chapter 7). Band structures allow to distinguish
insulators and conductors.

Figure 4.14: Occupied states and band structures giving (1) an insulator with a large band gap
> 1 eV, (2) a semiconductor, with a small band gap ∼ 1 eV)) and (3) a metal, with a band
gap below Fermi level and partially occupied conduction band. Also see Fig. 11 in [2], p.181,
showing the case for "semi-metal".
Chapter 5

Recap of thermodynamics and

statistical mechanics

Statistical mechanics and Thermodynamics are well introduced and summarized, for example,
in [4, 5].

5.1 Basic thermodynamic quantities

For convenience, here follows a brief summary of your thermodynamics lectures. Thermody-
namics is essential to characterize materials, because it shows us how state variables depend on
each other and how we can derive material properties from state functions or thermodynamic
potentials. In simulations one can often ‘measure’ some of the variables and thermodynamics
helps us to derive other important properties. Many variables that describes states ("state vari-
ables") exist, such as volume V or pressure P , which can be extensive or intensive, respectively.
Thermodynamics provides (often using differential forms) useful relations between these state
variables. In particular, derivatives of thermodynamic potentials with respect to their natural
variables deliver other important state variables. Thermodynamic potentials are linked with
Legendre Transforms. If you have forgotten about this, it would be useful to recall the basics
by looking into a standard thermodynamics book (e.g., Greiner et al. [4]).

number of particles

N 

volume

V




K kinetic energy



 extensive
U potential energy
∼ scale linearly with system size
internal energy (U+K)

E 


entropy

S




enthalpy (E+PV)

H 

density

ρ 

pressure

P  intensive
T temperature 
 independent of system size
chemical potential

µ 

50
Chapter 5. Recap of thermodynamics and statistical mechanics 51

Thermodynamic potentials natural variables standard differential forms

Internal energy dE = T dS − P dV + i µi dNi
P
E NV S
Helmholtz free energy F = E −TS dF = −SdT − P dV + i µi dNi
P
NV T
Gibbs free energy G = H −TS dG = −SdT + V dP + i µi dNi
P
NPT
Grand potential Ω = E − T S − µN = P V dΩ = −SdT − P dV − i Ni dµi
P
µV T

Table 5.1: Thermodynamic potentials and their natural variables related by the standard dif-
ferential forms obtained by Legendre transformation. The index i here goes over particle types
in a multi-component system. The particle number per type i is Ni .

Useful relations using the example of the Helmholtz free energy:

∂F ∂F ∂F
     
S=− P =− µi = (5.1)
∂T V,{Ni } ∂V T,{Ni } ∂Ni T,V,{Ni }

Maxwell relations follow from the second (mixed) derivatives, e.g.:

∂S ∂µi
   
=− (5.2)
∂Ni T,V,{Ni } ∂T V,{Ni }

The Gibbs-Duhem relation

SdT − V dP + Ni dµi = 0 (5.3)
X

is an useful relationship between intensive parameters, e.g., for fixed T and only a one-component
∂µ
system, i = 1, ∂P =ρ
These relations are particularly useful in computer simulations. For example, some variables,
such as free energy or entropy, are not easy to access directly. But we can access better, e.g.,
the pressure as a function of volume and then integrate it to get a free energy. Or, if one knows
how the free energy changes in the simulation with temperature, one can access the entropy by
the derivative S = −(∂F/∂T )N,V .
All material descriptors (even structure in density functional theory (DFT) framework) can
in principle also be derived from the free energy or related state variables. For example, the
isothermal compressibility χT
1 ∂V
 
χT = − (5.4)
V ∂P T
is the inverse of the bulk modulus K = χ−1 T and related to the linear elasticity V = χT P .
∆V

Generalization to anisotropic stress (tensors) and nonlinear elasticy are contents of the field
of ‘Continuum Mechanics’ and important in finite-element method (FEM) simulations in engi-
neering.

Figure 5.1: Deformation of a volume by ∆V under constant pressure P can be described by

the bulk modulus K = χ−1
T .
Chapter 5. Recap of thermodynamics and statistical mechanics 52

Another important example is the heat capacity:

∂E
 
cV = isochoric heat capacity (5.5)
∂T V
∂E
 
cP = isobaric heat capacity (5.6)
∂T P
from which the relation
α2
cP − cV = V T (5.7)
χT
 
with α being the thermal expansion coefficient, α = ∂V
∂T can be acquired [4].
P

5.2 Macro- vs. microstates

What we just discussed are the "macrostates" as thermodynamical states of a system, e.g., they
describe the average properties of a system with some natural variables, for example N , V and
T ("thermodynamical constraints") for the Helmholtz free energy F in a closed system.
In contrast, the "microstate" describes an instantaneous realization, e.g., a state with the kinetic
energy N i pi /2m for identical particles of mass m and momentum pi and the interaction
P 2

potential energy Hint = 2 i

1 PN PN
j V (|~
ri − r~j |) at a time t and given positions r~i of the particles.
Recall, we try to denote microstate operators with .̂., that is, describing the value of a microstate
and not an average. Note that in macroscopic thermodynamics, for reasons of notation brevity
explicit average symbols which we introduce in the following, h..i, are usually dropped. For
example, E = hÊi.
The connection between both macro- and micro-states is given by statistical mechanics. A
microstate is defined as a point {~rN (t), p~N (t)} in the 6N -dimensional phase space with:

~rN = {~
ri , ..., r~N } and p~N = {~
pi , ..., p~N } (5.8)

Given an instantaneous realization of the quantity X({~rN }; {~

pN }) we can define the time average
X of the phase space trajectories as:
1
Z τ
X= X(t)dt (5.9)
τ 0

whereas τ should be long enough to sample a representative region of the phase space. This is
what you learned in statistical mechanics in terms of ‘Ergodicity’. If a system is ergodic (which
is a good assumption for most systems as demonstrated by Liouville’s theorem in statistical
mechanics [4]), we do not have to follow the dynamic trajectory because we know the system
‘samples’ phase space in some representative fashion. For example, all states of the same fixed
energy are equally frequently visited in isolated systems (in stat. mech. the ‘microcanonical
ensemble’) which is the most fundamental theorem of statistical mechanics. In closed systems
(‘canonical ensembles’), phase space is sampled according to the Boltzmann weight.
Hence, for equilibrium properties we do not care about the path in time, just the probability in
being a microstate and corresponding representative sampling. We define a "statistical ensemble"
as a large number of possible realization of the system and its states the system can realize
with a probability w({~rN }; {~
pN }). The important ensemble average hXi is then given by:
 
hXi = tr X({~rN }; {~
pN })w({~rN }; {~
pN })
1
Z Z Z Z
with trace tr = dr~1 ... dr~N dp~1 ... dp~N (5.10)
h3N N !
whereas w is normalized by tr(w) = 1. Recall, the normalization prefactor 1/(h3N N !) originates
from two things: the phase space has a quantum nature and is discrete with a coarseness
Chapter 5. Recap of thermodynamics and statistical mechanics 53

given by Planck’s constant h. For N particles in three dimensions, thus the unit h3N . In
addition, in classical statistical mechanics (Boltzmann) we have to correct for the quantum
indistinguishability of particles and consider the Gibb’s factor 1/N ! to not overcount microstates
in phase space.
Depending on the thermodynamical natural variables we have different statistical ensembles.
Here are three important examples:

• Microcanonical ensemble (isolated system with E =const. and natural variables N V S)

1
w = δ(E − H) z = tr (δ(E − H))
z
S = kB ln z (Boltzmann entropy) (5.11)

• Canonical ensemble (closed system, N V T )

1 −βH  
w= e Z = tr e−βH
Z
Boltzmann weight and canonical partition sum
F (N, V, T ) = −kB ln Z (free energy) (5.12)

• Grand canonical ensemble (open system, µV T , particle exchange with a reservoir

governed by the chemical potential µ)

1 −β(H−µN )  
w= e Ξ = tr e−β(H−µN )
Ξ
Ω(µ, V, T ) = −kB ln Ξ (grand potential) (5.13)

In general, the system is always described by a Hamiltonian H = H({~rN }; {~

pN }), which de-
scribes the energy of a microstate. Quite generally, we can write
N
p2i
H= + Hint {~rN } + Hext {~rN }, (5.14)
X

i
2m

where the first term is the microstate kinetic energy K̂ = i p2i /(2m) (assuming all particles are
P

identical with mass m), an interaction energy Hint , depending only on positions, and an external
potential Hext , also depending only on the positions. The interaction potential is often assumed
to be a sum of pair potentials, that is, Hint = (1/2) i j V (rij ), where rij = |~ri −~rj | is the pair
P P

distance. Both, interactions and external potential, contribute to the total (instantaneous)
potential energy Û .
In a isolated system, represented by the microcanonical ensemble, the system energy is fixed to
H and conserved. The sum of the potentials Hint and Vext we can identify as operator Û , the
potential energy of a microstate, while Ê = H includes also the kinetic energy. In the canonical
ensemble, the state probability is weighted according to the Boltzmann weight ∝ exp(−βH) at
fixed (bath) temperature T [4].
As an example, we sketch the results for the ideal gas in the canonical ensemble in the absence
of an external field. Here, all interactions vanish in the Hamiltonian and only the kinetic term
remains. The partition sum (from eqs. (5.10), (5.12), and (5.14)) factorizes for the N particles
and can be easily integrated out to

1 VN
Zid = (5.15)
N ! Λ3N
Chapter 5. Recap of thermodynamics and statistical mechanics 54

The momenta in particular were integrated out, as

3N
h
Z Z 
3N
N −β K̂
d~
p e = d~
p wp = (2πmkB T )
N 2 = (5.16)
Λ

defining the thermal (de Broglie) wavelength Λ

s
h2
Λ= . (5.17)
2πmkB T

In between we identify the Maxwell-Boltzmann velocity distribution in equilibrium for the

absolute velocity v = p/m:
3/2 2
m

− 2kv
p(v) = 4π v2e BT , (5.18)
2πkB T

that is, p(v)dv telling us the probability of finding a particle with absolute velocity v in the
intervall dv. Similar distributions can be written for momenta, cartesian components of velocity,
or the kinetic energy [4].
The free energy of the ideal gas follows as F = −kB T ln Zid . This can be approximated for large
N  1 using the Stirling approximation ln N ! ' N ln N − N leading to
 
Fid = N kB T [ln ρΛ3 − 1]. (5.19)

Using known thermodynamic relations, such as S = −∂F/∂T and E = F + T S, the ideal gas
entropy and energy follows easily [4]. For the ideal gas, E = K = hK̂i, that is, we have only
kinetic energy that contributes to the total energy.

5.3 Ergodicity and averages

Ergodic hypothesis:
The time average X equals the ensemble average hXi for long times τ :

1
Z τ
X = X(t)dt ≡ hXi (5.20)
τ→
−∞ τ 0

Ergodicity is challenging to achieve in most complex systems! In particular, in computationally

expensive simulations, e.g., with many millions of atoms, often the system does not move out of
(meta)stable minima in the energy landscape. Then only a sub-space of phase space is sampled.
Typically this is defined via X = hXilocal , sampled from a "local equilibrium" (= local region
in phase space). Hence, one has to be careful with the interpretation of averages. One should
also critically think about systematic and statistical errors, see later in the simulation part.
In molecular dynamics simulations the application of the average is typically done in the canon-
ical ensemble, where N V T is fixed. Typically, mean values and standard deviations are calcu-
lated, but more generally whole probability distributions are interesting to look at. Simple and
typical examples for average mean values are the potential and kinetic energy in the canonical
ensemble, via
1
Z Z
U = hÛ i = d~rN Û e−β Û with ZU = d~rN e−β Û
ZU
1
Z Z
K = hK̂i = d~rN K̂e−β K̂ with ZK = d~rN e−β K̂ (5.21)
ZK
Chapter 5. Recap of thermodynamics and statistical mechanics 55

where in equilibrium statistical equipartitioning dictates that hK̂i = 23 N kB T . In simulations it

is important to watch these averages of Ê = Û + K̂ and K̂ because they tell us about behavior
(conservation and drift) of total energy and temperature, respectively. We will reconsider the
latter when we talk, for example, about thermostatting a simulation. Other quantities are of
course of interest as well in simulations. Since all trajectories (particle positions and velocities
at time t) are in principle available, most observables can be directly averaged. In the next
sections we will talk about averages of fluctuations but also about averages of structures, which
have important physical meanings as well.
Note again, the entropy S and the free energy F = U −T S are not directly accessible by averages,
but from probability distributions (F = −kB T ln P ) or T -dependencies (S = − ∂F ∂T ).
Chapter 6

Structure-property relationships in
simple liquids

Liquid state theory is the overarching statistical mechanics framework to describe correlated,
interacting systems, such as liquids. It is well discussed in [5].

6.1 From fluctuations to material properties

Materials properties can also be derived from the fluctuations of physical observables. Let us
first recall what is a statistical variance and standard deviation (SD) of a variable x:

Ensemble mean hxi

Variance h(hxi − x) i = hx i − hxi = σx2
2 2 2
q
Standard deviation σx = σx2 (6.1)

When we talk about ‘fluctuations’ we mean exactly this, a variance or SD of an observable.

6.1.1 Heat capacity and compressibility

How are fluctuations related to macroscopic properties? Consider, for example, the definition
of the average total energy E = hÊi in the canonical ensemble:
1
 
E = tr Êe−β Ê (6.2)
Z(T )
Thermodynamics tell us that the heat capacity is then obtained by the derivative of E with
respect to T , at fixed volume:

1 1 ∂Z(T )
!
∂E Ê 2 −β Ê
  
= tr e + tr Êe−β Ê
∂T V Z(T ) kB T 2 Z(T )2 ∂T
Ê −β Ê ∂ ln Z
!
hÊ 2 i

= 2
+ tr e
kB T kB T 2 ∂β
1  
= hÊ 2 i − hÊi2
kB T 2
σ2
= E 2 = cV (6.3)
kB T
The heat capacity is thus given by the energy fluctuations
√ in the canonical ensemble! It also
scales with the size of the system ∼ N , as σE ∼ N .

56
Chapter 6. Structure-property relationships in simple liquids 57
√
Relative Gaussian fluctuations σEE ∼ NN = √1N are thus very small for large systems. That is
why we can talk about ‘sharp’ macroscopic state variables in thermodynamics.
Similarly one can show in the N P T -ensemble, that the compressibility χT is given by the
volume fluctuations in this ensemble:

V kB T χT = σV2 (6.4)

whereas in an µV T -ensemble, it is given by the number fluctuations:

N2
kB T χT = σN
2
(6.5)
V
Note again, standard deviation describes physical fluctuations of the system, not the error!
Many other examples exist. Calculating mean AND fluctuations in theory and simulations is
thus a very important way to connect to macroscopic observables.

6.1.2 Time correlation function

Fluctuations have a life time, i.e., it takes some time before a perturbation in the system relaxes
away. These times are interesting for physical reasons (they tell us how fast a perturbation is
‘forgotten’) but also to estimate errors in our system, which can be particularly large if the
simulation time is comparable to the system’s fluctuation or ‘memory’ time scales.
To reveal in which way and how fast microscopic fluctuations decay, we define a time correlation
function between two observables A and B:

CAB (t, t0 ) = hA(t), B(t0 )i

CAB (τ ) = hA(τ ), B(0)i
1
Z T
= lim A(t + τ )B(t)dt (6.6)
T →∞ T 0

In practice, we average over all initial times in order to increase statistics. In other words, every
time of the simulation can be taken as t = 0 from which the ACF can be defined as origin. This
of course only makes sense for stationary systems, where observables depend only on t − t0 and
not on absolute time t.
More general the correlation is given in reference to the mean:

CAB (t) = hA(t) − hAiihB(0) − hBii (6.7)

For convenience, it is often normalized to their means, i.e., C̃AB (τ ) = CAB (τ )/(hAihBi)
The auto correlation function (ACF ) is defined as CAA (τ ).
Generally, time correlation functions are important to calculate macroscopic transport proper-
ties [4, 5]. One of the most important example for a correlation function is the velocity auto
correlation function (VACF):
Cvv (t) = h~v (t)~v (0)i (6.8)
with h~v i = 0 and h~v 2 i = 3kB T
m in equilibrium.
The normalized auto correlation function is then given by:
h~v (t)~v (0)i 3kB T
C̃vv (t) = = h~v (t)~v (0)i. (6.9)
h~v 2 i m
Time correlation functions are important for various reasons. Apart from telling us about how
fast and in what manner fluctuations and correlations decay in time, they are also helpful to
estimate statistical errors of time series. Integration over correlation functions also yield useful
macroscopic dynamic features such as diffusion and friction, and also provide definition of linear
response functions (in presence of perturbative fields).
Chapter 6. Structure-property relationships in simple liquids 58

6.1.3 Einstein diffusion law (random walk)

The Einstein diffusion law is defined in three dimensions (3D) as
1 2
D = lim h~r i (6.10)
t→∞ 6t
with h~r2 i being the mean square displacement (MSD) of the particles as seen in Fig. 6.1. More
generally, we can define the MSD for a particle i as h(~ri (t) − r~i (0))2 i with initial position ri (0)
and also average this over all particles, and over all initial times, to increases the statistics
in evaluating the simulations. Note that the random walk is independent in every cartesian
direction and we can also evaluate diffusion in every space directions, e.g., Dx = limt→∞ 2t 1
hx2 i,
with a prefactor 2 in every cartesian direction.

Figure 6.1: The MSD of the particles is given by averaging h(~r(t) −~r(0))2 i over all ~r. Depending
on time t, a simple system, such as a free underdamped Brownian (or free ‘Langevin’) parti-
cle [6], can either behave ballistic (short-time) or diffusive (long-time). It typically converges to
the long-time self-diffusion constant D for long times. For complex and dense systems, however,
there can be many different dynamic regimes.

One can show that the MSD can be expressed by the VACF through
Z t
h~r i = 2t
2
h~v (t0 )~v (0)idt0 (6.11)
0

which leads to the so called "Green-Kubo" relation [5]:

1
Z t Z t
kB T 0 0
D = lim C̃vv (t )dt = lim h~v (t0 )~v (0)idt0 (6.12)
t→∞ m 0 t→∞ 3 0

with D being the long-time self diffusion constant. (D can not be defined this way for ideal,
non-interacting systems as they behave fully ballistic. This definition is also problematic for
solids and glasses, where diffusion is not possible on observation time scales, only local positional
fluctuations.)
One can also show that the ACF of the force acting on a particle i leads to the friction constant
1
Z ∞
ξi = hF~i (t0 )F~i (0)idt0 (6.13)
3kB T m 0

with the forces F~i = − j ∇V (|~

ri − r~j |) as seen in Fig. 6.2. Through the fluctuations-dissipation
P

theorem (FDT) [6], the friction constant is related to the noise in Brownian motion.
Chapter 6. Structure-property relationships in simple liquids 59

Figure 6.2: The force ACF is calculated using the forces derived from the pair potential F~i =
− j ∇V (|~ ri − r~j |)
P

In particular, in the simplest form of the FDT for memoryless (Markovian) stochastic processes,
the diffusion D and the friction constant ξ are related by the Einstein relation (Einstein 1905!)

kB T
D= (6.14)
ξm

that is, system fluctuations (represented by the diffusion constant) are directly proportional to
temperature, and also directly related (inversely proportional) to the dissipation in the system
quantifed by friction. The latter is a response to external perturbations and not a priori
connected to fluctuations. However, the FDT and other historically important developments
(Einstein, and, for example, Onsager’s regresssion hypothesis [6]) have suggested and finally
proven this fundamental connection.
Note:

• Similar Green-Kubo relations exist for bulk viscosity, shear viscosity and thermal conduc-
tivity (see Appendix)

• The short-time behaviour is linked to static averages and fluctuations, e.g.

3kB T h|F~i |2 i 2
!
h~v (t)~v (0)i ' 1− t + ... , (6.15)
m 6mkB T

definining the so-called Einstein frequency in solids [5].

• Again, fluctuations connect to macro-observables, such as transport by diffusion or fric-

tion!

6.2 From the structure to material properties in the canonical

ensemble
6.2.1 The one-body density profile ρ(~r)
Let us first define and discuss how we can describe the microscopic structure of a many-body
system. Then we can try to relate it to a macroscopic material property. First, we consider the
"one particle density operator":
N
ρ̂0 = ri − ~r) (6.16)
X
δ(~
i=1
Chapter 6. Structure-property relationships in simple liquids 60

with r~i denoting the position of particle i and δ being the Dirac-function, defined as:
Z
δ(x)dx = 1
Z
δ(x − x0 )f (x0 )dx = f (x). (6.17)

The ensemble average of the so-called "one particle density operator" leads to the equilibrium
one-body density distribution:
N
ρ0 (~r) = h ri − ~r)i ≡ ρ(1) (~r) ≡ ρ(~r) (6.18)
X
δ(~
i=1

Note that without any external potential Vext = 0, the density distribution is isotropic ρ(~r +~a) =
ρ(~r) and constant ρ = NV . With an applied external potential though, the density becomes
inhomogeneous, cf. Fig. 6.3A.
In a typical model or simulation setup, often one considers inhomogeneities in one dimensions,
such as a fluid confined between two walls. Then the profiles can be obtained statistically by
averaging the number of particles in small bins (with meaningfully chosen size), cf. Fig. 6.3B.
By dividing by bin size and lateral box area A, one gets the correct units of number density
(number per volume).

Figure 6.3: A: The density distribution ρ(z) of a fluid in an uniform external force F . Due
to packing effects one has a non-trivial layering and inhomogeneities. (This can be compared
to an ideal gas in the same field, where the profile is given by the ‘barometric height law’
ρid (z) = ρ0 e−βF z .) B: The binning representation with bin size ∆z. The density distribution
hni
is obtained in a simulation by averaging the number of particles per bin hni(z) → ρ(z) = A∆z ,
and normalizing by bin volume with bin size ∆z and lateral box (wall) area A.

For an ideal gas in an external potential, one can show ρ(z) ∝ exp(−βVext (z)) (see further
below). For Vext (z) = mgz, with mass m and acceleration g this would be the barometric
height law.
Explicitly, for the ideal gas in the grand-canonical ensemble, the equilibrium one-body density
distribution is obtained by performing explicitly the ensemble average of the "one particle
density operator" given by Eq. (6.16). One obtains:
∞
) N
1 X 1 zN
Z Z (
ρ0 (~r) = dr13 ... drN exp −β Vext (~ri ) + Hint δ(~r − ~ri ) (6.19)
X X
3
Ξ N =1 λ3N N ! i=1 i

Considering an ideal gas, Hint = 0, in the external field Vext (~(r), Eq. (6.19) reduces to:
Chapter 6. Structure-property relationships in simple liquids 61

Z Z ( )
1 P∞ zN
ρ0 (~r) = dr23 ... drN exp −β Vext (~ri ) exp{−βVext (~ri )}
X
N 3
Ξ N =1 λ3N (N −1)! N
i=2
| {z }
Z N −1
dr3 exp{−βVext (r)} (6.20)
| {z }
Ṽ N −1

hN i
= Ṽ
exp{−βVext (~r}

which is the generalized barometric height law, with normalization

Z
→ dr3 ρ0 (~r) = hN i.

6.2.2 The radial distribution function, g(r)

The density operator can be generalized for the two-particle density:
N
ρ(2) (~r,~r0 ) = h rj − ~r0 )i
ri − ~r)δ(~ (6.21)
X
δ(~
i6=j

This method works for ρ(n) analogously. For ρ(2) , the pair distribution function (PDF) can be
derived via a normalization:
ρ(2) (~r,~r0 )
g(~r,~r0 ) = (6.22)
ρ0 (~r)ρ0 (~r0 )
If no external potential is applied, Vext = 0, then the PDF g(~r,~r0 ) is isotropic and it follows:

g(~r,~r0 ) ≡ g(|~r − ~r0 |) ≡ g(r)

N
1 X
= h δ(~r − (~
ri − r~j )i with ρ = const. (6.23)
ρN i6=j

where g(~r,~r0 ) only depends on the distance rij , or simply r. This defines the "radial distribution
function" (RDF), g(r). The latter is the most important function that describes the structure of
a classical many-body system. We can interpret it as the probability to find a second particle at
distance r if the first one is fixed at the origin. In another view, it describes the one-body density
profile ρ(r) of particles around a particle of same type, i.e., if the latter would hypothetically
act as a fixed external potential to the others. It can also easily generalized to multi-component
systems, then gij (r) is defined between particles of type i and j [5]. Typical examples for liquids
and solids are shown in the Appendix.
In simulations, it is calculated by creating a histogram around particles in shells of size ∆r (see
Fig. 6.4):
hni 1
g(r) = (6.24)
∆V ρ
Chapter 6. Structure-property relationships in simple liquids 62

Figure 6.4: Division in shells of size δr that cover a volume of ∆V = 4πr2 ∆r each used to
calculate the RDF in simulation.

Important features:

• g(r) = 1 for an ideal gas, as there exist no correlations.

• g(r → ∞) = 1 in liquids, as correlations do not occur far away.

• g(r) = e−βV (r) is the low density limit ρ → 0 for the pair potential V (r).

• ρ0 (r) = ρg(r) is the one-particle density distribution for one particle of the same type.
We introduce the (running) "coordination number" nc (r):
Z r Z r
0 3 0
nc (r) = ρ0 (r )d r = ρ4π dr0 g(r0 )r02 (6.25)
0 0

When integrated to r = r1 being the position of the first minimum of g(r), we get the
number of particles in the "first solvation shell" (see Fig. 6.5):
Z r1
n1 ≡ n(r1 ) = ρ4π dr0 g(r0 )r02 (6.26)
0

• g(rmax ) defines the first peak of the RDF and can be related with an association (binding)
affinity ∆G = −kB T ln (g(rmax )).

(a) (b)

Figure 6.5: (a) Pair distribution function, g(r), and (b) corresponding coordination number.

For the ideal gas, the radial distribution function can be exactly derived by explicitly calculating
the ensemble average of the two-particles density operator in the grand-canonical ensemble.
Assuming a homogeneous ideal gas, one obtains:
Chapter 6. Structure-property relationships in simple liquids 63

ρ(2)
g(r) = ρ2
1 P∞ ZN 1
= N =2 Λ3N Ξ (N −2)! dr3 ...
1
drN
R 3 R 3
ρ2 (6.27)
= z2
ρ2
with ρ = hN i
V =z
= 1
An analogous calculation of only two particles (infinite dilution) interacting with V (r) yields
the important limit g(r) = e−βV (r) .
The radial distribution function g(r) is a key quantity! It can be experimentally measured,
e.g., by a diffraction (indirect, reciprocal k-space) or confocal microscopy (direct, real-space).
A key function to compare to k-space measurements is the static structure factor, which will
be introduced below. Additionally, the g(r) provides structure-property relationships as also
demonstrated in the following.

6.2.3 From g(r) to macroscopic observables

Potential energy: One can easily show that the mean potential energy of the system can be
obtained through the knowledge of the g(r) and the pair interaction potential V (r). Consider
the average potential energy of a ‘bulk’ system, that is, in the absence of any external field:
1X
hÛ i = h V (|~
ri − r~j |)i
2 i6=j
1X
Z
=h d3 rδ(~r − (~
ri − r~j ))V (r)i
2 i6=j
ρN
Z
= d3 rg(r)V (r)
2
Z ∞
hÛ i
→ = 2πρ drr2 g(r)V (r) (6.28)
N 0

This equation provides the mean potential energy per particle which is macroscopic observable.
Hence, we have here a simple realization of a ‘quantitative structure-property relationship’
(QSPR), relating micro-structure to a macro-property, by a simple integration.
A similar expression can be derived for the pressure in the absence of external fields (or osmotic
pressure):
2
Z ∞
P = ρkB T − πρ2 drr3 g(r)V 0 (r) (6.29)
3 0
This equals to the virial equation of the pressure expressed by g(r), where

dV
V 0 (r) = (6.30)
dr
is the pair force. In the low density limit, the virial equation is
Z ∞
P = ρkB T − 2πρ2 dr(e−βV (r) − 1) = ρkB T + B2 ρ2 (6.31)
0

with B2 being the second virial coefficient. Eq. (6.31) represents the virial expansion of the
thermal equation of state (EOS) up to second order in ρ. The function exp(−βV (r)) − 1 is also
called the Mayer-f function [5].
Pressure: In classical mechanics, the mean kinetic energy hKi via the Clausius virial theorem
can be described as

hK̂i = − 21 hF~itot~ri i = − 12 hF~iint~ri i − 12 hF~iext~ri i = − 12 θint − 12 θext (6.32)

X X X

i i i
Chapter 6. Structure-property relationships in simple liquids 64

with θint and θext defining the internal and external interactions. The external interactions are
imposed on the system through a pressure P0 as can be seen in Fig. 6.6. This results in

dF~iext = −P0~ndA
Z
F~iext~ri = −P0 ~n~rdA (6.33)
X

where ~n is the normal vector in respect to the surface.

Figure 6.6: Applied external interactions on the volume V by the pressure P0 .

If we apply the Gauss theorem to the above statement

Z Z
~n~rdA = ∇~rdV = 3V, (6.34)

the external interactions are given by

hθext i = −3P0 V (6.35)

This derivation is very general and independent of the box shape!

The kinetic energy can also be defined via the equipartitioning theorem as

hKi = N 12 mhv 2 i = 32 N kB T (6.36)

and since the pressure is everywhere the same: P = hP i, we can derive the virial equation using
Eq. (6.35):
N
N 1 X
P = kB T − ~ri ∇i U (6.37)
V 3V i=1
Finally, one can show that:

2
Z ∞
P = ρkB T − πρ2 drr3 g(r) V 0 (r) (6.38)
| {z } 3 0 | {z }
ideal gas −dV /dr

This equation provides osmotic pressure which is macroscopic observable. Hence, we have here
again a QSPR, relating micro-structure to a macro-property, by an integration of the radial
distribution function.
Chapter 6. Structure-property relationships in simple liquids 65

A final example is the relation between g(r) and the system’s thermal compressibility
1 ∂V
 
χT = − . (6.39)
V ∂P T
One can proof that the compressibility χT can be expressed in the grand canonical ensemble
through the RDF as:
hN 2 i − hN i2
Z
ρkB T χT = = 1+ρ d3 r(g(r) − 1)
hN i
Z
= 1+ρ d3 rh(r) (6.40)
| {z }
Nexc

with h(r) = g(r) − 1. The Nexc can be called the excess adsorption around a particle with
respect to the ideal gas.

6.2.4 Relation of g(r) to scattering

The scattering structure factor is an experimentally important structure function. It is a result
of neutron or X-ray scattering at materials [5, 6]. It is formally defined as:
1
 
S(~k) = δ~ δ ~ (6.41)
N k −k
via the average of the density fluctuations with the wave vector ~k which is defined by the Fourier
transformation of the density:
Z Z
~ ~
δ~k = d3 rρˆ0 e−ik~r = d3 r δ(~r − r~j )e−ik~r (6.42)
X

It can be shown that the structure factor S(~k) also be rewritten in forms of the RDF [5]:
Z
~
S(~k) = 1 + ρ d3 r (g(r) − 1) e−k~r + (2π)3 ρδ(~k) (6.43)
| {z } | {z }
h(r) unscattered

We see that the experimentally accessible structure factors is, in principle, only the Fourier
Transform of the RDF! Hence, all oscillations and prominent wavelengths appear as peaks in
the structure factor. The information is not different from that in the RDF but presented in
reciprocal space from a different perspective. In particular, long wavelength effects such as
clustering can be better seen in the structure factor.

(A) (B)

Figure 6.7: (A) Scattering experiments can give insight to the structure of the macromolecular
liquid. Depending on the lengthscale probed, we can access the internal structure of the particles
(defined by Vscatt ) which defines the form factor, or we can access the liquid structure in terms of
the RDF for larger scales. (B) Representative structure factor, S(~k), in a homogeneous liquid.
Chapter 6. Structure-property relationships in simple liquids 66

However, in experiments only the intensity of the scattered wave can be measured which also
includes information on the internal structure of a (macro)molecule. In general, we have for
the intensity
I(~k) = N |F (~k)|2 S(~k) (6.44)
with F (~k) being the ‘form factor’
m
Z
~
F (~k) = − d3 rVscatt (~r)e−ik~r (6.45)
2π}2
defined by the internal structure of the particles expressed by an internal scattering potential
Vscatt . The form factor can be measured at larger wavevectors in diluted systems, where S(k) '
1. An important feature is also that
Z
~
S(~k → 0) = 1 + ρ d3 r(g(r) − 1)e−k~r = ρkB T χT , (6.46)

In words, the small wavevector behavior of the structure factor gives us direct information
about the compressibility of the system. Close to a phase transition and/or critical behavior,
S(~k → 0) is diverging due to long wavelength fluctuations in the system [5, 6].
Numerically, a Fourier Transform can be solved by given routines as in Python or Numeri-
cal recipes. Note that for a radially symmetric function, the FT simplifies into a quasi-1D
transform. One can show that the 3D trafo of f (r)
Z
f˜(~k) =
~
d3 rf (r)e−k~r (6.47)

can be simplified into a 1D-sinus Trafo of f (r)r of the form:

4π
Z
f˜(k) = dr[f (r)r] sin(kr) (6.48)
k
Chapter 7

Classical DFT (Density Functional

Theory)

Classical density functional theory (DFT) is a theoretical framework used to study the proper-
ties of classical many-particle systems. It is based on the idea that the equilibrium properties
of a system can be described by the minimization of a free energy functional that depends on
the particle density.
In this chapter, the mathematical background of functional calculus is introduced, as well as
functional derivatives. We will also discuss variational principle, which is used to derive the
free energy functional.
Several approximations used in classical DFT are presented, including the local density ap-
proximation (LDA), which assumes that the free energy density is a function only of the local
particle density, and the weighted density approximation (WDA), which includes non-local
effects by using a weighted average of the density. The Orsntein-Zernike equation, which de-
scribes the correlation function between particles, is also introduced. The chapter also covers
the use of thermodynamic integration to calculate the excess free energy in cases where no exact
expressions for correlation functions are available.
We will also learn basics of the theory of correlation functions, which is crucial to DFT, as
it provides a means of calculating the excess free energy of a system in terms of the density
distributions. Density distributions describe the probability of finding a particle in a particu-
lar region of space, while correlation functions describe the degree of correlation between the
positions of two particles in a system.

7.1 Mathematical background, functional calculus

Consider functional F as a functional of f (~x) ∈ L2 – space of square integral functions. Note fol-
lowing notations, which are used throughout this chapter: F ≡ F[f (~x)] ≡ F[f ], δ = b variation of F
(see below Eq. (7.1))
δF
Functional derivative is the generalization of the multi-dimensional derivative of f (x1 ,..,xN ).
δf
∂f
Analogously to classic derivatives: = Ai (x) with total differential: df = f (~x + d~x) − f (~x) =
∂xi
N
Ai (~x)dx~i and ~x = x1 , ..., xN . in functional space transforms into:
X

i=1
δF
= A[f ;~x].
δf (~x)
A[f ;~x] – functional of f and function ~x = x1 , ..., xN , with:

67
Chapter 7. Classical DFT (Density Functional Theory) 68

Z
δF = F[f + δf ] − F[f ] = dxN A[f ;~x]δf (~x). (7.1)

Below is a brief list of calculation rules, used in functional calculus and some specific examples.

1. Identity.
∂f
As we know, if = 1 → f (xi ) = xi . In functional calculus it is similar:
∂xi
δF
Z
= δ(x − x0 ) ⇒ F[f ] = f (x0 ) = dxδ(x − x0 )f (x).
δf (x0 )

Here, Dirac δ-function and in L2 -space δ(x − x0 ) = 1. One can prove this expression by using
definition from Eq. (7.1):
Z
δF = δ(x − x0 )δf (x)dx = δf (x0 );

δF δf (x0 )
= = δ(x − x0 ). (7.2)
δf (x) δf (x)

δF(~r0 )
In 3D for a functional F[f (~r)] this becomes = δ(~r − ~r0 ).
δf (~r)
2. Linear functional.
N
In classical calculus we can use definition of a linear function f (~x) = ai xi and its differential
X

i=1
∂f
df = ai dxi , then its partial derivative equals = ai with ai = const. Functional analogue
X

i=1
∂xi
with a(x) – some fixed function then:
Z
F[f ] = a(x0 )f (x0 )dx0 ,
Z
δF = a(x0 )δf (x0 )dx0 .

Using expressions above and Eq. (7.2) we get:

δF δf (x0 ) 0
Z Z
= a(x0 ) dx = a(x0 )δ(x − x0 )dx0 = a(x). (7.3)
δf (x) δf (x)

δF
Z
In 3D for F[f ] = d3 r0 f (~r0 )a(~r0 ) functional derivative is = ~a(~r0 ).
δf (~r)
δF
by Prof.: Note: often = ~a(~r0 ) but then ~r0 → ~r again.
δf (~r)
3. Chain rules.
Consider F ≡ F[g(f (x))], where g(f ) is an arbitrary function. Then:

δF[g] δF dg(x0 )
Z
= δf (x0 )dx0 , (7.4)
δf (x) δg(x0 ) df (x)

dg
with δg ∼ δf . We can consider following example:
df
Chapter 7. Classical DFT (Density Functional Theory) 69
Z Z
0 0 0
Example: F[f ] = dx a(x )g(f (x )) = dx0 a(x0 )f 2 (x0 ), with g(f ) = f 2 .
δF
Combining Eq. (7.1), a(x0 ) = , variation of the functional is:
δg(x0 )
Z Z
δF = dx0 a(x0 )δg = dx0 a(x0 )2f (x0 )δf (x0 ),

δg dg
with = being classic derivatives (where we can also apply product rule, like g = f ·h(f )
δf (x ) df
0
and dg = df · h + f · dh. Then for the functional derivative:

δF δf (x0 )
Z
= dx0 a(x0 )2f (x0 ) = 2a(x)f (x).
δf (x) δf (x)
| {z }
δ(x − x0 )

4. Mixed variables.
Z Z
Typically in statistical mechanics the functionals take form of F[f ] = dx1 ·...· dxN ·a(x1 , .., xN )·
Z Z
f (x1 ) · ... · f (xN ) or F[f ] = dr13 · ... · drN
3
· a(~r1 , ..,~rN ) · f (~r1 ) · ... · f (~rN ) (linear functional),
where i = 1, .., N label particle positions.
In classical statistical mechanics the particles are all identical, hence:

δF
Z
=N· dx2 · .. · dxN a(x1 , ...xN )f (x2 )...f (xN ) = H[f ; x1 ]. (7.5)
δf (x1 )

5. Higher derivatives.
δ (n) F
For the n-th dimensional derivative we basically apply the above steps n times.
δf (~r1 )...δf (~rN )
n-th order functional derivatives can be used to perform a functional Taylor-expansion around
f0 :

δF 1 δ2F
Z ZZ
F[f ] = F[f0 ]+ (f (x)−f0 (x))dx+ [f (x)−f0 (x)][f (x0 )−f0 (x0 )]dxdx0 +...
δf 2! δf (x)δf (x0 )
f =f0 f =f0
(7.6)

7.2 Functionals in statistical mechanics: fluids in external fields

While liquids are uniform in their bulk, non-uniformities can occur near walls or other physical
boundaries, as well as at the points of coexistence of different phases. The grand canonical
ensemble provides a general approach to studying these inhomogeneous fluids. The approach
involves using a Hamiltonian that includes a term representing the interaction of the particles
with an external field that varies in space (see Eq. (7.8) below). This term breaks the transla-
tional symmetry of the system, but by taking the limit where the external field vanishes, results
for uniform fluids can be obtained. A critical component of this theory is a variational principle
for the grand potential. Further reading and more detailed derivation can be found in Chapter
3 in [5].
If we look at the case of homogeneous fluid, then density distribution ρ(~r) = ρ = const. and
external potential Vext (~r) = 0.
Chapter 7. Classical DFT (Density Functional Theory) 70

In case of the grand canonical ensemble (GCE), ρ is uniquely determined by chemical potential
µ. However, in case Vext (~r) 6= 0, ρ(~r) is inhomogeneous. Then one can proof that ρ(~r) is uniquely
determined by the "local chemical potential":

ψ(~r) = µ − Vext (~r), (7.7)

with δψ = −δVext .
N
DX E DZ X E
Note that for Hamiltonian expression: hHext i = Vext (~ri ) = δ(~r − ~ri )Vext (~ri )d3 r =
Z i=1 i=1
d rρ(~r)Vext (~r) and the final expression is:
3

Z
hHext i = d3 rρ(1) (~r)Vext (~r). (7.8)

Now considering grand-canonical (GC) sum in functional framework:

∞ N N
1
Z Z Y  Y 
∗
Ξ = Ξ[µ, V, T ] = dr13 ... drN Z (~rk ) × e(~ri ,~rj ) , (7.9)
X
3

N =0
N! k=1 i<j

eβµ −βVext (~r) 0

with local activity Z ∗ (~r) =e and exponential of interactions: e(~ri ,~rj ) = e−βV (|~r−~r |) .
Λ3
Therefore we consider Ξ a functional of Z ∗ , meaning Ξ = Ξ[Z ∗ ]. The same applies to the grand
potential Ω = Ω[Vext ]).
One can show using functional derivatives:

δ(ln Ξ) δΩ
ρ(~r) = =− . (7.10)
δ ln Z ∗ δVext (~r)

Thus, we arrive to the alternative definition of ρ(~r), defining it as a response of grand potential
Ω[Vext ] to changes of external potential.
Analogously for the n-particle density profile we can define:

Z ∗ (~r1 )...Z ∗ (~rN ) δ (n) Ξ

ρ(n) (~r1 , ...,~rn ) = . (7.11)
Ξ δZ ∗ (~r1 )...δZ ∗ (~rN )
Fluctuations around the local density ρ(~r) can be described by the density correlation function
in the general form described as:
*  X  +
(~r1 , ...,~rn ) = δ(~ri − ~r1 ) −ρ(~r1 ) · δ(~ri − ~r2 ) − ρ(~r2 ) · ... . (7.12)
X
(n)
H
i i
| {z }
ρ̂(~
r1 )

For example, in case of particle pair (n = 2), the density-density correlation function is defined
as following, showing that essentially H 2 is equivalent to particle distribution function (PDF):
h i h i
0 0 0
H (2)
(~r,~r ) = ρ̂(~r) − ρ(~r) · ρ̂(~r ) − ρ(~r ) = ρ(~r)ρ(~r0 )h(~r,~r0 ) + ρ(~r)δ(~r − ~r0 ). (7.13)

Using definition of the intrinsic chemical potential ψ(~r) = µ − Vext (~r) one can show:

δ 2 βΩ δ 2 βΩ
H (2) (~r1 ,~r2 ) = − =− . (7.14)
δVext (~r1 )δVext (~r2 ) δψ(~r1 )δψ(~r2 )
And in general form for n particles in the system (see full derivation in Chapter 3.3 in [5]):
Chapter 7. Classical DFT (Density Functional Theory) 71

δ n βΩ
H (n) (~r1 , ...,~rn ) = − for n ≥ 2. (7.15)
δψ(~r1 ) · ... · δψ(~rn )

As a short summary of this section here are the key-points:

1. The grand potential is the generating functional for n−fold density correlations.

2. All spatial distributions and correlation functions follow from functional derivatives of the
the energy functionals.

3. To find the optimal density distribution, we can use the density functional framework, which
involves minimizing the grand potential functional using the calculus of variations. This typi-
cally requires solving the Euler-Lagrange equation, and computational methods based on this
approach are known as density functional theory.

7.3 DFT (classical)

Z
Consider following expression for mean energy of external field hHext i = d3 rρ(~r)Vext (~r), which
is essentially a functional of ρ(~r) or Vext (~r) (see Eq. (7.8)). To formulate a suitable free energy
functional we recall some basic thermodynamics.
Firstly, let us consider the fundamental relation – the first law of thermodynamics: dU =
T dS + i Fi dqi + µdN , where the term i Fi dqi is external field (e.g. −P dV ). Generalizing
P P

this expression to functional space we obtain:

Z
δU = T δS + ρ(~r)δVext (~r)d3 r + µδN, (7.16)

where δVext (~r)d3 r defines accessible volume for particles, where Vext is limited.
Secondly, we use definition of the Helmholtz free energy: F = U − T S with dF = dU − dT S −
T dS being its total differential. Again considering this energy a functional yields:
Z
δF = −SδT + d3 rρ(~r)δVext (~r) + µδN. (7.17)
Z
We can define an "intrinsic free energy" F = F − d3 rρ(~r)δVext (~r) = F − hHext i. Then, using
expressions found in Eq. (7.16) and (7.17), variation of this functional can be written as:
Z Z
δF = δF − d rδρVext (~r) −
3
d3 rρδVext (~r)
Z
= −SδT − d3 rδρ(~r)Vext (~r) + µδN (7.18)
Z
= −SδT + d3 rδρ(~r)ψ(~r),

where in the second term ρ(~r) and ψ(~r) are conjugated field variables. From this we can obtain
the functional derivative with respect to ρ(~r):

δF[ρ]
= ψ(~r), (7.19)
δρ(~r)
with local chemical potential ψ(~r) = µ − Vext (~r), which is sometimes also called "intrinsic chemi-
cal potential", because external field contributions are subtracted from total chemical potential
µ.
Chapter 7. Classical DFT (Density Functional Theory) 72

For the grand potential functional and its variation, we find:

Z Z
Ω = F − µN = F + d3 rρ(~r)Vext (~r) − µN = F − d3 rρ(~r)ψ(~r), (7.20)
Z Z
δΩ = δF + d rδρ(~r)Vext (~r) +
3
d3 rρ(~r)δVext (~r) − µδN − N δµ,
Z
δΩ = −SδT − ρ(~r)ψ(~r)d3 r, (7.21)

δΩ[ψ] δΩ[ψ]
=− = ρ(~r). (7.22)
δψ δVext (~r)
Looking at the Eq. (7.19) and (7.22) it is natural to take F and Ω as functionals of ρ(~r) and
ψ(~r) , respectively, which allows us to write following expression for the grand potential:
Z
Ω[ψ] = F[ρ] − ρ(~r)ψ(~r)d3 r, (7.23)

where F[ρ] – intrinsic, meaning no external field dependence, Helmholtz free energy.
For example, in case of the ideal gas F is typically split into ideal and "excess" parts as F[ρ] =
Fid [ρ] + Fexc [ρ].
Recall from thermodynamics, that for homogeneous ideal gas Fid [ρ] = N kB T [ln (ρΛ3 ) − 1], with
ρ = const. Since for ideal gas there are no interactions Zbetween particles, we can straight-
forwardly extend it to the inhomogeneous case with N = d3 rρ(~r), yielding:
Z
  
Fid [ρ(~r)] = kB T d3 rρ(~r) ln ρ(~r)Λ3 − 1 , (7.24)


Excess part of free energy, Fexc , however, typically is unknown.

7.3.1 DFT: basic theorems and variational principle

In the grand canonical ensemble, the condition for equilibrium is that the grand potential of
the system Ω is minimized at constant temperature, volume, and chemical potential.
In 1964 physicists W. Kohn and P. Hohenberg introduced two DFT theorems:
1. The first theorem states that the external potential of a system Vext (~r) uniquely determines
its single-particle density ρ(~r) for a fixed T, µ, V (see proof of the theorem in Appendix B in
[5]).
2. The second theorem provides a practical route to calculating the particle density by showing
that the total energy of a system is a unique functional of the single-particle density, given the
external potential. For more detailed derivation see Appendix B in [5].
Z
Consider Ω(T, µ, ψ) = F[n(~r)] − d3 rn(~r)ψ(~r) with n(~r) being some arbitrary density profile.
By requiring Ω is minimal for equilibrium ρ(~r), meaning:

δΩ δ
 Z 
= F[n(~r)] − d3 rn(~r)ψ(~r) = 0,
δn(~r) δn(~r)
n(~
r)=ρ(~
r) n=ρ

which can be also written in the following form:

δF[n]
= µ − Vext (~r). (7.25)
δn
n(~
r)=ρ(~
r)
Chapter 7. Classical DFT (Density Functional Theory) 73

For a fixed local chemical potential Vext (~r) n(~r) plays the role of an order parameter and
minimizes Ω[ψ] with the equilibrium distribution
Z ρ(~r). Using ideal gas case again, we consider
  
the Helmholtz free energy Fid = kB T d3 rρ(~r) ln ρ(~r)Λ3 − 1 , note that for convenience we


switch to the notation n(~r) → ρ(~r). The functional derivative and ideal gas density profile are
calculated as follows:

δFid  
= kB T ln ρ(~r)Λ3 ,
δρ

eβµ
ρid (~r) = · exp(−βVext (~r)).
Λ3
For interacting systems we get an additional term of excess energy F = Fid + Fexc . We need to
define so-called "first direct correlation function" as the first functional derivative of Fexc with
respect to ρ(1) (~r) :

δFexc
c(1) (~r) = −β . (7.26)
δρ
From minimization principle it follows:

eβµ −βVext +c(1) (~r)

ρ(~r) = ·e , (7.27)
Λ3
eβµ
where ≡ ρ0 from normalization conditions. Comparison with the corresponding ideal gas
Λ3
result (the barometric law) ρ(1) (~r) = ρ0 exp[−βVext (~r)] shows that the effects of particle in-
teractions on the density profile are wholly contained in the function c(1) (~r). It is also clear,
that −kB T c(1) (~r) is the excess contribution to intrinsic chemical potential ψ. Note that in
homogeneous situation, c(1) = const. = µexc .
One common approach to approximating the direct correlation function is through the use of
integral equations, specifically the Ornstein-Zernike (OZ) equation (see later in Section 7.3.2).
This equation relates the direct correlation function to the pair distribution function, which
describes the probability of finding two particles at a given separation. The OZ equation is a
transcendental equation, which can be solved numerically.
Chapter 7. Classical DFT (Density Functional Theory) 74

7.3.2 Typical examples (approximations) for Fexc [ρ] and applications

Local density approximation (LDA)
For more detailed description see Chapter 6 in [5].

Figure 7.1: Schematic plot ρ(x) for LDA approximation.

∂ρ(x) 1 1
Consider weak inhomogeneities (see Fig. 7.1) Assume  or ∇ρ(~r)  , where ξ is a
∂x ξ ξ
typical correlation length. Then we can write the function simply as local function of space,
Fexc
and with fexc = denoting the excess free energy per particle of the homogeneous case:
N
Z
Fexc [ρ] = d3 rρ(~r)fexc (ρ(~r))

Fexc is approximated by adding up local, nearly homogeneous regions. Also, fexc is typically ob-
tained from integration of equation of state (EOS) (see Eq. (7.28) below). Further approximat-
βP
ing, we can use a virial expansion: = 1+B2 ρ+B3 ρ2 +..., where the terms after unity are the
ρ
∂F ∂(Fid + Fexc )
excess part, besides the ideal gas contribution. One can show from P = − =− ,
∂V ∂V
that Fexc (ρ) = N kB T (B2 ρ + B3 ρ ) + ...
2

Strongly inhomogeneous ρ(~r) leads to high local densities which cause the local density approx-
imation to break down.
As a short note, good EOS for hard spheres is the Carnahan-Starling EOS (see Chapter 3.9 in
[5]):

βP 1 + η + η2 − η3
= , (7.28)
ρ (1 − η)3
π 3
where η = σ ρ is the packing fraction and σ is the hard sphere diameter.
6

Weighted-density approximation (WDA) & non-local forms

Here again only a brief summary of the approximation principles is presented, refer to Chapter
6 in [5] for more details.
Chapter 7. Classical DFT (Density Functional Theory) 75
Z
We will use the same Ansatz as in LDA: Fexc [ρ] = d3 rρ(~r)fexc (ρ̃), but now we define "weighted"
or coarse-grained density as follows:
Z
ρ̃(~r) = d3 r0 ρ(~r0 )ω(~r − ~r0 ), (7.29)
Z
where ω(~r − ~r0 ) is the weight function, normalized as d3 rω(|~r|) = 1. For example, consider
the second virial coefficient of a real gas:
1 1
Z Z
−βV (r)
B2 = − d r(e
3
− 1) = − d3 rf (r). (7.30)
2 2
Since the excess energy Fexc [ρ] = kB T d3 rB2 ρ2 (~r) in LDA, to be consistent we need to rewrite
R

this expression with respect to ρ̃(~r) from Eq. (7.29):

1
Z Z
Fexc [ρ] = − kB T d r 3
d3 r0 f (~r − ~r0 )ρ(~r)ρ(~r0 ). (7.31)
2
Then weighted function takes form of:

2 kB T f (~
r − ~r0 ) −1
1 1
 −βV (r) 
0 2 kB T e
ω(~r − ~r ) = − =− , (7.32)
B2 B2
where f (~r) – Mayer-function. Consider example of hard spheres (HS), where
1 1, r ≤ 0

ω(r) = θ(r − σ) with θ(r) = , then excess free energy of the HS system is:
2B2 0, r > 0

1
Z Z Z Z
3 0 0 0
βFexc [ρ] = d r
3
d r ρ(~r)ρ(~r )θ(|~r − ~r | − σ) = d r3
d3 r0 B2 ρ(~r)ρ(~r0 )ω(~r − ~r0 ). (7.33)
2
Many more sophisticated weight functions can be derived (see Chapter 6.5 in [5] and [14]).
WDA forms also follow from functional expansions or thermodynamic integration, which we
will discuss next.

Taylor expansion in ∆ρ(~r) and Ornstein–Zernike equation

More derivations and examples can be found in Chapters 3-4 in [5].
Consider the Taylor expansion of Fexc [ρ] in ∆ρ = ρ(~r) − ρ̄ around homogeneous state ρ(~r) ≡
ρ̄ = const.

δβFexc [ρ] δ 2 βFexc [ρ]

Z Z Z
βFexc [ρ] ≈ βFexc
0
+ d3 r ∆ρ(~r) + d3 r d3 r 0 ∆ρ(~r)∆ρ(~r0 ) =
δρ ρ̄ δρ(~r)δρ(~r0 ) ρ̄

1
Z Z Z
= βFexc
0
+ βµexc d3 r d3 r0 ∆ρ(~r)∆ρ(~r0 )c(2) (~r,~r0 ; ρ̄),
∆ρ(~r)d3 r − (7.34)
2
with µexc = const. and "direct correlation function" of second order defined as

δ 2 βFexc [ρ]
c(2) (~r,~r0 ) ≡ c(~r,~r0 ) = −β .
δρ(~r)δρ(~r0 ) ρ̄
Z
Also note, that ∆ρ(~r)d3 r = 0 if number of particles in the system N is conserved.
Before we continue onto the DFT minimization, we need to consider the so-called Ornstein–Zernike
(OZ) relation (also consider referring to Wikipedia page).
Remember from Section 7.3.1, that Fexc [ρ] is the generating functional for a hierarchy of direct
correlation functions (DCF), for example see Eq. (7.26) for 1st order DCF. One can show that
following important (and exact) relation holds between h(~r,~r0 ) and c(~r,~r0 ):
Chapter 7. Classical DFT (Density Functional Theory) 76

Z
0 0
h(~r,~r ) = c(~r,~r ) + ρ(~r00 )c(~r,~r00 )h(~r0 ,~r00 )d3 r00 . (7.35)

This equation isZ called Ornstein–Zernike relation. For homogeneous systems it follows as:
h(r) = c(r) + ρ d3 r0 c(|~r − ~r0 |)h(r0 ). We can also rewrite OZ relation by taking the Fourier
transform on both sides: h(k)
e = ce(k) + ρce(k)h(k),
e which yields:

ce(k)
h(k)
e = . (7.36)
1 − ρce(k)
Technically, if h(r) is known, then c(r) is known as well and vice versa. OZ equation is also a
defining equation for c(r). We can consider that total correlation equals to the sum of direct
correlation and convolution of direct correlation over total correlation. Note that, in the low-
density limit (LDL) c(r) = h(r). Historically, OZ has led to the development of the "integral
equation theory" (IET), where OZ is "closed" by a "closure-relation" between g(r) and c(r) to
solve the homogeneous problem.
Now we can go back to the idea of minimization of the energy functional. Consider the density
profile
Z
−βV (r)−β d3 r0 ρ(r0 )c(|~r − ~r0 |)
ρ(r) = ρ̄e . (7.37)
We will consider so-called "test particle" trick. The idea of this hypothetical "test particle" is
used to simplify the complex many-body problem of interactions between all the particles in
the system, and replace it with an effective one-body problem.

Figure 7.2: Illustration for the "test particle".

Consider one particle of the same kind as external potential for the others, then the density
profile yields:
 Z 
3 0 0 0
ρ(r) = ρ̄g(r)
A = ρ̄
A · exp −βV (r) − β ρ̄ d r g(r )c(|~r − ~r |) , (7.38)

which gives us the relation between g(r) and c(r). It can be solved with OZ Eq. (7.35) to obtain
g(r, ρ̄) and c(r, ρ̄) for the homogeneous case, which can then be used to calculate inhomogeneous
problems.
Chapter 7. Classical DFT (Density Functional Theory) 77

With the help of Percus’ test particle "trick" a self-consistent field theory can be developed.
Using Eq. (7.35) and Eq. (7.38) the correlation function c(r) can be rewritten as:

c(r) = h(r) − ln [h(r) + 1] − βV (r).

| {z }
g(~
r)

1. We see, that for large r : c(r) ≈ −βV (r), then for potentials with short-ranged repulsion
(g(r) ≈ 0) and long-range attraction following: g(r) = 0, r < σ and c(r) = −βV (r), r ≥ σ is a
good closure, so-called "mean spherical approximation" (further reading can be found in Chapter
4.5 in [5]).

2. In the low-density limit: c(r) = h(r) = e−βV (r) − 1 = f (r) (Mayer-function), thus we can
recover the WDA form in the LDL in Section 7.3.2.

3. Historically another closure was used – "hypernetted chain approximation" (HNC) in the
area of integral equation theories (IET). For more information refer to Chapters 4.3 and 10.4
in [5] and Wikipedia page.

7.3.3 λ-Integration and mean-field ("Vlasov") approximation

Thermodynamic integration (TI). Main goal is to find the free energy difference ∆F (pathway-
independent) between system in state I → II.
Potential energy function of known reference system is U0 , potential energy of target system is
denoted as UI . We shall introduce λ-parameter such that: U (λ = 0) = U0 and U (λ = 1) = UI ,
which we interpolate through the λ-path.
Remember the basic Zexpression for Helmholtz free energy F = −kB T ln Z and its partition
1
function Z = 3N d3N re−βU (λ) ≡ Z(λ). Then for the derivative with respect to λ at
Λ N!
constant N, V, T we obtain

∂F (λ) 1 ∂U −βU (λ) D ∂U (λ) E

  Z
= d3N r e = ,
∂λ N,V,T Z ∂λ ∂λ λ
∂U
with being generalized thermodynamic "force". The expression for thermodynamic in-
∂λ
tegration (TI) yields:
Z λ=1 D ∂U E
F (λ = 1) − F (λ = 0) = dλ . (7.39)
λ=0 ∂λ λ

We now introduce the pair potential Vλ (r) = λV (r). One example of a reference potential is
the hard-sphere interaction, where a weak, attractive tail can be considered as a perturbation.
The gradual increase of the parameter λ from 0 to 1 can be thought of as "switching on" the
perturbation. Z 1D
∂U E
Then the corresponding free energy functional for the system of interest is Fexc [ρ] = dλ
0 ∂λ
with potential U = λV (|~ri − ~rj0 |). Rewriting this expression:
X

i<j
1
Z 1 DX E Z 1
0
Fexc [ρ] = dλ λV (|~r − ~r |) = dλ ρ(2) (~r,~r0 , λ)V (~r − ~r0 ) =
0 i<j
2 0

1
Z 1
= dλ ρ(~r)ρ(~r0 )g(|~r − ~r0 |; λ)V (|~r − ~r0 |). (7.40)
2 0
The integration is over the λ-path. Expression obtained in Eq. (7.40) is formally exact; however,
g(r; λ) has to be known. Further reading can be found in Chapter 3.4 in [5].
Chapter 7. Classical DFT (Density Functional Theory) 78

We shall briefly introduce the Vlasov or "mean-field" approximation (used later in Chapter 8
on electrostatics). We consider g(r) = 1 (meaning no correlation), then:
Z Z
Fexc [ρ] = d r
3
d3 r0 ρ(~r)ρ(~r0 )V (|~r − ~r0 |). (7.41)

The Vlasov approximation is a fundamental concept in the kinetic theory of plasmas. It is a col-
lisionless approximation, which means that collisions between particles are neglected. Instead,
the motion of the particles is described by a single-particle distribution function.
In the low-density limit: g(r) = e−βV (r) we can perform explicit λ-integration. Result leads
consistently to LDL results of WDA (Sec. 7.3.2) and integral expansion (Sec. 7.3.2). Mean-field
approximation is quasi-exact for very soft potentials, such as Gaussian, at high densities. For
2
example, with V (r) = V0 e−(r/RG ) we get weak correlations for high densities (g(r) ≈ 1).

Combinations
For example, consider Lennard-Jones potential VLJ (r) (n = 6). We can also retrieve its form
approximated by the combination of hard-sphere and long-ranged attraction (Coulomb (n = 1))
potentials:

Figure 7.3: Schematic plot for the approximated Lennard-Jones potential (green line), obtained
by combining the hard sphere and Coulomb (red line) potentials.

The hard sphere potential takes into account the excluded volume effect of the particles, while
the attractive tail represents the long-range van der Waals attraction between the particles.
This approximation is useful because it is simpler to work with than the full Lennard-Jones
potential and can provide reasonable approximations for many physical systems.

Fexc = F HS
exc + Fexc
MF
,
| {z } | {z }
Hard sphere Mean field
(e.g. FMT) functional

ρ(~r0 )ρ(~r0 )
Z ∞ Z ∞
where MF
Fexc ∼ d r
3
d3 r 0 , form of which is similar as in the MSA and the abbre-
σ σ rn
viations stand for FMT – Fundamental Measure Theory (see Chapter 6.5 in [5]), MSA – Mean
Spherical Approximation (see 7.3.2 and also Chapter 10.3 in [5]). Electrostatic functionals are
very similar in classical and electronic DFT.
Chapter 7. Classical DFT (Density Functional Theory) 79

7.3.4 More example applications

To be completed - see handouts.
Chapter 8

Electrostatics of liquid electrolytes

and the Poisson-Boltzmann
Equation

Electrostatics is a branch of physics that deals with the study of electric charges at rest. It de-
scribes the behavior of charges and their interactions with electric fields. In molecular systems,
electrostatics plays a critical role in determining the stability, conformation, and reactivity
of molecules. The electrostatic interaction between charged molecules or ions can affect the
behavior of enzymes, the binding of ligands to proteins, and the transport of ions across cell
membranes. The challenge in liquid systems (electrolytes in solvent, e.g., water) is that ions
are mobile and they can respond to the field by moving into arrangements so that electrostatic
(free) energies are minimized. This requires a self-consistent theoretical formulation, which is
approximately provided by the Poisson-Boltzmann (PB) equation. It can be derived from a
mean-field DFT approximation for Coulombic interactions but also motivated well heuristically.
The PB equation can be linearized leading to Debye-Hückel-like theories. Their concepts are
fundamental in the modeling of charged liquids. Charges in liquids are the rule, rather than
exception and have large implications in biophysics and energy materials, such a batteries or
electrochemical supercapacitors.
Hence, the PB equation is a mathematical model that is widely used to describe the electro-
statics of molecular systems. The PB equation takes into account the Coulombic interactions
between charged particles in the implicit surrounding solvent molecules (dielectric environ-
ment). It provides this a way to calculate the electrostatic potential and the distribution of
ions around charged molecules or surfaces in solution. The PB equation is particularly useful
for studying biomolecular systems, such as proteins, DNA, and membranes, where the charges
are distributed over a complex surface.
The application of PB equation in different geometries involves solving the PB equation for
different boundary conditions. We will briefly cover the spherical, planar and cylinders cases.

8.1 Electrostatics of molecular systems

Typically, ions, molecules, and macromolecules carry charges due to functional groups, amino
acids, or nucleic acids. These charged particles move in an environment that is also ionic or
charged to some degree. The electrostatic interactions between these charged particles and the
surrounding charged environment determine the behavior and stability of molecular systems,
and play a critical role in many biological and chemical processes. Examples:

1. Dissolution/hydration: for example, when sodium chloride dissolves in water, the sodium
Na+ and chloride Cl− ions and the polar water molecules are strongly attracted to one another

80
Chapter 8. Electrostatics of liquid electrolytes and the Poisson-Boltzmann Equation 81

by ion-dipole interactions (see Fig. 8.1).

The solvent molecules (water in this case) surround the ions removing them from the crystal
and forming the solution. As the dissolving process proceeds, the individual ions are removed
from the solid surface becoming completely separate, hydrated species in the solution.

Figure 8.1: Dissolution process of Na+ Cl− crystal. Image source.

2. Cell membrane/charged cathode with simple geometries:

(a) "sphere" (more details in Section 8.2.1),

(b) "plane" (more details in Section 8.2.2).

3. Micelle.
A micelle is a self-assembled structure composed of amphiphilic molecules, dispersed in a liquid,
forming a colloidal suspension. In a micelle, the hydrophobic tails of the amphiphilic molecules
cluster together in the center of the structure, shielding themselves from the surrounding polar
solvent, while the hydrophilic head groups face outward and interact with the solvent (see
Fig. 8.2). Micelles are typically spherical or cylindrical in shape, with a diameter ranging from
a few nanometers to several hundred nanometers. They are important in a wide range of
applications in chemistry (micellar catalysis), bioengineering and medicine (drug delivery).
Chapter 8. Electrostatics of liquid electrolytes and the Poisson-Boltzmann Equation 82

Figure 8.2: Schematic view of a micelle structure, containing hydrophilic/polar region (head)
and hydrophobic/nonpolar region (tail). Adapted from: source.

4. Charged colloids.
Charged colloids are dispersed systems consisting of particles with sizes ranging from nanome-
ters to micrometers that carry a net electrical charge (see Fig. 8.3). The interaction potential
between two charged colloidal particles can be described by the DLVO theory [5, 6]., which
takes into account the repulsive electrostatic force due to the charged surface of the particles
and the attractive van der Waals force between them (see later in Chapter 9).

Figure 8.3: Schematic view of a charged colloid. This example can be used for spherical
geometry case of electrostatic description.

5. Charged "rod" (e.g. model for DNA or charged liquid crystals). In this case it is also useful
Q
to define line charge λ = (see Fig. 8.4).
l
Chapter 8. Electrostatics of liquid electrolytes and the Poisson-Boltzmann Equation 83

Figure 8.4: Schematic view of a charged rod.

All charges interact with Coulomb pair potential:

Z1 Z2 e 2 1
VC (r) = · , (8.1)
4πo r r

where Zi - valencies (for Na+ , Cl− : Z1 = 1 ≡ Z+ , Z2 = −1 ≡ Z− ) and 0 – dielectric constant,

r – relative permittivity.
Even with simple geometry case, such as planar case, the description of the phenomena is still
complicated and raises nontrivial questions (see Fig. 8.5).

1. The ionic movement and fluctuation of charges can affect the accuracy of the solutions.

2. The long-range nature of the electrostatic interactions proportional to the inverse distance
 1
∼ also presents challenges, as it can lead to slow convergence of numerical methods.
r
3. The global electroneutrality constraint and the correlations between positive and negative
charges can also complicate the solution.

Figure 8.5: Schematic plot ρ(z) in planar case. See expression for κ−1 in Eq. (8.6).
Chapter 8. Electrostatics of liquid electrolytes and the Poisson-Boltzmann Equation 84

Theoretical treatment is usually done by using the Poisson-Boltzmann (PB) equation. Below
you can see its heuristic derivation.
As the first step, we use the Poisson equation, which describes the electrostatic potential at a
given point in space due to a distribution of charges:

λ(~r)
∇ · [r (~r)∇φ(~r)] = − , (8.2)
0
where φ(~r) – electrostatic potential, r (~r) – position-dependent dielectric constant, λ(~r) – charge
density (per volume). If λ and r are known, integration of Poisson equation leads to electric
field and potential. Also note that often r = const. for simplicity.
As the ions can freely move, we can use the Boltzmann distribution, which describes the statis-
tical distribution of ions in solution based on their charges and the electrostatic potential. This
distribution accounts for the thermal energy of the system and the tendency of ions to move
towards regions of low electrostatic potential.

λ(~r) = λext (~r) + λion (~r),

where λext (~r) – external, fixed charge distribution and λion (~r) – ion distribution.
Boltzmann assumption, also known as mean-field approximation, is an assumption used in
statistical mechanics to simplify the calculations of the properties of a system with a large
number of interacting particles.
We treat the system as if each particle in the system interacts with the average effect of all
other particles, rather than considering the detailed interactions between each pair of particles.

λion = Zi eρ0i e−βZi eφ(~r) = Zi eρi (~r), (8.3)

X X

i i

where Zi – ion valencies and ρi – bulk (reservoir) density.

Combining Eq. (8.2) and (8.3) we obtain the Poisson-Boltzmann equation, which is a
nonlinear differential equation for φ(~r) and ρi (~r). With assumption of λext = 0 and r = const.,
we get:

λion (~r) 1 X
∆φ(~r) = − =− Zi eρ0i e−βZi eφ(~r) . (8.4)
0 r 0 r i

Eq. (8.4) can be linearized by assuming that the electrostatic potential is proportional to the
charge density. This is a useful approach when the electrostatic potential is small compared to
the thermal energy of the system (valid for eβφ  1).
In this case, the PB equation takes form of:

∆φ(~r) = − Zi eρ0i (1 − βZi φ) = + βZi2 e2 ρ0i φ(~r).

X X X
− Zi eρ0i
i i i
| {z }
=0 (electroneutral)

Linearized Poisson Boltzmann equation (LPB):

∆φ = (8.5)
X
βZi2 ρ0i e2 φ(r).
i

The term in front of φ(r): κ2 = β Zi2 e2 ρ0i → κ – is inverse screening length. Though more
X

i
widely used is λD = κ−1 .
Debye-Hückel theory is linearized Poisson Boltzmann model (Eq. (8.5)), that neglects higher-
order terms in the ion concentration and assumes that the electrostatic interactions are screened
Chapter 8. Electrostatics of liquid electrolytes and the Poisson-Boltzmann Equation 85

by the solvent and that the ion distribution is homogeneous. It is a good approximation for
weakly charged systems in which the ionic strength is low and the screening length is much
larger than the characteristic length scale of the system. Debye-Hückel theory introduces two
important parameters:

1) Debye-Hückel (screening) length, which is defined by

1 1
λD = =s . (8.6)
κ
4πλB
X
Zi2 ρ0i
i

The Debye-Hückel screening length describes the extent to which electrostatic interactions
between ions in solution are screened or attenuated by the solvent. It is also an important
parameter in the formation of double layers at charged surfaces and the screening of electrostatic
interactions between molecules in solution.
For monovalent, symmetric salts, like Na+ Cl – , where ρ0+ = ρ0− = cs - salt concentration
√ (e.g.,
1M =b 1 mol/l) and |Z1 = | − Z2 | = 1 the expression simplifies to: κ = λ−1
D = 8πλ c
B s .
Debye-Hückel length λD can vary from mm to sub-mm, depending on valency, permittivity
and salt concentration. For 0.1 mol/l of monovalent salt in water at 300 K λD ' 1 nm. In
the presence of a relatively strong electrolyte, the electrostatic interactions are exponentially
screened and can be effectively neglected for lengths larger than the Debye–Hückel screening
length.

2) Bjerrum length – electrostatic coupling parameter:

e2 1
λB = · . (8.7)
4π0 r (kB T ) kB T

The Bjerrum length is a length scale that characterizes the strength of the electrostatic in-
teractions between charged particles in solution. It is defined as the distance at which the
electrostatic interaction energy between two unit charges is equal to the thermal energy kB T .
The Bjerrum length depends on the dielectric constant of the solvent and is typically on the
order of a few angstroms (λB = 0.71 nm = 7.1Å for water (at normal conditions)).

8.2 Applications of PB in various geometries.

In this section we will take a look at the concepts of solving the Poisson-Boltzmann equation
for spherical and planar geometries as well as discussing cylindrical case.
Chapter 8. Electrostatics of liquid electrolytes and the Poisson-Boltzmann Equation 86

Figure 8.6: Schematic view of of planar, spherical and cylindrical cases.

8.2.1 Spherical case

Consider LPB equation in the following form:

1 d2 (rφ(r)) φ(r)
= 2 .
r dr2 λD
Derivation with respect to potential φ(r) yields:

rφ(r) = Ae−r/λD + Be+r/λD ,

with φ(r) = 0 and setting r → ∞ leads to

A −r/λD e−r/λD 1 1
 
φ(r) = e , then φ0 (r) = A +
r r λD r
Q
⇒ φ0 (r)|r=R ≡ (electrostatic field at r = R).
4π0 r R2
Then we arrive to the so called Debye-Hückel potential :

Q λB e−(r−R)/λD
eβφ(r) = φ(r)
e = , (8.8)
e r 1 + R/λD
!
Q λB e−(r−R)/λD Q λB
φ(r)
e = −1 + .
e r 1 + R/λD |e {zr }
Coulomb
| {z }
screening part

In the expression above first term represents the short-range interactions due to the presence
of the solvent molecules, the second one – the long-range electrostatic interactions between the
ions in the solution.
We can also calculate the work done on the screening ion cloud by charging the ions:
Chapter 8. Electrostatics of liquid electrolytes and the Poisson-Boltzmann Equation 87

1 λB 1 1
Z e
λB
  
βw = −1 Q0 dQ0 = − .
e2 R 1 + r/λD 0 2 λD + R
This expression is referred to as "Born self energy". In the case of a salt solution, the Born
self energy of an ion determines the electrostatic energy of the ion in the presence of its own
electrostatic field and the electrostatic field of the other ions in the solution.
Introducing the activity coefficient γ = eβw or kB T ln γ = w. Let us rewrite this expression using
λB and λD :
λB 1
 
ln γ = − .
2 λD + R
λB
In case of λD  R ' we arrive to the Debye-Hückel limiting law, name of which
2λD
signifies that its application is limited to dilute solutions of strong electrolytes only.
Deviations from linear electrostatics are attributed to charge renormalization (condensation)
caused by collapse of counterions from the ion atmosphere.

8.2.2 Planar case

Without loss of generality, we assume that the membrane is negatively charged and take the
Q
surface charge density (per unit area) as a negative constant, σ = < 0.
A

A single planar charged surface

In terms of the ionic solution itself, two different electrostatic situations will be distinguished.

1. Counterions (cations here) only.

In this case PB equation takes form:

eρ0 −eβφ(z)
φ00 (z) = − e , (8.9)
r

where ρ0 is the reference density, taken at zero potential in the absence of a salt reservoir.
σ
Eq. (8.9) with boundary condition: φ0 (z = z0 ) = − > 0 and vanishing electric field at infinity,
r
can be integrated analytically twice, yielding:

eβφ(z) = 2 ln(z + b) + φ0 ,

so that ionic density are (see Fig. 8.7):

1 1
ρ(z) = ,
2πλB (z + b)2

e
where b = – "Gouy-Chapman length".
2πλB σ
Chapter 8. Electrostatics of liquid electrolytes and the Poisson-Boltzmann Equation 88

Figure 8.7: Schematic plot of ρ(z), solution of the PB equation in planar geometry for "only
counterions" (cations) case.

Physical meaning behind the value of Gouy-Chapman length is following: within the layer of
thickness b close to the membrane, half of the counterions ( 21 |σ|) are accumulated. At the length
scale of b, the thermal energy is equal to the Coulombic energy between a unit charge and a
planar surface with a constant surface-charge density σ. For strongly charged membranes, b is
rather small, on the order of ∼ 1 nm.
Z b Z ∞
σ
Note: e ρ(z)dz = – bound, but zρ(z)dz = ∞ – "loose".
0 2 0

2. Added electrolyte.
In many biological situations, the charged membrane is in contact with a reservoir of electrolyte.
Two types of charge carriers are present in the solution: co-ions and counterions and both types
are in thermal equilibrium with the reservoir.
Since the surface charge was taken as negative, the potential tends to zero from below at large
z. Hence, it is negative for all z values. At larger z, both densities ρ± tend to the reservoir
(bulk) value ρ0 , where the potential is zero. However, their distributions are quite different
close to z = 0 since the counterions are attracted to the charged surface whereas the co-ions are
repelled from it.
For a symmetric electrolyte and the same boundary condition of constant surface charge σ,
solution of the PB equation takes form:

1 + γe−z/λD
!
eβφ(z) = −2 ln .
1 − γe−z/λD

The two ionic profiles then (see Fig. 8.8):

1 ± γe−z/λD
!
ρ± (z) = ρ0± ,
1 ∓ γe−z/λD
s
2
b b e

with γ = − + + 1 and b = .
λD λD 2πλB σ
Chapter 8. Electrostatics of liquid electrolytes and the Poisson-Boltzmann Equation 89

Figure 8.8: Schematic plot of ρ(z) in in planar geometry for added electrolyte case.

For constant surface potential, the parameter γ can be obtained by setting z = 0:

eβφ(z = 0) = −4arctanh(γ)

or rewriting the expression for surface charge, we yield the Grahame equation:

eβφ(0)
q  
σ= (8ρ0 0 r kB T ) sinh . (8.10)
2

This equation describes the relation between surface charge and surface potential, with an
assumption of electroneutrality condition: the total charge of the double layer must be equal
to the negative of the surface charge. For the case of lower potentials sinh x can be expanded
3
to sinh x = x + x3! + ... ' x and with λD – Debye length, we can obtain the linearized Grahame
equation (see exercise sheet):

0 r
σ= φ(0). (8.11)
λD
Chapter 8. Electrostatics of liquid electrolytes and the Poisson-Boltzmann Equation 90

Figure 8.9: Compare to the plate capacitor situation, at low potentials capacitance, derived
dσ 0 r
from Eq. (8.11): C = ' .
dφ(0) λD

Two planar charged surfaces

The results for a single flat membrane can be extended to include the case of two identically
charged planar membranes at a separation L, immersed in an aqueous solution as is illustrated
in Fig. 8.10.
One of the interesting and measurable physical quantities is the electrostatic pressure felt by
the membranes. For two flat boundaries, it can be shown, that is directly proportional to the
increase in the concentration of ions at the midplane. Namely, this pressure is the excess in
osmotic pressure of the ions at the midplane over the bulk pressure. See more information and
derivation in Chapter 14 in [3].

Figure 8.10: One membrane is located at z = −L/2, while the other is at z = L/2. The surface
potential on both membranes is denoted by φ(z ± L/2), and the midplane one with φ(0) ≡ φ0 .

Ionic densities can be written as:

ρ± (z) = ρ0± e∓βφ(z) ,

with ρ0± ≡ ρ0 . Also consider ρN (z) = ρ+ (z) + ρ− (z) – total number density.
Chapter 8. Electrostatics of liquid electrolytes and the Poisson-Boltzmann Equation 91

Using this and the Boltzmann distribution:

h i
ρ0N (z) = ρ0 βeφ0 (z) e+eβφ(z) − e−eβφ(z) ,
and then plugging in the PB we obtain:

0 r d 0
0 r βφ00 (z)φ0 (z) = β (φ (z))2 ,
2 dz
dφ
with E = , integration leads to:
dz
0 r 2
kB T ρN (z) − Posm = E (z). (8.12)
2

Eq. (8.12) represents the mechanical balance between excess osmotic pressure and electrostatic
pressure, it is so called "hydrostatic equilibrium". At any position z between the membranes, the
osmotic pressure has two contributions. The first is a negative Maxwell electrostatic pressure
proportional to (φ0 (z))2 . The second is due to the entropy of mobile ions and measures the
local entropy change with respect to the ion entropy in the reservoir.
Considering symmetry and boundary condition:

1. z = 0 → E(z) = 0, then:

βPosm = ρN (0) = 2ρ0 (in equilibrium).

L
2. z = ± , then:
2
 L σ2
βPosm = ρN z = ± − . (8.13)
2 20 r

If the electric field and ionic densities are calculated right at the surface, we obtain the contact
theorem that gives the osmotic pressure acting on the surface (Eq. (8.13)). This result also
shows that the concentration of counterions at the surface depends only on the surface charge
density σ and the counterion concentration at the midplane.
Consider more examples and details in chapter "Introduction to electrostatics in soft and bio-
logical matter" in [12] and in [6].

8.2.3 Cylindrical case

• Manning condensation
Linearized version of the PB equation is an appropriate approximation in cases of low surface
charge densities, however when dealing with particles of high surface charge density, and a
radius comparable to the Debye length (e.g. DNA molecules), this approximation is no longer
valid. For the subclass of solution electrostatics problems that involve charged rods, a popular
alternative has been the Manning theory of counterion condensation (CC).
This approach treats the distribution of counterions around highly charged polyelectrolytes in
terms of the linear charge density parameter ξ = λB /d, with the Bjerrum length λB , and d is
the axial distance between successive charges fixed to the polyelectrolyte chain.
Manning considered highly charged polyelectrolytes (ξ > 1) and counterions with valency Zi . He
argued that the PB ion atmosphere around the polyion is unstable for (ξ > 1/Zi ) and proposed
that, as a result, the fraction (1 − 1/Zi ξ) of the fixed charges on the rod becomes completely
neutralized by counterions that condense onto them, effectively reducing the density of the fixed
Chapter 8. Electrostatics of liquid electrolytes and the Poisson-Boltzmann Equation 92

charges on the rod from ξ to the value 1/Zi . Counterion condensation theory then assumes that
the non-condensed counterions are distributed according to the linearized PB equation. It is
partly the nature of Manning’s arguments that has made his condensation theories controversial.
More detailed description can be found in [7] and [8].

• Mathematical description is relevant not only for interaction of DNA with other molecules,
but also for charged colloidal particles (cylindrical viruses and nanotubes, etc.), as well as
charged membranes and ion channels.
Chapter 9

Coarse-graining of multi-component
systems: effective interactions

For modeling larger systems and larger length- and timescales we need to integrate out smaller
degrees of freedom (DOFs) to arrive at effective mesoscale Hamiltonians. This can reduce the
computational cost by orders of magnitude because instead of many microscopic DOFs, one
simulates only a few effective ones of interest. For example, consider salt ions in water and
we want to access the osmotic pressure of the salt. Instead of an all-atom simulations of 100
salt pairs in 10,000 water molecules with a microscopic Hamiltonian, one can define effective
water-mediated corrections to the ion-ion pair potential to take water effects approximately into
account. This can be done by inverting ion-ion RDFs (see Appendix) to define effective pair
potentials [11]. Such a coarse-graining strategy applies to all kind of multi-component systems,
where small and large-scale DOFs are present and the small (and quickly relaxing) DOFs are
not of explicit interest. Some examples are shown in Fig. 9.1. In the following we discuss how
formally a ‘coarse-graining’ can be done by integrating out DOFs and defining a ‘potential of
mean force’ or ‘effective interaction’ or ‘effective pair potentials’ [5, 6].
Examples for two-component mixtures are presented in Fig. 9.1

A B C
Figure 9.1: Examples for two-component mixtures: A ions in water, B colloids and polymers,
C charged colloids and salt.

93
Chapter 9. Coarse-graining of multi-component systems: effective interactions 94

9.1 Potential of mean force: effective interaction

Figure 9.2: A binary system with NM macroparticles at distances R

~ i and Nm microparticles at
distances ~ri .

Consider the Hamiltonian of a two-component system as in Fig. 9.2, formally given as:

H = KM + VM M + Km + Vmm + VM m . (9.1)

with Ki being the kinetic terms and Vij being the interactive terms, e.g., based on a sum of
pair potentials
N Nm
MX
VM m = V (|R
~ i − ~rj |) (9.2)
X

i j

The goal now is to integrate out the DOFs of the small, microparticles (m) to define effective
interactions between the macroparticles (M). The total partition sum is defined as:

Z = he−βH i = hhe−βH im iM = he−βHeff iM (9.3)

where h...i is based on the trace over all coordinates

Z Z Z Z
dP~1 ...dP~NM d~ pNm ×
p1 ...d~ dR
~ 1 ...dR
~N
M
d~r1 ...d~rNm (9.4)

and h...im is just the integral over the m coordinates. Eq. (9.3) is the defining equation for the
effective Hamiltonian: the exact partition sum (or free energy) follows from integrations over
all configuration of the macroparticles using their effective Hamiltonian. With the expression

he−βH im = e−β(KM +VM M ) he−β(Km +Vmm +VM m ) im , (9.5)

the effective Hamiltonian is then written as

 
Heff = KM + VM M − kB T ln he−β(Km +Vmm +VM m ) im (9.6)
| {z }
free energy of m-particles in external field of M -particles

Heff = KM + VM M + F(T, [ρm (~r)], {R

~ i }) (9.7)
| {z }
~ i })=VM M ({R
tot ({R
Veff ~ i }+Veff ({R
~ i })

The effective interaction Veff

tot is then the sum of the intrinsic (vacuum) M − M interactions and

the microparticle (solvent)-induced free energy Veff . This is an exact result that is generally
valid.
Chapter 9. Coarse-graining of multi-component systems: effective interactions 95

The effective force, F~i is derived by:

F~i = −∇R~ i Veff ({Rj }) = −∇R~ i VM M

tot ~
−∇R~ i Veff
| {z } | {z }
pair forces induced many-body forces

hβ∇R~ i VM m e−β(KM +Vmm +VM m ) im

= −∇R~ i VM M − kB T
he−β(KM +Vmm +VM m ) im
= −∇R~ i VM M − h∇R~ i VM m im (9.8)

The effective force of a macroparticle i is thus identical to the sum of the ‘vacuum’ pair force
and the mean force induced by the microparticles. That’s why the integrated force, leading to
a potential Veff
tot , is called the "potential of mean force" (PMF). Again, this is a formally exact

potential (including many-body effects) but in practice almost impossible to calculate because
Veff depends on the specific configuration ({R ~ j }).
Important: If we perform such a calculation for only NM = 2, we obtain an effective pair
potential (see, for example, Figure 9.5):

({Ri }) → V (r)
tot ~
Veff (9.9)

This is also how it is done in practice. The mean force or effective potential, see next section, is
only calculated for two macroparticles in the sea of microparticles. However, then many-body
interactions are neglected. (That is, if a third particle influences the pair potential between the
first two). In practice that is a popular approximation (which validity should be scrutinized
but is often not considered). Per construction it should be always good at low densities of the
(effective) macroparticles in the coarse-grained system.

9.1.1 Two-body approximation

If we perform such a calculation in the two-body approximation (NM = 2), the functional form
of the free energy (see Eq. (9.7)) reduces to:

N
~ i } ' F0 NM 1X M
   
F T, [ρm (~r)]; {R + φ(~ri ,~rj ) (9.10)
V 2 i6=j
| {z }
const

and we obtain the effective two-body pair potential (note that it is usually denoted at V (r):

φ(~ri ,~rj ) = φ(| ~ri − ~rj |) = φ(r) (9.11)

Then, the liquid structure follows from equilibrium statistics according to Heff .
*N +
M
ρ(1) (~r) = δ(~r − ~ri ) ≡ ρ(~r) (9.12)
X

i=1 Heff
*N +
M
(~r, r~0 ) = δ(~r − ~ri )δ(r~0 − ~rj ) ≡ g(r) RDF (9.13)
X
(2)
ρ
i6=j Heff

This is also how it is done in practice. The mean force or effective potential, see next section, is
only calculated for two macroparticles in the sea of microparticles. However, then many-body
interactions are neglected. (That is, if a third particle influences the pair potential between the
first two). In practice that is a popular approximation (which validity should be scrutinized
but is often not considered). Per construction it should be always good at low densities of the
(effective) macroparticles in the coarse-grained system.
Chapter 9. Coarse-graining of multi-component systems: effective interactions 96

9.1.2 Internal degree of freedom

We can now consider that the M macroparticles have m internal degree of freedom as repre-
sented in Fig. 9.3, for example monomers in a polymer network (hydrogel).

(a)

(b)

Figure 9.3: Coarse graining a one component system, with internal degree of freedom

Since these particles are so soft and deformable, it would be desirable to still resolve some
properties in the C.G model, e.g. the size or shape. In that case, one would need to enrich the
effective Hamiltonian with some collective variables that would describe the change of properties
such as:
Heff (~ri ) → H̃eff (~ri ;~σi ) (9.14)
• ~σi =: new CG (collective variable) that can assume different equilibrium values according
to some distribution.
• ~σi or σi can be the particle size, but in principle general (shape, charge density, orientation,
function, ...)
Details of such a coarse-graining of internal degrees of freedom can be found in Lin et al., Phys.
Rev. E 102, 042602 (2020).

9.2 Typical coarse-graining procedure

As a practical example, we look at a binary system with a species of big particles b (the previous
macroparticle) and a species of small particles (solvent) s (the previous microparticle), as seen
in Fig. 9.4, with densities ρb , ρs and interactions of Vbb , Vss and Vbs .
Practical question: what structure and phase diagram do the big particles have?
In the low density limit of the big particles, the interactions reduce to an effective pair potential
(see Fig. 9.4). By coarse-graining the pair potential, a one-component system of "effective" big
particles is derived.
Chapter 9. Coarse-graining of multi-component systems: effective interactions 97

Figure 9.4: Typical coarse-graining procedure: A binary system with species of big and small
particles which needs to be investigated; B consider only the low density limit with Nb = 2; C
Coarse-graining (‘integrating out’) results into an effective pair potential, leading to D a one-
component system of "effective" big particles. The final system has much reduced complexity
but due to the pair potential approximation is not exactly representing the original system.

How do we calculate Veff in practice?

1. Simulation

Figure 9.5: Calculating the effective pair potential Veff through simulation. Two big particles
are at fixed positions R
~ 1 and R
~ 2 in a fixed distance. The mean force from the small ones can be
averaged and will provide typically a net force in direction of the connection axis. Integrating
the force along the big particle distance wil provide the PMF, which in this case is equal to the
effective (coarse-grained) pair interaction.

In simulations, there are two standard ways for obtaining the effective pair interaction.
The first route is based on the calculation of the mean force. The two particles are fixed
in their distance, cf. Fig. 9.5. The mean force of all small particles is calculated, formally
through

F~eff (R ~ 2) = h
~ 1, R F~bs (R ~ 2 ,~ri )i
~ 1, R
X

i
Z ∞Z 1
= −2π r2 Fbs (r)ρs (~r, R ~ 2 ) × ωdωdr
~ 1, R (9.15)
0 −1

where ρs (~r, R ~ 2 ) is the triplet-density, ie., the one-body density profile of the small
~ 1, R
particles if the two big ones are placed at R ~ 1 and R~ 2 . In practice, the density profile
has not to be averaged, the mean force can simply be averaged ‘on-the-fly’ during the
Chapter 9. Coarse-graining of multi-component systems: effective interactions 98

simulation. Due to the symmetry of the problem, the scalar force in direction of the
distance vector is sufficient. In orthogonal directions the mean force must average out to
zero (in practice, this is a small number which gives an indication of the statistical error).
Due to actio = reactio, the mean force must be anti-symmetric on both particles. Finally,
integrating this mean force over many distances (typically 10-20 well interpolated values)
gives the PMF. In this case of only two particles, this is identical to the effective pair
potential.
Other sampling methods are possible, too. For example, umbrella sampling (US) (see
also chapter 11) is very popular, see also section 11.1.5 in this script. The US method
does not calculate forces but reconstructs the effective pair potential (free energy along
the distance coordinate) through probability distributions.
There are also other approaches [5, 6] which we just briefly mention:

2. Classical density functional theory (DFT)

This approach uses the liquid state theory. With the approximation ρb → ∞, the triplet
correlation ρs (~r, R ~ 2 ) is derived and the mean force calculated.
~ 1, R

3. Superposition approximation One can approximate the triplet correlation ρs (~r, R ~ 2)

~ 1, R
by a product of one-body profiles of only one particle:

ρs (~r, R ~ 2 ) ' ρ0 g(|~r − R

~ 1, R
s
~ 1 |)g(|~r − R
~ 2 |) (9.16)

This is only an exact treatment for the assumption of the species of small particles being
an ideal gas.

9.3 Effective macromolecular interactions: examples

In principle, all pair potentials are in some sense effective interactions. Examples for effective
pair interactions are: Lennard-Jones (electrons integrated out), Dispersion (Hamaker: atomic
van der Waals integrated out), DLVO theory and potential (mobile charges and ions integrated
out), hard core colloids (all chemical DOFs integrated out so that just a ‘shape’ remains),
soft repulsion between polymers (polymer monomers integrated out), hydrophobic interactions
(water integrated out), etc.

9.3.1 Depletion Potential

An important example in mixtures or macromolecules in solvents is the so-called "Depletion
potential" due to solvation forces.
The simplest example of a depletion potential is provided by the Asakura-Oosawa (AO)
model for colloid-polymer mixtures, cf. Fig. 9.6. It can be calculated analytically, so the
formal procedure of integrating out DOFs can be performed and followed explicitly!
Chapter 9. Coarse-graining of multi-component systems: effective interactions 99

Figure 9.6: A binary system with NM macroparticles (colloids) and Np polymers with a density
N
of ρp = Vp and V being the total volume of the system.

For this, we consider two macroparticles NM = 2 surrounded by polymers as in Fig. 9.7.

Figure 9.7: NM = 2 macroparticles (colloids) of radius R in a distance rM M . Np polymers of

interaction size σ are surrouning the colloids. The polymers are excluded from spheres of a
radius R + σ/2.

The interactions in the AO model system are:

∞ for rM M < 2R ∞ for rM p < R + σ2

( (
VM M = Vpp = 0 ideal VM p = (9.17)
0 else 0 else

Note that all interactions are repulsive and entropic! The polymer-polymer interaction is ideal,
which is a good approximation because these chain coils can easily interpenetrate each other.
However, if a polymer approaches the big sphere (colloid) there will be an entropic (elastic)
penalty and one can assume a hard repulsive excluded-volume interaction between polymers
and colloids. Importantly, this creates an excluded volume shell around the colloids for the
polymers, see again Fig. 9.7.

Figure 9.8: If we consider the possible overlap volume of the macromolecules, the accessible
volume of the polymer increases.
Chapter 9. Coarse-graining of multi-component systems: effective interactions 100

Now, when the colloids comes closer, the inaccessible volumina start overlapping, cf. Fig. 9.8.
We have an overlap volume V0 . Hence, the overall accessible volume for the polymers increases.
More volume means more configurations in space and this means more entropy for the system:
the colloids will effectively attract to minimize the system free energy! The accessible polymer
volume is given exactly as (see Fig. 9.8)

3rM M rM
!
π 3
Vacc = V − 2VM + V0 (rM M ) = V − D3 1 − + M
(9.18)
6 2D 2D3

with VM the volume of the two colloids, and D = 2R + σ their surface-to-surface distance, and
2R ≤ rM M < D.
The free energy of the polymers is purely entropic and given as:
N !
Vaccp
F = −T S = kB T ln = −kB T ln Z
Ω3Np Np !
Vacc (rM M )
! !
V Np

= kB T ln + Np ln
Ω Np !
3Np V
Vacc (rM M )
 
= 0
Fideal + kB T Np ln (9.19)
V

where Z is the partition sum. For V  π6 D3 , we can linearize via ln (1 − ) ' −:

3rM M rM
!
Np πD3 3
F= 0
Fideal − kB T 1− + M
(9.20)
V 6 2D 2D3

We define the pair potential as Vdepl (r) → 0 for rM M → ∞.

3rM M rM
!
πD3 3
Vdepl (r) = −ρp kB T 1− + M
(9.21)
6 2D 2D3

with 2R < r < D.

The effective pair interaction is then given as Vtot = VM M + Vdepl as seen in Fig. 9.9.

Figure 9.9: Total effective pair interaction as Vtot = VM M + Vdepl .

Features: the depletion potential

• has an attraction which is proportional to polymer density and attraction range propor-
tional to polymer size which leads to tuneable interactions
Chapter 9. Coarse-graining of multi-component systems: effective interactions 101

• is always attractive within the AO-model.

• is purely entropic (proportional to kB T ).
• is also derivable from a superposition approximation by integrating unbalanced forces.
• is an important interaction that drives particle aggregation or even phase separation

9.3.2 DLVO interaction (Hamaker and Debye-Hückel)

(Derjaguin, Landau, Verweg, Overbeck around 1940 [6]) DLVO developed an important concept
based on effective interactions to describe the stability of charged colloids in solution. With
‘stability’ it is meant that the colloids stay well dispersed in the solvent. If they become ‘unsta-
ble’, it means they cluster and flocculate, precipitating out of solution (a little like in heating
sour milk). This is often unwanted. DLVO showed that this stabilization can be controlled
by the salt concentration in solution. For low salt concentration, the repulsive electrostatic
interaction can then overcome the ubiquitous van der Waals attraction (which alone would lead
to clustering and instability).
The DLVO assumption is that the effective interaction between two charged macromolecules
(colloids) is given as a sum of effective van der Waals (vdW) and electrostatic potentials:

VDLVO (r) = VvdW (r) + Velec (r) (9.22)

Van der Waals (Hamaker) interaction

Figure 9.10: Homogeneous atoms with density τ in the sphere volumina V1 = V2 = V .

 6

The atom-atom van der Waals attraction is ∼ − Cr66 = − 4σ
r6
as in the Lennard-Jones potential
VLJ (r). If we assume that the constituting atoms are all homogeneously distributed over the
two colloids with radius R, we can simply add (integrate) all their interactions. This defines
the integration for the effective interaction
C6 τ 2
Z Z
3 00
VvdW (r) = − d r d3 r0 (9.23)
V V |r~00 − r~0 |6
for fixed center-to-center distance r between the spheres of same volume V . Hamaker (1936)
introduced a general formula for two differently sized spheres of radii R1 and R2 :

1 2R1 R2 2R1 R2 r2 − (R1 + R2 )2

!
VvdW (r) = −π τ C6 · 2
2 2
+ + ln (9.24)
6 r − (R1 + R2 )2 r2 − (R1 − R2 )2 r2 − (R1 − R2 )2

This equation goes ∼ r16 for r → ∞ (derivable through Taylor expansion), which makes sense
because far away the spheres look like atom-like points again and we obtain the usual form
of dispersion interactions. In the near-field for D  R = R1 = R2 , the equation of Hamaker
simplifies to:
R
VvdW (r) ' −A (9.25)
12D
Chapter 9. Coarse-graining of multi-component systems: effective interactions 102

where A = π 2 τ 2 C6 is the Hamaker constant, and D the surface-to-surface distance of the two
colloids.
The Hamaker integration is possible for all kind of geometries, see the Appendix. Typically we
use indexes Aijk to refer to interactions between i and k across a medium j as seen in Fig. 9.11.
The Hamaker Aijk are thus specific material constants and can be tabulated (see Appendix).

Figure 9.11: Interaction between i and k across medium j.

As a special geometry we also consider a wall (half-space) interacting with an atom. This is
important for simulations to define effective interactions between the simulated particles and
a confining wall. The Hamaker interaction between an atom and a planar surface as seen in
Fig. 9.12 is:
2πhτ
Z −∞ Z
VvdW (z0 ) = dz dh 2 (9.26)
z0 (z + h2 )6

Figure 9.12: Hamaker interaction between a planar surface of homogeneous density τ and an
atom.

The result of this Hamaker integration is of the ‘9-3’ form:

1 1
9 3 !
σ σ
 
V93 (z0 ) = 8πτ σ 3 − (9.27)
z0 90 z0 12

This potential is very suitable to model coarse-grained atom-surface interactions, e.g., for liquids
confined between smooth planar surfaces.
Chapter 9. Coarse-graining of multi-component systems: effective interactions 103

Electrostatic interactions

Figure 9.13: Charged colloids and salt in a water solution.

Now we consider two charged colloids with charge Qi = Zi e, i = 1, 2 (with e the elementary or
electron charge) as in Fig. 9.13. Typically, these colloids are immersed in water and salt (mobile
ions) with salt concentration Cs which screen the electrostatic interactions between the colloids.
The system can be coarse-grained by integrating out the water and salt contributions. Often
for the water, the coarse-graining is not done explicitly, only a continuum background with
an appropriate dielectric constant r is assumed. This is a good approximations if the colloids
are large, much larger than water molecules. Then, one can define the important Bjerrum
2
length λB = 4π0ekB T . The Bjerrum length indicates the separation at which the electrostatic
interaction between two elementary charges is comparable in magnitude to the thermal energy
scale. It is for example about 0.7 nm for water at normal conditions. Electrostatic theories,
like Debye-Hückel or Poisson-Boltzmann (discussed briefly in the simulation part), then lead to
the effective interaction between the colloids written as

Z1 Z2 λB e−κD (r−2R)
βVelec (r) = (9.28)
(1 + κD R)2 r
√ 1/2
with κD = λ−1
D = 8πλB Cs being the inverse screening length which scales ∝ Cs with the salt
concentration Cs . The latter has units per nm3 or mol/liter in chemistry. Eq. (9.28) shows
that the Coulomb interaction between the colloids is exponentially screened by the ions. The
strength and range of the electrostatic interaction can thus be tuned by the salt concentration.
The complete DLVO term is typically given as:
AR λB −κD r
VDLVO (r) = − + Z̃ 2 e , (9.29)
12(r − 2R) r
κ 2R
where a proper derivation [6] shows that the valencies are effective ones, given by Z̃ = (1+κ
Ze D
D R)
2.
The reason is that for highly charged colloids, some ions condense on the surfaces of the colloids
and renormalize the colloidal charge, so that the colloids are seen far away with effective,
reduced charges. As we see in Fig. 9.14 the salt now tunes the attraction versus repulsion of
the interaction and can thus stabilize/destabilize the dispersion.
Chapter 9. Coarse-graining of multi-component systems: effective interactions 104

Figure 9.14: The complete DLVO potential plotted over the distance r. The salt concentration
Cs tunes the magnitude and strength of the electrostatic interaction. The total interaction can
thus range from very repulsive (low salt) to very attractive (high salt).
Chapter 10

Polyatomic molecules and polymers

10.1 Introduction

10.1.1 Molecules
Until now we mostly talked about atoms. But obviously complex materials consists of polyatoms
(molecules) and polymers. Examples for molecules often occurring in complex systems are:
O

Water (H2 O) H H
H H
H

Ethanol (C2 H6 O) H C C O

H H
H H
 
 
 
n
 
Alkanes 
 C C 
 
 
 
H H
A good description of the statistical mechanics of small molecules (e.g., partition sums of
vibrations and rotations) can be found in the book by McQuarrie [9]. (We could discuss this
deeper in this lecture on demand by students). Resulting force fields for simple molecules we
will discuss later in Part II. Water, for example, can be built by three charged atoms (LJ spheres
plus a Coloumb partial charge) which are connected by rigid bonds of fixed length. Since bond
vibrations are very fast and small scale the rigidity is a good assumption for water in a larger
scale simulation containing thousands of water molecules.

10.1.2 Polymers
A good overview of polymer physics can be found in the book by Rubinstein and Colby [10].
Everything found in this script can be found there in more detail, and of course many more
things.
The term "polymers" is derived from the greek words "poly" and "mers" meaning ‘many parts’.
The elementary unit of a polymer is a so-called "monomer", a small molecule that is capable of
polymerizing into a large molecule, e.g.,

105
Chapter 10. Polyatomic molecules and polymers 106

(A) (A) (A) (A) (A) ... (A) = (A)n (A)n+1

where (A) denotes the monomer and n the degree of polymerization which ranges typically
between 102 ≤ n ≤ 104 . A polymer with n ≤ 20 is called "oligomer".

If only one form A of monomer occurs it is referred to as a "homopolymer" An . Apart from

that, many other forms exist, such as "copolymers" having two units of monomers A and B:

Alternating copolymer A B A B A B A...

Random copolymer A A B A B B A B A A...
Di-block copolymer (A)n (B)m
Tri-block copolymer (A)n (B)m (A)l

Notes:
Polymers with three different units are referred to as "terpolymers". Proteins have 20 different
monomers whereas DNA has four different monomers (GTAC).
Also other architectures such as "grafted polymers" are possible:

A A∗ A A∗ A

B B

B B

B B

10.1.3 Structures and architectures

In general, many different structures and architectures are possible as seen in Fig. 10.1 and
10.2.

Figure 10.1: Different possible structures and architectures of polymers. 1

Chapter 10. Polyatomic molecules and polymers 107

Figure 10.2: Structure of a dendrimer, derived form the greek word "dendron" meaning tree. 2

10.1.4 Example of specific polymers

Polyethylene
h i
CH2 CH2 CH2 CH2 n

Polyvinyl chloride (PVC)

 
CH2 CH CH2 CH n
 
 
 
Cl Cl

Polypeptide (protein)

Figure 10.3: Polypeptide consisting of monomers with a backbone of an amino acid and specific
side chains of the polymer Ri .

1
https://en.wikipedia.org/wiki/Polymer_architecture#/media/File:RAFT_Architecture.png
2
https://en.wikipedia.org/wiki/Polymer_architecture#/media/File:Graphs.jpg
Chapter 10. Polyatomic molecules and polymers 108

10.1.5 Definitions

In this course, we are focusing on conformation of ideal chains. In ideal polymers, there are no
interactions between the monomers.

Figure 10.4: Bond vector ~ri , bond angle θj , torsion angle φk and unit tangent vector ~e(s) where
s denotes the arc length for a given polymer.

(i)
First, we introduce the bond vector ~ri with a bond length of lb = |~ri | as seen in Fig. 10.4. In
polyethylene, the bond length is lb ' 1.54Å.

θi denotes the bond angle which has a value of θ = 68◦ for polyethylene and φi denotes the
torsion angle. Typically, a change of φ is accompanied by a change in energy as seen in
Fig. 10.5.

Figure 10.5: The energy landscape of a polymer depends on the torsion angle Φ.

The typical energy difference due to the torsion angle at the Gauche conformation for polyethy-
lene is ∆ ∼ 0.8kB T .
A change of the torsion angle φi leads to a change of flexibility of a chain on large scales. The
flexibility or stiffness of a chain is characterized by the "persistence length" lp which describes
the decay length of directional correlations. Typically, correlations decay exponentially:
|s0 −s|
−
D(|s0 − s|) = h~e(s)~e(s0 )i = e lp
(10.1)

Some typical persistence lengths are:

• lp ' 1 nm for unfolded proteins or single-stranded DNA

• lp ' 50 − 70 nm for double-stranded DNA

• lp ' 10 − 20 µm for actin filaments

Chapter 10. Polyatomic molecules and polymers 109

The stiffness/flexibility of a polymer is obtained by comparing the persistence length to

the contour length which can be defined as the end-to-end vector, R ~ n , given by

n Z lc !
~n = cont. limit of R
~n = ~e(s)ds (10.2)
X
R ~ri
i=1 0

with the maximum |R

~ n |max = lc being the contour length as seen in Fig. 10.6.

Figure 10.6: Contour length of a simple chain: For a constant bond length, lb , and a constant
bond angle, θ, the contour length is given as lc = nlb cos 2θ .

A polymer is stiff if lc . lp and flexible if lc  lp .

How do we define the size (conformational length) of a polymer?
n
~ ni = h~ri i = 0 (in isotropy), (10.3)
X
hR
i=1

but the mean-square is non-zero:

~2i = h ~rj i = h~ri2 i + h~ri · ~rj i = nlb2 + (10.4)

X X X X X
hRn ~ri hcos θij ilb2
i j i i6=j i6=j

where the second contribution depends on the angular correlations (considering small θ). If the
angular correlation is zero, the equation for the mean square reduces to:
q
√
~2i =
hR nlb (10.5)
n

This is however not a good measure for ring or branched polymers.

We define the radius of gyration Rg :
n
~ 2i = 1
Rg2 = hR ~ com )2 i
~i −R (10.6)
X
g h(R
n i=1

The radius of gyration denotes the average square distance of monomers at position R
~ i to the
polymer center-of-mass as seen in Fig. 10.7 where the center-of-mass is

~ com = 1 ~i (10.7)
X
R R
n i
Chapter 10. Polyatomic molecules and polymers 110

Figure 10.7: The center-of-mass of the polymer R

~ com in relation to the distances of the monomer
~ i.
R

The radius of gyration can be rewritten as:

1 X ~2
Rg2 = hRi − 2R ~ com + R
~ iR ~2 i
com
n i
1 X ~2 ~i X
R ~j + 1
= hRi − 2 ~ jR
~ ki
X
R R
n i n j n2 j,k
1 X ~2 1 X ~ ~
=h R − Ri Rj i
n i i n2 i,j
1 X ~ 2
= ~j i
h Ri − R
2n i,j
2

1 X ~ 2
= h R i − ~j i
R (10.8)
n2 i<j
 2  2
whereas 1 ~ ~
i,j h Ri − Rj i= ~ ~
i<j h Ri − Rj i holds due to the symmetry in i and j.
P P
2

10.2 Chain models

10.2.1 Freely jointed chain (FJC) model

Figure 10.8: The freely jointed chain model with a freely fluctuating bond angle θ and an
arbitrary torsion angle φi .

In this model, the bond angle is freely fluctuating as seen in Fig. 10.8 which leads to no bond
correlations, e.g. hcos θij = 0i for i 6= j.
The root-mean-square end-to-end distance is then given by
q s
√
~2i = nlb2 + hcos θij lb2 i = (10.9)
X
hRn nlb
i6=j
Chapter 10. Polyatomic molecules and polymers 111

indicating a random walk behavior.

For the FJC, the contour length is given as lc = nlb and the persistence length equals the bond
length lp = lb , leading to an end-to-end mean-square of
~ 2 i = lc lp .
hR (10.10)
n
R lc
The radius of gyration of the FJC model for the continuous limit ds results in:
Pn
i → 0
1 X ~ 2
Rg2 = 2 ~j i
h Ri − R
n i<j
1 lc
Z Z lc  2
→ ds ~ − R(s)
h R(t) ~ i
lc2 0 s
hR~ n2 i
... = (10.11)
6

10.2.2 Freely rotating chain (FRC) model

In contrast to the FJC model, the bond angles of the FRC model are constant θi = θ = const.
as seen in Fig. 10.9.

Figure 10.9: The freely rotating chain model with a constant bond angle θ and an arbitrary
torsion angle φi .

This leads to angular correlations:

h~ri · ~ri+1 i = cos θlb2
h~ri · ~ri+2 i = cos2 θlb2
.
.
h~ri · ~rj i = lb2 cos|i−j| θ (10.12)
The mean-square end-to-end distance is given by
n n n−i
~ n2 i = h~ri · ~rj i = (10.13)
X X X
hR h~ri · ~ri+k i
i,j=1 i=1 k=−i+1

where k = j − i. For large n, we can neglect end effects:

∞ X
∞
~2i ' lb2 cosk θ (10.14)
X
hRn
i=1 k=−∞

This inner sum is a geometrical sum resulting in

∞ ∞
1 1 + cos θ
!   
lb2 cos2 θ = 1+2 cos θ = 1+2 −1 = lb2 (10.15)
X X
lb2 2
lb2
k=−∞ k=1
1 − cos θ 1 − cos θ
Chapter 10. Polyatomic molecules and polymers 112

The mean-square end-to-end distance of the FRC can thus be approximated as:

~ n2 i = nlb2 1 + cos θ
hR (10.16)
1 − cos θ

Another way of denoting the mean-square ~ n2 i = n˜l2 with ˜lb being the
end-to-end distance is hR b
˜
q
effective segment length lb = lb 1+cos θ
.
1−cos θ
Comparing both the FJC and the FRC model, we see:

FJC: ~ n2 i = nlb2
hR lc = nlb
~ 2 i = nl2 1 + cos θ θ
FRC: hR lc = nlb cos (10.17)
n b
1 − cos θ 2

In a more general way, the mean-square end-to-end distance can be denoted as

n→∞
~ n2 i = nlb2 Cn → nlb2 C∞
hR (10.18)

where C∞ is called the (Flory) characteristic ratio of the FRC. It has an n dependence for finite
chains because of end-effects but converges quickly to the limiting value C∞ for increasing chain
length. Introducing the Kuhn length, lk , the mean-square end-to-end distance can be rewritten
to
~ n2 i = nlb2 C∞ = N lk2
hR (10.19)
with N being the renormalized degree of polymerization, such that the contour length is lc =
N lk . This way {N, lk } in the FRC model equals the {n, lb } in the FJC model. This is a very
general finding in polymer physics. Polymers are self-similar, i.e., fractal objects with universal
scaling behavior of conformations where microscopic DOFs (small scales) can be renormalized
away. The Kuhn length is thus synonymous for the effective bond length, irrespective of any
specific microscopic model details.
With the contour length, lc = N lk , given, the Kuhn length can be determined:
~ n2 i
hR hR~2i C∞
lk = = n
= lb (10.20)
lc nlb cos 2
θ
cos 2θ

examples C∞ lk [Å] lb [Å]

polyethylene '7 ' 12 1.5
poly alanine '9 ' 36 4
DNA (double) ' 90 ' 300 ' 3.5

Table 10.1: Kuhn lengths and bond lengths for example polymers

The persistence length lp of the FRC model can be derived via:

h~ri · ~rj i = lb2 cos|j−i| θ

( )
|j−i| |j − i|
cos θ = exp{|j − i| ln cos θ} = exp − (10.21)
sp

with sp = − ln cos
1
θ being the persistence segment. The persistence length is then defined as

lb π
lp = lb · sp = with θ≤ (10.22)
| ln cos θ| 2
Chapter 10. Polyatomic molecules and polymers 113

Note: In literature, the stiffness is often defined as

˜lp = h ~r1 · R
~ ni
|~r1 |
n n
1
= ~ri i = lb (cos θ)i−1
X X
h~r1 ·
lb i i=1
1 − (cos θ)n lb
= lb → ˜lp = (10.23)
1 − cos θ n→∞ 1 − cos θ

10.2.3 Hindered rotation model

We consider freely rotating chains and a torsion angle φi that has a probability distribution of
P (φi ) = φ0 e−βU (φi ) .

Figure 10.10: The energy landscape U (φ) as a function of the torsion angle of the polymer φi .

~ n2 i = nlb2 C∞ 1 + cos θ 1 + hcos θi

hR with C∞ = (10.24)
1 − cos θ 1 − hcos θi
The term hcos θi is meant to be the ensemble average of the probability distribution.

10.2.4 The Gaussian (or Rouse) chain

Figure 10.11: The Gaussian chain model with harmonic springs between the monomers.

The Gaussian chain is a model of monomers connected to each other via a harmonic spring as
seen in Fig. 10.11.
The next neighbour potential is given as:

U (R ~ i+1 ) = 1 k(R
~i −R ~ i+1 )2
~i −R with k=
3kB T
(10.25)
2 lb2

In the Gaussian chain, the mean square distance equals the squared bond length: h~ri2 i = lb2 .
This results in the mean-square end-to-end distance being random walk like:
~ n2 i = nlb2
hR (10.26)

All distance distributions of the model are of Gaussian nature.

Chapter 10. Polyatomic molecules and polymers 114

End-to-end distributions
The probability distribution φ(R
~ ee , n) that the end-to-end distance of n segments is |R
~ ee |:

1 sin klb
Z  n
~ ee , n) = 3 i~k·R
~ ee
φ(R dk e (10.27)
(2π)3 klb

with lb being the effective segment length. For long chains n → ∞:

!3/2
3 3R
~2
( )
~ ee , n) '
φ(R exp − ee2 (10.28)
2πnlb2 2nlb

While being an approximation for the FJC and FRC model, it is the exact solution for the
Gaussian chain model.
Note that the Gaussian probability distribution implies that the corresponding free energy
is harmonic. As a result, the stretching response of the polymer (pulling the polymer by a
constant force) is linear like Hook’s law for simple linear elasticity, with a spring constant
given by kB T /(nlb2 ). For ideal chains and low forces, this is indeed what one experimentally
observes [10]. However, for non-ideal (real) chains and large stretching forces the response is
very different. [Check our our computational lab "nano-tug-of war" https://www.compmat.
uni-freiburg.de/Nano_Tug-of-War].

10.2.5 Form factor

The form factor of an ideal polymer (of non-interacting monomers), is defined as:
2 −x
F (k) = N · f (k 2 Rg2 ) with f (x) = e −1+x (10.29)


x 2

with f (x) being the Debye function. For extreme values of kRg , the form factor can be simplified
to   
2 2
N 1 − k R g

if kRg  1
F (k) = 3
(10.30)
2N/k 2 R2

if kR g  1
g

with Rg being the radius of gyration, which is directly accessible from low- temperature scat-
tering of dilute solutions.

10.2.6 Real chains

In reality monomer-monomer interactions are present, e.g., excluded volume and dispersion
attraction etc. Instead of the random walk behavior of idealistic polymers, we see "self avoiding
walk" behavior or polymers or attractions that can even lead to collapses of the chain to a
compact globule (like in protein folding).
We recall the virial expansion:
βP 1 1
Z Z  
= 1 + B2 ρ + ... with B2 = − d3 rf (r) = − 4πr2 dr e−βV (r) − 1 (10.31)
ρ 2 2

where V (r) denotes the effective monomer-monomer pair potential. We define the effective
interaction volume as being 2B2 = v.
Typically, the effective monomer-monomer pair potential is split up in a repulsive and an
attractive part:
V (r) = Vrep + Vatt (10.32)
Chapter 10. Polyatomic molecules and polymers 115

Z  
B2 = d3 r 1 − e−βVrep e−βVatt (10.33)

Small attraction terms can be approximated as:

Vatt
 
−βVatt
e ' (1 − βVatt ) = 1 − (10.34)
kB T
which leads to
B
v = 2B2 = A − (10.35)
T
where the terms A and B are temperature independent. A is the hard core contribution
(entropic) whereas B is the attractive contribution (energetic). We can rewrite by defining v0 ,
via
θ
 
v = v0 1 − (10.36)
T
Typically v0 ' lb3 with θ = B/A and v0 = A. When modeling with a Lennard-Jones potential,
then v0 ' σ 3 . As we see, for T = θ, the effective interaction volume vanishes, v = 0, and the chain
behaviour becomes ideal (on a 2-body level). Hence, θ is an important temperature or solvent
condition, because we can apply ideal chain statistics. The solvent at the θ temperature is called
θ-solvent. Hence, in a θ-solvent, Rn ∝ n1/2 . If the temperature is lower, we name the solvent
‘poor solvent’ because the polymer collapses and does not walk freely random. Here, rather
Rn ∝ n1/3 because the monomers roughly fill a compact sphere. For higher temperatures, we call
the solvent ‘good’, because the self-avoiding polymer swells and fluctuates happily around, more
extended than an ideal random walk. Here, one finds the famous Flory scaling Rn ∝ n3/5 [10].
These scaling laws have been fundamental to describe structure and dynamics of polymeric
systems in materials science and especially in biophysics.
Chapter 11

Free energy calculations and phase

diagrams

11.1 Free energy calculations and entropy

The free energy is not directly accessible as simple averages from simulations (i.e., trajec-
tory data), while it is fundamental to characterize the material properties, phases and states!
There are various more or less sophisticated methods to calculate F = U − T S available. These
methods are important to get the free energy from numerical data or as starting points for
approximative perturbation theories.
Let’s consider some of the standard ones:

11.1.1 Thermodynamic Integration

Consider the potential energy function of a known reference system U0 and the potential energy
function of a known target system U1 . Let’s introduce a parameter λ such that:

U (λ = 0) = U0 U (λ = 1) = U1 (11.1)

where both constraints are interpolates through a λ-path.

The Helmholtz free energy is given as:

F = −kB T ln Z (11.2)

with the partition sum being

1
Z
Z(λ) = d3N re−βU (λ) (11.3)
Ω3N N !
From the free energy, we can calculate the derivative with respect to the λ-parameter:

∂F (λ) 1 ∂U −βU (λ) ∂U

  Z  
= d3N r e = (11.4)
∂λ N,V,T Z ∂λ ∂λ λ

with the thermodynamic "force" ∂λ .

∂U
The energy difference is then given as:
Z λ=1
∂U
 
F (λ = 1) − F (λ = 0) = dλ (11.5)
λ=0 ∂λ λ

The notation h...iλ means that we calculate the ensemble average for a constant value of λ. For
example, if simply we interpolate linearly between two potentials, V (r, λ) = λV (r)+(1−λ)V0 (r),

116
Chapter 11. Free energy calculations and phase diagrams 117

with the reference pair potential V0 (r) and the target pair potential V (r), then we need to
average V (r) − V0 (r)|λ , i.e., the energy difference of the system interacting with V (r, λ) vs.
V0 (r) and integrate it up.
Another example, If λ = r (i.e., simply the distance between two particle), then ∂U ∂λ is simply
the mean force. Integration of the latter provides the free energy along r, also often called the
potential of mean force (PMF), see also section 9.1.
In practice, the λ-path should be chosen in such a way to be numerically convenient. This
can prove to be tricky for (discrete in λ) simulation sampling due to badly behaving U (λ)
(see Fig. 11.1 for example). λ could also be a physical ‘reaction coordinate’, see section 11.1.4
further below. Then one obtains from TI directly the ‘free energy landscape’ as a function of λ
which could be useful to study equilibrium populations and kinetics along λ.

Figure 11.1: An example for a badly behaving sample of the potential U (λ).

This happens for instance if λ = σ or λ =  for "blowing" up (solvation) of a Lennard-Jones

sphere in water (see Fig. 11.2)

Figure 11.2: Solvation of a Lennard-Jones sphere in water with the chemical potential µex being
the free energy difference.

Free energy of a solvation of a Lennard-Jones sphere:

Consider the free energy of charging a LJ atom in water to an ion of charge Q at ~rc ≡ 0.
NX
H2O
λQqi λB ∂UC
U = U0 + UC (λ) UC = = Φ(~r ≡ 0) − Φ (11.6)
i
|~r − ~ri | ∂λ
This results in a free energy F of:
Z
F = F0 + Q dλhΦ(~r ≡ 0))λ (11.7)
Chapter 11. Free energy calculations and phase diagrams 118

from the average of the electrostatic potential at ~r ≡ 0.

11.1.2 Particle insertion method

(Also called Widom’s insertion method). The particle insertion methods is a very useful and
convenient method to calculate the free energy of inserting a particle into a complex medium.
With that we can obtain the probability that a particle enters a systems. It is a convenient
method because it can done in a ‘post’processing way for many systems, i.e., it can be used
already with pre-stored available trajectory data of an ergodic equilibrium system.
We consider the excess chemical potential µex as the free energy of solvation when adding a
particle to the system (see Fig. 11.3):

µex = F ex (N + 1, V, T ) − F ex (N, V, T ) (11.8)

F ex = F − Fideal is the "excess" over the ideal gas.

Figure 11.3: Insertion of a particle to the reference system consisting of N particles.

Consider the excess chemical potential

V Z̃N
µex = kB T ln (11.9)
Z̃N +1

where Z̃N is the configurational integral for N particles:

exp{−βUN +1 (~r1 , ...,~rN +1 )}d3(N +1) r

R
Z̃N +1
= (11.10)
exp{−βUN (~r1 , ...,~rN )}d3N r
R
Z̃N
Now the potential energy equals

UN +1 (~r1 , ...,~rN +1 ) = UN (~r1 , ...,~rN ) + (~r1 , ...,~rN ) (11.11)

where  is the interaction energy of particle (N + 1) fixed at ~rN +1 ≡ 0 with all others ("test
particle limit").
exp{−β} exp{−βUN }d3(N +1) r
R
Z̃N +1
= (11.12)
exp{−βUN }d3N r
R
Z̃N
For a homogeneous solution, this reduces to:

exp{−β} exp{−βUN }d3N r

R
Z̃N +1 V
= = V he−β i (11.13)
exp{−βUN }d3N r
R
Z̃N
"Widom’s test particle insertion" approach:

µex = −kB T ln hexp{−β}i0 (11.14)

Chapter 11. Free energy calculations and phase diagrams 119

where 0 is in reference to N particles of the system. In practice, one does average over many
random insertions of "test particles" into the ensemble of trajectories of N particles as seen in
Fig. 11.4. The ensembles can be generated before, hence the whole procedure can be performed
as post-processing. This is particularly convenient if one wants to sample different particles
(i.e. solutes of different size or charge): one can always use the same pre-generated trajectories
of the underlying liquid.

Figure 11.4: Insertion of many test particles to a reference system consisting of N particles.

11.1.3 Free energy perturbation approach

Consider the perturbation of a reference Hamiltonian H1 by ∆H:

H1 = H0 + ∆H (11.15)

This results in a difference in free energy of:

d3N re−β(H0 +∆H)

R
Z1
−β∆F = ln = ln R 3N −βH
Z2 d re 0

= ln hexp{−β∆H}i0 = ln hexp{−β∆U }i0 (11.16)

| {z }
for equal masses
∂U0
 
' −βh∆U i ∼ (11.17)
| {z } ∂λ ∆λ
β∆U 1

This method is directly related to the Thermodynamic Integration as well as the Particle
Insertion for ∆U = . The average of quantity A in the perturbed system is calculated as:

d3N rAe−β(H0 +∆H) hA exp{−β∆H}i0

R
hAipert = R 3N −β(H +∆H) = (11.18)
d re 0 hexp{−β∆H}i0

One might wonder why this is a ‘perturbation’ approach because Eqs. (11.16) and (11.18)
are in principle exact for every perturbation. Practical problems in evaluating the integrals
appear in fact for larger perturbations: the significant contributions of phase space of the per-
turbed systems then lie in the only very rarely samples tails of the equilibrium distribution.
Hence, a FEP simulation with large perturbations will in practice never converge. This also
relates to important sampling in Monte-Carlo where Markov chains are generated to move in
phase space to well represent the to-be-sampled system (and not just randomly throwing darts).

A good book for more details and more free energy calculation methods is Chipot and Po-
horille, "Free energy calculations", Springer.

Example: Charging free energy of a sphere of radius R

Chapter 11. Free energy calculations and phase diagrams 120

∆U (0 → q) = Φ0 q (11.19)
with φ0 being the electrostatic potential at q.
∆F = −kB T ln hexp{−β∆U i0 }
β2
' −kB T ln h1 − β∆U + ∆U 2 i
β∆U 1 2 0
β β
' h∆U − ∆U 2 i0 = hφ0 i0 q + hφ20 i0 q 2 (11.20)
2 2
Thus if we calculate the potential and fluctuations in the reference system, we get the charging
free energy for small charges q.

11.1.4 Reaction coordinate

An important concept for free energy calculations, and thus characterizing a complex system,
is the projection of the many degrees of freedom of the original system to a very small number
(low dimension) of coordinates, so called Reaction Coordinates.
A simple example of such a reaction-coordinate would be the end-to-end distance of a random-
walk (or ideal polymer), where many different realizations of the walk can lead to the same
distance. Another prominent example is the distance of a molecule to a protein binding pocket
in water, where also many solvation and protein conformational states are possible for a fixed
distance.
Consider the N V T ensemble. The partition sum Z, leading to F = E − T S = −kB T ln Z, sums
over all states Γ in phase space. Now project the states on a ‘collective variable’ or ‘reaction
coordinate’ q̂(Γ). Many states {Γ}q can lead to the same value q.
Formally, the probability of finding a state Γ in N V T is P (Γ) = exp{−βH(Γ))/Z}. Then the
probability of finding q is:
1
Z
P (q) = dΓ exp{−βH(q̂)}δ(q − q̂(Γ)) (11.21)
Z
The Dirac-δ function serves as a projection operator, collecting all states with prescribed value
q = q̂. Hence, we can write P (q) = hδ(q − q̂)iN V T . Note that from this follows the Boltzmann
entropy of the ensemble of value q: S(q) = +kB ln P (q) + S0 , and the total entropy
Z
S = kB dQP (q) ln P (q) + S0 = kB hln P (q)iP + S0 (11.22)

is of Gibbs/Shannon form and we require normalization dqP (q) = 1. The entropy difference
R

between two distributions is then

P (q)
∆S(q) = kB ln P (q) − kB ln P0 (q) = kB ln (11.23)
P0 (q)
The total ‘relative entropy’ follows from:
P (q)
Z
∆S = kB dqP (q) ln (11.24)
P0 (q)
Also the energy can be projected on q, via
1
Z
E(q) = dΓH(q̂) exp{−βH(q̂)}δ(q − q̂(Γ)) = hH(q̂)δ(q − q̂)iN V T (11.25)
Z
F (q) = E(q) − T S(q) is then the ‘free energy landscape’ along reaction coordinate q. The F (q) is
also often called the ‘potential of mean force’ (PMF), because it can be obtained from integrating
a mean thermodynamic force along the q-pathway like we showed in the λ-integration in the
Thermodynamic Integration (TI) approach above. (We will also see this in the next chapter
when we define effective potentials which are also free energies along distance coordinates.)
F (q) is important but not always easy to get!
Chapter 11. Free energy calculations and phase diagrams 121

11.1.5 Importance sampling & Umbrella sampling

Now consider the free energy along a "reaction coordinate" or "order parameter" q. The free
energy along q is then calculated as:
1
Z
F (q) = −kB T ln P (q) = kB T ln d3N δ(q − q̂{~r1 , ...,~rN })e−β{~r1 ,...,~rN } (11.26)
Z
where P (q) is the probability distribution and the function is integrated over all {~ri } for which
q = q̂{~r1 , ...,~rN } with q̂ = i qi δ(~r − ~ri ).
P

Figure 11.5: Sampling unlikely states can prove to be challenging as they are very sensitive to
q.

Importance sampling

P (q)
∆F (q) = F (q) − F (q0 ) = −kB T ln (11.27)
P (q0 )

Trδ(q − q̂({~ri }))e−βU ({~ri })

P (q) = (11.28)
Tre−βU ({~ri })
with the trace being Tr = Ω3N1 N ! d3 r1 ... d3 rN . If we add a biasing potential V (q̂) to facilitate
R R

sampling, we get an adjusted probability distribution of

Trδ(q − q̂)e−β(U +V (q̂)) e−βV (q) Trδ(q − q̂)e−βU

P 0 (q) = = (11.29)
Tre−β(U +V (q̂)) Tre−β(U +V (q̂))
The calculation of P 0 (q0 ) is analogous and the solution between P 0 (q) and P 0 (q0 ) is then:

e−βV (q) P (q) 0

P 0 (q) = P (q0 ) (11.30)
e−βV (q0 ) P (q0 )

Applying this formula to Eq. (11.27), we finally get:

P 0 (q)
 
∆F (q) = −kB T ln 0 + V (q0 ) − V (q) (11.31)
P (q0 )

The desired free energy function can be obtained by sampling the "non-Boltzmann" distribution
P 0 (q) and subtracting the bias V (q) afterwards. A challenge is the fact, that a good choice of
V (q) is a priori not known. Often the system is piecewise constrained with an "umbrella
potential" (typically harmonic) at values qi as seen in Fig. 11.6.
Chapter 11. Free energy calculations and phase diagrams 122

Figure 11.6: Umbrella sampling by piecewise constraining the system with umbrella potential
at qi .

The sampling is done over all overlapping umbrella windows. Using Eq. (11.31) for every
window and matching overlapping branches, the free energy F (q) can be reconstructed. (In
practice, one would use WHAM 1 = weighted histogram analysis method)
The fact that equidistant windows may not be efficient or sufficient may be challenging and the
procedure needs iterations (adding stiff windows with long sampling) to obtain the accurate
F (q)!

11.2 From free energies to phase diagrams

Figure 11.7: The free energy per volume f = F/V in dependence of the density ρ at different
temperatures T3 < T2 < T1 .

Free energies of a system can be used to calculate a system’s stability, coexisting phases, and
phase diagrams. For this, we define the free energy per volume:
F Fid Fexc
f= = + = fid + fex (11.32)
V V V
For hard spheres (being in the following a reference ‘0’ system) this is given as:
  6 η 2 (4 − 3η)
0
βfexc = ρ ln Λ3 ρ − 1 + (11.33)
πσ 3 (1 − η)2
| {z }
Carnahan-Starling eq. of state

1
https://www.youtube.com/watch?v=E8gmARGvPlI
Chapter 11. Free energy calculations and phase diagrams 123

with σ being the diameter of the sphere, η = π6 ρσ 3 being the packing fraction and ρ = N
V being
the density.
Now we consider a pair potential of

Vλ (r) = Vλ0 (r) + Wλ (r) (11.34)

with the switching parameter λ like in the TI approach we introduced above. Let Vλ0 be our
known reference system and Wλ be some (attractive) potential like for example a van-der-Waals
or depletion potential. The energy U (λ) of the system is:

U (λ) = Vλ (|~ri − ~rj |) (11.35)

X

i<j

Applying the thermodynamic integration

∂βFexc 1
Z Z
= d3 r1 ... d3 rN e−βU (λ) βU 0 (λ) = βhU 0 (λ)iλ (11.36)
∂λ Z(λ)

we get the ensemble average for a system interacting with Vλ with U 0 (λ) = ∂λ .
∂U

How we use λ as a switching parameter should not be important, free energy differences are
not path-dependent. Hence, we can chose the λ-integration as convenient as possible: let us
assume simply that Wλ (r) = λW (r), i.e., λ switches on the potential linearly:

Vλ (r) = V0 (r) + λW (r) (11.37)

with λ0 = 0 and λ1 = 1. This results in

∂U X
U 0 (λ) = = W (|~ri − ~rj |). (11.38)
∂λ i<j

The integration of Eq. (11.36) yields:

Z 1
βFexc (λ = 1) = +β dλh W (|~ri − ~rj |)iλ (11.39)
X
0
βFexc
0 i<j

The Taylor expansion up to the first order of U 0 (λ) is given as:

∂
hU 0 (λ)iλ ' hU 0 (λ)i0 + (λ − λ0 ) hU 0 (λ)i0 + O(λ2 ) (11.40)
∂λ
with

d rU (λ)e−βU (λ)
R 3 0
0
hU (λ)i = (11.41)
d re−βU (λ)
R 3

∂  2

hU 0 (λ)i = hU 00 (λ)i − β hU 0 (λ)i − hU 0 (λ)i2 (11.42)
∂λ
Applying the Taylor expansion to Eq. (11.39), we get the high temperature expansion:
 
1
βFexc (λ = 1) = βFexc + β h W (|~ri − ~rj |)i0 − β h W (|~ri − ~rj |)2 i0 − h W (|~ri − ~rj |)i20 
X X X
0

i<j
2 i<j i<j
| {z } | {z }
mean of perturbation variance of perturbation
(11.43)
Chapter 11. Free energy calculations and phase diagrams 124

The Fexc
0 term can be identified with the hard sphere reference system, which we perturb by

W . The pairwise additivity yields:

XZ ρN
Z
W (|~ri − ~rj |)iλ = h d rδ(r − |~ri − ~rj |)W (r)iλ = d3 rgλ (r)W (r) (11.44)
X
3
h
i<j i<j
2

with gλ (r) = g(r) for a system interacting with Vλ . This results in:
Z 1
βρN
Z
βFexc = βFexc
0
+ dλ d3 rgλ (r)W (r) (11.45)
2 0

Also the pair density can be expanded in powers of λ:

(2)
∂ρλ (r)
!
(2) (2)
ρgλ (r) = ρλ (r) = ρ0 (r) + λ + O(λ2 ) (11.46)
∂λ λ=0

First-Order perturbation theory:

0
βFexc βFexc βρ2
Z
' + d3 rg0 (r)W (r) (11.47)
V V 2

Let’s now assume that W (r) is a long-ranged potential and its contribution dominates over the
correlations of g(r). We can neglect the effect of correlations. This mean-field approximation
then yields:
0
βFexc βFexc βρ2
Z
βfexc = ' + d3 rW (r)
V V 2
0
βFexc
= + βρ2 a (11.48)
V
where the parameter a < 0 for an attractive potential W (r). Thermodynamics (P = −∂F/∂V )
tells us that the equation of state can be obtained from fexc = Fexc /V = (F − Fid )/V through

βP ∂ βfexc
−1 = ρ (11.49)
ρ ∂ρ ρ

For hard spheres of diameter σ and small packing fractions η = (π/6)σ 3 ρ, the Carnahan Starling
expression Eq. (11.33) reduces to the second virial contribution only, and Fexc 0 ' (2/3)πσ 3 ρ.

Inserted in Eqs. (11.48) and (11.49) one can bring the latter (by using 1/(1 − x) ' 1 + x for
small x) into the form:
βP 1
' + aβρ (11.50)
ρ 1 − 3 πσ 3 ρ
2
| {z }
4η=bρ

Finally, substituting v = ρ−1 and rearranging, we get the real gas equation (van-der-Waals
equation of state):
a
 
P + 2 (v − b) = kB T (11.51)
v

In general, attraction can lead to concave paths in the free energy in dependence of the density,
−1
ρ, meaning f 00 (ρ) ≤ 0. But as χT ∼ f 00 (ρ), this can lead to a diverging compressibility for
f 00 (ρ) = 0 and thus instability! The loci of f 00 (ρ) = 0 defines the spinodal line.

Equilibrium for coexisting phases: (T, µ, P = − VΩ )

Chapter 11. Free energy calculations and phase diagrams 125

Let’s assume two states in mechanical, chemical, and thermal equilibrium:

P1 = P2 µ1 = µ2 T1 = T2 = T (11.52)

The pressure P and the chemical potential µ can be written in terms of f = F

V = f (ρ):

∂F ∂  N
     
P =− =− Vf N V = −f − V f 0 (ρ) − 2 = −f − ρf 0 (11.53)
∂V N,T ∂V N,T V
∂F ∂
    
µ= = Vf NV = f0 (11.54)
∂N T,V ∂N T,V

The conditions imposed on the free energy f (ρ) are:

f10 (ρ1 ) = f20 (ρ2 ) f1 (ρ1 ) − ρ1 f10 (ρ1 ) = f2 (ρ2 ) − ρ2 f20 (ρ2 ) (11.55)

This implies that for the same slope f 0 and the same f -axis interception point, there are two
coexisting densities ρ1 and ρ2 . With the so-called Maxwell’s double-tangent construction, ρ1 , ρ2
define the binodal line (see Figs. 11.7 and 11.8).

Figure 11.8: Phase diagram with binodal and spinodal curves. There may be coexisting states
indicated by the horizontal lines with a density difference ∆ρ 6= 0 (e.g. ∆ρ = ρ2 − ρ1 for the red
line between liquid and vapor).
Chapter 12

Appendix: Introduction Talk

126
 PHYSIKALISCHES KOLLOQUIUM
PHYSIKALISCHES
Theoretical KOLLOQUIUM
ANTRITTSVORLESUNG
Condensed Matter Physics
AM 12. NOVEMBER 2018 UM 17 UHR C.T.
A
IM
NTRITTSVORLESUNG
Lecturer:
GROßENJoachim
HÖRSAALDzubiella
12. NOVEMBER
AM Tutors: 2018
Sebastien Groh and
UM 17 U
Sven HR C.T.
Pattloch
IM GROßEN HÖRSAAL

Advanced Physics Course, Institute of Physics, Freiburg, SoSe 2024

COMPUTATIONAL PHYSICS OF SOFT AND FLUID MATERIALS

PROF. DR. JOACHIM DZUBIELLA

PHYSIKALISCHES INSTITUT, UNIVERSITÄT FREIBURG
Macromolecules and complex fluids serve as integral components in the development of modern
soft and 'smart' functional materials, e.g., for applications in medicine, liquid phase catalysis, or
COMPUTATIONAL PHYSICS OF SOFT AND FLUID MATERIALS
energy storage devices. From a statistical physics perspective these liquids are typically mod-
elled as classical, interacting many-body systems of varying complexity. While the structural and
thermodynamic properties of a ‘simple’ liquid can still be studied by analytical (statistical me-

PROF. DR. JOACHIM DZUBIELLA

chanics) theory to a wide extent, applied fluid materials display much higher complexity and one
typically resorts to particle-resolved (molecular) computer simulations for an accurate treatment.
Here, the challenge is to set-up and analyse these 'computer experiments' in a meaningful way

PHYSIKALISCHES INSTITUT, UNIVERSITÄT FREIBURG

and, if possible, to reduce and interpret the large amount of generated data by simple and trans-
parent models. In this way, computer simulations establish an important bridge between theoret-
ical physics and applied material science, providing useful structure-property-function relation-
ships for the rational design of new materials. In this talk I will introduce and discuss these ideas
Macromolecules and complex
by means of a few illustrative fluids
examples serve
of soft and as
fluidintegral
systemscomponents in the development
of varying complexity, i.e., start- of modern
ing from
soft and simple
'smart'liquids like hard-sphere
functional materials,colloids to fully
e.g., for atomistic models
applications of multi-component
in medicine, liquid phase catalysis, or
electrolytes
energy or polymers
storage relevant
devices. for modern
From applications
a statistical .
physics perspective these liquids are typically mod-
elled as classical, interacting many-body systems of varying complexity. While the structural and
thermodynamic properties of a ‘simple’ liquid can still be studied by analytical (statistical me-
 Condensed Matter Physics

Wikipedia:
Condensed matter physics is the field of physics that deals with the
macroscopic and microscopic physical properties of matter, especially the
solid and liquid phases which arise from electromagnetic forces between
atoms.

Condensed matter physicists seek to understand the behavior of these

phases by experiments to measure various material properties, and by
applying the physical laws of quantum mechanics, electromagnetism,
statistical mechanics, and other theories to develop mathematical models.

Most active field of contemporary physics; overlaps with chemistry,

materials science, engineering and nanotechnology, and relates closely to
atomic physics and biophysics.

https://en.wikipedia.org/wiki/Condensed_matter_physics
 Elements: periodic table

https://en.wikipedia.org/wiki/Periodic_table
 Elements: orbitals
• Chemical behavior of same ‚groups‘
(columns in table) similar
• Recall quantization and Pauli
exclusion: each orbital has 2
electrons
• Most important for interactions
‚valence electrons‘ in orbitals
• Complete shells are least reactive
elements, see noble gases

https://en.wikipedia.org/wiki/Periodic_table
 Elements: atomic characteristics
Atomic sizes Energy release by electron uptake

10-9 m = 1 nm = 1000 pm = 10 Å

Ionisation energy: energy required to remove electron

 Elements: bonds and interactions

Electronegativity: tendency to attract electrons in a bond

 But also noble atoms have interactions!

Example: Lennard-Jones interaction between two atoms

⇣ ⌘ ⇣ ⌘6
12
V(r) V (r) = 4✏
r r
12
/r
r

21/6s
s r
6
/r
e van der Waals (vdW)
attraction
Bonded / nonbonded / electrostatic atomic interactions
à Structure of molecules, solids and liquids
 Also complex molecules form crystals
(at sufficiently low temperature T)

C36H26
p-conjugated organic
molecule (COM) T = 300 K

Wiki: conjugated system is a system of connected p-orbitals

with delocalized electrons in a molecule, which in general lowers
the overall energy of the molecule and increases stability.
https://en.wikipedia.org/wiki/Conjugated_system
10
Simulated annealing
 Good reproduction of bulk (liquid) crystal
phase transitions in molecular simulations

cooling

heating

K. Palczynski, G. Heimel, J. Heyda, and JD, Cryts. Growth & Design 14, 3791 (2014).
 Interactions à Condensed (solid and liquid) phases

Simple phase diagram

https://en.wikipedia.org/wiki/Phase_diagram
Condensed (solid and liquid) phases

Schematic phase diagram of a

simple liquid, showing the
boundaries between solid (S),
liquid (L), vapor (V), or fluid (F)
phases.

(Hansen&McDonald, Theory of Simple

Liquids)
 Material properties

Material Properties
• Optical (Spectra, adsorption, etc.)
• Reaction / catalysis (Chemical rates, products, etc.)
• Mechanical (elasticity, durability, etc.)
• Flow / mass transport (Viscosity, etc.)
• Magnetic (Susceptibility, etc)
• Electrostatic (Permittivity, conductivity, etc.)
• Heat / energy (thermal conductivty, heat diffusion, etc.)
Depend on electronic / atomic / molecular structure

à Need for different methods on various scales

 Soft condensed Matter

Soft matter is a subfield of condensed matter comprising a variety of

physical systems that are deformed or structurally altered by thermal or
mechanical stress of the magnitude of thermal fluctuations. They include
liquids, colloids, polymers, foams, gels, granular materials, liquid crystals,
and a number of biological materials.

These materials share an important common feature in that predominant

physical behaviors occur at an energy scale comparable with room
temperature thermal energy. At these temperatures, quantum aspects are
generally unimportant.

Applications:
- Foams and adhesives, detergents and cosmetics, paints, food
additives, lubricants and fuel additives, rubber in tires, etc.
- Biological materials: blood, muscle, milk, yogurt, gelatine, etc.
- Liquid crystals are an important subfield (LCDs, etc.).
 Why 'soft' matter?

Equilibrium at temperature T : kB Boltzmann's constant (1.39 10-23 J/K)

Equipartition theorem: typical energy scale: kB T

typical length scale: a
typical surface tension ⇠ kB T /a2
3
typical elastic modulus G ⇠ kB T /a
a 0.3 nm much larger than atomic sizes
à macromolecular liquids are many orders of magnitude softer
than molecular liquids/crystals!

a >> thermal (de Broglie) wavelength lB < 0.1 nm

à no direct quantum effects, classical statistical mechanics
applies (equilibrium and nonequilibrium).
 2 Theory of Simple Liquids

Soft Matter and Classical mechanics

kB Boltzmann's constant (1.39 10-23 J/K)
Equilibrium at temperatureFIGURE
T : 1.1 Schematic phase diagram of a typical monatomic substance, showing the boundaries
between solid (S), liquid (L) and vapour (V) or fluid (F) phases.
Equipartition theorem: typical energy scale: kB T
✬ ✩
typical length scale: a
TABLE 1.1 Test of the classical hypothesis.

Liquid Tt (K) Λ (Å) Λ/a

a >> thermal (de Broglie) H2
Ne
14.1
24.5
3.3
0.78
0.97
0.26
wavelength L < 0.1 nm CH4 91 0.46 0.12
N2 63 0.42 0.11
Li 454 0.31 0.11
Ar 84 0.30 0.083
HCl 159 0.23 0.063
Na 371 0.19 0.054
Kr 116 0.18 0.046
CCl4 250 0.09 0.017

✫ ✪

à no direct quantum effects, classical statistical mechanics

t ≫ 10−14 s. This second condition is somewhat more restrictive than the first,
applies (equilibrium and nonequilibrium).
but where translational motion is concerned the problem is again severe only
in extreme cases such as hydrogen.
Use of the classical approximation leads to an important simplification
insofar as the contributions to thermodynamic properties arising from thermal
motion can be separated from those due to interactions between particles.
 Modeling classical many-body systems 1.1 F

4 An introduction to liquid matter P T

4 An introduction to liquid matter

Examples: Figure 1.2. Typical atomic

Liquid
Figure 1.2. Typical atomic
configurations in a gas configurations in a gas
(left), liquid (middle) and

4
crystalline solid (right).
(left), liquid (middle) and
configurations in a gas
Figure 1.2. Typical atomic
(left), liquid (middle) and
crystalline solid (right). crystalline solid (right).
L

g(r) 1
(a)

4 An introduction to liquid matter

• Metals / Simple atomic solids 0.5

1.5
0

2
0
1
Distance from centre of sphere - r/σ

Figure 1.2. Typical atomic

2

(e.g., solid argon, carbon)

configurations in a gas
2
2 Solid
(left), liquid3 (middle) and
V
3

3
crystalline 2.5 solid (right). 5
2.5 1.5 5
4
oscillations are rapidly damped, showing the gradual smearing out of short-range
figure 1.3(b); the maxima may be associated with shells of neighbours, but the
ameters, so that the radial distribution would behave as shown schematically in
correlated, leading to a modulation of ρ(r ) extending over a few molecular di-
In the liquid, however, the positions of neighbouring molecules are strongly

β=
diameter, and will be modulated by the Boltzmann factor exp(−βv(r )), where
density around a fixed molecule will rapidly vanish for r less than the molecular
will only depend on the centre-to-centre distance r ; even in a dilute gas, the local
In reality, molecules interact via a pair potential v, which, for spherical molecules,
density, as seen from any one fixed molecule, is everywhere equal to its average ρ.

as a function of the distance r from that molecule.

characterizes the modulation of the local density ρ(r ) around a given molecule,
scale is provided by the radial (or pair) distribution function g(r ), which

thermodynamic states are: T = 2ϵ/kB , ρσ 3 = 0.05 (gas), T = ϵ/kB , ρσ 3 = 0.8 (liquid),

superimposed on the gas phase distribution function (dashed curve). The
figure 1.4). The Boltzmann factor for the Lennard-Jones potential has been
section 1.6) of atoms interacting through a Lennard-Jones potential (equation (1.9),
These functions have been generated using a molecular dynamics simulation (see
Figure 1.3. Typical pair distribution functions for (a) a gas, (b) a liquid and (c) a solid.

An introduction to liquid matter

T = 0.2ϵ/kB , ρσ 3 = 0.9 (solid).

1.5 2 4
2 4

• Molecular liquids / solvents

In the limit of an ideal gas of non-interacting (point) molecules, the local

g(r) 1 g(r) 1.5 g(r) 3

Vapor
g(r) 1 g(r) 1.5 g(r) 3
5
1/kB T .

1 2
1 0.5 2
6

0.5 0.5 1
g(r) 1.5

0.5 1
(b)

0 0 0
In other words

0.5

2.5

0 1 2 3 4 5 6 0 2 4 6 0 2 4 6
(c)
0

0 0 (a) 0Distance from centre of sphere - r/σ (b)

• Ions and electrolytes

Distance from centre of sphere - r/σ Distance from centre of sphere - r/σ
0

0 1 2 3 4 5 6 0 2 4 6 0 2 4 6
(a) (b) (c)
T
Distance from centre of sphere - r/σ Distance from centre of sphere - r/σ Distance from centre of sphere - r/σ

Theseand
functions have been generated using a molecular dynamics simulation (see
Phase diagram
Figure 1.3. Typical
2
‚simple system‘
pair distribution functions for (a) a gas,
3 (b) a liquid and (c) a solid.

Figure 1.3. Typical pair distribution functions for (a) a gas, (b) a liquid (c) a solid.
Distance from centre of sphere - r/σ

2.5
section 1.6) 1.5
of atoms interacting through a Lennard-Jones potential (equation (1.9),
These functions have been generated using a molecular dynamics simulation (see
2
ρ(r ) ≡ ρg(r ) = ρ exp (−βv(r ))

figure 1.4). The Boltzmann factor for the Lennard-Jones potential has been
Figure 1.1. Schematic phase diagrams of a simp
2
section 1.6) of atoms interacting through a Lennard-Jones potential (equation (1.9),

• Macromolecules and polymers

superimposed on the gas phase distribution function (dashed curve). The
figure 1.4). The Boltzmann factor for the Lennard-Jones potential thermodynamic
has beeng(r) 1 g(r) 1.5
states are: T = 2ϵ/kB , ρσ 3 = 0.05 (gas), T = ϵ/kB , ρσ 3 = 0.8 (liquid),
temperature (T)–pressure (P), density (ρ)–T and
superimposed on the gas phase distribution function (dashed curve). The , ρσ 3 = 0.9 (solid).
T = 0.2ϵ/k B
thermodynamic states are: T = 2ϵ/kB , ρσ 3 = 0.05 (gas), T = ϵ/kB , ρσ 3 = 0.8 (liquid),
1
4

0.5
3
T = 0.2ϵ/kB , ρσ = 0.9 (solid).
the middle and right panels indicate regions of t 0.5

• Colloidal solids and fluids

0 0

in the right panel

scale is provided
is an of the isotherm
by the radial (or pair) distribution
whilea giventmolecule,
and c sh
function g(r ), which 0 1 2 3 4 5 6 0 2 4 6
(a) (b) Distance from centre of sphere - r/σ Distance from centre of sphere - r/σ
6

characterizes the modulation local density ρ(r ) around

g(r) 3

scale is provided by the radial (or pair) distribution function g(r ), which
(c)

characterizes the modulation of the local densitycritical

ρ(r ) around apoints.
Figure
as a function of the distance r from that 1.3. Typical pair distribution functions for
molecule.
0

given
In themolecule, These functions havemolecules,
been generated using a m
0

limit of an ideal gas of non-interacting (point) the local

as a function of the distance r from that molecule. section is
density, as seen from any one fixed molecule, 1.6) of atomsequal
everywhere interacting through
to its average ρ. a Lenn
In the limit of an ideal gas of non-interacting (point) molecules, the local interact via a pairfigure 1.4). The Boltzmann factor for the Lennard
Distance from centre of sphere - r/σ

In reality, molecules potential v, which, for spherical molecules,

superimposed on the gas phase distribution fun
density, as seen from any one fixed molecule, is everywhere equal willtoonly
its average
depend onρ. the centre-to-centre distance r ; even in a dilute gas, the local
2

thermodynamic states are: T = 2ϵ/kB , ρσ 3 = 0.05

density around
In reality, molecules interact via a pair potential v, which, for spherical a fixed molecule willTrapidly
molecules, vanish
= 0.2ϵ/k for
3 r less than the molecular
B , ρσ = 0.9 (solid).
and T the critical point temperature, above w
will only depend on the centre-to-centre distance r ; even in a c diameter,
dilute gas, and will be modulated by the Boltzmann factor exp(−βv(r )), where
the local
1/kBmolecular
β = the T . In other words
density around a fixed molecule will rapidly vanish for r less than
4

into a single fluid phase. The shaded areas in

diameter, and will be modulated by the Boltzmann factor exp(−βv(r )), where ρ(r ) ≡ ρg(r ) = ρ exp (−βv(r ))
scale is provided by the radial (1.2) (or pair) d
(1.2)

β = 1/kB T . In other words

figure Incorrelated,
1.1 are two-phase
the liquid, however, the positions
regions:
characterizes
thearecorrespo
the molecules
of neighbouring modulation ofstrongly
the local dens
6

ρ(r ) ≡ ρg(r ) = ρ exp (−βv(r )) (1.2)to a modulation as

leading of a
ρ(rfunction of the
) extending overdistance r fromdi-
a few molecular that molec
ther metastable
In the liquid, however, the positions of neighbouring molecules are strongly or may
figure 1.3(b); the maxima unstable,
ameters, so that the radial distribution would
and
In thebehave
be associated with
density, as seen
will
limit asofshown
shells of eventually
schematically
an ideal in
gas of non-intera
from neighbours,
any one fixedbut the
molecule, is
correlated, leading to a modulation of ρ(r ) extending over a oscillations
few molecular di- damped, showing
are rapidly the gradual smearing out of short-range
isting
ameters, so that the radial distribution would behave stable
as shown schematically
figure 1.3(b); the maxima may be associated with shells of neighbours, but the
phases.
in Examples
In reality, molecules
of such
interact via
states a
a pair potentia
will only depend on the centre-to-centre distan
oscillations are rapidly damped, showing the gradualliquids
smearing outor supersaturated
of short-range vapours. Depending
density around a fixed molecule will rapidly v
diameter, and will be modulated by the Boltz
the latter will either form liquid droplets by a β = 1/k T . In other words B

) ≡ ρg(r ) = ρ exp
(to be discussed in section 10.6), orρ(rundergo r
In the liquid, however, the positions of neig
sidered in section 9.3).correlated, Glasses leadingconstitute
to a modulation of ano ρ(r ) e
ameters, so that the radial distribution would
metastable materials, often obtained
figure 1.3(b); the maxima may byberapidassociatedc
oscillations are rapidly damped, showing the g
well below its freezing temperature. Most su
rather extreme cooling conditions, but silicat
excellent glass-formers, as may be readily ob
shop. Glasses are amorphous (‘structureless’
 Molecular ‘fluids’

Molecular systems: water, salt, (conjugated) organic

molecules, peptides, proteins, etc. (~ 0.1-10 nm)

+ Cl- Na+
-

~0.3
~ 2.5 nm
nanometer

(p-hexaphenyl, C36H26) ~ 5 nm
 Polymeric systems

Long chain molecules (often in solution);

polyelectrolytes, polymer brushes & gels, etc. (~10-100 nm)

(thermosensitive)
core-shell hydrogels

hydrophilic hydrophobic
swollen collapsed
polymer
(polyethyleneglycole) switchable at critical temperature,
brushes
size and properties tuneable
 Colloidal fluids

Macromolecules (~10 nm -1 µm) dispersed in a solvent

Examples: milk, ink, paint, blood, bacteria, big proteins,

model colloids, etc.
 Condensed Matter Theory:
The multi-scale challenge
time
Continuum
s mechanics
Dissipative
dynamics
ms Molecular
Dynamics
Ċ = Dr2 C
µs Quantum
<latexit sha1_base64="EwFNuRiaQPMTqBKIPFrzxGKpxsE=">AAACAHicbVBNS8NAEJ3Urxq/oh48eFksgqeSVEEvQrEePFawrdDGstlu2qWbTdjdCKX04l/x4kERr/4Mb/4bN20O2vpg4PHezO7MCxLOlHbdb6uwtLyyulZctzc2t7Z3nN29popTSWiDxDyW9wFWlDNBG5ppTu8TSXEUcNoKhrXMbz1SqVgs7vQooX6E+4KFjGBtpK5z0OnFGtXQJbruCBxw/FBBNdu2u07JLbtToEXi5aQEOepd58u8RNKICk04VqrtuYn2x1hqRjid2J1U0QSTIe7TtqECR1T54+kBE3RslB4KY2lKaDRVf0+McaTUKApMZ4T1QM17mfif1051eOGPmUhSTQWZfRSmHOkYZWmgHpOUaD4yBBPJzK6IDLDERJvMshC8+ZMXSbNS9k7LlduzUvUqj6MIh3AEJ+DBOVThBurQAAITeIZXeLOerBfr3fqYtRasfGYf/sD6/AGKBZPH</latexit>

mechanics
⇠ ẋ = Fx + Rx
ns
<latexit sha1_base64="HVtN/MLRc0BQilkeXzbTaj5+9u8=">AAACA3icbVBNS8MwGE7n16xfVW96CQ5BEEY7Bb0IQ0E8TnEfsJWSZukWlqYlSWWjDLz4V7x4UMSrf8Kb/8Z060E3Hwh5eJ73fZP38WNGpbLtb6OwsLi0vFJcNdfWNza3rO2dhowSgUkdRywSLR9JwigndUUVI61YEBT6jDT9wVXmNx+IkDTi92oUEzdEPU4DipHSkmftdYYUdrqRgkN4Aa89fR3DO28ITdOzSnbZngDOEycnJZCj5llfehBOQsIVZkjKtmPHyk2RUBQzMjY7iSQxwgPUI21NOQqJdNPJDmN4qJUuDCKhD1dwov7uSFEo5Sj0dWWIVF/Oepn4n9dOVHDuppTHiSIcTx8KEgZVBLNAYJcKghUbaYKwoPqvEPeRQFjp2LIQnNmV50mjUnZOypXb01L1Mo+jCPbBATgCDjgDVXADaqAOMHgEz+AVvBlPxovxbnxMSwtG3rML/sD4/AFwwpTN</latexit>

F = ma
<latexit sha1_base64="Z1pqXevY779Vf6W8QH5eGiu0e7E=">AAAB/HicbZDLSsNAFIYnXmu9Rbt0M1gEVyWpgm6EoiAuK9gLtKFMppN26MwkzEyEEOKruHGhiFsfxJ1v4yTNQlt/GPj4zzmcM78fMaq043xbK6tr6xubla3q9s7u3r59cNhVYSwx6eCQhbLvI0UYFaSjqWakH0mCuM9Iz5/d5PXeI5GKhuJBJxHxOJoIGlCMtLFGdi0d+gG8zeAV5AWirDqy607DKQSXwS2hDkq1R/bXcBzimBOhMUNKDVwn0l6KpKaYkaw6jBWJEJ6hCRkYFIgT5aXF8Rk8Mc4YBqE0T2hYuL8nUsSVSrhvOjnSU7VYy83/aoNYB5deSkUUayLwfFEQM6hDmCcBx1QSrFliAGFJza0QT5FEWJu88hDcxS8vQ7fZcM8azfvzeuu6jKMCjsAxOAUuuAAtcAfaoAMwSMAzeAVv1pP1Yr1bH/PWFaucqYE/sj5/ALGIk4A=</latexit>

ps H
<latexit sha1_base64="Bn7cFiPhOWt4wWMFpDZN3499FVQ=">AAAB9XicbVDLSsNAFL2pr1pfVZduBovgqiRV0I1QFKHLCPYBbSyT6aQdOpmEmYlSQv/DjQtF3Pov7vwbJ20W2nrgcg/n3MvcOX7MmdK2/W0VVlbX1jeKm6Wt7Z3dvfL+QUtFiSS0SSIeyY6PFeVM0KZmmtNOLCkOfU7b/vgm89uPVCoWiXs9iakX4qFgASNYG+mh0XMVQ1foNuulfrliV+0Z0DJxclKBHG6//NUbRCQJqdCEY6W6jh1rL8VSM8LptNRLFI0xGeMh7RoqcEiVl86unqITowxQEElTQqOZ+nsjxaFSk9A3kyHWI7XoZeJ/XjfRwaWXMhEnmgoyfyhIONIRyiJAAyYp0XxiCCaSmVsRGWGJiTZBZSE4i19eJq1a1Tmr1u7OK/XrPI4iHMExnIIDF1CHBrjQBAISnuEV3qwn68V6tz7mowUr3zmEP7A+fwB06ZEu</latexit>
=E length

10-12 10-9 10-6 10-3 100

pico nano micro milli 1 meter
 The multi-scale challenge: Battery example
QM; e Classical particle-resolved Continuum & device

10-12 10-9 10-6 10-3 100

pico nano micro milli 1 meter
 The multi-scale challenge: biological example
QM; e Classical particle-resolved Continuum & ‚device‘

10-12 10-9 10-6 10-3 100

pico nano micro milli 1 meter
 The multi-scale challenge: this lecture
time QM; e Classical particle-resolved Continuum & device
Continuum
s mechanics
Dissipative
dynamics
ms Molecular
Dynamics
Ċ = Dr2 C
µs Quantum
<latexit sha1_base64="EwFNuRiaQPMTqBKIPFrzxGKpxsE=">AAACAHicbVBNS8NAEJ3Urxq/oh48eFksgqeSVEEvQrEePFawrdDGstlu2qWbTdjdCKX04l/x4kERr/4Mb/4bN20O2vpg4PHezO7MCxLOlHbdb6uwtLyyulZctzc2t7Z3nN29popTSWiDxDyW9wFWlDNBG5ppTu8TSXEUcNoKhrXMbz1SqVgs7vQooX6E+4KFjGBtpK5z0OnFGtXQJbruCBxw/FBBNdu2u07JLbtToEXi5aQEOepd58u8RNKICk04VqrtuYn2x1hqRjid2J1U0QSTIe7TtqECR1T54+kBE3RslB4KY2lKaDRVf0+McaTUKApMZ4T1QM17mfif1051eOGPmUhSTQWZfRSmHOkYZWmgHpOUaD4yBBPJzK6IDLDERJvMshC8+ZMXSbNS9k7LlduzUvUqj6MIh3AEJ+DBOVThBurQAAITeIZXeLOerBfr3fqYtRasfGYf/sD6/AGKBZPH</latexit>

mechanics
⇠ ẋ = Fx + Rx
ns
<latexit sha1_base64="HVtN/MLRc0BQilkeXzbTaj5+9u8=">AAACA3icbVBNS8MwGE7n16xfVW96CQ5BEEY7Bb0IQ0E8TnEfsJWSZukWlqYlSWWjDLz4V7x4UMSrf8Kb/8Z060E3Hwh5eJ73fZP38WNGpbLtb6OwsLi0vFJcNdfWNza3rO2dhowSgUkdRywSLR9JwigndUUVI61YEBT6jDT9wVXmNx+IkDTi92oUEzdEPU4DipHSkmftdYYUdrqRgkN4Aa89fR3DO28ITdOzSnbZngDOEycnJZCj5llfehBOQsIVZkjKtmPHyk2RUBQzMjY7iSQxwgPUI21NOQqJdNPJDmN4qJUuDCKhD1dwov7uSFEo5Sj0dWWIVF/Oepn4n9dOVHDuppTHiSIcTx8KEgZVBLNAYJcKghUbaYKwoPqvEPeRQFjp2LIQnNmV50mjUnZOypXb01L1Mo+jCPbBATgCDjgDVXADaqAOMHgEz+AVvBlPxovxbnxMSwtG3rML/sD4/AFwwpTN</latexit>

F = ma
<latexit sha1_base64="Z1pqXevY779Vf6W8QH5eGiu0e7E=">AAAB/HicbZDLSsNAFIYnXmu9Rbt0M1gEVyWpgm6EoiAuK9gLtKFMppN26MwkzEyEEOKruHGhiFsfxJ1v4yTNQlt/GPj4zzmcM78fMaq043xbK6tr6xubla3q9s7u3r59cNhVYSwx6eCQhbLvI0UYFaSjqWakH0mCuM9Iz5/d5PXeI5GKhuJBJxHxOJoIGlCMtLFGdi0d+gG8zeAV5AWirDqy607DKQSXwS2hDkq1R/bXcBzimBOhMUNKDVwn0l6KpKaYkaw6jBWJEJ6hCRkYFIgT5aXF8Rk8Mc4YBqE0T2hYuL8nUsSVSrhvOjnSU7VYy83/aoNYB5deSkUUayLwfFEQM6hDmCcBx1QSrFliAGFJza0QT5FEWJu88hDcxS8vQ7fZcM8azfvzeuu6jKMCjsAxOAUuuAAtcAfaoAMwSMAzeAVv1pP1Yr1bH/PWFaucqYE/sj5/ALGIk4A=</latexit>

ps H
<latexit sha1_base64="Bn7cFiPhOWt4wWMFpDZN3499FVQ=">AAAB9XicbVDLSsNAFL2pr1pfVZduBovgqiRV0I1QFKHLCPYBbSyT6aQdOpmEmYlSQv/DjQtF3Pov7vwbJ20W2nrgcg/n3MvcOX7MmdK2/W0VVlbX1jeKm6Wt7Z3dvfL+QUtFiSS0SSIeyY6PFeVM0KZmmtNOLCkOfU7b/vgm89uPVCoWiXs9iakX4qFgASNYG+mh0XMVQ1foNuulfrliV+0Z0DJxclKBHG6//NUbRCQJqdCEY6W6jh1rL8VSM8LptNRLFI0xGeMh7RoqcEiVl86unqITowxQEElTQqOZ+nsjxaFSk9A3kyHWI7XoZeJ/XjfRwaWXMhEnmgoyfyhIONIRyiJAAyYp0XxiCCaSmVsRGWGJiTZBZSE4i19eJq1a1Tmr1u7OK/XrPI4iHMExnIIDF1CHBrjQBAISnuEV3qwn68V6tz7mowUr3zmEP7A+fwB06ZEu</latexit>
=E length

10-12 10-9 10-6 10-3 100

pico nano micro milli 1 meter
 Complexity of a material
H. Löwen, J. Phys. Condens. Matter 13,
R415 (2001)

‚Applied‘ and
biological
materials

polymers,
… water, salt, …
proteins,

THEO4

“simple”
Some applied (research) examples
Molecular Dynamics (MD)

Movie: 50 NaCl, 1000 water

molecules, ~2 mol/l, ~ 100 ps

Method:
Molecular dynamics (MD)
computer simulations
~ 4 nm
 Molecular Dynamics (MD)

Basic principle: integrate Newton’s

X
equation of motion: mr̈ = F = ⇥i H(r1 , .., rN )
i ij
j
classical Hamiltonian
X X X
H(r1 , .., rN ) = Vintra (ri , rj ) + VLJ (ri , rj ) + VCoul (ri , rj )
ij ij ij
 Molecular Dynamics (MD)

Basic principle: integrate Newton’s

X
equation of motion: mr̈ = F = ⇥i H(r1 , .., rN )
i ij
j
classical Hamiltonian
X X X
H(r1 , .., rN ) = Vintra (ri , rj ) + VLJ (ri , rj ) + VCoul (ri , rj )
ij ij ij

r ion

⇣ ⌘ ⇣ ⌘6
12
VLJ (r) = 4✏
r r
 Molecular Dynamics (MD)

Basic principle: integrate Newton’s

X
equation of motion: mr̈ = F = ⇥i H(r1 , .., rN )
i ij
j
classical Hamiltonian
X X X
H(r1 , .., rN ) = Vintra (ri , rj ) + VLJ (ri , rj ) + VCoul (ri , rj )
ij ij ij
qH
qO Coulomb’s law, electrostatics
ion
qion q1 q2
VCoul (r) =
4⇡✏0 r
qH
MD: From ionic structure to osmotic pressure

pressure semipermeable
membrane
+
+ +

+ +
+
+ + +
salt +
+
water fresh
+ +
+
water
+ +

Z
2⇡kB T ⇢2 X 1
⇧ = kB T ⇢ Fij (r)gij (r)r3 dr
3 ij 0
Theory, stat. Mech:
Structure-property
relationship
I. Kalcher and JD, J. Chem. Phys. 130, 134507 (2009)
MD: From ionic structure to osmotic pressure

osm = ⇧/(2kB T ⇢)
osm

I. Kalcher and JD, J. Chem. Phys. 130, 134507 (2009)

 MD of functional liquid/solid interfaces:
ion-specific adsorption on Au nanoparticles
How does ionic
adsorption affect
electrostatic
properties?

à Important for
catalysis and
colloidal stability in
solution.

Li, Lopez, Kanduc, and JD, Langmuir 36, xxxxx (2020).

 MD of functional liquid/solid interfaces:
ion-specific adsorption on Au nanoparticles
pubs.acs.org/Langmuir
Langmuir Article
pubs.acs.org/Langmuir

NaCl

NaPF NaNip NaHCF

6 of water and ions shown by isodensity surfaces in different colors. NaCl solutions for (a)
(3D-resolved) density distribution profile
NP2, and (c) AuNP3 systems. (d) NaPF6, (e) NaNip, and (f) Na3HCF solutions for the AuNP2 system (red: cations, green:
3D density (iso-surfaces)
: water oxygens). The isodensity surface for water shows the density values of 60 nm−3 and for cations and anions of 0.6 nm−3 for
t NaNip, where 10 nm−3 was used for the ions.

tionic and anionic number adsorptions) follows

Angular average:
apparently different adsorption mechanisms is exhibited by
. PF6−, where the adsorption is quite pronounced mostly on the
ensional (3D) Spatial Distribution of Water flat (111) facets, but the first water layer stays more intact than
Li, Lopez, Kanduc,were
and JD, Langmuir 36,for
xxxxx Figure 2. Radial number density distributions ρi(r) normalized by bulk density ρ0i as a fun
e spatial distribution functions calculated the (2020).
case of Nip−. profiles around the three different nanoparticles. The radii of the Gibbs dividing surface
IS program,70 and the plots were created by the Intermediate Discussion:
dashed lines and Possible Effects
vertical dotted of Ion- (b) Water density profiles around Au
lines, respectively.
71
The dense water layers observed in Figure 2a Specific Adsorption on(c)Functional
legend). Anion and (d) cationInterfaces. The around AuNP2 in the presence of diffe
density profiles
tured adlayers in space surrounding all three location and strengthTableof ion1. binding onto an AuNP play an
lized in the spatial density distributions (Figure important role in its surface reaction processes for catalytic
below), which we will abbreviate as ϕMD GDS = ϕ(RGDS) and ϕeff =
MD
can also see
mation of an ordered, hexagonal water structure applications.9−13,15,16 ϕ(R
Theeff),radial number density (in Figure
respectively. 2) radius RGDS
n the extended planar (111) surfaces, while combined with the 3D Inspatialorder to distribution
gain insight into(in Figure 3) show
the effective properties seen in the far roughly corr
field,55,56 −
more disorder on the more “edgy” (100) and distinct ionic adsorptions. In particular,
i.e., effectiveasurface
simplecharges
anionand as Cl is as measured in
potentials the width of
Water is indeed known to have an energetically known to preferentially electrophoretic
adsorb, albeit measurements
relativelyor weakly,
derived from scattering experiments,
to flat Now, by
we consider 28 the Debye−Hü ckel (DH) model for the radial rewritten as5
hydration layer on the bare planar (111) gold gold electrodes at zero potential.
electrostatic Moreover,
potential around athe observation
sphere 57
2,73
The spatial adsorptions of Na+ and Cl− of the strongest adsorption of Nip− at the interface is probably
e−κr structure,
e κRplatelike ln|re0βϕD
e0βϕ (r ) = Zeff lB
eff
eterogeneous behavior with their preferential due to its aromatic character DHalong with its

y on the disordered water regions (probably being favorable for adsorbing at flat, 1 +relatively
κR eff r nonpolar (9)
− where the ri
m broken hydrogen bonds, which may then surfaces. Nip was found where R particularly strongly adsorbed with
eff is the effective radius and Zeff the charge valency of the slope of the
in ionic hydration shells) and hardly on the epitaxial matching onparticle,
the (111)e0 is thefacets (see charge,
elementary Figureβ S3 inBTthe
= 1/k is the inverse thermal far-field de
2
s with the highly packed hexagonal water Supporting Information),energy, lB = βeto
owing the0ϵsoft
0/4πϵ r is the length of the medium, and κ =
Bjerrumadsorption
epitaxial simulations
(8πlBI)1/2 is the inverse Debye length (λD = κ−1), which is ZefflB(eκR )/
mechanism as proposed by Feng and co-workers, where the
proportional to the square root of the bulk ionic strength I, defined
eff

free parame
o the NaCl systems, the (3D) spatially resolved coordination of morebypolarizable
eq 1. atoms (such as C, N, and O) Thus, Zeff c
ucture of water and other salts are shown in in peptides or surfactants Thereof ismultiple
no exact epitaxial
definition sites
of Refffavors thethe effective surface
. Usually, which we de
potentialfacets. 74,75 by electrophoretic experiments where the surface
is obtained
d Figure S2. Similar adlayer structures of water adsorption on gold (111) On the one hand, the given effecti
potential
in Figure 3b are maintained for systems hydrophobic nature of Nip− isenablesdefined by the zeta potential, read off where the water
these32anions to favorably Note, in the
 partial charges of all species, including io
towhere
calculate
the sum runselectric
over allfields and
atom species potentials
j (j = O,
molecules, are H, in+,into
Na
taken anaccount
...), implicit-solve
ρj(r) for calcula
approach,
and zj are thewhere only density
radial number the non-aqueous
distribution charge
the partialcontributions
electrostaticandproperties. charge
This ent
approach assum
of atom species
Poisson’s j, respectively.
equation ϵ0 is thepermittivity
(i.e., without vacuum
the permittivity,
O and εr =H partial
1 becauseand ofϵcharges
is vacuum
r the of wat
refe
the relative permittivity of the medium.The Thetotal
latterelectric
is takenfield
to becan ϵr =be1 decomposed
molecules). This will be useful
(i.e., vacuum) for a full explicit treatment
for the
accordingof all
evaluation
to the
eq 7:charges
of
the firstinpart
the electrostat
the is the ionic co
potential
system, i.e.,inincluding
the far thefield where
solvent water
(water) onlycharges
partial
the second has homogeneous
of Ocomes
contribution and H.from the screeni
water
54
effects.
Alternatively, we will also use the value ϵr = system,
theofNaCl 71 for SPC/E water
the decomposed electric field
toThecalculate
potentialelectric
ϕ(r) and potentials
is obtained
fields ionicby in an implicit-solvent
contribution
integrating to theeqtotal one is negligible
6 twice. After t
approach, where only the non-aqueous that charge
from thecontributions
water molecules enter(Figure 4a). In th
first integration, we obtain the electric
Poisson’s equation (i.e., without the Oelectrostatic
and H partial
field as
potential
a function
chargescalculated
of rad
of water by eq 8, the
distance

∫
molecules). This will be useful for thecontribution
evaluation fromof thethe integration of the electri
electrostatic
potential in the far field where r water only and has
negative, while water
homogeneous molecules control t
screening
1 functional liquid/solid oscillatory structure of the total potential (Fig
= ′ ′ ′
effects. MD of 2 interfaces:
E(r )potential ρ ( r )r d r ≳
ϵ0ϵwe
The ϕ(r)2 is obtainedq by far field
integrating(r eq 2
6 nm),
twice. water
After and
the ionic contrib
ion-specific
first integration, adsorption on Au nanoparticles
r r obtain
0 the electric field as a function of radial (
distance

∫
where Integrate
ρq(r) = charge∑jzr jρjdensity:
(r) is the radial charge density in spheric
1
E(r ) = with
coordinates no ρ ( ′ ′
angular
r )r 2
dr′ dependence. The second integratio
leads to theϵ0ϵelectrostatic
2 q
rr 0 potential around the particle(7)

∫
where ρq(r) = ∑jrzjρj(r) is the radial charge density in spherical
ϕ(r ) = with
coordinates − noE(angular
r′)dr′ dependence. The second integration
0
leads to the electrostatic potential around the particle (

ϕ(r ) = − ∫ E(r′)dr′
à Insights into ion-specific near- and
which fulfills
far-field the
r
boundary condition ϕ(∞) = 0. In order to obta
electrostatics
the surface potential,
0 we can read off the potential (8) at a predefin
Theory: Poisson-Boltzmann
radius,
which fulfills RGDS
e.g., the or some
boundary otherϕ(∞)
condition effective order toRobtain
= 0. In radius eff (to be defin
theory (derivable from
the surface potential, we can read off the potential at a predefined
classical DFT)
radius Reff (to be defined
radius, e.g., RGDS or some other effectivehttps://dx.doi.org/10.1021/acs.langmuir.0c020
Langmuir XXXX, XXX, XXX−X
Li, Lopez, Kanduc, and JD, Langmuir 36, xxxxx (2020).
C https://dx.doi.org/10.1021/acs.langmuir.0c02097
Figure 4.Langmuir XXXX, XXX,
Electrostatic XXX−XXX
properties of
two representa
AuNP2 system. Electric fields E(r) calculated by eq
and (b) NaNip solutions. (c, d) Correspondi
potentials ϕ(r) obtained by eq 8. The total value
each of them is decomposed into two parts, i.e., io
(green line) and contribution from water molecules
vertical dotted lines represent the position of the
(Reff).
 Brownian Dynamics simulations of a simple
colloid model

Pair potential:
⇢
1 if r 
V (r) =
0 if r >

s
number density ⇢ = N/V0
⇡ 3
volume fraction = ⇢
6
N colloids in volume V0
 Brownian Dynamics simulations of a simple
colloid model

Pair potential:
⇢
1 if r 
V (r) =
0 if r >

s
dr
Brownian dynamics (BD): ṙ =
dt
N
X
⇠ ṙi = Fij + ✓i (t)
j=1
N colloids in volume V0 F(r) = rV (r)
 Brownian Dynamics simulations of a simple
colloid model

Structure described by the radial

(pair) distribution function g(r):

Brownian dynamics (BD)

 Hard sphere (fcc) crystals for f > 0.49

F =U TS

P. Pusey, Edinborough

D. Pine, Physics,
Bragg reflection NYU
 Brownian Dynamics simulations of a simple
colloid model

Structure described by the radial

(pair) distribution function g(r):

Brownian dynamics (BD)

 Statistical mechanics:
Structure – property relationships

thermal energy kBT : Boltzmann’s constant * Temperature

✓ Z 1 ◆
compressibility T = kB T ⇢ 1 + 4⇡⇢ [g(r) 1]r2 dr
0

Z 1
internal energy U/N = 2⇡⇢ V (r)g(r)r2 dr
0
Z 1
2⇡kB T ⇢2
pressure P = kB T ⇢ F (r)g(r)r3 dr
(à free energy) 3 0

F (r) = dV /dr
 Statistical mechanics:
Structure – property relationships

statistical mechanics, in general:

the integration of microscopic spatial or temporal

correlation functions yields macroscopic observables.

more examples:
surface tension, electrostatic potential, dielectric constant,
elasticity, viscosity, diffusion constant, transport coefficients, etc.
 g(r): direct connection to (light / X / neutron) -
scattering experiments

static structure factor:

Z
S(k) = 1 + 4⇡⇢ g(r) exp( ik · r)r2 dr

sample k
k0
k0
incident ray
 Shear response of colloidal plastic crystals

Brownian Dynamics

shear magnitude (strain)

~ 1µm
F. Chu, N. Heptner, Y. Lu, M. Siebenbürger, P. Lindner, JD, and M. Ballauff, Langmuir (2015).
(downwards) with velocity (x), velocity gradient (y) and vorticity (z) directions is defined to

Shear response
describe the scattering of geometries.
planes for available colloidal plastic
The experiments crystals
in this work have
been performed in the velocity-vorticity (x-z) scattering plane with the incident beam along
the velocity gradient direction.
Viscoelastic response from experiments:
a  Hard  spheres,   eff
 =  0.57   b  Hard  dumbbells,   eff
 =  0.60  
ⅡⅢ
Ⅰ
Shear modules (Pa)

Shear modules (Pa)

0 0
10 10

Polycrystalline    
phase  
-1 -1
10 10
G' storage modules
G'' loss modules
0 1 2 0 1 2
10 10 10 10 10 10
Shear strain (%) Shear strain (%)

Figure 2 Yielding behavior of hard spheres and hard dumbbells with L* ~ 0.24 in the
fully crystalline phase. Dependence of storage modules (G’) and loss modules (G’’) of hard
F. Chu, N. Heptner,
spheres  at   Y. Lu,
  =  0.57 (a)M.
andSiebenbürger, P. Lindner,
hard dumbbells at   eff  JD, and (b)
=  0.60 M. Ballauff, Langmuir
on increasing (2015).
shear strains at a
eff

fixed frequency (f = 1Hz). (●)   and   (◌)   stands   for   the   storage   modules   G’   and   loss   modules  
G’’,   individually.   To   facilitate   the   discussion   on   the   yielding   behavior   of   hard   dumbbells,  
positions  (I),  (II),  (III)  are  denoted  along  G  curve  according  to  the  position  of  yielding  peaks.  
 Contents of the course
ü Intro: Elements; forms of condensed Matter ✔
ü Solids: binding, crystal structure, diffraction
ü Solids: phonons, sound, thermal properties
ü Solids: electrons (brief)
ü Liquids: interactions, thermodynamics and stat. mech
ü Liquids: structure of Liquids, correlations functions, scattering
ü Liquids: classical DFT (density functional theory)
ü Liquids: Electrostatics (Poisson-Boltzmann)
ü Liquids: Coarse-graining, effective interactions
ü Phase transitions (critical phenomena?)
? Interfaces
? Polymers
? Linear response theory (dynamical response)
• C. Kittel, "Introduction to Solid State Physics"
Literature/ • J.P. Hansen, I.R. McDonald, "Theory of Simple Liquids"
Books: • P. Chaikin, T. Lubensky, "Principles of Condensed Matter Physics"
 Methods
Ø Lecture:
- Concepts and theory of solids, reciprocal space, phonons
- Liquid state theory, correlation functions; advanced statistical mechanics
- Density functional theory
- Poisson-Boltzmann theory
- Free energy calculations
- Construction of phase diagrams
- Linear response theory

Ø Tutorials:
- 60% Analytical work on theory in lecture
- 20% numerical questions
- 20% applied and knowledge question
Bibliography

[1] IBACH and LÜTH, Solid-State Physics: An Introduction to Principles of Materials Sci-
ence. Berlin, Heidelberg: Springer, (2009). doi: 10.1007/978-3-540-93804-0.

[2] KITTEL, Introduction to Solid State Physics, 8th ed. Hoboken, N.J.: Wiley, (2005).

[3] J. N. Israelachvilli, Intermolecular and Surface Forces. Academic Press, Elsevier. Third
Edition, (2011).

[4] GREINER, NEISE, STÖCKER, Thermodynamics and Statistical Mechanics, Springer

(1995).

[5] HANSEN and McDONALD, Theory of Simple Liquids, 4th Edition, Elsevier (2013).

[6] BARRAT and HANSEN, Basic Concepts for Simple and Complex Liquids, Cambridge
Univ. Press (2010).

[7] D. Stigter, “Evaluation of the counterion condensation theory of polyelectrolytes.,” Biophys

J, vol. 69, no. 2, pp. 380–388, Aug. 1995. doi: 10.1016/S0006-3495(95)79910-6.

[8] J. A. Bertolotto and J. P. Umazano, “Counterion condensation theory for finite polyelec-
trolyte and salt concentrations,” J. Phys.: Condens. Matter, vol. 34, no. 35, p. 354003,
Jun. 2022. doi: 10.1088/1361-648X/ac792e.

[9] McQuarrie: Statistical Mechanics, University Science Books (2000).

[10] RUBINSTEIN and COLBY: Polymer Physics, Oxford University Press (2003).

[11] KALCHER and DZUBIELLA : J. Chem. Phys. 130, 134507 (2009).

[12] D. Andelman, “Chapter 12 - Electrostatic Properties of Membranes: The Poisson-

Boltzmann Theory” in Handbook of Biological Physics, R. Lipowsky and E. Sackmann,
Eds., in Structure and Dynamics of Membranes, vol. 1. North-Holland, 1995, pp. 603–642.
doi: 10.1016/S1383-8121(06)80005-9.

[13] W. C. K. Poon and D. Andelman, Eds., Soft Condensed Matter Physics in Molecular and
Cell Biology. Boca Raton: CRC Press, 2006. doi: 10.1201/9781420003338.

[14] I. Kalcher, J. C. F. Schulz, and J. Dzubiella, “Electrolytes in a nanometer slab-confinement:

Ion-specific structure and solvation forces,” The Journal of Chemical Physics, vol. 133, no.
16, p. 164511, Oct. 2010. doi: 10.1063/1.3490666.

175