Molecular Interactions

Molecular Interactions

Concepts and Methods

David A. Micha
University of Florida, Gainesville, FL, USA
This edition first published 2020
© 2020 John Wiley & Sons, Inc.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or
transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or
otherwise, except as permitted by law. Advice on how to obtain permission to reuse material from
this title is available at http://www.wiley.com/go/permissions.
The right of David A. Micha to be identified as the author of this work has been asserted in
accordance with law.
Registered Office
John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, USA
Editorial Office
111 River Street, Hoboken, NJ 07030, USA
For details of our global editorial offices, customer services, and more information about Wiley
products visit us at www.wiley.com.
Wiley also publishes its books in a variety of electronic formats and by print-on-demand. Some
content that appears in standard print versions of this book may not be available in other formats.
Limit of Liability/Disclaimer of Warranty
In view of ongoing research, equipment modifications, changes in governmental regulations, and the
constant flow of information relating to the use of experimental reagents, equipment, and devices,
the reader is urged to review and evaluate the information provided in the package insert or
instructions for each chemical, piece of equipment, reagent, or device for, among other things, any
changes in the instructions or indication of usage and for added warnings and precautions. While the
publisher and authors have used their best efforts in preparing this work, they make no
representations or warranties with respect to the accuracy or completeness of the contents of this
work and specifically disclaim all warranties, including without limitation any implied warranties
of merchantability or fitness for a particular purpose. No warranty may be created or extended
by sales representatives, written sales materials or promotional statements for this work. The fact
that an organization, website, or product is referred to in this work as a citation and/or potential
source of further information does not mean that the publisher and authors endorse the information
or services the organization, website, or product may provide or recommendations it may make.
This work is sold with the understanding that the publisher is not engaged in rendering professional
services. The advice and strategies contained herein may not be suitable for your situation. You
should consult with a specialist where appropriate. Further, readers should be aware that websites
listed in this work may have changed or disappeared between when this work was written and when
it is read. Neither the publisher nor authors shall be liable for any loss of profit or any other
commercial damages, including but not limited to special, incidental, consequential, or other
damages.
Library of Congress Cataloging-in-Publication Data applied for
ISBN: 9780470290743
Cover image: © Photobank gallery/Shutterstock
Cover design by Wiley
Set in 10/12pt Warnock by SPi Global, Pondicherry, India
Printed in the United states of America
10 9 8 7 6 5 4 3 2 1
 v

Contents

Preface xi

1 Fundamental Concepts 1
1.1 Molecular Interactions in Nature 2
1.2 Potential Energies for Molecular Interactions 4
1.2.1 The Concept of a Molecular Potential Energy 4
1.2.2 Theoretical Classification of Interaction Potentials 6
1.2.2.1 Small Distances 7
1.2.2.2 Intermediate Distances 8
1.2.2.3 Large Distances 8
1.2.2.4 Very Large Distances 8
1.3 Quantal Treatment and Examples of Molecular Interactions 9
1.4 Long-Range Interactions and Electrical Properties of Molecules 21
1.4.1 Electric Dipole of Molecules 21
1.4.2 Electric Polarizabilities of Molecules 22
1.4.3 Interaction Potentials from Multipoles 23
1.5 Thermodynamic Averages and Intermolecular Forces 24
1.5.1 Properties and Free Energies 24
1.5.2 Polarization in Condensed Matter 25
1.5.3 Pair Distributions and Potential of Mean-Force 26
1.6 Molecular Dynamics and Intermolecular Forces 27
1.6.1 Collisional Cross Sections 27
1.6.2 Spectroscopy of van der Waals Complexes
and of Condensed Matter 28
1.7 Experimental Determination and Applications of Interaction Potential
Energies 29
1.7.1 Thermodynamics Properties 30
1.7.2 Spectroscopy and Diffraction Properties 30
1.7.3 Molecular Beam and Energy Deposition Properties 30
1.7.4 Applications of Intermolecular Forces 31
References 31
vi Contents

2 Molecular Properties 35
2.1 Electric Multipoles of Molecules 35
2.1.1 Potential Energy of a Distribution of Charges 35
2.1.2 Cartesian Multipoles 36
2.1.3 Spherical Multipoles 37
2.1.4 Charge Distributions for an Extended System 38
2.2 Energy of a Molecule in an Electric Field 40
2.2.1 Quantal Perturbation Treatment 40
2.2.2 Static Polarizabilities 41
2.3 Dynamical Polarizabilities 43
2.3.1 General Perturbation 43
2.3.2 Periodic Perturbation Field 47
2.4 Susceptibility of an Extended Molecule 49
2.5 Changes of Reference Frame 52
2.6 Multipole Integrals from Symmetry 54
2.7 Approximations and Bounds for Polarizabilities 57
2.7.1 Physical Models 57
2.7.2 Closure Approximation and Sum Rules 58
2.7.3 Upper and Lower Bounds 59
References 60

3 Quantitative Treatment of Intermolecular Forces 63

3.1 Long Range Interaction Energies from Perturbation Theory 64
3.1.1 Interactions in the Ground Electronic States 64
3.1.2 Interactions in Excited Electronic States and in Resonance 68
3.2 Long Range Interaction Energies from Permanent and Induced
Multipoles 68
3.2.1 Molecular Electrostatic Potentials 68
3.2.2 The Interaction Potential Energy at Large Distances 70
3.2.3 Electrostatic, Induction, and Dispersion Forces 73
3.2.4 Interacting Atoms and Molecules from Spherical Components of
Multipoles 75
3.2.5 Interactions from Charge Densities and their Fourier
Components 76
3.3 Atom–Atom, Atom–Molecule, and Molecule–Molecule Long-Range
Interactions 78
3.3.1 Example of Li++Ne 78
3.3.2 Interaction of Oriented Molecular Multipoles 79
3.3.3 Example of Li++HF 80
3.4 Calculation of Dispersion Energies 81
3.4.1 Dispersion Energies from Molecular Polarizabilities 81
3.4.2 Combination Rules 82
 Contents vii

3.4.3 Upper and Lower Bounds 83

3.4.4 Variational Calculation of Perturbation Terms 86
3.5 Electron Exchange and Penetration Effects at Reduced Distances 87
3.5.1 Quantitative Treatment with Electronic Density Functionals 87
3.5.2 Electronic Rearrangement and Polarization 93
3.5.3 Treatments of Electronic Exchange and Charge Transfer 98
3.6 Spin-orbit Couplings and Retardation Effects 102
3.7 Interactions in Three-Body and Many-Body Systems 103
3.7.1 Three-Body Systems 103
3.7.2 Many-Body Systems 106
References 107

4 Model Potential Functions 111

4.1 Many-Atom Structures 111
4.2 Atom–Atom Potentials 114
4.2.1 Standard Models and Their Relations 114
4.2.2 Combination Rules 116
4.2.3 Very Short-Range Potentials 117
4.2.4 Local Parametrization of Potentials 117
4.3 Atom–Molecule and Molecule–Molecule Potentials 119
4.3.1 Dependences on Orientation Angles 119
4.3.2 Potentials as Functionals of Variable Parameters 124
4.3.3 Hydrogen Bonding 124
4.3.4 Systems with Additive Anisotropic Pair-Interactions 125
4.3.5 Bond Rearrangements 125
4.4 Interactions in Extended (Many-Atom) Systems 127
4.4.1 Interaction Energies in Crystals 127
4.4.2 Interaction Energies in Liquids 131
4.5 Interaction Energies in a Liquid Solution and in Physisorption 135
4.5.1 Potential Energy of a Solute in a Liquid Solution 135
4.5.2 Potential Energies of Atoms and Molecules Adsorbed
at Solid Surfaces 139
4.6 Interaction Energies in Large Molecules and in Chemisorption 143
4.6.1 Interaction Energies Among Molecular Fragments 143
4.6.2 Potential Energy Surfaces and Force Fields in Large Molecules 145
4.6.3 Potential Energy Functions of Global Variables Parametrized with
Machine Learning Procedures 148
References 152

5 Intermolecular States 157

5.1 Molecular Energies for Fixed Nuclear Positions 158
5.1.1 Reference Frames 158
viii Contents

5.1.2 Energy Density Functionals for Fixed Nuclei 160

5.1.3 Physical Contributions to the Energy Density Functional 162
5.2 General Properties of Potentials 163
5.2.1 The Electrostatic Force Theorem 163
5.2.2 Electrostatic Forces from Approximate Wavefunctions 164
5.2.3 The Example of Hydrogenic Molecules 165
5.2.4 The Virial Theorem 166
5.2.5 Integral Form of the Virial Theorem 168
5.3 Molecular States for Moving Nuclei 169
5.3.1 Expansion in an Electronic Basis Set 169
5.3.2 Matrix Equations for Nuclear Amplitudes in Electronic States 170
5.3.3 The Flux Function and Conservation of Probability 172
5.4 Electronic Representations 172
5.4.1 The Adiabatic Representation 172
5.4.2 Hamiltonian and Momentum Couplings from Approximate Adiabatic
Wavefunctions 173
5.4.3 Nonadiabatic Representations 174
5.4.4 The Two-state Case 175
5.4.5 The Fixed-nuclei, Adiabatic, and Condon Approximations 176
5.5 Electronic Rearrangement for Changing Conformations 180
5.5.1 Construction of Molecular Electronic States from Atomic States:
Multistate Cases 180
5.5.2 The Noncrossing Rule 181
5.5.3 Crossings in Several Dimensions: Conical Intersections
and Seams 184
5.5.4 The Geometrical Phase and Generalizations 189
References 192

6 Many-Electron Treatments 195

6.1 Many-Electron States 195
6.1.1 Electronic Exchange and Charge Transfer 195
6.1.2 Many-Electron Descriptions and Limitations 198
6.1.3 Properties and Electronic Density Matrices 203
6.1.4 Orbital Basis Sets 205
6.2 Supermolecule Methods 209
6.2.1 The Configuration Interaction Procedure for Molecular Potential
Energies 209
6.2.2 Perturbation Expansions 215
6.2.3 Coupled-Cluster Expansions 218
6.3 Many-Atom Methods 222
6.3.1 The Generalized Valence-Bond Method 222
 Contents ix

6.3.2 Symmetry-Adapted Perturbation Theory 225

6.4 The Density Functional Approach to Intermolecular Forces 228
6.4.1 Functionals for Interacting Closed- and Open-Shell Molecules 228
6.4.2 Electronic Exchange and Correlation from the Adiabatic-Connection
Relation 232
6.4.3 Issues with DFT, and the Alternative Optimized Effective Potential
Approach 238
6.5 Spin-Orbit Couplings and Relativistic Effects in Molecular
Interactions 243
6.5.1 Spin-Orbit Couplings 243
6.5.2 Spin-Orbit Effects on Interaction Energies 245
References 247

7 Interactions Between Two Many-Atom Systems 255

7.1 Long-range Interactions of Large Molecules 255
7.1.1 Interactions from Charge Density Operators 255
7.1.2 Electrostatic, Induction, and Dispersion Interactions 258
7.1.3 Population Analyses of Charge and Polarization Densities 260
7.1.4 Long-range Interactions from Dynamical Susceptibilities 262
7.2 Energetics of a Large Molecule in a Medium 265
7.2.1 Solute–Solvent Interactions 265
7.2.2 Solvation Energetics for Short Solute–Solvent Distances 268
7.2.3 Embedding of a Molecular Fragment and the QM/MM
Treatment 270
7.3 Energies from Partitioned Charge Densities 272
7.3.1 Partitioning of Electronic Densities 272
7.3.2 Expansions of Electronic Density Operators 274
7.3.3 Expansion in a Basis Set of Localized Functions 277
7.3.4 Expansion in a Basis Set of Plane Waves 279
7.4 Models of Hydrocarbon Chains and of Excited Dielectrics 281
7.4.1 Two Interacting Saturated Hydrocarbon Compounds: Chains and
Cyclic Structures 281
7.4.2 Two Interacting Conjugated Hydrocarbon Chains 284
7.4.3 Electronic Excitations in Condensed Matter 289
7.5 Density Functional Treatments for All Ranges 291
7.5.1 Dispersion-Corrected Density Functional Treatments 291
7.5.2 Long-range Interactions from Nonlocal Functionals 294
7.5.3 Embedding of Atomic Groups with DFT 297
7.6 Artificial Intelligence Learning Methods for Many-Atom Interaction
Energies 300
References 303
x Contents

8 Interaction of Molecules with Surfaces 309

8.1 Interaction of a Molecule with a Solid Surface 309
8.1.1 Interaction Potential Energies at Surfaces 309
8.1.2 Electronic States at Surfaces 314
8.1.3 Electronic Susceptibilities at Surfaces 319
8.1.4 Electronic Susceptibilities for Metals and Semiconductors 321
8.2 Interactions with a Dielectric Surface 324
8.2.1 Long-range Interactions 324
8.2.2 Short and Intermediate Ranges 329
8.3 Continuum Models 332
8.3.1 Summations Over Lattice Cell Units 332
8.3.2 Surface Electric Dipole Layers 333
8.3.3 Adsorbate Monolayers 335
8.4 Nonbonding Interactions at a Metal Surface 337
8.4.1 Electronic Energies for Varying Molecule–Surface Distances 337
8.4.2 Potential Energy Functions and Physisorption Energies 341
8.4.3 Embedding Models for Physisorption 347
8.5 Chemisorption 349
8.5.1 Models of Chemisorption 349
8.5.2 Charge Transfer at a Metal Surface 354
8.5.3 Dissociation and Reactions at a Metal Surface from Density
Functionals 359
8.6 Interactions with Biomolecular Surfaces 363
References 367

Index 373
 xi

Preface

The temperatures and densities of the chemical elements present in our natural
world lead to the formation of molecules, which interact to create complex
many-atom systems. Many materials and the components of living organisms
are made up of aggregates of atomic and molecular units. Fluids, molecular
and atomic solids, polymers, and proteins are examples of those aggregates.
The properties of these objects can be described from first principles of quan-
tum mechanics and statistical mechanics, after their composition in terms of
electrons and atomic nuclei have been specified.
Molecular encounters are governed by intermolecular forces, which can be
measured and calculated. This book presents concepts and methods needed
to obtain energies of a molecular system, as they change with interatomic dis-
tances to provide intermolecular forces. Once the interaction forces in a molec-
ular system are known, the equations of motion of classical or quantum
mechanics, implemented with physical boundary conditions, can be used to
obtain thermodynamic equilibrium and nonequilibrium properties from first
principles, and the response of the system to external factors, such as light.
Interactions of molecular species can be described by their electromagnetic
properties and chemical reactivity. The reverse is also true. Information about
molecular interaction energies can be derived from measurements of thermo-
dynamic, kinetic, and electromagnetic properties of matter, and from experi-
ments specifically devised to extract interaction energies, such as crossed
molecular beam and photodissociation experiments. This information has been
incorporated into the selection of concepts and methods for the book.
The advent of quantum theory and its applications during the first half of the
twentieth century has led to many of the quantitative concepts about molecular
structure and properties employed today, and how to calculate them, as well as
ways to describe their interaction energies. Molecular dynamics and statistical
mechanics have also provided thermodynamical and kinetics values of proper-
ties to be compared with experimental measurements. Comparisons and tests of
theoretical and experimental results require extensive computational work.
Computational methods, software and utilities, and the ever-increasing speed
xii Preface

and data storage power of computers have been essential for these purposes.
Some of these methods are covered in the following chapters.
Intermolecular forces are essential in many applications of molecular and
materials properties to technologies contributing to the needs of society. To
illustrate the enormous impact of the subject, some of their subjects (and their
applications) are storage of hydrogen in solids (fuel cells), storage and transport
of ions in solids (batteries), synthesis of thermally stable and conducting sur-
faces (solar energy devices), delivery of compounds through biological cell
membranes (pharmacology), catalysis and photocatalysis in electrochemical
cells (sustainable fuel production), atmospheric reactions (environmental
sciences), efficient fuel combustion (transportation and energy), and solvation
and lubricants (machinery). Furthermore, as the quantitative tools of chemistry
and physics in this book have become more useful and common in biology,
pharmaceutics, and medicine, its contents should also be of interest in these
new areas of applications.
This book is based on the author’s lectures given over many years at the Uni-
versity of Florida in Chemical Physics courses taken by graduate and advanced
undergraduate students in Chemistry, Physics, Chemical Engineering, and
Materials Sciences, with working knowledge of quantum and statistical
mechanics as usually covered in Physical Chemistry or Modern Physics courses.
The lectures have emphasized concepts and methods used in calculations with
realistic models, to be compared with empirical data. They have been expanded
to cover recent developments with advanced theoretical methods, as presented
by researchers at conferences and particularly at many Sanibel Symposia on the
quantum treatments of atoms, molecules, and materials, which this book’s
author has helped to co-organize.
The text contains hundreds of citations, but these are only a small portion of
the tens of thousands of relevant publications on the subject of molecular
interactions. To compensate, much can be done to gain additional knowledge,
including physical and chemical data, by searching for information in the
World Wide Web. The concepts and methods described here are meant to also
provide a broad vocabulary as needed to search the Web. To keep the book
length within reasonable bounds, its contents have been restricted to cover
the energetics of molecular interactions and the potential energy surfaces that
are generated as interatomic distances are varied. A short introduction has been
given about how these interactions affect the spectroscopy, dynamics, and
kinetics of molecular interactions, but these subjects are not covered in detail
here and have been left for possible future treatments. References have been
limited to fundamental publications, monographs, reviews, and comprehensive
reports. Apologies are advanced here to the authors of many relevant publica-
tions that could not be mentioned due to length limitations.
Chapters in this book have been organized to cover first the concepts and
methods for simpler systems, and next to cover more advanced subjects for
 Preface xiii

complex systems. Each chapter begins with qualitative aspects before these are
treated in quantitative fashion. The first four chapters include the well-
established concepts and quantitative aspects of long-range (electrostatic,
induction, and dispersion) forces and how they extend to intermediate and
short ranges, for ground and excited states. They are followed by chapters deal-
ing with recent developments including electronically nonadiabatic interac-
tions, correlated many-electron treatments, generalized density functional
theory, decomposition and embedding of molecular fragments for large sys-
tems, and very recent developments using artificial intelligence with network
training for many-atom systems. The first four chapters and the first sections
of the following four chapters can provide an introductory course on molecular
interactions, for students with a knowledge of the fundamentals of quantum and
statistical mechanics and of molecular structure. A more advanced course can
be based on material starting with the fourth chapter and continuing with the
following ones.
This author thanks his science teachers at the Physics Institute, Universidad
de Cuyo, Bariloche (Argentina), presently Instituto Balseiro, and at the Quan-
tum Chemistry Group, Uppsala University (Sweden), and particularly Dr. Jose
Balseiro in Bariloche for his introduction to quantum theory and Dr. Per-Olov
Lowdin in Uppsala, for his mentoring in quantum chemistry. Much was learned
from colleagues and researchers at the University of Wisconsin (Madison) and
at the University of California, San Diego, and at institutions visited during sab-
baticals at Gothenburg University in Sweden, Harvard University, Max Planck
Institutes for Stroemungsforschung and for Astrophysics in Germany, Imperial
College in London, England, Institute of Theoretical Physics at the University of
California Santa Barbara, JILA at the University of Colorado in Boulder, the
Weizmann Institute in Rehovot, Israel, Florida State University, the Ecole Nor-
mal Superieur, Paris, France, and the Institute for Mathematics and its Applica-
tions, University of Minnesota, Minneapolis. This author is grateful also to the
many graduate and undergraduate students who did research under his direc-
tion and to postdoctoral associates and visiting scientists working with him at
the University of Florida. They are too numerous to be listed here but their
names can be found at the website https://people.clas.ufl.edu/Micha.
This author is greatly appreciative of the financial support by the Foundations
that provided essential funds for his research and collaborations: the Swedish
International Development Agency, the Alfred P. Sloan Foundation, the
National Science Foundation of the USA for many years of Principal Investiga-
tor research and also for support of US-Latin American workshops, NASA, the
Alexander von Humboldt Foundation Program of U.S. Senior Scientists Awards
twice, and the Dreyfus Foundation.
Thanks are also due to reviewers at the early stages when this book was being
planned, particularly Bernard Kirtman, George Schatz, and Victor Batista. Sev-
eral of the author’s research colleagues have been kind enough to read and
xiv Preface

comment on some of the book’s chapters, in particular Rod Bartlett, Adrian

Roitberg, John Stanton, Akbar Salam, Pilar de Lara-Castells, and Benjamin
Levine. All of them had helpful comments at early and later times, but they
are not to be blamed for any shortcomings in this book!
This author appreciates the patience and support of the editors at the pub-
lisher, John Wiley and Sons, Inc., over the years it took to have this book com-
pleted. As the times for its production and publication approach, special thanks
go to the project editor, Ms. Aruna Pragasam, and the senior editor Jonathan T.
Rose for their guidance. A special thanks is also due to the author’s wife, Rebecca
A. Micha, for her help keeping the English grammar right in many paragraphs
and for her unwavering support.

David A. Micha,
December 2018 Gainesville, FL, USA
 1

Fundamental Concepts

CONTENTS

1.1 Molecular Interactions in Nature, 2

1.2 Potential Energies for Molecular Interactions, 4
1.2.1 The Concept of a Molecular Potential Energy, 4
1.2.2 Theoretical Classification of Interaction Potentials, 6
1.2.2.1 Small Distances, 7
1.2.2.2 Intermediate Distances, 8
1.2.2.3 Large Distances, 8
1.2.2.4 Very Large Distances, 8
1.3 Quantal Treatment and Examples of Molecular Interactions, 9
1.4 Long-Range Interactions and Electrical Properties of Molecules, 21
1.4.1 Electric Dipole of Molecules, 21
1.4.2 Electric Polarizabilities of Molecules, 22
1.4.3 Interaction Potentials from Multipoles, 23
1.5 Thermodynamic Averages and Intermolecular Forces, 24
1.5.1 Properties and Free Energies, 24
1.5.2 Polarization in Condensed Matter, 25
1.5.3 Pair Distributions and Potential of Mean-Force, 26
1.6 Molecular Dynamics and Intermolecular Forces, 27
1.6.1 Collisional Cross Sections, 27
1.6.2 Spectroscopy of van der Waals Complexes and of Condensed Matter, 28
1.7 Experimental Determination and Applications of Interaction Potential Energies, 29
1.7.1 Thermodynamics Properties, 30
1.7.2 Spectroscopy and Diffraction Properties, 30
1.7.3 Molecular Beam and Energy Deposition Properties, 30
1.7.4 Applications of Intermolecular Forces, 31
References, 31

Molecular Interactions: Concepts and Methods, First Edition. David A. Micha.

© 2020 John Wiley & Sons, Inc. Published 2020 by John Wiley & Sons, Inc.
2 1 Fundamental Concepts

1.1 Molecular Interactions in Nature

Many materials and the components of living organisms in nature are made up
of aggregates of atomic and molecular units. Fluids, molecular and atomic
solids, polymers, and proteins are examples. The properties of these objects
can be described from first principles of quantum mechanics and statistical
mechanics, after their composition in terms of electrons and nuclei have been
specified. This requires a theoretical framework to describe the structure of
atoms and molecules and the way the atoms and molecules interact. In some
cases, the interactions affect but do not change the conformations of molecules,
which can then be taken as the basic building units of the objects being studied.
Once the interaction forces in a molecular system are known, the equations of
motion of classical or quantum mechanics, implemented with physical bound-
ary conditions, can be used to derive thermodynamic equilibrium and nonequi-
librium properties from first principles. The response of the system to external
factors, such as light and interacting species, can be described by their electro-
magnetic and chemical reactivity properties. The reverse is also true. Informa-
tion about molecular interaction energies can be derived from measurements of
thermodynamic, kinetic, and electromagnetic properties of matter, and from
experiments specially devised to extract interaction energies, such as crossed
molecular beam and photodissociation experiments.
Early treatments that proceed from atomic and molecular structure to the cal-
culation of intermolecular forces and properties of molecular systems have been
covered in several books going back over 50 years [1–4] and have been
expanded to incorporate results of more extensive calculations [5–13] made
possible by continuous improvements in computational power. The present
work introduces and updates theoretical concepts and methods needed to
model structures and properties and provides links to the more recent compu-
tational developments.
A chronological Table 1.1 follows with some important early discoveries on
molecular interactions. They start with the acceptance of the existence of mole-
cules and understanding of how they interact, as early as the nineteenth century.
The advent of quantum theory and its applications during the first half of the
twentieth century has then lead to many of the quantitative concepts about
molecular structure and properties and how to calculate them, as well as ways
to describe their interaction energies. Statistical mechanics has also provided
thermodynamical and kinetics values of properties to be compared with exper-
imental measurements.
 1.2 Molecular Interactions in Nature 3

Table 1.1 Chronology of some early discoveries on molecular interactions.

Year Authors Subject

1857 R. Clausius Distance dependence of interaction potentials

1868 J.C. Maxwell Simple molecular transport theory
1872 L. Boltzmann Transport theory for fluids
1873 J.D. van der Waals Equation of state for real gases
1905 P. Langevin Ion–molecule interaction potential
1912 P. Debye Dielectric properties of fluids
1924 J.E. Lennard-Jones Analytic interaction potentials
1925 J. Franck and E.U. Condon Molecular photoexcitations
1927 W. Heitler and F. London Chemical bonding in H2
1927 M. Born and J.D. Quantum theory of molecules
Oppenheimer
1930 F. London Quantal calculation of dispersion forces
1931 J.C. Slater and J.G. Kirkwood Variational calculation of dispersion forces
1932 J.H. Van Vleck Electric and magnetic susceptibilities of
molecules
1933 P.K.L. Drude Optical properties of fluids
1939 R. Feynman Hellmann–Feynman or force theorem
1943 B.M. Axilrod and E. Teller Three-atom interaction potentials

The link between electrons and nuclei and potential energy functions related
to molecular structure is provided by quantum chemistry, and thermodynam-
ical properties follow by using statistical mechanics. Steady-state (or transport)
properties as well as nonequilibrium properties, including reactivity, can be
obtained from molecular dynamics, as shown in the block diagram illustrated
in Figure 1.1. Spectroscopic properties follow from the electrodynamics of
molecular systems. Transport in gases follow from molecular collision cross
sections.
4 1 Fundamental Concepts

Electrons, nuclei,
and photons

Molecular properties Quantum chemistry

Potential energy functions

Intramolecular
Intermolecular
Statistical mechanics Molecular dynamics

Collisional cross sections

Thermodynamical properties
Molecular kinetics
Virial coefficients
Molecular reactivity
Heat capacities
Compressibilities
Dielectric coefficients
Equilibrium constants Transport coefficients
Relaxation times
Reaction rates
Spectral line shapes
Time-resolved spectra

Figure 1.1 Properties derived from interacting electrons, nuclei, and photons.

1.2 Potential Energies for Molecular Interactions

1.2.1 The Concept of a Molecular Potential Energy
The large difference between electron and nuclear masses leads to a qualitative
difference between electronic and nuclear motions. Nuclei move slowly com-
pared to electronic motions and this allows the formation of molecules, with
positively charged nuclei held together by the negative electron distributions.
This leads to stable molecular systems when total energies are not too high
compared to electronic binding energies. We consider electrons bound by Cou-
lomb forces inside a finite region of space, and separately deal with (i) bound
nuclei, where all particles are restricted by their own Coulomb forces to a finite
region of space, and (ii) unbound nuclei with attached electrons, where some of
the particles (atoms, molecules, etc.) are involved in a collision event.
For molecules occupying a finite region of space, electrons and nuclei move
under their own Coulomb forces. Since the magnitude of electron and nuclear
charges are comparable, their Coulomb forces are similar so that Fe ≈ Fn and in
terms of masses and accelerations, meΔve/Δt ≈ mnΔvn/Δt. In the time interval
Δt one therefore finds comparable momentum changes, meΔve ≈ mnΔvn. Pro-
vided electronic and nuclear momenta are comparable to begin with and a sys-
tem is observed over short times, one finds that over time pe = meve ≈ pn = mnvn.
Hence, since mn > 2000 me, one finds that velocities satisfy
 1.2 Potential Energies for Molecular Interactions 5

pe pn
ve = vn =
me mn
and to a first approximation nuclei can be assumed to be at rest while electrons
move around them. Fixing the nuclear positions, the molecular energies become
functions of the nuclear coordinates and provide the potential energies for the
nuclear motions; hence, they can be referred to as molecular potential energies.
This is the Born–Oppenheimer picture of molecular structure [14, 15]. These
authors showed, using quantum mechanical perturbation theory, that for bound
molecular states, the potential energy correction due to nuclear motion goes as
(me/mn)1/2, while the correction to molecular wavefunctions goes as (me/mn)1/4,
and therefore are relatively small and acceptable for most applications.
To be more specific, we can compare electronic and nuclear kinetic energies;
for the electronic kinetic energy Kel we find, assuming hydrogenic states for an
electron moving in a framework of atomic cores with an effective (screened)
positive charge, and since then Kel ≈ − Eel for the bound states,
me v2e 1 Zscr
2
e2 1 Zscr
2
Kel = ≈ = 27 2 eV
2 2 n2 aB 2 n2
where e, aB, Zscre, and n are the electron charge in esu units, the atomic Bohr
radius, the screened nuclear charge around which the electron orbits, and the
principal quantum number of the orbital, respectively. The energy units conver-
sion is that 1 au(E) = 27.2116 eV, and 1 eV = 1.60218 × 10−19 J = 96.485 kJ mol−1
for future reference in this work.
The magnitudes of nuclear kinetic energies depend on whether the nuclei are
bound or not. Letting vn ≤ ve/10 be our criterion for separation of electronic and
nuclear motions, we find for the nuclear kinetic energy,
2
mn v2n mn ve 10 1 mn 1 e2 Zscr
2
Kn = ≤ =
2 2 100 me 2 aB n2
where mn ≈2000mn A in terms of an average atomic mass number A. This
expression shows that as the electronic state is excited, and n increases, the
upper energy bound for motion separation decreases, and it becomes more dif-
ficult to satisfy the inequality; the separation of electronic and nuclear motions
is more doubtful for electronically excited states.
For ground and low excited electronic states, we can separately estimate
nuclear kinetic energies for molecular systems with bound and unbound nuclei.
A) Bound nuclei
The nuclei in molecules vibrate and rotate. We can estimate the vibrational
and rotational kinetic energies by means of
1 2
Kvib ≈hνv nv , νv = k μn 2π
Krot ≈ℏ l l + 1
2
2I
in a system where the vibrational frequencies are of magnitude νv (given in
terms of a vibrational force constant k and a reduced mass μn), I is a
6 1 Fundamental Concepts

moment of inertia, and the vibrational and rotational quantum numbers are
nv and l, respectively. These terms would typically add to about 1.0 eV and
would be much smaller than the required upper bound to the nuclear
kinetic energy. However, the bound can be exceeded if the vibrational–
rotational levels are very high, signifying less accuracy in the separation
of electronic and nuclear motions.
B) Unbound nuclei
For colliding atoms and molecules, nuclear momenta include the center of
mass components acquired in the collision, which can be very large for
example in atomic beams. Potential energy functions can then be defined
only provided as before, in eV units, Kn ≤ 272 eV A Zscr
2
n2 For a hydrogen
atom, a potential energy function can safely be defined for kinetic energies
below 272 eV, while for deuterium (A = 2) the upper limit would be 544 eV.
The potentials can also be defined above these values, but one must consider
the possibility that electronic and nuclear motions will be coupled.

1.2.2 Theoretical Classification of Interaction Potentials

Interaction potential energies between species A and B are obtained from the
Coulomb interaction of all electrons and nuclei making up the whole system,
and their quantal expectation value in given electronic states, for fixed nuclear
positions. The nuclear position variables are collected in a set {Qj} = Q, where
the Qj are for example atomic position vectors, or bond coordinates (distances,
angles), and the distance R between the centers of mass of the two molecules A
and B. The electronic variables (positions and spin) are collected in X. The
potential energy VΓ(Q) of the system in a quantum state ΦΓ(X; Q) is obtained
from a quantal expectation value by integrating over electronic variables. It can
be treated separating nuclear variables as Q = (R, Q ) and concentrating first on
its dependence on R for a given quantum state.
Total pair potential energies are the sum of long-range, short-range, and
intermediate-range contributions, V(R) = VSR(R) + VIR(R) + VLR(R). Short-range
potentials follow from electronic charge distributions obtained in quantum
chemical calculations or by parametrizing measured properties. This can also
be done for the intermediate range potential energies. Long-range potential
energies follow from properties of the two species.
A theoretical analysis of the electronic states of a molecular system allows us
to classify the interactions in accordance with the separation between interact-
ing fragments A and B. We work in a reference frame where the z-axis goes
through the centers of mass of A and B, at relative distance R. Indicating with
aA and aB the approximate radii of the two fragments, the classification can be
made with respect to ranges of R/(aA + aB) omitting, to simplify matters here,
the dependences on internal and orientation coordinates. In each region, one
 1.2 Potential Energies for Molecular Interactions 7

Table 1.2 Classification of molecular interaction potentials.

Separation R/(aA + aB) Potential type R-dependence Treatment

Small < 1.0 Repulsive A(R) exp(−bR) Variational

Intermediate 1.0–10 Valence F(R; Rm, … Vm, …) Variational–
perturbative
−n
Large 10–50 Electrodynamic nCnR Perturbative
−n
Very large >50 Retardation nDnR Perturbative

can consider the types of contributions to the potential energy, its dependence
with distance R, and the way it can be calculated with theoretical methods.
Table 1.2 summarizes this information.
The calculation of molecular interaction potential energies requires a quan-
tum mechanical treatment starting from the structure of the whole collection of
electrons and nuclei. The Schroedinger equation of motion of the whole system
can be solved to obtain stationary state solutions ΦΓ(X; Q) where the collection
of nuclear position variables Q is considered, to begin with, as fixed at chosen
values. The index Γ is a collection of quantum numbers that also labels the
energy EΓ(Q) of the steady state. These steady states can be calculated with a
variety of methods, including variational and perturbation treatments of the
equations of motion with given boundary conditions [16–19].

1.2.2.1 Small Distances

Short-range interactions are repulsive due to the so-called Pauli repulsion of
closed electron shells, resulting from the requirement of single occupancy of
electron orbitals imposed by the antisymmetry of electronic wavefunctions.
This leads to distortion of overlapping shells that acquire opposing dipoles
so that their repulsion goes as VSR(R) = CSRR−3 at short distances R ≥ Rmin
and contains larger powers of R at even shorter distances where distortion mul-
tipoles are formed. This can be represented by an exponential function A exp
(−R/a) , more approximately by an inverse power form Cn/Rn, or instead by a
combination ( nAn/Rn) exp(−R/a) with some coefficients chosen to reproduce
intermediate-region energies. However, for two atoms A and B with nuclear
charges Za and Zb and at very short distances that exclude electronic charge
in between them, the potential energy goes as ZaZb/(4πε0R) as R 0.
Variational methods can be used to calculate potential energies at small dis-
tances with relatively simple trial electronic wavefunctions of the Hartree–Fock
type, including Coulomb and exchange electronic interactions. Contributions
are the electronic kinetic, Coulomb, and exchange energies, and the potential
can be obtained by minimization of an energy functional of the trial wavefunc-
tion. Energy functionals of the electronic density can also provide useful infor-
mation for short distances.
8 1 Fundamental Concepts

1.2.2.2 Intermediate Distances

At intermediate distances, the potential can be parameterized in terms of the
position Rm and energy Vm at the potential minimum, and other parameters
referring, e.g. to the zero of the potential, inflection point, etc. Results can be
fit to expansions in powers of R − Rm, or can be fit by functions suitable for near
equilibrium regions, such as the Morse potential [2] within a finite region, con-
nected by splines to short- and long-range energy functions.
Elaborate electronic wavefunctions are required for trial functions in varia-
tional methods, including the effects of electron correlation, for example by
using superpositions of configurations, or wavefunctions must be obtained from
perturbation expansions properly symmetrized to account for overall antisym-
metry with respect to electron exchange. This will be covered in the chapter on
many-electron treatments.

1.2.2.3 Large Distances

At large distances, the potential results from the interaction of permanent and
fluctuating charge distributions, leading to electrostatic, induction, and disper-
sion energies,

VLR = Vel + Vind + Vdisp = n≥1

Cn R −n

They can be derived from perturbation theory using electronic wavefunctions

appropriate for the separated species and can be expressed in terms of molec-
ular properties such as charge, electric multipole, and multipolar dynamical
polarizabilities.

1.2.2.4 Very Large Distances

It is convenient to describe very long-range interactions in the language of quan-
tum electrodynamics, with interaction energies resulting from the exchange of
photons between the two molecular species. When the propagation time of
photons exchanged between the fragments, tprop = R/c, with c the speed of light,
is comparable to or larger than the periods of charge oscillation τ = λrad/c = h/
ΔE, with λrad the radiation wavelength and h the Planck constant, or equivalently
when the distance R is larger than or equal to the wavelength of light emitted and
absorbed in the interaction, one must in addition consider retardation effects. We
find, using the smallest excitation energies (ΔE)A, B of a pair of fragments, that for
ch
R
ΔE A, B
we must include retardation effects in molecular potentials. The result from
perturbation theory is also an expansion in R−1, with different powers from
before. For two neutral molecules, the leading term goes as R−7, instead of
R−6 in the absence of retardation [2].
 1.3 Quantal Treatment and Examples of Molecular Interactions 9

At very large distances, the electromagnetic contributions are very small and
may become less important than magnetic (relativistic) contributions coming
from spin couplings, and these must be accounted for.
Effects resulting from the electron spin coupled to its orbital angular momen-
tum are in principle present at all distances, particularly for open-shell systems.
They can be treated by means of a quantal perturbation treatment for the states
ΦΓ(X; Q) due to spin-orbit energy couplings, which lead to splittings of inter-
action potential energies.
Combining short- and long-range functional forms for two neutral and
isotropic species gives model potentials, with two common ones being
n 6
6 Rm n Rm
V R = Vm −
n−6 R n −6 R

the Lennard–Jones n − 6 potential with parameters n, and Vm and Rm for the

minimum energy and radius, and
6
6 R α Rm
V R = exp α −1 −
α− 6 Rm α −6 R

the exponential-6 potential, with parameters α, Vm, and Rm. The Lennard–Jones
potential function is usually chosen to have n = 12, with Vm = ϵ, and is also writ-
ten using instead the distance R0 = Rm/21/6, where the potential energy is zero so
that V(R) = 4ϵ[(R0/R)12 − [(R0/R)6].

1.3 Quantal Treatment and Examples of Molecular

Interactions
Energies can be obtained within a quantal treatment of molecular states
[16–19]. We can to begin with, fix nuclei in space at positions r n or alterna-
tively using variables {Qj} = Q, where the Qj are bond coordinates (distances,
angles) and the distance between centers of mass of two molecules, to define
a Hamiltonian operator H Q containing all energies except the nuclear kinetic
energy. It depends on all the electronic variables (positions and spin) collected
in X. We have for the total energy Hamiltonian operator, using carats to
signify operators on functions of coordinates, and adding the nuclear kinetic
energy K n ,
H = Kn + HQ
H Q = K el + Vne + Vee + Vnn
10 1 Fundamental Concepts

and the equation for stationary molecular states,

H Q Φ X; Q = E Q Φ X; Q
where Φ(X; Q) = X Φ(Q) is the electronic eigenstate of energy E, a function of
all the electronic variables and dependent on the nuclear coordinates, shown as
parameters. Here H Q is the Hamiltonian operator for fixed nuclear positions, so
that it includes the electronic kinetic energy, nuclear–electronic, electronic–
electronic, and nuclear–nuclear Coulomb charge energies but no nuclear
kinetic energy. Alternatively, the energy E(Q) of the molecular system can be
obtained as the expectation value of H Q , using the Dirac bra-ket notation,

E Q = Φ Q HQ Φ Q = dX Φ X; Q ∗ H Q Φ X; Q

where a normalization Φ(Q) Φ(Q) = 1 has been used. This energy may be
decomposed into contributions from the several terms in the Hamiltonian oper-
ator and is the so-called Born–Oppenheimer or adiabatic potential energy.
Other properties (such as the electric dipole), represented by an operator AQ ,
are also obtained for fixed nuclei as averages,
A Q = Φ Q AQ Φ Q
An accurate treatment of the electronic state must describe the electrons as
moving in a self-consistent field shaped by the antisymmetry properties of the
wavefunction, as given usually by the Hartree–Fock formulation to incorporate
electronic exchange energies [16], and must in addition contain corrections for
the correlation energy of electronic motions, spin-orbit coupling, and magnetic
(relativistic) effects, so that
E Q = Eel Q + Enn Q
Eel Q = EHF + Ecorr + ESO + Emagn
where Enn(Q) is the nuclear–nuclear repulsion energy, and all the terms depend
on the nuclear coordinates.
The electronic state also provides information on electronic density functions
of electronic positions, ρ r ; these can be decomposed into atomic core ρc and
valence ρv electron terms, with the first leading to nuclear screening and to
core–core interactions. Electron exchange leads to a “Pauli pressure,” which
explains the distortion and repulsion of atomic cores as they overlap.
The energy contributions can be described for the case of H2+ in ground and
excited electronic states. Here, we have
K el + Vne φμ r; R = εμ R φμ r; R
e2
Eμ R = εμ R +
4πε0 R
 1.3 Quantal Treatment and Examples of Molecular Interactions 11

where εμ(R) and φμ r; R are electronic energy and orbital function at internu-
clear distance R, and we have added the electronic energy of the electron orbital
to the nuclear Coulomb repulsion energy to obtain the total adiabatic potential
energy Eμ(R). Figure 1.2a and b, from [20], shows the electronic orbital
energy εμ(R) and the diatomic potential energy Eμ(R) = Uμ(R) versus the inter-
atomic distance R in atomic units of energy (Hartree energy), with 1 au(E) =
4.35979 × 10−18 J, and length (Bohr radius), with 1 au(L) = 5.2918 × 10−11m
= 52.918 pm. The electronic binding energies are negative and increasing as
the nuclei move apart, with the limits at R = 0 and at infinite R given by the elec-
tronic energies for He+ and H + H+, respectively, and are labeled by molecular
point group symmetry symbols. After adding the positive repulsive energy of the
nuclei in Figure 1.2b, the lowest (or ground) total potential energy Eg(R) shows a
minimum that defines the static equilibrium distance Re and displays a net pos-
itive repulsion energy at short distances.
Similarly, for H2 in ground and excited electronic states, energies depend only
on the internuclear distance R. Figure 1.3 shows the total, electronic plus n–n
repulsion contributions for the ground state Σg+ . Here, the separated atoms
(S.A.) limit is H(1s) + H(1s), while the electronic united atom (U.A.) limit is
He(1s2) and has a lower electronic energy. The contributions are again found
to add to a potential with a minimum, attractive at long distances and repulsive
at short distances. The contributions to the excited Σu+ electronic state add up
instead to a purely repulsive potential and here the united atom limit is
He(1s2p), with a larger electronic energy [22].
A different situation arises for LiF, where potential curves show avoided
crossings between states 11Σ+ and 21Σ+ as seen in Figure 1.4. The electronic
ground state and five electronically excited states have been obtained with a
many-electron treatment including electronic correlation energies. Electron
transfer occurs as the distance R increases, and couples the state 11Σ+ going
asymptotically to Li + F with the state 21Σ+ going to Li+ + F−. A full treatment
of the atomic interactions here requires that one also specifies the interstate
coupling energy due to atomic displacements, leading to transitions between
the two nearly degenerate potential energies.
Potential energies for more than two atoms require special treatments to
account for energy functions of many atomic position variables. Here, we give
some general examples and postpone details to the following chapters. Conti-
nuing with polyatomic intermolecular potentials, consider a molecular system
AB breaking up into fragments A and B. The potential energies are functions of
the nuclear position vectors. The molecular energy (electrons plus fixed nuclei)
is E AB Q = EelAB Q + Enn AB
Q , where the first term is the sum of electron
kinetic energies plus electron–electron repulsion and electron–nuclear attrac-
tion, and the second term is the nuclear–nuclear repulsion energy. The nuclear
position variables are collected in Q = R, q , with R the relative position vector
(a)
0
3σu
–1–
8 1πg 3σg 1
3s, 3p – 2
– to – –
9 2σg 8
(3d) 2σu

2s, 2p –0.5 1πu

1σu to – 0.5
E (hartree)

–1.0

1σg

–1.5

1s –2.0
0 1 2 3 4 5 6 7
He+ R (bohr) H + H+

(b)

0.3

0.2

0.1
3σu
1πg
0
2σu
2σg
U (hartree)

–0.1

–0.2
1πu 3σg

–0.3

–0.4
1σu = σu*1s

–0.5

1σg = σg1s
–0.6

0 2 4 6 8 10 12 14
R (bohr)

Figure 1.2 (a) Electronic orbital energy εμ(R) for ground and excited states of an electron in
the H2+ molecular ion versus the distance R between the two protons, labeled by symmetry
symbols. The limits at R = 0 and at infinite R are the electronic energies for He+ and H + H+,
respectively. Source: adapted from Slater [20]. (b) Total potential energy Eμ(R) = U(R) of ground
and excited states of H2+ versus interatomic distance R. Source: adapted from Slater [20].
 1.3 Quantal Treatment and Examples of Molecular Interactions 13

–0.5

–0.6 a3 Σ+g

Σ1 Σ+g
–0.7 B1 Σ +u
U (hartree)

–0.8

–0.9
b3 Σ +u

–1.0

–1.1 X1 Σ +g

–1.2
0 1 2 3 4 5 6 7 8 9 10
R (bohr)

Figure 1.3 Ground and excited total potential energies for several states of H2 versus the
internuclear distance R. The limiting electronic energies for infinite R are ground and excited
H + H energies. Source: from Kolos and Wolniewicz [21]. Reproduced with permission of AIP
Publishing.

between the centers of mass of fragments A and B, and with q = (qA,qB) their
internal and orientation coordinates. We can work in a reference frame with
its z-axis passing through the centers of mass of A and B, so that the interaction
potential energy is given by
VAB R,q = E AB R,q −E A q A − E B q B
where R is the distance between centers of mass, which becomes infinitely large
as the pair breaks up.
Specific examples can be given for polyatomic intramolecular and intermo-
lecular potentials. Starting with intramolecular potentials, we describe cases
where only two nuclear coordinates or degrees of freedom Q1 and Q2 are chan-
ging, which leads to potential energy surfaces (PESs) U(Q1, Q2) in a three-
dimensional depiction. The energy can alternatively be represented by contours
of equal energy, or equipotential isocontours, in the (Q1, Q2) plane.
14 1 Fundamental Concepts

1.00

31Σ+
0.75
2 1Π Li–(1S) + F+(1D)
11∆
0.50
E (hartree)

0.25

21Σ+ Li+(1S) + F–(1S)

0.00
11Π Li(2S) + F(2P)
11Σ+
–0.25
2.0 4.0 6.0 8.0 10.0 12.0
R (bohr)

Figure 1.4 Potential energies E versus internuclear distance R for ground and excited states
of LiF. The electronic ground state and five electronically excited states have been obtained
with a many-electron treatment including electronic correlation energies. Source: from Yagi
and Takatsuka [23]. Reproduced with permission of AIP Publishing.

(a) (b) (c)

r BC

r A,BC r AB r AB

Figure 1.5 Typical PESs versus atomic positions for a triatomic ABC system with a fixed bond
angle between AB and BC. (a) Isocontours for the interaction energy of a noble gas atom A
interacting with a diatomic BC; (b) isocontours for the reaction A + BC AB + C, showing the
location of an activation energy barrier; (c) isocontours for a stable compound ABC in a
potential well, connected to dissociation valleys. Source: from Figure 2.3 of Smith [7].
Reproduced with permission of Elsevier.

Figure 1.5 shows typical energy isocontours of PESs: (i) with a single valley, as
for He + H2, versus the center of mass relative and internal coordinates R = rA, BC
and rBC; (ii) showing a rearrangement, e.g. H + H2 H2 + H, with a saddle point
between two valleys in a PES versus bond distances; and (iii) a well for a stable
species such as H2O connected by a saddle point to valleys for dissociations [9].
 1.3 Quantal Treatment and Examples of Molecular Interactions 15

3.0

2.0
yH (Å)

S
1.0

C N
A B
0 • •
–3.0 –2.0 –1.0 0.0 1.0 2.0 3
xH (Å)

Figure 1.6 HCN PES isocontours for the HCN HNC isomerization, with H changing position
on a x–y plane and C and N fixed on the x-axis. The locations of minima are shown as A and B,
for the two isomers, and a saddle point as S. Source: from Hirst [10]. Reproduced with
permission of Taylor & Francis.

Figure 1.6 is an alternative description and shows the HCN HNC isomer-
ization by displaying the interaction energy of the hydrogen atom with the CN
fragment as the atom is displaced between minima near C and N.
Plots of potential energies versus torsion angles for macromolecular systems
also offer examples of surfaces with several minima. For example Figure 1.7
shows several minima of the potential energy of a polypeptide chain versus
the consecutive dihedral angles ψ and φ. In all these cases, figures are only show-
ing the lowest PES. The PESs of electronically excited states would have differ-
ent shapes.
Figures 1.8 and 1.9 give specific examples of dissociation in HCN H + CN
and N + CH, versus bond distances, and of atomic rearrangement from reac-
tants to products in F + H2 FH + H.
More than one PES may be involved in electron transfer or in electronically
excited systems, involving two position variables Q1 and Q2, or more. Two
potential energy functions Vg(Q1, Q2) and Ve(Q1, Q2) for ground and excited
electronic states, drawn in a 3D picture of V versus (Q1, Q2) with isocontours
of constant energy, typically show intersections where couplings due to atomic
displacements can be large [26]. Depending on how the electronic states couple
through those displacements, the surfaces can intersect along a line forming a
seam, or at a point giving a conical intersection. Seams are features of interac-
tion potentials involving electron transfer in collisions between ions and neutral
molecules, such as found in collisions of H+ + H2 [27] and of H2+ + Ar [28]. Con-
ical intersections appear in electron exchange between molecules [29], in iso-
merization [30], and in photoinduced chemistry [31].
16 1 Fundamental Concepts

180° Figure 1.7 Peptide link (a glycyl

residue of a polypeptide chain)
isocontours of constant energy
versus the torsion angles phi
around the C–N bond and psi
around the C–C bond (a
Ramachandran diagram). Ovals
lead to energy minima. Source:
𝜓 from Hovmoeller [24]. Reproduced
with permission of John
Wiley & Sons.

–180°
–180° 𝜙 180°

3.0

14
RCN (Å)

2.0
12

10

v1 8
v3
1 2 3 4 5 6
1.0

1.0 2.0 3.0

RCH (Å)

Figure 1.8 HCN PES versus bond distances C–H and C–N for the dissociation HCN H + CN.
Contour 1 is for the energy – 13.75 eV, and other contours are drawn at intervals of 1 eV.
Source: from Hirst [10]. Reproduced with permission of Taylor & Francis.
 1.3 Quantal Treatment and Examples of Molecular Interactions 17

Figure 1.9 F + H2 FH + H PES 20

for the collinear FHH 0.4
configuration and contour
energies given in kcal mol−1 (with
1.0 kcal mol−1 = 4.184 kJ mol−1).
Source: from Bender et al. [25]. 0.3
Reproduced with permission of
AIP Publishing.

rHH (nm)
0.2

+27
0.1 +5
+1.6

0.1 0.2 0.3 0.4

rHF (nm)

Figure 1.10 shows the intersection of two potentials for collinear Ar(1S) + H2+
( Σ) and Ar+ (2P)+ H2 (1Σ) and higher potentials versus the H–H distance when
2

Ar is far removed, and also potentials for ArH+ and ArH as a H is far away. The
potentials for H2+ and H2 versus the H–H distance are seen to cross, but when
Ar is brought in this becomes a seam between two PESs, and their coupling
leads to a linear avoided crossing of the PESs versus the Ar–H distances.
A similar situation arises for H+ + H2 H2 + H+, or H2+ + H, involving
two PESs, versus bond distances, but here the lowest one shows an energy well
for H3+ . The new feature in both cases is that the PESs have an avoided crossing
or “seam” running along the entrance valley, corresponding to electron transfer
as an approaching atom moves in.
Figure 1.11 shows two PESs interacting to give conical intersections in the
isomerization of the C2H2 ethylene molecule. The picture at the top corre-
sponds to ground and excited electronic states for isomerization into HC2H3,
ethylidene, while the one at the bottom shows the ground and excited PESs
for the pyramidalization intermediate C2H2∗ as obtained in [32]. Here, the vari-
ables are g and h, themselves functions of all the atomic positions, with g mea-
suring the strength of the coupling of the states by the atomic displacements,
and h giving the magnitude of energy splitting between the two PESs [31].
Similar treatments and images apply to interactions of atoms and molecules
with an extended system, such as a solid surface or a cavity in a liquid.
Figure 1.12 shows the PES of He approaching the surface LiF(001) of the solid,
with a face-center-cubic (or fcc) structure for a lattice of F− and Li+ ions.
The surface is perpendicular to a z-axis with the origin of a surface plane
18 1 Fundamental Concepts

H2 and H2+ ArH+ and ArH

6 6
2Σ 1Σ
3Σ u 1Π
u

4 4 3Σ
3Π

2Σ
Ar++H+H
Energy (eV)

2 2

1Σ
0 g 0 Ar+H++H

2Σ
g
–2 –2
1Σ

2 4 6 2 4 6 8
RH–H (bohr) RAr–H (bohr)

Figure 1.10 Potential energies for collinear ArH2+ showing their change with atom–atom
distances, while the third atom is far away, with energies given relative to Ar + H+ + H. The
intersection in H–H curves leads to the appearance of a seam between potential energy
surfaces as the Ar atom is brought in. Source: from Baer and Beswick [28]. Reproduced with
permission of American Physical Society.

(x, y) at a F− ion location, a cell cube side of 2.84 Å and a Li+ at the center of the
face. The isocontours reflect the periodicity of the surface and show deep wells
on the x–z plane above each ion, for two chosen y values corresponding to lines
along the negative and positive ions. A similar figure can be drawn for an atom
approaching a molecule adsorbed on a solid surface.
All these PESs display special points where forces are null, defined by
∂E
=0 Q0j = Q 0
∂Qj
for at least one set {Q0j }, which locates maxima, minima, or saddle points; these
can be distinguished by calculating the second derivatives ∂ 2E/(∂Qj ∂Qk) at the
special points. The dynamics of motion on PESs are closely related to their
topography as defined by the geometry of the special points. For example, min-
ima are connected by paths of steepest ascent and descent along which one finds
activation barrier energies, which in turn account for reaction rate magnitudes.
A different conceptual approach is needed to describe the energetics of react-
ing molecules in a medium such as a liquid solvent or a solid matrix, and in par-
ticular electron transfer in liquid solutions [34, 35]. Sets r n of many atomic
 1.3 Quantal Treatment and Examples of Molecular Interactions 19

Ethylidene
0.4

0.3
Energy (eV)

0.2

0.1
0.06
0.03
0.0 0.00

or
ct
–0.03

ve
0.03
0.00 –0.06
g
–0.03
h vecto –0.06
r

1.25 Pyramidalization

1.00
Energy (eV)

0.75

0.50

0.25
0.06
0.03
0.00 0.00
r
to
c

–0.03
ve

0.03
0.00 –0.06
g

–0.03
h vecto –0.06
r

Figure 1.11 Conical intersections in the isomerization of the C2H2 ethylene molecule. The
picture at the top corresponds to ground and excited electronic states for isomerization into
HC2H3, ethylidene, while the one at the bottom shows the ground and excited PESs for the
pyramidalization intermediate C2H2∗ as obtained in [32]. Source: from Ben-Nun and Martinez
[32]. Reproduced with permission of Elsevier.

position variables are involved and one must instead work with collective struc-
ture variables Y r n which account for many atom rearrangements for given
thermodynamical constrains such as temperature T, density ρ, and pressure p.
Potential energies VΓ r n ,Γ = g,e, must be replaced by thermodynamical
internal energies UΓ(Y, T, ρ) or Gibbs free energies GΓ(Y, T, p).
20 1 Fundamental Concepts

3.0
b –2.0

–3.0
Z(Å)
–4.0
2.0 –5.0
–6.0
–7.0
–8.0
–9.0
–10.0
1.0 –11.0
–7.0 –12.0 –7.0
–5.0 –5.0

0 0
10.0 10.0
0
0 2.84 X(Å) 5.68
3.0
c –2.0

Z(Å) –3.0

–4.0
2.0
–5.0
–6.0

–7.0
–5.0 –5.0
1.0

0 0
10.0 10.0
20.0 20.0
30.0 30.0
0
0 2.84 X(Å) 5.68

Figure 1.12 He approaching the surface LiF(001) of the fcc solid. The surface is perpendicular
to a z-axis with the origin of a surface plane (x, y) at a F− ion location, a cell cube side of 2.84 Å
and a Li+ at the center of the face. The isocontours in meV values reflect the periodicity of the
surface and shows deep wells on the x–z plane above each ion, for two chosen y values of
0.00 Å (upper panel) and 1.42 Å (lower panel) corresponding to lines along the ions. Source:
from Wolken [33]. Reproduced with permission of Springer Nature.

Figure 1.13 shows sketches of two Gibbs free energy surfaces for ground and
excited electronic states Gg(Y, T, p) and Ge(Y, T, p) of reactants donor–acceptor
pair D–A undergoing charge transfer into products pair D+–A−, such as a
Fe2+/Fe3+ mixed valence compound, in a liquid solution, with avoided conical
 1.4 Long-Range Interactions and Electrical Properties of Molecules 21

Excited state

MEP Photochemical funnel

(conical intersection)
R*
Reactive trajectory R*
MEP
Nonreactive
trajectory
hv P CI

Free energy
Transition hv
state P
TS
Ground state R

MEP Excited state

Ground state
R Reaction coordinate

Figure 1.13 Gibbs free energy surfaces (FESs) for ground and excited electronic states in a
system containing reactants R donor and acceptor compounds D and A in a liquid solution
and the products P electron transfer ions D+ and A−. (a) Contours of FESs versus collective
variables Y1 and Y2 chosen to be average liquid cavity radii around fragments D and A,
showing minimum energy paths. (b) Free energy along the Ys reaction path coordinate
through a transition state between the two minima. Source: from Schapiro et al. [36].
Reproduced with permission of Royal Society of Chemistry.

intersections similar to the ones mentioned above now shown instead for free
energies. The variables chosen here are average liquid cavity radii RD = Y1 and
RA = Y2 around donor and acceptor fragments. The ground surface has two
wells, for ground electronic states of the D–A and D+–A− arrangements, con-
nected by a path Ys at the surface through a saddle point. The reaction energy is
ΔGR and the splitting between electronically adiabatic free energy surfaces is
given by 2GR0 . The upper surface displays a conical shape with a single minimum
for the average radii of cavities around the two fragments interacting when the
electronic state is excited.

1.4 Long-Range Interactions and Electrical Properties

of Molecules
1.4.1 Electric Dipole of Molecules

Molecules usually have permanent electric dipole vectors D, except when they
have specific symmetries, such as inversion symmetry, which make the
dipole null. Dipole physical units are Cm in the SI system, or the Debye, with
1 D = 10−18 cgs-esu = 3.335 × 10−30 Cm. Molecular dipoles are anisotropic and
22 1 Fundamental Concepts

Monopole can be estimated by adding the dipoles of its

atom–atom bonds [16]. Examples are HF, HCl,
Dipole HI, H2O, CO2, O3, C6H6, and CH4.
More generally, multipoles can be constructed
as point charge models, with the monopole or
20 − pole given by a single charge, a dipole or
Quadrupole 21 − pole given by two opposite charges at a dis-
tance L, and so on, as shown in Figure 1.14.

1.4.2 Electric Polarizabilities of

Molecules
Octupole
A molecule in a static electric field of compo-
nents ξ , ξ = x, y,z, acquires an induced dipole
Dη = ξ αηξ ξ along a direction η proportional
Figure 1.14 Point charge to the field strength, where αηξ is a dipolar
models of electrical multipoles.
polarizability tensor. To simplify here, we con-
sider that both field and dipole are along the
z-direction with D = α , where α is the zz-dipolar polarizability. For large fields,
an expansion in the field strength gives also hyperpolarizabilities, with
D = α + β 2 2+ … involving higher polarizabilities.
Dipoles and polarizabilities can be obtained from quantum mechanical
perturbation theory as detailed in following chapters. Second-order perturba-
tion theory shows that the polarizability is inversely proportional to molecular
excitation energies. Quantum mechanical perturbation theory gives, in
0
terms of unperturbed energies En and molecular wavefunctions Φn0 , static
polarizabilities
2
Dn0
α=2 n 0 0 0
En −E0
0
where Dn0 = Φn0 D Φ0 is a matrix element of the dipole operator. This
shows that electronic polarizabilities tend to be larger for molecules with small
electronic excitation energies, such as small highest occupied/lowest unoccu-
pied (HOMO/LUMO) molecular orbital excitation energy. This gives larger
polarizabilities for larger molecules with delocalized electrons.
The polarizability volume is given by αv = α/(4πε0), where ε0 = 8.854 ×
10−12 J−1 C2m−1 = 1/(4π) cgs-esu is the vacuum electrical permittivity. The value
of αv can be estimated for a molecule of size L using an average excitation energy
ΔE ≈ e2/(4πε0L), with e the electron charge in the SI units and a transition dipole
Dn0 ≈ eL, giving αv ≈ L3.
 1.4 Long-Range Interactions and Electrical Properties of Molecules 23

The response of a molecule to an oscillating electric field with frequency ω is

given by the dynamical polarizability α(ω), obtained from second-order time-
dependent perturbation theory in the form

αω = 2 ℏ n 0
Dn0 2 ωn0 ωn0 2 −ω2
0 0
with ℏωn0 = En −E0 . As ω increases, this function goes through infinite values
at the transition frequencies and shows regions of nearly constant values in
between for each type of motion: rotational, vibrational, and electronic, corre-
sponding to the range of frequencies for radio, microwave, infrared, visible, and
UV waves. A more detailed treatment accounts for the lifetime of excited states
and gives finite polarizability values at transition frequencies, as shown in the
following chapter.

1.4.3 Interaction Potentials from Multipoles

Long-range potential energies for two interacting molecules are obtained from
their charge, dipole, quadrupole, etc., interactions both permanent or induced.
Interaction between a 2m-pole of species A and a 2n-pole of species B can be
called a (2m)A−(2n)B interaction.
Interaction potential energies between multipoles can be obtained from
charge models for molecules of size L when the distance between centers of
mass satisfies R L, with each one containing a collection of charges at given
positions, by expanding the molecule–molecule interaction potential energy,
written as a sum of charge pair interactions, in powers of L/R.
One finds that for the monopole–monopole interaction of charges CA and CB
at distance R,
CA CB
20 A
− 20 B
V R =
4πε0 R
For a charge–dipole interaction, with the dipole of B consisting of charges CB
and −CB at distance L, oriented along angles ωB = (θB, φB),
C A DB
20 A
− 21 B
V R = fB ω B
4πε0 R2
with DB = CB L. The dipole–dipole interaction energy is
DA DB
21 A
− 21 B
V R = fAB ωA , ωB
4πε0 R3
m
In general, the (2m)A − (2n)B interaction between multipoles of magnitude A
n
and B is of the form
24 1 Fundamental Concepts

m n
m, n
2m A− 2 R = fAB ωA ,ωB
n A B
B VAB
4πε0 Rm + n + 1
Induction energies are obtained from the interaction of an induced multipole
with a permanent multipole. For the (induced dipole)–dipole interaction,
B
this involves the induced dipole DA = αA A and the electric field
B
A = −DB 4πε0 R3 ) created at A by dipole B. The interaction energy is built
B
increasing the field by increments d A as the polarized species is brought
B
closer to the permanent multipole so that in increments dVD D = − DA d A ,
which after integration over field values gives, using the polarization volume
of A,
2
B
αA A αv, A DB 2
VD D R = − =−
2 2πε0 R6
Dispersion energies originate in (induced multipole)–(induced multipole)
interactions, when charge fluctuations create transient dipole. For the interac-
tion between transient dipoles D A = αA A and D B = αB B , where each field is
created by the transient dipole at the other molecule, the energy builds up to
2 2
B A
αA A αB B CDD
VDD R = − − =−
2 2 R6
which is the van der Waals attractive long-range interaction, with CDD > 0. This
coefficient can be obtained from a quantum mechanical treatment and can be
approximated in terms of the polarizabilities and ionization potentials of species
A and B, as shown in following chapters.

1.5 Thermodynamic Averages and Intermolecular

Forces
1.5.1 Properties and Free Energies
The total interaction energy of a macroscopic system such as a gas or condensed
matter (liquid, solid, surface, etc.) at a temperature T must be obtained by doing
statistical averages for given statistical distributions of the system components
and their physical properties. The thermodynamical internal energy or the
Gibbs free energy can be obtained from a classical treatment with probability
distributions of the Maxwell–Boltzmann type, or from quantal statistical
averages using thermal distributions of quantum states [37–39].
 1.5 Thermodynamic Averages and Intermolecular Forces 25

The classical statistical average of a many-atom property F(Q, PQ), where Q is

the collection of all internal molecular position variables and center-of-mass
locations, and the PQ are their related momenta, is obtained from the Max-
well–Boltzmann distribution wcl Q,PQ ; T ,V , , for a system at temperature
T and with = N V molecules per unit volume in a volume V, in the form

F T , V , = dQ dPQ F Q,PQ wcl Q,PQ ; T ,V ,

In particular, the average energy E T , V , = ,V , T is the internal energy. It is

usually possible to obtain correct thermal averages of potential energies for a
collection of molecules with a classical procedure, provided the system is not
electronically excited.
In a quantal description, one first obtains the expectation value of the
property, now given by the operator F Q,P Q , a function of position and
momentum operators, in each stationary quantum state Φn(Q) with energy
En V , and the quantal statistical probability wn T ,V , . The statistical
average follows from

FT,V , = n
F n wn T , V ,
i∂
F = Φn F Φn = dQ Φn Q F Q, − Φn Q
n
∂Q
When the operator is the energy of the system, the above procedure leads to
expressions for the free energy and internal energy. For a closed system of
molecules per unit volume, at given temperature and with volume V, the Helm-
holtz free energy ,V , T is obtained from the thermal partition function
Qth ,V ,T = n exp − En kB T by means of [37]
= −kB T ln Qth ,V ,T
from which the internal energy follows as ,V ,T = −T ∂ ∂T V .
Changing variables from volume in to pressure p gives the Gibbs free energy
,p,T .

1.5.2 Polarization in Condensed Matter

A molecular fluid subject to an external electric field acquires a polarization
per unit volume = χ el ε0 , where χ el is the electrical susceptibility and ε0 is the
vacuum permittivity. This polarization can be obtained from an average molec-
ular polarizability α per unit volume, which depends on the temperature T and
from the density (number of molecules per unit volume ) of the fluid as an
statistical average = α loc , where loc is an average local electric field at
the position of the molecule and depends on intermolecular forces. It is
26 1 Fundamental Concepts

approximately given by the Lorentz formula loc = + 3ε0 . For polar mole-
cules one must also obtain the average dipole per unit volume, related to the
permanent molecular dipole. This gives D T = D2 loc 3kB T for the thermal
average of the permanent molecular dipole D, to which one must add the
induced dipole. The average permanent dipole is the limit, for small fields ,
of a more general Langevin formula for the average [16].
In a dielectric (a fluid which does not conduct electricity), the presence of
polarization alters the Coulomb interaction between charges, by changing
the electric permittivity from ε0 to ε = ε0εr. The relative permittivity εr is given
by the Debye equation
εr − 1 ρ m
=
εr + 2 M
NA D2
m = α+
2ε0 3kB T
where NA is the Avogadro number, ρ is the mass density, M the molar mass, and
m is the molar polarization. This formula can be used also for dynamical pert-
mitivities when the field oscillates with a constant frequency, and also gives the
electrical susceptibility insofar as χ el = εr − 1. When the permanent dipole is
null, the equation gives a way to measure the polarizability, extracting it from
optical measurements of the refractive index nr(ω) = c/cmed(ω) = εr(ω)1/2, where
c is the speed of light in vacuum and cmed in the fluid medium. The average
polarization m ω depends on the intermolecular forces in the fluid and pro-
vides information about them.

1.5.3 Pair Distributions and Potential of Mean-Force

The potential energy VN of a fluid containing N molecules can frequently be
expressed as a sum over interactions v of pairs (jk) of molecules so that

VN = v
1 ≤ j < k ≤ N jk

in which case the internal energy of the fluid can be obtained from a pair cor-
relation function g2(R, Ωj, Ωk) for molecules j and k at a distance R and with
orientations given by angles Ωj and Ωk. Introducing a thermal average g 2 R
over a distribution of orientations (Ωj, Ωk; T), the average total potential energy
Epot = pot ,V ,T is a function of the temperature and density which can be
given in terms of the pair correlation function g 2 R; T , by [37]

pot N2 3
, V ,T = d R v R g 2 R; T ,
2V
It is also useful to introduce a statistical pair potential of mean force
w2 R; T , by means of g 2 R; T , = exp −w2 R; T , kB T . It can be
 1.6 Molecular Dynamics and Intermolecular Forces 27

interpreted as the interaction between two molecules at a distance R when a

thermal average is done over the remaining N − 2 molecules so that the forces
between them are both direct and also indirect through the medium.
For the dipole–dipole interaction of a pair at fixed distance R, the orientation
average gives the Keesom formula,
2 2
m n
f2 sph A B
vDD R T = vDD 2
R =−
4πε0 R3 2 kB T
which involves an orientation factor f squared after the leading thermal average
term disappears. The original dipole–dipole interaction going as R−3 has become
an average interaction going as R−6 as a result of orientational averaging [1].
Using a sum of pair potentials also helps to derive an equation of state. The
pair potential can be written as a sum of a short-range hard sphere repulsion
term v(SR)(R) = 0, R ≥ R0, plus a long-range attraction term v(LR)(R) = − C6/R6.
The first one entails that molecular motions occur outside an excluded volume
per pair 2b = 4πR30 3, and the second term means that the total attraction energy
∞
is attr = aN 2 V with a = − 2π R0 dR R2 v LR R [37]. The equation of state is
found then to be the well-known van der Waals equation, (p + aN2/V2)(V −
Nb) = NkBT.

1.6 Molecular Dynamics and Intermolecular Forces

1.6.1 Collisional Cross Sections
Molecular species A and B can be prepared in molecular beams with given rel-
ative velocities and in specific internal states. Their collision in a crossed beam
region leads generally into new species C and D emerging from the collision
region at solid angles ΩP where they are detected. Measurements can be stated
L
in terms of differential collisional cross sections Iβα ΩP ,vrel from reactants A +
B moving with relative velocity vrel in a reactant channel α to products C + D in
channel β, and with species R = C or D detected at angles ΩP [8, 40, 41].
Cross sections can be calculated from a knowledge of the interaction potential
energy between reactant and product species, in terms of the incoming reactant
flux Jα0 , as the number of incoming reactants A relative to B per unit area and
unit time, and the rate N βα of the number of transitions from reactants to pro-
ducts per unit time. For an increment dΩP in detection angle, the rate increment
L L L
is dN βα = Jα0 dσ βα with dσ βα = Iβα dΩP an increment of cross sectional area. The
integral cross section is obtained from
L L
σ βα vrel = dΩP Iβα ΩP , vrel
28 1 Fundamental Concepts

Calculations of these cross sections as functions of ΩP, vrel and the channel
states, α, starting with the interaction potential energies and the structures of
reactants and products, can be compared to experimental measurements to
provide very detailed information on potential energy functions and
parameters.
These calculations can be very demanding of computational resources. In
some cases, when transitions among rotational, vibrational, and electronic
states of the molecules have been averaged, calculations can be done with a clas-
sical mechanics treatment using equations of motion for classical positions and
momentum variables Q(t), PQ(t) of the constituent atoms. More accurately, the
state-to-state cross sections must be obtained from quantal treatments which
may involve time-dependent wavepackets or time-independent (stationary)
wavefunctions. The stationary wavefunction ΦE for a total energy
α
E = Erel + Eint is a function of nuclear an electronic position variables. Its asymp-
totic form as the distance R between the products C and D becomes large can be
written as ΦE ≈ Φ(0) + Φ(sc) for R ∞, with Φ(0) giving the incoming flux Jα0 and
the scattered wave Φ(sc) giving scattering amplitudes fβα(ΩP, vrel) from which the
L 2
differential cross section Iβα ΩP ,vrel = pβ pα fβα ΩP , vrel can be calculated
[40]. The wavefunction ΦE satisfies the time-independent Schroedinger equa-
tion with a Hamiltonian operator which contains the potential energy of the
interacting reactant and product species, and therefore is shaped by the inter-
molecular forces.
A related subject is transport rates of molecular momentum (viscosity),
kinetic energy (thermal conductivity), and mass (diffusion) in real gases and
liquids. They are linked to intermolecular forces through collision integrals
which can be calculated in many realistic situations, particularly when thermal
averages allow use of classical mechanics and deflection functions [1, 8, 33].

1.6.2 Spectroscopy of van der Waals Complexes

and of Condensed Matter
The spectroscopy or diffraction pattern of a molecule interacting with light of
frequencies extending from radiowaves to UV waves provides extensive infor-
mation on molecular parameters such as rotational and vibrational constants
and oscillator strengths for electronic transitions [16]. Calculations of absorp-
tion–emission light shapes and diffraction intensities for given molecular
structures, starting from their electronic structure, can be compared with
results of measurements to obtain details of potential energy surfaces for
atoms near equilibrium positions. A similar procedure can be followed for
molecular complexes where molecules are held together by intermolecular
forces to form van der Waals molecular pairs, but in this case, the theoretical
treatment is complicated by the appearance of floppy motions and quantal
 1.7 Experimental Determination and Applications of Interaction Potential Energies 29

tunneling effects, due to small energy variations among minima, maxima, and
barriers in the intermolecular potentials.
The treatment of large amplitude motions in van der Waals complexes can be
done separating the nuclear position variables into a set of position values {uj} =
u which give small displacements from equilibrium positions for rigid structural
regions of the complex, and a remaining set {qk} = q of position variables which
can range over large distances. The complex dynamics and related spectra are
treated starting from a Hamiltonian containing the potential energy V(u, q) and
kinetic energy terms Ku, Kq, and Kuq. Terms containing u are expanded in its
powers and treated as done for standard molecular spectra, while Kq and Kuq
can be expressed in terms of a mass tensor G = Gkk q,q and generalized
momenta p = − iℏG − 2 ∂ ∂q G2 conjugate to q, with G = det[G]. Examples of
1 1

applications can be found for van der Waals complexes like ArCH4 and ArC6H6
and hydrogen bonded complexes like (H2O)2 and (HCl)2 [42, 43]. This approach
is also applicable to the study of transient (or collision-induced) complexes
detected through light absorption–emission in a collision region. The subject
of van der Waals complexes continues to be quite active and has been recently
reviewed in [44].
The mentioned treatment can be generalized to include scattering boundary
conditions applicable to photodissociation of van der Waals complexes, which
can be considered to be half-collisions and can be described as mentioned above
with channel amplitudes for collisional phenomena [45].
Intermolecular forces participate in molecular spectra in condensed phases
such as real gases, liquids, and molecular solids. They shape absorption–
emission lines, Raman spectra, and electron transfer spectra through solvation
effects. These spectra can be calculated and compared with experimental mea-
surements to obtain information about the functional form and parameters of
involved interaction potential energies [46–48]. Some of these aspects are
considered in Chapter 7 on interactions of two many-atom systems.

1.7 Experimental Determination and Applications

of Interaction Potential Energies
Knowledge on interaction potential energies can be extracted from experiments
measuring properties that depend in a known way on molecular interactions.
This requires introduction of realistic theoretical models and the ability to cal-
culate measurable properties, for a range of choices on the form of the potential
energies and their parameters. Experimental methods can be classified in three
categories: (i) thermodynamic; (ii) spectroscopic and diffraction; and (iii) beam
methods. Of these the last one provides the most detailed results when cross
sections can be obtained as functions of deflection angles and relative velocities,
30 1 Fundamental Concepts

and for state-to-state collisional transitions, insofar it deals with isolated sys-
tems of one or two molecules, but it is limited by the necessity to produce
and detect beams of interacting species in known internal states. Spectroscopic
methods are even more accurate, but they mostly provide information around
equilibrium conformations. This is changing with the recent introduction of
time-resolved spectroscopy methods. Following are lists of some relevant meas-
urable properties.

1.7.1 Thermodynamics Properties

• Virial coefficients in equations of state of real gases versus temperature T and

pressure p.
• Heat capacities and cohesive energies of solids and liquids versus temperature
T and pressure p.
• Compressibilities, sound absorption, and dispersion in solids and liquids
versus temperature T and pressure p.
•• Adhesive energy of solid surfaces versus temperature T and pressure p.
Viscosity, diffusion, and heat conduction transport coefficients of real gases
versus temperature T and pressure p.
• Relaxation rates of temperature or pressure of real gases near equilibrium
temperature T and pressure p.

1.7.2 Spectroscopy and Diffraction Properties

• Molecular spectra of weakly bound (van der Waals) atomic and molecular
complexes formed in beams.
• Collision-induced spectra in gases versus light wavelengths at given temper-
ature T and pressure p.
• Line shapes in condensed matter versus light wavelengths at given tempera-
ture T and pressure p.
• Relaxation times of quantum state populations in condensed matter versus
temperature T and pressure p.
• Electron, neutron, and light diffraction intensities in condensed matter versus
temperature T and pressure p.
• Time-resolved spectral intensities versus light wavelengths in pulsed laser
(pumping–probing) experiments.

1.7.3 Molecular Beam and Energy Deposition Properties

• Cross sections for elastic, inelastic, and reactive collisions in crossed molec-
ular beams.
• Energy loss cross sections for beams traversing fluids or solids.
 References 31

• Photodissociation of atomic and molecular complexes and of clusters formed

in beams.
• Cross sections for scattering of atoms and molecules by solid surfaces and by
adsorbates.

1.7.4 Applications of Intermolecular Forces

Intermolecular forces are essential in many applications of molecular and mate-
rials properties to technologies contributing to social needs. Some properties
(and their applications) are as follows:

•• Storage of hydrogen in solids (fuel cells)

Storage and transport of ions in solids (batteries)
• Synthesis of thermally stable and conducting surfaces for solar devices (solar
panels)
•• Delivery of compounds through biological cell membranes (pharmacology)
Catalysis and photocatalysis in electrochemical cells (sustainable fuel production)
•• Atmospheric reactions (environmental sciences)
Efficient fuel combustion (transportation and energy)
• Solvation and lubricants (machinery)

References
1 Hirschfelder, J.O., Curtis, C.F., and Byron Bird, R. (1954). Molecular Theory of
Gases and Liquids. New York: Wiley.
2 Hirschfelder, J.O. (ed.) (1967). Intermolecular Forces (Adv. Chem. Phys. vol. 12).
New York: Wiley.
3 Margenau, H. and Kestner, N.R. (1971). Intermolecular Forces, 2e. Oxford,
England: Pergamon Press.
4 Dainton, F.S. (ed.) (1965). Faraday Society Discussions (Intermolecular Forces
vol. 40). London, England: Royal Society of Chemistry.
5 Lawley, K.P. (ed.) (1980). Potential Energy Surfaces (Adv. Chem. Phys. vol. 42).
New York: Wiley.
6 Pullman, B. (ed.) (1982). Intermolecular Forces. Dordrecht, Holland: Reidel
Publishing Company.
7 Smith, I.W.M. (1980). Kinetics and Dynamics of Elementary Gas Reactions.
London, Great Britain: Butterworth.
8 Maitland, G.C., Rigby, M., Brain Smith, E., and Wakeham, W.A. (1981).
Intermolecular Forces. Oxford, England: Oxford University Press.
9 Murrell, J.N., Carter, S., Farantos, S.C. et al. (1984). Molecular Potential Energy
Functions. New York: Wiley.
32 1 Fundamental Concepts

10 Hirst, D.M. (1985). Potential Energy Surfaces. London, England: Taylor &
Francis.
11 Stone, A.J. (2013). The Theory of Intermolecular Forces, 2e. Oxford, England:
Oxford University Press.
12 Israelachvili, J. (2011). Intermolecular and Surface Forces, 3e. New York: Elsevier.
13 Kaplan, I.G. (2006). Intermolecular Interactions, 81. New York: Wiley.
14 Born, M. and Oppenheimer, J.R. (1927). Zur Quantentheorie der Molekeln. Ann.
Phys. 84: 457.
15 Born, M. and Huang, K. (1954). Dynamical Theory of Crystal Lattices, Appendix
VIII. Oxford, England: Oxford University Press.
16 Atkins, P.W. and Friedman, R.S. (1997). Molecular Quantum Mechanics.
Oxford, England: Oxford University Press.
17 Schatz, G.C. and Ratner, M. (1993). Quantum Mechanics in Chemistry.
Englewood, New Jersey: Prentice-Hall.
18 Cohen-Tanoudji, C., Diu, B., and Laloe, F. (1977). Quantum Mechanics, vol. 1
and 2. New York: Wiley.
19 Messiah, A. (1962). Quantum Mechanics, vol. 1 and 2. Amsterdam: North-
Holland.
20 Slater, J.C. (1963). Quantum Theory of Molecules and Solids, vol. 1. New York:
McGraw-Hill.
21 Kolos, W. and Wolniewicz, L. (1965). Potential‐energy curves for the X Σg ,
1 +

b Σu , and C Πu states of the hydrogen molecule. J. Chem. Phys. 43: 2429.

3 + 1

22 Levine, I.N. (2000). Quantum Chemistry, 5e. New Jersey: Prentice-Hall.

23 Yagi, K. and Takatsuka, K. (2005). Nonadiabatic chemical dynamics in an intense
laser field: Electronic wave packet coupled with classical nuclear motions.
J. Chem. Phys. 123: 224103.
24 Hovmoeller, T. (2002). Conformations of amino acids in proteins. Acta Cryst
D58: 768.
25 Bender, C.F., Pearson, P.K., O’Neil, S.V., and Schaefer, H.F. III (1972). Potential
energy surface including electron correlation for the chemical F+H2 FH+H I.
Preliminary surface. J. Chem. Phys. 56: 4626.
26 Baer, M. (2006). Beyond Born-Oppenheimer: Electronic Nonadiabatic Coupling
Terms and Conical Intersections. New York: Wiley.
27 Preston, R.K. and Tully, J.C. (1971). Effects of surface crossing in chemical
reactions: the H3+ system. J. Chem. Phys. 54: 4297.
28 Baer, M. and Beswick, J.A. (1979). Electronic transitions in the ion-molecule
reaction Ar+ + H2 to Ar + H2+ and ArH+ + H. Phys. Rev. A 19: 1559.
29 Worth, G.A. and Cederbaum, L.S. (2004). Beyond Born-Oppenheimer:
molecular dynamics through a conical intersection. Annu. Rev. Phys. Chem.
55: 127.
30 Levine, B.G. and Martinez, T.J. (2007). Isomerization through conical
intersections. Annu. Rev. Phys. Chem. 58: 613.
 References 33

31 Domcke, W. and Yarkony, D.R. (2012). Role of conical intersections in molecular

spectroscopy and photoinduced chemical dynamics. Annu. Rev. Phys. Chem.
63: 325.
32 Ben-Nun, M. and Martinez, T.J. (2000). Photodynamics of ethylene: ab initio
studies of conical intersections. Chem. Phys. 259: 237.
33 Wolken, G.J. (1976). Scattering of atoms and molecules from solid surfaces.
In: Dynamics of Molecular Collisions Part A (ed. W.H. Miller), 211. New York:
Plenum Press.
34 Marcus, R. (1964). Chemical and electrochemical electron-transfer theory.
Annu. Rev. Phys. Chem. 15: 155.
35 Hush, N.S. (1961). Adiabatic theory of outer sphere electron-transfer reactions
in solution. Trans. Faraday Soc. 57: 557.
36 Schapiro, I., Melaccio, F., Laricheva, E.N., and Olivucci, M. (2011). Using the
computer to understand the chemistry of conical intersections. Photochem.
Photobiol. Sci. 10: 867.
37 McQuarrie, D.A. (1973). Statistical Mechanics. New York: Harper & Row.
38 Gray, C.G. and Gubbins, K.E. (1984). The Theory of Molecular Fluids. I.
Fundamentals. Oxford, England: Clarendon Press.
39 Kittel, C. (2005). Introduction to Solid State Physics, 8e. Hoboken, NJ: Wiley.
40 Child, M.S. (1974). Molecular Collision Theory. New York: Academic Press.
41 Lawley, K.P. (ed.) (1975). Molecular Scattering (Adv. Chem. Phys. vol. 30).
New York: Wiley.
42 Wormer, P.E.S. and van der Avoird, A. (2000). Intermolecular potentials,
internal motions, and spectra of van der Waals and hydrogen-bonded
complexes. Chem. Rev. 100: 4109.
43 Beswick, J.A. and Jortner, J. (1981). Intramolecular dynamics of van der Waals
molecules. In: Photoselective Chemistry, Part 1 (Adv. Chem. Phys. vol. 47) (eds. J.
Jortner, R.D. Levine and S.A. Rice), 363. New York: Wiley.
44 Hobza, P. and Rezak, J. (eds.) (2016). Non-Covalent Interactions. Special Issue of
Chemical Review, vol. 116. Washington DC: American Chemical Society.
45 Lawley, K.P. (ed.) (1985). Photodissociation and Photoionization (Adv. Chem.
Phys. vol. 60). New York: Wiley.
46 Mukamel, S. (1995). Principles of Nonlinear Optical Spectroscopy. Oxford,
England: Oxford University Press.
47 May, V. and Kuhn, O. (2000). Charge and Energy Transfer Dynamics in
Molecular Systems. Berlin: Wiley-VCH.
48 Nitzan, A. (2006). Chemical Dynamics in Condensed Phases. Oxford, England:
Oxford University Press.
 35

Molecular Properties

CONTENTS

2.1 Electric Multipoles of Molecules, 35

2.1.1 Potential Energy of a Distribution of Charges, 35
2.1.2 Cartesian Multipoles, 36
2.1.3 Spherical Multipoles, 37
2.1.4 Charge Distributions for an Extended System, 38
2.2 Energy of a Molecule in an Electric Field, 40
2.2.1 Quantal Perturbation Treatment, 40
2.2.2 Static Polarizabilities, 41
2.3 Dynamical Polarizabilities, 43
2.3.1 General Perturbation, 43
2.3.2 Periodic Perturbation Field, 47
2.4 Susceptibility of an Extended Molecule, 49
2.5 Changes of Reference Frame, 52
2.6 Multipole Integrals from Symmetry, 54
2.7 Approximations and Bounds for Polarizabilities, 57
2.7.1 Physical Models, 57
2.7.2 Closure Approximation and Sum Rules, 58
2.7.3 Upper and Lower Bounds, 59
References, 60

2.1 Electric Multipoles of Molecules

2.1.1 Potential Energy of a Distribution of Charges
The size of a molecule is of the order of a nanometer, while the wavelength of
visible light is about 500 nm, and larger for infrared or radio waves. Therefore,
the amplitude of an externally applied electric field can be assumed to be constant
over space except in two situations: (i) for large atomic systems such as clusters,
polymers, surfaces, or proteins; or (ii) for inhomogeneous fields created by a

Molecular Interactions: Concepts and Methods, First Edition. David A. Micha.

© 2020 John Wiley & Sons, Inc. Published 2020 by John Wiley & Sons, Inc.
36 2 Molecular Properties

neighbor molecule. These cases can be included in a general treatment by letting

the electric field r,t be a function of position r and time t. We ignore here the
magnetic field associated to it because the interaction energy of a molecule with
a magnetic field is much smaller than the interaction with the electric field, and
will concentrate on electric multipoles and polarizabilities. Magnetic field
effects can be considered as perturbations involving the magnetic properties
of the molecules, such as their electronic spin and orbital magnetic moments.
We consider a molecule at rest with its center of mass located at the origin of a
coordinate system . To start with we treat all charged particles as classical ones
with well-defined positions and will later on allow for their quantal nature. The
presence of electric charges in space creates an inhomogeneous electric
potential ϕ r (also called U r in some books). If a molecule with its own
distribution of charges {CI} at locations r I (with I = a indicating a nucleus
and I = i an electron) enters this region of space, the potential energy function
of particle positions is given by VMF = I CI ϕ r I , due to the molecule-field
interaction. This energy can be expressed in terms of the electric potential
and its space derivatives around the origin of coordinates. In effect, using
Cartesian coordinates ξ = x, y, z, and expanding in a series around ξ = 0,
∂ϕ 1 ∂2 ϕ
ϕ r =ϕ 0 + ξ + ξη +
ξ ∂ξ 0 2 ξ, η ∂ξ∂η
∂ϕ 1 ∂2 ϕ
VMF = ϕ 0 C +
I I
ξC +
I I I
C I ξI η I +
ξ ∂ξ 0 2 ξ, η ∂ξ∂η 0
I

where summations over I display in order from the left: total charge C = ICI ,
total dipole Cartesian component Dξ = IξICI , total quadrupole, and so on. We
use the relation Ce = e/(4πε0)1/2 between the electron charge Ce and its symbol e
in esu units, with ε0 the vacuum permittivity.
These expansions can be rewritten in terms of the electric field vector com-
ponent ξ = − ∂ϕ ∂ξ, and its space derivatives, and are valid whether the electric
field is internal to a molecular system, or externally applied. Simplifications
arise when the electric field is external and the molecule is not large so that
the field can be assumed to be constant over its extent. Then the quadrupolar
and higher terms can be omitted and a quantal treatment of the molecular
properties can be derived from a quantized Hamiltonian operator
H MF = ϕ 0 I CI − ξ ξ I ξI CI , where operators are shown with car-
ats [1–4].

2.1.2 Cartesian Multipoles

The distribution of nuclear and electronic charges in a molecule can be
described in terms of electrical multipoles [1, 2, 5]. These can be constructed
from classical mechanics and then converted into quantal operators.
 2.1 Electric Multipoles of Molecules 37

Let particle I (a nucleus or electron) have charge CI and be located at r I with

components ξI = xI, yI, zI. Then the dipole vector component along direction ξ is
given by

Dξ = I
C I ξI

and its units are Coulomb × meter (or Cm) in the SI system or e.a0 in atomic
units; another usual unit is 1 Debye = 1 D = 10−18 esu. The relation among them
is 1 au(dip) = 2.5418 D = 8.478 × 10−30 Cm. The quadrupole tensor component
along (ξ, η) is defined by

Qξη = I
CI 3 ξI ηI − rI2 δξη

where δξη equals one for ξ = η and zero otherwise, and rI2 = x2I + y2I + zI2 ,
constructed so that the trace ξQξξ = 0. An alternative definition of the
quadrupole in the literature is Θξη = Qξη/2.
In general, the Cartesian components ξ, η, … ζ of the n-th electric multipole
are given by
n
n −1 ∂ ∂ ∂ 1
Mξ, η, …ζ = CI rI2n + 1
n I ∂ζ ∂η ∂ξ r I

which is traceless when adding over a pair of identical indices, as follows from
the relation ∇2(1/r) = ξ ∂ 2 r−1/∂ξ2 = 0.

2.1.3 Spherical Multipoles

The analytical form of electric multipoles changes as the molecule is rotated. To
describe that change, it is more convenient to work with spherical components
of the multipole, and the corresponding spherical components of the electric
potential gradients. Components of rotational number l ≥ 0 and projection
−l ≥ m ≥ l are given by

1 2
4π
Qlm = CI rIl Ylm ϑI ,φI
2l + 1 I

written in terms of the spherical harmonic function Ylm of position angles, with
well-known transformation properties under rotation. From Y00 ϑI , φI =
1 4π one finds that Q00 is the total charge. Dipole and quadrupole multipoles
follow from l = 1 and l = 2, respectively, and are related to Cartesian compo-
nents, so that, for example Q11 = Dx + iDy 2 = Q∗1, −1 and Q10 = Dz. The
potential energy of interaction takes the form VMF = lm(−1)mQlmFl, −m, where
the coefficients are related to the electric field and its derivatives. See [4, 6, 7] for
details.
38 2 Molecular Properties

2.1.4 Charge Distributions for an Extended System

Here one introduces the density of charge per unit volume c r = I CI δ r −r I ,
as a sum of Dirac delta functions over all the point charges in the system, elec-
trons (with I = i) and nuclei (with I = a). Alternatively, the sum is over valence
electrons and ionic cores. One can analyze the distribution in terms of Cartesian
or spherical components and relate them to the previous results. For example
the dipole ξ-components of the extended system are Dξ = d 3 r ξ c r and other
multipoles can be obtained from higher moments of the charge density with
respect to position components. However, such multipole description is not
practical for an extended system, but possibly only for its molecular fragments.
Instead, it is more general to work with the total charge density.
When particles are treated within quantum mechanics, the charge density
becomes an operator c r = I CI δ r −r I , and its average values must be
obtained from expectation values over quantal states. These are functions
Φk(X, Q) of all the electronic variables (positions and spins) X and all nuclear
(or ionic core) positions Q. Provided electronic and nuclear motions are not
very strongly coupled, as it happens for low electronically excited states, the
state functions are given in the Born–Oppenheimer approximation by products
nu
Φk X, Q = ΦJ Q Φαel X; Q of a many-electron wavefunction for fixed
nuclei and a nuclear motions wavefunction describing vibrations and rotations
of the molecule. The electronic functions Φαel X; Q must be antisymmetric
with respect to electron exchanges and can be constructed as superpositions
of electronic determinantal wavefunctions. The nuclear wavefunctions
nu
ΦJ Q depend on variable distances between nuclei and on molecular orien-
tation angles and describe vibrational and rotational motions for each given
electronic state.
It is convenient to separate nuclear and electronic charge densities and to use
the Born–Oppenheimer approximation where electronic states are generated for
fixed nuclear positions Q, with the nuclei treated to begin with as classical par-
ticles so that c r = c nu r + c el r with c nu r = a Ca δ r −r a and c el r =
Ce iδ r −r i This gives for an electronic state Φαel X; Q = X Φαel Q ,
using the Dirac bra-ket notation and integrating over electronic variables, the
quantal expectation value for the averaged electronic density cαel r; Q =
Φαel Q c el r Φαel Q , and a total charge density cα r,Q = c nu Q +
cαel r;Q .
This can be treated by decomposition into physically or mathematically con-
venient terms, introducing a variety of approaches: (a) A decomposition into
atomic and bond densities (or atomic charge- and bond-order sum) suitable
 2.1 Electric Multipoles of Molecules 39

for large molecules; (b) decomposition into fragments densities for molecules in
liquid solutions and at surfaces; (c) plane-wave decompositions for extended
systems with many atoms; and (d) decompositions into finite elements for con-
tinuum models. These are described in some detail as follows:
a) The charge density operator can be represented by a matrix in a basis set of
atomic orbitals, and matrix elements can be classified as atomic- or bond-
like. If an electronic structure can be accurately described by a basis set of
atomic orbitals χ νn centered at nucleus n and with atomic quantum numbers
ν, then a related basis set of localized orthonormal orbitals φνn can be intro-
duced using the overlap matrix Δ with elements Δμm, ν, n = χ μm χ νn and the
m, n
symmetric transformation orbital φνn = m, μ χ μm Δ−1 2 , which is
μν
large at location n and contains small contributions of nearby atomic orbi-
tals. It provides a partial completeness relation I = mμ φμm φμm . With
the identity I operating to right and left of the charge density operator, this
becomes a sum over single atom and atom-pair terms,
2
c el r = m, μ
φμm φμm r φμm
∗
+ mμ nν
φμm φμm r φνn r φνn

The expectation value cαel r for a given electronic state α can be interpreted
as an atomic charge- and bond-order sum, and terms can be given physical
meaning [8].
b) Molecular fragments are identified by a subset of nuclear positions, usually
chosen with well-known structure, each with its own multipoles and polar-
izations constructed to reproduce the fragments density, and the total charge
density can be given for the combined fragments in the form of a sum over all
the distributed multipoles and polarizations [4]. In each fragment, the multi-
poles and polarizations are located at chosen points a and are expressed in
terms of localized functions. These are conveniently written using Gaussian
exponentials, which can be readily translated between locations, of the form
1 2
4π
La lm r = ral Ylm ϑa , φa exp −ζra2
2l + 1
where (ra, ϑa, φa) are angular variables for r − a.
c) The density can be expanded in plane waves for a collection of vectors
selected to form a grid in a reciprocal space. This is convenient when the
charge densities are smoothly varying, which excludes ion core electrons.
It is usually done enclosing the molecule in a volume Ω = L1L2L3 with
40 2 Molecular Properties

periodic boundary conditions which restrict the reciprocal vectors to a denu-

merable set qn with n = (n1, n2, n3) a triplet of integers. The density oper-
ator is Fourier transformed from space to reciprocal space using that
qn qn = Ω −1 d3 r exp − iqn r exp iqn r = δnn , as
Ω
d3 r
cr = exp iqn r cn , cn = exp − iqn r c r
Ω Ω
n

and the operator coefficients are used in interactions. This expansion is com-
mon in treatments of solid-state properties and is particularly useful for poly-
mers, surfaces, and solids with periodic atomic structure [9].
d) The density can be numerically represented by a decomposition of a contin-
uous distribution of charges into three-dimensional finite elements. Each
element is a solid tetrahedron P3 = a0 + a1x + a2y + a3z with coefficients
fixed by density values at the four corners. A set of tetrahedra can cover a
general distribution of the density and can be used in the treatment of
two interaction densities. Finite elements have been introduced in treat-
ments of optical properties of atomic clusters [10] and are convenient also
for biomolecular distributions.
Molecular interactions can be described in terms of these density decomposi-
tions. They are all approximations, and their accuracy must be tested by system-
atic improvement of the decomposition procedures.

2.2 Energy of a Molecule in an Electric Field

2.2.1 Quantal Perturbation Treatment
The energy of a molecule in an electric field must be obtained in quantum
mechanics as the expectation value of the energy operator for the system.
The starting point is the quantal Hamiltonian for a set of charged particles in
an external electromagnetic field (e.m.) described by the vector and scalar
potentials A r,t and ϕ r,t at each point in space. The electric field follows
from r,t = − ∂ A ∂t − ∂ϕ ∂ r. For a homogeneous external field without
external charges, ϕ = 0 and the divergence ∇ A = 0, which simplify the molec-
ular internal charge-field interaction as H MF = − I CI A pI mI to lowest order
in the field strength, with pI the charged particle momentum. The Hamiltonian
can be further simplified when the wavelengths of the e.m. field is long com-
pared to molecular dimensions so that only a dipolar coupling to an external
electric field is needed [6].
This gives a Hamiltonian operator H = H M + H MF with two terms, corre-
sponding to the energy operator of the molecule in free space plus the operator
 2.2 Energy of a Molecule in an Electric Field 41

form of the molecule-field interaction energy H MF = − ξ Dξ ξ t for an exter-

nally applied electric field of components ξ . This field can vary with position
for a large molecule, but is otherwise constant over the molecular extension.
When the maximum applied external electric field amplitude ext is small
compared to internal electric fields in the molecule, of magnitude int , one
can use the parameter λ = ext int as the perturbation expansion variable, with
0 1
HM = H and H MF = λH . Perturbation solutions for the wavefunction can be
0
obtained using a basis set of unperturbed molecular states Φl , solutions of
0 0 0 0
the Hamiltonian equation H Φl = El Φl with the orthonormalization con-
0 0 0
dition Φl Φm0 = δlm and assuming completeness, or l Φl Φl = I,
the identity operator [6]. The molecular wavefunctions Φk(X, Q) = X, Q Φk
can be taken as Born–Oppenheimer (adiabatic) products Φk X,Q =
nu
ΦJ Q Φαel X; Q , with the bracket notation here indicating integration over
both electronic and nuclear position variables.
The perturbation treatment depends on whether the perturbing field is static
or changing with time. In the first case, it involves solving for the time-
0 1
independent Schroedinger equation, H + λH 0 Φl = El Φl to obtain sta-
tionary states of fixed energy, while for a time-dependent field, the wavefunction
0 1
must satisfy the equation of motion H + λH t Ψ t = iℏ∂Ψ ∂t, with an
initial condition Ψ(0) = Ψ determined by the state of the molecule before it
(in)

is perturbed.
Solutions can be generated using operator methods, or in more detail intro-
ducing an expansion in the basis set of unperturbed states. This expansion gives
physically meaningful expressions involving matrix elements of the multipole or
charge density operators. The ground and excited states must be generated and
the treatment is accurate provided enough excitations can be considered.
A general and compact perturbation treatment can also be done introducing
resolvent operators and a basis set partitioning method. This also facilitates treat-
ment of perturbation of a degenerate energy eigenvalue, with several unperturbed
states of this same energy [6, 11–13]. An operator-based treatment is needed
when the molecule is very large and its relevant unperturbed states are not all
known. In this case, it is yet possible to account for property values by expanding
in chosen basis sets, or with an expansion in a set of operator amplitudes.

2.2.2 Static Polarizabilities

For a field constant in time, properties can be obtained from the eigenvalue
solutions of the quantal Hamiltonian equation HΦk = Ek Φk . The stationary
molecular states Φk and their energies Ek for each state k depend on the
42 2 Molecular Properties

magnitude of the applied field and are given by the products Φk X, Q =

nu
ΦI Q Φαel X; Q with the electronic functions Φαel X; Q antisymmetric
with respect to electron exchanges, and constructed as superpositions of elec-
tronic determinantal wavefunctions.
The response of a molecule to a time-independent electric field can be
expressed in terms of static polarizabilities, which can be obtained from a quan-
tum mechanical treatment using perturbation theory, insofar as external fields
are usually small compared to internal ones [5, 6, 14]. Expanding states and their
energies in powers of the field strength parameter λ one has
0 1 n p q
H + λH n≥0
λ n Φk = p≥0
λ p Ek q≥0
λ q Φk

and equating coefficients of each power one finds for n = p + q ≥ 1 that, after
0∗
multiplication by Φk and integration over its variables,

1 0 1 0 1 0 1 0 0 0 0
Ek = Φk H Φk , Φk = l k
Φk H Φl Φl Ek − El

for first-order corrections, and that to second order

1 2
2 0 0 0 0
Ek = l k
Φk H Φl Ek − El

with similar expressions for higher-order perturbation corrections. Replacing

1
λH = − ξ ξ Dξ in these equations one obtains the field dependence of the
energy Ek at each order.
The total average dipole component Dξ equals the permanent molecular
dipole plus a dipole induced by the field, and this contains as the leading terms
the molecular dipole polarizability plus higher polarizabilities. It is obtained as
the derivative of the total energy with respect to the corresponding electric field
component so that for a chosen state k = a it is Dξ a = − ∂Ea ∂ ξ . This can be
expanded in powers around ξ = 0 to obtain

0 1
Dξ a
= Dξ , a + α
η ξη, a η
+ β
η, ζ ξηζ , a η ζ+
2
∂ 2 Ea
αξη, a = −
∂ ξ∂ η 0

∂ 3 Ea
βξηζ, a = −
∂ ξ∂ η∂ ζ 0

where αξη is a dipolar polarizability of the molecule and βξηζ is a dipolar

hyperpolarizability. These tensor components are usually related by molecular
 2.3 Dynamical Polarizabilities 43

symmetry and depend on the molecular orientation with respect to a reference

frame. The polarizability is found from the second-order correction to the
energy as

0 0
Φa0 Dξ Φl Φl Dη Φa0 +c c
αξη, a = l a 0 0
El − Ea

which is positive valued when a is the ground state, of lowest energy. Here c.c.
stands for complex conjugate of the preceding form. The hyperpolarizability
requires calculation of the third-order correction to the energy and can be found
from recursion relations.
The static polarizability is related to the transition dipole for each state-to-
state transition, and the largest contributions to the polarizability come from
large transition matrix elements and small excitation energies. They can be cal-
culated separating nuclear and electronic dipole operators and using the Born–
Oppenheimer factorization of states.
The Qξη a Cartesian components of the total quadrupole are similarly
obtained differentiating the total energy with respect to the field gradient tensor
components ξη = −∂ 2 ϕ ∂ξ∂η, to obtain the permanent molecular quadrupole
plus its quadrupolar polarization components.
An alternative description is based on the expansion of the molecule-field
interaction in terms of the spherical components of the dipole operator and
of the electric field. Derivatives of the energy with respect to the spherical com-
ponents of the electric field give the total spherical components of the dipole, as
sums of a permanent value plus a value derived from the polarizability and con-
tributions from hyperpolarizabilities [5]. Using spherical components, it is pos-
sible to give compact expressions for higher multipoles [4].
Molecular multipoles can be calculated as quantal averages so that the total
n n
molecular n-pole average is Mξ, η, …ζ = Φa M ξ, η, …ζ Φa , a function of the
a
applied electric field. Permanent n-pole and polarizabilities follow from the per-
turbation expansion keeping up to second-order terms.

2.3 Dynamical Polarizabilities

2.3.1 General Perturbation
The response of a molecule to a time-dependent electromagnetic field can
instead be expressed in terms of dynamical polarizabilities, which follow from
a quantal treatment and time-dependent perturbation theory [6]. As men-
tioned, the Hamiltonian can be simplified when the wavelengths of the e.m. field
44 2 Molecular Properties

is long compared to molecular dimensions so that only a dipolar coupling to an

external electric field is needed [6]. The associated Schroedinger equation of
motion must be considered to obtain the time-dependent wavefunction Ψ(t)
of the molecule plus field, or more generally the time-evolution operator
describing a general perturbation. It will be done here in particular for an
external electric field which is initially null at times t < 0 and grows to a given
oscillating field of constant amplitude and is homogeneous over the molecular
region. The perturbation expansion at time t leads to expressions involving
integrals over earlier time t and integral kernels which are functions of t and
t , with Fourier transforms providing response functions. The treatment can
also be conveniently done introducing time-delayed propagators [13].
The Hamiltonian is H = H M + H MF t , with H M a time-independent Hamil-
tonian for the isolated molecule in its stationary states, and H MF t = −Dz z t .
As before, when the maximum value 0 of a time-dependent applied external
electric field amplitude is small compared to internal electric fields in the
molecule, of magnitude int , one can use the parameter λ = 0 int as the per-
0 1
turbation expansion variable, with H M = H and H MF t = λH t , so that the
time-dependent equation of motion for the state is

0 1
H + λH t Ψ t = iℏ∂Ψ ∂t

and it can be solved with the wavefunction expansion Ψ(t) = n ≥ 0λnΨ(n)(t)

when the molecule at times t < 0 is in a steady state Φa exp(−iEat/ℏ). After a
short interval when the transient effects of excitation have disappeared, this
can be taken to be the unperturbed state at times t ≥ 0 with Ψa0 0 = Φa .
The perturbative treatment for dynamical polarizabilities can be done for a
0
time-independent unperturbed Hamiltonian H using the time evolution
0 0
operator U t = exp − i H t ℏ for the unperturbed problem, which satis-
fies the equation
0 0 0 0
H U t = iℏ∂U ∂t, U 0 =I

−1
t † , its
0 0
This operator is unitary, and its inverse satisfies U t =U
0
adjoint. Writing Ψ t = U t Ψ I t and solving the full equation of motion
for the interaction picture states Ψ(I)(t), which are constant in the absence of
a perturbation so that Ψ(I, 0)(t) = Ψ(I)(0) and have the initial value Ψ(I)(0) = Ψ(0),
gives the integral equation
t
i
t †H
0 1 0
Ψ I t =Ψ I 0 − λ dt U t U t ΨI t
ℏ 0
 2.3 Dynamical Polarizabilities 45

A solution by iteration of this integral equation provides the desired expan-

sion in powers of λ. The first term to the right is the unperturbed solution of the
time-dependent Schroedinger equation. The solution to first order is
t
i
t †H
0 1 0
Ψ I, 1 t = − dt U t U t ΨI 0
ℏ 0

and one has in the interaction picture to n-th order the recursive relation
t
i
t †H
0 1 0
Ψ I, n t = − dt U t U t Ψ I , n− 1 t
ℏ 0

Replacing results in the average total dipole, one finds the results for
permanent and induced dipoles. Writing for the dipole the ex-
0 1 2
pansion Dz t a = Dz, a t + λ Dz, a t + λ2 Dz, a t 2 + … one finds that
0
Dz, a t = Ψa0 t Dz Ψa0 t , and that

Dz1, a t = Ψa0 t Dz Ψa1 t + Ψa1 t Dz Ψa0 t

This first correction gives the induced dipole (or polarization) Pz, a t =
1
λDz, a t as

t † Dz Ψa0 t
0 0
Pz, a t = λ Ψa0 t Dz U t ΨaI , 1 t + λ ΨaI , 1 t U

and can be expressed in terms of the integral over time as

t
i
Pz , a t = − dt Ψa0 0 Dz t Dz t −Dz t Dz t Ψa0 0 z t
ℏ 0

where the time-dependent dipole operator Dz t = U t † Dz U t appears

0 0

multiplying at two different time, in a quantal average giving a dipole-dipole

time-correlation function. The induced linear polarization is found to be
dependent on earlier values of the applied field and contains a delayed response
to its application. This treatment is valid for a general perturbation and takes a
special form for a harmonic perturbation.
In practice, calculations are usually done introducing the basis set of unper-
turbed states Φk with energies Ek, to expand time-dependent states. In a com-
pact Dirac notation with kets Φk = k , bras k , and brackets indicating
integration over state variables,

Ψt = k
k exp −i Ek t ℏ ck t
dck
iℏ = k H t l −Ek δkl exp iωkl t cl t
dt l
46 2 Molecular Properties

where ωkl = (Ek − El)/ℏ is a transition frequency, and the coupled equations
must be solved with the initial conditions ck(0) = δka for each initial state a.
n
Expanding in powers of the field strength with ck t = n ≥ 0 λ n ck t it follows
0 0 n
that ck t = ck 0 = δka and therefore ck 0 = 0 for n ≥ 1. The changes of
higher-order terms for n ≥ 1 satisfy the iterative equations
n
d ck 1 n −1
iℏ = k H t l exp iωkl t cl t
dt l

1 t 1
which gives ck t = 0 dt k H t a exp iωka t , linear in the field and
from an integral to be obtained for specific fields.
The average dipole takes values obtained from the coefficient. To the lowest
0
order, Dz, a t = a Dz a which is usually null by molecular symmetry of the
state a. To the next order
1 1 ∗
Dz1, a t = k
a Dz k ck t exp − iωka t + ck t k Dz a exp iωka t

is linear in the field strength and provides the polarizability. The induced dipole
or polarization is given by
t
Pz, a t = k
dt a Dz k k Dz a z t exp iωka t − t + c c
0
which can be integrated for a given time-dependent field.
More generally, the polarization component Pξ, a(t) arises in response to an
applied field η t present from time t = 0 at all times 0 ≤ t ≤ t, and can be
expressed in terms of delayed response functions for phenomena both linear
and nonlinear in the applied field, as [15]
t
1
Pξ, a t = η
dt1 αξη, a t,t1 η t1
0 t t1
2
+ η, ζ
dt1 dt2 αξηζ, a t,t1 ,t2 η t1 ζ t2 +
0 0
written here in general using tensor components of response functions. To sim-
plify, consider only the linear response in the first term, usually sufficient for stud-
ies of intermolecular interactions. For a time-independent unperturbed state a, it
is a function of the difference t = t − t1 and it is convenient to introduce the
+ 1
retarded susceptibility αξη, a t = θ t αξη, a t ,0 , where θ(t ) is the step function
null at negative times. Furthermore, the lower integration limit can be replaced
by −∞ insofar as the electric field is null for negative times, and changing the
integration variable from t1 to t gives the useful expression for the linear term,
∞
+
Pξ, a t = η
dt αξη, a t η t −t
−∞
 2.3 Dynamical Polarizabilities 47

which is a convolution form and has a Fourier transform from time to fre-
+
quency, giving P ξ, a ω = η α ξη, a ω η ω in terms of the transforms of the
three expressions, using the definitions
∞ ∞
dω
f ω = dt exp iωt f t , f t = exp −iωt f ω
−∞ − ∞ 2π
+
for a function f(t). The dynamical polarizability α ξη, a ω can be obtained from a
perturbation calculation of the average polarization. The retarded susceptibility
is in general
+ i
αξη, a t = − Ψ0 0 Dξ t Dη 0 − Dη 0 Dξ t Ψa0 0 θ t
ℏ a
where the time-dependent dipole density operator Dξ t = U t † Dξ U t
0 0

appears multiplying at two different time, in a dipole–dipole time-correlation

function.

2.3.2 Periodic Perturbation Field

So far the field is a general function of time and may be a superposition of har-
monic terms or a light pulse. When the applied field is harmonic of form
z t; ωL = 0 cos ωL t for positive times, the linear polarization acquires, after
a transient interval, the same frequency but has two components, one in phase
and one out of phase with the applied field. It can be written as

Pz, a t; ωL = αzzc , a ωL 0 cos ωL t + αzzs , a ωL 0 sin ωL t

∞
αzzc , a ωL = dταzz+, a τ cos ωL τ
0
∞
αzzs , a ωL = dταzz+, a τ sin ωL τ
0
l
The dynamical polarizabilities αzz, a ωL , l = c, s, can be derived from a pertur-
bation treatment.
We start a perturbation treatment with an electric field ξ t; ωL =
0 cos ωL t + φ fr t oscillating with the harmonic light frequency ωL and turned
on by a rising function fr(t) null at times t < 0 and tending to 1.0 for long times t
of interest, with a rise time parameter τ much smaller than measuring times of
interest. To begin with, we assume that the field is along the z-axis and that the
molecule is isotropic in state a. The polarization is then along the z-axis and
the total dipole Dz(t) a is the sum of a permanent dipole plus terms induced
by the field. Once transient polarizations have disappeared, and the
system has reached a steady state for the given light frequency, it is of the form
48 2 Molecular Properties

Dz(t; ωL) a = Dz, a(0) + Pz, a(t; ωL), the sum of the permanent dipole plus a polar-
ization changing with the same frequency and given in general by [15]
t
Pz, a t; ωL = dt1 χ zz1, a t,t1 ; ωL z t1 ; ω L
0
t t1
+ dt1 dt2 χ zzz
2
, a t,t1 ,t2 ; ωL z t1 ; ω L z t2 ; ω L +
0 0

which displays a dipolar polarizability in the term linear with the field and a first
hyperpolarizability in the quadratic term, as functions of the field frequency. To
obtain explicit expressions for them, we perform the integrations over time.
One can introduce a specific electric field along the z-direction in the form
z t; ωL = 0 cos ωL t 1−exp − t τ and integrate the previous expression
1
for ck t which give for t τ > 0
ωka + ωL t ωka − ωL t
1 0 ei −1 ei −1
λck t = k Dz a +
2ℏ ωka + ωL ωka − ωL

Replacing this in the form of the average dipole, it appears as a permanent

0
dipole Dz, a plus an induced dipole or polarization Pz,a(t; ωL), and the result
allows identification of the dynamical polarizability αzz(ωL)a, in
0
Dz t a = Dz, a + Pz, a t; ωL , as

Pz, a t; ωL = αzz ωL a 0 cos ωL t

2
2 ωka k Dz a
αzz ωL a =
ℏ k a ω2ka −ωL 2
This shows the polarizability as a function of the field frequency ωL with infi-
nite values as this goes through each transition frequency.
More accurately, a solution for the coefficient ca(t) to all orders in the cou-
pling of molecule and field reveals that it must decay exponentially over time
as state a makes spontaneous light-induced transitions to all states of lower
energy. This gives its magnitude as a decay function |ca(t)| = exp(−γ at/2) where
γ a is a decay rate. To account for this decay, it is convenient to return to the
1 0
iterative coupled equations for ck t and replace ck t = δka to the right with
d
ck t = δka exp − γ a t 2 [16]. It is then found that to second order in the mol-
ecule-field coupling
2
γa = k <a
2π ℏ2 a H MF k ρL ωka

where k < a signifies adding over states of lower energy, and ρL is a density of
light per unit frequency. With the same time-dependent coefficient changes
 2.4 Susceptibility of an Extended Molecule 49

used in the dipole average, one finds its previous stationary term replaced by a
d
decaying dipole Dz, a t = a Dz a exp − γ a t , and that the transition fre-
quency is replaced by ωka + iγ ka/2 with γ ka = γ k + γ a the sum of initial and final
decay rates, in the first-order coefficients. With this replacement, the dynamical
polarizabilities becomes
2
2 ωka k Dz a
αzz ωL a=
ℏ k a ωka − ωL 2 + γ 2ka 4

which avoids the function’s singularity at the transition frequencies and gives
instead peaks near resonance ωL = ωka excitations.
The present treatment can be readily generalized to the case where the field
and dipole components are not in the same direction. For a general orientation
of the field, the dynamical polarizabilities become tensors and the total dipole is
0
Dξ t a
= Dξ, a + Pξ, a t; ωL

Pξ, a t; ωL = α
η ξη, a
ωL 0η cos ωL t

1 ωka a Dξ k k Dη a
αξη, a ωL =
ℏ k a ωka −ωL 2
+ γ 2ka 4

with similar expressions for the quadrupolar and higher dynamical polarizabil-
ity tensors, replacing the dipole operator with a higher electrical multipole.
More generally, the decay rates originate in the interactions of the molecule
with its medium, with light, and with other molecules, and the rates can be
obtained extending the expansion basis set to include states with a continuum
of energies, which describes the photons and the molecules in its medium.
Solving the larger set of coupled equations for the expansion coefficients, the
interactions are found to give in addition an energy shift ΔEka and new transition
frequencies ωka = ωka + ΔEka ℏ [6].
An alternative treatment of decay can be developed quantizing the electro-
magnetic field, which introduces photons instead of the field and can describe
the interaction of the molecule with photons, including stimulated absorption-
emission and also spontaneous emission [14, 17].

2.4 Susceptibility of an Extended Molecule

A large molecule, an aggregate of molecules, a polymer, or a solid surface are
examples of systems where there are very large numbers of states which are usu-
ally unknown and which would make summations over states an unpractical
approach. Furthermore, the applied electric field may vary over the lengths
50 2 Molecular Properties

of the molecular system, and it is necessary to generalize the treatment to

include a position- and time-dependent electric potential ϕ r,t .
For an extended system of charges of electrons and nuclei (or ionic cores) at
operator locations r I , the total charge density operator per unit volume at space
location r is c r = I CI δ r −r I . The dipole operator is D = dr 3D r with

dipole density D r = r c r , and higher multipoles can be similarly defined

and used to expand the charge density operator as a series containing multipolar
polarizations. However, many such multipoles may be needed for an extended
system, and the treatment is best done avoiding expansions in multipoles.
Instead, it is better to work with the average, or expectation value, of the total
charge density, and to return to the potential energy expression given by charges
interacting with an electric potential, rewriting VMF = I CI ϕ r I , t as
V MF = dr 3 c r ϕ r,t and working directly with the charge density operator
[15, 18].
For a time-independent applied electric potential, the average charge density
0
in state a is a sum ca r = ca r + ca r of a permanent value plus an induced
charge density, and the latter can be derived theoretically using a perturbation
treatment with resolvent operators, which can be expressed in any suitable basis
set instead of having to use unknown unperturbed states.
For a time-dependent electric potential ϕ r, t , the average charge density in
0
state a is a sum ca r,t = ca r + ca r,t of a permanent value plus an induced
charge density changing over time, with the latter obtained from the density
operators c r,t , and response functions involving the time correlation of these
densities at two times.
It is convenient to proceed within a formal operator approach to obtain
expressions for a physical system in the form of response functions and to cal-
culate these without expanding in the usually unknown states for an extended
system. The average induced charge density ca r,t arises in response to an
applied electric potential ϕ r, t present from time t = 0 at all times 0 ≤ t ≤ t,
and can be expressed in terms of delayed response functions for phenomena
both linear and nonlinear in the applied field, as [15]
t
ca r,t = dt1 d 3 r1 χ a1 r,t; r 1 ,t1 ϕ r 1 ,t1
0
t t1
+ dt1 dt2 d 3 r1 d 3 r2 χ a2 r, t; r 1 ,t1 ; r 2 ,t2 ϕ r 1 , t1 ϕ r 2 , t2
0 0
+
written here using response functions of time and positions. To simplify, con-
sider only the linear response in the first term, usually sufficient for studies of
 2.4 Susceptibility of an Extended Molecule 51

intermolecular interactions. For a time-independent unperturbed state a, it is

a function of the difference t = t − t1, and it is convenient to introduce the
+ 1
retarded susceptibility χ a r,r 1 ; t = θ t χ a r, t ; r 1 , 0 where θ(t ) is the step
function null for negative arguments. Furthermore, the lower integration limit
can be replaced by −∞ insofar as the electric field is null for negative times, and
changing the integration variable from t1 to t gives the useful expression for the
linear term,
∞
ca r,t = dt d 3 r1 χ a+ r, r 1 ; t ϕ r 1 ,t − t
−∞

which is a convolution form and has a Fourier transform from time to frequency
+
giving c a r,ω = η d 3 r1 χ a r,r 1 ; ω ϕ r 1 ,ω in terms of the transforms of
the three expressions.
+
The dynamical susceptibility χ a ω can be obtained from a perturbation
calculation of the average polarization. Proceeding as before for the dipolar
response, now the charge polarization is found to be
i t
ca r, t = − dt d 3 r Ψa0 0 c r, t c r , t − c r , t c r, t Ψa0 0 ϕ r , t
ℏ 0

where the time-dependent density operator c r,t = U t † c r U t appears

0 0

multiplying at two different time. The bracket is in fact a function of only the
difference τ = t − t because the time evolution is done by a time-independent
Hamiltonian, and this provides a result with a real valued retarded response
function
i
χ cc+, a r, r ,τ = − Ψ0 0 c r, τ c r ,0 − c r ,0 c r,τ Ψa0 0
ℏ a
+
for τ > 0 and χ cc τ = 0 for τ < 0. The lower integral limit can be replaced with
−∞, and the integration variable can be changed to τ giving
∞
ca r,t = dτ d 3 r χ cc+, a r, r ,τ ϕ r ,t − τ
−∞
+
and the two sides can be Fourier transformed into c a ω = χ cc, a ω ϕ ω , which
are complex valued functions of frequency.
+
The response function χ cc, a t can be calculated in a variety of ways, using
expansions of operators in a known basis set, or generating numerically a solu-
tion over time. It can also be constructed semiempirically to incorporate known
features of the response. The response functions can be calculated introducing a
convenient basis set { μ } of states to expand operators and the reference unper-
turbed state Ψa0 , in which case they can be expressed in terms of the amplitudes
μ c r,t μ = cμμ r,t . The response can alternatively be obtained from states
52 2 Molecular Properties

Υa r, t = c r,t Ψa0 0 , which satisfy an equation of motion generated by

0
U t . This can be accomplished propagating the states numerically in time, or
obtaining them from a variational procedure [19].
+
The connection between the previous polarizabilities αξη, a and the
+
present susceptibilities χ cc, a can be obtained writing the polarization
components as Pξ, a t = d 3 r ξca r,t and the electric potential as
ϕ r , t − t = − η η η r ,t − t , and using the equation given above for
ca r,t to find that

+
αξη, a τ = d 3 r d 3 r ξ η χ cc+, a r,r ,τ

which are moment integrals of the susceptibility with respect to positions. Sim-
ilar expressions can be obtained for higher multipolar polarizabilities from
higher moment integrals. In an extended system, the integrals over space vari-
ables can be calculated with coarse-grained numerical procedures that can be
adapted to the physical distribution of electronic densities and localized polar-
izabilities in atomic or bond components.

2.5 Changes of Reference Frame

So far the reference frame has been chosen with its origin at the center of mass
of the molecule and with axes attached to the molecule. It is also convenient to
choose the origin of coordinates to be at the center of mass of the nuclei (CMN)
to simplify the treatment of electronic transitions for fixed nuclei. This is suf-
ficient for most applications, but for higher accuracy it is necessary to include
the masses of electrons with a so-called mass-polarization correction. The axes
attached to the molecule form a body-fixed (BF) set, and the molecule is
described in the CMN −BF frame. Interacting molecules are, however, rotated
and displaced in a common laboratory space, and we need to know what hap-
pens to multipoles of a molecule when its BF frame has been translated or
rotated with respect to a common frame L [5, 20]. An alternative description
of changes moves a molecule from one initial position to another, both
described in the same L frame, and gives different relations between multipoles
insofar as it is equivalent to keeping the molecule fixed and changing the refer-
ence frame with the reverse movements.
Let L = and CMN −BF = have the same origin and orientation of axes to
begin with, and consider first a translation of by a displacement vector L
 2.5 Changes of Reference Frame 53

while keeping the axes parallel, so that the origin position R CMN =RCMN + L, and
charge positions relate instead by r I =r I − L. This gives for the average dipole
component ξ in the BF frame,
Dξ = Dξ − CLξ
which shows it has changed due to the translation, if the total charge C = IcI is
different from zero. For the quadrupole Qξη = I CI 3ξI ηI − rI2 δξη one finds

Qξη = Qξη + C 3Lξ Lη −L2 δξη + 2 D L δξη

which changes with frame translation if either total charge or total dipole is not
zero. The conversions between frames when the molecule is moved, instead of
moving the frame, follow by changing L into − L in the above equations.
Changes of the n-th multipole involve the n-th power of the Lξ components,
and therefore its changes can be large, and they must be considered together
with all others.
A rotation of the BF while keeping the origins at the same location can be
done moving the frame by the set of Euler angles Ω = (α, β, γ) with respect
to , by angle α around the z-axis of (a right-handed) , followed by a rotation
–β around an intermediate y -axis, and after this a γ rotation around the final z -
axis [20]. For a column 3 × 1 matrix of elements (x, y, z) transformed into a new
column with (x , y , z ), the 3 × 3 transformation matrix A α,β,γ =
Rz γ Ry β Rz α , a product of three axial rotation matrices, with components
Aξη, can be used to transform dipole and higher multipole components and also
polarizations. From Dξ = η Aξ η α,β,γ Dη and ξ = η Aξ η α,β,γ η for the
electric field, using the inverse of the matrix A, one finds for the polarizability
tensor components in the rotated frame,
αξ η = ξ, η
Aξ η αηζ A −1 ζη

which is a function of the Euler angles. The permanent quadrupole has the same
angle dependence. If instead one rotates the molecule keeping the same refer-
ence frame, then the transformations of multipoles and polarizabilities are sim-
ilar but involve the inverse rotation and Euler angles. The usual notation in this
case is (α, β, γ) = (φ, ϑ, χ) for the orientation of the main axes of a rotated
rigid body.
The spherical components of multipoles are simply transformed under a
rotation operation R α,β,γ by means of the Wigner rotational matrices
l
Dm m α,β,γ , so that the transformed multipole is [4, 7]
l
R α,β,γ Qlm = m
Qlm Dm m α,β,γ
54 2 Molecular Properties

The effect on a spherical multipole Qlm of a translation by L is more

complicated but can also be given in a compact way as a sum of multipoles
of order 0 ≤ l ≤ l with coefficients obtained from solid spherical harmonic
functions of the translation components [4].

2.6 Multipole Integrals from Symmetry

Molecular symmetry can be used to identify which components of the average
multipole tensors and polarization tensors must vanish, or are interrelated. For
finite systems, this is done introducing point symmetry groups and symmetry
adapted wavefunctions, and the symmetry of multipole and polarization com-
ponents [5, 21]. An extension to include infinite systems with translational sym-
metry, such as crystalline solids, regular polymers, and surface lattices, is done
with space groups or crystallographic groups [22]. Symmetry arguments can
also be used to determine which transitions between molecular states will be
forbidden, leading to selection rules. Here, we only summarize the extensive
group theory for properties of molecules and solids.
Molecular symmetry is introduced identifying movements (or operations)
with respect to geometric elements that leave the nuclear framework with a
structure equivalent to the original one. The elements are the point of inversion
(named i), the axis of rotation by an angle 2π/n (or Cn), the planes of reflection
(σ), and the combination of a rotation axis and a perpendicular plane of reflec-
tion (improper axis Sn).
The set of symmetry movements G for a finite atomic system that leave
unchanged at least one point of the structure is collected in a point group
of transformations. This contains h symmetry movements (the group order).
A physically meaningful symmetry–conserving transformation moves not only
the nuclear structure but also the electronic charge distribution bound to it. In
the spirit of the Born–Oppenheimer treatment, it is convenient to focus on the
symmetry of electronic states Φ(X; Q) for fixed nuclear positions. When the
location of the nuclei uniquely determines the electronic state, as is usually
the case, it is sufficient to inspect only the atomic structure to identify symmetry
elements. States of molecules with open electronic shells or nonzero spin mul-
tiplicity may require additional symmetry considerations. The movement or
operation G is defined by changes of position variables of all electrons and
G
nuclei so that using Cartesian coordinates x,y,z x , y , z G.
These movements generate changes in the electronic state functions,
represented by symmetry operators OG for each movement given by
Φ GX; GQ = OG Φ X; Q , which defines the symmetry operators in the space
of states. These form a point group of operators isomorphic with the original
point group of movements in real space. The theory of point groups and space
 2.6 Multipole Integrals from Symmetry 55

groups and their matrix representations provide powerful and general methods
for analyzing multipole and polarization components.
For each element G in a group of symmetry movements, the equation for sta-
el
tionary states H Φ X; Q = E Q Φ X; Q is also satisfied by Φ(GX; GQ), which
el el
means that the electronic Hamiltonian satisfies OG H = H OG , and
OG Φ X; Q can be expressed as a combination of stationary eigenstates of
the Hamiltonian. Stationary states Φμ with the same eigenenergy Er(Q) of
degeneracy lr form a basis set for an irreducible representation (or irrep) Γ(r)
of the symmetry operator by means of matrices of the same order as the energy
degeneracy. A generated set of irreducible matrix representations Γμνr G can be
r
used to construct symmetry-adapted electronic states Φλ with projection
operators so they undergo prescribed changes for given movements [5]. The
projection operators can be simply constructed from the characters
χ r G = μ Γμμr G as
lr
G ∗ OG
r
P = χ r
h G

r r
and project the adapted state Φλ = P Φλ from a given state Φλ. Products of
r
irreps generated by basis set products Φλ Φμs can be decomposed into sums
p
shown schematically as Γ r ⨂ Γ s = p ars Γ p with coefficients derivable from
the related characters [5, 22]. This is done using Tables of characters and the
orthogonality properties of the characters, Gχ (r)(G)∗χ (s)(G) = hδrs.
Symmetry considerations allow determination of what permanent multipole
or polarization components must be null, and if a transition matrix elements
between states must vanish indicating a selection rule. The integrals involved
in quantum brackets are numbers I = dXdQ f(X, Q) clearly invariant under
symmetry operations. Therefore, f(GX, GQ) must give the same integral value
or this must be zero. Consequently, nonzero integrals must involve fully invar-
iant integrands, and these can be identified using symmetry. The identification
of null tensor components is frequently obvious from simple symmetry consid-
erations, especially when the electronic state is unchanged by symmetry
movements.
A homonuclear molecule such as H2 has a center of inversion and as a
results all three components of its average dipole in its ground electronic state,
which is nondegenerate and invariant, must be null because they are identical
to their opposite when the inversion movement is applied. The average dipole
in the H2O molecule must be located along the z-axis through O and the mid-
point between H’s because the perpendicular components are equal to their
opposite upon a 180 rotation around the z-axis, and so on, for other
molecules.
56 2 Molecular Properties

More complicated symmetries require the mathematical group treatment and

this relies on the basic result that Dirac brackets involving the product of two
electronic functions are null if they belong to different irreducible representa-
tions. In more detail, a theorem states that the direct product of two irreps con-
tains the totally symmetric representation only if they are identical, and then
only once. This can be applied as follows: using Tables of characters of point
groups and of direct product decompositions [5].
r n r
The permanent value Φλ αβ Φλ of a multipole in a state belonging to
the r representation is found to likely differ from zero only if the multipole com-
ponents change under symmetry operations so that the components span a rep-
resentation that contains the same irreducible representations found in
p
Γ r ⨂ Γ r = p arr Γ p . Otherwise, the integral in the bracket expression
does vanish. Similarly, the matrix elements of the transition multipole
r n
Φλ αβ Φμs vanish if the multipole transformation gives a representa-
tion which does not contain one of the irreducible representations in Γ(r) ⨂ Γ(s).
As an example, consider again the H2O molecule with symmetry movements
E (none), C2, σ(xz), and σ(yz) giving the point group symmetry 2v and the fully
symmetric irrep designated as A1. From the group characters in Table 2.1 for
2v , and knowing that the ground state function is of type A1 one concludes that
the z-component of the permanent dipole, which also belongs to A1, is different
from zero because the product of ground states belongs to Γ A1 ⨂ Γ A1 = Γ A1 ,
the same symmetry as the z-component. But the x- and y-components have
symmetry B1 and B2, which are absent, and vanishing expectation value inte-
grals. Similar arguments lead to the conclusion that the x2, y2, and z2 compo-
nents of the permanent quadrupole are not zero, but that the xy, yz, and xz
components must vanish. With regard to transition multipoles appearing in
the polarization, one finds that the x-, y-, and z-components of the dipole
r
may only induce excitations to states Φλ of the same symmetry B1, B2, and
A1, respectively, because these are the ones equal to the product of irreps
Γ A1 ⨂ Γ r = Γ r of the involved states. Similar considerations apply to other
molecules with different point group symmetries.

Table 2.1 Group characters for the point group symmetry 2v .

C2v or 2 mm E C2 σ(xz) σ(yz) Basis set

A1 1 1 1 1 z; x2, y2, z2
A2 1 1 –1 –1 Rz; xy
B1 1 –1 1 –1 x, Ry; xz
B2 1 –1 –1 1 y, Rx; yz
 2.7 Approximations and Bounds for Polarizabilities 57

The treatment can also be done introducing the spherical components of

multipoles. If the set {Qlm} with −l ≥ m ≥ l spans a representation Γ(l), this can
be decomposed using characters into irreps to consider the values of permanent
and induced components. For example in H2O, the irrep decomposition for the
dipole set {Q1m} is A1 + B1 + B2, and as before one finds that there is one perma-
nent dipole (from Q10) and that dipolar excitations from the ground state can
occur into states of symmetry B1 and B2. The decomposition shows the same
symmetries as the rotational basis functions in the Table 2.1.
Conditions under which transition matrix elements of multipole operators
are null constitute selection rules which exclude many transitions. In addition
to the ones relating to molecular symmetry, there are selection rules derived
from the rotational and spin dynamics of involved states. For fixed nuclear posi-
tions, if electronic initial and final transition states have angular momentum
quantum numbers (Li, Mi) and (Lf, Mf), respectively, then a spherical multipole
component with an angular distribution given by quantum numbers (l, m) will
only lead to allowed transitions insofar as |Li − l| ≤ Lf ≤ Li + l and Mf = Mi + m.
Also, the multipole operators do not depend on electronic spin variables and
cannot lead to transitions which change spin quantum numbers (S, MS), so that
we have also the selection rules Sf = Si and MSf = MSi [6].

2.7 Approximations and Bounds for Polarizabilities

2.7.1 Physical Models
Useful parametrized models of response functions, valid over a homogeneous
region of space, have been introduced in the literature, such as the Debye model
of dielectric relaxation with a decay time τD, where the dipolar time correlation
+
is parametrized as αzz, a t = D2z a exp − t τD , and one finds

1 − iωτD
α zz+, a ω = D2z a 1 + ω2 τD 2

Other physical choices for the dipole time correlation lead to dynamical sus-
ceptibilities with Lorentzian or Gaussian distributions of frequencies. For a dis-
tribution of charges undergoing an oscillation of frequency ωR, and a dipole
+
correlation relaxing over time τR, the choice αzz, a t = D2z a exp − t τR
cos ωR t leads to a Lorentzian susceptibility as above but with ω replaced by
ω − ωR. Another instance involves random changes over time in the charge dis-
tribution and is given by a Gaussian dipole correlation function like
+
αzz, a τ = D2z a exp −t 2 τG 2 from which a Gaussian distribution of frequen-
cies is found for the dynamical susceptibility [23].
58 2 Molecular Properties

2.7.2 Closure Approximation and Sum Rules

The summation over states in the polarizability extends over molecular states
λ = (I, α) composed of an electronic state α and rotational-vibrational states I for
each state α. Fixing the nuclear positions Q and concentrating on the electronic
polarizability, this can be expressed in terms of the electronic oscillator strength
for electrons with mass me and charge Ce, defined as the dimensionless value
2
el el
fξ, κλ Q = 2me e2 ℏ ωκα Q κ Dξ λ Q , involving elements of the elec-
tronic dipole between electronic states κ and α and excitation ener-
0
gies ℏωκλ Q = Eκ0 Q −Eλ Q = Δκλ .
The static polarizability for a field only along the z-axis is
2
0
Φλ Dz Φκ0
αzz, λ = 2 κ λ Δκλ

and for a general field the orientation average is αλ = 1 3 αxx, λ + αyy, λ + αzz, λ
given by
el
e2 ℏ2 fκλ
αλ = κ λΔ 2
me κλ

el 2
el
where the averaged oscillator strength fκλ = 2me 3e2 ℏ2 Δκλ κ D λ Q
contains the electronic dipole vector. A compact expression is obtained defining
a weighted excitation energy Δ by means of
2 2
el el
κ λ
κ D λ Δκλ = κ λ
κ D λ Δ
Q Q

where the numerator to the right, using the closure (or completeness) relation
κ λ κ κ = I − λ λ of the electronic basis set, is simply the dipole standard
2 2
el el
deviation value ΔD 2 = λ D λ − λ D λ . This gives
Q Q

αλ = 2 3 ΔD 2 Δ, which can be estimated taking (ΔD)2 to be of the order of

the square of electron charge times the average atomic displacement δ Q in
the electronic state λ, and taking Δ to be of the order of the vertical ionization
energy for fixed Q. Alternatively, one finds αλ = e2 ℏ2 me Δ − 2 κ λ fκλ and
el

from the sum rule for the oscillator strength [5] which gives el
κ α fκα = Nel∗ ,
 2.7 Approximations and Bounds for Polarizabilities 59

equal to the effective total number of independent electrons partaking in exci-

tation, one finds that αλ = e2 ℏ2 me Δ −2 Nel∗ , which provides another rough
estimate.
For the dynamical polarizability, the denominator is instead Δκλ2 = (ωkλ −
ωL)2 + γ kλ2/4 and provided the decay rate is negligible compared to excitation
frequencies, it agrees with the static value when ωL 0. In the limit for ωL ∞
ωL 2 = e2 ℏ2 me Nel∗ ωL 2 approaching zero.
el
one finds αλ ≈ e2 ℏ2 me κ λ fκλ
These two limits can be incorporated into a generic form for the dynamical
polarizability extended to complex values ω + iu of the frequency [24], which
appear, for example in the intermolecular dispersion energy [25, 26]. The inter-
polation form α(iu) = a/(b2 + u2) can be rewritten using the static polarizability
and the polarizability limit for ωL ∞ as α iu = α 0 1 + α 0 u2 Nel∗ .
More general sum rules can be applied to the calculation of dynamical polar-
izabilities of the form α iu = N 2 2
n = 1 an bn + u , for example starting from the
k
sum Gλ ω = κ λ Δκλ − ω k λ A κ Q κ B λ Q , which agrees with the
dipolar polarizability for k = –2 and when the operators are dipoles [25, 27].
These sums can be evaluated with convenient basis sets instead of a set contain-
ing usually unknown excited states of the unperturbed molecule.

2.7.3 Upper and Lower Bounds

Static polarizabilities can be obtained from time-independent, second-order
perturbation theory of the energy of a molecule in an external field. This can
be treated using the basis set completeness relation and projection operators,
to provide upper and lower bounds to the energy and from this to get bounds
to the polarizability [28, 29].
A similar treatment can be provided for upper and lower bounds to the
dynamical polarizability extended to imaginary values of the complex fre-
quency, β(u) = α(iu), as it appears in formulas for the intermolecular dispersion
forces [30]. This has also been done using Pade approximants and Gaussian grid
integration [31].
The charge C, dipole D, and mean polarizability α of Li+, Ne, H2, HF, CO2,
H2O and other selected species, are given in SI units in Table 2.2. Electric quad-
rupole moments are second-order tensors with components Qξη, and the main
quadrupole component Q = max(Qξξ) take values (in units of 10−40Cm2) of
0.237, 0.0, 2.12, 8.7, –14.3, and 9.087, respectively for the mentioned species.
Multipole and polarizability values, however, depend on the chosen origin of
coordinates for the charge and polarization distribution, which are assumed
here to refer to the molecular center of mass.
60 2 Molecular Properties

Table 2.2 Charge C, electric dipole D, and mean polarizability α of selected atoms and
molecules in SI units.

SI units C/(10−19 C) D/(10−30Cm) α 10 −40 J −2 C2 m2

Li+ 1.602 0.0 0.189

Na+ 1.602 0.0 0.994
Ne 0.0 0.0 0.437
H2 0.0 0.0 0.911
N2 0.0 0.0 1.97
HF 0.0 6.37 0.57
CO 0.0 0.390 2.20
CO2 0.0 0.0 2.93
H2O 0.0 6.17 1.65
CH4 0.0 0.0 2.89

References
1 Jackson, J.D. (1975). Classical Electrodynamics. New York: Wiley.
2 Landau, L.D. and Lifshitz, E. (1975). Classical Theory of Fields, 4e. Oxford,
England: Pergamon Press.
3 Hirschfelder, J.O., Curtis, C.F., and Bird, R.B. (1954). Molecular Theory of Gases
and Liquids. New York: Wiley.
4 Stone, A.J. (2013). The Theory of Intermolecular Forces, 2e. Oxford, England:
Oxford University Press.
5 Atkins, P.W. and Friedman, R.S. (1997). Molecular Quantum Mechanics.
Oxford, England: Oxford University Press.
6 Cohen-Tanoudji, C., Diu, B., and Laloe, F. (1977). Quantum Mechanics, vol. 2.
New York: Wiley-Interscience.
7 Zare, R.N. (1988). Angular Momentum. New York: Wiley.
8 McWeeny, R. (1989). Methods of Molecular Quantum Mechanics, 2e. San Diego,
CA: Academic Press.
9 Martin, R.M. (2004). Electronic Structure: Basic Theory and Practical Methods.
Cambridge, England: Cambridge University Press.
10 Kelly, K.L., Coronado, E., Zhao, L.-L., and Schatz, G.C. (2003). Optical properties
of metal nanoparticles: influence of size, shape, and dielectric environment.
J. Phys. Chem. B 107: 668–677.
11 Messiah, A. (1962). Quantum Mechanics, vol. 2. Amsterdam: North-Holland.
12 Lowdin, P.O. (1962). Studies in perturbation theory. IV. Projection operator
formalism. J. Math. Phys. 3: 969.
 References 61

13 Micha, D.A. (1974). Effective Hamiltonian Methods for Molecular Collisions

(Adv. Quantum Chem.), 231. New York: Academic Press.
14 Loudon, R. (1973). Quantum Theory of Light. Oxford, England: Oxford
University Press.
15 Mukamel, S. (1995). Principles of Nonlinear Optical Spectroscopy. Oxford,
England: Oxford University Press.
16 Davydov, D.S. (1965). Quantum Mechanics. Reading, MA: Pergamon Press.
17 Cohen-Tannoudji, C., Dupont-Roc, J., and Grynberg, G. (1992). Atom-Photon
Interactions. New York: Wiley.
18 May, V. and Kuhn, O. (2000). Charge and Energy Transfer Dynamics in
Molecular Systems. Berlin: Wiley-VCH.
19 Karplus, M. and Kolker, H.J. (1963). Variation-perturbation approach to the
interaction of radiation with atoms and molecules. J. Chem. Phys. 39: 1493.
20 Arfken, G. (1970). Mathematical Methods for Physicists, 2e. New York:
Academic Press.
21 Cotton, F.A. (1990). Chemical Applications of Group Theory, 3e. New
York: Wiley.
22 Tinkham, M. (1964). Group Theory and Quantum Mechanics. New York:
Mc-Graw-Hill.
23 McQuarrie, D.A. (1973). Statistical Mechanics. New York: Harper & Row.
24 Mavroyannis, C. and Stephen, M.J. (1962). Dispersion forces. Mol. Phys. 5: 629.
25 Dalgarno, A. and Davison, W.D. (1966). The calculation of Van der Waals
interactions. Adv. At. Mol. Phys. 2: 1.
26 Kramer, H.L. and Herschbach, D.R. (1970). Combination rules for van der Waals
force constants. J. Chem. Phys. 53: 2792.
27 Hirschfelder, J.O., Brown, W.B., and Epstein, S.T. (1964). Recent developments
in perturbation theory. Adv. Quantum Chem. 1: 255.
28 Lowdin, P.O. (1966). Calculation of upper and lower bounds of energy
eigenvalues in perturbation theory by means of partitioning techniques. In:
Perturbation Theory and its Applications in Quantum Mechanics (ed. C.H.
Wilcox), 255–294. New York: Wiley Publication.
29 Lindner, P. and Lowdin, P.O. (1968). Upper and lower bounds in second order
perturbation theory and the Unsold approximation. Int. J. Quantum Chem. 2
(S2): 161.
30 Goscinski, O. (1968). Upper and lower bounds to polarizabilities and Van de
Waals forces I. General theory. Int. J. Quantum Chem. 2: 761.
31 Langhoff, P., Gordon, R.G., and Karplus, M. (1971). Comparison of dispersion
force bounding methods with applications to anisotropic interactions. J. Chem.
Phys. 55: 2126.
 63

Quantitative Treatment of Intermolecular Forces

CONTENTS

3.1 Long Range Interaction Energies from Perturbation Theory, 64

3.1.1 Interactions in the Ground Electronic States, 64
3.1.2 Interactions in Excited Electronic States and in Resonance, 68
3.2 Long Range Interaction Energies from Permanent and Induced Multipoles, 68
3.2.1 Molecular Electrostatic Potentials, 68
3.2.2 The Interaction Potential Energy at Large Distances, 70
3.2.3 Electrostatic, Induction, and Dispersion Forces, 73
3.2.4 Interacting Atoms and Molecules from Spherical Components of
Multipoles, 75
3.2.5 Interactions from Charge Densities and their Fourier Components, 76
3.3 Atom–Atom, Atom–Molecule, and Molecule–Molecule Long-Range Interactions, 78
3.3.1 Example of Li++Ne, 78
3.3.2 Interaction of Oriented Molecular Multipoles, 79
3.3.3 Example of Li++HF, 80
3.4 Calculation of Dispersion Energies, 81
3.4.1 Dispersion Energies from Molecular Polarizabilities, 81
3.4.2 Combination Rules, 82
3.4.3 Upper and Lower Bounds, 83
3.4.4 Variational Calculation of Perturbation Terms, 86
3.5 Electron Exchange and Penetration Effects at Reduced Distances, 87
3.5.1 Quantitative Treatment with Electronic Density Functionals, 87
3.5.2 Electronic Rearrangement and Polarization, 93
3.5.3 Treatments of Electronic Exchange and Charge Transfer, 98
3.6 Spin-orbit Couplings and Retardation Effects, 102
3.7 Interactions in Three-Body and Many-Body Systems, 103
3.7.1 Three-Body Systems, 103
3.7.2 Many-Body Systems, 106
References, 107

Molecular Interactions: Concepts and Methods, First Edition. David A. Micha.

© 2020 John Wiley & Sons, Inc. Published 2020 by John Wiley & Sons, Inc.
64 3 Quantitative Treatment of Intermolecular Forces

3.1 Long Range Interaction Energies from

Perturbation Theory
3.1.1 Interactions in the Ground Electronic States
Basic concepts of intermolecular forces can be extracted from models of elec-
tronic structure chosen here to simplify the mathematical treatment [1, 2].
The main simplification derives from the Born–Oppenheimer separation of elec-
tronic and nuclear motions whereby we consider, as a first good approximation,
how electrons behave while keeping fixed the nuclear positions. This amounts to
treating nuclei as point charges while describing electrons by means of their elec-
tronic wavefunctions or in terms of their electronic charge densities. A reference
frame, called here the CMN-SF frame, is chosen for two interacting species A and
B, with its center at the position of the center-of-mass of the nuclei and with axes
orientations fixed in the space common to A and B. Interaction potential energies
between A and B become functions of the relative position vector R of the center
of mass of B relative to that of A, and of the set of internal nuclear position coor-
dinates Qint = (QA, QB) including internal position variables of A and B.
We consider here a system of two molecules A and B with fixed nuclei and
electronic particles labeled by I and J, respectively, and with charge distributions
as obtained from their electronic wavefunctions ΨκAB for a state κ. The
Hamiltonian operator H A of species A, written in the coordinate representation
in terms of the position and momentum operators r a and p a = −iℏ∇a for nuclei
A A
a = 1 to Nnu and operators r i and p i = − iℏ∇i for electrons i = 1 to Nel , is given
for fixed nuclei by the sum of the kinetic energy of electrons plus the Coulomb
potential energies of interaction between nuclei and electrons and among
electrons, as
e e, n e, e
HA = KA + V A +VA
e ℏ2 e, n 1 Ca Ce
KA = − ∇2i , V A = ,
2me i 4πε0 a, i
r a −r i
e, e 1 Ce2
VA =
4πε0 i<j
r j −r i

where Ca and Ce are the electric charges of nucleus a and the electron, in the SI units.
The electrostatic energy operator must be added to the Hamiltonian opera-
tors of A and B to describe their interaction, given in general for R > Rmin by
int
H AB = H A + H B + H AB R

int −1 CI CJ
H AB R = 4πε0 I J
R −r IA +r JB
 3.1 Long Range Interaction Energies from Perturbation Theory 65

Figure 3.1 Charges and center-of-

mass position vectors for species A
and B.
→ → → →
rA r s sB

A → B
R
→ →
RA RB

Charges I of A are located at r I =RA +r IA and charges J of B are at sJ =RB +sJB ,

withRA and RB the center-of-mass of nuclei in A and B. The definitions of charge
and CM position vectors are shown in Figure 3.1. Each of the Hamiltonians H S ,
S = A, B, contain the kinetic energy and Coulomb interactions of all particles in
species S. The interaction or coupling term between A and B is written assuming
that there are no interchanges of particles (electrons or nuclei) between them.
The form of the coupling can be written in different ways depending on
whether the two interacting systems contain only few atoms, or they are
extended many-atom systems. For many atoms, it is advantageous to introduce
the charge density operators per unit volume, containing particles I = a, i for
nuclei and electrons in species A and J = b, j for B. The charge density operators
are cA r = I CI δ r −r I in terms of the Dirac delta function, and
cB s = J CJ δ s−sJ , and the Coulomb interaction energy is given by

int −1 −1
H AB R = 4πε0 d 3 r d 3 s cA r r − s cB s

States and energies of the whole system satisfy the eigenvalue equation
H AB ΨκAB = EκAB ΨκAB , with energies and states dependent on the intermolec-
ular distance R. This can be solved with a perturbation expansion to a conven-
ient order, assuming that the interaction is small, introducing the internal
electronic states of A and B, and the zeroth order electronic states
Ψκ0 = jA kB = jA kB of the unperturbed (non-interacting) pair AB with
A B
zeroth-order energies Eκ0 = Ej + Ek . Here, it is assumed that insofar as A
and B are electronically bound states, their wavefunctions of electronic
66 3 Quantitative Treatment of Intermolecular Forces

coordinates fall exponentially with distance and their overlap is negligible at

large R so that electronic exchange and charge transfer between A and B can
be neglected. It is then sufficient to start with an interaction energy Hamiltonian
written in terms of the sets of unperturbed states of A and B assumed to form
complete systems, as
1 int
λH AB R = j, k jk
jA kB jA kB H AB jA k B jA k B
R

int
The matrix element jA kB H AB jA kB can be re-expressed as a sum of pro-
ducts of factors relating separately to A and B and has the meaning of an inter-
action where a transition in A from state jA to jA is coupled to a transition in B
from state kB to kB .
For two molecules A and B enclosed by spheres of radius aA and aB , the inter-
action potential energy is small at distances larger than a Rmin and a suitable
expansion parameter is λ = (aA + aB)/Rmin. Expanding the energy and wavefunc-
tion in powers of λ the Hamiltonian equation is
0 1
H + λH n≥0
λ n Ψκn = p≥0
λ p E κp q≥0
λ q Ψκq

0 1 1
where H = H A + H B and λH = H AB R . Equating on both sides the factors
multiplying λp, a recursion procedure follows as
0
H −Eκ0 Ψκ0 = 0
0 1
H −Eκ0 Ψκ1 + H Ψκ0 = Eκ1 Ψκ0
0 1 p− 1 q
H −Eκ0 Ψκp + H Ψκp −1 = Eκp Ψκ0 + E
q=1 κ
Ψκp− q

for p ≥ 2, and with the energy to p-th order given by the recursion relation
1 p −1 q
Eκp = Ψκ0 H Ψκp− 1 − E
q=1 κ
Ψκ0 Ψκp− q

where the normalization Ψκ0 Ψκ0 = 1 has been chosen. Higher-order wave-
function terms can alternatively be constructed so that Ψκ0 Ψκp −q = 0 for p
− q ≥ 1, which imposes the intermediate normalization Ψκ0 Ψκ = 1. Addi-
tional relations can be used to show that knowledge of Ψκp is sufficient to obtain
1
the Eκ2p + 1 energy [3]. In particular, Eκ1 = Ψκ0 H Ψκ0 ,
1 1
Eκ2 = Ψκ0 H Ψκ1 , and Eκ3 = Ψκ1 H −Eκ1 Ψκ1 .
 3.1 Long Range Interaction Energies from Perturbation Theory 67

We consider first perturbation energies when both A and B are separated by a

distance R and in their ground electronic state (0A, 0B). The first-order energy is

1 1
E0, 0 R = 0A ,0B H 0A , 0B

with A in its ground state interacting with B also in its ground state. This is the clas-
els 1 int
sical electrostatic interaction energy E0, 0 R = λE0, 0 R = 0A , 0B H AB 0A ,0B .
To second-order

1 2
0A ,0B H jA ,kB
2
E0, 0 R = j, k 0, 0 0 0
E0, 0 − Ej, k

and the double sum over j and k, which must not be both ground states, can be
separated into two expressions: one where only j or k is the ground state, called
an induction potential energy; and a second double sum where both j and k dif-
fer from the ground state, called a dispersion potential energy. The induction
energy is therefore

2
int
0A ,0B H AB jA ,0B
ind
E0, 0 R =− j 0 A A
Ej − E0
2
int
0A ,0B H AB 0A , kB
− k 0 B B
Ek − E0

and has the meaning of B in its ground state interacting with A in a transition
state for the first term, and the reverse in the second term. For the dispersion
energy one has a double sum where A in a state transition interacts with B also
in a state transition, in accordance with

2
int
0A ,0B H AB jA , k B
dsp
E0, 0 R =− j 0 k 0 A A B B
Ej − E0 + Ek − E0

This can be made more specific by introducing details of the AB pair inter-
action. Here, the denominators have been changed to show them as always pos-
itive, with the minus sign in front of expressions indicating that these are
negative quantities corresponding to attraction between A and B when these
are in their ground states. This is not necessarily the case for the perturbation
energies of their excited states.
68 3 Quantitative Treatment of Intermolecular Forces

3.1.2 Interactions in Excited Electronic States and in Resonance

The perturbation treatment requires reconsideration when the reference state,
ground, or excited, has the same, so-called degenerate, energy as other states. In
this case, a useful procedure is to lift the degeneracy by including the part of
int
H AB within the subspace of degenerate functions in the Hamiltonian operator
to zeroth order and solving for new eigenstates and eigenenergies within the
subspace of the previous degenerate states with the expectation that the new
energy values will not be degenerate. The perturbation treatment can then pro-
ceed as before with the new zeroth order states and energies [3, 4].
When the two species are the same and one of them is in an excited electronic
state A∗, the interactions in A + A∗ lead to quantum coherence of its states with
those of A∗ + A and consequent resonance energy transfer. The coherent states
replacing jA, kB are now j,k AA∗ = 2 − 1 2 jA ,kA∗ ± kA ,jA∗ with zeroth order
±

A A∗
energy Eκ0 = Ej + Ek and the first-order resonance interaction energy is for
R > Rmin,

EAA∗ = 2 −1
± int int
jA , kA∗ H AB jA ,kA∗ + kA ,jA∗ H AB kA , jA∗

int
± 2 jA , kA∗ H AB kA ,jA∗

so that the first two terms are again electrostatic interaction energies from the
charge distributions of A and A∗ in states j or k, but the last term is an addition
involving transitions between states j and k for each species. This involves an
energy of interaction between two electronic transition densities. The last term
contains state transfer integrals which obey different selection rules and lead to
longer range interactions, absent when there is no state coherence.
This subject is closely related to the theory of molecular excitons [5] and
intermolecular electronic energy transfer [6].

3.2 Long Range Interaction Energies from Permanent

and Induced Multipoles
3.2.1 Molecular Electrostatic Potentials
We model the interaction of two molecules A and B with their center of mass (or
CM) located at positions RA and RB and with relative position R =RB −RA . We
need the Coulomb, or electrostatic potential, ϕels s created by the charges in
A at a field position s near B where its charges may be found.
The interaction energy of charges in A with those in B must be separately trea-
ted depending on whether the distance between A and B is large compared to
 3.2 Long Range Interaction Energies from Permanent and Induced Multipoles 69

their size, or not. In the first case, we deal with long-range (LR) interactions, and
we can express the energy as an expansion in inverse powers of R and in terms of
the quantal electronic states of the isolated compounds. The short-range inter-
actions must, however, be treated differently starting from the electronic struc-
ture of the whole pair and its quantal states, allowing for electronic
rearrangement. Indicating the electrostatic, or Coulomb, interaction of the
int
two sets of charges as HAB R , its long-range form can be written as
LR int
HAB R = fd R HAB R , with fd(R) a function changing from zero at short
distances into 1.0 at long distances with a transition around a minimum dis-
tance Rmin and a transition region of width aA + aB, the sum of radii of spheres
enclosing the charge distributions of A and B. The total Hamiltonian is then
int LR
H AB = H A + H B + HAB R 1 −fd R + HAB R , and one can concentrate on
the LR part to begin with.
We consider here situations where the two charge distributions do not over-
lap, with R > Rmin, a minimum distance, large, and with rI R and sJ R. We
work in the CMN-SF frame, take the origin of coordinates at the CM of A and
−1 −1
expand the inverse distance s− r = R +sB −r A between generic field and

charge locations in a power series for small rA/R , where R = R +sB , with com-
ponents Rξ = X ,Y ,Z . This is done using that ∂ R −r A ∂ξA = Rξ R
ξA = 0

= − ∂ R −r A ∂Rξ which leads to

1 ∂ 1 ∂2 1
= 1− ξ
ξ A
+ ξ η
ξ, η A A
+…
R −r A ∂Rξ 2 ∂Rξ ∂Rη R

−1
−1
and an expression for ϕels R = 4πε0 I CI R −r IA like

1 A ∂ 1 A ∂2 1
ϕels R = C A
− D
ξ ξ
+ Q
ξ, η 6 ξη ∂R ∂R
+…
4πε0 ∂Rξ ξ η R

written in terms of total charge, dipole, and quadrupole components

C= I
CI , Dξ = I
C I ξI
Qξη = I
CI 3 ξI ηI − δξη rI2

for compound A. These can also be given in terms of spherical multipole func-
tions of spherical coordinates, which are useful in the description of two inter-
acting few-atom systems.
70 3 Quantitative Treatment of Intermolecular Forces

This is next illustrated in a simple case where we describe the interaction of

compound A with charges 1 and 2, interacting with compound B containing
charges 3 and 4. For two charges C1 and C2 at locations z1 and z2 on a z-axis
along direction nz , the Coulomb electrostatic potential at field location s = R,
the CM of B, with components (X, Y, Z) is

ϕels R = C1 4πε0 R1 + C2 4πε0 R2

where Rj = R −r j | and r j = zjnz , j = 1, 2.

Expanding Rj−1 in a power series for small rj/R, and using that
∂R1 ∂z1 z1 = 0 = Z R = −∂R ∂Z , we have

1 ∂ 1 ∂2 1 ∂ 1 2 ∂2 1
= 1 + z1 + z12 2 + … = 1 − z1 + z1 2 + …
R1 ∂z1 2 ∂ z1 R1 ∂Z 2 ∂Z R
and similarly for charge 2; further using that Z/R = cos(Θ) we find
0 1 2
ϕels R = ϕels R + ϕels R + ϕels R

0 C
ϕels R =
4πε0 R
1 Dcos Θ
ϕels R =−
4πε0 R2
2 Q 2 3cos2 Θ −1 2
ϕels R =
4πε0 R3
where we have introduced the total charge C = C1 + C2 (a 20-pole), dipole D =
C1z1 + C2z2 (a 21-pole), and quadrupole Q = 2 C1 z12 + C2 z22 (a 22-pole) of the
system. This shows the electrostatic potentials created by charge, dipole, and
quadrupole distributions. Higher-order terms include octupoles, and higher
2n-poles. The electric potential of a 2n-pole is found to vary with large R as
R−(n + 1).
The electrostatic field vector = −∂ϕels ∂ R has spherical components (for
the above charges on the z-axis) given by

∂ϕels 1 ∂ϕels
ℇR = − , ℇΘ = − , ℇΦ = 0
∂R R ∂Θ

3.2.2 The Interaction Potential Energy at Large Distances

Continuing with the simple model, we apply the previous expansion to the elec-
trostatic potential due to the A charges, and find the interaction Coulomb
 3.2 Long Range Interaction Energies from Permanent and Induced Multipoles 71

energy of a system B of charges in that potential [7, 8]. We consider the system B
with two charges 3 and 4 located at positions R3 = R +r 3 and R4 = R +r 4 in the
potential of charges 1 and 2 (system A). The interaction energy function is now
int A A
HAB = C3 ϕels R3 + C4 ϕels R4
This can be further expanded now for small rj/R, j = 3, 4, which brings in the
multipoles of B, and we introduce CB = C3 + C4 and DB = C3r 3 + C4r 4 , and a sim-
ilar notation for the A multipoles. Using a vector notation and noticing that the
interaction of a 2m-pole of A with a 2n-pole of B gives a term with an R−(m + n + 1)
dependence, we find for R > Rmin
−1
R−
int m+n+1
HAB R = 4πε0 F AB
m, n m, n
AB
F0, 0 = CA CB
AB AB
F0, 1 = − CADB n , F1, 0 = −CBDA n
AB
F1, 1 =DA DB − 3 DA n DB n

where n = R R is a unit vector pointing from A to B. Additional terms contain

higher-order multipoles and can be derived by inspection. These expressions
have the same form in a quantal treatment where the electrostatic energy
and dipoles are operators. The electrostatic potential energy of a quantal state
of the A–B pair of compounds is obtained from the quantal expectation value of
this potential energy operator for the given state.
A general treatment requires an expansion of the inverse of R = R +sB for
small sB/R, which brings in all the multipoles of B, multiplying those of A.
The electrostatic energy operator
int B A
HAB = J B
CJ ϕels R +sJB
can be expanded now for small sJB/R as
LR B ∂ 1 B ∂2 A
HAB = C B
+ D
ξ ξ
+ Q
ξ, η ξη
+ … ϕels R fd R
∂Rξ 6 ∂Rξ ∂Rη
and written for large distances in a very general form introducing Cartesian ten-
sors t constructed from unit vectors nξ = Rξ/R from
∂ 1 Rξ nξ tξ
=− 3=− 2=− 2
∂Rξ R R R R
∂ 2 1 3nξ nη −δξη tξη
= = 3
∂Rξ ∂Rη R R3 R
72 3 Quantitative Treatment of Intermolecular Forces

and tξηζ = − 15nξnηnζ + 3(nξδηζ + nηδξζ + nζδξη). In general, the tensor with n
indices is

∂χ R − 1
n
tξη…χ = R n + 1 ∂ ξ ∂ η
for higher derivatives, with the notation ∂ ξ = ∂/∂Rξ [9]. These tensors depend on
the orientation angles of the relative position vector R =RB −RA and their
sign changes by (−1)n if this direction is reversed. The electrostatic potential
is then

A 1 CA A tξ 1 A tξη
ϕels R = + ξ
Dξ + ξ, η 6
Qξη +…
4πε0 R R2 R3
which is in turn differentiated n times and multiplied times the 2n-th multipole
of B to obtain the interaction energy operator as [9, 10]
A B
−1 C C tξ B A tξ
H AB LR
= 4πε0 + C A
D − Dξ 2 C B
R ξ R2 ξ R
1 A tξη B A tξη B
+ C Q + Qξη 3 C
6 ξ, η R3 ξη R
A tξη B
− Dξ D
ξ, η R3 η
1 A tξηζ B A tξηζ B
− Dξ Q + Qξη 4 Dζ + fd R
6 ξ, η, ζ R4 ηζ R

Alternatively, introducing multipolar tensor factors Tμ,mν, n R = fd R tμm

,ν
,n
dependent on orientation angles for each pair of interacting multipoles, the
interaction Hamiltonian is compactly given by

LR −1 Tμ,mν, n R
H AB = 4πε0 Mm A
nν μ
Mn B
mμ Rm + n + 1 ν
where μ designates the components of the 2m multipole, with three values for
the dipole, six for the quadrupole, and so on.
In a reference frame with the Z-axis along the intermolecular relative position,
the perturbation energies depend on R and are given as before in terms of the
1 1
contains a sum of R−(m + n + 1) terms, and
0
integrals Ψκ0 H Ψλ . Here, H
the energy term Eκp R is itself an expansion in powers of R−1, of the form

Eκp R = n≥n p
Cnp R − n

where n(p) is the lowest power of R−1 which appears to order p in the energy
expansion, and increases as p increases.
 3.2 Long Range Interaction Energies from Permanent and Induced Multipoles 73

For the ground state of the pair, the energy varies as the relative distance
decreases from infinity, and the intermolecular potential energy evolving from
the ground state g = (0A, 0B) of the pair is

Vg AB R = EgAB R −EgAB ∞
AB A B
where the asymptotic term is simply Eg ∞ = E0 + E0 , the sum of energies
of the isolated molecules in their ground state. This intermolecular potential
energy is therefore a double perturbation expansion of form

Vg AB R = p≥1
λp n≥n p
Cnp R −n

and the lower limits n(p) to the summations must be found by inspection of the
asymptotic expansion series in powers of R−1. The coefficient of R−n must be
obtained from all orders in λ relevant to a desired accuracy.
An extension of the perturbation treatment is needed when the states of A or
B are degenerate [3, 4, 11], as is frequently the case for excited electronic states.
The expansions in inverse powers of R can be expected to converge only
asymptotically, particularly when applied to atoms or compounds with open
electronic shells, as has been noticed in a survey of several examples [12].
AB
The terms An/Rn making the series expansion of Vg R are frequently such
that the convergence criterion based on the quotient of two adjacent terms gives
An + 1/(AnR) ∞ for n ∞ indicating series divergence. However, the sum
of the first N terms in the expansion usually gives a good approximation to the
AB
exact potential energy so that Vg R − N n = 0 An R
n
0 as N ∞ for R lar-
ger than a critical value dependent on N, indicating an asymptotic or semicon-
vergent series. This justifies using a sum of inverse power terms to calculate the
potential energy for large distances.

3.2.3 Electrostatic, Induction, and Dispersion Forces

We consider here the ground states of the pair. To first order, the total energy is
int
in general the expectation value λE κ1 = jA kB H AB jA kB , a sum where each
term contains the expectation values of the (2 )A and (2n)B multipoles of A and
m

B. In the ground state (jA, kB) = (0A, 0B). Therefore, each term is of the form
−1
Cm + n + 1 R −
els m+n+1
Vmels
, n R = 4πε0
els
where Cm + n + 1 is a constant obtained from the permanent multipoles in the
ground states.
To second order, the potential energy operator appears in two factors,
between the ground and excited states of the (A, B) pair, and again back from
excited states to the ground state. Depending on the nature of the excitation, we
74 3 Quantitative Treatment of Intermolecular Forces

can distinguish induction and dispersion contributions in the second-order

energies.
In the induction energy terms, the first factor involves the transition from (A,
B) to (A, B∗) in which B∗ is excited and acquires distortion multipoles. This fac-
tor is an interaction between the (2m)A permanent multipole of A and the 2n B∗
induced multipole of B∗. The second factor in the induction energy involves an
induced multipole 2n B∗ and a permanent (2p)A and represents the interaction
leading back to the ground state. Therefore, the interaction is of the form
Cm + p + n + n + 2 R − m + p + n + n + 2 . However, selection rules for allowed transi-
ind

tions into and out of states of B∗ usually requires that n = n , so that the induc-
tion potential energy is usually
ind −2 ind − m + p + 2n + 2
Vm, p, n R = − 4πε0 Cm + p + 2n +2 R

A similar term arises from B A∗ B, with the induced multipole 2m A∗

interacting with (2n)B up and (2q)B down in energy. The coefficient here is pos-
itive as can be seen from the second-order correction to the ground state energy.
The leading induction term is typically one with m = p and n = q. For ion-neutral
interactions, it usually is a (permanent monopole)–(induced dipole) interaction
going as R−4.
In the dispersion interaction terms, we have temporary transitions from (A, B)
to (A∗, B∗) and back to (A, B), in nonenergy conserving transient events allowed
by quantum mechanics. This can be interpreted as a fluctuation of charges in A
which creates a distortion multipole 2m A∗ ; this induces fluctuations in B,
which changes into B∗ with multipole 2n B∗ as both compounds are excited.
The same interacting multipoles bring the system back to (A, B). The interaction
potential is therefore of the form
−2
R −2 m + n
dsp dsp
Vm , n R = − 4πε0 C2 m +n +1
+1

with the leading term coming from (induced dipole)–(induced dipole) interac-
tions going as R−6. The three types of interaction processes, electrostatic, induc-
tion, and dispersion, are diagrammed in Figure 3.2.
Going to higher powers in the perturbation expansion in λ of the intermolec-
ular potential energy gives additional terms combining permanent and induced
multipole interactions. The total potential energy can be written as V(R) =
V(els)(R) + V(ind)(R) + V(dsp)(R) + V(pol)(R), an asymptotic expansion in powers
of R−1 with a varying radius of convergence, where a polarization term from
higher-order perturbations in λ combines induction and dispersion
interactions.
When the two species are the same, and one is electronically excited, there is
in addition a resonance potential energy V(res)(R) resulting from the quantal
 3.2 Long Range Interaction Energies from Permanent and Induced Multipoles 75

Figure 3.2 Interactions (dashed Electrostatic

lines) of electronic transitions
(arrows) in electrostatic, induction, 2m 2n
and dispersion phenomena
A 1 B
between A and B.
R m+n+1

Induction
2m
2n Excitation
A 1
R m+n+1 B

2mʹ
2nʹ De-excitation
A 1
R mʹ+nʹ+1 B

Dispersion

2m 2n Excitation
1
A R m+n+1 B

2mʹ 2nʹ De-excitation

1
R mʹ+nʹ+1

coherence of states in A + A∗ and A∗ + A. This leads to additional interactions

between transition multipoles with the same behavior as the electrostatic term,
and in the case of the (transition dipole)–(transition dipole) interaction, it gives
to first-order perturbation V (res)(R) = (4πε0)−1C(res) R−3, of longer range than
induction or dispersion terms.

3.2.4 Interacting Atoms and Molecules from Spherical Components

of Multipoles
The interaction energies involving atoms and small molecules can be conven-
iently treated using a few spherical multipoles per compound, with each com-
ponent containing charges I given by
1 2
4π
Qlm = CI rIl Ylm ϑI ,φI
2l + 1 I

for l ≥ 0 and −l ≥ m ≥ l, written in terms of the spherical harmonic function Ylm

of position angles, with well-known transformation properties under rotation.
76 3 Quantitative Treatment of Intermolecular Forces

−1
A −1
The electrostatic potential ϕels R = 4πε0 I CI R−r IA can be expanded
in the spherical harmonics for R and r IA as

A −1 1 rIA l 4π ∗
ϕels R = 4πε0 C
I I
Y
m lm
ϑAI , φAI Ylm ϑR ,φR
lR R 2l + 1
which gives in terms of multipoles
1
A −1 4π 2 1 ∗
ϕels R = 4πε0 Qlm r IA Ylm ϑR ,φR
l, m 2l + 1 Rl

This potential acts on the charge multipoles of B to give a pair interaction

energy with a relatively simple form provided the local reference frames A
and B are parallel and have their z-axes parallel to the intermolecular position
vector R [13], as in
−1 l ,l
R−
LR la + lb + 1
H AB = 4πε0 la ≥ 0
F a b
lb ≥ 0 AB
fd R
λ
l ,l μ=λ −1 la + lb
F ABa b = μ = −λ 1 2
Qla , μ r IA Qlb , − μ r JB
la − μ la + μ lb − μ lb + μ
where λ is the smallest between la and lb. The multipoles here can be readily
l
rotated in each frame using the Wigner rotational matrices Dm m α, β, γ , func-
tions of Euler angles for each rotation [14, 15].
Electrostatic, induction, and dispersion energies follow from perturbation
expansions containing integrals of multipoles over electronic states of A and
B. For atoms, integrals of the multipoles of A between electronic states with
given angular momentum quantum numbers, νALAMA and νA LA MA , are
zero except if the selection rules MA = MA + μ and LA −LA ≤ la ≤ LA + LA apply,
and similarly for B. These rules substantially decrease the number of excited
states needed in calculations of the interaction energies.

3.2.5 Interactions from Charge Densities and their Fourier

Components
It is possible to avoid expansions of interaction energies in terms of multipoles,
by instead using charge densities and their Fourier transforms [16, 17]. The
charge density operator of all the charges I in species A distributed over space
can be re-expressed in terms of its Fourier components QA k as in

d3 k
cA r = C δ r −r IA =
I I
exp − ik r QA k
2π 3
 3.2 Long Range Interaction Energies from Permanent and Induced Multipoles 77

with the inverse QA k = d 3 r exp ik r cA r = I CI exp i k r IA and simi-

larly for species B. The interaction energy operator
int −1 −1
H AB R = 4πε0 d 3 r d 3 s cA r r − s cB s

can be also re-expressed in terms of Fourier components using that

−1
r = d 3 k 2π 3
exp − ik r k −2 , to obtain the result

int −1 d3 k
H AB R = 4πε0 exp ik R QA − k QB k
2π 2 k 2
to be incorporated into the first- and second-order perturbation expansions
given above for the interaction energies. Here, we concentrate on the dispersion
component to second order, which contains denominators like
A A B B
Ej −E0 + Ek − E0 = a + b, and use the Casimir–Polder integral (for positive
a, b)
∞
2 du a b 1
=
π 0 a2 + u2 b2 + u2 a+b
to write for the dispersion energy in Section 3.1.1, with ω = iυ

dsp −2 d3 k d3 k
E0, 0 R = − 4πε0 exp i k − k R
2π 2 k 2 2π 2 k 2
∞
ℏ
dυ χ A k,k ; iυ χ B − k, −k ; iυ
2π 0
2 ωj
χ A k, k ; iυ = 0 QA − k j j QA k 0
ℏ jω 2
j + υ2
A A
with ℏωj = Ej − E0 , and similarly for B. This gives the dispersion energy as a
sum over grid points in k and k spaces and avoids an expansion in inverse
powers of R. The factors χ S are dynamical susceptibilities evaluated at imagi-
nary-valued frequencies for each species S.
Further, using the plane wave decomposition into spherical waves,
∞ l
exp i K R = l=0 m = −l
i l 4π Ylm ϑK ,φK ∗ Ylm ϑR , φR jl KR

where the spherical Bessel function jl(x) satisfies asymptotic conditions jl(x) ≈
Alx−l − 1 sin(x) for x ∞ and jl(x) ≈ Blxl for x 0, it is possible to provide a
detailed analysis of the two limits for large and small R [17]. It is found that
for large R one recovers the inverse R multipole expansion previously generated,
and that now the present treatment gives finite values also for small R, so that
the dispersion energy for all R can be written as
78 3 Quantitative Treatment of Intermolecular Forces

−2
R −2 m
dsp dsp dsp
Vm , n R = − 4πε0 f2 m +n +1 R C2 m + n +1
+ n +1

with f(R) ≈ 1 for R ∞ and f(R) ≈ 0 for R 0 [17]. A detailed analysis here
shows that for R 0 one finds Vm, n R ≈ −AR2 m −n + BR2 m −n + 2 . Therefore,
dsp

using the Fourier transform of charge distributions provides a rigorous proce-

dure for avoiding the divergences of the inverse R expansions. For the (induced
dipole)–(induced dipole), interaction, with m = 1 and n = 1, it gives
dsp
V1, 1 R ≈ − A + BR2 when R 0.

3.3 Atom–Atom, Atom–Molecule, and Molecule–

Molecule Long-Range Interactions
3.3.1 Example of Li++Ne
The ground electronic state of Li+ in the atomic LS coupling scheme, which
neglects spin-orbit coupling, is 1s2 1S, and that of Ne is 1s22s22p6 1S. Their states
are labeled as 2S+1L(ML, MS). For a charged Li+ and neutral Ne, and since their
ground state parity (−1)L = 1 is even, their permanent dipole is zero by the parity
selection rule, and they have quadrupoles and higher multipoles. Locating the
atoms along the z-axis at a distance R, with Li+ as A and Ne as B, the interaction
−1
F 01 R −2 + F 11 R − 3 , plus higher-order
AB AB AB
Hamiltonian is H LR = 4πε0
terms in the inverse distance variable. In their ground state (0A, 0B), the first-
order energy perturbation expression gives V(els)(R) = 0.
Second-order perturbation terms involve excited atomic states. For Li+ they
are 1sns 1S, 1snp 1P, 1snd 1D, …, with n ≥ 2; for Ne, and ignoring core electrons,
they are 2p53p 1S, …, 2p53s 1P, 2p53p 1P, 2p54s 1P, 2p53d 1P, …, 2p53p 1D, 2p53d
1
D, …, all of which are coupled to the ground states by multipole transitions
allowed by selection rules. All the excitations are to total spin singlet states, with
S = 0. Considering that a 2m multipole has parity (−1)m and angular momentum
quantum numbers (L, M) = (m, μ), it follows that transitions from initial to final
states must satisfy the parity rule −1 Li = − 1 m + Lf and the angular momen-
tum rules |Li − m| ≤ Lf ≤ Li + m and MLf = MLi + μ. The rules can be imposed
going from Cartesian components ξ = x, y, z to spherical ones = 0, ± 1, and using
the dipole spherical components D0 = Dz and D±1 = (Dx ± Dy)/ √ 2.
AB B B
A A
For the induction potential energy we find, with F 01 = C nz D =C D0 ,
2
−2
V ind
= − 4πε0 C A
αB 0 2R4
2
B B B
α B 0 =2 k 0
0 B D0 kB Ek − E0
 3.3 Atom–Atom, Atom–Molecule, and Molecule–Molecule Long-Range Interactions 79

where the second line gives the static dipolar polarizability of B, and the excited
states are kB = nB1P(ML = 0). The dispersion potential energy follows from
AB A B A B
F 11 = D + D −B + D −A D + −2 D0 D0 , and its matrix elements between
0A0B and jAkB . Noticing that selection rules restrict the three terms to be
nonzero only one at a time, the dispersion energy is given by three σ = ± , 0 com-
ponents like
2
1 DσA 0j
B
D −σ 0k C6
dsp
Vσ dsp = − =−
4πε0 2 R6 j 0 k 0 A
ϵj0 + ϵk0
B 4πε0 2 R6

A A A
where DσA 0j
= 0A DσA jA and the ϵj0 = Ej − E0 are excitation energies,
dsp dsp
as V dsp = V + + V −dsp + 4 V0 . Here, the excited states for A mediated by
A A
D0 and D ± are instead jA = nA 1P(ML = 0, ± 1) and similarly for B.
For a pair of neutral atoms such as He+Ne, the induction energy disappears
and only the dispersion terms remain. The (induced dipole)–(induced quadru-
pole) dispersion energy goes as R−8.

3.3.2 Interaction of Oriented Molecular Multipoles

Interaction energies depend on the Euler angles ΓA = (α, β, γ)A giving the orien-
tation of multipoles of A, and similarly for B, situated in a common space fixed
reference frame for the A–B pair. Their transformation from the local body-
fixed (BF) frames of A and B to their common space-fixed (SF) frame can be
generally done in terms of matrix representations of the rotations operators
for each multipole.
A rotation operator R ΓA applied to the Cartesian coordinates ξ, η in the SF
frame generates a 3 × 3 matrix A(ΓA) giving the new components
ξ = η Aξ η ΓA η in the BF frame of A, with a similar expression for the Carte-
sian coordinates of species B. The dipole components are given by
A B
Dξ = η Aξ η ΓA DηA with a similar expression for Dξ in terms of its Euler
angles ΓB. This is generalized to other multipoles, which contain two or more
matrix factors A(ΓA) for quadrupoles or higher multipoles. An alternative com-
A
pact derivation can be based on the spherical components Qlm of species A,
which is transformed from the BF to the SF frame by means of the Wigner
l
rotational matrices Dm m ΓA as shown in Section 2.5.
Induction potentials V(ind) depend in these cases on the multipole orientation
angles through the static polarizability α(B)(0; ΓB ), and dispersion potentials
dsp
Vσ dsp depend on the orientations of the two species, through C6 ΓA ,ΓB .
80 3 Quantitative Treatment of Intermolecular Forces

3.3.3 Example of Li++HF

The treatment for Li++HF goes along similar lines, with the same electronic
states for Li+. The main differences are that now HF is a heteropolar molecule
with a permanent electric dipole, and with electronically excited states which
have only a good angular momentum quantum number Λ along the diatomic
axis so that applicable selection rules for HF are different. The ground state of
HF is a 1Σ, for Λ = 0, and dipolar excitations can occur to Λ = 0, ± 1. Only spin
singlet states are allowed. The interaction must be treated in a space fixed (SF)
XYZ reference frame, while the selection rules for the diatomic are applicable in
a body fixed (BF) X Y Z frame attached to the diatomic.
The interaction potentials depend on the relative orientation of the diatomic.
Letting the diatomic be the compound B with its center of mass (CM) at the
origin of coordinates in the SF frame, the BF frame with the same origin is
located by the Euler angles (α, β, γ) of rotation from SF to BF. Choosing the three
atoms to be on the SF XZ-plane, the relative position vector from A to the CM of
B along the SF Z-axis, and the diatomic nuclei on the BF Z’-axis, the Euler angles
are (0, β, 0). The transformation of multipole components is done with the 3 × 3
rotation matrix A(α, β, γ) and in this case, this simply gives for the dipole in the
B B B
SF frame Dξ = η Dη Aη ξ 0,β,0 = η Dη RY β η ξ , a rotation around the
Y = Y axis, in terms of dipole components in the BF frame. These give the oper-
B B B B B B
ator relations DX = cos β DX + sin β DZ , DZ = − sin β DX + cos β DZ ,
B B
and DY = DY . Similar transformations apply for the quadrupoles and involve
two rotational matrix factors.
To obtain the potential energies, we need the matrix elements of permanent
and transition dipole operators. In the BF frame, the diatomic electronic states
have axial symmetry and are labeled by the axial angular momentum quantum
numbers. Matrix elements satisfy selection rules so that for the permanent com-
B B B B
ponents we have DZ = 1
Σ DZ 1
Σ = D B , DX = DY = 0. Transition dipole
B 1
matrix elements of the form 1
Σ DZ Λ = 0 except for Λ = 0, ± 1, with
parallel (for Λ = 0) and perpendicular components, respectively. This leads to
parallel and perpendicular polarizabilities appearing in the induction and dis-
persion energies. Taking the angle between the diatomic dipole vector and
the position vector from B to A to be θ = π − β, the atom–diatom (A–D) poten-
tials are, with P2(u) = (3u2 − 1)/2 [18]
−1
D B R −2 cos θ
els A
VA − D = 4πε0 C
2 2
ind C A
αB 2 α − α⊥ B αA DB
VA − D =− 1+ P2 cos θ − P2 cosθ
4πε0 2 2R4 3 α 4πε0 2 R6
 3.4 Calculation of Dispersion Energies 81

B
with α B = α + 2α⊥ 3, an average static polarizability for the HF diatomic,
while the dispersion potential becomes
dsp dsp
dsp
C6 + 2 C6⊥ 1 α − α⊥ B
VA −D = − 1+ P2 cosθ
4πε0 2 6
R 3 α
in terms of parallel and perpendicular dispersion coefficients involving corre-
sponding transition dipoles of HF. Additional terms must appear if interactions
involving permanent and transient quadrupoles were included, containing
higher powers of R−1 and of cos(θ). The interactions of a neutral atom with a
heteropolar diatomic are obtained letting C(A) = 0 above, while the interaction
of a charged atom with a homopolar molecule follows from D(B) = 0.
The charge C, dipole D, and mean polarizability α of Li+, Ne, H2, HF, H2O, and
other selected species have been given in SI units in Table 2.2 of Chapter 2, and
can be used to calculate electrostatic, induction, and dispersion energies, how-
ever keeping in mind that multipole and polarizability values depend on the
chosen origin of coordinates for the charge and polarization distribution, which
refer here to the molecular center of mass.

3.4 Calculation of Dispersion Energies

3.4.1 Dispersion Energies from Molecular Polarizabilities
A dispersion potential energy for A and B in their ground states, of form
2
1 DσA 0j
B
D −σ 0k
Vσ dsp =−
4πε0 2 R6 j 0 k 0 A
ϵj0 + ϵk0
B

can be conveniently rewritten as an integral with factors relating separately to

species A and B, which allows calculations of the pair A–B dispersion potential
from the polarizabilities of the components. This is done following [19] using
again the Casimir–Polder integral
∞
2 du a b 1
=
π 0 a2 2 2
+u b +u 2 a+b
valid for a > 0, b > 0. The right-hand side denominator is like the denominator in
A B
the dispersion potential, and choosing a = ϵj0 , b = ϵk0 , the integral form leads to

∞
1 1 A iu B iu
Vσ dsp = − du ασ, 0 α
4πε0 2 R6 2π 0 ℏ −σ, 0 ℏ
82 3 Quantitative Treatment of Intermolecular Forces

where
2
A
iu ϵj0 DσA 0j
A
ασ , 0 =2
ℏ j 0 A
2
ϵj0 + u2
is a positive valued function decreasing and vanishing as u goes from zero to
u ∞. It follows that the dispersion potential energy can be constructed from
the pair polarizabilities evaluated at imaginary-value frequencies ω = iu/ℏ.
The dispersion energy written for Cartesian or spherical dipoles can also be
re-expressed as shown, and the procedure can also be applied to dispersion
interactions involving quadrupoles or higher multipoles of A or B, introducing
the corresponding multipolar polarizabilities.
An alternative expression for the orientation averaged polarizability is
A, el
A iu c2e ℏ2 fj0 A
α0 = = β0 u
ℏ me j 0 A
2
ϵj0 + u2

where the averaged electronic oscillator strength is

2
A, el A A
fj0 = 2me 3c2e ℏ2 ϵj0 j Del 0 from the dipole vector at fixed
Q
nuclear positions.

3.4.2 Combination Rules

Approximate expressions for the dynamical polarizabilities translate into useful
approximations for the dispersion coefficients. They also lead to combination
rules which provide values of the coefficient for the pair A–B in terms of the
coefficients of A–A and B–B.
Provided the optically active electrons in compounds A and B, typically their
valence electrons, can be described as independent on the average, the well-
A, el A∗
known Kuhn–Thomas sum rule j fj0 = Nel , the number of active elec-
trons of A, and similarly for B, gives for u ∞ the asymptotic value
∗
A A A A
u2 β0 ≈ c2e ℏ2 me Nel . This together with the static value β0 0 = α0 0
at zero frequency gives an interpolation function
2 −1
A A u
β0 u = α0 0 1+
uA
2 ∗
A A
with a new parameter satisfying α0 0 u A = c2e ℏ2 me Nel , that can be
used to evaluate the integral over u [20]. It gives the approximate dispersion
coefficient
 3.4 Calculation of Dispersion Energies 83

A B
AB 3 α0 0 α0 0
C6 = ∗ 1 2 ∗ 1 2
2 A A B B
α0 0 Nel + α0 0 Nel

which can be calculated from static polarizabilities and the known number of
valence electrons for A and B.
A slightly modified version of this procedure provides useful combination
AA
rules [21, 22]. The parameter u A appears in C 6 and it can be extracted as
AA A 2
u A = 4 3 C6 α0 0 which involves two measurable quantities. Its
values are given in [22] for atoms and small molecules and are (in au’s) for
He, Ar, Na, H2, and CH4 equal to 1.02, 0.706, 0.0774, 0.610, and 0.649, respec-
tively, with larger values of 1.71 and 1.401 for He+ and Na+, all relating to elec-
A
tronic excitation energies. Using the form of u A in β0 u , and performing the
integration over u for the A–B pair gives the combination rule
AA BB
AB 2 C6 C6
C6 = AA BB
α B α A C6 + α A α B C6
which allows calculation of the dispersion coefficient for two different interact-
ing compounds A and B, given the coefficients for the two identical pairs A–A
and B–B, and has been found to be accurate within a few percent for many com-
binations of atoms [22].

3.4.3 Upper and Lower Bounds

The second-order perturbation energies providing induction and dispersion
interactions require sums over excited states, most of which are usually not
known. Approximate expressions can however be obtained that involve only
a few excited states, and that provide upper and lower bounds to the second-
order energies. Here, we follow an operator-based treatment [23, 24] which
is also closely related to a method of sum over states [13].
Writing the unperturbed, zeroth order states of pair A–B as a product of states
of A and B like k AB = p A q B, the second-order energy for the ground state
0 AB = 0 A 0 B can be written in terms of the resolvent operator
k k
R0 = − k ≥1 0 0
Ek − E0
where unperturbed pair energies have been ordered in the sequence
0 0 0
E0 ≤ E1 ≤ E2 ≤ and so on, and all denominators are positive valued giving
a negative valued operator. The second-order energy is now
2 1 1
E0 = 0 H R0 H 0
AB
84 3 Quantitative Treatment of Intermolecular Forces

0 0 0 0
Replacing in the resolvent Ek − E0 with the smaller E1 −E0 in all the quo-
tients, so they all become larger, and using the completeness relation
k ≥ 0 k k = I, the identity operator, and the complementary projector
P = I − 0 0 it follows that
P
0 ≥ R0 ≥ −
0 0
E1 − E0

which gives the energy bounds

1 1
0 H PH 0
2
0 ≥ E0 ≥− AB
0 0
E1 − E0

This can be calculated from a knowledge of only the 0 AB = 0 A 0 B state of

1 1 1
the pair and the integrals in 0 H 0 and 0 H H 0 , and has a
1
numerator containing the squared standard deviation of H in the state 0 AB.
For the dispersion energy of two hydrogen atoms in their ground state 0 = 1s0 ,
1 −1
the dipole–dipole interaction Hamiltonian is H = 4πε0 Ce2 xA xB + yA yB
1 1 1
− 2zA zB R − 3 and its averages are 0 H 0 = 0 and 0 H H 0 =
−2 2
4πε0 Ce4 R −6 2 r 2 1s 3, after consideration of spatial symmetry and selection
0 0
rules. This together with r 2 1s = 3a20 and E1 − E0 = C E2p0 − E1s0 = 3Ce2 4a0
gives a dispersion energy lower bound of V dsp R C − 8 0 Ce2 a0 a60 R6 , with
Ce and a0 the electron charge and the Bohr radius both equal to 1.0 in atomic
units, where 4πε0 = 1.0. This lower bound contains a numerical factor of the
correct magnitude and not far from the actual value of nearly 6.50 au. The next
1
term in the multipole expansion of H comes from the dipole–quadrupole
interaction and the dispersion energy can be similarly calculated to find in
V (dsp) a factor 1 + 22 5a20 R2 [1]. This indicates that asymptotic convergence
for large R must start around R ≈ 5.0a0.
Better bounds can be found if one knows ground and excited states up to k =
N, in which case one can define a partial resolvent and write

N N k k N ∞ k k
R0 = − k =1 0 0
≥ R0 ≥ R0 − 0 0
Ek − E0 k =N +1
EN + 1 − E0

which gives the better bounds

 3.4 Calculation of Dispersion Energies 85

1 N 1 2
0 H R0 H 0 ≥ E0
N 0 0
≥ 0 H 1
R0 H 1
0 − 0 H 1
I− N
H 1
0 EN + 1 −E0
AB

N
with = N k = 1 k k a projection operator on the known subspace of N
states excluding the ground state k = 0. A calculation can now proceed involving
integrals of states only within this subspace. Induction and dispersion energies
can be obtained as done above, by letting the states k contain only one ground
state of the A–B pair, to generate the induction interaction of one permanent
multipole and a transient one, or by excluding the two ground states, to generate
the dispersion interaction of two transient multipoles.
Calculations can be done quite generally introducing a new basis set suitable for
N N
expansion of the R0 and operators [23, 24]. Using a notation with the 1 × N
row matrix of chosen states f = [ f1 , … , fN ], and constructing the projector
N −1
= f f f f where f is a N × 1 column matrix, a positive-valued
1 2 N 1 2
operator A is found to satisfy the inequality g A g ≥ g A A g .
1 1 2
This can be used with g = H 0 , A = − R0 , and − R0 f = h , all
orthogonal to 0 , to obtain the upper bound
1 0 −1 1
2 0
E0 ≤ 0 H h h E0 − H h h H 0
−1 1
0
0 0 0
Using instead A = R0 −P E0 − E1 and A2 f = H − E1 k , with k
in the orthogonal complement space of P, one obtains a lower bound from [23]
1 1
0 H PH 0
2
g A g = E0 − 0 0
E0 − E1
1 0 0 0 0
≥ 0 H H − E1 k k H −E0

0 0 −1 0 0 1
× H − E1 k k H − E1 H 0

The calculation of bounds to the induction energy in the pair A–B in its
ground state can be done selecting the functions in h and k to generate
the interaction of a permanent dipole with a transition dipole, with 0 = 0A,
0B , hj = 0A, hBj or hk = hAk, 0B , the hAk orthogonal to 0A , and similarly
for B and for the kj . The dispersion energy calculations must involve states hj =
hAj, hBj where both components are orthogonal to the ground states of A and B.
86 3 Quantitative Treatment of Intermolecular Forces

The sets h and k can be also chosen to be a single variational state χ = χ A, χ B

which can be varied to maximize or minimize bounds as needed.
These procedures allow the introduction of any convenient basis set, which
need not be made of unperturbed eigenfunctions. In an application to the dis-
persion energy of two hydrogen atoms, it is found that using just a few hydro-
genic states for each atom is not accurate because of the neglect of the hydrogen
atom continuum. Better results are obtained with a new basis set of functions
r nlm of the electronic coordinates constructed from Laguerre radial func-
tions, and sequentially adding new basis functions [23]. In this way upper
and lower bounds to the dispersion energy for the ground state of H + H can
be made to converge to the highly accurate value C6 = 6.4990267 au using only
seven basis functions.
Dispersion interaction energies can also be obtained doing Gaussian quadra-
tures for the integrals over imaginary frequencies appearing in the Casimir–
Polder expressions, using Pade approximants for the polarizabilities versus fre-
quency, or choosing the expansion states in h or k to generate energies as
sums over states of terms containing oscillator strengths. Some of these proce-
dures have been compared in the literature [4, 13, 25].

3.4.4 Variational Calculation of Perturbation Terms

A computationally convenient alternative to the calculation of perturbation cor-
rections to the energy by means of expansions in basis sets is to relate the wave-
function equations for each perturbation order to variational functionals
optimized to obtain perturbed functions and energies. This was done first for
the dispersion interaction energy of H + H and He + He [26], and was later
expanded within the perturbation theory of intermolecular forces [4, 13, 27].
0 1
The functional Φ = Φ H + λH −E Φ of a variational function
Φ can be made stationary, with variations δ = 0 giving an equation for Φ or its
parameters, and then = 0 giving its energy E. These two can first be expanded
in powers of λ to extract the functional at each order. To simplify here, one can
0
assume that the exact solution to H Φ 0 = E 0 Φ 0 is known and normalized,
0 p
and let Φ = Φ 0 while solving for the higher orders so that Φ Φ0 =0
for p ≥ 1. This in particular gives to second order,
1 0 0 1 0 1 1 1
2
Φ = Φ H −E Φ + Φ H −E Φ
1 1 1 0 2
+ Φ H −E Φ −E
1 0 1 0 0
with E = Φ H Φ known. With accurate functions Φ XA , XB of
electronic coordinates and choices of variational functions of the form
 3.5 Electron Exchange and Penetration Effects at Reduced Distances 87

Table 3.1 Dispersion coefficients from variational wavefunctions.

Atom pair C6(DD)(au) C6(DQ) (au) C6(QD) (au)

H+H 6.4990 124.399 124.399

H + He 2.817 13.14 28.64
He + He 1.459 7.05 7.05

1 0
Φ XA , XB = Φ XA ,XB f XA ,XB H AB XA ,XB where f(XA, XB) is a sum of
factors with powers of electronic coordinates in A and B, it has been possible
DD
to obtain this way highly accurate values of dipole–dipole C6 and dipole–
DQ
quadrupole C8 dispersion energies for interacting atoms with one and two
electrons, including H + H, H + He, and He + He [13]. Their collected values
are shown in Table 3.1, in au’s.

3.5 Electron Exchange and Penetration Effects

at Reduced Distances
3.5.1 Quantitative Treatment with Electronic Density Functionals
As the interacting species A and B are brought closer with decreasing R, elec-
tronic clouds start to penetrate each other, and it is necessary to account for
quantal electron exchange. Effects depend on whether the electronic shells of
the species are closed, with full occupation of electron orbitals, or open due
to partial occupations. For a species with closed electronic shells, a resulting
physical effect is the Pauli exclusion of electrons from occupied orbitals, which
amounts to a repulsive force, with electronic density moving away from the
intermolecular region and into the outer regions. If one or both species have
an open shell with partly occupied orbitals, electrons can move into the inter-
molecular region to form covalent bonds and create attraction forces. In addi-
tion there is a probability of electron transfer forming pairs A+ + B− or A− + B+,
which brings in electrostatic attraction forces. To obtain these forces as func-
tions of the distance R one must properly antisymmetrize the total electronic
wavefunctions with respect to electron exchange, and must allow for electronic
redistribution. The electronic wavefunction Φ(AB)(X; Q) of all electronic posi-
tion and spin variables in X = (XA, XB), for fixed nuclear (or ion core) positions
Q = Ra , can be constructed to contain a set λ = {λj} of parameters chosen to
be optimized. This is done requiring that the total energy functional
Φ AB = Φ AB H Q Φ AB Q Φ AB Φ AB Q = E AB λ; Q must have a
88 3 Quantitative Treatment of Intermolecular Forces

ε
unocc

occ
A B A B A B A B
Electrostatic Polarization Electron Electronic
interactions (induction and exchange excitation
dispersion) and transfer

Figure 3.3 Schematic illustration of electrostatic, polarization (induction and dispersion),

electron exchange, and electronic excitation phenomena in transitions involving closed
(shown dashed) and open (shown empty) electronic shell interactions.

minimum as a function of the parameters, for the ground state or for each state
of lowest energy with a given symmetry given by fixed Q. The optimized state
Φ(AB) can be taken as the reference or zeroth order state to be further corrected
by combining it with a basis set of orthonormal states to be used in a perturba-
tion expansion.
A qualitative discussion can be based on electronic orbitals and their charge
distributions. The several phenomena, appearing at short and long distances for
closed–closed, closed–open, and open–open shell interactions are schemati-
cally illustrated in Figure 3.3.
For each species A or B, electronic energy levels ε of orbitals are found to be
occupied as shown by the dashed bands in the figure, or to be unoccupied. Elec-
trostatic interactions are always present and obtained from electronic charge
densities, polarization (induction and dispersion) occurs at both long and short
distances, electron exchange appears as the distance is reduced and orbitals start
to overlap, and electronic excitation and charge transfer occur more likely at
short distances due to electron correlations and the Pauli exclusion of electrons
from occupied orbitals.
A great deal about interactions at intermediate and short distances can be
understood and estimated from the relationship between energies and elec-
tronic densities established by the Hohenberg–Kohn theorem [28]. Given a
set of nuclei a (or ion cores) at fixed positions Ra , electrons are distributed
among them and show unique spikes at their locations. This means that the
external potential energy Vext r provided by the nuclei uniquely creates an
electron number distribution ρ r per unit volume. The theorem states that
the reverse is also verified for the nondegenerate electronic ground state and
lowest energy electronic state of each symmetry, and that there is a unique func-
tional ρ r which reaches a minimum when a trial density function is varied
to reach the correct one. The density derives from a many-electron
 3.5 Electron Exchange and Penetration Effects at Reduced Distances 89

wavefunction Φ[X; ρ r which is constrained to provide the density but may be

unknown in detail. Letting the potential energy of repulsion among nuclei be
VNN(Q) and the interaction energy between an electron and all the nuclei
vNe r; Q , the total energy is decomposed as

ρ r ; Q = VNN Q + d 3 r ρ r vNe r; Q + F ρ r ; Q

where F ρ r ; Q = Φ ρ r K el + H ee Φ ρ r Q in terms of the

electronic kinetic energy and electron–electron potential energy operators.
Further separating in F the Coulomb electron–electron interaction energy
VCoul ρ r the remainder is a functional containing kinetic energy, exchange
and correlation effects, so that
F ρ r ; Q = VCoul ρ r + Gkin ρ r ; Q + Gxc ρ r ; Q
with the last term describing a combined exchange-correlation energy that
can further be decomposed into separate exchange and correlation terms
as Gxc ρ r ; Q = Gexc ρ r ; Q + Gcor ρ r ; Q .
The electron–electron Coulomb energy functional follows from the Hamilto-
nian operator
ee 1 c2e
H = d 3 r1 d 3 r2 δ r 1 −r m δ r 2 −r n
2 r 1 −r 2 m n

m n
δ r 1 −r m δ r 2 −r n

= m, n
δ r 1 −r m δ r 2 −r n −δ r 1 −r 2 m
δ r 1 −r m

where the double sum in the last line includes m = n. This gives

m n
δ r 1 −r m δ r 2 −r n = ρ 1 r 1 ρ 1 r 2 − δ r 1 −r 2 ρ 1 r 1

with ρ 1 r = 1≤m≤N δ r −r m , to be replaced in the double integral. The

ee
expectation value of H for a given wavefunction then leads to the functional
of the electronic number densities ρ r at locations r and r ,
1 3 c2e
VCoul ρ r = d r d3 r ρ r ρ r −ρ r δ r − r
2 r−r
with c2e = e2 4πε0 in SI units or c2e = e2 in cgs statCoul units.
This provides a framework for discussion of the intermediate and short-
distance interaction energies for species with closed or open electronic shells.
The G ρ r ; Q functionals have been extensively treated and constructed from
90 3 Quantitative Treatment of Intermolecular Forces

information about the inhomogeneous electron gas [29, 30] and by parametri-
zations designed to reproduce the energetics of sets of molecules [31]. The so-
called DFT approach has been widely applied to molecular and solid-state
phenomena. Separating nuclear position variables as Q = R,QA ,QB and doing
calculations for varying R, the most recent functionals give quite acceptable
results for ground electronic state equilibrium distances Re and dissociation
energies De = E0(∞) − E0(Re) in many applications involving intermediate and
short distances.
However, in applications of intermolecular forces to the A–B pair, it is neces-
sary that as R ∞, one must find ρ AB r; R ,QA ,QB ≈ρ A r; QA + ρ B r; QB ,
and also that the functional must satisfy the limiting form

FAB ρ r ; R,QA ,QB ≈FA ρ A r ; QA + FB ρ B r ; QB

This is a demanding restriction which is not obeyed in the available functionals

and is the subject of active research at this time.
For the intermediate and short ranges (or reduced range, RR), two useful com-
putational procedures have been devised for a system with Ne electrons. One is
suitable for electronic rearrangement and bonding [32], with electrons of spin
states η = α, β described by independent (Kohn–Sham, or KS) spin-orbitals
η
ψ j r,ζ = ψ j r η ζ , chosen as the trial functions to be varied in the func-
tionals, in terms of which the electron density for each spin is
η 2
Ne η
ρη r = j=1
ψj r

and the total density is ρ r = ρ α r + ρ β r . The other (Gordon–Kim, or GK)

[33] is suitable for the interaction of two species where the densities of A and B
do not change much over distances R, so that ρ AB r; R ,QA , QB = ρ A r; QA +
ρ B r; QB is a good approximation at all R. These are complementary
approaches which can be applied to species with closed or open electronic
shells.
For a pair of species both of which have closed shells, such as Ne + Ne, the
total density can be simply chosen as ρ AB r = ρ A r + ρ B r . Each species
density is obtained from its electronic structure and the functional
F AB ρ r does not need variation to obtain an optimal minimum. Results
are acceptable for the short and intermediate ranges and provide a good esti-
mate of equilibrium distance and dissociation energy. It must however be
extended to include the long-range polarization terms, and this has been done
to some extent in following work, by allowing displacement and distortion of
the densities [34]. Applications with some improvements in the functionals
 3.5 Electron Exchange and Penetration Effects at Reduced Distances 91

have included diatomic and polyatomic species for which the dispersion energy
terms have been added at large distances [35–37] Calculation of interaction
energies at very short ranges also requires adding distortion densities for very
short distances [38].
For a pair of species, such as Ne + Na, where one has closed shells and the
other has open shells, it is necessary to allow for a partial redistribution of
the electronic density of the open shells and to work with spin densities to allow
for the unpaired electrons. If these are in species B, a possible choice of a
∗
variational density is ρ AB r = ρ A r + ρ B r where the second term can bet-
2
ter be treated within the KS formalism by choosing ρ B, η r =
B
j cjη φj r ,a
sum including orbitals at species B with electron spin η, forming hybrid atomic
orbitals, or more general distortion orbitals. If B is a molecule, coefficients of
molecular orbitals can be optimized by minimizing its total energy functional.
For two species with open shells, such as Na + Na, the KS choice is suitable for
intermediate and short distances, and the functional must be minimized with
AB
respect to variations of KS spin-orbitals φjη with occupation numbers νjη
2
AB
that extend over the two species, as ρ AB r = jη νjη φjη r with
AB
φjη r = μ cμ, jη χ μ r a linear combination of atomic orbitals χ μ centered at
the nuclei. This is best done in the case of chemical bonding by introducing spin
densities to account for open shells of separated species [39].
The procedure for pairs of atoms was implemented in early work using a den-
A B
sity ρ AB r = ρHF r + ρHF r with the atomic densities from Hartree–Fock
calculations, and with a simple Thomas–Fermi–Dirac functional for
F ρ r ;R [40]. Results for the short-range interaction energy of many (Z =
2–36) atom pairs could be well fitted at short distances with the Born–Mayer
function V(R) = A exp(−bR), and it was found that reasonable accuracy, of the
order of 1%, could be obtained for different atoms A and B from the combina-
tion rule V(AB) = (V(AA)V(BB))1/2. This approach was substantially extended with
the GK procedure to include both short and intermediate distances and to
obtain the location and depth of van der Waals potential minima. Using
Hartree–Fock atomic densities, and kinetic energy, exchange, and correlation
energy functionals for the inhomogeneous electron gas provided by
Thomas–Fermi, Dirac, and a local energy density expansion, respectively
[33], remarkably good results could be obtained by comparison with ab initio
calculations and experiment. This is shown in Figure 3.4 for Ar + Ar, and in
Table 3.2 where GK values for Rm and ϵm are compared to experimental ones.
For Ne–Ne it gives Rm = 2.99 Å = 299 pm and ϵm = 56 × 10−16erg = 56 × 10−23 J
for the minimum location and well energy, compared with experimental values
of Rm = 303 pm and ϵm = 63 × 10−23 J.
92 3 Quantitative Treatment of Intermolecular Forces

10
x
x
x
x
1
x

10–1
eV

10–2
V(R)

10–3
0
10–16 erg

–100

–200

2.0 3.0 4.0 5.0

R (Å)

Figure 3.4 The Ar–Ar potential energy versus interatomic distance: full line GK calculations
[33]; dashed line from gas experiments, and dotted line from beam scattering experiments.
Source: from Gordon and Kim [33].

Table 3.2 Calculated and measured values of potential energy parameters (from [33]) in
SI units.

Atom Rm (pm) Rm (pm) ϵm (×10−23 J) ϵm (×10−23 J)

pair calculated measured calculated measured

Ne–Ne 299 303 56 63

Ar–Ar 363 370 175 195
Ne–Ar 342 338–364 81 84–115
Source: Reproduced with permission of AIP Publishing.
 3.5 Electron Exchange and Penetration Effects at Reduced Distances 93

Similar accuracy is obtained for other pairs of atoms. The treatment was later
improved with a more accurate functional correcting for the electronic self-
energy and was applied to molecular species with similar encouraging results,
also for interaction potential anisotropies [35].
At present, there is no satisfactory density functional treatment of intermo-
lecular forces to cover all ranges with a consistent formulation, and it seems
best to construct the exchange-correlation potential energy in the form

Vxc R = Gxc ρ AB r ; Q −Gxc ρ A r ; QA − Gxc ρ B r ; QB

= Vxcnr R 1 − fd R; Rd , wd + fd R; Rd ,wd Vxcpol R

nr
where Vxc describes the exchange-correlation of electrons at short and inter-
mediate ranges, or near-range (nr), as modeled by a density functional, and
pol
Vxc includes long-range induction and dispersion energies, with a transition
between the two provided by a damping function fd going smoothly from 1.0 at
large R to zero at small R, with a drop around a damping location Rd of width wd.
pol
At large distances, Vxc R must correctly go into an exchange-correlation
expression constructed from properties of the separate species A and B. The
nr
Vxc R extended to large distances contains some of the long-range polariza-
pol
tion interaction, and conversely Vxc R at short distances contains some
electronic exchange-correlation. Therefore, it is necessary to dampen their
magnitudes as shown to avoid doubly counting of exchange-correlation. Vari-
ous choices have been proposed and tested for functions such as this, which
damp with a function fd R each long-range energy varying as R−n, containing
n

values of the parameters suggested by electronic structure treatments or chosen

for computational convenience to fall more rapidly than Rn as distances
decrease [41, 42]. The damping function must be chosen so that potential
energy surfaces or derived forces are smooth over all distances.
Alternatively, the near-range term may contain also the long-range electro-
static and induction energies when these appear in the DFT treatment, with
the remaining polarization term equal to a calculated dispersion energy. Ongo-
ing efforts to provide useful treatments at all distances have been presented in
several recent reviews [43–46].

3.5.2 Electronic Rearrangement and Polarization

A deeper and more general understanding of interaction energies requires treat-
ment of the electronic states of many-atom systems. It is easier to follow the
related mathematical treatment of electronic rearrangement in interactions,
by focusing on simple systems such as homonuclear and heteronuclear dia-
tomics, and results for H + H+, H + H, H + He, and He + He. The building
94 3 Quantitative Treatment of Intermolecular Forces

components are atomic orbitals (AOs) χ μ r , with μ = (Nnlm) for nucleus N = a,

b and atomic quantum numbers n, l, m. The treatment can conveniently be
done in a body-fixed (BF) reference frame where the z-axis points from nucleus
a to b. Wavefunctions depend then only on the distance R, and the Hamiltonian
has axial symmetry around the z-axis, as well as reflection symmetry through a
plane containing the z-axis.
For an electron in the field of two ion cores at positions Ra and Rb with ion
charges ZaCe and ZbCe, a simple molecular orbital written as a linear combina-
tion of atomic orbitals (a MO-LCAO) is
φj r = caj χ a r + cbj χ b r
to be normalized to φ φ = 1 and with two variational parameters cμj(R) which
can be obtained minimizing the electronic energy ϵ(el) for fixed nuclei, obtained
el 2
from the Hamiltonian for an electron (1), h 1 = − ℏ2 2me ∂2 ∂ r 1
− Za ce 2 ra1 −Zb ce 2 rb1 . This leads for each j = 1, 2 to the matrix equation

c
μ = a, b μ
Hμν − ϵ el Sμν = 0
el
with ν = a, b, Sμν = χ μ χ ν the overlap integral, and Hμν = χ μ h χ ν an inte-
gral for the electronic Hamiltonian matrix element. The AOs satisfy certain
combination rules: (i) They combine only when they have the same spatial sym-
metry, since otherwise Sμν = Hμν = 0; (ii) they combine more strongly (with com-
parable coefficient values) if Haa Hbb as seen from (ca/cb)2 = (Hbb − ϵ(el)Sbb)/
(Haa − ϵ(el)Saa); (iii) they combine more strongly when there is a larger overlap
between atomic charge distributions at a and b, which increases the magnitudes
of coupling terms Hab and Sab.
Solving for the coefficients requires that the determinant of the matrix they
multiply must be zero, or (α − ϵ)(β − ϵ) − (γ − ϵS) = 0 in the notation α = Haa,
β = Hbb, γ = Hab, ϵ = ϵ(el), S = Sab, and with Saa = Sbb = 1. This gives energy roots [47]

1 2
ϵ ± = − A ± A2 − B ,

− α + β + 2γS αβ −γ 2
A= , B =
1 − S2 1 −S 2
el el
and energy eigenvalues ϵ1 = ϵ − ≤ ϵ2 = ϵ + . The corresponding linear combina-
tion coefficients follow from
el el
cb γ − ϵj S α− ϵj
=− el
=− el
ca j β − Ej γ − Ej S
 3.5 Electron Exchange and Penetration Effects at Reduced Distances 95

and the normalization condition c2a + c2b + 2ca cb S = 1 for each j. Furthermore, the
MOs can be labeled by the axial rotation quantum number mz = 0, ± 1, ± 2, …, as
being of symmetry λ = mz and type σ, π, δ, …. The coefficients change with R
and in the limit R = 0 the MO becomes a united atom AO of nuclear charge
number Za + Zb and atomic quantum numbers (nlm), with m = mz. Further-
more, the energy functions of R for MOs of the same symmetry must not cross
as R changes, except by numerical coincidence, because any electronic coupling
el
term U in the Hamiltonian must then lead to coupling integrals
el
φj U φj different from zero and the energy roots of a new eigenvalue
equation must be different, indicating their repulsion. All this leads to a MO
energy correlation diagram as R changes from separated atoms at large values
to the united atom, as shown in Figure 3.5.

3dδ
3d
3dπ σ3sA 3sA
3dσ

3pπ π2pB 2pB

3p
3pσ σ2pB
2pA
π2pA
3s 3sσ
σ2pA
2sB
σ2pB
σ2sA 2sA
2pπ
2p
2pσ

2s 2sσ

σ1sB 1sB

σ1sA 1sA

1s 1sσ
United atom Separated atoms (ZA > ZB)
(R= 0) (R → ∞ )
R→

Figure 3.5 Energy-level correlation diagram (not to scale) for the one-electron heteronuclear
diatomic molecules. Source: Figure 4.1, page 96 of Bransden and McDowell [48]. Reproduced
with permission of Oxford University Press.
96 3 Quantitative Treatment of Intermolecular Forces

In H + H+, or for two identical ion cores, the molecular reflection symmetry
across a plane perpendicular to the internuclear axis leads to ca = ± cb for real
valued MOs, and the normalized LCAOs are φg , u r = χ a r ± χ b r
2 1 ± Sab 1 2 , with the gerade and ungerade forms for the + and − signs.
The nlm hydrogenic orbital for an electron at position r at a nucleus a with
nuclear charge Za is χ a, nlm r = r anlm = Rnl ra Ylm ϑa ,φa . For the 1s orbital
−1
χ a, 1s0 r = r a1s = Za a0 3 2 2 exp −2Za ra a0 2π 1 2 with a0 the Bohr
radius, of energy E1s = − Za2 e2 2a0 , when using Gaussian units [49]. Here,
Sab = S decreases exponentially for large R as exp(−R/a0). An energy correlation
diagram from the separated atoms to the united atom for homonuclear dia-
tomics is shown in Figure 3.6.
The total energy is the sum of nuclear repulsion plus electronic energy,
el
Eg R = c2e 4πε0 R + Eg , with

σu3s
3s
σg3s
3dδg
3d 3dπg
σg σu2p
3pπu
3p πg2p 2p
3pσu
3sσg πu2p
3s
σg2s
σu2s
2s
2pπu
2p
2pσu
σu2s
2s 2sσg

σu1s
1s

σg1s

1s 1sσg

United atom Separated atoms

(R = 0) R→ (R → ∞)

Figure 3.6 Energy-level correlation diagram (not to scale) for the one-electron homonuclear
diatomic molecules. Source: Figure 4.3, page 98 of Bransden and McDowell [48]. Reproduced
with permission of Oxford University Press.
 3.5 Electron Exchange and Penetration Effects at Reduced Distances 97

j+k
Egel R = ϵ1s −
1+S
2
Ce2 χ r Ce2 χ r χb r
jR = d3 r a , k R = d3 r a
4πε0 r −Rb 4πε0 r −Ra
where j is called the Coulomb integral and describes the attraction between the
electron in orbital a and the nucleus at b, a term expected from classical elec-
trostatic. The integral k, however, is purely quantal and accounts for the inter-
action of the overlap of charges centered at a and b with the nucleus at a, a
quantal penetration effect. For large distances, one has j R ≈Ce2 4πε0 R plus
a term exponentially decreasing as exp(−2R/a0), while k(R) ≈ 0 exponentially
as exp(−R/a0), so that the results from pure electrostatic are recovered at
large R.
The choice of atomic orbitals with a nuclear charge number Za = 1.0 gives a
potential energy V0(R) = E0(R) − E1s repulsive at short distances and attractive at
large ones with a minimum at Re = 2.50 au and a dissociation energy De = E0(∞)
− E0(Re) = 0.065 au, to be compared with the accurate values of De = 0.1026 au =
2.79 eV at Re = 2.00 au. This is qualitatively correct and can be improved mini-
mizing the energy functional φ with respect to parameters in a trial function.
A flexible trial function must describe the polarization of atomic orbitals, and
the effective nuclear charge at short distance keeping in mind that as the inter-
nuclear distance decreases the charge changes from Za = 1.0 au at large dis-
tances to the united atom limit at R = 0 of value Za = 2.0 au. This can be
done allowing for variable charge numbers Z in exponents to get the energy
optimized. Polarization of H by H+ can be treated adding to the 1s0 orbital a
portion of the 2pz orbital, pointing from one nucleus to the other, to form a
hybrid orbital χ a r = χ a, 1s0 r + λχ a, 2pz r , with
3 2 −1
Za Za ra Za ra
χ a, 2pz r = exp − 2π 1 2
cos ϑa
2a0 a0 2a0
and λ treated as a variational parameter which depends on R. Varying para-
meters for each R one finds optimal values Z1s(R), Z2p(R), and λ(R) [50] and a
minimum at Re = 2.01 au with De = 0.1004 au = 2.73 eV when Z1s = 1.246,
Z2p = 2.965, and λ = 0.138, a much better result, showing the importance of
introducing effective nuclear (or ion core) charges and orbital hybridization
as functions of a decreasing distance.
A connection can be made with the interaction energy at large distances.
Keeping the variation parameter λ but letting Z1s = Z2p = 1.0 at large distances,
the minimum of λ provides the analytical form of the parameter and the
energy, which is found to agree with the long-range charge-induced dipole
interaction energy going as R−4 with a transition dipole and a polarizability
obtained from the integral 1s0 z 2pz . The long-range dependence on R−1,
98 3 Quantitative Treatment of Intermolecular Forces

however, is altered in this homonuclear system when there is electronic excita-

tion, by the quantal coherence between the two equivalent states of the electron
at large distances, which leads to a R−3 resonance interaction in first order of
perturbation theory.
The MO constructed for the ground state of H + H+ can be labeled by its sym-
metry as the 1σ g MO φg with energy E0(R) = Eg(R), and describes how the elec-
tronic distribution extends into the internuclear region. An excited 1σ u with the
MO φu r = χ a r − χ b r 2 1 − Sab 1 2 has the energy
Ce2 j−k
Eu R = + ϵ1s −
4πε0 R 1−S
and describes an electron distribution that has instead moved from inside the
internuclear region to the outside. When more than one electron is present out-
side closed subshells, their correlation leads to interactions between electronic
configurations made up of both φg and φu MOs. In general for a homonuclear
diatomic one can locate an electron at each nucleus with the MO combinations
that reconstruct AOs, χ a, b r = 1 + Sab 1 2 φg r ± 1 − Sab 1 2 φu r .

3.5.3 Treatments of Electronic Exchange and Charge Transfer

Turning to H + H, with two interacting electrons, the new features as distances
decrease are electron exchange and correlation changing with R. Given the AOs
χ a r and χ b r , two electron states Φ(x1, x2) depend on position and the spin
variable ζ in each xj = r j ,ζj , and must be constructed to satisfy antisymmetry
with respect to electron exchange. This can be done applying the antisymmetry
projection operator = 2 − 1 I − 12 written in terms of the permutation oper-
2
ator 12 of electrons 1 and 2, that is idempotent insofar = . A valence bond
(VB) function with the electrons in spin up and spin down states α(ζ) and β(ζ)
coupled to form a singlet state (of total spin quantum numbers S = 0, MS = 0)
Θ0,0(ζ1, ζ 2) = [α(ζ1)β(ζ2) − α(ζ 2)β(ζ1)]/21/2 is, normalized and with a short form
for arguments,

Φ cov x1 ,x2 = Ccov χ a 1 χ b 2 Θ0, 0 1, 2

χa 1 χb 2 + χa 2 χb 1
= 1 2
Θ0, 0 1, 2
2
2 + 2 Sab
which describes the covalent bonding of the atoms. This form can be used in
calculations with the AOs chosen as 1s orbitals or as hybrids, to form
the H2(1Σ) state. The electronic Hamiltonian can be decomposed as
el
H = H a 1 + H b 2 + H ab 1, 2 with H a 1 χ a 1 = ϵa χ a 1 involving the
 3.5 Electron Exchange and Penetration Effects at Reduced Distances 99

atomic Hamiltonian and its 1s AO, and H ab 1,2 for the interatomic Coulomb
interactions. The corresponding electronic energy is given by the variational
function for this state, as

cov Qab + Aab

E = 2ϵa +
1 + Δab
where Δab = Sab 2
, Qab = d 1 d 2 χ a 1 χ b 2 H ab 1,2 χ a 1 χ b 2 is an intera-
tomic Coulomb integral, and Aab = d 1 d 2 χ a 1 χ b 2 H ab 1, 2 χ a 2 χ b 1
is an interatomic exchange integral containing the differential overlap
χ a(1)χ b(1). The Aab integral has negative values for H + H and contributes most
of the binding energy.
Numerical results show a repulsive potential energy at short distances and
attraction at long distances with a minimum giving values for Re and De. Using
AOs with a variable orbital exponent corresponding physically to an effective
nuclear charge number, which varies from 1.0 at large distances to 2.0 at the
united atom limit, the so-called Heitler–London–Wang wavefunction gives
Re = 0 744 A and De = 3.78 eV with an effective charge number ζ = 1.166, to
be compared with the accurate values Re = 0 741 A and De = 4.75 eV. The dis-
crepancy is due to a shortcoming of the wavefunction at short distances, where
the potential energy is found to be too repulsive because electron transfer at
short distances has been ignored. Results are much improved if one adds an
ionic state describing H+ + H−,
χa 1 χa 2 + χb 1 χb 2
Φ ion x1 ,x2 = 1 2
Θ0, 0 1, 2
2
2 + 2 Sab
to form a covalent-ionic extended VB function given by

Φ EVB x1 ,x2 = Acov Φ cov x1 ,x2 + Aion Φ ion x1 ,x2

This covalent-ionic trial function can be used to minimize the electronic energy
by varying the parameters Acov, Aion for each distance R, and provides more accu-
rate values De = 4.03 eV for ζ = 1.19 even when using only 1s atomic orbitals.
Introducing hybrid AOs which combine 1s0 and 2pz atomic orbitals further
allows treatment of induction and dispersion forces which then appear at large
distances and are given in terms of the transition dipole integrals 1s0 z 2pz .
Extensions to many-electron homonuclear or heteronuclear diatomics and to
polyatomics can be more readily implemented using an alternative description
based on MO-LCAO molecular orbitals, which for a diatomic is
φ r = ca χ a r + cb χ b r such as we found for H + H+. Antisymmetry is incor-
porated through the formation of Slater determinants, and a linear combination
of determinants, or configuration interaction (CI), can be formed to construct a
wavefunction of high accuracy. The procedure is to identify each element in a
100 3 Quantitative Treatment of Intermolecular Forces

N × N matrix [Akj] with the j-th one-electron spin-orbital ψ j(xk), a function of

the k-th electron position and spin variables, so that the variable index identifies
the k-th row, and the function index identifies the j-th column. The determinant
of the matrix is then det[Akj] = det[ψ j(xk)] and a determinant formed with a set
of N spin-orbitals ψ j1 ,ψ j2 ,…,ψ jN can simply be written as ψ j1 ψ j2 ,…,ψ jN with an
implicit understanding of the electron variables, designating a chosen electronic
configuration.
Early work on few-electron diatomics, including also discussions of physical
aspects important in the treatment of intermolecular forces, can be found in
Refs. [1, 10]. The coefficients of AOs in MOs, the coefficients in the linear com-
binations of configurations, and the exponents of AOs corresponding to effec-
tive nuclear charges have all been used as variational parameters to obtain
accurate results. An alternative to variation of AO exponents is to introduce lin-
ear combinations of AOs with several fixed Z-values (called double-Z, triple-Z,
…, basis sets), chosen to allow for electronic shell splittings and for variations of
effective nuclear charges with distance R [47, 51, 52].
el el el −1
For H + H the electronic Hamiltonian is H 1, 2 = h 1 + h 1 + c2e r12
and a single determinant trial wavefunction ΦRHF(1, 2) = 1σ gα(1)1σ gβ(2)|, in a
compact notation for determinantal wavefunctions, generates a restricted
Hartree–Fock state where the MO can be written as a LCAOs. This is accept-
able at short and intermediate distances, but incorrect at large distance. It is
found already for H2 in its determinantal state |1σ gα1σ gβ|, that a single deter-
minant gives the wrong asymptotic behavior, insofar it describes at large dis-
tances a mixture of H + H and H+ + H− states.
To avoid this, a useful procedure is to start from the VB states and re-express
its AOs in terms of MOs, in which case a linear combination of configurations,
or configuration interaction (CI), is usually found [47, 51]. This leads to the
appearance of two electronic configurations, |1σ gα1σ gβ| and |1σ uα1σ uβ|.
A linear combination of these two in a configuration interaction (CI)
wavefunction

Φ CI x1 ,x2 = AG 1σ g α 1 1σ g β 2 + AU 1σ u α 1 1σ u β 2
with coefficients to be varied, leads to exactly the same states and results
obtained from the covalent-ionic treatment [47]. Induction and dispersion
interactions can be described by introducing AO hybrids, or alternatively by
adding configurations constructed with an atomic basis set containing polari-
zation atomic orbitals.
The two-electron heteronuclear diatomic He + H+ can be treated using MOs
constructed with He 1s and H 1s AOs, with the simplest wavefunction of form
ΦRHF(1, 2) = 1σα(1)1σβ(2)|. Long-range induction and dispersion interactions
can again be incorporated by means of basis sets of AOs containing at least the
2pz type in hybrid AOs, or by means of CIs. An accurate treatment here involves
 3.5 Electron Exchange and Penetration Effects at Reduced Distances 101

several configurations, capable of describing the electronic structures going to

both He(1s2) and He+(1s) + H(1s) at large distances, and structures that go to the
united atom with configuration 1s2 for a nuclear charge number Z = 3 for short
distances. Other configurations are also relevant to the description of this sys-
tem at intermediate distances [48].
The four-electron system He+He is an example of two closed shell species
which can also be described by a CI. The leading configuration involves MOs
as linear combinations of 1s orbitals at each He, and a single determinant
ΨRHF(1, 2, 3, 4) = 1σ gα(1)1σ gβ(2)1σ uα(3)1σ uβ(4)| suffices to qualitatively
describe the short-range interaction and to go to the correct asymptotic limit
He(1s2) + He(1s2) at large distances. However, to obtain an accurate description
at all distances, it is necessary to work with a larger basis set of AOs and at least
include atomic polarization to generate induction and dispersion interactions as
R decreases. The united atom limit is 1s22p02.
The accurate treatment of two interacting many-electron species at all dis-
tances is very challenging. Theoretical methods and computational algorithms
are being developed to account for many related properties describing systems
as supermolecules. In addition to CI treatments, many-electron perturbation
theory of the Moeller–Plesset type, with proper accounting of linked diagrams
to assure electronic size consistency [52], coupled cluster theory [53], and sym-
metry adapted perturbation theory are important methods available and being
developed to do ab initio calculations of intermolecular forces [54].
For systems with two or more electrons, interacting species may form states
of higher total spin. For example in H + H, the H2(3Σ) is formed from the
spin states ΘS, Ms ζ 1 , ζ2 with S = 1, MS = 0, +1, −1 which are given
by Θ1,0(ζ1, ζ2) = [α(ζ1)β(ζ 2) + α(ζ2)β(ζ 1)]/21/2, Θ1,1(ζ1, ζ2) = α(ζ1)α(ζ2), and
Θ1,− 1(ζ1, ζ2) = β(ζ 1)β(ζ2), all of which are even under electron permutation and
must be matched by odd permutation functions of positions. A simple state of
valence-bond type is C[χ a(1)χ b(2) − χ a(1)χ b(2)]Θ1,0(1, 2) which can be re-expressed
in terms of MOs φg , u r as done for the bonding H2(1Σ). The electronic energy
expression for the new triplet state, however, contains the VB atomic exchange
integral with the opposite sign and its energy is purely repulsive at all distances,

rep Qab − Aab

E = 2ϵa +
1 −Δab

and goes to the H(1s) ground states at large distances.

In a MO description, the state H2(3Σ) is given by the determinantal wavefunc-
tion |1σ gα1σ uβ| and this goes in the united atom limit to the higher atomic con-
figuration 1s12p01, which alternatively explains its repulsive energy at short and
intermediate distances. This H2(3Σ) state does also have a shallow well at large
distances as the result of attraction due to long-range polarization, which can be
described with hybrid AOs in a VB treatment, or with a CI in a MO treatment.
102 3 Quantitative Treatment of Intermolecular Forces

3.6 Spin-orbit Couplings and Retardation Effects

A more detailed treatment of molecular interactions requires consideration of
quantum relativistic phenomena. The imposition of relativistic invariance of
quantum mechanical and electromagnetism equations with respect to the Lor-
entz transformation means that one must consider coupled electric and mag-
netic fields in accordance with the Maxwell equations, and the Dirac equation
for the electron in an electromagnetic field [55]. However, for the purpose of
calculating interaction energies it is sufficient to consider the limit of small elec-
tron velocities v relative to the speed of light c, and to obtain results by mean of
perturbation theory with the small expansion parameter v/c. Two main conse-
quences are the appearance of new Hamiltonian terms involving the electron
spin, and retardation effects in the very long range interaction of species.
Molecules with heavy atoms of large atomic number Z have their electronic
structure affected by relativistic phenomena associated to large electron speeds
with a fraction of the speed of light. In these cases, electron states must be
obtained from a relativistic equation. For states of a single electron described
by the Dirac equation, states are given by spinors with four spatial components,
with two classified as strong and two as weak ones insofar they are smaller to
order (v/c)2. Molecular interactions can be accurately treated with a formalism
which keeps only the strong components and gives rise to the electron intrinsic
angular momentum, or spin, s with spin-up and spin-down electronic states
[11]. This spin creates a magnetic dipole μs which interacts with a magnetic field
to add an energy −μs to the Hamiltonian operator for the electron. An elec-
tron moving in the field of a nucleus also acquires a spin-orbit energy coupling
hSO = ξ r s l dependent on the electron distance r to the nucleus and its
momentum p through the orbital angular momentum l = r × p. The Hamilto-
nian contains in addition other terms dependent also on the electron momen-
tum: a mass-polarization term and the so-called Darwin term involving the
electronic density in contact with an electron-positron distribution at the
nucleus, both of which are of the order α2 ≈ (1/137)2, with α = e2/ℏc the fine
structure constant, compared to the nonrelativistic 1s orbital energy, and are
therefore usually insignificant in calculations of intermolecular energies.
However, the spin-orbit coupling energy can make a difference for heavy
atoms and in the interaction of systems at very large distances R ≥ 100 a0. Rel-
ativistic effects in systems with two or more electrons can be treated with the
Breit–Pauli equation, as given in appendix A of Ref. [10] and in [56]. The appli-
cations to intermolecular forces are reviewed in Ref. [57]. They are based on the
Hamiltonian H = H NRel + α2 H Rel and a perturbation treatment with a basis set of
zeroth-order state eigenfunctions of the nonrelativistic operator H NRel . The
 3.7 Interactions in Three-Body and Many-Body Systems 103

relativistic interaction operator decomposes as H Rel = H LL + H SS + H SL + H MP +

H D , where in order the energy terms are coupling of magnetic dipoles created by
orbiting electrons; coupling of magnetic dipoles from electron spins; spin-orbit
couplings within each species and between spin in one and orbital angular
momentum in the other; relativistic, or mass polarization, correction due to var-
iation of the electron mass with its speed; and the Darwin-like contact term at
the atomic nuclei. The calculation of interaction energies in the simpler case of
atom–atom systems is done by constructing diatomic states of specific
electronic angular momenta, depending on the strength of spin-orbit cou-
plings [58], and proceeding to do perturbation theory for the relativistic
interactions [59].
The other mentioned important relativistic effect on intermolecular forces
appears at very large distances R between species A and B, so that photons of
wavelength λ emitted during electronic density fluctuations in one must travel
a large distance R ≈ λ to be absorbed by the other and are delayed by the finite
speed of light. The first treatment of this effect by means of quantum electro-
dynamics showed that the previous R−6 dependence derived for the dipole–
dipole dispersion interaction turns into R−7 due to retardation [19]. This can
be derived in a perturbation treatment where radiation is quantized and inter-
acting with A and B, using a Hamiltonian H = H A + H B + H L + H AL + H BL where
H L and H AL are the Hamiltonians of quantized light and of its interaction with
A. Perturbation terms must be added up to the fourth order (for photon emis-
sion and absorption at both A and B) to obtain a result valid for distances R
going from very large to the previous large distances and showing the change
in R−n behavior [60, 61]. An alternative to the perturbation treatment is pro-
vided by response theory including density fluctuations of both electrons and
light, that bypasses the need for high order perturbation theory and is conven-
ient for the treatment of very long-range forces in large systems [62]. Retarda-
tion effects in van der Waals interactions for two and three atoms have also been
considered within molecular quantum electrodynamics [63] in a treatment that
includes both short-range and long-range interactions. This describes the inter-
actions as resulting from the exchange of photons between two species and has
also been extended to three body interactions.

3.7 Interactions in Three-Body and Many-Body

Systems
3.7.1 Three-Body Systems
The long-range interactions of a system of three atoms A, B, and C was treated
using perturbation theory and was developed for a Hamiltonian operator con-
taining dipole–dipole interaction potentials between pairs of neutral atoms in
104 3 Quantitative Treatment of Intermolecular Forces

S = 0 states [64]. It was pointed out that the magnitude of a three-body disper-
sion interaction as obtained from third-order perturbation theory is of magni-
−1
tude Iat α3at R3AB R3BC R3CA where Iat and αat are atomic ionization and dipolar
polarization magnitudes, and RLM is the distance between atoms L and M. The
angular dependence of the energy on the interior angles of the triangle formed
by the atoms is 3 cos(γ A) cos(γ B) cos(γ C) + 1, with γ L the angle at the location of
atom L. This factor changes sign depending on the shape of the triangle and
indicates that the three-atom dispersion energy can be either positive or nega-
tive. Perturbation theory shows that the dispersion energy is pairwise additive to
second order, and that a three-atom interaction term appears first in the energy
function to third order. This can be expressed in terms of dipolar polarizabilities
for each atom. However, when species contain permanent multipoles, induction
energies appear, and it is found that pairwise additivity is lost even to second
order [12].
More generally A, B, and C can be molecules, and their pair interactions may
contain not only monopole and dipole operators but also higher order multi-
poles. It is convenient to introduce pair interactions in terms of charge distribu-
tions, and to give a more general treatment using the resolvent operator. The
simplest case involves three species sufficiently separated so that their proper-
ties are not affected by electron exchange. This allows allocation of charges I of
electrons and nuclei to each separate species L = A, B, C.
The Hamiltonian is now, with species labeled by indices L and M,

int int int

H ABC = H A + H B + H C + H ABC , H ABC RAB ,RBC ,RCA = H LM RLM
L M

−1
int −1
H LM RLM = 4πε0 d 3 r d3 s cL r L RLM +r L −sM cM s M

where the interaction is a sum of pairs, and the charge density operators are
functions of positions with respect to the center of mass of each species L or
M. The equation for three-body states can be solved expressing the three-body
resolvent operator in terms of two-body operators in a way similar to what has
been done for three-body transition operators [65–67]. In a compact notation,
arrangement a = 0 is A + B + C with all species noninteracting, and arrange-
ments a = 1, 2, and 3 refer to cases where A, B, or C are, respectively, noninter-
acting while the other two atoms interact. The zeroth-order Hamiltonian is
0
H = H A + H B + H C , and interaction Hamiltonians in each arrangement are
0 int 1
V = 0 and V a , so that the perturbation Hamiltonian is H ABC = λH = a V a .
Perturbation theory expressions give wavefunctions and energies to second and
third order. For each of the three species starting in its ground state 0L , the
0 0 A B C
unperturbed state is ΨG = 0A 0B 0C with energy EG = E0 + E0 + E0 .
 3.7 Interactions in Three-Body and Many-Body Systems 105

The first-order energy and state satisfy, for the given G state, λE 1
=
int
Ψ 0
aV a Ψ 0
= a Ea and
0
H −E 0
λΨ 1 = a
Eaint −V a Ψ 0
1
and second- and third-order energies are obtained from E 2
= Ψ0 H Ψ1 ,
1
and E 3
= Ψ1 H −E 1
Ψ 1 , and involve only Ψ(1). This satisfies an
inhomogeneous differential equation that can be solved by inspection in terms
of two-body solutions as in

Ψ1 = a
Ψa1
0
H −E 0
λΨa1 = Eaint − V a Ψ 0

where each arrangement term in the wavefunction involves excited states of two
1
species and the ground state of the third one so that Ψ1 is a combination of
states 0AjBjC and similarly for the other two arrangements. It follows from this
and from the orthonormality of the states jL , L = A, B, C, that one can separate
in each arrangement the induction interaction integrals, where one of the spe-
cies remains in its state and the other is excited, from the dispersion interaction
integrals where both are excited. This gives after analysis of all integrals of the
1
type Ψ 0 V a Ψb , containing products with arrangement interactions V a
and V b [1]
2 2 2
E ABC = Eind ABC + Edsp ABC ,
2 2 2 2
Edsp ABC = Edsp AB + Edsp BC + Edsp CA

and indicates that the dispersion interaction of the three species is additive to
the second order, while the induction interaction is intrinsically a three-body
function.
The analysis can be extended to third order where one has that

E 3
= Ψ1 a
V a − Eaint Ψ1

1
contains 27 integrals of the type Ψb V a Ψc1 involving products with the
three arrangement interactions V 1 ,V 2 , and V 3 . When the arrangement
interactions are expanded in the basis set jAjBjC , many of the integrals are zero
and the remaining ones can be classified as giving three-body induction or dis-
persion energies. In particular, the dispersion energy arises from the interaction
of density fluctuations of the three species. They undergo a sequence of
106 3 Quantitative Treatment of Intermolecular Forces

transitions from the ground state for each pair first to excited states which trans-
fer energy in a second interaction and next return to the ground states in a third
interaction, all involving transition charge densities jL cL r L jL . A special
case detailed in the literature considers the dipole–dipole interaction energy
operator
dd −1 L tξη
H LM RLM = − 4πε0 Dξ DM
ξ, η R3LM η
which when entered into the third-order dispersion energy leads to [1]
3 dsp 1 + 3 cos γ A cos γ B cos γ C
Edsp ABC = C9 ABC
4πε0 3 R3AB R3BC R3CA
dsp
with a factor C9 ABC which can be expressed in terms of the dynamical
polarizabilities of the three species, as [13, 68]
∞
dsp 3ℏ
C9 ABC = du α A iu α B iu α C iu
π 0
3
Some calculated values of Edsp for H and He systems, and discussions
of the role of nonadditivity in condensed matter are given in the literature [2,
13, 69].

3.7.2 Many-Body Systems

A system composed of N species A, B, C, …, N has a total interaction energy
Eint(AB … N) which can be decomposed into two-body, three-body, and further
energy terms, writing
Eint AB … N = E2 AB …N + E3 AB …N + … + EN AB …N
with E2(AB … N) calculated as the sum of all two-body interaction energies,
E3(AB … N) calculated from all groupings of three-body energies excluding
two-body ones, and so on. The last term is a single quantity obtained from
the interaction of the N bodies when contributions from all smaller groupings
have been subtracted.
The sum over many-body terms is not necessarily rapidly convergent. The
convergence can be improved introducing statistical concepts such as the
two-body potential of mean force for a collection of N identical bodies, which
applies to the interaction of two bodies when the other N − 2 bodies are
described as a medium with a statistical spatial distribution for given temper-
ature and average density. This has been done to obtain the energies of atomic
clusters, and cohesive energy of condensed matter systems such as molecular
liquids and crystals [2, 70, 71].
 References 107

References
1 Margenau, H. and Kestner, N.R. (1971). Intermolecular Forces, 2e. Oxford,
England: Pergamon Press.
2 Maitland, G.C., Rigby, M., Brain Smith, E., and Wakeham, W.A. (1981).
Intermolecular Forces. Oxford, England: Oxford University Press.
3 Dalgarno, A. (1961). Stationary perturbation theory. In: Quantum Theory, vol. 1
(ed. D. Bates), 171. New York: Academic Press.
4 Hirschfelder, J.O., Byers Brown, W., and Epstein, S.T. (1964). Recent
developments in perturbation theory. Adv. Quantum Chem. 1: 255.
5 Davydov, A.S. (1971). Theory of Molecular Excitons. New York: Plenum.
6 Foerster, T. (1948). Intermolecular energy migration and fluorescence. Ann.
Phys. 2: 55.
7 Jackson, J.D. (1975). Classical Electrodynamics. New York: Wiley.
8 Landau, L.D. and Lifshitz, E. (1975). Classical Theory of Fields, 4e. Oxford,
England: Pergamon Press.
9 Stone, A.J. (1996). The Theory of Intermolecular Forces. Oxford, England: Oxford
University Press.
10 Hirschfelder, J.O., Curtis, C.F., and Bird, R.B. (1954). Molecular Theory of Gases
and Liquids. New York: Wiley.
11 Cohen-Tanoudji, C., Diu, B., and Laloe, F. (1977). Quantum Mechanics, vol. 2.
New York: Wiley-Interscience.
12 Kaplan, I.G. (2006). Intermolecular Interactions, 81. New York: Wiley.
13 Dalgarno, A. and Davison, W.D. (1966). The calculation of van der Waals
interactions. Adv. At. Mol. Phys. 2: 1.
14 Zare, R.N. (1988). Angular Momentum. New York: Wiley.
15 Edmonds, A.R. (1960). Angular Momentum in Quantum Mechanics. Princeton
NJ: Princeton University Press.
16 Linder, B. and Rabenold, D.A. (1972). Unified treatment of van der Waals forces
between two molecules of arbitrary sizes and electron delocalization. Adv.
Quantum Chem. 6: 203.
17 Koide, A. (1976). A new expansion for dispersion forces and its applications.
J. Phys. B: Atomic Mol. Phys. 9: 3173.
18 Buckingham, A.D. (1967). Permanent and induced molecular moments and
long-range intermolecular forces. Adv. Chem. Phys. 12: 107.
19 Casimir, H.B.G. and Polder, D. (1948). The influence of retardation on the
London-van der Waals forces. Phys. Rev. 73: 360.
20 Mavroyannis, C. and Stephen, M.J. (1962). Dispersion forces. Mol. Phys. 5: 629.
21 Tang, K.T. (1969). Dynamical polarizabilities and van der Waals coefficients.
Phys. Rev. 177: 108.
22 Kramer, H.L. and Herschbach, D.R. (1970). Combination rules for van der Waals
force constants. J. Chem. Phys. 53: 2792.
108 3 Quantitative Treatment of Intermolecular Forces

23 Linder, P. and Lowdin, P.O. (1968). Upper and lower bounds in second order
perturbation theory and the unsold approximation. Int. J. Quantum Chem.
Symp. Series 2 S: 161.
24 Goscinski, O. (1968). Upper and lower bounds to polarizabilities and van der
Waals forces. Int. J. Quantum Chem. 2: 761.
25 Langhoff, P., Gordon, R.G., and Karplus, M. (1971). Comparison of dispersion
force bounding methods with applications to anisotropic interactions. J. Chem.
Phys. 55: 2126.
26 Slater, J.C. and Kirkwood, J.G. (1931). van der Waals forces in gases. Phys. Rev.
37: 682.
27 Dalgarno, A. (1967). New methods for calculating long-range intermolecular
forces. In: Advances in Chemical Physics, 143. New York: Wiley.
28 Hohenberg, P. and Kohn, W. (1964). Inhomogeneous electron gas. Phys. Rev. B
136: 864.
29 Lundqvist, S. and March, N.H. (eds.) (1983). Theory of the Inhomogeneous
Electron Gas. New York: Plenum Press.
30 Trickey, S.B. (Special Editor) (1990). Advances in Quantum Chemistry: Density
Functional Theory of many-Fermion Systems. San Diego, CA, USA:
Academic Press.
31 Parr, R.G. and Yang, W. (1989). Density Functional Theory of Atoms and
Molecules. Oxford, England: Oxford University Press.
32 Kohn, W. and Sham, L.J. (1965). Quantum density oscillations on an
inhomogeneous electron gas. Phys. Rev. A 137: 1697.
33 Gordon, R.G. and Kim, Y.S. (1972). Theory for the forces between closed shell
atoms and molecules. J. Chem. Phys. 56: 3122.
34 Kim, Y.S. and Gordon, R.G. (1974). Unified theory for the intermolecular forces
between closed shell atoms and ions. J. Chem. Phys. 61 (1).
35 Parker, G.A., Snow, R.L., and Pack, R.T. (1976). Intermolecular potential surfaces
from electron gas methods. I. Angle and distance dependence of the He + CO2
and Ar + CO2 interactions. J. Chem. Phys. 64: 1668.
36 Clugston, M.J. (1978). The calculation of intermolecular forces. A critical
examination of the Gordon-Kim model. Adv. Phys. 27: 893.
37 Cohen, J.S. and Pack, R.T. (1974). Modified statistical method for intermolecular
potentials: combining rules for higher van der Waals coefficients. J. Chem. Phys.
61: 2372.
38 Smith, F.T. (1972). Atomic distortion and the combining rule for repulsive
potentials. Phys. Rev. A 5: 1708.
39 Gunnarsson, O. and Lundqvist, B.I. (1976). Exchange and correlation in atoms,
molecules, and solids by the spin density functional formalism. Phys. Rev. B
13: 4274.
40 Gaydaenko, V.I. and Nikulin, V.K. (1970). Born-Mayer interaction potential for
atoms with Z=2 to Z=36. Chem. Phys. Lett. 7: 360.
 References 109

41 Knowles, P.J. and Meath, W.J. (1986). Non-expanded dispersion and induction
energies, and damping functions for molecular interactions with application to
HF-He. Mol. Phys. 59: 965.
42 Tang, K.T. and Toennies, J.P. (1984). An improved simple model for the van der
Waals potential based on universal damping functions for the dispersion
coefficients. J. Chem. Phys. 80: 3726.
43 Dobson, J.F. and Gould, T. (2012). Calculation of dispersion energies. J. Phys.
Condens. Matter 24: 073201.
44 Berland, K., Cooper, V.R., Lee, K. et al. (2015). van der Waals forces in density
functional theory: a review of the vdW-DF method. Rep. Prog. Phys. 78: 066501.
45 Grimme, S., Hansen, A., and Brandenburg, J.G.C. (2016). Dispersion corrected
mean-field electronic structure methods. Chem. Rev. 116: 5105.
46 Zhao, Y. and Truhlar, D.G. (2008). The M06 suite of energy functionals:
systematic testing of four M06 functionals and 12 other functionals. Theor.
Chem. Accounts 120: 215.
47 Levine, I.N. (2000). Quantum Chemistry, 5ee. New Jersey, USA: Prentice-
Hall Inc.
48 Bransden, B.H. and McDowell, M.R.C. (1992). Charge Exchange in the Theory of
Ion-Atom Collisions. Oxford, England: Clarendon Press.
49 Atkins, P.W. and Friedman, R.S. (1997). Molecular Quantum Mechanics.
Oxford, England: Oxford University Press.
50 Weinhold, F. (1971). Dickinson energy of H2+. J. Chem. Phys. 54: 530.
51 Szabo, A. and Ostlund, N.S. (1982). Modern Quantum Chemistry. New York:
Macmillan.
52 Hehre, W.J., Radom, L., Schleyer, P.v., and Pople, J.A. (1986). Ab initio Molecular
Orbital Theory. New York: Wiley.
53 Bartlett, R.J. (1981). Many-body perturbation theory and coupled cluster theory
for electronic correlation in molecules. Annu. Rev. Phys. Chem. 32: 359.
54 Jeziorski, B., Moszynski, R., and Szalewicz, K. (1994). Perturbation theory
approach to intermolecular potential energy surfaces of van der Waals
complexes. Chem. Rev. 94: 1887.
55 Messiah, A. (1962). Quantum Mechanics, vol. 2. Amsterdam: North-Holland.
56 Bethe, H.A. and Salpeter, E.E. (1957). Quantum Mechanics of One- and Two-
electron Atoms. Berlin: Springer-verlag.
57 Hirschfelder, J.O. and Meath, W.J. (1967). The nature of intermolecular forces.
Adv. Chem. Phys. 12: 3.
58 Herzberg, G. (1950). Molecular Spectra and Molecular Structure I. Spectra of
Diatomic Molecules, 2e. Princeton, NJ: Van Nostrand.
59 Chang, T.Y. (1967). Moderately long range intermolecular forces. Rev. Mod.
Phys. 39: 911.
60 Power, E.A. (1967). Very long-range (retardation effect) intermolecular forces.
In: Intermolecular Forces (Advan. Chem. Phys. vol. 12), 167. New York: Wiley.
110 3 Quantitative Treatment of Intermolecular Forces

61 Levin, F.S. and Micha, D.A. (eds.) (1993). Long-Range Casimir Forces: Theory and
Recent Experiments on Atomic Systems. New York: Plenum Press.
62 McLachlan, A.D. (1963). Retardation dispersion forces between molecules. Proc.
Royal Soc. A 271: 387.
63 Salam, A. (2011). Molecular quantum electrodynamics of radiation-induced
intermolecular forces. Adv. Quantum Chem. 62: 1.
64 Axilrod, B.M. and Teller, E. (1943). Interaction of the Van der Waals type
between three atoms. J. Chem. Phys. 11: 299.
65 Micha, D.A. (1972). Collision dynamics of three interacting atoms: the Faddeev
equations. J. Chem. Phys. 57: 2184.
66 Micha, D.A. (1985). Rearrangement in molecular collisions: a many-body
approach. In: Theory of Chemical Reaction Dynamics, vol. II, Chap. 3 (ed.
M. Baer). Boca Raton, Florida: CRC Press.
67 Micha, D.A. (2017). Quantum partitioning methods for few-atom and many-
atom dynamics. In: Advances in Quantum Chemistry, vol. 74, 107. London,
England: Elsevier.
68 Salam, A. (2016). Non-Relativistic Quantum Electrodynamics of the van der
Waals Dispersion Interaction. Cham, Switzerland: Springer.
69 Kihara, T. (1958). Intermolecular forces. Adv. Chem. Phys. 1: 267.
70 Kaplan, I.G. (1999). Role of electron correlation in nonadditive forces and
ab initio model potentials for small metal clusters. In: Advances in Quantum
Chemistry, vol. 31, 137. San Diego, CA: Academic Press.
71 McQuarrie, D.A. (1973). Statistical Mechanics. New York: Harper & Row.
 111

Model Potential Functions

CONTENTS

4.1 Many-Atom Structures, 111

4.2 Atom–Atom Potentials, 114
4.2.1 Standard Models and Their Relations, 114
4.2.2 Combination Rules, 116
4.2.3 Very Short-Range Potentials, 117
4.2.4 Local Parametrization of Potentials, 117
4.3 Atom–Molecule and Molecule–Molecule Potentials, 119
4.3.1 Dependences on Orientation Angles, 119
4.3.2 Potentials as Functionals of Variable Parameters, 124
4.3.3 Hydrogen Bonding, 124
4.3.4 Systems with Additive Anisotropic Pair-Interactions, 125
4.3.5 Bond Rearrangements, 125
4.4 Interactions in Extended (Many-Atom) Systems, 127
4.4.1 Interaction Energies in Crystals, 127
4.4.2 Interaction Energies in Liquids, 131
4.5 Interaction Energies in a Liquid Solution and in Physisorption, 135
4.5.1 Potential Energy of a Solute in a Liquid Solution, 135
4.5.2 Potential Energies of Atoms and Molecules Adsorbed at Solid Surfaces, 139
4.6 Interaction Energies in Large Molecules and in Chemisorption, 143
4.6.1 Interaction Energies Among Molecular Fragments, 143
4.6.2 Potential Energy Surfaces and Force Fields in Large Molecules, 145
4.6.3 Potential Energy Functions of Global Variables Parametrized with Machine
Learning Procedures, 148
References, 152

4.1 Many-Atom Structures

The structure of a many-atom system with Nat atoms is defined by three Carte-
sian coordinates per atom and has 3Nat degrees of freedom. A rigid structure in
vacuum has invariant properties under overall translations and rotations so that a
description in a reference frame with its origin at the center of mass of the nuclei
Molecular Interactions: Concepts and Methods, First Edition. David A. Micha.
© 2020 John Wiley & Sons, Inc. Published 2020 by John Wiley & Sons, Inc.
112 4 Model Potential Functions

and its Cartesian axes fixed in space, a CMN-SF frame, decreases the required
number of degrees of freedom. Overall, translation can eliminate three degrees
of freedom, and in addition three (for a nonlinear system) or two (for a linear
system) variables can also be eliminated with overall rotational invariance so that
the rigid structure requires only Ndf = 3Nat − 6 variables, or Ndf = 3Nat − 5 for a
linear system. They can be chosen as atom–atom distances, of which there are
Ndf (Ndf − 1)/2. For a two atom system, this is the distance R between centers
of mass, and for a nonlinear three atom system, A–B–C one can choose
atom–atom distances RAB, RBC, RCA, or sometimes two bond distances RAB,
RBC and the common bond angle αABC. Generally, for a system with an interac-
tion energy invariant under overall rotations and described by a single potential
energy function (which must go to zero as any atom is removed to infinity) this
potential energy is a V({RIJ}) function which depends on the interatomic distances
between all atom pairs IJ. As distances are varied the energy generates a potential
energy surface (PES) in a space of Ndf (Ndf − 1)/2 degrees of freedom, with peaks
and valleys which can be pictured in three dimensions or as isocontours in two
dimensions, by freezing all but two variables at a time.
In addition to this distances dependence, the potential energy may also
depend on electron spin variables if the system has open electronic shells.
It may also be multivalued for a given atomic structure if more than one PES
is needed to account for electronic rearrangement due to electronic charge
transfer or excitation. This requires consideration of couplings between PESs
and is treated in a following chapter.
The potential can be decomposed in a variety of ways, some with more phys-
ical content than others. A number of criteria can be imposed on the form of
V({RIJ}) to make it useful in the treatment of energetics and dynamics properties.
They are the following:

a) Asymptotic consistency, meaning that as a fragment is removed to infinity,

the total potential energy must become a sum of the potential energies for
the parts.
b) Size consistency so that if the whole system is a collection of N identical
units, the total potential energy must change linearly with a changing num-
ber N of units.
c) Proper spatial and spin symmetry.
d) Physical continuity of potential values and gradients with respect to dis-
tances, as distances change from short ranges to long ranges.

These PESs contain parameters to be fixed by means of calculated or meas-

ured procedures, and this requires:
a) An algebraic form allowing good fits as the structure varies, and practical for
given computational resources.
 4.1 Many-Atom Structures 113

b) Parameters which can be determined with a relatively small number of data

points.
c) Flexible shape so that a fitted PES converges to an accurate one as the
number of data points are increased.
d) Physically meaningful parameters providing insight as structure or
composition are varied.
Model potential functions can be constructed with functional forms suggested
by physical properties and parametrized to reproduce interaction energies known
from measurements or from atomistic calculations. They can be classified in
accordance with the atomic and electronic structure of the involved species.
These can be atoms, molecules, clusters, solid surfaces, polymers, biomolecules,
and complexes formed from them. Their interactions depend on whether they are
in their ground electronic state or electronically excited, and whether they are
stable during interactions or reactive. All these possibilities require a variety of
functional forms, with parameters obtained from model calculations [1–6], or
from experimental measurements. Experimental extraction of potential energy
parameters from spectroscopy, molecular beam scattering, and molecular kinet-
ics have been covered, for example in Refs. [7–9].
For three atoms A, B, C, we can use atomic orbitals χ aλ, χ bμ, χ cν centered at
positions Rn of nuclei n = a, b, c, to construct molecular orbitals φj as linear
combinations of atomic orbitals (or LCAOs). These compose many-electron
determinantal wavefunctions DK from which an optimized configuration
interaction (CI) state Φ = KcKDK is obtained by minimizing the system’s
energy [10]. The total energy E can be calculated from a Hamiltonian given
by a sum over a one-electron and a two-electron operator, as
1 2
E= Φ H +H Φ
The expectation value involves integrals with one-electron operators between
two atomic orbitals or two-electron operators with four atomic orbitals, with up
to three centers for this system. The coefficients cK of the CI expansion, how-
ever, depend on all atom locations. Collecting energy terms dependent on one,
two, and three nuclear positions, one can write

E = EA + EB + EC + EAB + EBC + ECA + EABC

V = E − EA + EB + EC

with one-center EJ Rj , two-center EIJ Ri ,Rj , and three-center energies vary-

ing with nuclear positions. Hence, the potential energy V for three atoms can be
written as a sum of two-center and three-center terms, V = V (2) + V (3), each a
sum of pairs and triplets.
114 4 Model Potential Functions

Similarly for N atoms we have VN = V (2) + V (3) + … + V (N), with terms

dependent on the positions of the N atoms. Here, it may happen that electrons
are localized over few near-neighbors or next-to-near-neighbor centers, in
which case many of the multicenter integrals can be neglected, and the CI coef-
ficients and energies can be organized into sums for interacting molecular frag-
ments. A two-fragment interaction term can be written as a sum of short-range
and long-range interactions, and similarly for three- and four-fragment terms.
For a system in its electronic ground state, a useful decomposition is, labeling
atoms in numerical fashion,

V Rlm = l<m
V 2
Rlm + l<m<n
V 3
Rlm , Rmn ,Rnl + …

from which a general force field can be generated. The two-atom terms can be
parametrized as shown in the next section, while the three atom terms can be
added as corrections which are dampened as interatomic distances increase
beyond some boundary values R0lm , and have adjustable parameters at short
distances. Labeling the pair (lm) = μ = 1, 2, 3, …, and with Rlm = rμ, one such
function is

1
V 3
rμ ,rν ,rπ = μ
1 −tanh γ μ rμ − rμ0 b0 + b
μ μ
rμ − rμ0
2
+ b
μ < ν μν
rμ − rμ0 rν − rν0 + …

where the factor with a tanh goes to zero outside the boundary rμ0 values, and the
coefficients b can be chosen to shape the region within boundaries.
Alternative forms including also four-atom terms can be written in terms of
bond distances, bond angles, and dihedral (torsion) angles, as well as distances
between nonbonding atoms, and are the basis for treatments of the molecular
dynamics of polymers and biomolecules.

4.2 Atom–Atom Potentials

4.2.1 Standard Models and Their Relations
To begin with, we consider the simplest case of two interacting atoms in their
ground electronic states, classified by their electronic shell structures. For atoms
they are (i) two closed shell atoms with a van der Waals interaction; (ii) two
closed shell ions forming an ionic bond; (iii) an open shell ion and one closed
shell ion in a transition complex; and (iv) two open shell atoms forming a chem-
ical bond [11]. Potential energy functions V(R) of the interatomic distance R can
 4.2 Atom–Atom Potentials 115

be given in terms of their values V = 0, Vm, Vi, … at special locations R0, Rm, Ri, …
at the zero-value or core, minimum, and inflection points, as shown in
Figure 4.1, with magnitudes and functional forms dependent on their physical
origin.
We deal here first with pairs where one or both atoms have a closed electronic
shell and are not charged, so that they do not form a chemical bond, but may
form a van der Waals complex with an interaction potential energy VvdW(R).
The potential minimum energy Vm = − ϵ and location Rm are chosen here as
units of energy and length, to introduce the reduced variables x = R/Rm and
f(x) = V(R)/ϵ. Analytical forms usually employed are the Lennard Jones (n,6),
Exp-6(α), also called the Buckingham or Slater potential, and the Morse(a)
potential. They all have three parameters, or only one in reduced variables.
Their form and range of the remaining parameter are

−1
fLJ x; n = n− 6 6x −n −nx −6 , 8 ≤ n ≤ 20
fE6 x; α = 6 exp − α x−1 − αx − 6 , 8 ≤ α ≤ 20
fM x; a = exp −2a x −1 − 2 exp − a x −1 , 4 ≤ a ≤ 8

From these, we can find the curvatures at the minimum to be, respectively,

d2 f α− 7
κ= = 6n, = 6α , = 2a2
dx2 x=1 α− 6

Figure 4.1 Special points R0, Rm, V

Ri, in the potential energy
function V versus the interatomic
distance R between atoms A and
B, for the core radius, minimum,
and inflection distances.

A +B
R0 Rm Ri R

Vm = –ε

AB
116 4 Model Potential Functions

and the location x0 of the zero of the potential,

1 n− 6
6 ln α 6
x0 = , ≈1 − , = 1 − ln 2 a
n α− 6
The location Ri and energy Vi of the inflection point can also be obtained from
the root xi of the second derivative d2f/dx2 = 0. The Exp-6 and Morse potentials
are unphysical at short distances and show an inner inflection point as R
decreases. Therefore, these potential functions must be limited to values of R
larger than an ion core RC, and may better be written as functions of R − RC with
V(RC) kept much larger than relevant energies. Alternatively the Exp-6 potential
can be modified multiplying its short range term times a R−n factor with n > 6.
The Morse potential is frequently used to treat chemical bond interactions and
is parametrized with results from spectroscopy. An alternative is the extended
Rydberg potential
k
fRyd x = − 1 + a
k k
x− 1 exp −a1 x− 1

with 1 ≤ k ≤ 3, which gives a potential energy VRyd = − ϵ at the minimum Rm and

a value at the origin of VRyd(0) = ϵ[ kak(−1)k + 1 − 1] exp(a1) > 0.
Other potential functions apply to interactions of atomic and molecular ions.
For an interaction between a neutral species and an ionic one, it is necessary to
add a charge-induced dipole term going as R−4. The interaction of two closed
shell charged ions such as in Na+Cl− can be described with the potential,
Vion(R) = A exp(−αR) − Z+Z−e2R−1, with e = ce/(4πε0)1/2 in SI units, and para-
meters extracted from rotational and vibrational spectroscopy. More accurately,
the potential function must include the van der Waals dispersion term going as
R−6, the charge-induced dipole term going as R−4, and the interaction between
each distorted closed shell dipole with the other ion going as R−7, as can be also
found from third-order perturbation theory.

4.2.2 Combination Rules

These rules allow us to obtain parameters for a heteronuclear diatomic pair AB
from the homonuclear pairs AA and BB. For the long range −CABR−6 dispersion
energy coefficients, we have [12]

2CAA CBB
CAB =
αB αA CAA + αA αB CBB

where αA is the static polarizability of atom A. For the short range we can use the
Born–Mayer potential

VAA R = AAA exp −bAA R

 4.2 Atom–Atom Potentials 117

and the combination rule

1 2
VAB R = VAA R VBB R

giving AAB = (AAAABB)1/2 and bAB = (bAA + bBB)/2. The parameter bAA derives
from the electronic charge distribution of the outer shell and the size of the
atomic core of inner shells; if the former is given (in a hydrogenic description)
by the effective nuclear charge ζ A, it follows that bAA = b0AA + c ζA , with the first
term given by the A core radius.

4.2.3 Very Short-Range Potentials

These are determined by the distortion of overlapping electron distributions,
which happens due to the “Pauli exclusion” repulsion arising from the require-
ment that electronic wavefunctions must be antisymmetric. It leads to distor-
tion multipoles which oppose each other and therefore lead to repulsion;
their R-dependence relates to the overlap Δ(A) of charge densities, which intro-
A
duces a factor Δdst R = Δ A RAc exp −λA R− RAc for distances larger than a
core value RAc in the interaction function. If the highest distortion multipole at
each species is an n-pole, then we have an interaction for AA of the form
−N
R
VAA R = VAA RAc exp −2λA R− RAc
RAc

with N = 2n + 1. More elaborate models for very short distances have also been
introduced [13]. Interactions VAB(R) for heteronuclear systems follow from
combination rules there.

4.2.4 Local Parametrization of Potentials

Local parametrizations can be done turning on and off potentials in certain
regions, or constructing them in piecewise fashion. Local parametrizations
are useful when semiempirical information is introduced for certain ranges
of distances.
Short-range damping functions gs(R) are constructed to change around a
damping location Rd and go from 1.0 at short distances to zero at infinite values
of R, within a range Rd/λ. For example,
−1
Rd − R
gs R; Rd ,λ = 1 − exp − λ 1 − exp −λ
Rd

can be used to select the short-range potential using Rd = R0 and to dampen

the long range past that point, while its complement g s = 1 −gs dampens the
118 4 Model Potential Functions

potential inside its inflection point using Rd = Ri. Choosing Rd = R0, and λ R0
the decomposition
S L
V R = gs R V R + g s R V R = V R +V R

separates short-range and long-range components, with the second term usu-
ally small at all distances and suitable for perturbative treatments of thermody-
namical and transport properties. This separation is shown in Figure 4.2.
Another use for these functions is in the switching from ionic to covalent
potential energy functions as distance increases, with a crossing at a location
Rd = Rx for example in alkali–halide interactions, where one can use
V R = Vion R gs R + Vcov R g s R
The damping functions, however, have large derivatives around Rd that can
affect the calculation of forces in studies of molecular dynamics, and their
change with R must be locally balanced in the transition region before deriva-
tives with respect to position variables are taken, to avoid artificial forces. Fur-
thermore, if the damping function is combined with a long-range interaction
going as R−n, then care must be taken to compensate for, or exclude, the diver-
gence of the interaction as R 0. This can be done, for example using instead

V Figure 4.2 Separation of a short-range

(or SR) potential energy component
from intermediate and long-range (or
VSR LR) components, using a switching (or
long-range damping) function gs(R)
and its complement gs R .

A+B
R0 R

VLR

AB

gS

1.0

R
0.0
 4.3 Atom–Molecule and Molecule–Molecule Potentials 119

(R + δR)−n with a small δR parameter, or multiplying the long-range interaction

n
times a new damping function fd R = g s R given by [14]

n n
fd R; an , bn = 1 − exp −an R− bn R2

which leads to a constant value at the origin for the potential energy containing
R−n, and has two parameters as needed to fix the transition location and its
width. An alternative proposed form contains only the Born–Mayer parameter
b and can be used if the origin is excluded [15].
Piecewise potentials can be put together from exponentials, Morse, and
van der Waals potentials at short, intermediate, and long-range distances,
respectively, and joining them with spline functions chosen to assure a smooth
function and smooth derivative; this is the so-called Exponential–Spline–
Morse–Spline–van der Waals (ESMSV) potential [16, 17]. The spline function
between points xj and xj + 1 is

k
3 x −xj
S x = a
k =0 k xj + 1 −xj

introduced with its bounds chosen to bracket the zero and inflection points and
can be used to construct very accurate potential functions and forces over a wide
range of distances. The four parameters are fixed by imposing continuity of the
function and its derivative at the two ends of its range. The spline can also be
used to smoothly join avoided crossing potentials, for example from an outer
covalent potential to an inner ionic one as in NaCl.

4.3 Atom–Molecule and Molecule–Molecule Potentials

4.3.1 Dependences on Orientation Angles
For an atom A interacting with a diatomic M = BC in a CMN-BF frame with its
z-axis along the two centers of mass, let R be the position of A with respect to the
CMN of BC, r the position vector from B to C, and θ the angle between the two
directions, as shown in Figure 4.3, also for given interatomic distances Rac and
Rab and a chosen angle at A. For a nondegenerate state, the potential energy is a
unique function of only the three variables R, r, θ (after eliminating the rigid
body three translational and three orientational degrees of freedom, a total of
six variables).
The angular dependence can be treated differently depending on whether the
interaction anisotropy is weak or strong. In the first case, it is expanded in
Legendre polynomials Pl(cosθ), and the coefficient functions for each term
120 4 Model Potential Functions

(a) (b)
B B
Rab

R
θ r
A A
Rac

C C

Figure 4.3 (a) Center-of-mass distance variables in the triatomic A + BC. (b) Variables for a
description in terms of atom-pair potential energies, for a chosen distance between B and C.

are parametrized as given functions of R, and r, with parameters expanded

around the equilibrium bond value of r,

V R,r, θ = l
Vl R, r Pl cosθ

and the expansion functions can be further decomposed into short-range and
long-range terms as
S L
Vl R,r = Vl R,r + Vl R,r
S
Vl R,r = Al r exp − bl r R
R R−n
L n
Vl R, r = vBC r δl0 + n Cl , n r fd
Only a few Legendre polynomials are needed for weak anisotropy and, if M is a
homonuclear BB molecule, symmetry eliminates all odd polynomials from the
expansion. The A, b, and C parameters can be expanded in powers of the dis-
placement ρ = r − re from the equilibrium bond length re. For each angle, it is
possible to show potential energies as isocontours of V versus R and r. Isocon-
tours of potential energy surfaces describing the atomic interactions are shown
in Figure 4.4. The top panel is the same as Figure 1.5 in Chapter 1: Part (a),
including also the internal diatomic potential, vBC (r) which must be added
to the long-range interaction for l = 0 (Figure 4.4). Part (b) shows isocontours
for a reaction A + BC AB + C with an activation barrier, such as H + H2
H2 + H, while Part (c) shows isocontours for a stable intermediate ABC such
as H2O with a fixed bond angle, and varying atom–atom distances Rab and
Rbc. The bottom panel is the same as Figure 1.6, and shows the potential energy
of H approaching CN in HCN, in a (x,y) plane with the coordinate origin at the
midpoint between C and N.
An alternative parametrization avoids the expansion in polynomials and
instead introduces parameters dependent on r and θ, for each value of the latter.
 4.3 Atom–Molecule and Molecule–Molecule Potentials 121

(a) (b) (c)

r BC

r A,BC r AB r AB

3.0

2.0
yH (Å)

S
1.0

C N
A B
0 • •
–3.0 –2.0 –1.0 0.0 1.0 2.0 3
xH (Å)

Figure 4.4 Isocontours for three-atom systems. Top panel: generic isocontours for (a) an inert
gas A interacting with a diatomic BC, (b) the reaction A + BC AB + C with an activation
barrier, and (c) a stable intermediate ABC with a fixed bond angle, from [18]. Bottom panel:
potential energy of H approaching CN in HCN, in a (x,y) plane with the coordinate origin at the
midpoint between C and N, from [19]. The locations of minima are shown as A and B, for the
two isomers, and a saddle point is shown as S.

For example the three atoms may be on a line or in a “T” conformation, with
different parameters. Then we have
S L
V R,r,θ = V R,r,θ + V R,r,θ
V S
R,r, θ = A r,θ exp − b r,θ R
R R −n
n
V L
R,r,θ = vBC r + n Cn r, θ fd
This may be described as a method of variable parameters, where the atom–
atom functional forms are used, but the parameters are allowed to change with
internal coordinates and orientation of the molecules.
122 4 Model Potential Functions

The case of a linear system with θ = 0 is frequently described with a model in

which the repulsion depends only on the distance from A to the nearest (B)
atom, so that V (S)(R, r) = A exp[−b(R − γr)], γ = mC/(mB + mC), in terms of
atomic masses. More generally, parameters can be expanded in Legendre poly-
nomials Pl(cosθ) to account for angle variations. For example results of exten-
sive electronic configuration interaction calculations for He + H2 have been
used to parametrize its interaction as

V S
R,r,θ = C exp − α0 R + α0 ρR A θ + B θ ρ

V L
R,r,θ = −R − 6 C6 θ fd R

with ρ = r − re the bond length displacement from equilibrium, the expansions

A(θ) = A(0) + A(2)P2(cosθ), B(θ) = B(0) + B(2)P2(cosθ), and C6(θ) = C(0) +
C(2)P2(cosθ), and fd(R) damping the long range potential [20].
For diatom–diatom interactions and interactions involving polyatomic mole-
cules we must specify the orientation of each species in a common reference
frame, and either expand the interaction energy in a basis set of Wigner rota-
tional functions [21], or let parameters depend on orientation angles. The inter-
action potential for a molecule A with internal degrees of freedom QA, for
example normal mode coordinates, and orientation given by the local
(CMN)A Euler angles ΓA = (α, β, γ)A interacting with molecule B in the joint
CMN-BF frame with the z-axis along R, is a function V (AB)(R, QA, ΓA, QB,
ΓB). This function can be expanded in a basis set of rotational functions of angles
or it can be considered as a functional of parameters which depend on orien-
tation angles.
Alternatively, the molecule–molecule interaction in a CMN-BF frame can be
written as a function of the locations Ra of distributed charges ca for molecule A
and similarly for B,
SR ca cb
VAB = VAB R + a, b 4πε R
0 ab

with the first term giving the short-range repulsive interaction between the
CMN of A and B, and with the summation containing locations and values
of charges chosen to reproduce molecular charge densities and their interaction
and suitable only for large distances Rab. This information can be extracted from
an electronic structure calculation of the AB pair, which would provide a charge
and bond order matrix from which charges and locations can be defined [22].
The set of charges at each molecule can alternatively be replaced by a set of
distributed multipoles, which then give the total potential energy in terms of
multipole–multipole interactions [5]. Figure 4.5 shows the TIP5P model for
water [23], with four charges a = 1–4 representing hydrogens with positive point
 4.3 Atom–Molecule and Molecule–Molecule Potentials 123

0.70 Å

109.47° 0.9572 Å
+

– 104.52°

Figure 4.5 The H2O monomer TIP5P charge model used for H2O–H2O interactions.

charges of 0.241e and lone pairs with balancing negative values, in addition to
the oxygen atom which interacts with other oxygen atoms only through a pair
potential.
The water–water interaction can then be constructed from an LJ(12,6) poten-
tial energy function between the two oxygen atoms with parameters σ 0 = 3 12 A
for its core radius and εOO = 0.16 kcal mol−1 for its well depth, plus point charge
interactions as shown in Figure 4.5 for the hydrogen atoms and lone pairs, and is
of the form

12 6
σ0 σ0 ca cb
VWW = 4εOO − +
ROO ROO a, b
4πε0 Rab

which leads to a physically acceptable description of the local tetrahedral struc-

ture of liquid water, and other properties. This function is suitable only for large
distances Rab but can be modified introducing damping factors for the Coulomb
a,b interactions, or adding bond dipoles and quadrupoles, to improve
properties.
124 4 Model Potential Functions

4.3.2 Potentials as Functionals of Variable Parameters

For an atom-diatomic A–D system, we can start from the atom–atom V(R)
potential expression with parameters ε, Rm, γ and make these dependent on
internal coordinates and relative orientation,

VAD R,r,θ = V R; ε r,θ ,Rm r,θ ,γ r,θ

with the expansion

ε r, θ = ε
l l
r Pl cos θ

and similarly for the other parameters for small anisotropies. For example with
the Exp-6(α) parametrization, the short-range exponential parameter α can be
made to vary with the BC distance and its orientation relative to A, and can be
expanded in Legendre polynomials if the anisotropy is small. For interacting
molecules A and B, this generalizes to VAB(R,QA,ΓA,QB,ΓB) = V[R; ε(QA,ΓA,
QB,ΓB), Rm(QA,ΓA,QB,ΓB), γ(QA,ΓA,QB,ΓB)].

4.3.3 Hydrogen Bonding

Attraction between two species A and B mediated by a hydrogen atom plays an
important role in aqueous solutions and in binding of nucleic acids such as
DNA and RNA. A simple description involves covalent bonding to one of
the species (like A) and noncovalent bonding to the other (as B), in a A─H B
structure. Given the electronic density distribution of this system, it is possible
to construct models of its energetics based on distributed multipoles and polar-
izabilities, such as the interaction dipoles of the A─H bond and of the lone elec-
tron pair in :B, plus repulsion energies at short distances [5]. More accurately, a
many-electron treatment mixes its electronic states in a quantum superposition
of the states of A─H B and A H─B. The potential for H moving between A
and B may show a double well or a wide shallow well, and depends on the bond
distances RAH, RHB and the bond angle αAHB. For H covalently bonded to A, its
interaction with B in A─H B can be parametrized as the function of the dis-
tance between H and the nonbonding B

m 12 m 10
RHB RHB
VAHB = εHB 5 −6 fang cosαAHB
RHB RHB

The overall attraction between organic molecules A and B mediated by H is of

the order 10–25 kJ mol−1.
 4.3 Atom–Molecule and Molecule–Molecule Potentials 125

4.3.4 Systems with Additive Anisotropic Pair-Interactions

It is sometimes accurate to use sums of atom-pair potentials, cutting off the
long-range potentials at short distances to avoid over-binding there. For A +
BC, we add model pair potentials (A,I) over each atom I = B,C interacting with
A, but let the potential depend on the orientation ωi of the BC bond relative to
the direction of AI, a function of θ in Figure 4.3 (a),

V Ra ,Rb ,Rc = v
i = b, c AI
Rai ,ωi
S L
vAI Rai , ωi = vAI Rai ,ωi + vAI Rai , ωi fd, i Rai

where Rai = Ra −Ri , and the long-range term has been dampened at short
interatomic distances.

4.3.5 Bond Rearrangements

In situations where bonds can be broken or formed, it is possible to obtain PESs
from a Valence-Bond (VB) treatment starting from many-electron functions
constructed with one orbital per active electron in the rearranging bonds.
For a diatomic AB, such as H2, with two active electrons in atomic orbitals
χ a(1) and χ b(2) centered at the A and B nuclei, and a diatomic Hamiltonian writ-
ten as H AB = H A + H B + H AB , the interaction potential energy for singlet and tri-
plet states 1,3Σ can be written as [22]
QAB ± AAB
VAB 1, 3 Σ =
1 ± ΔAB
QAB = d 1 d 2 χ a 1 ∗ χ b 2 ∗ H AB χ a 1 χ b 2

AAB = d 1 d 2 χ a 2 ∗ χ b 1 ∗ H AB χ a 1 χ b 2

where the + sign corresponds to the singlet, and the VB Coulomb integral QAB
and exchange integral AAB are shown in terms of the atomic orbitals while
ΔAB = Sab2
is the square of the integral of the orbitals overlap. Reverse relations
give expressions for the integrals in terms of the singlet and triplet energies,
which are known from accurate calculations or from spectroscopic measure-
ments. The singlet and triplet energy functions can be parametrized as Morse
fM(x; a) = exp[−2a(x − 1)] − 2 exp[−a(x − 1)] and anti-Morse fAM(x; b) = exp[−2b
(x − 1)] + 2 exp[−b(x − 1)] potential functions respectively, with adjustable
parameters. The overlap integral follows from the form of atomic orbitals,
1 2
which for 1s orbitals of the form χ a r = ζ 3a πa30 exp − ζa r −Ra a0 gives
2
Sab(Rab) = [1 + (ηRab/a0) + ((ηRab/a0) /3)] exp(−ηRab/a0) and can be used as an
adjustable function of the parameter η for each atom pair, to shape the PES.
126 4 Model Potential Functions

For the three-atom system ABC with three active electrons described with
three atomic orbitals, a London-like interaction potential energy function suit-
able for treatments of reactive atom-diatom systems undergoing a rearrange-
ment such as A + BC AB + C can be constructed from a generalized VB
description [22, 24]. Two three-electron VB wavefunctions 2ΦI and 2ΦII can
be formed to describe the doublet states corresponding to two independent
bondings A─B C and A B─C. The lowest PES is obtained from the Ham-
iltonian eigenvalues in this two-state description and can be given a simple form
provided orbital overlaps are neglected. It is however made more flexible, so it
can be applied to a variety of triatomics, by reintroducing overlap terms para-
metrized to reproduce desired features such as the location of a surface saddle
point at an energy barrier for reaction, in which case it is the so-called London–
Eyring–Polanyi–Sato (LEPS) function [3, 25, 26]
−1
VABC = 1 + ΔABC QAB + QBC + QCA − 2 − 1 2
AAB − ABC 2

2 2 1 2
+ ABC −ACA + ACA − AAB
and the integrals can be obtained from known singlet and triplet potential ener-
gies for each pair, while the triatomic overlap ΔABC can be used as an adjustable
parameter to obtain a desired activation barrier in a rearrangement A + BC
AB + C. In the Polanyi version [27], the term ΔABC is omitted and replaced with
three distance dependent parameters ΔIJ = SIJ2 introduced in the expressions for
the singlet and triplet energies of each IJ pair, from which the VB Coulomb and
exchange integrals have been obtained, to allow for desired energy barrier and
added attraction or repulsion between reactants or products. This gives
VIJ + 3 VIJ + ΔIJ 1 VIJ − 3 VIJ
1
QIJ =
2
in terms of singlet and triplet potential energy functions of R and an overlap inte-
gral, and with a similar expression for AIJ replacing 3VIJ with −3VIJ. A generic iso-
contour for a reaction with an activation barrier is found in the top Figure 4.4a,
while Figure 4.6 shows the case of the collinear F+ H2 FH + H reaction with an
activation barrier in the entrance channel. An analogous PES expression can be
written for systems with four active electrons [22, 26] such as H2 + D2.
Spin-dependent pair potentials may instead be used to write the interaction
of open shell atoms in terms of spin operators, which can be parametrized
using pair interaction functions for specific multiplet states. These
potentials can be derived from electronic structure states in a VB formulation
[30]. For three monovalent atoms the pair potential operators are V IK RIJ =
c s
VIJ RIJ + VIJ RIJ S I S J where S I is the spin operator of atom I with two spin
states. For three monovalent atoms, the spin operator V AB + V BC + V CA can be
diagonalized in a basis set of two doublet three-atom spin states to recover the
LEPS expression.
 4.4 Interactions in Extended (Many-Atom) Systems 127

2
–35

–30
RH–H(Å)

–25

–20
1 –15
–10 +5
–5 +1.6
0 +1.6

1 2
RH–F(Å)

Figure 4.6 Energy isocontours for collinear FH2 versus atom–atom distances, showing an
early activation barrier as F approaches H2. Source: from Bender et al. [28], almost identical to
the figure from Bender et al. [29].

4.4 Interactions in Extended (Many-Atom) Systems

4.4.1 Interaction Energies in Crystals
We consider to begin with atoms or molecules forming a lattice with nuclear
positions at rest, and surrounded by electrons, and a solid with a number den-
pot
sity ρsld. The lattice energy is then a potential energy Usld ρsld for the system
at a temperature T = 0 K. As the temperature is increased, one must add to this
the kinetic energy of nuclear motion and the excitation energy of thermal elec-
kin
trons, Usld ρsld ,β to obtain the total thermodynamical internal energy
Usld(ρsld, β), with β = 1/(kBT) the inverse temperature in units of the Boltzmann
constant kB, that gives the cohesive energy at the given density and
temperature.
In crystals made up of inert gas atoms, such as Ne(s), or of ions, such as
NaCl(s), a good description of their crystal structure and cohesive energy can
be extracted from models containing only the atom–atom or ion–ion
interactions.
Crystals with covalent bonding, such as Si(s), must be treated as many-
electron systems with electronic densities localized at ion cores and along
chemical bonds.
128 4 Model Potential Functions

In metals, such as Cu(s), some electrons are localized into ion cores and others
are delocalized over the whole crystal with the exclusion of the ion cores, and
treatments can be based on the concept of embedded atoms [31]. This is also
helpful in the treatment of unsaturated hydrocarbon chains with delocalized
electrons.
The cohesive energy of a molecular crystal with Nat atoms at rest in a sample
volume Ω, such as crystals composed of identical noble gas atoms, can be writ-
ten as a sum of pair potentials v(Rij) between atoms i and j in a lattice, with suit-
able parameters which must depend on the atomic density ρsld = Nat/Ω. The
solid potential energy is then approximated by

1
Vsld = v Rij
2 i j

with pair potential parameters vm(ρsld) for the well depth, and l(ρsld) for the near-
est neighbors distance, at the minimum of energy between them, as functions of
the solid density. It can be conveniently re-expressed in terms of a relative atom–
atom distance pij = Rij/l, so that the lattice sum for a long range R−n interaction
would be (1/2) i j(Rij)−n = l−nαn, with a lattice sum αn = (1/2) i j(pij)−n
characteristic of the lattice symmetry and independent of density. A starting
point for calculation of the cohesive energy is a sum of pair potentials for the
isolated pair, with ρsld = 0.
For the LJ(12,6) pair potential written in terms of its energy minimum vm|
and the location R0 of its zero energy value, and with a close packed (fcc) lattice
structure at equilibrium, the potential energy for identical noble gas atoms is
12 6
1 R0 R0
Vlatt = Nat 4 vm α12 fcc − α6 fcc
2 l l

The lattice summations have been evaluated and are α12(fcc) = 12.13188 and
α6(fcc) = 14.45392, not far from the number 12 of nearest neighbors. The equi-
librium distance leq is obtained setting the force dVlatt/dl = 0 to get leq/R0 = 1.09,
indicating that the internuclear distance has increased going from an isolated to
an embedded atom pair. The measured values for Ne, Ar, Kr, and Xe are 1.14,
1.11, 1.10, and 1.09, in quite good agreement. The remaining discrepancy can be
corrected by including in the cohesive energy the zero-point energy of vibra-
tional motion of each atom as it oscillates around its equilibrium position [31].
Ionic crystals, such as NaCl(s) with the fcc lattice symmetry, contain short-
range repulsions and long-range Coulomb electrostatic attractions. They are
treated introducing a diatomic ion, like Na+Cl−, in a unit cell of a lattice, and
adding over atom pair interactions. The solid cohesive energy is
Z + Z − c2e
Vlatt = Ndiat v SR l − α1 latt
4πε0 εr l
 4.4 Interactions in Extended (Many-Atom) Systems 129

where l is the distance between adjacent positive and negative ions, εr is the
dielectric constant of the medium, the charge numbers Z are positive, and
α1 = αM is the Madelung lattice constant. It is obtained from two interpene-
trating fcc lattices for Na+ and Cl−. Taking one of the negative ions as a refer-
ence center in the solid, one can write the constant as a sum of terms for its
interaction with positive and negative ions, and the sum can be done partition-
ing the lattice around the reference ion into nearly neutral groups of ionic
charges in cubes of increasing size until convergence. The result for the fcc
lattice is 1.747558. For comparison, the value for the bcc lattice, such as in
the CsCl(s) crystal, is 1.762670 [31]. The short-range term can be written
as a Born–Mayer exponential repulsion A exp(−α l) and the equilibrium value
leq of the interionic distance can be found from the location of the zero-value
force. Replacing this into the potential energy one finds the Born–Mayer equi-
librium form

Z + Z − c2e 1
Vlatt = −Ndiat 1− αM latt
4πε0 εr leq αleq

This shows that cohesion increases for more highly charged ions and for more
steeply repulsive short-range interactions.
More accurately, the cohesive energy must also contain three- and many-
atom terms such as l < m < nV(LR)(Rlm, Rmn, Rnl ), to account for nonadditivity
of long-range interactions. The many-atom effects can be included in a variety
of ways which depend on the nature of the structure. For solid crystalline Si, a
useful functional form has been given by Tersoff [32], which incorporates a
many-atom factor in the long-range interaction. The total energy can be
written as

V Rab = f
l<m c
Rlm Aexp −λ1 Rlm − B Rab exp −λ2 Rlm

The first term in the square parenthesis describes the repulsion energy
between atoms l and m, and the second term gives a modified attraction energy
where the function B({Rab}) depends on all the atom–atom distances needed to
describe the environment of the (lm) pair. This attraction factor is constructed
as a monotonically decreasing function of the number of bonds of l and m with
other atoms. It also depends on the strength of the bonds to those two atoms
competing with the (lm) bond, and on the angles between the (lm) bond and the
other bonds. The cut-off function fc(R) is chosen to equal 1.0 at short distances
and to smoothly fall to 0.0 at large distances. Results for various phases of solid
structures can be fitted with suitable parameters this way.
The zero-point vibrational motion energy of the lattice atoms must be added
even at the temperature of 0 K. If the temperature is increased, then parameters
like the atom–atom distance in the expressions for Vlatt depend on the density
130 4 Model Potential Functions

pot
ρsld and temperature in β, and become thermal potential energies Ulatt to
which one must add the thermal kinetic energy of vibrating atoms in a lattice,
to obtain the internal energy Ulatt, or cohesive energy, of the solid.
The solid dielectric properties alter the strength of the electrostatic and polar-
ization interactions, decreasing them by the magnitude of the dielectric con-
stant εr of the medium and affecting the values of the Vlatt potential energy.
A general treatment of dielectric effects can be based on the dynamical suscep-
tibility of the solid, by considering the response of the whole system to an
applied external electric field ext [31]. The local electric field lcl = εr ε0 ext at
the position of an atom in the solid can be constructed adding fields coming
from the surfaces of the solid with an inner idealized cavity, plus the field of
the atoms inside it, as

lcl = ext + out + inn + atm

where out is the depolarization field from charges at the outer surface of the
solid, inn is the field from the charges at the inner surface of the cavity, and
atm is created by all the polarized atoms in the cavity. The sum
ext + out = appl , also called the Maxwell field, depends on the shape of the outer
surface and gives the total applied field. The field from the inner surface charges
is the so-called Lorentz field which for a spherical surface can be obtained from
the polarization vector P = j ρjd j = j ρj αj lcl of the sample, where ρj is the
number of atoms of type j per unit volume, with atomic induced dipole d j
and polarizability αj, and equals inn = P 3ε0 [31]. The last term atm depends
on the distribution of atoms in the cavity and can be seen by symmetry that it
equals zero for a cubic lattice. This gives lcl = appl +P 3ε0 in SI units, with
the dielectric susceptibility χ = εr − 1 introduced by P = ε0 χ appl . This assumes
an isotropic medium or a cubic crystal structure. More accurately, one must use
a susceptibility tensor χ ξη and extend the treatment using Cartesian compo-
nents in Pξ = ε0 η χ ξη appl, η . Considering a single component and isotropy, with
P= j ρj αj appl + P 3ε0 , and solving for χ = P ε0 appl one obtains the
relation [31]

εr − 1 1
= ρα
j j j
εr + 2 3ε0

which provides the dielectric constant needed in the solid potential energy Vlatt.
 4.4 Interactions in Extended (Many-Atom) Systems 131

4.4.2 Interaction Energies in Liquids

The structure of a fluid made up of atoms or molecules can be described in
terms of a probability distribution of its component particles, which depends
on the average density ρ and temperature in β. To simplify, we first consider
a liquid of N atoms and a probability P N R1 ,…,RN d 3 R1 …d 3 RN for finding
each atom j (=1 to N) in volume element d3Rj. This allows us to define a reduced
density for a set of n atoms irrespective of their label, as [33]

ρ n R1 ,…,RN = N N − n P n R1 ,…,RN

where the factor to the right comes from integration of the whole probability
over the position variables of N–n particles. In particular, when n = 1 we have
−1
for liquid volume Ωliq that Ωliq d 3 R1 ρ 1 R1 = N Ωliq = ρ, while ρ 2 R1 ,R2
describes the distribution of atom pairs and ρ(3) gives the triplets distribution.
Liquid structure correlation functions g n
R1 , …,RN are defined by ρ(n) = ρng(n).
In an isotropic liquid, these functions are invariant under reference frame rota-
tions and depend only on relative distances, so that, for example the relative dis-
tribution of atoms at a distance R = R1 −R2 | around a fixed one at position
R1 = 0 averaged over orientations is ρ 2 R ρ = ρg 2 R , which leads to
g(2)(∞) ≈ 1.
Interaction energies in liquids and in liquid solutions can be obtained from
thermal averages and are constructed from statistical pair distribution functions
g 2 = g R; ρ,β , dependent on the liquid number density ρ and an inverse tem-
perature β = 1/(kBT), with ρg(R; ρ, β)R2dR counting the number of atoms found
in a spherical shell of width dR. The simplest dependence, only on R, is for
closed shell atoms, and this must be generalized to include variables for orien-
tation angles of molecules, or for anisotropies of open shell atoms in covalent
bonding. The function g(R; ρ, β) can be measured with X-ray diffraction meth-
ods, or calculated using molecular dynamics.
For a closed thermodynamical system described by a statistical canonical
ensemble with N 1 atoms in a volume Ω, and density ρ = N/Ω, with up to
n atoms interacting with a potential energy Vn R1 ,…,RN , the n-ple distribution
function is given by [33]
−1
g n
R1 ,…,RN ; ρ,β = Ω N ZN Ω, β d 3 RN + 1 …d 3 RN exp −βVn R1 , …,RN

where ZN = d 3N R exp −βVn R1 ,…,RN is the configuration integral.

132 4 Model Potential Functions

The pair distribution g(R; ρ, β) is small for small R where repulsion takes place,
followed by a large peak around the distance Rm for the pair potential energy
minimum, and shows subsequent minima and maxima corresponding to
atomic shells, as distance increases [33]. A pair distribution function is shown
in Figure 4.7 obtained from a molecular dynamics simulation with the Lennard–
Jones pair potential function with parameters ε and R0, as a function of the
reduced distance R/R0 and for the reduced temperatures T ∗ = kBT/ε and den-
sities ρ∗ = R03ρ, with R0 the core radius.
The potential energy of interaction in the simplest case, when the density is
low and many-atom effects can be ignored, is given as a sum of pair v(R) poten-
tial energies as
∞
pot 1 3 N
Uliq ρ,β = d R1 d3 R2 v R12 ρ 2 R1 ,R2 = ρ dR 4πR2 v R g R; ρ,β
2 2 0
kin
To this one must add the thermal kinetic energy Uliq ρ,β = 3NkB T 2 =
3N 2 β to obtain the thermodynamical internal energy
∞
3N N
Uliq ρ,β = + ρ dR 4πR2 v R g R; ρ,β
2β 2 0

or cohesive energy, at the given density and temperature. Different choices for
the atom pair potential have been used in the calculations of correlation
functions.
pot
A perturbation theory for calculating Uliq ρ,β can be based on the assump-
tion of additive pair potentials and separation of short-range and long-range
S L
contributions in v(R) = v(S)(R) + v(L)(R), so that VN = VN + VN , with the

3.0 Figure 4.7 Radial pair

distribution function g(R; ρ, β)
for an atomic fluid with a
2.5
Lennard–Jones potential at
reduced temperatures
2.0 T ∗ = kBT/ε and densities
ρ∗ = R30 ρ, with R0 the core
g(R)

1.5 radius, as shown. Calculated

for T ∗ = 0.71 and ρ∗ = 0.844.
Source: from Wikipedia, 2009,
1.0 public domain.

0.5

0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
R/R0
 4.4 Interactions in Extended (Many-Atom) Systems 133

0
short-range term used to construct a reference potential energy Uliq ρ,β =
S
Uliq ρ,β , and the long-range part treated as a perturbation to obtain a correc-
L
tion Uliq ρ,β = Uliq ρ,β . The short-range summations over pairs can be usu-
ally limited to the average number zNN of nearest neighbors, while the long-
range part is treated as a small perturbation [33]. Restricting the nearest neigh-
bors to a narrow distance around the maximum of the pair distribution func-
tion, which coincides with the minimum of the pair potential, provides a
S
rough estimate of the short-range potential energy as Uliq ρ, β = Nv Rm zNN 2.
Contributions from the LR interactions can be described by means of the
Fourier components of the new function h R = g R − 1, given by the structure
factor

−3 2
h k = ρ 2π d 3 R exp ik R g R −1

from which the pair distribution function is obtained by the inverse Fourier
transform, as

−3 2
g R = 1 + 2π d 3 k exp −ik R h k ρ

L
Replacing this in the integrals over R for Uliq ρ,β we find the long-range
potential energy of the liquid solution expressed in terms of the Fourier trans-
L
form v k of the long-range part of the pair interaction potentials, and of the
structure factor, as
∞
L N
Uliq = d3 k v L
− k h k −ρ dR 4πR2 v L R
2 0

with

−3 2
v L
k = 2π d 3 R exp ik R v L R

For the present potential and distribution functions dependent on only the
radial variable, the transforms are
∞
L −3 2 −1
v k = 2π k dR 4πR sin kR v L R
0
∞
− 3 2 −1
h k = ρ 2π k dR 4πR sin kR g R −1
0
134 4 Model Potential Functions

involving long-range electrostatic and polarization energy terms, cut-off at

short distances, and dependent only on the magnitude k. The structure factor
can be measured by means of X-ray diffraction techniques, in which case it
would be available for calculation of the LR internal energy. It can also be gen-
erated from an expansion in powers of the density ρ to generate an equation of
state for real gases and liquids. Alternatively, a perturbation theory can be devel-
oped for liquids, using the SR internal energy as a reference to zeroth order, and
the LR interaction potential as a perturbation energy. The van der Waals equa-
tion of state for the pressure readily emerges from this treatment when the pair
distribution function is approximated by a hard sphere expression [33].
The function g(R; ρ, β) can also be used to introduce the concept of the potential
of mean force w(2)(R; ρ, β) = − β−1 ln[ g(R; ρ, β)] which goes to zero at infinite
distances and is shaped by the direct pair interaction v(R), but in addition
accounts for changes of the interaction of an atom at the origin with another
at distance R, due to their indirect interactions through surrounding medium par-
ticles. The shape of the potential of mean force as R changes follows from its def-
inition. It shows in particular a minimum at the location of the first maximum of
the radial distribution, followed by decreasing oscillations as R increases through
the medium. It can be obtained at low densities for real gases from the expansion
in powers of the density, g(R; ρ, β) = g0(R; β) + ρg1(R; β) + … and the Boltzmann
thermal distribution for a dilute system which gives g0(R; β) = exp[−βv(R)] [33].
2
More accurate treatments for g R1 ,R2 are derived from the triplets distri-
3
bution g R1 ,R2 ,R3 . To cut a many-molecule chain of equations, it is some-
times accurate to assume that the potential of mean force for triplets satisfies
w 3 R1 ,R2 ,R3 = w 2 R1 ,R2 + w 2 R2 ,R3 + w 2 R1 ,R3 , a sum of pair inter-
actions, even if this is not the case for the original triplet potentials. In this case
3 2 2 2
g R1 ,R2 ,R3 = g R1 ,R2 g R2 ,R3 g R1 ,R3 and all thermodynamical
properties can be described in terms of pair distribution functions and pair
potentials of mean force.
Generalizations are possible for a liquid of anisotropic molecules, and for liq-
uid solutions. The pair distribution function of two polyatomic molecules A and
B can be described in a space fixed (SF) frame in terms of their relative position
RAB and their Euler angles of orientation, ΓJ = (α, β, γ)J, J = A, B. It is frequently of
interest to select certain orientations with respect to the intermolecular posi-
tion, and to average over the remaining angles of the molecular orientations.
For two identical linear molecules A1 and A2 with orientation angles
Ωj = (ϑ, φ)j, j = 1, 2, the averaged pair distribution function can be expanded
in spherical harmonics as

g R12 ,Ω1 ,Ω2 = 4π g

l, l , m l, l , m
R12 Yl, m Ω1 Yl, −m Ω2
 4.5 Interaction Energies in a Liquid Solution and in Physisorption 135

and the coefficients can be related to partial potentials of mean force for mole-
cules with fixed orientation. Expansions in terms of the Wigner rotational func-
l
tions Dmm Γj [21] can instead be done for nonlinear molecules.
An alternative for interacting polyatomics is to represent each one with a set
of potential centers a and b, possibly anisotropic, for A and B, respectively, and
to introduce center-pair distribution functions so that, for example for interact-
ing H2O molecules one would define pair distributions gHH, gOO, and gHO and
use them to construct the internal energy of liquid water.
As in the case for solids, the long-range electrostatic and polarization inter-
action energies depend on the dielectric constant of the liquid. This can be
obtained from the polarization P liq = j ρjd j = j ρj αj lcl given by the sum of
induced atomic dipoles per unit volume in an external electric field. In the sim-
ple case of a liquid of average density ρ with a single atomic species of induced
dipole d R , the local polarization is P liq R = d R ρ R , where
ρ R = jδ R−Rj is the local density of atoms. Its thermal average is usually
a smooth function over space within the liquid, which introduces the liquid pair
distribution function in ρ R = ρg R; ρ, β , giving

P liq R = ρ α R g R; ρ,β lcl R

in terms of the local atomic polarizability α R . From this, one can obtain the
dielectric constant εr of the liquid as done before for a solid, given by
εr − 1 1
= d 3 Rρ α R g R; ρ,β
εr + 2 3ε0
which however must be generalized for anisotropic species requiring a polariz-
ability tensor, and for polar molecules with a permanent dipole [34].

4.5 Interaction Energies in a Liquid Solution and in

Physisorption
4.5.1 Potential Energy of a Solute in a Liquid Solution

The interaction energy between a solute molecule M at location RM and solvent

molecules Sn at location Rn can be constructed from electronic structure calcu-
lations of the energy of M perturbed by surrounding solvent molecules plus the
energies of interaction of the solvent molecules among themselves, obtained by
considering the solution as organized into solute and solvent fragments [35].
136 4 Model Potential Functions

The treatment is simpler when there are no chemical bondings between solute
and solvent species, and one can use already constructed pair and triplet inter-
action energies. The liquid solution total interaction energy takes the form
Coul pol rep Coul pol rep
Vsol = VMS + VMS + VMS + VSS + VSS + VSS
which includes Coulomb (or electrostatic), polarization (induction and disper-
sion), and repulsion (electron penetration and exchange) contributions between
int
the molecule and solvent species through the pair potential energies vMS , with
int
int = Coul, pol, rep, and solvent–solvent interaction potential functions vSS ,
plus possibly contributions from triplet interactions. The interactions among sol-
vent species have been described in the previous section. Here we concentrate on
Coul pol rep
the total interaction potential energy Vslt = VMS + VMS + VMS acting on the
solute molecule M.
It helps to distinguish short-range (SR) and long-range (LR) interactions, with
SR sums containing only nearby molecules, while sums of LR terms are averaged
over large distances, are smoother, and can be expanded in a Fourier series or
averaged over space. To proceed, it is convenient to separately treat short-range
rep
interactions in VMS as differently from the other, long-range interactions.
Short-range interactions between solute or solvent molecules are important
only between nearest neighbors or next nearest neighbors. We assume to sim-
plify first that the molecules are isotropic on average. The short-range (repul-
SR rep
sion) interaction energy Vslt = Vslt for the solute is a sum of short-range pair
potentials for atoms in the solvent S, which can be written as
SR SR
Vslt = v
n S MS
RM −Rn

However, the SR potential extends only a short distance away from the origin
and involve only the regions of pair distributions coming mostly from nearest
neighbors, and next-nearest-neighbors shells to a smaller extent. Restricting the
nearest-neighbors to a narrow distance around the maximum of the pair distri-
m
bution functions, which coincide with the minima RMS of the pair potential, a
rough estimate of the short-range potential energy near thermal equilibrium at
the solute molecule M is
SR SR m NN
Vslt ≈ vMS RMS zMS
eql

NN
where zMS is the number of nearest neighbor species S. The SR term can also
be written as

SR SR
Vslt = d 3 R vMS RM − R ρ S R
 4.5 Interaction Energies in a Liquid Solution and in Physisorption 137

where ρS R = nδ R−Rn is the local density of solvent species. Further-

more, the solute potential energy can be statistically averaged over a thermal
configuration distribution of the solvent, which contains the average density

−1
ρS R = ZN Ω,β d 3N R exp − βVN R1 , …,RN δ R −Rn
n

to obtain

SR SR
V slt RM = ρS d 3 R vMS RM − R gMS R; ρS , β

in terms of the pair distribution function gMS for the M and S species insofar
ρS R = ρS gMS R; ρS , β , with ρS the number of solvent molecules per unit vol-
ume, that can be obtained from a potential of mean force. More generally the
pair potential function of R must be replaced by a function containing internal
distances and orientation angle variables for the M and S molecules, and statis-
tical averages can be performed over all internal variables.
The remaining long-range interactions due to electrostatic and polarization
terms are again given by sums of long-range pair potentials suitably cut-off
LR Coul pol LR
at short distances, vMS = vMS + vMS , which contribute to an energy Vslt .
The long-range term becomes

LR LR LR
Vslt = v
n Mn
RM −Rn = d 3 R vMS RM − R ρS R

This integrand is usually a smooth function over space within the solid. It can
be thermally averaged to obtain

LR LR
V slt RM = ρS d 3 R vMS RM − R gMS R; ρ, β

Here, the radial integral is nearly zero for values smaller than the radius RMS,0,
where the LR potential is large but the pair distribution function vanishes.
Beyond this radius the integrand is a smooth function of R which can be
approximated by an analytical expression. It can be obtained for an attractive
polarization potential of form vMS R = −CMS R −s , with s = 6 for a dispersion
LR s

potential or s = 4 for induced polarization, approximating gMS by its hard sphere

c c
form gMS(R) = 0 for R ≤ RMS and gMS(R) = 1 for R > RMS . This follows from a
c
potential of mean force wMS strongly repulsive inside a cavity radius RMS of
138 4 Model Potential Functions

the order of the sum of van der Waals radii for M and S. With the origin of coor-
dinates at the CMN of M and for an isotropic solvent, introducing the relative
position vector r =RM − R we find
∞
LR LR LR
V slt = ρS d3 r vMS r gMS RM − r ; ρ, β ≈ρS 4πr 2 dr vMS r
pol c
RMS

c −s + 3
s
4πρS CMS RMS
≈
−s + 3
giving a negative (attractive) solute potential energy which is larger for a smaller
solute cavity radius. A similar approximation for the solvent density is obtained
c
assuming that solvent atoms are packed with hard sphere radius RSS so that
3 −1
c
ρS = 4π RSS 3 , the inverse of the volume per solvent species.
A more accurate treatment can be derived using what is known about the pair
distribution function and related structure factor, from measurements or from
molecular dynamics calculations. Introducing the Fourier transform of the LR
pair potential and its inverse transform,

LR −3 2 LR
v MS k = 2π d 3 R vMS R exp ik R ,

LR −3 2 LR
vMS R = 2π d 3 k v MS k exp − ik R

we find from the sum over pair potentials

LR −3 2 LR
Vslt = 2π d 3 k v MS k exp − ik RM n
exp ik Rn

Further, considering that the transform of the thermally averaged number den-
−3 2
sity ρS R of the solvent is ρ S k = 2π n exp ik Rn we find for the
th
LR part,

LR LR
V slt = d 3 k v MS k ρ S k exp − ik RM

A general way to evaluate this involves the introduction of the Fourier transform
hMS k of the M–S pair correlation function hMS R = gMS R − 1, that is a
structure factor measurable by X-ray scattering. In detail,
−3 2
hMS k = ρS 2π d3 R exp ik R gMS R − 1
 4.5 Interaction Energies in a Liquid Solution and in Physisorption 139

from which the pair distribution function is obtained by the inverse Fourier
transform, as
−3 2
gMS R = 1 + 2π d 3 k exp − ik R hMS k ρS

Replacing this in the integrals over R, we find the solute long-range potential
energy expressed in terms of the Fourier transforms of the long-range part of
the pair interaction potential and of the structure factors, as
LR LR LR
V slt RM = d 3 k v MS k hMS k exp −ik RM −ρS d 3 R vMS RM − R

Further simplification follows from observing that for the present isotropic
potentials and distributions, the potential Fourier transforms and the
present structure factors depend only on the wavevector magnitude k and
are given by
∞
L − 3 2 −1 L
v MS k = 2π k dR 4πR sin kR vMS R
0
∞
−3 2 −1
hMS k = ρS 2π k dR 4πR sin kR gMS R − 1
0
L
which allows calculation of Vslt from one-dimensional integrals over k. The
structure factor hMS(k) can also be calculated from pair potentials by means
of perturbation theory [33].
The extension to nonspherical, rotating, and vibrating M and S molecules
requires reintroduction of internal orientation angles and atomic displacement
variables in pair potentials and in pair distribution functions, and statistical
averages over all internal variables of M and S components of the liquid solution.
The mean interaction wMM(R) between two solute molecules in the solution
differs from the bare interaction pair potential vMM(R) due to medium effects
and is a potential of mean force which depends on the dielectric constant of
the solvent. A general treatment can be based on the susceptibility function
of the solvent, by considering the response of the whole system to an applied
external electric field ext [36].

4.5.2 Potential Energies of Atoms and Molecules Adsorbed

at Solid Surfaces

To begin with we consider an atom A physisorbed at position RA on the surface

of a solid made up of atoms or ions located at positions Rn , n = 1 to Nnu, with all
positions given in a reference frame with an origin at a chosen point RO and a
140 4 Model Potential Functions

z-axis perpendicular to the solid surface and pointing into a vacuum. The solid
may contain several types of atoms, and may have a crystalline structure or be
amorphous, such as a silicon solid with its surface dangling bonds saturated with
hydrogen (Si(s):H), or titanium dioxide (TiO2(s)). In the absence of chemical
bonding between the adsorbate and the surface atoms, it is usually accurate
to construct the total adsorption potential energy Vads as a sum of pair potential
energies vAn RA −Rn , so that for fixed positions of atoms in the solid,
Vads RA = n vAn RA −Rn . The treatment is then similar to the one for a sol-
ute in a solvent. However, for metal and semiconductor surfaces with deloca-
lized electrons, the pair potentials vAn must be constructed to account for
embedding effects as treated in Chapter 8, and are also functionals of the elec-
tronic densities at the surfaces. What follows here applies to nonbonding adsor-
bates and localized surface electrons.
Separating short range (SR) and long range (LR) terms in the pair potential
SR LR
one can write Vads = Vads + Vads to derive approximations. The short range
potential can be obtained from the nearest neighbors (NTs) of the adsorbed atom,
so that
SR SR
Vads RA ≈ v
n NT An
RA −Rn

which can be evaluated for a known location of the adsorbate on the surface, and
from the atomic structure of the surface and inner layers. This can be improved
by adding next-NTs.
The long range term becomes

LR LR
Vads RA = v
n An
RA −Rn

which is usually a smooth function over space within the solid. For a solid con-
taining atoms of type J = B, C, … it can be written in terms of the number den-
J
sities ρsld R = n Jδ R −Rn of the solid as

LR LR J
Vads RA = d 3 R v
J AJ
RA − R ρsld R

Introducing the Fourier transform of the LR pair potential and its inverse
transform,
LR −3 2 LR
v AJ k = 2π d 3 R vAJ R exp ik R ,

LR −3 2 LR
vAJ R = 2π d 3 k v AJ k exp − ik R
 4.5 Interaction Energies in a Liquid Solution and in Physisorption 141

we find

LR −3 2 LR
Vads RA = 2π d 3 k exp −ik RA v
J AJ
k n J
exp ik Rn

Further considering that the transform of the thermally averaged number den-
J −3 2
sity of the solid is ρ sld k = 2π n J exp ik Rn we get for the ther-
th
mally averaged LR part,

LR LR J
V ads RA = d 3 k exp − ik RA v
J AJ
k ρ sld k

Special cases follow when the atoms form a crystalline lattice, and when one can
assume that on the average the solid has a constant density which vanishes past
the surface.
The treatment simplifies when all atoms in the solid are the same, with vAJ = vA.
The total energy operator for a solid in a volume Ω is then
Vads RA = Ωd
3
R vA RA − R ρsld R and this can be statistically averaged over
a thermal distribution of atom locations for a solid at temperature T giving a
density ρ R; T = ρsld g R; ρsld ,β in terms of the pair distribution function of
atoms in the solid relative to the origin of coordinates, and an average potential
energy of adsorption

V ads RA ; T = ρsld d 3 R vA RA − R g R; ρsld ,β

Ω

which can be further treated separating SR and LR terms. For the SR term one
S S
again has Vads RA ≈ n NT vA RA −Rn , while for the LR we can estimate the
potential energy on the adsorbate by assuming that the solid is crystalline, or
homogeneous with a smooth surface at distance D from the adsorbed atom.
Consider the case where the solid surface is on a (xy)-plane with atoms
labeled by integers n = (nx, ny, nz) in a cubic crystalline lattice generated by
translations of a unit cell with axes along orthogonal unit vectors ux ,uy ,uz
and edges of length ax, ay, az, with one atom per cell at locations Xn = X0 + nxax
and Yn = Y0 + nyay with − ∞ < nξ < ∞. The locations along the z-direction
are Zn = Z0 − nzaz with 0 ≤ nz < ∞. The density is now a lattice periodic
s
function ρsld R = ρlatt R = ρlatt R + t n , with a surface periodic translation
s
t n = nx axux + ny ayuy . The condition that the density must be invariant under
translations along x- and y-directions by steps ax and ay means that now an
expansion in related wavevector components must only contain values
142 4 Model Potential Functions

x
kx = 2π mx ax = gmx and similarly for ky, with integers − ∞ < mξ < ∞ [31]. The
density is given by

s +∞
ρlatt R = m
exp ig ms R dk z exp ikz Z ρ latt g ms ,kz
−∞

s s
where g m = 2π mxux ax + myuy ay and R = xux + yuy are surface vectors. The
density ρ latt in the reciprocal space of k vectors can be extracted by taking its
inverse transform and using it in the adsorbate expression

L s L
Vads RA = m
exp ig ms RA dkz v A g ms ,kz ρ latt g ms ,kz exp −ikz ZA

s s
with the lattice density obtained for R = t n as

−3 2 −3 2
ρ latt g ms ,kz = 2π n
exp ig ms t ns exp ikz Zn = 2π n
exp ikz Zn

The integral and summations over lattice points can be evaluated given the
Fourier transform of the pair interaction potential.
For the solid treated as a homogeneous structure, and a pair interaction
vA R = − CA R − s between the adsorbed atom and an atom within a solid ele-
L s

ment of volume, integration within the solid with a surface perpendicular to a

z-axis can be done adding over its elementary rings of radius r and width dr par-
allel to the surface with volume dΩrng = 2πrdrdz, and with the number of atoms
in this ring volume being dNrng = ρslddΩrng. Taking the adsorbate distance to
1 2
that element of volume to be RA = r 2 + zA2 , with the zA variable originating
at A and growing into the solid, the thermally averaged LR potential energy
is then

∞ ∞ s
s s r 2πρsld CA
V A D = − 2πρsld CA dr dzA =−
0 D r 2 + zA2
s 2 s− 2 s −3 D s−3

This gives, for the van der Waals dispersion potential with s = 6, the well-
known D−3 dependence on the distance from the adsorbate to the surface.
This result must be modified when the adsorbate is surrounded by a medium
(such as a liquid) outside the solid, to account for the dielectric constant of that
medium [36]. Here, we can assume that the dielectric effects have been incor-
s
porated into the constant CA .
 4.6 Interaction Energies in Large Molecules and in Chemisorption 143

4.6 Interaction Energies in Large Molecules

and in Chemisorption
4.6.1 Interaction Energies Among Molecular Fragments
Large molecules such as polymers, proteins, and nucleic acids can be described
as made up of atomic groups or of molecular fragments = , , , … usually
selected on the basis of their chemical structure and bonding strength, interact-
ing among themselves as described by their electronic density distributions. The
assumption is that to zeroth order the total electronic wavefunction Φ(0) is a
0
superposition of products of fragment wavefunctions ΦJ for each fragment
, antisymmetrized for the exchange of electron variables among all fragments.
0
Each fragment has an electronic density ρJ r distributed among atomic cen-
ters, bonds, and localized electronic lobes. As a consequence the total electronic
density ρ 0 r is a sum of intra-fragment and inter-fragment density terms, or
0 0
ρ 0 r ≈ J ρJ r + JK ρJK r , if one discards additional smaller terms invol-
ving more than two fragments. A large two-fragment term indicates that there is
chemical bonding between those fragments.
Interfragment interactions contain short-range repulsion, electronic
exchange, electrostatic (or Coulomb), polarization, and dispersion terms. For
the calculation of electrostatic interaction energies, fragment electronic densi-
ties can be alternatively represented by a collection of weighted electronic point
0
charge densities as in ρJ r = j J wj δ r −Rj located at sites Rj of its atoms,
bonds, and lobes. The total density ρ r can be decomposed into terms
0

labeled by j and k, and interfragment electrostatic interaction energies can be

obtained from the interactions of sites j in and k in [5].
To improve on that zeroth order approximation when molecular fragments
and display chemical bonding, it is necessary to account for the electronic
rearrangement at the bonds in terms of more accurate molecular wavefunctions
ΦJK calculated for the bonded pair. An improved density is now of the form
0
ρ r ≈ J ρJ r + JK ρJK r and includes intrafragment ΔρJ r = ρJ r − ρJ r
0
and interfragment ΔρJK r = ρJK r −ρJK r density changes, which can be
represented by additional electrical multipoles at the locations of point charges.
Repulsion and exchange interaction energies can be extracted from quantal
electronic structure calculations [1, 37], or from mixed (quantum mechan-
ics)/(molecular mechanics) (QM/MM) treatments [38, 39]. Electrostatic induc-
tion and dispersion interaction energies can be obtained from a set of electrical
multipoles and a set of polarizabilities for each fragment and their pair interac-
tions. Improved fragment electronic density and polarizability distributions
accounting for bonding can be calculated from more accurate wavefunctions
ΦJK [5, 35].
144 4 Model Potential Functions

When intramolecular fragment–fragment interactions are thermally aver-

aged by atomic motions and solvent effects, less than complete details are
needed in the description of properties of fragments, and these can be treated
in an average sense with continuum models, or with coarse graining methods. It
also helps to distinguish short-range and long-range fragment interactions, with
SR sums containing only nearby fragments, while sums of LR terms can be
expanded in a Fourier series or simply averaged over space.
Given the electronic wavefunction ΨM of the large molecule M for fixed
nuclear positions, possibly in a medium of solvent molecules with their charge
and polarization densities, the liquid solution total potential energy takes
the form

0 Coul pol rep

VM, sol = ΨM H M + H MS + H MS + H MS ΨM
Coul pol rep
+ VSS + VSS + VSS

which includes Coulomb (or electrostatic), polarization (induction and dispersion),

and repulsion (electron penetration and exchange) potential energy contributions
between the molecule and solvent species through the Hamiltonian operators
int
H MS , with int = Coul, pol, rep, and solvent–solvent interaction potential func-
int 0
tions VSS . The Hamiltonian operator H M contains the electronic kinetic ener-
gies and all the molecular electronic and nuclear Coulomb interactions.
The wavefunction ΨM can be obtained from perturbation theory starting from
0
a wavefunction ΨM of the isolated M, and adding a perturbation term ΨM due
to the presence of solvent species. This additional term depends parametri-
cally on the positions of solvent species and their electronic charge densities.
The expectation values of the Hamiltonian operators for M, and the M–S
interaction energies can be constructed from information about bonding
between fragments and charge and polarization densities in the fragments,
and also from charge and polarization densities in the solvent and how they
interact with the electrons and nuclei in the solute M described by its atomic
groups or molecular fragments. The interactions among solvent molecules
can be modeled with the functions already described for interacting non-
bonding pairs. But the interactions in the solute M and between molecule
M and the solvent require a full treatment of the electronic structure of M
and how it is perturbed by the solvent. This is the subject of a following
Chapter 7.
A similar treatment is applicable to a system where a molecule is chemically
bonded to a surface, and the chemisorption complex can be considered to be a
supermolecular system by analogy with the way energies are described in large
molecules. The treatment is simplified when the atomic structure of the surface
shows periodicity as in crystal surfaces.
 4.6 Interaction Energies in Large Molecules and in Chemisorption 145

4.6.2 Potential Energy Surfaces and Force Fields in Large Molecules

Using knowledge from electronic structure calculations, thermochemistry, and
molecular spectroscopy of molecular fragments and their interactions, it is
possible to construct potential energy functions of all the relevant atom posi-
tions in the system and to generate force fields from the gradients of the poten-
tial with respect to atomic displacements. These forces are basic to the
calculation of the structure of new compounds, and to calculation of their
molecular properties.
A large molecule containing many atoms of different types, as found espe-
cially in biomolecules, complex materials, and nanostructured surfaces, show
complicated deformations when subject to forces resulting from interactions
with other molecules or from electromagnetic radiation. Deformations may
involve a bond stretching between two atoms, an angle bending between two
adjacent bonds involving three atoms, an angle torsion between two adjacent
atomic planes defined by four atoms, and buckling angles for an atom displaced
relative to a plane defined by three atoms.
As long as the whole system is in a given electronic state and there are no elec-
tronic excitations involved, the potential energy can be given as a function of
atomic positions, or as a function of internal variables dependent on several
atomic positions, which when varied generate a potential energy surface
(PES). The choices of variables depend on whether the molecular fragments
composing the total system are chemically bonding or nonbonding. The poten-
tial energy for an isolated large molecule M can be written as

0 0 0
VM = ΨM H M ΨM =V B
+V NB

with the first term a function of internal bond variables describing bonding frag-
ments, and with the second term a function of atomic group or fragment posi-
tions for interacting nonbonding fragments. In detail, the bonding potential
energy can be given in terms of the bond distances dab between atoms a and
b, bending angle ϑabc, torsion angle γ ab, cd between two planes with a common
edge, and buckling angle γ a, bcd between a plane and a line to a vertex. Provided
displacements from equilibrium values are small, changes in lengths or angles
can be used as variables, and deformation energies can be expanded in power
series of displacements from equilibrium. We write

B
V = Vstr + Vbnd + Vtor + Vbck

where Vstr is the stretch component involving sums over two atoms, Vbnd is the
bending components for sets of three atoms, Vtor the torsion component with
sums over four atoms, and Vbck the buckling component also for four atoms in
146 4 Model Potential Functions

each term. These terms can be constructed from calculated or measured

properties of isolated fragments. In detail, introducing force constants K for
each type of displacement, equilibrium variable values, and a torsion integer
index n,

1 eq
2
Vstr = a<b2
Kab dab − dab

1 eq
2
Vbnd = K
a < b < c 2 abc
ϑabc − ϑabc

1 eq
Vtor = a<b<c<d 2
Kab, cd 1+ cos n γ ab, cd − δab, cd

1 eq
2
Vbck = a<b<c<d 2
Ka, bcd γ a, bcd − γ a, bcd

where the summations involve only neighbor atoms. To this one must add
energy terms for hydrogen bonding, if present, taken to involve atoms
a, b (for H), and c, and expressed as a function of a variable Sac =
d cb −d ab = dab dbc cos ϑabc of the form

1 2
vaHc = KH Sac −Saceq + KH Sac − Saceq
2

or alternatively as a potential function of the distance daH between the hydrogen

and a nonbonding atom a, with a repulsive term varying as the inverse distance
to the power of 12 and an attractive term with the inverse distance to the power
of 10, with suitable parameters. When needed, improvements in accuracy of the
PESs can be obtained by allowing coupling of two adjacent stretching bonds, in a
potential energy Vstr,str(dab, dbc) containing three atom terms, and coupling of
stretch and bending in a Vstr,bnd(dab, ϑabc) function.
The nonbonding potential energy involves short-range repulsive and long-
range polarization terms that can be constructed for pairs of atomic groups
NB SR LR
or molecular fragments , as vIJ RIJ = vIJ RIJ + vIJ RIJ for a pair at
relative distance RIJ, with the repulsive SR term an exponential or inverse power
LR
function, and the LR interaction vIJ RIJ containing induction, dispersion, and
electrostatic (or Coulomb) terms deriving from charges and polarizabilities
embedded in the group fragments. Using in particular the van der Waals
LJ(12,6) function combining repulsion and dispersion energies with energy
minimum ϵIJ at the distance σ IJ, and collections of point charges cj and
 4.6 Interaction Energies in Large Molecules and in Chemisorption 147

polarizabilities αj at each fragment chosen to replicate the induction and Cou-

lomb interactions, the nonbonding potential energy is
12 6
σ IJ σ IJ Coul ind
V NB
= ϵ
I , J IJ
−2 + vIJ Rij + vIJ Rij
RIJ RIJ
Coul ci cj
vIJ Rij = fd Rij
i I, j J 4πε0 εr Rij
ind ci αj + αi cj
vIJ Rij =− i I, j J 4πε ε 2 R 4 d
f Rij
0 r ij

where εr is the dielectric constant of the medium. The Coulomb and induction
terms have been given as sums of pair interactions between point charges and
polarizabilities located at fragments and at the relative distance Rij. They
can also be cut-off at short distances using as shown suitable damping functions
fd(Rij), previously introduced. The present force fields V (B) + V (NB) and similar
variations have been extensively parametrized on the basis of electronic struc-
ture calculations and empirical information [40–44]. The energy landscapes
generated by these force fields can be shown as energy isocontours versus
two bond variables, keeping the other ones constant. Figure 4.8 shows the def-
inition of torsion angles φ and ψ for a polypeptide strand, and potential energy
isocontours versus those angles.

(a) (b)

180°

–180°
–180° 𝜙 180°

Figure 4.8 (a) Planes showing the definition of torsion angles ψ and φ between two peptide
units in a polypeptide. (b) Potential energy isocontours for the polypeptide glycyl residue energy
versus the torsion angles phi around the C─N bond and psi around the C─C bond
(a Ramachandran diagram). Ovals lead to energy minima. Source: from Hovmoeller [45] and [46].
148 4 Model Potential Functions

More accurately, the LR terms must be obtained from the perturbed wave-
0
function ΨM instead of ΨM , to account for interfragment interactions, and must
also include electric multipole interactions between permanent and induced
charges, which arise from intramolecular and intermolecular polarization of
atoms and bonds. For the isolated molecule M, the resulting more accurate
interactions are so-called polarizable force fields [47–51]. In addition, the total
potential energy for the M–S liquid solution must account for the interaction of
the molecule M with solvent molecules.
The M–S interaction requires knowledge of the wavefunction of M and cal-
Coul pol rep
culation of the expectation value ΨM H MS + H MS + H MS ΨM , where
the Hamiltonian operators depend on the locations and charge distribution of
each solvent species interacting with M. These interactions perturb M and
involve repulsion, polarization, and Coulomb (electrostatic) terms as shown.
To first order in the M–S interaction potentials and assuming that the solvent
species have unchanged charge distributions, the expectation value can be
0
obtained from the wavefunction ΨM of the isolated M with its own permanent
and induced charge distribution. Higher-order terms in a perturbation expan-
0 1 2
sion ΨM = ΨM + ΨM + ΨM + … involve excited states of M and describe solute
polarization, which include additional induction and dispersion terms, to first
and second order in the M–S interaction. Further changes in potential energies
arise from polarization of the solvent species.
Thermal equilibrium properties of the molecular system can be obtained
from internal energies derived from the above potential energies plus the system
kinetic energy at a given temperature. More detailed properties, such as isomer-
ization rates, follow from molecular dynamics calculations involving force fields
given by gradients of the potential functions with respect to atomic displace-
ments [52, 53].

4.6.3 Potential Energy Functions of Global Variables Parametrized

with Machine Learning Procedures
Given the structure of a many-atom system by a set of all its atomic position
vectors Q = Ra , a potential energy function V(Q) for the ground electronic
state can be written as a sum over embedded atoms, V Q =
emb
a Va Q ,a = 1 to Nat , where each term depends not only on the position
of atom a but also on the positions of atoms in the environment in which it
is embedded. The PES may be known from calculations or measurements
at a number of conformations. It can be given as a sum over embedded
atom terms as shown, but each term must contain fitting parameters which
describe their environment, and the total number of parameters can be
very large.
 4.6 Interaction Energies in Large Molecules and in Chemisorption 149

It is possible to use artificial intelligence (AI) procedures making use of

machine learning to train artificial neural networks (NN) which find all the
needed parameters to fit the known PES V(Q) at calculated conformations
and to provide an interpolation for energies at other conformations. This
was done for molecules chemically bonded to solid surfaces [54], for weakly
bonded molecules [55], and for reaction PESs containing activation energies
[56]. A number of improvements on the methodology and many more recent
applications have been recently reviewed [57, 58], and will be briefly presented
here with more details given in Chapter 7.
The early AI-NN procedure has been recently modified and extended to make
it more efficient and general. Instead of using position coordinates as variables,
the recent treatments have introduced global variables Gα(Q) forming a new set
G = {Gα(Q)} specifying the embedded atom positions and its environment with
the correct symmetry, so that

V Q = E emb
a a
G Q

The global variables can be chosen so that fittings to known values are phys-
ically invariant under collective translation and rotation of the rigid many-
atom system, and can also include environmental symmetries such as they
arise from the presence of identical atoms there. For example global (symme-
try) variables X1, … , X8 for the H2/Pd(100) adsorbate have been chosen as
functions of the six diatomic position variables (three for the center-of-mass
of H2 and three for its distance to the surface and the diatomic orientation
angle) with X4, X5, X7, X8 taken as Fourier spatial components oscillating with
the reciprocal lattice vector periodicity of the substrate surface [59]. Symmetry
adapted global variables have also been constructed using permutational sym-
metry of identical atoms in treatments of reaction PES for H + H2 and Cl +
H2 [60].
More generally, global symmetry functions for each embedded atom a have
been defined in terms of the position coordinates of neighbor atoms within a
rad ang
finite sphere of chosen radius Rc. They are radial Ga and angle Ga variables
dependent on two or three atom positions and given by [61]

sph
Garad = j a
exp − ηR2aj fc Raj

sph
Gaang = 21− ζ j, k a
1+ cosθajk exp − η R2aj + R2ak + R2jk fc Raj fc Rak fc Rjk

with the sphere cut-off function fc(R) = [1 + cos(πR/Rc)]/2 for R ≤ Rc and zero
past Rc, angle θajk between the bonds aj and ak, and parameters η and ζ to
be found.
150 4 Model Potential Functions

The AI-NN procedure as presently used [57] involves at least an input NN

μ
layer with nodes μ containing each Ga value, outputting their values into a hid-
den layer (number 1) of nodes m (as many as needed for training) coding partial
1
energy components xa, m given by

xa1, m = bm1 + μ
Gaμ wμ01
,m

01
Here, the weights wμ, m are parameters to be found, connecting the global var-
iable in the input layer to node m in the hidden layer. This partial energy com-
ponent is accepted with an activation (or acceptance) probability biased by the
1
magnitude of bm . The bias parameter is introduced to allow for an acceptance
somewhere between zero and one for the output of node m, given by a sigmoid-
1 1 1
shape (activation) function f xa, m = ya, m varying from zero at minus infinity
1
to one at plus infinity, with bm setting its value in the slope region and its
acceptance probability. This function can be chosen, for example as [1 + tanh
(x)]/2.
When only one hidden layer is used in the NN, the final stage constructs the
total energy of each embedded atom with

Eaemb = y1
m a, m
Gaμ ; bm1 , wμ01
,m

as a sum of the outputs from all the nodes of the hidden layer. The parameters
1 01
bm and wμ, m in the acceptance function follow from a NN training that opti-
emb
mizes the fitting of a set of reference PES values Ea, ref for given input atomic
coordinates, using usually the mean absolute deviation (MAD)
1 emb emb
MAD = Ea, ref −Ea, NN
Nat a

which must be minimized by the selection of parameters. An alternative is to

minimize the root-mean-square error. Once the parameters are known, the
interpolation of values to new atomic positions must be checked, and this is
done using a validation set of previously generated PES values different from
the ones used for training the NN. The total PES is given by the sum of
emb
embedded atom energies, V = a Ea . The architecture of the NN is illus-
trated in Figure 4.9. Listing the number of nodes per layer and sizes of layers,
this is a 2-3-1 NN.
Alternative AI-NN architectures have been explored, with different numbers
of nodes in the input layer and more hidden layers in a deep-learning procedure.
The training and validation sets of PES values have been generated with a variety
 4.6 Interaction Energies in Large Molecules and in Chemisorption 151

Figure 4.9 Architecture of a neural Input Hidden Output

network with an input layer with layer layer layer
N = 2 nodes containing initial
assigned values of global variables
coming from chosen atomic W11
1 Y11
2
conformations. A hidden layer with W11
M = 3 nodes and inputs from those G1
variables, gives outputs with
2
acceptances biased by an activation W21
function. The outputs of the hidden Y12 E
layer nodes are added to generate 2
the energy of an embedded atom in G2 W31
an output layer. Source: from [61].
1
W23
Y13 2
W1
1
b1
1
b3

Bias

of empirical and many-electron methods to test accuracies of the NN. As an

example to test the AI-NN procedure in recent calculations, energies of the
adsorbate in H2/Pd(100) were generated with an empirical PES with 1560 train-
ing points and 5200 validation points and a 8-50-50-1 NN, which gave a vali-
dation error of 0.150 eV [59]. The energies of several solid structures of Si
have been obtained using DFT [61], with 8200 PES training points and 800 val-
idation points, for a NN containing 48 nodes in the input layer and two hidden
layers with 40 nodes each (a 48-40-40-1 NN), giving a validation accuracy of 5
meV atom−1. Another example is the calculation of the PES of vinyl bromide
undergoing unimolecular decomposition [56], done generating energy sets with
the MP4 many-electron treatment, with 1120 training points and 140 validation
points, a 12-20-1 NN, and validation error of 0.001 eV.
The procedure just described can give very accurate fits to PESs with errors of
the order of a few meV atom−1. However, it needs to be reconsidered when
charge transfer between atoms occurs as interatomic distances change, for
example when an ionic bond changes into a neutral atom–atom interaction
at large distances, such as in the ground electronic state of NaCl. Reconsider-
ation may also be needed when long-range interactions are important, because
dispersion energies may not be accounted by the training done with spheres of
relatively small radius. A possible procedure in these cases is to increase the
radius of the embedded atom spheres to verify that the training incorporates
charge transfer and dispersion energies. This is however computationally
demanding and other approaches need consideration. This is an active area
of research.
152 4 Model Potential Functions

Forces Fa, ξ at atom a along the ξ-direction in the many-atom structure can be
obtained as needed in molecular dynamics calculations from
emb
∂V N M ∂ Ea ∂Gα
Fa, ξ = − =−
∂Ra, ξ a=1 α = 1 ∂G
α ∂R a, ξ

where N is the number of atoms and M the number of global symmetry func-
tions. The fraction in the first factor is given by the NN architecture and the
second factor follows from the choice of the global variables.
The procedure given here for interpolation of PESs allows for accurate fit-
tings, improvement if additional calculated points are available for training of
the NN, and is quite general for ground electronic states. It allows for descrip-
tion of bonding and nonbonding (weak or van der Waals) interactions and for
systems undergoing reactions with broken and formed bonds. And it can pro-
vide results with the accuracy of the training set (such as DFT or CCSD) at the
same computational cost as force-field calculations. The AI-NN procedure,
however, requires preliminary calculations of PESs with standard treatments
to be used for training and validation so that the AI-NN procedure is comple-
mentary to standard treatments of intermolecular forces using molecular prop-
erties or many-electron methods.
Given calculated PESs for a large class of compounds and interacting pairs of
molecules, an AI-NN procedure can also be used to calculate new compounds
and their interaction, provided they contain the same embedded atoms [62, 63].
This however must be done with caution and extensive validation, and some of
these aspects are treated in Chapter 7.

References
1 Hirschfelder, J.O., Curtis, C.F., and Byron Bird, R. (1954). Molecular Theory of
Gases and Liquids. New York: Wiley.
2 Margenau, H. and Kestner, N.R. (1971). Intermolecular Forces, 2e. Oxford,
England: Pergamon Press.
3 Murrell, J.N., Carter, S., Farantos, S.C. et al. (1984). Molecular Potential Energy
Functions. New York: Wiley.
4 Schatz, G.C. (1989). The analytical representation of electronic potential energy
surfaces. Rev. Mod. Phys. 61: 669.
5 Stone, A.J. (2013). The Theory of Intermolecular Forces, 2e. Oxford, England:
Oxford University Press.
6 Kaplan, I.G. (2006). Intermolecular Interactions, 81. New York: Wiley.
7 Herzberg, G. (1950). Molecular Spectra and Molecular Structure I. Spectra of
Diatomic Molecules, 2e. Princeton NJ: Van Nostrand.
8 Lawley, K.P. (ed.) (1975). Molecular Scattering (Advan. Chem. Phys. vol. 30).
New York: Wiley.
 References 153

9 Maitland, G.C., Rigby, M., Brain Smith, E., and Wakeham, W.A. (1981).
Intermolecular Forces. Oxford, England: Oxford University Press.
10 Jensen, F. (1999). Introduction to Computational Chemistry. New York: Wiley.
11 Karplus, M. and Porter, R.N. (1970). Atoms and Molecules. New York: W. A.
Benjamin.
12 Kramer, H.L. and Herschbach, D.R. (1970). Combination rules for van der Waals
force constants. J. Chem. Phys. 53: 2792.
13 Smith, F.T. (1972). Atomic distortion and the combining rule for repulsive
potentials. Phys. Rev. A 5: 1708.
14 Koide, A., Meath, W.J., and Alltnatt, A.R. (1981). Second order charge overlap
effects and damping functions for isotropic atomic and molecular inetractions.
Chem. Phys. 58: 105.
15 Tang, K.T. and Toennies, J.P. (1984). An improved simple model of the van der
Waals potential based on universal damping functions for dispersion
coefficients. J. Chem. Phys. 80: 3726.
16 Siska, P.S., Parson, J.H., Schafer, T.P., and Lee, Y.T. (1971). Intermolecular
potentials from crossed beam differential elastic scattering measurements III. He
+He and Ne+Ne. J. Chem. Phys. 55: 5762.
17 Buck, U. (1975). Elastic scattering. Adv. Chem. Phys. 30: 313.
18 Smith, I.W.M. (1980). Kinetics and Dynamics of Elementary Gas Reactions.
London, Great Britain: Butterworth.
19 Hirst, D.M. (1985). Potential Energy Surfaces. London, England: Taylor & Francis.
20 Gordon, M.D. and Secrest, D. (1970). Helium-atom-hydrogen-molecule
potential energy surface employing the LCAO-MO-SCF and CI methods.
J. Chem. Phys. 52: 120.
21 Edmonds, A.R. (1960). Angular Momentum in Quantum Mechanics. Princeton
NJ, USA: Princeton University Press.
22 Levine, I.N. (2000). Quantum Chemistry, 5e. New Jersey: Prentice-Hall.
23 Mahoney, M.W. and Jorgensen, W.L. (2000). A five-site model for liquid water
and the reproduction of the density anomaly by rigid, non-polarizable potential
functions. J. Chem. Phys. 112: 8910.
24 Karplus, M. (1970). Potential energy surfaces. In: Molecular Beams and Reaction
Kinetics (ed. C. Schlier), 320. New York: Academic Press.
25 Sato, S. (1955). Potential energy surface of the system of three atoms. J. Chem.
Phys. 23: 2465.
26 Truhlar, D.G., Steckler, R., and Gordon, M.S. (1987). Potential energy surfaces
for polyatomic reaction dynamics. Chem. Rev. 87: 217.
27 Polanyi, J.C. (1972). Some concepts in reaction dynamics. Acc. Chem. Res. 5: 161.
28 Bender, C.F., Pearson, P.K., O’Neil, S.V., and Schaeffer, H.F. III (1972). Potential
energy surface including electron correlation for F + H2 FH + H. I.
Preliminary surface. J. Chem. Phys. 56: 4626.
29 Bender, C.F., O’Neil, S.V., Pearson, P.K., and Schaeffer, H.F. III (1972). Potential
energy surface including electron correlation for F + H2 FH + H: refined linear
surface. Science 176: 1412.
154 4 Model Potential Functions

30 Micha, D.A. (1985). General theory of reactive scattering. In: Theory of Chemical
Reaction Dynamics, vol. II (ed. M. Baer), 181. Boca Raton, USA: CRC Press.
31 Kittel, C. (2005). Introduction to Solid State Physics, 8e. Hoboken, NJ: Wiley.
32 Tersoff, J. (1986). New empirical model for the structural properties of silicon.
Phys. Rev. Lett. 56: 632.
33 McQuarrie, D.A. (2000). Statistical Mechanics. Sausalito CA: University
Science Books.
34 Hansen, J.P. and McDonald, I.R. (1986). Theory of Simple Liquids, 2e. Orlando
FL, USA: Academic Press.
35 Gordon, M.S., Fedorov, D.G., Pruitt, S.R., and Slipchenko, L.V. (2011).
Fragmentation methods: a route to accurate calculations on large systems.
Chem. Rev. 112: 632.
36 Israelachvili, J. (1992). Intermolecular and Surface Forces. San Diego CA:
Academic Press.
37 Jeziorski, B., Moszynski, R., and Szalewicz, K. (1994). Perturbation theory
approach to intermolecular potential energy surfaces of van der Waals
complexes. Chem. Rev. 94: 1887.
38 Friesner, R.A. and Guallar, V. (2005). Ab initio quantum chemical and mixed
quantum mechanics/molecular mechanics (QM/MM) methods for enzymatic
catalysis. Annu. Rev. Phys. Chem. 56: 389.
39 Albaugh, A., Boateng, H.A., Bradshaw, R.T. et al. (2016). Advanced potential
energy surfaces for molecular simulation. J. Phys. Chem. B (Feature Article)
120: 9811.
40 Case, D.A., Cheatham, T.E.I., Darden, T. et al. (2005). The AMBER biomolecular
simulation programs. J. Comput. Chem. 26: 1668.
41 Brooks, B., Brooks, C.L., Mackerell, A.D. et al. (2009). CHARMM: the
biomolecular simulation program. J. Comput. Chem. 30: 1545.
42 Arnautova, Y.A., Jagielska, A., and Scheraga, H.A. (2006). A new force field
(ECEPP-05) for peptides, proteins, and organic molecules. J. Phys. Chem. B
110: 5025.
43 Van Der Spoel, D., Lindahl, E., Hess, B. et al. (2005). GROMACS: fast, flexible,
and free. J. Comput. Chem. 26: 1701.
44 Allinger, N.L., Chen, K., and Lii, J.-H. (1996). An improved force field (MM4) for
saturated hydrocarbonds. J. Comput. Chem. 17: 642.
45 Hovmöller, S., Zhou, T., and Ohlson, T. (2002). Conformations of amino acids in
proteins. Acta Cryst D58: 768.
46 Atkins, P. and De Paula, J. (2010). Physical Chemistry, 9e. New York: W. H.
Freeman and Co.
47 Kaminski, G.A., Stern, H.A., Berne, B.J. et al. (2002). Development of a
polarizable force field for proteins via ab initio quantum chemistry: first
generation model and gas phase test. J. Comput. Chem. 16: 1515.
48 Warshel, A., Kato, M., and Pisliakov, A.V. (2007). Polarizable force fields: history,
test cases, and prospects. J. Chem. Theory Comput. 3: 2034.
 References 155

49 Shi, Y., Xia, Z., Zhang, J. et al. (2013). Polarizable atomic multipole based
AMOEBA force field for proteins. J. Chem. Theory Comput. 9: 4046.
50 Baker, C.M. (2015). Polarizable force fields for molecular dynamics simulations
of biomolecules. WIREs Comput. Mol. Sci 5: 241.
51 Xie, W., Song, L., Truhlar, D.G., and Gao, J. (2008). The variational explicit
polarization potential and analytical first derivative of energy: towards a next
generation force field. J. Chem. Phys. 128: 234108.
52 Brooks, C.L.I., Karplus, M., and Montgomery-Pettitt, B. (1988). Proteins:
A Theoretical Perspective of Dynamics, Structure, and Thermodynamics. New
York: Wiley-Interscience.
53 Warshel, A. (1991). Computer Modeling of Chemical Reactions in Enzymes and
Solutions. New York: Wiley.
54 Blank, T.B., Brown, S.D., Calhoun, A.W., and Doren, D.J. (1995). Neural network
models of potential energy surfaces. J. Chem. Phys. 103: 4129.
55 Brown, D.F.R., Gibbs, M.N., and Clary, D.C. (1996). Combining ab initio
computations, neural networks, and diffusion Monte carlo: an efficient method
to treat weakly bound molecules. J. Chem. Phys. 105: 7597.
56 Raff, L.M., Malshe, M., Hagan, M. et al. (2005). Ab initio potential energy
surfaces for complex, multichannel systems using modified novelty sampling
and feedforward neural networks. J. Chem. Phys. 122: 084104–084101.
57 Behler, J. (2011). Neural network potential energy surfaces in chemistry: a tool
for large scale simulations. Phys. Chem. Chem. Phys. 13: 17930.
58 Behler, J. (2016). Perspective: machine learning potentials for atomistic
simulations. J. Chem. Phys. 145: 170901–170901.
59 Lorenz, S., Scheffler, M., and Gross, A. (2006). Description of surface chemical
reactions using a neural network representation of the potential-energy surface.
Phys. Rev. B 73: 115431.
60 Jiang, B. and Guo, H. (2013). Permutation invariant polynomial neural network
approach to fitting potential energy surfaces. J. Chem. Phys. 139: 054112.
61 Behler, J. and Parrinello, M. (2007). Generalized neural network representation
of high-dimensional potential-energy surfaces. Phys. Rev. Lett. 98: 146401–
146401.
62 Smith, J.S., Isayev, O., and Roitberg, A.E. (2017). ANI-1: an extensible neural
network potential with DFT accuracy at force-field computational cost. Chem.
Sci. 8: 3192.
63 Ramakrishnan, R., Dral, P.O., Rupp, M., and von Lilienfeld, O.A. (2015). Big data
meets quantum chemistry approximations: the delta-machine learning
approach. J. Chem. Theory Comput. 11: 2087.
 157

Intermolecular States

CONTENTS

5.1 Molecular Energies for Fixed Nuclear Positions, 158

5.1.1 Reference Frames, 158
5.1.2 Energy Density Functionals for Fixed Nuclei, 160
5.1.3 Physical Contributions to the Energy Density Functional, 162
5.2 General Properties of Potentials, 163
5.2.1 The Electrostatic Force Theorem, 163
5.2.2 Electrostatic Forces from Approximate Wavefunctions, 164
5.2.3 The Example of Hydrogenic Molecules, 165
5.2.4 The Virial Theorem, 166
5.2.5 Integral Form of the Virial Theorem, 168
5.3 Molecular States for Moving Nuclei, 169
5.3.1 Expansion in an Electronic Basis Set, 169
5.3.2 Matrix Equations for Nuclear Amplitudes in Electronic States, 170
5.3.3 The Flux Function and Conservation of Probability, 172
5.4 Electronic Representations, 172
5.4.1 The Adiabatic Representation, 172
5.4.2 Hamiltonian and Momentum Couplings from Approximate Adiabatic
Wavefunctions, 173
5.4.3 Nonadiabatic Representations, 174
5.4.4 The Two-state Case, 175
5.4.5 The Fixed-nuclei, Adiabatic, and Condon Approximations, 176
5.5 Electronic Rearrangement for Changing Conformations, 180
5.5.1 Construction of Molecular Electronic States from Atomic States:
Multistate Cases, 180
5.5.2 The Noncrossing Rule, 181
5.5.3 Crossings in Several Dimensions: Conical Intersections and Seams, 184
5.5.4 The Geometrical Phase and Generalizations, 189
References, 192

Molecular Interactions: Concepts and Methods, First Edition. David A. Micha.

© 2020 John Wiley & Sons, Inc. Published 2020 by John Wiley & Sons, Inc.
158 5 Intermolecular States

5.1 Molecular Energies for Fixed Nuclear Positions

5.1.1 Reference Frames
We must specify the positions of all the particles in our system in a suitable ref-
erence frame, with a chosen origin of coordinates and with axes oriented in a
specified way. We begin with a laboratory reference frame or laboratory system
L attached to a measuring device in the laboratory with axes x, y, and z oriented
along directions given by unit vectors nx , ny , nz . It is however convenient in the-
oretical studies to work in a center-of-mass reference frame. To simplify,
assume that only two nuclei a and b are present, of masses ma and mb, at posi-
tions Ra and Rb , and that electron masses can be ignored to start with. The cen-
ter of mass and relative positions are then

maRa + mbRb
RCMN =
ma + mb

R =Rb −Ra

and we can translate the reference frame keeping axes parallel so that in a new ref-
erence frame CMN called the center-of-mass of the nuclei, space-fixed (CMN-
SF) frame, we have RCMN = 0 and the (x, y, z) axes pointing along the previous
unit vectors. A further convenient transformation, when we describe the prop-
erties of a molecule in an equilibrium conformation, is to reorient the axes to
coincide with the principal axes of inertia of the nuclear framework, along new
unit vectors n x , ny , nz . This can be done specifying three Euler angles (α, β, γ) = Γ
of the new frame called now the center-of-mass of the nuclei, body-fixed (CMN-
BF) frame, designated as CMN . New coordinates (x , y , z ) are related to the old
ones by a 3 × 3 rotation matrix A α,β,γ = Rz γ Ry β Rz α , a function of the
Euler angles constructed from three axial rotations [1]. The three mentioned
reference systems are shown in Figure 5.1.
In the case of intermolecular potentials, one usually finds two molecules A
and B, each with its own CMN-BF reference frame I , I = A, B, with origins
at their centers of mass and oriented by Euler angles ΓI with respect to an overall
CMN-SF frame located along the line joining the centers of mass of A and B.
This is shown in Figure 5.2.
The remaining nuclear coordinates qI are internal ones, such as bond dis-
tances, bond angles, and torsion angles. Therefore, the whole system of Nnu
nuclei is described by the set of 3Nnu variables Q = R, ΓA ,qA , ΓB ,qB .
 5.1 Molecular Energies for Fixed Nuclear Positions 159

Figure 5.1 Reference frames for

laboratory (L), center-of-mass of
nuclei with space-fixed (CMN-
SF) orientation of orthogonal 𝒮CMN–SF
axes, and center-of-mass of
→ 𝒮CMN–BF = 𝒮′
nuclei with body-fixed (CMN-BF) Rb
orientation of axes.

→
Ra

𝒮L

Figure 5.2 Two molecules A

and B oriented in a common
CMN-SF reference frame and
with their own local reference 𝒮A′
frames. 𝒮CMN–SF
𝒮′B

𝒮L

The stationary state wavefunction Ψ of two interacting molecules is obtained

from the Schrödinger equation
1
P †Q P Q + H Q Ψ Q,X = E Ψ Q,X
2M
where we have introduced the electronic position and spin variables xi = r i ,ζ i
of electron i and the sets X = (x1, x2, …) for Nel electrons, the kinetic energy
operator for the nuclei written in terms of 3Nnu mass weighted momentum
operators P Q in a column 1 × 3Nnu matrix and its adjoint P †Q row matrix, with
M a 3Nnu × 3Nnu matrix of masses, and the Hamiltonian operator H Q for fixed
nuclear positions. The wavefunction must satisfy normalization conditions for
bound states or asymptotic conditions for scattering states. The solution to this
equation can best be found with a preliminary treatment for the nuclei at fixed
positions, leading to the equation
H Q Φ X; Q = E el
Q Φ X; Q
160 5 Intermolecular States

where the electronic wavefunction Φ(X; Q) depends parametrically on nuclear

position variables, and the electronic energy varies with those variables. This
seems to indicate that all the details of a many-electron wavefunction must
be known to obtain the system energy, but this is not the case since as shown
next all what is needed is knowledge of the electronic density.

5.1.2 Energy Density Functionals for Fixed Nuclei

Starting with the electronic number density function ρ r; Q per unit volume in
a molecular system with fixed nuclear positions Q, a knowledge of the electronic
cusp positions and isocontour slopes of the density provide enough information
to locate the nuclei and to find their nuclear charges. Therefore, the density
uniquely defines the Coulomb potential energy due to Nnu nuclei with charges
cα, α = 1 to Nnu, interacting with Nel electrons and among themselves, given by
the Hamiltonian terms in
H nu = H ne + H nn
and appearing in a Hamiltonian operator containing also the electronic kinetic
energy and electron–electron Coulomb interactions,
H Q = K el + H ee + H nu
Since the electronic kinetic and electron–electron repulsion terms are known,
we expect on physical grounds that knowing the electronic density of the ground
electronic state of the system will determine its Hamiltonian and therefore all its
properties. This can be proved by the following theorem and its corollary [2].
The theorem states that the electronic energy E(Q) of a nondegenerate ground
electronic state is a unique functional of the electronic density ρ. To reproduce
the proof, we introduce first the separation of electronic variables in the set X =
(xi, X(i)) for Nel electrons, the normalized electronic wavefunction Φ(X; Q), and
the electronic density as
∗ Nel
ρ r; Q = dX Φ X; Q i=1
δ r −r i Φ X; Q = Nel P r

2
Pr = ζi
dX i
Φ X; Q r i =r

and note that the Coulomb interaction between nuclei and electrons
1 Nnu Cα Ce
H ne = v ri , v ri = α=1
i
4πε0 Rα −r i
has an expectation value

H ne = d 3 r ρ r v r
 5.1 Molecular Energies for Fixed Nuclear Positions 161

To prove the theorem by contradiction, we assume that for a given density

ρ r , there may be two different nuclear potential energy operators H nu and
H nu , which differ only in the magnitudes of the nuclear charges and lead to dif-
ferent states and energies. We use the variational theorem that gives an upper
bound to the exact ground state energy, written for the primed quantities,
to find

E Q = Φ Q HQ Φ Q < Φ Q HQ Φ Q

E Q < ΦQ H Q − H nu + H nu ΦQ

E Q < E Q + d 3 r ρ r v r −v r + Hnn − Hnn

Reversing primed and unprimed quantities, the last three terms only change
sign and when we add the two expressions they cancel and we find
E Q +E Q <E Q +E Q
which is a contradiction resulting from the assumption that two different
energies could arise from the same density. This establishes the one-to-one
correspondence between the energy functional and the density. As a corollary,
we find that if E ρ r; Q is the ground state energy functional of the electronic
density for a given nuclear conformation, then minimizing this functional leads
to an upper bound to the exact ground state energy, which is yet a function of
the nuclear position variables.
The potential energy function for species A and B at relative position R is
obtained as

VAB Q = E ρ r; Q −E ρ r; Q R ∞

provided the energy functional has been constructed to satisfy E ρ r; Q R ∞ =

EA ρA r; Q + EB ρB r; Q .
This theorem justifies efforts to construct energy functionals of the electronic
density and to obtain from them upper bounds to the exact ground state energy
with a constrained variational procedure. A related alternative statement is that
the energy can be written as a functional of a wavefunction that has been con-
strained to give the electronic density while minimizing the sum of electronic
kinetic and Coulomb interaction energies [3]. This further justifies a procedure
based on construction of a wavefunction constrained to give the density of a
collection of independent electrons interacting in accordance with a Hamilto-
nian containing electron correlations [4]. Functionals can be constructed in dif-
ferent ways depending on whether atoms and molecules are chemically bonded
or not. Two ways of generating them are the Kohn–Sham procedure [4], which
162 5 Intermolecular States

makes use of a total density written as a sum of one-electron terms, and the
Gordon–Kim procedure [5], giving the total density as a sum of the densities
of molecular components of the whole system. Many details of the DFT
treatments can be found in reference [6].

5.1.3 Physical Contributions to the Energy Density Functional

The Hohenberg–Kohn decomposition of the molecular energy of electrons and
nuclei for fixed nuclear positions is [2]

E ρ = d3 r ρ r v r + F ρ + Hnn

Fρ = ΦQ K e + H ee ΦQ ρ
= VCoul + G ρ
1 3 3 Ce 2
VCoul = d rd r ρ r ρ r −δ r − r ρ r
2 4πε0 r −r
G ρ = Gkin ρ + Gxc ρ
where VCoul is the classical Coulomb energy of interaction among all the elec-
trons, a Hartree-like (purely electrostatic) energy, with the second term in VCoul
introduced to prevent electron–electron self-interactions. The two terms in G[ρ]
come from the electronic kinetic energy, and combined electronic exchange and
correlation energies, with the second term containing the difference between the
exact energy of the electron system and the Hartree energy. Once this functional is
constructed, one can proceed to minimize it for variations of the electron density
that conserve the number of electrons so that d 3 r ρ r = Nel . The variation of the
constrained functional E ρ − μ d 3 r ρ r , where μ has the meaning of a chemical
potential at absolute zero temperature, leads to the equation
δE
−μ = 0
δρ r
The minimum for E[ρ] provides an upper bound to the ground state elec-
tronic energy and the optimal density ρ r; μ .
In the Kohn–Sham (K–S) procedure, the density is written as a sum of terms
containing the spin-orbitals ψ j from each electron, for spin up and down vari-
ables ζ = (+, −),
Nel 2
−
ρr = ζ j=1
ψ j r, ζ =ρ +
r +ρ r

and these K–S functions are used to construct an antisymmetrized determinan-

tal (independent-electron) wavefunction Φ(KS)(X; Q) = X Φ(KS)(Q) for fixed
nuclear positions, from which single-electron kinetic energy and total func-
tionals can be calculated as
 5.2 General Properties of Potentials 163

Φ KS X; Q = C j
ψ j r j , ζj

Ks ρ = Φ KS Q K e Φ KS Q
ρ

F ρ = VCoul ρ + Ks ρ + Exc ρ
The last term is an exchange-correlation functional (which also includes a
contribution to correct the kinetic energy functional). This is parametrized
and used in calculations. Improvements follow using separate functionals for
up and down spins, needed to describe bonding between open shell species,
and using functionals of both the density and its gradient, to account for inho-
mogeneous charge distributions.
The K–S spin-orbitals are obtained solving a one-electron eigenvalue
equation

ℏ2 2
− ∇ + v KS r,ζ ψ j r,ζ = ϵj ψ j r,ζ
2me

where v KS r, ζ = δE δρ r,ζ is the K–S one-electron potential energy func-

tion. It is helpful to have in mind that the functional derivative of the energy
can be defined in integral form by E = d 3 r δρ r δE δρ r so that the physical
dimensions of δE δρ r are energy divided by density times volume element,
again an energy quantity.
The Gordon–Kim (G–K) procedure, suitable for interacting systems that do
not show chemical bonding, constructs the density of the pair AB from given
separate densities ρA r and ρB r , usually calculated at the Hartree–Fock level,
as ρ r = ρA r + ρB r , and obtains potential energy functions for short and
intermediate distances from functionals G[ρ] of this density. Extensions have
been introduced to also add long-range van der Waals interactions in terms
of additional density changes Δρ r generated by displacing individual density
components, and introducing restoring forces. More details of the K–S and
G–K procedures can be found in Chapter 6 on many-electron treatments.

5.2 General Properties of Potentials

5.2.1 The Electrostatic Force Theorem
The electrostatic force theorem shows that once the electronic charge distri-
bution is known from molecular (quantal) calculations, it is possible to calcu-
late the forces on nuclei using classical electrostatics, applied to a collection of
positive nuclear charges plus the negative charge distribution of the elec-
trons [7].
164 5 Intermolecular States

Consider two nuclei a and b on the z-axis, with relative position R, surrounded
by electrons in a state Φk with energy Ek. This can be a ground or excited
electronic state, and here we assume it is noninteracting with other states.
We find the force on nucleus a by differentiation of the expectation value of
the energy operator with respect to its position coordinate Za. We begin with
the Schrödinger equation and
∂ ∗
dX Φk X;R H R − Ek R Φk X;R = 0
∂Za
Expanding this expression, displaying the Coulomb terms in the Schrödinger
equation and identifying the force on a, we have with r 1a =r 1 −Ra ,
∂Ek Ca Cb Zab Ce ρ r 1 Ca z1a
Fa, z = − =− + d 3 r1
∂Za 4πε0 R3 3
4πε0 r1a
This is the electrostatic force theorem, formally identical to the corresponding
classical mechanical expression, but now constructed from a quantal electronic
density. The first term to the right is negative and shows the repulsive force on a
from the nucleus b; the integrand in the second term is the Coulomb force
between an electronic distribution at position r 1 and nucleus a, with positive
or negative integrand values depending on the location of the element of volume
d3r1 along the z-direction. If the element of volume of electrons is found
between two parallel planes perpendicular to the axis through the nuclei, then
it gives a positive force increment contributing to bonding. Far outside these
planes, the volume will oppose bonding. The first region is called the bonding
region, the second one is the antibonding region. They are separated by a sur-
face where the electronic force densities in Fμ, z = d 3 rfμ, z r at each nucleus μ =
a, b are equal, or
fa, z r = fb, z r

This defines boundary surfaces, B r = fa, z r −fb, z r = 0, showing the transi-

tion between an inner bonding region and two outer antibonding regions, one
for each nucleus. Given an electronic charge distribution, this tells us whether
the nuclei undergo a net attractive or repulsive force. The potential energy fol-
lows by integration of the relative force over the internuclear distance.

5.2.2 Electrostatic Forces from Approximate Wavefunctions

Electronic wavefunctions are usually known only within approximations
obtained from a variational procedure or from perturbation theory. In these
cases, the theorem must be applied considering the nature of the
approximations.
 5.2 General Properties of Potentials 165

For a trial wavefunction Φ X; R, λ dependent on a set of variational para-

meters λ = {λn}, the variational energy functional is

Φ R, λ = Φ R, λ H R Φ R, λ Φ R, λ Φ R, λ

optimized to obtain a minimal energy when the parameters are varied and cal-
culated at the minimum for each distance R. This gives an energy Emin R =
Φ R,λmin from which the force at nucleus a follows from the gradient as

∂Emin ∂Emin ∂Emin ∂λn

Fa , z = − =− −
∂Za ∂Za λmin
n ∂λn ∂Za
where the first term to the right gives the force from a frozen trial function,
while the second term gives an additional force contribution due to changes
of parameters with the nuclear distance. The parameters may include coeffi-
cients of atomic orbitals in molecular orbitals, giving a MO distortion force;
exponents in atomic orbitals, giving an AO distortion force; or coefficients in
a combination of electronic configurations, giving a configuration interac-
tion force.
When the electronic state Φ derives from a perturbation expansion, with the
energy given by a perturbation series E = E(0) + E(1) + E(2) + …, the force appears
0 1 2
also as a perturbation sum Fa, z = Fa, z + Fa, z + Fa, z + … obtained from the energy
gradient, with the leading term as in the theorem expression but with the unper-
turbed electronic density ρ 0 r 1 . Additional corrections involve density
changes ρ 1 r 1 , ρ 2 r 1 , … obtained from the perturbed wavefunction.

5.2.3 The Example of Hydrogenic Molecules

Consider a diatomic molecule with two electrons such as H2, HD, HHe+, …, or a
diatomic with only two active electrons outside closed shells of ion cores, such
as LiH, with charge numbers ζμ, μ = a, b, for the two cores. We can construct a
two-electron wavefunction to describe the whole range of distances R and with
the correct asymptotic form for R ∞. This can be done with molecular orbi-
tals (MOs) written as linear combinations of atomic orbitals (AOs) with variable
exponents. Two-electron determinants describe configurations and their com-
bination provide the required total wavefunction with the correct asymptotic
form. The electronic energy can be obtained from the expectation value of
the electronic Hamiltonian in a variational minimization of the function of
all the parameters, or fixing some parameters and varying the remaining ones.
The electrostatic force is then obtained as shown above from the derivatives of
all the varied parameters with respect to atomic distances.
166 5 Intermolecular States

The AO at center a in the two-electron systems contains an effective nuclear

charge as the distance-dependent exponential ζa(R), varying from the united
atom value for the diatomic to the value for the isolated atom. For a 1s orbital
−1
it is given by χ a r = ζ a a0 3 2 2 exp −2ζ a ra a0 2π 1 2 , where a0 is the
Bohr radius, and with ra the distance from the electron to nucleus a.
A molecular orbital formed as a linear combination of atomic orbitals
(a MO-LCAO) is φj r = caj χ a r + cbj χ b r , j = 1, 2, normalized to φj φj = 1
and with two variational parameters cμj(R) which can be obtained minimizing
the diatomic electronic energy for fixed nuclei. A two-electron determinantal
function is of form ΦJ(x1, x2) = |φjϑj(1)φkϑk(2)|, where ϑj is a spin state (α or β),
and a general two-electron state can be written as a superposition of configura-
tions (a configuration interaction, CI)

Φ CI x1 , x2 = J
AJ ΦJ x1 , x2

with all terms of the same space and spin symmetry. The energy follows from
the variational functional Φ CI = E ζ μ ,cμj ,AJ and forces are given by

∂E ∂E ∂ζμ ∂E ∂cμj ∂E ∂AJ

Fa, z = − − − −
∂Za λ
μ ∂ζ
μ ∂Za μj ∂Za
μj ∂c J ∂AJ ∂Za

with λ = {ζ μ, cμj, AJ} the set of variational parameters. This expression shows how
the forces are affected by changing AO exponents, MO coefficients, and coeffi-
cients of CIs when these parameters have not been optimized by minimization
of the energy. If instead they have been optimized, then the last three summa-
tions are zero and the force is given by the gradient of the energy calculated at
the optimized parameter values.

5.2.4 The Virial Theorem

When interaction among particles are restricted to Coulomb forces, it is possi-
ble to find general relations between expectation values of kinetic and potential
energies, known as virial theorems, applicable to situations where a physical sys-
tem is in a stationary state. They have been applied to treatments of chemical
bonding for cases where interatomic distances are fixed, or where the system is
at equilibrium in an energy minimum.
The starting point is the Hellmann–Feynman theorem which follows imme-
diately from the Schrödinger equation for a normalized quantum state Ψ(λ)
dependent on a free parameter λ and satisfying the equation H λ Ψ λ =
E λ Ψ λ for a molecular Hamiltonian H λ and with E(λ) the eigenenergy.
Differentiation of both sides with respect to λ, projecting on the left into Ψ(λ) ,
and using the equation and its adjoint, gives the Hellmann–Feynman relation
 5.2 General Properties of Potentials 167

∂E ∂H
= Ψλ Ψλ
∂λ ∂λ
kin Coul
We consider a special case where the Hamiltonian is H el = H el + H el , a
sum of electronic kinetic and Coulomb electron–electron and electron–nucleus
Coul
energies in H el = H ee + H en , for fixed nuclear positions and excluding the
Coulomb interaction among nuclei. The corresponding electronic eigenstate
is Φel(R) with eigenenergy Eel(R) for a relative position vector R = Rnz along
the z-axis. A parameter λ is introduced to scale the position coordinates of elec-
trons, but not of the nuclei, writing λr j =sj for electron j and changing integra-
tion variables in expectation values to the set sj . The electronic energy
kin Coul
becomes Eel λ = λ2 H el R + λ H el λR R and differentiation with respect
to the parameter λ, using that

∂ 1 ∂ 1 ∂ λR
=
∂λ s− λRnz ∂ λR s−λRnz ∂λ

in the expectation value for the Coulomb Hamiltonian, gives for λ = 1 a virial
theorem valid for any fixed internuclear distance,
kin Coul
2 H el + H el R = − R∂Eel ∂R
R R

where the Hellmann–Feynman theorem has been used for the right-hand side.
kin Coul
This can be combined with H el R + H el R R = Eel R to obtain expecta-
tion values of separate electronic kinetic and Coulomb energies,
kin
H el = −R∂Eel ∂R −Eel R
R
Coul
H el R = 2Eel R + R∂Eel ∂R
R

showing how they depend on the internuclear distance. At equilibrium for the
minimum of the electronic energy, where ∂Eel/∂R = 0, one recovers the virial
equality for Coulomb interactions,
kin Coul
2 H el = − H el Rmin
min

relating the electronic kinetic and Coulomb energies for equilibrium nuclear
positions. The expectation values for kinetic and Coulomb energies obtained
from the virial relations, in the case of variable distances, are more accurate than
their values when calculated from variationally optimized state averages. This
follows because the errors in the wavefunction obtained in a variational calcu-
lation are of first order with respect to the exact wavefunction, but errors are
168 5 Intermolecular States

only of second order for the optimized variational energy that appears in the
virial relations.
The derivation for a diatomic system can easily be generalized to a polyatomic
system where atoms are located at positions Ra , a = 1 to N, in which case the
virial relation is
kin Coul
2 H el + H el R =− R
a a
∂Eel ∂Ra
R R

with R = Ra the set of all atomic positions, and with electronic energy gradi-
ents calculated for each atomic displacement.
When the exact wavefunction is not known (as usual), then it is necessary
to instead differentiate the previously introduced functional Φ R, λ .
Assuming that the approximate wavefunction has been normalized to
Φ R,λ Φ R, λ = 1, one finds that

∂ ∂H ∂Φ ∂Φ
= Φλ Φλ + H λ Φλ + Φλ H λ
∂λ ∂λ ∂λ ∂λ
which can again be analyzed in terms of scaled kinetic and Coulomb compo-
nents but with results depending now on how the states Φ depend on the
parameter λ by construction.

5.2.5 Integral Form of the Virial Theorem

The Hellmann–Feynman theorem can be used in its integral version to obtain
changes in a total energy due to changes in the Hamiltonian operator. Let the
Hamiltonian between interacting species A and B be written as H λ =
H A + H B + λH AB . Then provided the state Ψ(λ) is an exact eigenstate with energy
E λ = Ψ λ H λ Ψ λ , one finds that ∂E ∂λ = Ψ λ H AB Ψ λ and
integrating,
1
ΔE = E −E0 = dλ Ψ λ H AB Ψ λ
0

for the energy change due to the interaction of A and B, with E(1) = E and
E(0) = E0.
However, when the state Ψ(λ) is only approximately known as Φ(λ) from a
variational or perturbative procedure, the relevant derivative ∂ ∂λ contains
additional terms and the integral form becomes
1
∂Φ ∂Φ
Δ = dλ Φ λ H AB Φ λ + H λ Φλ + Φλ H λ
0 ∂λ ∂λ
 5.3 Molecular States for Moving Nuclei 169

which cannot be further generally transformed insofar as ∂Φ/∂λ depends on

how the wavefunction was constructed. In the simplest special case, with the
approximation Φ = AB ΦA ΦB , the antisymmetrized product of normalized
noninteracting states of A and B, the result is Δ = AB − A − B =
AB ΦA ΦBA H AB AB ΦA ΦB , as expected.
The integral expressions can also be used when the wavefunction has been
obtained from a density functional approximation, in which case the integrals
become functionals of the electron density and of its derivative with respect to
the parameter λ.

5.3 Molecular States for Moving Nuclei

5.3.1 Expansion in an Electronic Basis Set
A complete description of molecular states involves moving nuclei. These must
be rigorously treated using quantum mechanics and wavefunctions dependent
on the nuclear variables. The full Hamiltonian operator must contain also the
kinetic energy operator for the nuclei, and one must allow for the interaction of
electronic and nuclear degrees of freedom due to moving nuclei [8–14].
We work in a reference frame with space-fixed axes and its origin at the center
of mass of the nuclei, assuming that electron masses are negligible (the center of
mass of the nuclei frame CMN-SF). To simplify matters, we work here with only
two nuclei a and b of masses ma and mb. Their reduced mass is M = mamb/
(ma + mb) and their relative position is R. The set of electron variables is given
by X = r i , ζi in terms of the position vector and spin variables of electrons i.
Molecular states are described by solutions of the Schrödinger equation

HΨ = EΨ

where H, E, and Ψ are the Hamiltonian operator, energy, and state of the whole
system of electrons and nuclei. The state Ψ must be normalized to describe a
bound system, or must satisfy asymptotic boundary conditions for an unbound
system at large R.
The Hamiltonian operator can be separated into the kinetic energy operator
of the relative nuclear motion plus the Hamiltonian H R for fixed nuclei, with the
latter including the electronic kinetic energy, the Coulomb attraction of elec-
trons to nuclei, the Coulomb repulsion among electrons and the repulsion
among the nuclei, as
ℏ2 2
H=− ∇ + HR
2M R
H R = K e + H ee + H en + H nn
170 5 Intermolecular States

If some of the electrons are found to form rigid ion cores, then the electron-
nucleus interaction must be replaced by an electron-core interaction which is
Coulombic at large distance and repulsive at short distance due to electronic
exclusion; similarly the nucleus–nucleus interaction must be replaced by the
core–core interaction.
Many-electron states of molecules can be constructed from many-electron
atomic states, in an approach known as the valence-bond (VB) method, or from
atomic orbitals combined to form molecular orbitals, in the so-called molecular
orbital (MO) method, as we have seen with the example of H2 in the previous
Chapter 3.
The molecular wavefunction Ψ can be expanded in a basis of orthonormal
electronic states Φk which are functions of the electronic variables and depend
parametrically on the nuclear positions. Using the Dirac notation for electronic
states we write Φk X;R = X Φk R and work in a linear space of K states,
introducing an orthonormal and complete basis set of such states for fixed
nuclei, the expansion is
K
Ψ R,X = k =1
Φk X;R ψ k R

Φk R Φl R = δkl
K
k =1
Φk R Φk R =I

The coefficients ψ k are functions of the nuclear positions that describe the
nuclear motions in the electronic states k. The bracket notation here and in
what follows signifies integration and sum over all the electron variables, for
fixed nuclei as shown. The last line gives an assumed completeness relation,
with I the identity operator.
Several choices are possible for the basis set of electronic states, which lead to
different electronic representations of potential energies and nuclear ampli-
tudes. These representations are related by matrix transformations which can
be conveniently introduced to simplify treatments of molecular spectroscopy
and dynamics [15–19].

5.3.2 Matrix Equations for Nuclear Amplitudes in Electronic States

Introducing a matrix notation with rows and columns shown as boldface sym-
bols, we have that

Ψ R = Φ R ψ R

where the first factor is a 1 × K row of states Φk and the second factor is a K × 1
column of amplitudes ψ k. The Schrödinger equation can now be put in a matrix
 5.3 Molecular States for Moving Nuclei 171

form. Operating on Ψ with the momentum operator PR = ℏ i ∇R , we find a

sum of two terms, and using the completeness relation of the electronic basis
in front of the second term, we have
ℏ ℏ ℏ
∇R Ψ R = Φ R ∇R ψ R + ∇R Φ R ψ R
i i i
ℏ
= Φ R I ∇R + G R ψ R
i
ℏ
G R = Φ R ∇R Φ R
i

where I is the identity matrix. The symbol Φ R represents a K × 1 column

matrix with the complex conjugate elements Φ∗k . The elements Gkl of the
momentum-coupling matrix G are vectors that have the meaning of momen-
tum fluctuations resulting from the electronic transitions induced by nuclear
displacements. This matrix is Hermitian, as can be seen by taking the gradient
of the normalization condition for the electronic basis, giving
ℏ ℏ ∗ ∗
Gkl R = Φk R ∇R Φl R = − Φl R ∇R Φk R =Glk R
i i
Its diagonal elements are null when the electronic functions are real valued, as
follows from ∇R Φk R Φk R = 0, but they are not necessarily zero when
the functions are complex-valued.
Since the effect of operating with the gradient is to introduce a square matrix
with vector-valued elements between the two factors making Ψ, operating twice
with the gradient simply repeats the effect and one ends up with the vector
product of two square matrices; the elements of this product matrix are scalar
products of two vectors. The original Schrödinger equations takes the
matrix form
2
1 ℏ
I ∇R + G R + HR − IE ψ R = 0
2M i
HR = Φ R H R Φ R

where the second power of the square bracket signifies a scalar product of matri-
ces with vector-valued elements. This is a set of coupled differential equations
that must be solved with the appropriate boundary conditions for the ψ ampli-
tudes, for bound or unbound (decaying or scattering) states.
It is of interest to analyze the vector product of the total momentum appear-
ing in this equation. Expanding it,
2
ℏ ℏ
I ∇R + G R ψ R = −ℏ2 ∇R 2 + 2G ∇R + ∇R G + G G ψ R
i i
172 5 Intermolecular States

we find that the two terms in the square bracket to the right add up to a Her-
mitian operator but each of them is, separately, non-Hermitian. Hence, it is safer
to keep them together as a sum.

5.3.3 The Flux Function and Conservation of Probability

†
The total nuclear density at position R is given by ψ R ψ R , with the dagger
meaning transpose and conjugate, which is time-independent. Consequently,
one can define a matrix flux function J(R) whose divergence is zero, indicating
conservation of the nuclear density at each position. To do this, first write the
matrix differential equation for ψ R in the form DR ψ R = 0, where DR is a
†
matrix differential operator. Then multiplying to the left by ψ R and rearran-
ging the equation
† †
ψ R DR ψ R − DR ψ R ψ R =0

with help of the relation f ∇R2g = ∇R. (f ∇Rg) − (∇Rf ). ∇Rg and cancelling oppo-
site terms, we find the equation for the divergence of the intermolecular flux
matrix J R , with vector valued elements J kl , as

∇R J R = 0

1ℏ † 1
J R = ψ ∇R ψ − ∇R ψ † ψ + ψ † Gψ
Mi M

showing that probability is conserved, provided however that one must now
include a contribution from the electronic momentum fluctuation in the defi-
nition of the flux.

5.4 Electronic Representations

5.4.1 The Adiabatic Representation
a
In this representation, we choose states Φk = Φk so that they diagonalize the
Hamiltonian for fixed nuclei,
a a a
Φk H R Φl = Ek R δkl
 5.4 Electronic Representations 173

where we have specified the representation with a superscript a. The total

molecular wavefunction is now given by
K a a
Ψ R, X = k =1
Φk X;R ψ k R

and the nuclear amplitudes satisfy corresponding matrix equations with the
momentum and Hamiltonian matrices G a , HR calculated in the adiabatic
a

representation. The second one is diagonal by construction of the basis set.

5.4.2 Hamiltonian and Momentum Couplings from Approximate

Adiabatic Wavefunctions
The adiabatic wavefunctions can be approximated by expanding them in a given
basis set of ND many-electron determinantal functions DJ X;R = X DJ R ,
J = 1 to ND, generally dependent on the intermolecular position variables, with
the expansion coefficients determined by a variational procedure. We have that
a
Φk R = J
DJ R CJk R = D R C k R

C †k Δ C l = δkl

where D is a 1 × ND row matrix and C k R is a ND × 1 column matrix repre-

senting the adiabatic functions, with elements dependent on the intermolecular
position variables, while Δ R = D D indicates the overlap matrix of the
given basis. The coefficients follow from the diagonalization of the electronic
Hamiltonian in the original basis,
a
H R C k R = Δ R C k R Ek R

and the matrix of the electronic Hamiltonian is diagonal because

Hkl = C †k HR C l = Ek δkl . The momentum coupling matrix has elements
a a

ℏ †
C l + C †k Δ ∇R C l
a
G kl = C Dk R ∇R Dl R
i k
where the first term describes the momentum coupling contributed by the dis-
placement of the basis functions, and the second term gives the contribution by
the changes of coefficients with positions. The gradients of coefficients in the
second term can be obtained by differentiation of the secular equation for
the coefficients, as
−1
a a
∇R C l = H R − El I ∇R H R − El I Cl

which only involves gradients of adiabatic potential energies.

174 5 Intermolecular States

The equation of motion for the nuclear amplitudes in the adiabatic represen-
tation, with a diagonal matrix of adiabatic energy potentials E a R =
a
Ek R δkl , is

2
1 ℏ
I ∇R + G a
R +E a
R − IE ψ a
R =0
2M i

which shows that the amplitudes are interacting only through the momentum
couplings. Ways to generate nonadiabatic representations have been given in
the literature [12, 13, 15, 16, 20].

5.4.3 Nonadiabatic Representations

Nonadiabatic representations can be defined in a variety of ways. A specific
choice is the strictly diabatic or P-representation where new electronic states
d
Φk R are constructed from given ND basis functions such that

d d
Φk ∇R Φl = δkl

so that there is no coupling of different electronic states due to displacements of

nuclear positions. These new states can be related to the previous adiabatic ones
by means of a ND × ND matrix transformation A R . At a large intermolecular
distance R0, where neither electronic nor momentum interactions are present,
both representations must coincide. We write in matrix notation the row
equation

Φd R = Φa R A R;R0

with A R0 ;R0 = I and the unitarity condition A† = A−1. From the condition
defining the diabatic basis set, we have the differential equation
ℏ
∇R A R;R0 + G a
R A R;R0 = 0
i
which can be solved to obtain the transformation matrix from given adiabatic
momentum couplings. The total molecular wavefunction in the new basis set is
M d d
Ψ R,X = k =1
Φk X;R ψ k R = X Φd R ψ d
R

with nuclear motion amplitudes satisfying

ℏ2 d
− ∇R 2 + H R −IE ψ d
R =0
2M
 5.4 Electronic Representations 175

which shows them coupled through the nondiagonal electronic matrix

H R = A† H R A. These couplings are typically smoother that the ones arising
d d

from momentum couplings so that the diabatic representation is frequently

used to treat electronic rearrangements in molecular spectroscopy and
collisions, usually introducing semiclassical approximations for the nuclear
amplitudes [19, 21]. The transformation between adiabatic and diabatic
representations in a space of ND dimensions can be generally treated introdu-
cing an algebra of skew symmetric matrices [22].

5.4.4 The Two-state Case

A solution to the differential equation for A at all distances can be easily
obtained for a simple case with two real-valued coupled electronic
a
states, Φj , j = 1,2, and only one radial variable. Introducing the gradient cou-
a a
pling in the adiabatic representation, τ a R = Φ1 ∇R Φ2 , the momentum
a
coupling matrix has diagonal elements Gjj = 0 and off-diagonal ones
a a∗ a a a a
G12 = G21 = ℏ i τ a and by construction H12 = H21 = 0, Hjj R = Ej R .
We introduce a 2 × 2 transformation A(R; R0) for a large R0, where there are
no couplings between states 1 and 2, such that A(R0; R0) = I, the identity matrix.
When the two states and therefore τ(a) are real valued, one can choose

cos γ sin γ
A R; R0 =
−sin γ cos γ
It follows from the differential equation for A(R; R0) that the angle γ(R; R0) is a
function satisfying the differential equation dγ/dR = τ(a)(R) with the boundary
condition γ(R0; R0) = 0. Its solution
R
γ R; R0 = dR τ a R
R0

can be used to construct the Hamiltonian matrix elements in the strictly dia-
d
batic representation, where Gjk = 0 for all elements, as
d a a
H11 R = E1 R cos2 γ + E2 R sin2 γ ,
d a a
H22 R = E1 R sin2 γ + E2 R cos2 γ ,
d a a
H12 R = E1 R − E2 R sin γ cos γ

These expressions allow us to consider what happens when the adiabatic

potentials cross at a point R× < R0, with R0 large and τ(a)(R) negligible there.
d a d a
Then for large R, one finds H11 R = E1 R and H22 R = E2 R . Around
176 5 Intermolecular States

d
the crossing point, H12 R is small but the diagonal elements do not cross, and
for even shorter distances, the diabatic elements differ significantly from the
adiabatic energies.
d
We also find that the diagonal diabatic potential energies Hjj R will cross
whenever γ = ± (n/2 + 1/4)π and that this can happen more than once on the
way to small R if the adiabatic momentum couplings are large. This is not
physically meaningful and is a consequence of the definition of the strictly
diabatic representation. An alternative more physically meaningful diabatic rep-
resentation can be introduced by parametrizing the transformation angle to a
form γ(R; α) with the parameter α chosen to minimize the momentum coupling
in a new “m” representation [18]. This generates new smooth diabatic potentials
m
Hjk R and a new smaller momentum coupling τ(m)(R; α) which can be used
for molecular spectroscopy and collision treatments in the new representation.
Figure 5.3 shows an example constructed from Morse type potentials.
In any case, the new states are related to adiabatic ones through the transfor-
mation matrix as

d a a
Φ1 R = Φ1 R cos γ − Φ2 R sin γ

d a a
Φ2 R = Φ1 R sin γ + Φ2 R cos γ

This can be generalized when the coupling τ(a)(R) = τ(a)(R) exp[iχ(R)] is a

complex function, writing instead the diabatic states as combinations of the adi-
a a
abatic basis functions Φ1 R exp −iχ R 2 and Φ2 R exp iχ R 2 .
Extensions of the treatment to several variables and more than two electronic
states have been presented in the literature [14]. For example the NaH molecule
in its ground and excited 1Σ+ states shows multiple pairs of avoided crossings,
one for each excited potential energy that breaks into Na(nl) + H(1s) after it
pseudo-crosses the Na+ + H− attractive potential energy as seen in
Figure 5.4 [23].

5.4.5 The Fixed-nuclei, Adiabatic, and Condon Approximations

The previous description of alternative representations can give accurate results
for properties such as molecular spectra and scattering cross sections by
introducing a large number ND of functions in the basis set of many-electron
functions Φk X; R and by calculating nuclear amplitudes ψ k R as solutions
of coupled differential equations. However, in many cases, it is sufficient to
introduce approximations where some of the momentum couplings are
neglected.
 5.4 Electronic Representations 177

(a)
–0.11

–0.12

–0.13
V (a.u.)

–0.14

–0.15

–0.16
1.8 2.0 2.2 2.4
R (a.u.)

(b)
0.005 10.0

τa
0.004 8.0

0.003 6.0

0.002 4.0
τa (a.u.)
V (a.u.)

0.001 2.0

0.0 0.0
vd12

–0.001
vm
12
–0.002

–0.003
1.6 2.0 2.4 2.8
R (a.u.)

Figure 5.3 (a) Adiabatic (full line), diabatic (dash line) and minimum coupling (dash-dot line)
potential energies for two states and (b) their couplings in the three representations. Source:
from Ref. [18]. Reproduced with permission of John Wiley & Sons.
178 5 Intermolecular States

E (hartree)

Na H 1Σ+
162.25
Na++ H–

Na(4p) + H(1s)
0.30 Na(3d) + H(1s)
Na(4s) + H(1s)

Na(3p) + H(1s)
162.35

0.40
Na(3s) + H(1s)

162.45

0.5 1.0 1.5 logR

2 5 10 20 50 100 R(bohr)

Figure 5.4 NaH potential energies showing multiple pairs of avoided crossings. Source: from
Ref. [23]. Reproduced with permission of John Wiley & Sons.

The fixed-nuclei approximation (also called the Born–Oppenheimer approx-

imation) starts with the adiabatic representation and sets all the momentum
a
couplings G kl = 0 so that the equation for the nuclear motion in each electronic
state k is
ℏ2 2 a a
− ∇ + Ek R − E ψ k R = 0
2M R
In the adiabatic approximation, one again starts with the adiabatic represen-
tation, but one keeps only one term in the expansion of the total molecular
wavefunction so that
a a
Ψ ad R,X = Φk X;R ψ k R

a a a
with Φk a real-valued normalized function for which Φk ∇R Φk = 0.
When this is replaced in the Schrödinger equation one finds
 5.4 Electronic Representations 179

ℏ2 a a a
− ∇R 2 + Ek R + Wk R − E ψ k R = 0
2M
with

a ℏ2 a a ℏ2 a a
Wk R =− Φk R ∇2R Φk R = ∇R Φk R ∇ R Φk R
2M 2M
a repulsive potential energy addition, with the last equality obtained by partial
integration. Comparisons of the magnitudes of these approximations for the
energy of the ground molecular state of hydrogen molecule isotopes can be
found in [24]. There one finds that the ground state energy E0 of the X 1 Σg+
molecular term for H2, in cm–1 for E0/(hc) is 36 112.2, 36 118.0, and 36 114.7
for the fixed-nuclei, adiabatic, and nonadiabatic approximations, to be com-
pared with the experimental value of 36 113.6 cm–1. These differences are typ-
ical of corrections to equilibrium energy values. Considering that 1.0 cm–1 =
1.9864 × 10–23 J in the SI units and 1.0 cm–1 = 4.55634 × 10−6 au, it is found that
the nonadiabatic corrections are meaningful in very high precision spectros-
copy, but not essential for most work on intermolecular forces around equilib-
rium structures. The nonadiabatic couplings are however essential in many
treatments of photoinduced or collision-induced electronic rearrangement.
Another nonadiabatic approximation useful in molecular spectroscopy, for a
molecular system where internal variables remain near their equilibrium values
Qeq in the initial electronic state, is the so-called Condon approximation where a
C
basis set of electronic functions Φk X; Qeq is used in an expansion of the
molecular state. Here all the expansion functions have distances kept constant
at the equilibrium values Qeq of internal coordinates, and the variation of ener-
gies with displacements is described expanding the electronic Hamiltonian as a
function of displacements around the equilibrium conformation. This generates
a vibronic Hamiltonian displaying couplings of electronic and nuclear degrees
of freedom in molecular dynamics.
Given the set of coordinates Q = {Qν}, the total molecular wavefunction is
expanded in the basis set of Condon states as
K C C
Ψ Q, X = k =1
Φk X; Qeq ψ k Q,

while the electronic Hamiltonian is expanded around equilibrium values of

internal variables in powers of displacements q = Q − Qeq, giving
H Q = H eq −F eq q + q† G eq q 2

in a matrix notation where q is a column matrix, F eq is a row matrix of force

operators and G eq is a square matrix of potential energy curvatures, also opera-
tors. A vibronic description of dynamics can be based on a perturbation
180 5 Intermolecular States

C
calculation of the coefficient functions ψ k , starting with an expansion of the
C
equation for Ψ(Q, X) in the basis set Φk X; Qeq . This leads to equations
μ
containing force matrix elements Fkl multiplying the qμ displacements and cur-
μν
vature matrix elements Gkl multiplying the products qμqν.
For a simple case of a diatomic, this amounts to the expansion
K C C
Ψ R, X = k =1
Φk X;Req ψ k R

and calculations of molecular energies as functions of the displacement x = R −

Req doing perturbation theory with the Hamiltonian perturbation operator
∂H R ∂R eq x. This leads to vibronic coupling among electronic-nuclear states,
C C x
with electronic coupling matrix elements Φk ∂H R ∂R eq
x Φl = −Fkl x
C C
and corresponding coupling of nuclear motion states through ψ k x ψl .
x
The transition force is different from zero only when the product of point
Fkl
group symmetry representations of states k and l, and of x, contains the totally
symmetric representation.
The interaction of the molecular system with a light field coupled to the
molecular electric dipole leads to spectral intensities calculated in terms of
C C C C C C
Frank–Condon overlaps Φk ψ k D Φl ψl ≈Dkl Req ψ k ψl , where
C C
D is the electric dipole vector operator and Dkl Req = D Φl eq . When Φk
the photoexcitation occurs around equilibrium, the vibronic displacement fac-
C C C
tor ψ k x ψ l is small for k l and one can use amplitudes ψ k R
obtained from a Schrödinger equation with a potential energy function
C
Hkk R to obtain light absorption-emission rates.

5.5 Electronic Rearrangement for Changing

Conformations
5.5.1 Construction of Molecular Electronic States from Atomic States:
Multistate Cases
Given the state Ψ(Q, X) of a pair of molecules A and B as an expansion in a basis
set of electronic wavefunctions Φk(X; Q) (not necessarily adiabatic ones) which
vary with a set Q = R, Q of interatomic distances, its Schrödinger equation
contains a potential energy matrix with elements Hkl(Q) that are functions of
distances, as well as a matrix of momentum coupling elements Gkl(Q). It is
 5.5 Electronic Rearrangement for Changing Conformations 181

usually possible to construct the wavefunction of the pair of molecules so that at

large intermolecular distances, the molecular motions are independent and
occur along well-defined separate electronic potential energy surfaces
B
EκA QA and Eλ QB . However, as distances are decreased, the potential
energy for each electronic state k originating in (κ, λ) becomes a function
AB
Ek R,Qint , which changes with the intermolecular distance and internal
positions, excluding overall translations and rotations, and generates a potential
energy surface (PES) showing as isocontours of energy in the multidimensional
space of distances. Pairs of such PESs may show crossings or regions of minimal
energy differences, which depend on what electronic representation is being
used. They interact through nuclear momentum couplings, which also depend
on the representation.
Special points can be identified looking for minimal values of differences in
each pair of potential surfaces, as one varies the internal coordinates, so that
AB AB
Ek R,Qint −El R,Qint = Min. This can be zero if there is a crossing
along a surface in the hyperspace of internal coordinates, and its shape depends
again on what electronic representation one is using. When dealing with a dia-
tomic (and a single internal distance) one may find a crossing or single inter-
section at a distance R0 ; for two or more internal variables one may find a
conical intersection, crossing seams, or near crossings. All these are important
in descriptions of electronic rearrangement.

5.5.2 The Noncrossing Rule

The procedure for calculating intermolecular states by first fixing nuclear
positions leads to the construction of many-electron states Φk dependent
parametrically on all the internal nuclear position variables. Starting with
the total number 3N of atomic position variables for a system with N atoms,
and indicating with QRB the set of rigid body variables (six for overall transla-
tions and rotations, or five for a linear conformation), the remaining internal
variables form a set Qint = {Qν, ν = 1 to Nint} of Nint = 3N − NRB degrees of free-
dom. Matrix elements Hkl of the Hamiltonian operator for fixed nuclei, in an
original electronic basis set {Φk} which may describe adiabatic or nonadiabatic
electronic states and is taken to be orthonormalized, also depend on the
nuclear variables. As they vary the diagonal elements Hkk and Hll may cross
or show avoided crossings at specific points or lines. These locations are
important for the description of molecular spectra and interaction dynamics,
showing the structure conformations where light absorption-emission may
happen or where atomic rearrangement occurs for electronic transitions
between k and l.
182 5 Intermolecular States

General statements can be made about the location and nature of interaction
energies at these special points. To simplify, we consider first the case with only
one internal degree of freedom, for two atoms at the internuclear distance R,
and where only two states k = 1, 2 are involved. Examples are the lowest two1Σ
states of LiF or of NaH, where ionic electronic distributions give states of lower
energy around their ground equilibrium distance, but covalent distributions
give lower energy states at larger distances.
A special point Rc of a crossing or avoided crossing is located by finding where
the difference H11(R) − H22(R) shows a minimum as R is varied. Writing R =
c
Rc + x and Hkl Rc = Hkl , matrix elements can be expanded around x = 0 as

c ∂Hkl 1 2 ∂ 2 Hkl
Hkl x = Hkl + x + x +
∂Rc 2 ∂Rc 2
c c
with H11 = H22 if there is a crossing when working with the original basis set.
c
A nondiagonal Hamiltonian matrix element H12 appears if the representation is
nonadiabatic. Near the special point, an electronically adiabatic representation
a
is generated by the linear combinations Φk = Φ1 C1k + Φ2 C2k with the coeffi-
cients obtained by diagonalizing the Hamiltonian matrix at each R so that
a a
the new elements are Hkl R = δkl Ek x . These two energies are given by
the well-known roots of the second-order equation det[H − EI] = 0 where 2 ×
2 matrices are shown in boldface, as
a 1 1 2 2 1 2
E1, 2 x = H11 x + H22 x ± H11 x −H22 x + 4 H12 x
2 2
and become equal only when the radical L(x) = [H11(x) − H22(x)]2 + 4|H12(x)|2 = 0,
or when both H11(x) − H22(x) = 0 and H12(x) = 0. These are two equations in x to
be satisfied by a single solution root x0, and there is no real valued answer, except if
in a particular system one of the two equations is already obeyed for all x, such as
when the states 1 and 2 are degenerate and H11(x) = H22(x). For two different
diagonal element functions of x and in the absence of accidental vanishing of their
a
coupling H12, the two adiabatic energies E1, 2 x are different for all R = Rc + x
around x = 0. This is the noncrossing rule for adiabatic potential energies.
In this case, the potentials show an avoided crossing and the difference
a a
between the new potential energies, ΔE x = E1 x − E2 x , has a minimum
at the gap located at xm, where ∂(ΔE)/∂x = 0, equal to
2 2 1 2
ΔEm = ΔHm + 4 Wm
with ΔHm = H11(xm) − H22(xm) and Wm = H12(xm). Introducing derivatives
c c a
fkl = ∂Hkl ∂Rc and gkl = ∂ 2 Hkl ∂R2c , the energies E1, 2 x can be expanded up
to second order in x and are found to be opposing parabolas glancing at xm.
 5.5 Electronic Rearrangement for Changing Conformations 183

The momentum coupling between the two adiabatic states is a special case of the
a
previous relation described for approximate adiabatic states, here Φk R =
Φ R C k R in matrix notation with Ck a 2 × 1 column matrix, giving
ℏ †
C Φ R ∇R Φ R C 2 + C †1 Φ R Φ R
a
G12 = ∇R C 2
i 1
−1
a a a
∇R C 2 = E1 − E2 ∇R H R − E2 I C2

The denominator in the second line has a minimum value ΔEa at the avoided
crossing, showing that this is where the momentum coupling will be large.
Conditions under which H12(x) ≡ 0 give situations with crossings and can be
found from considerations of state symmetry. This happens if: (i) The two
original states have different electronic spin quantum numbers; or (ii) the
two original states belong to different irreducible representations of their point
group symmetry. The groups are ∞ h for homonuclear diatomics and C∞v for
heteronuclear diatomics. When the symmetry species (or labels) of the diatomic
states are different, then H12(x) ≡ 0, and if the Ek(R) = Hkk(x) are equal at a real
valued x0 in the original representation, it follows that a crossing of adiabatic
a
potential energies E1, 2 may occur at a distance R = Rc + x0.
If instead the two states have the same spin quantum numbers and point
group symmetry, then in general H12(x) 0. In this case, to find the root where
adiabatic energies are equal one must require that the radical L(x) = L0 + xL1 +
x2L2 = 0, giving in general a complex-valued root x0 + ix0 , and a crossing in the
complex plane of position variables.
For a complex valued coupling H12(x) = H12(x) exp[iα(x)], the two new adi-
abatic states around an avoided crossing are [25]
a iα iα
Φ1 R = Φ1 R exp − cos β + Φ2 R exp sin β
2 2
a iα iα
Φ2 R = − Φ1 R exp − sin β + Φ2 R exp cos β
2 2
with
tan 2β x = 2 H12 x H11 x − H22 x
taking a large value at the avoided crossing, leading there to a rapid change of
β(x) and rapid switch between states.
This analysis can be extended to cover interactions involving a spin-orbit
couplings in a Hamiltonian, in which case the relevant point symmetry groups
are extended groups with representations generated by the direct product of
spin dependent and space-dependent basis functions. In particular double
groups arise when orbital angular momentum states are extended to total
angular momentum states to include half-integer spin states, and their
184 5 Intermolecular States

representation labels can be used to derive conclusions from molecular sym-

metry arguments [26].

5.5.3 Crossings in Several Dimensions: Conical Intersections

and Seams
The search for special intersections of potential energy surfaces (PESs) can be
extended to systems with three or more atoms, involving more internal degrees
of freedom. Even for three atoms, there are Nint = 3 or 4 degrees of freedom
for the atomic positions and enough variables to satisfy the two equations that
indicate an intersection of PESs. This fixes only two internal degrees of freedom.
However, there are many rearrangements of atoms during electronic excitation
where most degrees of freedom remain inactive and only two are changing.
The two changing ones can be identified by searching for the configurations,
where H11({Qν}) − H22({Qν}) is a minimum for a set Qνm and where distor-
m
tion forces for two of them, called here Qa and Qb, are small around Qa and
m m
Qb . We first treat this case, with displacement variables x = Qa − Qa and
m
y = Qb − Qb , and all other internal variables set at their values for the minimum
and omitted from arguments for now. Consider again just two original states
Φj in a diabatic representation with diagonal Hamiltonian matrix elements
H11(x, y) and H22(x, y), and a nondiagonal H12(x, y) real valued for now but taken
to be complex later on. A transformation to the adiabatic representation states
a
Φk gives the potential energy surfaces

E1, 2 x, y = M x, y ± N x,y
1
M x,y = H11 x, y + H22 x, y
2
2 2 1 2
N x,y = H11 x,y − H22 x, y + 4 H12 x,y
The condition for intersection of PESs is N(x, y) = 0, or both H11(x, y) − H22(x, y)
= 0 and H12(x, y) = 0, which may be satisfied by roots (x0, y0), corresponding
to variables Qν0 expected to be close to Qνm , ν = a, b. When this happens for
the original (diabatic set) matrix elements, the new energies E1,2(x, y) of the
adiabatic states can be expanded around the roots up to second order in the
displacements so that

E1, 2 x, y = M 0, 0 + M 1, 0 x + M 0, 1 y + M 2, 0 x2 2 + M 0, 2 y2 2
+ M 1, 1 xy N 1, 0 x + N 0, 1 y + N 2, 0 x2 2 + N 0, 2 y2 2 + N 1, 1 xy

Each of these two energy functions gives an isocontour surface for each con-
stant energy in the space of E versus x, y. Since they are quadratic forms, the
 5.5 Electronic Rearrangement for Changing Conformations 185

(a) (b) (c)

Figure 5.5 Illustration of three cases of conical intersections between two potential energy
surfaces: (a) Bound-to-bound; (b) bound-to-unbound; (c) unbound-to-unbound.

isosurfaces are conical forms, and they touch at a degeneracy vertex, at the loca-
tion of the roots (x0, y0) in a (x, y) plane. They can be opposing cones or collateral
ones, depending on slopes along x and y. This in turn depends on what types of
bonding surfaces are interacting. Figure 5.5 illustrates conical intersections for a
triatomic system ABC, as can be expected coming from (a) isomerization, that is
a bound-to-bound structure interaction; (b) from dissociation in bound-to-
unbound structure changes; or also (c) from collisions as unbound-to-unbound
structure events.
Insight on the nature of these conical intersections can be gained constructing
simple models of the intersecting conical surfaces (or CISs) in a space E versus
(x, y), where x and y are bond distances or angles, and the other Nint − 2 internal
conformation variables are kept fixed. The intersection of two diabatic potential
energies leads to CISs when coupled, and as the other variables are changed the
conical shapes can be pictured as moving along seams in a hyperspace of Nint − 2
dimensions. Examples of models are given as follows.
a) Bound-to-bound CIS:
ω21 a 2 ω22 2 ω2 a 2 ω22 2
H11 = x+ + y , H22 = 1 x− + y − Δ, H12 = cy
2 2 2 2 2 2
This corresponds to two intersecting paraboloids and describes for exam-
ple SH2, of point group symmetry C2v, in the excited states 11A and 21A
[27], with x corresponding to the H-H distance r, and y for the distance R
from S to the center of mass of HH. The third variable is the angle between
r and R orientations, and is considered here to be chosen at a fixed value.
b) Bound-to-unbound CIS:
ω2 a 2 ω22 2 ω2
H11 = 1 x + + y , H22 = A exp − α x− x0 + 2 y2 −Δ,
2 2 2 2
x− x0 2 y −y0 2
H12 = cy exp +
λ2 μ2
where the unbound motion occurs as the x variable increases. This applies to
the dissociation of excited HNO(1Δ) into H(2S) + NO(2Π) [28], with the two
186 5 Intermolecular States

variables corresponding to the H to NO distance (x) and the HNO bond

angle (y).
c) Unbound-to-unbound CIS:

H11 = A exp −α x− x0 , H22 = B exp − β y − y0 − Δ,

2 2
x −x0 y − y0
H12 = cy exp +
λ2 μ2

This applies to a triatomic reaction A + BC AB + C with a fixed ABC angle

and x and y representing the bond distances of initial and final diatomics, such
as in collinear H + H2 H2 + H reaction. Also, and more generally, a triatomic
PES for three active electrons in an atom-diatomic reaction can be described by
the LEPS surface as mentioned in Chapter 4. That is the lower PES of a pair
generated by the combination of two valence-bond functions, and the two PESs
form a conical intersection when the radical in the LEPS expression is null, as
can be expected to also occur for other monovalent atom combinations at cer-
tain interatomic distances.
The diabatic to adiabatic transformation of electronic states is generated as
before by coefficients dependent now on the two variables x and y. For a com-
plex valued coupling H12(x, y) = H12(x, y) exp[iα(x, y)], the two new adiabatic
states are as before
a iα iα
Φ1 = Φ1 exp − cos β + Φ2 exp sin β
2 2
a iα iα
Φ2 = − Φ1 exp − sin β + Φ2 exp cos β
2 2
but now with
tan 2β x, y = 2 H12 x,y H11 x,y − H22 x,y
which is well defined away from the intersection point.
More accurately, one must consider all the internal degrees of freedom and
work with the displacements q = Q − Q(m), to allow for couplings of the displa-
cements in the two variables Qν, ν = a, b, with the remaining degrees of freedom,
as all atoms are displaced from the minimum conformation. Indicating with Q
the set of all internal coordinates other than Qa or Qb, the two potential energy
m
surfaces depend more generally also on the displacements q = Q − Q and
when these are not all zero the PESs have form E1, 2 x,y,q =
M x,y,q ± N x, y,q . One finds in general that N x0 , y0 , q 0 at the previous
roots, and the energy degeneracy at the vertex is lifted. This is likely to distort
the conical energy surfaces touching at the vertex, changing them into glancing
paraboloids. This avoided degeneracy is due to couplings to other variables
within the adiabatic state description, and therefore different in origin from
the avoided crossing in the one-dimensional case which arises from the two
 5.5 Electronic Rearrangement for Changing Conformations 187

original electronic states having the same symmetry. The magnitude of the split-
ting at the vertex can be found doing perturbation theory with the vibronic per-
turbation energy operator generated by displacing internal variables from the
reference values at the minimum conformation. Models can be presented in
terms of adiabatic or nonadiabatic PESs [29]. Two adiabatic PESs are also sub-
a
ject to nonadiabatic momentum couplings G kl , with momentum components
along the x and y directions, even if all other internal variables are assumed to
be fixed.
Depending on the nature of the two variables, such as whether they describe
distortions of a bond length or of a bond angle, the conical shapes may become
flattened and in the limit of a breaking bond the degeneracy may become a line,
showing as a seam for two intersecting PESs along the direction of the breaking
bond. This is what happens, for example in the case of the PES for collinear H+
+ H2 interacting with the PES for H + H2+ [20], and of the PES for Ar+ + H2
interacting with that of Ar + H2+ as shown in Figure 5.6a for the adiabatic
potential energies and Figure 5.6b for their nonadiabatic momentum coupling
a
τ = G kl ℏ along the diatomic axis [30]. The crossing between the two lowest
PESs can also be described as two intersecting potential energy lines in the plane
of the energy E and the bond distance r in H2 and in H2+, translated along a seam
in a direction perpendicular to that plane, as the distance R between Ar+ and H2
is changing. The crossing in Figure 5.6a below 1.0 Å (or 0.1 nm) is transformed
due to the momentum coupling along R, into an avoided crossing as shown in
Figure 5.6b.
The two adiabatic PESs are also subject to nonadiabatic momentum cou-
a
plings G kl , with momentum components along the x and y directions, even
if all other internal variables are assumed to be fixed.
A special situation arises when the two original electronic states, 1 and 2, are
degenerate with the same energy H11(x, y) = H22(x, y) for all argument values.
The degeneracy can be lifted due to a coupling H12(x, y) 0. Conditions under
which this happens can be found assuming vibronic coupling and can be stud-
ied by expanding the electronic Hamiltonian operator of the system in powers
of small displacements from the reference structure. Coupling matrix ele-
ments of the resulting force operator are different from zero if the direct prod-
uct of the three representations of the point groups for the two original states
and for any of the displacements contain the totally symmetric representation.
In this case, the degeneracy can be lifted with new states and PESs found from
the solution of the two-state eigenvalue equations. This analysis has been fol-
lowed to arrive at the Jahn–Teller theorem [31], stating that in any symmetric
nonlinear molecule with electronically degenerate states, displacement of
atoms lead to symmetry distortions that decrease the original symmetry.
188 5 Intermolecular States

(a)
5
+
Ar + H 2(2Σu)
4 Ar+ + H2(3Σu)
3
Ar+ + H + H
2

1
E (eV)

Ar + H+ + H
0 Ar+ + H2(4Σg)
–1
+
–2 Ar + H 2(2Σg)

–3

0 1 2 3 4
RH–H (Å)

(b)
Energy relative to Ar + H+ + H (eV)

–1.5 20

v1 v2
Interaction (Å–1)

–2.0 10

τr(R,r)
–2.5 0

–3.0
0.5 0.6 0.7 0.8 0.9 1.0
Distance (Å)

Figure 5.6 (a) Adiabatic potential energy surfaces for collinear Ar+ + H2 interacting with Ar + H2+
along the diatomic axis, as functions of the H to H distance and for large intermolecular Ar-H
distances; (b) non-adiabatic momentum coupling between the two lowest PESs at a fixed
intermolecular distance. Source: adapted from Ref. [30]. Reproduced with permission of Elsevier.

If the energy of the two degenerate states is, to begin with, a paraboloid show-
ing a single minimum versus x and y, the theorem indicates that this form will be
distorted leading to a more complex form with two minima in the (E, x) plane as
shown in Figure 1.6. But now as a circle is traced in the (x, y) plane around the
 5.5 Electronic Rearrangement for Changing Conformations 189

degeneracy point there are energy maxima and minima corresponding to lower
symmetry structures [32]. An example is provided by the radical C5H5, of symme-
try D5h to begin with, and in doubly degenerate states of label 2 E1 . The coupling
between the two states can be written in terms of the variables x = ρ cos(5φ) and
y = ρ sin(5φ) and as the angle φ varies over a circle this shows energies of struc-
tures with the lower symmetry C2v. The distortion may also arise from atomic
momentum couplings between the two original states, involving derivatives in
the x and y variables, again provided symmetry arguments indicate that the cou-
pling is not zero.
Conical intersections of PESs can be expected to appear when dealing with
photoexcited polyatomic molecules and may show up in large numbers when
the molecular system is complex, as in polymers and biomolecules. Recent
reviews have covered the role that CISs play in molecular spectroscopy and
photoinduced dynamics [33], in isomerization [34], organic photochemistry
[35], and inorganic compounds with transition metal atoms. The presence of
CISs in the primary isomerization event in vision has been experimentally con-
firmed using ultrafast optical spectroscopy in the visible and near infrared [36].
It can be expected that many CISs will be present and linked through seams in
photoinduced biomolecular dynamics.

5.5.4 The Geometrical Phase and Generalizations

The electronic wavefunctions introduced so far are real valued, but more gen-
erally, they can be constructed to also contain a phase factor making them com-
plex valued. The phase factor can be obtained from the Schrödinger equation
for the wavefunction, but it is nonunique. The phase appears to affect theoret-
ical results when adiabatic electronically excited states are involved and may be
needed to reproduce experimental results. This can be determined by doing cal-
culations of dynamical properties and comparing results with experimental
measurements of molecular spectra or collisional cross sections. The phase
of an electronic wavefunction has been considered and used in dynamics calcu-
lations under the various names geometric phase, or Berry phase, or molecular
Aharonov–Bohm phase [28, 37–41].
The role of the geometric phase is relevant at the level of the adiabatic approx-
imation. The electronic wavefunction can be allowed to take complex values
a
from a given phase factor ηk Q = exp iγ k Q ] with γ k a real-valued geometri-
cal phase, so that now
a a a
Ψ ad Q,X = ηk Q Φk X; Q ψ k Q
It appears as if a phase factor for the total molecular wavefunction should not
have physical content. But here one must consider what happens when the
nuclei undergo displacements through a hyperspace containing degenerate
190 5 Intermolecular States

electronic wavefunction points, where other states with l k have the same
energy Ek(Q) and interfere quantum mechanically with the state above, and also
what happens due to momentum couplings of electronic states.
Even when degeneracies and couplings are absent, the phase may alter
numerical values of PESs and the related dynamics. From the definition of
a a a
the momentum G kk Q , now for electronic state ηk Φk , one finds that it is
a
not any longer null but that instead G kk Q = ℏ∂γ k ∂Q, a hyper-momentum-
like quantity in the Q hyperspace, and that by integration along a line integral
from point Q0 to Q, the phase function factor is
Q
a i a
ηk Q = exp dQ G kk Q
ℏ Q0

Further, replacing the above wavefunction in the Schrödinger equation and

a ∗ a
projecting it on ηk Q Φk Q with integration over electronic variables,
one obtains
2
1 ℏ a a a a
∇Q + G kk Q ψk Q + Ek Q −E ψk Q =0
2M i
a
The effect of G kk Q on the nuclear wavefunction can be accounted for by
writing it as ψ k Q = φk Q ηk Q ∗ with φk real valued. Applying the
a a a a

momentum operator within the square bracket above, in sequence, one

finds that
ℏ2 a a
− ∇Q 2 + Ek Q − E φk Q = 0
2M
Therefore, introduction of the phase factor in the electronic state must be
compensated by a phase factor in the nuclear motion wavefunction describing
the system dynamics, to be able to recover the form of the total wavefunction. If
the electronic state is not degenerate, then the path in the line integral can be
closed and deformed to disappear, corresponding to a choice of a null phase
function. However, if the electronic state is degenerate with another one and
they interact to create an intersection of PESs, such as the conical one, then
there is a choice of paths around the intersection which will give nonzero
phases, and from these, it follows that there are related changes in the energies
of each of the interacting states. This has been carefully studied in examples of
adiabatic transformations of a state for three identical nuclei [28, 39] and in
many-state models analogous to a spin in a magnetic field [40].
An expression for the geometrical phase can be obtained introducing an inte-
gration path parameter s labeling positions along a chosen path Q(s). Writing
the equation for the electronic state Υk = ηkΦk as
 5.5 Electronic Rearrangement for Changing Conformations 191

H Q s Υk Q s = Ek Q s Υk Q s
Differentiating it with respect to s, one can construct a state change at position
s1 = s + Δs from
∂H Q s
∂Υk Υl ∂s Υk
= Υl 1
∂s 1
l k 1
Ek Q s1 −El Q s1
followed by a second step to s2 = s + 2Δs giving
∂H Q s ∂H Q s
∂Υk Υk ∂s Υl Υl ∂s Υk dγ k
Υk =− 1 2
=
∂s 2
l k Ek Q s1 − El Q s1 2 ds 2

This allows generation of the geometric phase in a step-wise procedure along

a path, for a given electronic Hamiltonian and adiabatic basis set.
As a path moves around a CIS, the adiabatic state in the lowest surface will
change with the parameter s and, after a closed loop in the x and y variables that
brings them to their original values in tan(2β) = 2 H12 /(H11 − H22), one finds
that 2β will have changed by ±2π and the adiabatic wavefunction coefficients
cos(β) and sin(β) will have changed the wavefunction by a factor −1. This is
an essential change indicating that one must incorporate the geometrical phase
γ k in electronic and nuclear wavefunctions in a treatment of the molecular
dynamics so they compensate each other, to make sure that results are invariant
when the physical system is distorted along a path until it returns to its original
conformation.
a
Inspection of the matrix equation given above for ψ k shows that there is a
similarity with the form of the Schrödinger equation for a charged particle in an
electromagnetic field described by vector and scalar potentials A and φ, to be
a a
compared to G kk Q and Ek Q . In the same way as a gauge transformation
of the electromagnetic potentials preserves the form of the particle equation of
motion, now the transformation generated by the geometric phase γ k plays a
similar role for the equation of a single adiabatic state. More generally, this
can be extended for coupled adiabatic states by considering the previous
coupled equations for moving nuclei and the column matrix ψ R , which is also
valid for a ψ(Q) in many variables, containing the square matrices G(a)(Q) and
H(a)(Q). This suggests that the geometric phase in a molecular system can be
generally treated using a unitary local gauge transformation U(a)(Q) applied
a
to the set of electronic states { Φk Q . Corresponding transformations of
a
G(a)(Q) and H(a)(Q) and of the nuclear motion states {ψ k Q with the adjoint
U(a)† and substitutions in the Schrödinger equation show that its form remains
invariant under the new gauge transformation [42]. This generalizes the
192 5 Intermolecular States

observation made above for a single adiabatic state and shows the way to con-
struct geometric phase factors for many coupled states.
It can be expected that geometric phases in an adiabatic representation will
play a quantitative role in the calculation of dynamical properties of electron-
ically excited systems, such as optical spectra and collisional phenomena show-
ing electronic rearrangement. This happens in the reaction D + H2 HD + H,
for which comparison of theory and experiment have shown the existence of a
conical intersection and a need for a geometric phase in the calculation of the
reaction cross section using adiabatic potential energy surfaces [43].

References
1 Arfken, G. (1970). Mathematical Methods for Physicists, 2e. New York:
Academic Press.
2 Hohenberg, P. and Kohn, W. (1964). Inhomogeneous electron gas. Phys. Rev. B
136: 864.
3 Levy, M. (1979). Universal variational functionals of electron densities. Proc. Nat.
Acad. Sci. U. S. A 76: 6062.
4 Kohn, W. and Sham, L.J. (1965). Self-consistent equations including exchange
and correlation effects. Phys. Rev. A 140: 1133.
5 Gordon, R.G. and Kim, Y.S. (1972). Theory for the forces between closed shell
atoms and molecules. J. Chem. Phys. 56: 3122.
6 Parr, R.G. and Yang, W. (1989). Density Functional Theory of Atoms and
Molecules. Oxford, England: Oxford University Press.
7 Levine, I.N. (2000). Quantum Chemistry, 5e. Upper Saddle River: Prentice-Hall.
8 Born, M. and Huang, K. (1954). Dynamical Theory of Crystal Lattices, Appendix
VIII. Oxford, England: Oxford university Press.
9 Hirschfelder, J.O., Curtis, C.F., and Bird, R.B. (1954). Molecular Theory of Gases
and Liquids. New York: Wiley.
10 Hirschfelder, J.O. (ed.) (1967). Intermolecular Forces (Adv. Chem. Phys., vol. 12).
New York: Wiley.
11 Nikitin, E.E. (1974). Theory of Elementary Atomic and Molecular Processes in
Gases. London, England: Oxford University Press.
12 Micha, D.A. (1974). Effective Hamiltonian Methods for Molecular Collisions
(Adv. Quantum Chem., vol. 71), 231. New York: Academic Press.
13 Garrett, B.C. and Truhlar, D.G. (1981). The coupling of electronically adiabatic
states in atomic and molecular collisions. In: Theoretical Chemistry, Part A,
vol. 6 (ed. D. Henderson), 216. New York: Academic Press.
14 Baer, M. (1985). The theory of electronic nonadiabatic transitions in chemical
reactions. In: Theory of Chemical Reaction Dynamics, vol. II (ed. M. Baer), 219.
Boca Raton, FL, USA: CRC Press.
 References 193

15 Smith, F.T. (1969). Diabatic and adiabatic representations for atomic collision
problems. Phys. Rev. 179: 111.
16 Baer, M. (1975). Adiabatic and diabatic representationsfor atom-molecule
collisions: treatment of the collinear arrangement. Chem. Phys. Lett. 35: 112.
17 Mead, A.C. and Truhlar, D.G. (1982). Conditions for the definition of strictly
diabatic electronic basis for molecular systems. J. Chem. Phys. 77: 6090.
18 Olson, J.A. and Micha, D.A. (1982). Electronic state representations at molecular
potential pseudocrossings. Int. J. Quantum Chem. 22: 971.
19 Micha, D.A. (1983). A self-consistent eikonal treatment of electronic transitions
in molecular collisions. J. Chem. Phys. 78: 7138.
20 Tully, J.C. (1976). Nonadiabatic processes in molecular collisions. In: Dynamics
od Molecular Collisions Part B (ed. W.H. Miller), 217. New York: Plenum Press.
21 Cohen, J.M. and Micha, D.A. (1992). Electronically diabatic atom-atom
collisions: a self-consistent eikonal approximation. J. Chem. Phys. 97: 1038.
22 Fernandez, F.M. and Micha, D.A. (1992). Time-evolution of molecular states in
electronically diabatic phenomena. J. Chem. Phys. 97: 8173.
23 Bruna, P.J. and Peyerimhoff, S.D. (1987). Excited-State Potentials (Adv. Chem.
Phys., vol. 67, Part I), 1. New York: Wiley.
24 Hirschfelder, J.O. and Meath, W.J. (1967). The nature of intermolecular forces.
Adv. Chem. Phys. 12: 3.
25 Cohen-Tanoudji, C., Diu, B., and Laloe, F. (1977). Quantum Mechanics, vol. 1,
Chap. IV. New York: Wiley.
26 Tinkham, M. (1964). Group Theory and Quantum Mechanics. New York:
McGraw-Hill.
27 Matsunaga, N. and Yarkony, D.R. (1997). Energies and derivative couplings in
the vicinity of a conical intersection II. CH2 and SH2. J. Chem. Phys. 107: 7825.
28 Herzberg, G. and Longuet-Higgins, H.C. (1963). Intersection of potential energy
surfaces in polyatomic molecules. Discuss. Faraday Soc. 35: 77.
29 Worth, D.A. and Cederbaum, L.S. (2004). Beyond Born-Oppenheimer: molecular
dynamics through a conical intersection. Annu. Rev. Phys. Chem. 55: 127.
30 Baer, M. and Beswick, J.A. (1979). Incorporation of electronically nonadiabatic
effects into bimolecular reactive systems. III. The collinear Ar + H2+ system.
Phys. Rev. A 19: 1559.
31 Jahn, H.A. and Teller, E. (1937). Stability of polyatomic molecules in degenerate
electronic states. I. Orbital degeneracy. Proc. R. Soc. A 161: 220.
32 Herzberg, G. (1966). Molecular Spectra and Molecular Structure III. Polyatomic
Molecules. New York: Van Nostrand Reinhold.
33 Domcke, W. and Yarkony, D.R. (2012). Role of conical intersections in molecular
spectroscopy and photoinduced chemical dynamics. Annu. Rev. Phys. Chem.
63: 325.
34 Levine, B.G. and Martinez, T.J. (2007). Isomerization through conical
intersections. Annu. Rev. Phys. Chem. 58: 613.
194 5 Intermolecular States

35 Robb, M.A. (2011). Conical intersections in organic photochemistry. In: Conical

Intersections: Theory, Computation, and Experiment (eds. W. Domcke, D.R.
Yarkony and H. Koeppel), 3. World Scientific Publishing Co.
36 Polli, D., Altoe, P., Weingart, O. et al. (2010). Conical intersection dynamics of
the primary photoisomerization event in vision. Nature 467: 440.
37 Longuet-Higgins, H.C. (1975). The intersection of potential energy surfaces in
polyatomic molecules. Proc. R. Soc. Lond. A 344: 147.
38 Stone, A.J. (1976). Spin-orbit coupling and the intersection of potential energy
surfaces in polyatomic molecules. Proc. R. Soc. Lond. A 351: 141.
39 Mead, C.A. and Truhlar, D.G. (1979). On the determination of Born-
Oppenheimer nuclear motion wavefunctions including complications due to
conical intersections and identical nuclei. J. Chem. Phys. 70: 2284.
40 Berry, M.V. (1984). Quantal phase factors accompanying adiabatic changes.
Proc. R. Soc. Lond. A 392: 45.
41 Aharonov, Y. and Bohm, D. (1959). Significance of the electromagnetic potential
in the quantum theory. Phys. Rev. 115: 485.
42 Pacher, T., Mead, C.A., Cederbaum, L.S., and Koeppel, H. (1989). Gauge theory
and quasidiabatic states in molecular physics. J. Chem. Phys. 91: 7057.
43 Kuppermann, A. and Wu, Y.-S.M. (1993). The geometric phase effect shows up
in chemical reactions. Chem. Phys. Lett. 205: 577.
 195

Many-Electron Treatments

CONTENTS

6.1 Many-Electron States, 195

6.1.1 Electronic Exchange and Charge Transfer, 195
6.1.2 Many-Electron Descriptions and Limitations, 198
6.1.3 Properties and Electronic Density Matrices, 203
6.1.4 Orbital Basis Sets, 205
6.2 Supermolecule Methods, 209
6.2.1 The Configuration Interaction Procedure for Molecular Potential
Energies, 209
6.2.2 Perturbation Expansions, 215
6.2.3 Coupled-Cluster Expansions, 218
6.3 Many-Atom Methods, 222
6.3.1 The Generalized Valence-Bond Method, 222
6.3.2 Symmetry-Adapted Perturbation Theory, 225
6.4 The Density Functional Approach to Intermolecular Forces, 228
6.4.1 Functionals for Interacting Closed- and Open-Shell Molecules, 228
6.4.2 Electronic Exchange and Correlation from the Adiabatic-Connection
Relation, 232
6.4.3 Issues with DFT, and the Alternative Optimized Effective Potential
Approach, 238
6.5 Spin-Orbit Couplings and Relativistic Effects in Molecular Interactions, 243
6.5.1 Spin-Orbit Couplings, 243
6.5.2 Spin-Orbit Effects on Interaction Energies, 245
References, 247

6.1 Many-Electron States

6.1.1 Electronic Exchange and Charge Transfer
A fundamental and general treatment of molecular interactions must be based
on the quantal description of the many-electron system for given nuclear con-
formations, including Coulomb energies for charged particles and electronic

Molecular Interactions: Concepts and Methods, First Edition. David A. Micha.

© 2020 John Wiley & Sons, Inc. Published 2020 by John Wiley & Sons, Inc.
196 6 Many-Electron Treatments

spin-orbit coupling energies. Here the emphasis is on aspects of the many-

electron theory for two interacting species A and B, as they are needed to obtain
correct intermolecular energies at all relative distances between them. The pres-
entation assumes familiarity with concepts of molecular electronic structure
and properties as found in the literature [1–5].
Occupied and unoccupied electronic orbitals of two molecules A and B at rel-
ative distance R change at intermediate and short intermolecular distances to
give energy levels of the supermolecule AB, as electronic orbitals of the two spe-
cies combine and allow electrons to spread over the whole system. Electronic
exchange involves combinations of the occupied molecular orbitals (MOs) of
A with occupied MOs of B, while electron transfer involves combinations of
occupied MOs of one species with unoccupied MOs of the other. In both cases,
the species can be in their ground electronic states or in excited ones, as shown
in Figure 6.1. The pair AB is mentioned here as a compound when A and B form
bonds, or as a complex if they are not bonded.
Expectation values of extensive properties, such as the electronic energy E(R)
and the electron density ρ(R), where R is the distance between the centers of
mass of A and B, must go in the limit of large intermolecular distances into
the sums of the fragments contributions. One must have for R ∞ that

E ≈ EA + EB and ρ ≈ ρA + ρB

This will be called here asymptotic consistency, but is usually termed size con-
sistency in the literature. Its requirement is essential in treatments of intermo-
lecular forces. It guarantees that the potential energy of interaction at all R
satisfies V(R) = E(R) − (EA + EB) ≈ 0.
When the number of interacting identical species S is increased, as one pro-
ceeds from clusters to solids or fluids, the extensive properties (in the sense of
statistical thermodynamics) must increase linearly. For a system SN with a large
number N of the stable species S, one has E(SN) ≈ NE(S), as N ∞, where SN
may be a compound or a complex. This is size extensivity, a property needed in
treatments of many-molecule systems.
Two general methods for intermolecular forces are the so-called supermole-
cule method, where the interacting species are considered to be a single many-
electron system in a field of all the nuclei, and the many-atom method where the
total many-electron system is instead treated as a collection of atomic many-
electron systems. In both cases, one must properly treat the overall symmetry
of the many-electron state with regard to electron exchange, its total electronic
spin state, and its total point group or crystal group symmetry.
The methods described in what follows are applicable also to chemical reac-
tions such as A + B C + D, where C and D are new species, involving the rear-
rangement of nuclei. Treatments of energetics require exploration of potential
energy surfaces over wide ranges of atomic positions, to identify potential
 6.1 Many-Electron States 197

Figure 6.1 Pictorial description of (a)

electronic energy levels and their
combination for interacting species
A and B: (a) closed shell–closed shell;
(b) closed–open shell; (c) open A B
shell–open shell. Their molecular
orbital (bonding and antibonding)
occupations are shown at
intermediate distances: (a, b1, c1)
electronic exchange; (b2, c2) (b1)
electron transfer.

A B

(b2)

A+ B–

(c1)

A B

(c2)

A+ B–

energy barriers between regions of low energy. Much can be learned about
allowed or disallowed reactions from qualitative treatments based on the shapes
of molecular orbitals and the interactions between highest occupied and lowest
unoccupied molecular orbitals (HOMOs and LUMOs) of the interacting
species [6–8]. But accurate calculations of reaction energy barriers, stable
conformations, and vibrational-rotational properties require many-electron
treatments as described in what follows.
198 6 Many-Electron Treatments

6.1.2 Many-Electron Descriptions and Limitations

In the supermolecule method, the basic idea is to introduce molecular spin-
orbitals, which spread over all the atoms in the whole system, and then to con-
struct many-electron states to calculate their properties. A self-consistent field
(SCF) approximation such as the restricted Hartree–Fock (or RHF) treatment
for the many-electron system in the supermolecule method, with doubly or sin-
gly occupied MOs constructed from linear combinations of atomic orbitals
(LCAOs), is of limited applicability to intermolecular forces. It works for two
interacting closed-shell systems such as He2 for which SCF leads correctly to
He + He asymptotically (although without their long-range van der Waals inter-
action), but gives incorrect results for two open-shell systems such as in H2, for
which the SCF wavefunction would asymptotically contain the electronic states
of both H + H and H+ + H− and would give incorrect potential energies. This
can be improved using an unrestricted HF (or UHF) description, but this does
not generate a molecular state of the correct spin symmetry. The SCF treatment
for the interaction between an open-shell species and another with closed shells
may also lead to asymptotically incorrect potential energies due to unphysical
contamination of a neutral state with a charge transfer state, such as in ArF
going to Ar + F and also Ar− + F+. These are examples of appearances of
“intruder states” at large distances, giving wrong asymptotic energies and
requiring extensions of the SCF treatment. More general treatments must allow
for asymptotic consistency and size extensivity and must be amenable to sys-
tematic improvement to a desired high accuracy.
Supermolecule treatments that do this are the configuration interaction (CI)
treatment, many-body perturbation (MBPT) for electrons, and many-electron
coupled clusters (CC) treatments. But even these must be carefully considered
in applications to intermolecular forces to satisfy constrains and provide accu-
racy and would not necessarily account for van der Waals (vdW) interactions.
They are treated in detail in Section 6.2.
In the many-atom treatment, electrons are located in spin orbitals, which
extend only over each molecular component of the total system. Component
many-electron functions are first constructed and then used to obtain the state
of the whole system. This must be done so that electron exchange antisymmetry
is properly incorporated in the whole many-electron function. Two such gen-
eral procedures are the extended valence-bond method (EVB) and the symme-
try-adapted perturbation theory (SAPT) of a many-atom system. They are
treated in Section 6.3.
Both treatments introduce a many-electron wavefunction of all the electronic
variables X = {xj : j = 1 to N} of N electrons, with xj = r j ,ζ j containing electron
position and spin variables, and is a function that depends parametrically on the
set of nuclear positions Q, given as Φ(X; Q). However, in what follows in this
chapter the nuclear positions will be fixed and omitted, and the focus will be
 6.1 Many-Electron States 199

on electronic aspects. The total electronic wavefunction of a stationary state

solution for a given electronic Hamiltonian H Q = H must be constructed to
be antisymmetric with respect to electron exchange and will typically describe
a system with given total spin quantum numbers S, MS(= − S, …, + S), and with
point group or crystal symmetry of type Γ, both obtained from the spin and
space symmetries of the electronic Hamiltonian for fixed nuclei [3]. These con-
ditions can be imposed by means of projection operators constructed from con-
stant of motion of the Hamiltonian [9].
Antisymmetry is imposed applying the projection operator
AS −1
= N P
−1 p P
where P is a permutation among the N electrons, of parity p, and the summation
extends over all different permutations, satisfying PH = HP. The self-adjoint
† AS † AS
property P = P implies that = as well. Insofar as a product of
two permutations is also a permutation in the set, it follows that AS is idem-
AS AS AS
potent, with = and that it also commutes with the Hamilto-
nian. Starting with an electronic function F(X), a proper antisymmetrized
N
state is usually constructed as Φ X = F X , with the antisymmetrizer
N 1 AS
= N 2 . A fundamental law of quantum mechanics for identical par-
ticles is that the wavefunction of identical fermions must be antisymmetric
under their exchange even if the species A and B are far away.
Therefore, even at very large distances, the total wavefunction must be of the
N
form ΦAB X ≈ FA X A FB X B involving the total antisymmetrizer, while
expectation values of an extensive property M AB must satisfy the asymptotic
condition M AB = ΦAB M AB ΦAB ≈ M A + M B as R ∞. This can be
verified considering that the total antisymmetrizer can be expressed as a prod-
uct of antisymmetrizers of the two species A and B times a complimentary oper-
AS
ator = I + T − 1 t T containing the identity I and transpositions T of
electrons between A and B,
AS −1 AS AS AS
= NA NB N A B
N A B N N
or alternatively = A B with a complimentary antisymmetri-
zer. When this operates at large distances R, the transpositions of electrons
from, say, A moves its electrons to regions of B where expectation values with
respect to states of A vanish because the wavefunctions of A fall exponentially
with distance. Cross terms disappear, and the sum of asymptotic property values
N
can be obtained. Therefore, as long as ΦAB X ≈ ΦA X A ΦB X B in terms
of wavefunctions for A and B with energies EA and EB, one finds for the total
200 6 Many-Electron Treatments

int
energy that E = ΦAB H A + H B + H AB ΦAB ≈ EA + EB and the potential
energy V(R) vanishes at large distances.
Spin and space symmetries can be imposed starting from normal many-
† †
electron operators N, with the property N N = N N, which generate symme-
try transformation and are constants of motion insofar as NH = HN. Their
eigenstates and eigenvalues in N k = νk k can be used to define projection
operators k = k k , which extract a component of correct symmetry from a
wavefunction as Φk X = X kΦ = X k k Φ X , where it is shown that
the second factor may also depend on some of the electronic variables. In appli-
cations where the constant of motion operator has only a discrete set of eigen-
values, a frequent situation, the projection operator can be conveniently
constructed as
N − νl
k = l kν
k −νl

with the product extending over only different eigenvalues. It satisfies the
orthogonality property k l = δkl k .
This is applicable to the angular momentum operator with components
M ξ , ξ = x,y, z, whether this is a component of the spin S or of the orbital angular
momentum L. For the total spin operator with components S ξ = j S ξj , of mag-
2
nitude and projection S and S z with eigenvalues S(S + 1) and MS = − S, − S + 1,
…, S, the projector SMS = S MS can be written as [10]

S + MS Smax −S μ S S−−MS + μ S S+− MS + μ

SMS = 2S + 1 −1
S − MS μ=0 μ 2S + μ + 1
in terms of spin-projection rising and lowering operators S ± = S x ± iS y , satisfy-
ing Sz S ± = S ± Sz ± 1 , in a form suitable for construction of spin-adapted
many-electron states. They satisfy the orthogonality condition SMS S MS =
δSS δMS MS SMS . At large distances, one can write S ± = SA ± + SB ± in terms of
the operators for the species, which leads to products such as SA + SB − in the
projector. Provided the asymptotic states of A or B are nondegenerate, whether
ground or excited states, the expectation value of these cross terms gives zero
and one recovers a sum of uncoupled terms for A and B. If, however, the asymp-
totic states of A and B are degenerate, then it is necessary to analyze their expec-
tation values to ascertain asymptotic behavior.
Relating to space symmetry, the constants of motion are operator movements
R such as inversion, reflections, or rotations, which commute with the Hamil-
tonian and which form a symmetry group . Representation matrices, or their
 6.1 Many-Electron States 201

characters χ k R for the k-th irreducible representation, can be used to con-

struct projection operators Γ k for each irreducible representation Γ(k) as a
sum over the group symmetry movements [3]. One such projector is
dk ∗
Γk = χk R R
h R

with h the order of the group and dk the dimension of the representation.
The orthogonality properties of characters ensures that Γk Γl =
δkl Γ k . The asymptotic form of this projector depends on how the overall
symmetry is expressed in terms of the symmetries of the Hamiltonians of A and
B. In some cases, such as for overall translation or reflection on a plane, the
group is a direct product of those of A and B, = A ⨂ B , with characters
χ k R = χ kA RA χ kB RB , which allow factorization of the projector, and a
proper asymptotic decomposition will follow. However, if the total group
involves also a complementary subgroup as in = ⨂ A ⨂ B , then a further
analysis of the asymptotic wavefunction and expectation values must be done to
establish whether cross terms vanish to give the correct behavior [3].
Many-electron wavefunctions are sometimes variationally obtained by mini-
mizing the total electronic energy but do not display the total symmetries of the
system. In such cases, it is yet possible to extract the correct symmetry by using
projection operators after the variation, in a postsymmetrization. Consider a
variational but nonsymmetrized wavefunction Φ, which would contain symme-
try-adapted components Φk as in Φ = kakΦk, assumed to be all normalized for
simplicity. The variational energy is Evar = k|ak|2Ek with factors |ak|2 ≤ 1.
Therefore postsymmetrized components are found to provide energies Ek lower
than the unsymmetrized original wavefunction. But while the original unsym-
metrized wavefunction gives an upper bound to the calculated energy, the post-
symmetrized one may not do so for the exact symmetrized energy and must be
tested for accuracy of the postsymmetrized associated energy. This can be done,
for example, by doing some calculations with prior symmetrization and other-
wise the same conditions at selected nuclear conformations, to compare ener-
gies with results from postsymmetrization.
The interaction energy of open-shell or of electronically excited species
presents additional complications because they frequently involve a number
ND of states ΦJ, J = 1 to ND, of the whole system that are degenerate or nearly
degenerate in energy, and are either eigenstates of the Hamiltonian for the
energy E(D), with H ΦJ = E D ΦJ or have very similar expectation values
ΦJ H ΦJ E D . They define a multireference, or configuration interaction
(CI), state space to be treated as a whole. In these cases, one must extend the
usual treatments to include linear combinations of the ΦJ states in the CI multi-
reference wavefunctions, or one must construct unperturbed multireference
Hamiltonians in perturbation and coupled cluster treatments.
202 6 Many-Electron Treatments

A wave operator treatment appears to be suitable for these extensions

[11–13]. Introducing a multireference projection operator = J ΦJ ΦJ
onto the space of the degenerate states with energy E(D), and its complement
= I − , the Schroedinger equation H Ψ = E Ψ can be formally
solved writing Ψ = Φ(D) + Ψ , with the first term a chosen reference state
satisfying Ψ = Φ D , a generalization of intermediate normalization
2
giving ΦD Ψ = J Φ D ΦJ =N D
. The energy is then
E= Φ D
H Ψ N . Defining a wave operator Ω to give Ψ = ΩΦ
D
as D

the solution of the Schrodinger equation, one finds that using the reduced resol-
vent operator

−1
RE = α + E −H

with α a convenient constant to be set to zero after calculations, the wavefunc-

tion is given by

Ψ = ΩΦ D = + RE H ΦD

as can be verified applying the inverse of R E to both sides of this equation. This
multireference wave operator can be constructed using CI methods or pertur-
bation expansions, described as follows.
Electronic antisymmetry and spin and space symmetries enforced by projec-
AS
tion operators and k can all be maintained when the wave operator is
applied, provided the projector has been symmetry-adapted so that
AS AS AS AS
= and k = k . Then one finds that Ω = Ω and
Ω k = k Ω insofar as the wave operator has been constructed from the full
Hamiltonian.
The energy of the whole system is obtained as E = Φ D H ΩΦ D N D ,
and the potential energy follows from V = E − (EA + EB). For the purpose of
generating potential energy surfaces going to the correct asymptotic limits,
it is convenient, but not necessary, that the reference state is chosen as
D D
ΦD = ΦA ΦB and that the wave operator for the pair of species A and B
D D
satisfies ΩAB Φ D ≈ ΩA ΩB ΦA ΦB as R ∞, with the complementary
antisymmetrizer. In this case, the expectation value of an extensive property
would go asymptotically to the sum of its values for the two separate species,
and in particular V = E − (EA + EB) ≈ 0, with asymptotic ener-
D
gies EI = Φ D H ΩΦ D I NI ,I = A,B.
 6.1 Many-Electron States 203

6.1.3 Properties and Electronic Density Matrices

Many-electron operators can be classified as being one-electron, two-electron,
or N-electron operators. Most physical properties are given by sums of one-,
and two-electron operators and their expectation values can be obtained from
one- and two-electron density operators and their matrices (DMs). They have
been thoroughly considered in Refs. [2] and [14]. The operator of a property A in
an N-electron system, such as the energy or an electric multipole, can be writ-
ten as
1 2 N
A=A +A + +A ,
1 2 1
A = a1
1≤m≤N m
,A = a2
1 ≤ m < n ≤ N m, n
= a 2 ,…
m n m, n
2

with am a one-electron operator containing in general the position and momen-

tum variables of electron m, and matrix elements in the coordinate representa-
tion with states xm = r m ,ζ m written as xm am xm = am x,x , and so on.
The expectation value of each term in a state with wavefunction ΦJ(1, 2, …, N)
is given by

1 1
ΦJ A ΦJ = d 1 d 1 a 1 1, 1 ρJ 1 ;1 ,

d N Φ∗ J 1 ,2, …,N ΦJ 1,2, …,N

1
ρJ 1 ;1 =N d 2

2 2
ΦJ A ΦJ = d 1 d 1 d 2 d 2 a 2 1, 2,1 ,2 ρJ 1 , 2 ; 1, 2 ,

N N −1
d N Φ∗ J 1 ,2 , 3,…,N ΦJ 1, 2,3, …,N
2
ρJ 1 ,2 ; 1, 2 = d 3
2
1 2
and so on. Here ρJ and ρJ are one- and two-electron reduced density func-
tions for this state in the coordinate representations of reduced density opera-
1 2
tors (RDOps) ρJ and ρJ with the argument (1) a short form for (x1) and with
d 1 … = d 3 r1 ζ1 = ± … . They have been defined here with normalization
tr 1 ρ 1 = d 1 ρ 1 1; 1 = N, tr 2 ρ 2 = d 1 d 2 ρ 2 1, 2; 1, 2 = N N − 1
2, the number of different electron pairs, and so on. This is a common normal-
n
ization [14] but not universal. Transition averages ΦJ A ΦK can similarly
n
be expressed in terms of transition density matrices [14]. The electronic ρJK
reduced density operators are usually constructed in a basis set of one-electron
functions for computational purposes.
204 6 Many-Electron Treatments

In particular, the N-electron Hamiltonian, including here only kinetic ener-

gies of electrons, and Coulomb electron–electron vm,n and electron–nuclei
interaction Vm energies, is given by
1 2 1 2
H =H +H ,H = h ,
1≤m≤N m
H = v
1 ≤ m < n ≤ N m, n

where hm = K m + Vm is the sum of kinetic energy plus electron attraction to all

nuclei (or ion cores), and vm, n = c2e r m −r n . Total electronic energies as func-
tions of atomic positions for a state J are found from
1 2
EJ = d 1 d 1 h 1,1 ρJ 1 ; 1 + d 1 d 2 v 1,2 ρJ 1,2; 1,2

insofar as the Coulomb potential energy is local so that 1, 2 v1, 2 1 , 2 =

1
δ(1 − 1 )δ(2 − 2 )v(1, 2). The density function ρJ 1 ; 1 can be obtained from
2
ρJ 1 , 2; 1,2 integrating over (2) so that the total energy is a linear functional
2
of ρJ .
When the interacting pair of species A and B move apart as R ∞, each con-
taining NA and NB electrons, the reduced densities for one electron must phys-
ically go to
1 1 1
h 1, 1 ρJ 1 ; 1 ≈ h 1, 1 ρJ 1 ;1 + h 1, 1 ρJ 1 ;1
A B
1 1
with tr 1 ρJ , A = NA and tr 1 ρJ , B = NB , and with the corresponding limit for
2
the two-electron function ρJ 1, 2; 1, 2 . One then finds that EJ ≈ EJ,A + EJ,B.
This, however, is not always the case in approximations, depending on how
the state J has been constructed, and it must be verified in applications to inter-
molecular forces.
In the particular case when the many-electron state is a single determinantal
function, as in the Hartree–Fock method, with ΦJ 1,2, …,N = D 1,2, …,N =
−1 2
ψ 1 1 ψ 2 2 …ψ N N = N det ψ j n , constructed from N orthonormal
spin orbitals (SOs) ψ j, one finds, recalling that the determinant of a product of
matrices equals the product of their determinants, the special results
1 N ∗
ρHF 1 ; 1 = j=1
ψj 1 ψj 1
2 1 1 1 1 1
ρ 1 ; 1 ρHF 2 ; 2 − ρHF 1 ; 2 ρHF 2 ; 1
ρHF 1 , 2 ; 1,2 =
2 HF
Here the first line shows that each SO in the determinant appears with an
occupation number of 1, and the second line indicates that the total energy
1
can be obtained from just ρHF , insofar as the total HF energy follows from
1 2
EHF = d 1 d 1 h 1, 1 ρHF 1 ; 1 + d 1 d 2 v 1, 2 ρHF 1,2; 1, 2
 6.1 Many-Electron States 205

The density operators ρ 1 and ρ 2 for a general state J can be expanded

in an orthonormal basis set {uk(xm)} = {uk(m)} of one-electron functions
uk(m) = m k . This gives in the Dirac notation

ρ1 = k, k
k ρ 1 k,k k

ρ2 = k1 , k1 k2 , k2
k1 k2 ρ 2 k1 ,k2 , k1 k2 k1 k2

providing reduced density matrix (RDM) elements in the basis set.

These operators have natural spin orbitals (or NSOs) eigenstates and eigen-
values (or occupation numbers), which can be obtained by diagonalizing the
1
matrices with a unitary transformation to a new basis set ul m [14]. For
the one-electron RDOp, this gives
1 1 1
ρ1 = n
l l
ul ul
1 1 1
where the NSO occupation numbers nl satisfy l nl = N so that 0 ≤ nl ≤ 1.
These occupation numbers provide a useful criterion for choosing a basis set of
optimal expansion SOs in many-electron systems. The molecular NSOs with
large occupation numbers can be expected to be the most relevant for descrip-
tion of electronic correlation. Given a subset of NSOs as a basis set, it can be
systematically improved by adding to the set the NSOs with subsequent large
occupation numbers, to achieve convergence of properties. The NSOs can be
obtained at different levels of accuracy from wavefunctions containing elec-
tronic correlation, and they can be tested to determine whether they stay stable
as levels of correlation are increased, to generate a computationally useful basis
set. This selection procedure can be done separately for the component species
A and B, and the NSOs from each one can be combined into a larger basis set to
be used for the whole AB system.

6.1.4 Orbital Basis Sets

The calculation of electronic orbitals as functions of electron positions requires
solving partial differential equations for the orbitals. This is usually done intro-
ducing basis sets to expand the orbitals, with expansion coefficients extracted by
solving coupled matrix equations. The choices of basis sets are guided by
requirements of accuracy and feasibility in the calculations of properties. Delo-
calized electron orbitals (MOs) ψ j in a molecular system can be expanded as
linear combinations of atomic orbitals (LCAOs) χ μ r , with the index μ contain-
ing the atom location and quantum numbers of core (inner shell) or valence
(outer shell) electrons, and may include AOs hybridized or polarized by inter-
atomic interactions. These have overlaps Sμν = χ μ χ ν components of the
206 6 Many-Electron Treatments

matrix S and, if they form in practice a complete set, then they also satisfy the
identity decomposition μ, ν χ μ S −1 μν χ ν = I. Orbitals of large systems like
solids or polymers can also be alternatively expanded in a basis set of plane
wavefunctions ϕk r for wavevectors k forming a grid in reciprocal space.
Important computational considerations in the choice of a basis set are how fast
can electronic integrals be done with the basis set and how large are the matrices
that must be stored and multiplied in calculations of properties. The size of the
basis set is important for achieving high accuracy of potential energies, and con-
sistent basis sets must be used when subtracting the energies of components A
and B from the total energy of a complex A + B, to obtain changes of energies
with structural parameters.
Numerous basis sets of AOs have been introduced in the literature and imple-
mented for computational work [3, 4, 15]. The original basis sets derived from
hydrogenic orbitals, products of radial functions times angular functions, and
have used Cartesian coordinates (x, y, z) or spherical coordinates (r, ϑ, φ) with
origin at the atomic nucleus positions. In the so-called (nlm) Slater-type orbitals
(or STOs), a radial factor Rn(r) = rn − 1 exp(−ζr) containing an exponential
parameter ζ multiplies spherical harmonics Ylm(ϑ, φ) for the angular depend-
ence. The nodeless radial functions are combined into contractions with fixed
coefficients to describe the physical radial functions. The more popular and
recent basis sets are made up of Gaussian radial functions (or GTOs), which
simplify the calculation of one- and two-electron integrals. The primitive radial
factors are of form exp(−ζr2) times a power of the radius, multiplying spherical
harmonics, or instead the exponential multiplies powers xaybzc of Cartesian
coordinates with a + b + c = l. The quantum number l corresponds to an occu-
pied l-subshell (s-, p-, d- …) or to a subshell with higher l if there is a need to
describe polarization or hybridization of shells.
Products of primitive GTOs ga, b, c(x, y, z; ζ) = exp(−ζr2)xaybzc located at dif-
ferent centers can be rewritten as combinations of GTOs centered at interme-
diate points, a fact that speeds the calculation of electron integrals. But GTOs do
not accurately describe the electronic wavefunction cusp at an atomic position,
and they decrease with r much faster than the physical exponential functions.
These shortcomings are compensated using: (i) primitives with more than one
value of ζ combined to give double-zeta (or DZ), triple- (TZ), quadruple- (QZ),
or n-zeta (n-Z) functions; (ii) contracted sums of gaussians (CGTO) where Nl
primitives of a given l are combined with fixed coefficients to form nl < Nl
CGTOs. A set such as 6-31G, suitable for compounds with first and second
row elements, contracts six primitive GTOs for inner shells and treats outer
shells with a contraction of three primitives plus another primitive with a smal-
ler exponent parameter ζ, for a more spread out orbital. Parameters are
obtained from atomic HF calculations. Hydrogen and He orbitals of 31G type
have exponentials further adjusted for molecular calculations.
 6.1 Many-Electron States 207

The introduction of primitives with several ζ values allows for electronic

rearrangement during molecular interactions where electronic densities
move closer to or further from atomic positions as interatomic distances
change. In addition, it is usually necessary to describe polarization of the elec-
tron clouds as it occurs in induction and dispersion forces. This requires add-
ing primitives with values of l larger than that for the highest occupied shells,
such as adding p-type outer shell AOs to a Na atom in a molecular interac-
tion. The basis sets are then called DZP, TZP, and so on, with additional
explicit nomenclature to indicate whether the polarization orbitals are added
only for atoms in the second and higher rows of the periodic table or also for
hydrogen atoms [4].
With so many considerations relevant to the choice of a basis set, it is con-
venient to devise a systematic method for improvement of a basis set. One
such method is to construct atomic natural orbitals (or ANOs) from accurate
atomic wavefunctions [4]. The procedure starts with diagonalization of the
first-order density matrix for each isolated atom as obtained from an accurate
treatment of its electron correlation, usually by means of an atomic configu-
ration-interaction calculation. This leads to ANOs with decreasing occupa-
tion numbers as measures of their importance. The ordering shows which
orbitals are more important to describe electronic correlation energy and
polarization of the atomic cloud. Convergence of atomic properties may
require a large number of ANOs, even though it can be expected that features
of molecular interactions can be accurately described with a relatively small
set of orbitals similar to the ANOs.
More compact basis sets of orbitals have been derived as alternatives to
ANOs, constructed to provide accurate atomic correlation energies, and
are called correlation-consistent (or cc) basis sets, or correlation- and polar-
ization-consistent (cc-p) basis sets [4, 16, 17]. When done for valence shells,
they are called cc-pVDZ, cc-pVTZ, and so on. Basis sets of this sort are suit-
able for calculations of intermolecular forces, while the ones that do not
contain polarization functions fail to account for induction and dispersion
forces. In addition, accurate calculation of intermolecular forces at large
distances may require a treatment including diffuse (widespread) electron
distributions of A and B. This can be done adding to the basis set other func-
tions with smaller exponent parameters ζ for each angular momentum
quantum number l. The resulting basis sets are labeled with aug as in
aug-cc-pVDZ.
Accurate calculations of intermolecular forces involve differences of large
energies for the pair AB minus energies of A and B, giving small potential energy
V(R) values changing with the relative distance R between A and B and also with
their internal conformations given by coordinates QA and QB. This requires that
the energy terms must have comparable accuracy. Two procedures must be
implemented to achieve this. One is that the energies of systems AB, A and
208 6 Many-Electron Treatments

B must be calculated with the same type of basis set. The other relates to the size
of the basis sets for the complex and for each component. Indicating with
ES[R, QS; {χ S}] an electronic energy obtained at relative distance R with the basis
set {χ S} suitable for species S = A, B, and AB, it would appear as if the potential
energies should be EAB[R, QA, QB; {χ A} {χ B}], obtained from the union of the
component basis sets, minus EA[{χ A}] and EB[{χ B}] from each separate basis set.
However, this means that at each distance R, the total energy is being calculated
with a more extensive (or more complete) basis set for the complex and less so
for the components, giving an inaccurate potential energy V (R, QA, QB). The
resulting error is called the basis set superposition error. It can be corrected
by calculating the components energy with the larger (union) basis set at the
conformations of the components for each distance R, as a counterpoise (CP)
correction energy
ΔECP(QA,QB) = EA QA ; χ A χ B + EB QB ; χ A χB
−EA χ A − EB χ B
giving a more accurate potential energy V(R, QA, QB) = V (R, QA, QB) +
ΔECP(QA, QB) [4, 18].
Basis sets suitable for extended systems frequently use instead plane wave
expansions for valence shells and delocalized orbitals, introduced together
with a description of inner shells in terms of atomic pseudopotentials [19].
−1 2
Plane waves ϕk r = Ω exp ik r for a system of volume Ω with a set of
wavevectors k forming a grid k n ,n = n1 ,n2 ,n3 , in reciprocal space, must
be dense enough to describe spatial variations of the electronic density. The
B B
upper limit Nj of the grid numbers nj, in 1 ≤ nj ≤ Nj , is increased to account
for highly oscillating orbitals in real space. This provides a basis set suitable for
calculations of interaction forces in large systems, and properties such as the
total energy of a crystalline solid as a function of the size of its unit cell volume.
Inner- and outer-shell atomic orbitals are differently treated, with inner-shell
PS
orbitals ϕl r introduced to construct atomic pseudopotential operators
PS
V l in which valence electrons move [20]. Expansions of one-electron wave-
functions in plane wave functions, and fast Fourier transformations between
grids in real and reciprocal spaces, provide one-electron valence states and ener-
gies for many-electron treatments of intermolecular forces in solids and at solid
surfaces. The same grid in reciprocal space can be used for a total system AB and
for its isolated components A and B to obtain accurate energy differences.
a
Expanding MOs ψ j r as linear combinations of AOs χ μ r , with Nbf basis
functions for atom 1 ≤ a ≤ Nat and a total number of basis functions Nbf,
means that
Nbf
ψj r = c χ
μ = 1 jμ μ
r
 6.2 Supermolecule Methods 209

and that MO integrals ψ j a 1 ψ k and ψ j ψ k a 2 ψ l ψ m for one- and two-

electron operators are sums with AO matrix elements

d 1 d 1 χ μ 1 ∗ a 1 1, 1 χ ν 1 = μ a 1 ν

d 1d 1 d 2 d 2 χ κ 1 ∗ χ μ 2 ∗ a 2 1, 2; 1 ,2 χ λ 1 χ ν 2 = κμ a 2 λν

which depend on two or four AO indices. This allows an estimation of how

computing times and data storage scale with the number Nbf of basis functions,
also considering invariance under permutation of integration variables, which
gives Nbf2/2 and Nbf4/8 for the number of integrals of one- and two-electron
operators. Transformation of two-electron integrals from AOs to MOs requires
adding over four loops with the transformation coefficients multiplying AO
integrals. This can be done in a sequence of four summations, each scaling
as Nbf5. It is therefore important to work with efficient basis sets to decrease
computing times and data storage for two-electron integrals. The scaling power
of Nbf can be decreased for local two-electron operators, such as the Coulomb
e-e energy, introducing criteria to discard integrals when the overlap products
χ κ(1)∗χ μ(1) are very small for distant atomic nuclei. This brings down the effec-
tive scaling for two-electron integrals to below Nbf3.

6.2 Supermolecule Methods

6.2.1 The Configuration Interaction Procedure for Molecular
Potential Energies
Here we present only an outline of the CI method as it relates to the calculation
of intermolecular forces. Extensive presentations can be found in references [5,
15, 21, 22]. Starting with supermolecular spin orbitals (SOs) as products of space
η
and spin functions, ψ j r n ,ζ n = φj r n η ζn = n j , with the notation
n = r n ,ζn for the space and spin variables of electron n = 1 to N, and
η = α, β for spin up and down states, a single reference state is given by an elec-
tronic determinantal wavefunction D0 1, 2,…,N = ψ 1 1 ψ 2 2 …ψ N N =
N −1 2 det n j for N electrons and occupied SOs j = 1 to N, also shown
in a short notation as D0 = ψ 1ψ 2…ψ N . As written, the SO ψ j displays a differ-
η
ent orbital factor φj for different spins. In applications to closed-shell systems,
with an even number of electrons, electrons with up and down spins are located
at the same space orbitals, and this leads to a wavefunction with good total spin
quantum numbers S = 0, MS = 0. But for different orbitals for different spins, the
unrestricted determinant does not generate an eigenfunction of total spin.
210 6 Many-Electron Treatments

However, application of the spin projection operator SMS gives a spin-adapted

configuration state function (or CSF) ΦSMS 0 = SMS D0 , which is in general a
combination of determinantal functions. The reference CSF for given spin
quantum numbers will be called Φ0.
Other CSFs are generated by removal of an occupied SO i and addition of an
unoccupied SO a from each determinant in the CSF to form a new singly excited
state Φia = c†a ci Φ0 , written in terms of fermion creation and annihilation
operators c†i ,ci with the anticommutation relations
c†j ci + ci c†j = δij , c†j c†i + c†i c†j = 0, cj ci + ci cj = 0

Doubly excited CSFs Φijab , and generally n-excited ones, can be similarly
constructed. A CI wavefunction is given by the combination of the reference
CSF plus singly excited (S) ones, plus doubly (D) excited ones, and so on. This
can be described within the wave operator formalism of Section 6.1.2, starting
with the projector = Φ0 Φ0 valid for a single reference state. Single exci-
1
tation determinants add up to a wave operator term Ω , doubly excited ones
2
add to Ω and so on, leading to

1 2
Ψ= I +Ω +Ω + Φ0

1 2
Ω = i, a
Aia c†a ci , Ω = i, j, a, b
Aijab c†b c†a cj ci , …

where the coefficients Aab…

ij… are functions of the energy E and must be found for
each intermolecular distance.
1 2
Keeping only Ω and Ω in the expansion leads to the CISD treatment. This
is sufficient for many applications at finite distances, but would not satisfy the
asymptotic consistency test. Indeed, if one treats the species A and B to the same
1 2 S
accuracy, with Ψ S = I + Ω +Ω + Φ0 , S = A, B, for each species, one
S
finds that the asymptotic product ΨAΨB contains also wave operator terms of
3 1 2 2 1 4 2 2
the form Ω = ΩA ΩB + ΩA ΩB and Ω = ΩA ΩB , which are absent in
the total wavefunction Ψ as constructed. This inconsistency can be avoided
N
by including all excitations in the expansion of Ψ up to Ω , the last nonzero
operator with the largest number of excitations. This generates the so-called
full-CI wavefunction, which, however, is usually too demanding of computing
time for applications.
To make calculations more affordable and yet asymptotically consistent, a
valid procedure is to use a complete active space self-consistent field (or
CASSCF) treatment [23]. Here we again consider, to begin with, the case of a
 6.2 Supermolecule Methods 211

single reference CSF constructed from SOs ψ j r n ,ζ n = ψ j n obtained now as

self-consistent field (SCF) solutions of the Hartree–Fock (or HF) equation
HF
f HF n ψ j n = ϵj ψ j n , f HF = h + J − K

HF
where f HF is the one-electron Fock operator and ϵj is the energy of the SO
[3]. The term h contains the electron n kinetic energy operator and the Cou-
lomb attraction energy of the electron to all the nuclei (or atomic ion cores).
The Coulomb and exchange HF operators are given by
occ 2
J n ψj n = k
d 1 v 1, n ψ k 1 ψj n

occ
K n ψj n = k
d 1 ψ k 1 ∗ v 1,n t 1n ψ k 1 ψ j n

with v(1, n) the Coulomb energy between electrons 1 and n, and t 1n the trans-
position operator exchanging the variables of electrons 1 and n in the electron
integrals: t 1n ψ k 1 ψ j n = ψ k n ψ j 1 . The integrals here are over space vari-
ables and also indicate sums over spin variables ζ = ± 1. The SO energies are
given by
HF occ
ϵj = j h j + k
jk v 1 − t jk

where alternatively jk v 1 − t jk = jk jk . A special case (RHF) follows

when the orbitals are doubly occupied with up and down spins and integrals
are over only space variables, in which case there are twice as many Coulomb
integrals as exchange integrals.
For the ground electronic state at each distance R, a reference determinantal
function will usually contain doubly occupied orbitals with up and down spins
for the lower energies, and singly occupied orbitals of higher energy, chosen to
generate the CSF with correct spin quantum numbers. The HF equation gives
not only physically meaningful occupied SOs for the supermolecule, but also
excited (although usually unphysical) SOs, which can be used to form a basis
set for expansion of the kinetic energy and Coulomb energy terms in the
Hamiltonian.
The SCF SOs can be obtained expanding them in a basis set of atomic SOs
χ μ r,ζ , both factorized into space and spin parts. Introducing the expansions
ψj = μaμjχ μ, the Fock operator matrix elements fμν = χ μ f χ ν and overlaps
Sμν = χ μ χ ν , the one-electron equations are solved finding the expansion coef-
ficients that satisfy
faj = ϵj Saj
212 6 Many-Electron Treatments

where aj is a column matrix with elements aμj for each supermolecular SO j.

This is self-consistently done for the occupied SOs with j = 1 to N, which then
provide the Fock matrix elements in an equation that can be solved for excited
(unoccupied) SOs, also called virtual orbitals.
The HF treatment can be applied to two interacting closed-shell species, such
as He + He or Ar + H2, by choosing as the reference state a determinantal func-
tion with doubly occupied molecular orbitals giving a restricted HF (or RHF)
wavefunction. Such a determinant can properly describe their physical ground
state also as R ∞. However, that choice would not be acceptable for a system
with open shells, such as H + H or H + Li. Then a doubly occupied determinant
like DRHF = σα, σβ for two valence electrons, with σ = cA χ A + cB χ B a
linear combination of atomic orbitals with the symmetry of the diatomic, and
with the same MO for up and down spins, contains a spurious determinantal
term at large distances corresponding to H+ + H− or Li+ + H−, and gives the
wrong asymptotic energy. This problem can be partly solved introducing an
unrestricted determinant in the unrestricted HF (or UHF) treatment. In this
example, it amounts to using orbitals σ η = cAη χ A + cBη χ B , with η = α, β to
form DUHF = σ αα, σ ββ . The four coefficients must satisfy normalization and
orthogonality of the two MOs. This leaves one free parameter to optimize
the UHF energy, and its value changes with R to represent the electronic rear-
rangement taking place as R grows. The resulting orbital densities are found to
be different for up and down spins, allowing each electron to move toward one
of the two nuclei, and the pair then breaks up into the correct ground state of the
neutral species.
The UHF determinant is not an eigenfunction of the total spin operators, but
this can be extracted with the corresponding projection operator in a postsym-
metry procedure. Alternatively, the projector can be first applied to a determi-
nantal function to construct an unrestricted CSF, to be optimized. Results of
calculations for H2 are shown in Figure 6.2. The UHF behavior is qualitatively
correct but must be improved to agree with accurate results, using CI or PT
treatments.
An alternative choice to UHF for open-shell systems is the restricted open-
shell HF (or ROHF) choice, where most electrons are located in doubly occu-
pied orbitals, which are the same for up and down spin. Remaining electrons are
located in orbitals that may depend on the spin orientation, in which case the
ROHF wavefunction is not an eigenfunction of the total spin operators, and its
spin components must be extracted by means of a projection operator, or con-
structed from a superposition of determinants.
The RHF and UHF procedures can be applied to a supermolecule as needed to
assure that the breakup product A + B at large distances is in the proper physical
states. Returning to the CI treatment, this can make use of SOs optimized with
RHF or UHF to construct excited configurations. The orbitals in the supermo-
lecule basis set are classified as inactive if they remain doubly occupied as the
 6.2 Supermolecule Methods 213

0.06

0.04

0.02
0.6 1.0 1.4 1.8 2.2 2.6 3.0 3.4 R (a.u.)
0.00

R RHF
–0.02
E(H2) – 2E(H) (a.u.)

UHF
–0.04

–0.06

Exact
–0.08
(Kolos–Wolniewicz)

–0.10

–0.12

–0.14

–0.16

–0.18

Figure 6.2 Results from RHF, UHF, and CI calculations of potential energy versus
interatomic distance for H2, using the basis set 6-31G∗∗. Source: from Ref. [5]. Reproduced
with permission of Dove Publishing.

molecular system changes conformation, typically core orbitals, and as active

SOs if they have single occupancy or are unoccupied to begin with at each
conformation.
A full-CI expansion is generated by promotion of electrons from all active
occupied SOs to all unoccupied and partly occupied ones. Doing this for all con-
formations involves all possible n-excitation wave operator terms and guaran-
tees that a supermolecule at intermediate distances and its asymptotic
molecular fragments are described with the same accuracy. The many-electron
equation HΨ = EΨ containing single excited CSFs Φia = c†a ci Φ0 , and similarly for
214 6 Many-Electron Treatments

higher excitations, must be solved for the lowest energy E and the associ-
ated state

Ψ = Φ0 + i, a
Aia Φia + i, j, a, b
Aijab Φijab + = ΦA

in a matrix notation with the row matrix Φ of CSFs and the column matrix A of
expansion coefficients. This can be done searching for the variational minimum
of Ψ H Ψ = A† Φ H Φ A = A† HA subject to the normalization Ψ
Ψ = A† Φ Φ A = 1 in a well-known procedure leading to the matrix eigen-
value equation (H − EI)A = 0, which needs to be solved only for the lowest
eigenvalue as a function of conformations and in particular as a function of
R, if one wants only the ground state PES.
A more flexible and accurate procedure is the multiconfiguration SCF (or
MCSCF) treatment where the CI coefficients A and also the HF coefficients
a are varied to achieve self-consistency at the lowest energy. This gives orbitals
adapted to optimize not only the reference configuration, but also useful to opti-
mize its mixture with excited ones, and leads to more realistic excited SOs for
the supermolecule.
The CASSCF expansion is a MCSCF treatment involving all the electron exci-
tations among active SOs, and guarantees that the potential energy V(R) from the
difference of total energies goes properly to zero at large distances, provided the
reference CSF Φ0 correctly describes the asymptotic states of A and B at large R.
The limitation here is that the number of configurations to be generated increases
exponentially with the number of active SOs. In fact, if Nact is the number of active
orbitals, N the number of active electrons, and S the total spin quantum number,
the number of complete active space functions to be generated is [23]
2S + 1 Nact + 1
NCAS =
Nact + 1 N 2 −S Nact + 1 − N 2 + S
Nact + 1
×
N 2+S+1 Nact + 1 − N 2 − S − 1
a very large number when Nact is larger than 10, which prevents calculations of
this type for large molecular systems or large electronic excitation.
The calculation of interaction potential energies gets more demanding when
there are state degeneracies or electronically excited states, or where a single
CSF Φ0 does not properly describe the asymptotic A and B states. In these cases,
one must use a treatment based on multireference states. Instead of a single ref-
erence CSF Φ0, one must consider a set {ΦJ, J = 1 to ND} of CSFs, introduce the
projection operator = J ΦJ ΦJ , and proceed to construct multireference
CI (or MR-CI) states, to calculate energies.
An example is provided by the treatment of H2 H + H in its ground elec-
tronic state 1 Σg+ with a minimal MR-CI containing only two configurations
 6.2 Supermolecule Methods 215

made up of diatomic orbitals σ g,u = Cg,u( 1sA ± 1sB ) constructed from

hydrogenic 1 second orbitals of A and B atoms, where g, u stand for gerade
and ungerade symmetry. The configurations Φ1 = σ gα, σ gβ and Φ2 = σ uα, σ uβ
are two determinants that can be taken as a multireference pair interacting to
give the correct neutral species asymptotically, with a ground (G) state

ΨG = A1G σ g α, σ g β + A2G σ u α,σ u β ≈ CG 1sA α,1sB β + 1sA β, 1sB α

insofar as A1G ≈ − A2G for R ∞, while a treatment based only on Φ1 leads to

an incorrect mixture containing a state of H+ + H−(1s2) in addition to the cor-
rect H(1s) + H(1s) state. Alternatively, the H2 H + H process can be described
with a single unrestricted determinantal function, which employs different orbi-
tals for different spins, with parameters to be varied.
The formalism presented above, based on the multistate projection operator,
shows how one can proceed to generate needed excited CSFs from the multi-
reference ones by transferring electrons from active occupied SOs into unoccu-
pied ones. The treatment has been made feasible by introducing energy
selection criteria so that only excited CSFs that are energetically coupled to
within a selected energy range are included in the expansion. This so-called
MRD-CI procedure has produced many potential energy surfaces of high accu-
racy [24]. The limitations of the MR-CI treatment, on account of the formal
absence of asymptotic consistency, have been covered in reference [25].

6.2.2 Perturbation Expansions

Many-body perturbation theory (MBPT) and its applications to an N-electron
system are based on the standard perturbation theory for a nondegenerate or a
0
degenerate reference state, and a choice of a zeroth-order Hamiltonian H that
0
allows construction of a basis set ΦJ of unperturbed states. Details of the
general theory can be found in references [3–5, 25]. The treatment can be
derived for a single reference determinantal wavefunction constructed from
SOs, D0 1, 2, …, N = ψ 1 1 ψ 2 2 …ψ N N obtained from f ψ j = ϵj ψ j , where f
is a convenient one-electron effective operator giving physically meaningful
energies ϵj, SOs ψ j orthonormalized with ψ j ψ k = δjk, and a suitable density
of occupied and unoccupied states for an expansion basis set. A standard choice
for f is the mentioned Fock operator f HF , which provides physically acceptable
occupied states but usually unsuitable unoccupied states. An alternative is the
f KS = h + J − vxc Kohn–Sham (KS) operator, which can be chosen to generate the
SOs of density functional theory [1]. Other basis sets may involve the occupied
HF SOs and a set of complimentary (orthogonalized) basis functions suitable for
the treatment of excited SOs [14].
216 6 Many-Electron Treatments

The unperturbed (zeroth-order) Hamiltonian is H 0 = F = 1≤m≤Nf m with

0
the eigenstate equation FD0 = E0 D0 and the zeroth-order energy
0 occ
E0 = j
ϵj

obtained from one-electron energies of occupied SOs. This is a sum of approx-

imate ionization energies and does not correspond physically to the total energy,
but must be corrected by higher order energy terms. The perturbation operator
1 2
H = H + H − F is the fluctuation (or correlation) potential energy operator
r
containing one- and two-electron terms and giving perturbation energies E0 of
1 2
order r. The PT generates the terms in Ψ = I + Ω +Ω + D0 and
0 1 2
E = E0 + E0 + + E0
by recurrence at each order, using for convenience
the intermediate normalization D0 Ψ = 1. These can be obtained from D0
and a set of determinants DJ = Dia , Dijab , …, with single, double, …, excitations.
The energies to first and second orders of a many-body PT (or MBPT) can be
MB
obtained for a pair of species A and B from a wave operator ΩAB and are
given by
1 occ occ
E0 = D0 H D0 = j
j h j −ϵj + j<k
jk v 1 − t jk
2
D0 H DJ
2
E0 = J 0 E0 − EJ
1 2 2
occ unocc
D0 H +H Dia
= i a ϵi −ϵa
2 2
occ unocc
D0 H Dijab
+ i<j a<b ϵi + ϵj − ϵa − ϵb
where t is the transposition operator exchanging the variables of electrons 1 and
2 in the electron integrals so that t 12 ψ j 1 ψ k 2 = ψ j 2 ψ k 1 , occupied SO indi-
ces run between 1 and N, and unoccupied ones between N + 1 and the largest
value in the basis set. The numerators can be evaluated in terms of SO using the
Condon–Slater rules for matrix elements of operators between determinantal
wavefunctions [1].
These and higher order energies can be analyzed in terms of Feynman dia-
grams and organized to contain only linked diagrams in each order. The wave-
function is then written as
∞ k
ΨMBPT = D0 + k =1
R0 H D0 = exp T cnc D0
link
 6.2 Supermolecule Methods 217

where R0 is the unperturbed resolvent, the summation is restricted to linked

Feynman diagrams, and the operator T cnc is a sum of connected terms. The cor-
responding MBPT energy then satisfies the size extensivity condition to all
orders [25].
As a special case, it is possible to consider the Moller-Plesset PT (or MPPT)
where the unperturbed Hamiltonian is the Fock operator, a sum of one-electron
effective Hamiltonians f = f HF , with
f HF m = h m + J m − K m
containing the HF Coulomb J) and exchange (K operators. The zeroth-order
energy
0 occ HF
E0 = j
ϵj

and the first-order correction

1 1 occ
E0 = − jk v 1 − t jk
2 j<k

0 1 HF
reconstruct the HF energy, E0 + E0 = E0 , a meaningful total energy.
The first summation in the second-order energy involving single excited
1 2
states is zero in the HF treatment, insofar as D0 H +H Dia =
a f HF i = 0, per the so-called Brillouin theorem [4].
The remaining summation in the second-order energy gives the leading cor-
relation energy involving excited SOs. However, the excited (unoccupied) SOs
generated in HF correspond formally to an electron moving in a field of N elec-
trons, instead of N-1 electrons as happens for the occupied SOs. As a result, they
usually describe electrons that are less bound and are more delocalized than
accurate excited SOs. Calculation of interaction energies may require a larger
basis set and inclusion of higher orders in the MPPT expansion, as compared
to expansions with more realistic excited SOs, or otherwise they may lead to
large errors in intermolecular forces at large distances. Furthermore perturba-
tion terms must be carefully constructed to avoid unlinked diagrams and non-
extensive results.
The asymptotic consistency requirement E ≈ EA + EB for a pair of species A
and B, essential for intermolecular forces, is not necessarily satisfied by MBPT.
This is because the reference state must satisfy D0 ≈ D0, A D0, B , a correct
asymptotic reference state made up of products of states of A and B, and in
MB
addition, the perturbation treatment must construct the wave operator ΩAB
MB MB MB
so that the energy as obtained from H AB ΩAB D0 ≈ EA + EB ΩA ΩB
D0, A D0, B gives accurate energies for A and B. This must be verified in
a treatment of the intermolecular forces.
218 6 Many-Electron Treatments

Another aspect affecting the accuracy of intermolecular forces relates to the

order of perturbation for the pair AB and each component. For example, if a
treatment is done up to second order in the MPPT2 treatment, one finds that
MB
each operator ΩS ,S = A, B will introduce two-electron excitations so that the
MB MB
product ΩA ΩB will generate up to four excitations. However, the operator
MB
ΩAB to the same order would only involve double excitations and would there-
fore miss some of the asymptotic states. This is avoided in the MBPT treatment
based on linked diagrams and the connected T cnc operator, such as the coupled-
cluster treatment, which consistently generates each perturbation order for the
pair AB.

6.2.3 Coupled-Cluster Expansions

Consistency in the treatment of excitations can be achieved within the coupled
cluster (CC) treatment [25, 26]. Here the wave operator takes an exponential
form and one writes
CC
Ψ=Ω D0 = exp T D0
1 2 N n
with the excitation operator T = T + T + + T where T contains only
n electronic excitations, and with the intermediate normalization D0 Ψ = 1.
Expansion of the exponential leads to the appearance of terms like
n m n m
T T for integers 0 ≤ m ≤ ∞, involving 1 ≤ (n. m) + (n . m ) ≤ N
excitations since no more than N SOs may be excited. The excitation operator
n
terms T can be expanded in a basis set of SOs classified as occupied or unoc-
cupied, introducing excitation amplitudes for each term as in
1 occ unocc a † 2 occ unocc ab † †
T = i a
ti c a c i , T = i<j
t c c c c ,…
a < b ij b a j i

One finds, with the expansion of the exponential operator, a sum of config-
urations as in the CI method, but here their superposition is generated by the
CC excitation terms in a convenient way, which involves fewer transition ampli-
tudes than the number of coefficients in the CI. The unknowns in the CC treat-
ment are E, tia , tijab ,…, and these can be extracted by projecting (or taking
moments of ) the equation exp −T Hexp T D0 = E D0 on the Dirac bras
D0 , Dia , Dijab ,…. This gives the same number of nonlinear algebraic
equations as there are unknowns, which can be solved by iteration. An analysis
of many-electron terms in E = D0 exp − T Hexp T D0 after an expansion
in commutators of H and T shows that only linked diagrams appear, assuring
size extensivity [25].
 6.2 Supermolecule Methods 219

Relating to asymptotic behavior, there are two convenient consequences of

the exponential form in the CC treatment. In the limit for R ∞, one finds that
CC CC CC
T AB ≈ T A + T B and ΩAB ≈ ΩA ΩB . This together with a choice of refer-
ence state satisfying D0 ≈ D0, A D0, B warrants that the potential energy
of interaction is V(R) = E(R) − (EA + EB) ≈ 0 as desired. It may require using an
unrestricted determinantal function with different orbitals for different spins,
although for a special system such as H2 a RHF reference determinant is suffi-
cient. Secondly, choosing a level of expansion such as CCSD, with single and
double excitations included in the excitation operator, it follows that all excita-
tions, up to the maximum N, NA and NB, will be respectively present in the
wavefunctions of the pair AB and the fragments A and B, so that the energy dif-
ferences should be better balanced than in the MPPT treatment. These reasons
make the CC treatment a more reliable procedure for the calculation of inter-
molecular forces using supermolecule SOs.
1 2
The expansion containing only T and T produces the CCSD treatment,
which gives accurate energies for many molecules near their equilibrium con-
3
formation. A better treatment also adds T to generate triplet excitations and
CCSDT wavefunctions and energies. However, this level of calculations is very
demanding of computer time and data storage, and an alternative is to incor-
porate only part of the triple excitations by means of perturbation theory, in
a CCSD(T) treatment. This improves potential energies over wider ranges of
R around equilibrium, but can lead to erroneous results at large R when exci-
tation energies appearing in denominators of the third-order PT become very
small at large R. The asymptotic behavior can be corrected using an UHF ref-
erence state, as shown in Figure 6.3 for N2.
The CC approach has been extended to excited states by means of the equa-
tion-of-motion CC (or EOM-CC) treatment. It preserves the role of the non-
Hermitian Hamiltonian operator exp − T Hexp T = H, introduced to obtain
the ground state Ψ0 of energy E0, in the calculation of excited states ΨK = RK Ψ0
with energies EK. These states can be generated by excitation operators

RK = r 0 I + r a c† c
i, a i a i
+ r ab c† c† c c +
i, j, a, b ij b a j i

where the terms are assumed to apply in the normal order (of creation operator
followed by annihilation operator to the right) as done with the terms in the
excitation operator T . Since operators in RK and T are all of excitation type,
this gives the commutator RK ,T = 0. It follows from H RK Ψ0 =
EK RK Ψ0 and from Ψ0 = exp T D0 that [25]

H,RK D0 = ΔEK − ΔE0 RK D0

220 6 Many-Electron Treatments

–108.95

–109.00

–109.05

–109.10
E (Hartree)

–109.15

–109.20

FCI
CCSD(T)
CCSDT
CCSDTQ
–109.25 UHF-CCSD(T)
UHF-CCSDT
UHF-CCSDTQ
UHF-CCSDTQP
UHF-CCSDTQPH

–109.30
1.5 2.0 2.5 3.0 3.5 4.0 4.5
R (a.u.)

Figure 6.3 Potential energy curves versus interatomic distance for the ground state of the N2
molecule. Results from CC are compared to full CI using a cc-pVDZ basis set. Incorrect
behavior of CC at large distances can be corrected using a UHF reference state. Source: from
Ref. [27]. Reproduced with permission of American Physical Society.

where relative energies are defined by ΔEK = EK − Eref with Eref = D0 H D0 .

Solving this equation for ΔEK and for the coefficients in RK as functions of inter-
molecular distances provides PESs for excited states. This is an active area under
development, involving excited states of open-shell systems and bond dissoci-
ation and related computational algorithms [27–31].
The CC derivations above have assumed that only a single reference state
is needed. When a reference state is degenerate or close to others in energy,
one can return to the treatment using the more general wave operator in
 6.2 Supermolecule Methods 221

Ψ = ΩΦ D = + RE H Φ D including a multireference projection

operator = J ΦJ ΦJ = J J for many-electron states J = 1 to ND. This
has been the subject of extensions of the CC treatment [32] and has been com-
putationally implemented [25]. The multireference set of states is made up of
determinants with double or single occupancy. The MR-CC extension consists
of introducing partial excitation operators called here T J to be applied to each
state ΦJ in the multireference set. Each of them contains a sum over many-
electron excitations. The wave operator is then of the form
Ω= J
exp T J J

and its application into the subspace can be done by first defining an effective
eff
Hamiltonian H operating only within the subspace with matrix elements
eff
HKJ = ΦK HΩ ΦJ = ΦK H exp T J ΦJ . The perturbation corrected
energies EJ in HΨJ = EJ ΨJ , J = 1 to ND , and excitation amplitudes like tia J in
each partial excitation operator follow from projections of the kets
eff
exp − T J H exp T J ΦJ = K
exp −T J exp T K ΦK HKJ
eff
onto the bras ΦJ , Φia , Φijab , …, and diagonalization of HKJ with a linear
transformation in the subspace.
This procedure allows consideration of dissociation and intermolecular forces
in the AB system when the subspace contains all the determinantal functions
needed to construct correct asymptotic states of A and B. For example, in the
case of H2 H + H, it is convenient to choose the subspace as made of states
with the configurations Φ1 = σ gα, σ gβ and Φ2 = σ uα, σ uβ , which are close in
energy for large internuclear distances. The MR-CC procedure can then give the
correct potential energy surface at all distances. Dissociation in other diatomic
systems including multiple bonds has been reviewed in reference [27]. Accurate
results over all interatomic distances have been obtained for F2, N2, and HF.
Results so far indicate that using a MR-CC approach, or alternatively using
an UHF reference state, provides ways to correctly treat bond dissociation. This
is an area of theoretical research yet under development.
The scaling of MBPT and CC calculations with the size Nbf of the basis set can
be described separating the MO basis set into two sets of occupied and unoc-
o u
cupied MO functions numbering Nbf ≤ N and Nbf , respectively. Excitation
operators fix the indices of occupied MOs so that transformations from the AO
basis to the MO basis involve fewer summation loops. Transforming a four
2
index (two-electron) integral like D0 H Dijab in MBPT or tijab in a CC treat-
u 2
ment, for fixed i and j, involves summations scaling only as Nbf , which must
222 6 Many-Electron Treatments

State
quality

Full CI

MCSCF

CASSCF
S
CCSDT PE
of
SAPT cy
ura
Acc
CCSD(T)

MBPT4

CCSD

CISD

MBT2 Basis
set size
DZP TZP ... ccVDZP cc-pVTZP aug-cc-pVDZP

Figure 6.4 Diagram with increasing basis set quality, containing polarization terms, going
from left to right on the horizontal axis, and increasing many-electron quality going up as
needed for the calculation of accurate potential energy surfaces (PESs).

2
o
be done Nbf times, an effort much smaller than (Nbf)4 for a general four-
index quantity. Analysis of the scaling in CCSD(T) shows it growing as
o 3 u 4
Nbf Nbf . Further analysis of scaling is found in the literature [4, 15, 25].
The quality of electronic structure calculations for intermolecular forces
depends on the type of treatment being used and the size of the basis set in
expansions. Figure 6.4 shows what treatments of electronic structure account
for asymptotic and size consistency and what basis sets are large enough to
account for electron transfer, electronic localization, and polarization.

6.3 Many-Atom Methods

6.3.1 The Generalized Valence-Bond Method
Alternatives to the supermolecule approach are provided by treatments that
start from the states and SOs of species A and B before interaction, and con-
struct the potential energies of the A + B pair from them. Electrons are assumed
to be localized around species A and B separately, and their charge densities are
 6.3 Many-Atom Methods 223

then allowed to distort, polarize, or transfer between species. One of these treat-
ments derives from the valence-bond (or VB) method for electronic structure,
generalized to deal with electron transfer and polarization. The GVB method
has been mathematically developed, and more recently has been reintroduced
as a viable computational method [2, 33, 34]. It is appealing for studies of inter-
molecular forces because it provides a direct route to construction of reference
states with the correct asymptotic behavior.
Total wavefunctions for N electrons are written as products of a spin function
J
ΘSMS ζ1 ,…,ζN of spin variables for quantum numbers (S, MS) times a space
function of position variables FJ r 1 ,…,r N for many-electron states J, describing
single or multiple bonds with paired electrons in the case of bonding between A
and B, or describing their nonbonding interactions. The space function is con-
structed from atomic orbitals of each species and the pairing of spin and space
functions is done so that as R ∞, one finds the correct asymptotic limit fac-
torized, with the total wavefunction
J
Φ 1, …,N = F
J J
1, …, N ΘSMS 1, …,N
K L
≈ KL
CKL FK 1, …, NA ΘSA , MSA 1, …,NA FL 1, …, NB ΘSB , MSB 1, …, NB

In a simple example with one valence electron per species A and B in atomic
orbitals (AOs) χ p r 1 and χ q r 2 forming a single bond, the standard covalent
VB wavefunction, written here for a homonuclear diatomic, is the single bond
function with F(cov)(1, 2) = χ p(1)χ q(2) and
χp 1 χq 2 + χp 2 χq 1
Φ cov 1, 2 = 1 2
Θ0, 0 1,2
2
2 + 2 Sab
Θ0, 0 ζ1 ,ζ 2 = α ζ1 β ζ 2 − α ζ 2 β ζ 1 21 2

with Spq = χ p χ q ≈ 0 the overlap integral of the two nonorthogonal AOs, going
to zero at large distances. This wavefunction clearly gives the correct asymptotic
behavior.
When the interacting species dissociate instead into ions with the two elec-
trons at the same site, the two-electron state contain F(ion)(1, 2) = χ u(1)χ u(2),
u = p, q, and is

χp 1 χp 2 + χq 1 χq 2
Φ ion 1, 2 = 1 2
Θ0, 0 1,2
2
2 + 2 Spq

again giving the correct limit. An accurate treatment of the potential energy for
all distances R combines the cov and ion states.
A pair of species with multiple bonds can be similarly described with a product
of bond functions constructed with the correct total electronic antisymmetry.
224 6 Many-Electron Treatments

Generalizations can be derived using more flexible orbitals ϕμ, which are yet cen-
tered at each atom or fragment μ, but have each AO mixed with the orbitals of
other fragments, present as small components. Another generalization involves
using all the spin functions, which can be generated for N electron spins, instead
of just the spin-pairing per bond.
The generalized VB (or GVB) treatment [33] replaces the AO χ p with a
more flexible localized orbital ϕp written as a linear combination of functions
with coefficients varied to optimize the system energy. It can be chosen to be a
combination of AOs all centered at the same fragment A, to account for
hybridization and polarization, or it can be a combination of orbitals at
center A with some mixing of orbitals at B to account for electron transfer
between them. Furthermore, orbitals localized at A can be the original ones
nonorthogonal to those of B, or they can be transformed by symmetric ortho-
normalization into orbitals predominantly centered at A or B [14]. Orthonor-
malization allows for simpler expressions in matrix elements of operators, but
dilutes the physical significance of spin pairing of orbitals in the description
of bonds.
The simple spin pairing shown above in Θ0,0(1, 2) is only one of several
pairings of electron spin that are acceptable in polyatomic systems with multiple
bonds. The treatment can be made rigorous introducing a set of N-electron
J
spin functions ΘSMS 1, …,N for given quantum numbers (S, MS) and making
use of properties of the symmetric group N of all permutation operators
N of electron variables in N-variable functions [2]. Insofar as these permuta-
tions and the projection operator SMS for given S, MS commute, it follows that
J K
N ΘSMS = K ΘSMS PJK , with PJK a matrix element of an irreducible matrix rep-
resentation of the symmetric group [2]. These irreducible representations and
their characters are well known, and they can be used to construct all the needed
spin functions for each MS. The space functions FJ(1, …, N) multiplying the spin
functions in the expansion of total states Φ can be written as products of N loca-
lized orbitals ϕu, u = a, b, of a basis set chosen to account for electronic
rearrangements.
Among the possible total states, the ones relevant to intermolecular forces go
A B A B
asymptotically to products ΦK ΦL with ΦK and ΦL states of the species A
and B constructed from the same orbitals. The number of spin functions created
in the unitary group procedure above is

2S + 1 N
fSN =
N 2 + S + 1 N 2− S

and it can get quite large for a large number N of active electrons, so that this
general procedure is in practice limited to systems with few active electrons for
all the intermolecular conformations of interest in A + B.
 6.3 Many-Atom Methods 225

6.3.2 Symmetry-Adapted Perturbation Theory

It is appealing to derive a treatment of molecular interactions at all relative dis-
A B
tances using component states like ΦK and ΦL , eigenstates of the Hamilto-
nians H A and H B , and functions of electron variables 1, …, NA and NA + 1, …,
NA + NB, respectively. They are the ones that appear at large distances and
are likely to be good approximations to component states also at intermediate
distances when one or both are closed-shell systems and their bonding is
nonexistent or weak. The main obstacle to such treatments, however, is that
the total Hamiltonian H for the pair AB has symmetries different from those
of the component Hamiltonians. The most essential is electronic antisymmetri-
AB
zation of the total states ΦJ , which for the pair AB must include not only com-
NS NS S S
ponent antisymmetrizers S ,S = A,B, satisfying S ΦK = ΦK , but also a
AS
complimentary antisymmetrizer = NA NB N −1 = NA NB N −1

I− containing electron transposition operators in between A and B.

N NA NB
The total antisymmetrizer = A B provides the symmetry-
AB A B
adapted ΦJ = ΦK ΦL . In addition, the total Hamiltonian has spatial
and spin symmetries generally different from those of the component Hamilto-
nians. Therefore, a PT treatment with the Hamiltonian partitioning
int
H = H 0 + H ,H 0 = H A + H B ,H = H AB , where H 0 is the unperturbed Hamilto-
nian, must be done explicitly imposing the total symmetry on the wavefunction
perturbation terms. This can be achieved by total postsymmetrization after a PT
has been done with only component symmetries, in a so-called weak symmetry-
adapted PT (SAPT), or more rigorously by total presymmetrization before the
perturbation expansion is done, in a strong SAPT [35, 36].
The total wavefunction can be obtained from the previously introduced wave
operator in

Ψ = ΩΦ D = + RE H ΦD
−1
with R E = α + E −H , which is as written valid for any partition of
the total Hamiltonian and implies an intermediate normalization. Provided the
multireference projection operator commutes with total symmetry projection
operators , which can be done by choosing a basis set of symmetry-adapted
total functions, this wave operator can be used to generate all perturbation
orders for the wavefunction and the energy, by recurrence. Expanding the resol-
vent R E in terms of the unperturbed resolvent R0 E0 , for the noninteracting
int 1
A + B pair with total energy E0, and using the perturbation H AB = λH AB [11],
the resulting Rayleigh–Schrodinger expansion of energy and wavefunction in
powers of λ can then be implemented computationally.
226 6 Many-Electron Treatments

At large distances between the components, it is possible to ignore the com-

plementary electron transpositions and to obtain electrostatic, induction, and
dispersion energies from PT as done in previous chapters. But the transposi-
tions must be accounted for if one wants to have smooth potential
energies as the distance R is decreased, maintaining the same PT for each
conformation of the AB pair. Indicating results from unsymmetrized PT as
Edir(R), the total energy with a generalized intermediate normalization is
E R = ΦD H Ψ N D = Edir R + Eexch R with the first term, also called
Epol, obtained from the perturbation treatment excluding the transpositions in
, and the second term arising from the effects of transpositions. The potential
energy for a many-atom conformation (R, QA, QB) is given by

V R,QA ,QB = E R,QA ,QB − EA QA + EB QB

= Vdir R, QA ,QB + Vexch R,QA ,QB

The direct and exchange energies have been analyzed in detail and developed
for computational work. To simplify here we consider a single reference state
A B A B
Φ0 Φ0 of energy E0 = E0 + E0 . The direct, or “pol,” term contains to first
order the classical electrostatic (or Coulomb) interaction energy and to second
order induction and dispersion energies,
1 2
Vdir AB = Velst AB ,Vdir AB = Vind A + Vind B + Vdisp AB
The corresponding terms in the exchange Vexch are gathered from perturba-
tion terms containing transpositions for each order in the perturbation as
1 2
Vexch AB = Vexch −elst AB , Vexch AB = Vexch −ind AB + Vexch −disp AB
where Vexch − elst is a correction to the classical Coulomb interaction containing
the effect of electron exchange. The second-order energies involving sums over
excited states of A and B can be rewritten in terms of their dynamical polariz-
abilities as done in the Casimir–Polder treatment of dispersion forces [37].
Higher order terms for direct and exchange energies have been similarly ana-
lyzed [35, 36].
The implementation of this SAPT to calculate intermolecular forces requires
S
in addition a procedure for correcting the component wavefunctions Φ0 and
S
energies E0 , S = A, B, when as usual these have been obtained in approximate
treatments for many-electron systems. They are intramonomer electronic cor-
relation corrections that can be incorporated by means of perturbation treat-
S
ments of each species. Starting with known eigenfunctions ΦK of an
S
approximate Hamiltonian F S with eigenenergies EK , an additional perturba-
tion treatment can be done with F = F A + F B as the zeroth-order Hamiltonian
 6.3 Many-Atom Methods 227

and the difference W = H 0 − F = W A + W B as the perturbation energy operator.

Usual choices for F are the Hartree–Fock Hamiltonian or a sum of one-electron
Hamiltonians derived from density functional theory. In these cases, the oper-
ator W is a kind of correlation energy giving corrections beyond HF or DFT
mean values. This intramonomer PT can be done separately for each species,
but it is more accurate to keep the same energy term order for both of them,
insofar they are equally present in energy differences. The whole computational
int 1
procedure therefore involves a double PT with perturbations H AB = κH AB and
1
W = λW giving expansions in the products of order parameters κ kλl, with k
and l integers. The double PT gives energies
K L
V R = k =1 l=0
κ k λl V k, l R

where K is the highest affordable perturbation order for the molecular interac-
tion energy, usually K = 3, and L is the largest order in the intramolecular per-
turbation, which within a CC treatment can include some multiple excitations
of the species A and B to all orders.
The SAPT combined with HF molecular functions, in SAPT(HF) treatments,
involves some simplifications resulting from the absence of singly excited
k, 1
matrix elements, insofar as a f HF i = 0, so that Vdir R = 0 Electronic
exchange between A and B and correlation in A and B, however, distort their
HF charge distributions and add to the intramolecular corrections. The
SAPT(HF) treatment has been successfully applied to closed-shell systems
such as (H2O)N, and Ar + H2O. They have been obtained going to second order
(K = 2) in the intermolecular interaction, which scales as Nbf7 with the number of
basis functions [36, 38]. The formalism is also suitable for interactions involving
open-shell species provided unrestricted HF states are chosen as reference ones
in a SAPT(UHF) treatment.
When F is chosen from a DFT treatment, the present PT is named
SAPT(DFT). This has been found to give accurate results provided that:
(i) the second-order interaction energies are obtained from dynamical polariz-
abilities, which correctly account for the coupling of one-electron KS excita-
tions, as in time-dependent DFT, called SAPT(CKS); (ii) high-quality basis
sets are used for atomic orbitals; and (iii) density functionals are preferably of
the hybrid type containing some exact long range electron exchange energy.
Results have been published for dimers (He)2, (Ne)2, (H2O)2 and (CO2)2 [39],
and also for the benzene dimers in a variety of conformations.
Nonadditivity in three-body systems has also been considered within SAPT,
which extends to intermediate and shorter distances the arguments presented
in Chapter 3 for long-range nonadditivity of pair interactions, by incorporation
of electron exchange [40]. The nonadditive terms originate already in first-order
228 6 Many-Electron Treatments

1 2
PT due to exchange, with Enon −add ABC = Eexch ABC + Eind ABC +
2 2 3 3
Eind −exch ABC + Edsp−exch ABC + Eind ABC + Edsp ABC + where each
term is an intrinsic three-body function. Combining SAPT with DFT, it has
been possible to calculate nonadditive energy terms for trimers of He, Ar, water,
and benzene [41].
In addition to giving accurate results for a variety of AB pairs and ABC triplets,
the SAPT treatment provides a systematic procedure for analyzing electrostatic,
induction, and dispersion contributions to intermolecular forces as functions of
atomic conformations, including electron exchange, even for many-body sys-
tems. This is useful in the development of approximations where accurate elec-
tronic energies are calculated first at short and intermediate intermolecular
distances, to which long-range induction and dispersion energies are added a
posteriori to cover all distances. In these treatments, it is necessary to avoid
overcounting electronic correlation energies, which are sometimes duplicated
in initial and added terms, something that can be avoided with a consistent
SAPT treatment over all intermolecular distances.

6.4 The Density Functional Approach to

Intermolecular Forces
6.4.1 Functionals for Interacting Closed- and Open-Shell Molecules
The original density functional theory, stating that the ground state energy of an
electronic system is a unique functional E ρ r of the total electronic density
ρ r at each position r [42], has been applied to the calculation of molecular
properties and to intermolecular forces involving one or two closed-shell
species, where chemical bond formation does not occur, within the KS and
Gordon–Kim treatments as described in previous chapters [43–45]. Those
treatments, involving local density (LD) functionals, have been subsequently
extended with semilocal functionals by including the density gradient and local
kinetic energy, and with hybrid functionals combining the exact Hartree–Fock
exchange energy with an approximate local or semilocal form. They give rea-
sonably accurate descriptions of short- and intermediate-range distances but
do not account for the long-range vdW forces, which physically arise from elec-
tronic density fluctuations within the interacting species. The long-range forces
can be added as corrections, but this must be done introducing short-range
damping factors multiplying the added dispersion energies in a physically mean-
ingful way. A more satisfactory procedure is to extend local or semilocal DFT
treatments to include nonlocal functional terms in a consistent way. This must
be done, however, while avoiding double counting the vdW forces, which are
partly present in semilocal functionals at intermediate distances and might
be erroneously added again when introducing long-range nonlocal terms in a
functional.
 6.4 The Density Functional Approach to Intermolecular Forces 229

Improvements to the semilocal treatments are needed to deal with dissoci-

ation of bonds between two open-shell species and with the long-range inter-
actions. These come from: (i) separating up and down electronic spin
densities, to be able to describe the asymptotic limit when a bond is dissociated
and the two fragments are open-shell systems with different numbers of up
and down spins; (ii) correcting the uniform gas expressions to account for
the inhomogeneous nature of the densities; (iii) requiring correct behavior
of the exchange one-electron potential to avoid electron self-interaction
errors and incorrect one-electron potential energies at large interelectronic
distances; and (iv) allowing for local electronic density fluctuations at large
distances. These improvements have been covered in many reviews, and here
the focus is on aspects directly related to intermolecular forces as they change
with distances all the way to separation of nonbonding species or to breaking
of bonds.
Recent developments relevant to molecular interactions have climbed the
accuracy ladder of functionals from the original LD approximation involving
only the electronic density ρ r , up to the generalized gradient approximation
(GGA) including the gradient ∇ρ r , and more recently the meta-GGA, which
also includes a local kinetic energy density τ r constructed from local elec-
tronic momenta. The resulting functionals can generally be called semilocal
(or sl). Also relevant are the hybrid functionals, which mix the exchange energy
sl
Ex as a semilocal functional with the accurate Hartree–Fock exchange energy
HF
Ex , in ways that can provide better thermochemical energies [46], correct
electronic excitation energies, crystal lattice constants and band gaps in
solids [47, 48], or asymptotically correct electronic long-range potential
energies [49–51].
The original treatment by Kohn and Sham (K–S) in terms of density ampli-
tudes (or KS spin-orbitals) ψ j r, ζ has been generalized to include interactions
of open-shell species where bonds are formed or broken, by instead introducing
a ground state functional of a local 2 × 2 spin density matrix ρηη r with spin
indices η = α, β, that is a unique functional provided the external potential
ext
energy matrix in which the electrons move is local, of the type Vηη r [52].
A simplified version with a diagonal external potential energy matrix was imple-
mented at the level of a local spin density functional (LSDF) of spin densities
ρηη r = ρ η r and was shown to correctly describe the dissociation of a H2
molecule into two ground state H atoms. This is done optimizing the LSDs
so that ρ α r and ρ β r go asymptotically into separate electronic densities
located at the H atoms as needed, something that is not possible using a single
LD. The formalism with an LSDF has been applied also to other molecules
to calculate their spectral properties and their dissociation into open-shell
species [53].
230 6 Many-Electron Treatments

The LSDF approach describes near-range interaction energies and also long-
range electrostatic and induction forces, but not the dispersion (or vdW) forces
that arise from electronic density fluctuations and require introduction of
dynamical susceptibilities. Efforts to extend DFT to include dispersion forces
have involved recasting the long range electron–electron interactions in terms
of charge density operators and dynamical susceptibilities for each species A
and B, and approximating the susceptibilities in a variety of ways. This provides
a treatment suitable for extended systems, which contains as special cases vdW
interactions derived from multipolar polarizabilities present in the susceptibil-
ities. We consider in what follows a case where electronic charge distributions of
A and B do not overlap, avoiding electronic exchange or transfer, and only inter-
act through Coulomb forces.
Dispersion energy functions of intermolecular and intramolecular position
variables can be combined with near-range functionals obtained from DFT
and containing all energies (electrostatic, exchange, near-range correlation,
long-range induction) except for the vdW dispersion energy. This is a very
active area of research where several DFT-plus-vdW treatments have been pro-
posed and tested by comparison with accurate results from CCSD(T) or SAPT
calculations for interaction potentials V(R, Q). Here the focus is on methods
applicable to large classes of chemical systems, described as many-atom struc-
tures A and B containing sets of atoms {a} and {b} with their centers of mass at a
relative distance R in a body-fixed reference frame, and with internal variables
Q = Ra ,Rb referred to their C-of-M.
We have seen in a previous chapter that the dispersion energy can be sepa-
rately treated and calculated from the dynamical susceptibilities of the two spe-
cies, or from their multipolar polarizabilities. The aim here is to add them to the
semilocal energies while avoiding double counting of long-range interactions
and to ascertain how important these are for development of potential energy
surfaces valid at all distances. A variety of ways to do this can be termed DFT-D
treatments, with the D label signifying that dispersion energies are to be added
to DFT results. This has led recently to treatments where the semilocal DFT
energies E(sl)(R, Q) at near-range distances (rapidly decreasing at large dis-
tances) and the nonlocal dispersion (vdW) energies Edsp(R, Q) at large distances
are combined to give total energies as
sl nl
E R,Q = E R,Q + Edsp R,Q fd R,Q

for all intermolecular distances. It restricts the dispersion contribution to the

long range by multiplying it times a suitable damping function fd (R, Q), whereas
before fd (R) ≈ 1, for R ∞ and fd (R) ≈ 0, for R 0 [54–57]. An alternative that
provides a more clear-cut separation of a dispersionless functional energy and
an energy obtained from the long-range dispersion energy uses results from
accurate SAPT to identify dispersion-like contributions to a density functional.
 6.4 The Density Functional Approach to Intermolecular Forces 231

Subtraction of those contributions provides a dispersionless density functional

(or dlDF) to which a long-range, separately calculated dispersion energy with an
appropriate cut-off, can be added. It is followed by fitting energy parameters to
reproduce interaction energies in groups of compounds [58]. This, however,
requires additional calculations done with the SAPT many-electron approach
as a preliminary task. Details on the DFT-D treatments relying on calculations
at the sl–functional level for two interacting many-atom systems are presented
in a following chapter.
An alternative to the explicit addition of long-range vdW energies to semilo-
cal energies is to develop heavily parametrized energy functionals containing
one-electron terms that can be adjusted to reproduce vdW energies at large dis-
tances. Functionals have been recently introduced with one-electron semilocal
sl
potential energy terms vxc r constructed to include long-range interaction
energies, also combined with meta-hybrid GGA functionals [59]. Hybrid func-
tionals contain a chosen amount of orbital-dependent exchange energy as
derived in the Hartree–Fock treatment, combined with the standard DFT
exchange as a functional of density, and are described in Section 6.4.3 [46].
Another approach has introduced, within functionals, additional one-electron
dispersion-corrected atom-centered potentials (DCACPs) v ECP r, r =
v sl r δ r − r + v nl r, r containing both semilocal and nonlocal terms
[60]. In both treatments, the one-electron potential energies have been
parametrized and tested by reproducing accurately known dispersion
energies for atom pairs and have been used in calculations for large many-atom
systems.
Semilocal potentials for electrons of a given spin σ = α, β depend not only on
the electronic density and its derivatives (as in the generalized gradient or GGA
approximations) that can describe long-range electrostatic and induction ener-
gies, but depend also on local electron momentum densities
KS KS KS
− iℏ∇φjσ r = π jσ r , derivatives of KS orbitals φjσ of a given spin compo-
nent. This describes to some extent density fluctuations as they appear in inter-
action dispersion energies. Exchange-correlation functionals are of the form
xc, σ ρ, s,τ with s r = ∇ρ 2kF ρ r a dimensionless function of the density
1 3
gradient ∇ρ r , where kF r = 3π 2 ρ r is a local Fermi wavenumber, and
KS 2
with τσ = j π jσr 2me a kinetic energy density. Fitting parameters in
the total energy functional tot, σ ρ, s,τ to reproduce values in large sets of com-
pound properties can then provide some of the polarization (induction and dis-
persion) interactions around equilibrium atomic conformations, including
nonbonding structures and molecular crystals. This program has been imple-
mented, for example, within the M06 [59] and MN12 [61] software packages,
which have also been extended in hybrid versions with screened electronic
exchange suitable for applications to large molecules and solids [62].
232 6 Many-Electron Treatments

6.4.2 Electronic Exchange and Correlation from the

Adiabatic-Connection Relation
Yet another approach introduces nonlocal functionals based on the adiabatic
connection fluctuation-dissipation (ACFD) relation giving a formally exact
expression for the electron–electron correlation energy in terms of the dynam-
ical susceptibility. It leads to the introduction of a nonlocal expression for the
sl nl
correlation energy. This can be written as Ec R,Q = Ec R,Q + Ec R,Q with
the first term coming from a semilocal functional and the second term given by
a nonlocal functional modified by subtraction of the homogeneous-electron
correlation component to avoid double counting it. This second term can be
constructed from the susceptibilities as a known functional of the electronic
density so that no parametrization from molecular properties is needed. It goes
at large distances to the known form of the dispersion energy in terms of sus-
ceptibilities, and in addition, it can be obtained as a smooth function of distance
for the near-range without introduction of a separate damping function. The
approach has generated a treatment called vdW-DF in various versions, suitable
for applications to large as well as small molecules and to physisorption and
chemisorption at solid surfaces [63–65]. It has also been implemented with a
more detailed and accurate parametrization of dynamical polarizabilities in
the so-called VV10 treatment [65].
It is of interest to summarize the ACFD formalism as it provides a compact
form for the exact exchange-correlation energy and may open a route to new
nonlocal density functionals. That can be done by means of the adiabatic con-
nection theorem of many-electron theory, closely related to the Hellmann–
Feynman (or H–F) theorem for molecular forces. The total Hamiltonian for
interacting electrons can be redefined to apply to an electron–electron Cou-
lomb energy where the squared charge c2e is replaced by λc2e , with a parameter
0 ≤ λ ≤ 1. This can be interpreted also as the interaction of electrons at far
positions r λ for small λ 1, coming in as the parameter increases to its value
1 2
λ = 1 for the physical system. The total Hamiltonian is H λ = H + λH , a sum
of one-electron and two-electron terms, with
1 2
H = h ,
1≤m≤N m
H = 1 2 c2
m n e
r m −r n

where hm contains the kinetic energy of the m-th electron, its attraction energy to
positive ion cores, and the energy coming from the repulsion among the ions,
with the latter independent of electron variables. When the one-electron term
also contains some mean-field contribution from electron–electron interactions,
2
such as a functional of electron density, this must be subtracted from λH .
Differentiating the eigenstate equation H λ Ψλ = Eλ Ψλ with respect to λ and
projecting on the normalized state Ψλ, we find as in Chapter 5 that
 6.4 The Density Functional Approach to Intermolecular Forces 233

2
∂Eλ ∂λ = Ψλ H Ψλ , the H–F relation. Its integral form for 0 ≤ λ ≤ 1 can be
1
written considering that Eλ = 0 = Es , the sum of single-electron kinetic and
nuclear attraction energies, and E = Eλ = 1 is the exact energy, so that the integral
form of the H–F relation is
1
2
E = Es 1 + dλ Ψλ H Ψλ
0
The exact energy E = Eλ = 1 can be expressed in terms of a two-electron density
2
ρλ 1,2; 1, 2 obtained as a function of the λ-parameter values. To derive its
2
form, one can start with the expression for H written in terms of the density
operator

ρ1 r = 1≤m≤N
δ r −rm

where the electron position is being treated as an operator and one performs
integrals over the electron variables in r. The double summation in
2
Ψλ H Ψλ must be constructed while avoiding the interaction of each elec-
tron with itself. To this effect, first write
2 c2e
H = 1 2 d 3 r1 d 3 r2 δ r 1 −rm δ r 2 −rn
r 1 −r 2 m n

m n
δ r 1 −rm δ r 2 −rn = m, n
δ r 1 −rm δ r 2 −rn

−δ r 1 −r 2 m
δ r 1 −rm

where the double sum to the right includes m = n. This gives

m n
δ r 1 −rm δ r 2 −rn = ρ 1 r 1 ρ 1 r 2 − δ r 1 −r 2 ρ 1 r 1

to be replaced in the double integral. Its expectation value in the state Ψλ can
2
also be obtained from the two-electron density function ρλ 1, 2; 1,2 as
previously done for λ = 1, with the short-hand notation 1 = r 1 , ζ1 including
space and spin variables. For the spin-independent electron–electron interac-
tion energy, a summation over spin variables can be carried out and the
expectation value is found to contain the two-electron pair density
2 2
Pλ r 1 ,r 2 = ζ1 , ζ2 ρλ 1, 2; 1, 2 in

2 c2e 2
Ψλ H Ψλ = 1 2 d 3 r1 d 3 r2 P r 1 ,r 2
r 1 −r 2 λ
2
Pλ r 1 ,r 2 = Ψλ ρ 1 r 1 ρ 1 r 2 Ψλ
234 6 Many-Electron Treatments

− δ r 1 −r 2 Ψλ ρ 1 r 1 Ψλ

From this and the integral form of the H–F relation, it follows that
1
1 c2e 2
E = Es 1 + dλ d 3 r1 d 3 r2 Pλ r 1 ,r 2
2 0 r 1 −r 2
This can be compared to the Hartree energy that contains the classical elec-
tron–electron interaction as
1
1 1 c2e 1
EH = Es 1 + dλ d 3 r1 d 3 r2 Pλ r1 Pλ r 2
2 0 r 1 −r 2
1
where Pλ r = Ψλ ρ 1 r Ψλ . The difference E − EH = Exc is the exchange-
correlation energy of DFT, which is therefore given by
1
1 c2e 2 1 1
Exc = dλ d 3 r1 d 3 r2 Pλ r 1 ,r 2 −Pλ r 1 Pλ r2
2 0 r 1 −r 2
Further introducing the density fluctuation operators
1
Δλ ρ 1 r = ρ 1 r − Pλ r
2
and the spatial correlation ΔPλ r 1 ,r 2 , one finds the equivalent expression
1
1 c2e 2 1
Exc = dλ d 3 r1 d 3 r2 ΔPλ r 1 ,r 2 −δ r 1 −r 2 Pλ r1
2 0 r 1 −r 2
2
ΔPλ r 1 ,r 2 = Ψλ Δλ ρ 1 r 1 Δλ ρ 1 r 2 Ψλ
2
so that ΔPλ r 1 ,r 2 is the spatial correlation function of two density
fluctuations.
The next step in the proof of the ACDF theorem relates the correlation of
fluctuations of a pair of electrons to the dynamical susceptibility of the pair
2 2 1
[66]. The function Pλ, xc r 1 ,r 2 = ΔPλ r 1 ,r 2 −δ r 1 −r 2 Pλ r 1 can be re-
expressed in terms of a dynamical susceptibility χ λ, xc r 1 ,r 2 ; ω of the many-
electron system, dependent on the frequency ω, as an integral over an imagi-
nary-valued argument iω, to obtain an expression used within DFT [52, 53,
63, 67, 68].
The relation follows from response theory applied to the calculation of
changes in ρ 1 r,t induced by an external electric potential ϕext r , t starting
at time t = 0 and null before then, as it couples to the density with energy
1
H ext t = d 3 r ϕext r ,t ρ 1 r ,t and with the system initially in its ground
electronic eigenstate 0λ of the Hamiltonian operator H λ . This has been done
 6.4 The Density Functional Approach to Intermolecular Forces 235

for the response of a molecular charge density in Chapter 2, solving for an

excited state Ψλ(t), which can be used here to write the density change
1
Δλ ρ 1 r,t = Ψλ t ρ1 r Ψλ t − Pλ r as

t
Δλ ρ 1 r,t = d 3 r dt χ λ r, t; r ,t ϕext r ,t
0

χ λ r,t; r ,t = − i ℏ 0λ ρ 1 r,t ρ 1 r ,t −ρ 1 r , t ρ 1 r, t 0λ

where χ λ is a scaled response function containing ρ 1 r,t = exp i ℏ H λ t

ρ 1 r,0 exp − i ℏ H λ t , a Hermitian operator. It has the same form if rewrit-
ten in terms of the density fluctuations Δλ ρ 1 instead of the densities ρ 1 . The
lower time integration limit can be extended to −∞, and insofar as the ground
state is stationary, the susceptibility depends only on τ = t − t . It can also be
written as an imaginary-part form
2
χ λ r, r ; τ = ℐm 0λ Δλ ρ 1 r,τ Δλ ρ 1 r ,0 0λ
ℏ
in a compact expression. Its Fourier transform is
∞
χ λ r, r ; ω = dτ χ λ r, r ; τ exp iωτ
−∞

It contains the Fourier transformed Δλ ρ 1 r,ω operator in

χ λ r, r ; ω = 2 ℏ ℐm 0λ Δλ ρ 1 r,ω Δλ ρ 1 r ,0 0λ

The calculation of density changes is conveniently done introducing a

+
retarded (time-forward) susceptibility χ λ τ = θ τ χ λ τ where θ(τ) is the step
function null for negative arguments. This gives
∞
+
Δλ ρ 1 r,t = d 3 r dτ χ λ r, r ; τ ϕext r ,t −τ
−∞

A Fourier transform from τ to ω of this time convolution leads to

+
Δλ ρ 1
r,ω = d 3 r χ λ r, r ; ω ϕ ext r ,ω

+ ∞ +
showing that χ λ r, r ; ω = − ∞ dτ χ λ r, r ; τ exp iωτ is the scaled dynam-
ical susceptibility, related to the transform of the response through
+ ∞
χ λ r, r ; ω = 0 dω χ λ r, r ; ω−ω θ ω . Therefore the dynamical
response χ λ r, r ; ω provides both total energies and also the time evolution
of the electronic density.
236 6 Many-Electron Treatments

The function χ λ r, r ; ω can be related to the density-fluctuation correlation

2
ΔPλ r 1 , r 2 appearing in the energy, in a series of steps starting with an expan-
sion of Δλ ρ 1 r,τ in the basis set { Jλ } of eigenstates of H λ with energies { Eλ, J }
[68], designed to extract 0 λ Δλ ρ 1 r Δλ ρ 1 r 0λ from an integral of
χ λ r, r ; ω over frequencies. Introducing the identity decomposition
I = J Jλ Jλ between the operators in the response function χ λ, the matrix
element

0λ Δλ ρ 1 r,τ Jλ = exp i Eλ, 0 − Eλ, J τ ℏ 0λ Δλ ρ 1 r,0 Jλ

can be Fourier transformed and introduced in χ λ r, r ; ω to obtain its imagi-

nary part. An integration over frequencies of χ λ r, r ; iω leads to
∞
ℏ
− dω χ λ r, r ; iω = 0λ Δλ ρ 1 r, 0 Δλ ρ 1 r , 0 0λ
π 0

as needed. This relation can alternatively be derived from the analytical proper-
ties of χ λ r, r ; iω in the complex frequency plane and its integration in a
closed path [66].
The final expression for exchange correlation is then
∞
1 1
c2e ℏ
Exc = dλ d 3 r1 d 3 r2 − dω χ λ r 1 ,r 2 ; iω
2 0 r 1 −r 2 π 0

1
− δ r 1 −r 2 Pλ r1

This allows calculation of Exc from the dynamical susceptibility, and if the lat-
1
ter is constructed as a functional of the electron density ρλ r = Pλ r , then a
formally exact expression has been obtained for the functional Exc ρλ r .
Applications to the calculation of intermolecular forces between species A
and B at relative distance R require a knowledge of χ λ and Exc over each distance
R and in particular for large R. In this case, it is known that the dispersion energy
Edsp(R) is given in terms of susceptibilities χ S r, r ; iω of S = A, B by a Casi-
mir–Polder expression, and Exc R; ρλ r must correctly give Edsp(R) at large
distances. An important distinction in the two asymptotic forms (the one above
and Casimir–Polder’s) is that each spatial integral in the above Exc extends over
all space including both species, while the spatial integrals in the Casimir–
Polder expression extend only into nonoverlapping regions containing A or
B. The connection can be done introducing localized basis functions around
A and B, to partition the spatial integrals in Exc. This is presented in some detail
in the next chapter on extended systems.
 6.4 The Density Functional Approach to Intermolecular Forces 237

The connection of the ACDF relation with DFT was pointed out and used in
early molecular calculations [52] and reviewed [53] some time ago and has been
the basis for the vdW–DF formulation of density functional, which has provided
many results in applications to complexes involving molecules and surfaces,
adsorbates, and their potential energy changes with distances [55, 63, 69].
One such result for two interacting benzene molecules in a complex with par-
allel planar structures is shown in Figure 6.5. Here the semilocal (near-range)
energies were obtained from generalized gradient density functional as shown,
and the vdW interaction was added as described in a vdW–DF treatment.
1 1
The factorization Δλ ρ 1 r = ρ 1 r − Pλ r = Pλ r Δλ φ r , which intro-
duces a relative density fluctuation operator Δλ φ r at all distances R between
2 1 2 1
A and B, leads to the expression ΔPλ r 1 , r 2 ; R = Pλ r 1 Φλ, xc r 1 , r 2 ; R Pλ r2
2
where the two-electron function Φλ, xc
r 1 , r 2 ; R accounts for exchange correla-
tion arising from density fluctuations at each relative distance R. This factori-
zation can be used to bring Exc to a form similar to the one developed to
account for long-range vdW interactions in terms of local polarizabilities,

5
CCSD(T)
4 MP2
vdW-DF
3 GGA(revPBE)
Interaction energy (kcal mol−1)

GGA(PW91)
2

–1

–2

–3

–4
3 3.5 4 4.5
Separation (Å)

Figure 6.5 Interaction energy versus distance between the planes of two benzene molecules
in a “sandwich” conformation. Source: from Ref. [63]. Reproduced with permission of Institute
of Physics.
238 6 Many-Electron Treatments

but here we have an expression that is applicable to all distances and contains
both semilocal and nonlocal contributions to the energy. Several treatments
have been based on this expression, which can in principle provide a seamless
connection between the long-range dispersion energy given by a Casimir–
Polder integral, and semilocal DFT energies for intermediate and short ranges
[55, 65, 67]. This is presently an active area of research, and the following chap-
ters present some related aspects relevant to interaction energies between
extended systems and at solid surface.

6.4.3 Issues with DFT, and the Alternative Optimized

Effective Potential Approach
The absence of a constructive proof for the density functional has led to the
development of many versions of DFT. The many density functionals in the
literature have been subjected to continuous tests and improvements within
two procedures: imposing a large number of constrains on the functionals; and
comparing DFT calculated properties of a large set of molecules with accurate
results generated by many-electron methods including correlation. An exam-
ple of imposition of many constrains is found in the construction of a recent
strongly constrained functional called SCAN [70]. Several comparisons of
DFT molecular structures and properties, and of interaction energies of
molecular pairs, have been done with known values from experiment or from
accurate non-DFT calculations, collected for several groups of compounds
[59, 71–73]. Comparisons of properties such as vibrational frequencies of
molecules, and polarizabilities of molecular dimers, that are relevant to calcu-
lation of intermolecular forces, give some idea of which functionals may work
for potential energy surfaces and which ones can be discarded. Generally, the
more accurate functionals are meta-GGA functionals containing a hybrid of
semilocal exchange and Hartree–Fock exchange terms. Much, however,
remains to be done to improve the accuracy of DFT methods for molecular
interactions.
The original Hohenberg–Kohn treatment of functionals is rigorous for the
ground electronic state of a nondegenerate energy level and for the lowest
nondegenerate energy of a given symmetry. It is therefore limited to states with
zero or maximal total electronic spin when the Hamiltonian is spin-free. Its
generalization including spin densities in addition to charge densities allows
treatments of open-shell molecular pairs in states with a good projection quan-
tum number MS for the electronic spin, but the electronic states do not corre-
spond to good total spin quantum numbers S [52]. A way around this limitation
is to construct a determinantal wavefunction with the K–S orbitals of lowest
energy and to project out of it the state with the correct spin quantum numbers
using the projector SMS introduced in Section 6.1, in a postsymmetrization
procedure.
 6.4 The Density Functional Approach to Intermolecular Forces 239

The issue of degeneracies or near-degeneracies can be treated within DFT

introducing a functional of a statistically averaged density function ρW r con-
structed as a weighted sum over a set of interacting states of similar energy. The
functional can be chosen to be the grand-canonical ensemble of a system of
electrons under thermodynamical constrains given by a medium at temperature
T = (kBβ)−1 and chemical potential μ [74], or can be built as an average over
energies E1, E2, … ED with weights W = {wd, d = 1 to D} [43, 75]. In each case,
the many-electron states are given, within a DFT treatment, as determinantal
KS
wavefunctions with K–S orbitals ψ d, j and the resulting average density func-
tion is a K–S sum
D ∞ 2
KS
ρW r = d=1
w ψ
j = 1 d, j d, j
r

with fractional occupation numbers wd, j of K–S orbitals, instead of the previous
occupation numbers of 1 or 0. These procedures provide average energies as
functions of interatomic distances, and they can be extended to include pertur-
bation energies arising from external fields. Properties relevant to intermolec-
ular forces, such as electric dipoles and polarizabilities, energy gradients, and
vibrational frequencies can then be obtained by taking derivatives of perturbed
energies with respect to applied fields [4].
An important concern in DFT treatments of intermolecular forces relates to the
appearance of electron self-interaction errors resulting from approximations in
density functionals. In a wavefunction treatment of a many-electron system at
the level of the Hartree–Fock approximation, it becomes clear that terms in
the HF one-electron potential energy operator, in the notation of Section 6.1,
occ ∗
v HF 1 = J 1 −K 1 = k
d 1 ψk 1 v 1 , 1 1 − t1 1 ψ k 1

HF
containing the Coulomb and exchange energy operators vCoul = J and
HF
vx = − K partly cancel so that an electron in an occupied orbital, generated
by HF or KS methods, does not interact with itself, and also that asymptotically
HF HF
for large r1, vCoul 1 ψ l 1 ≈ ψ l 1 Ne c2e r1 while vx 1 ψ l 1 ≈ −ψ l 1 c2e r1 , for
occupied ψ l, with the sum correctly containing Ne − 1 c2e r − 1 . But when the
exchange-correlation functional Exc[ρ] is approximated, that cancellation does
not occur because the one-electron DFT exchange potential vx(1) = δEx/δρ(1)
does not cancel the Coulomb self-interaction in the DFT one-electron potential
energy vCoul(1). As a consequence, the KS one-electron potential energy
v KS r = vCoul r + vx r + vc r deviates from the correct Ne − 1 c2e r − 1 behav-
ior at large distances and is unphysical near each ion core. This affects the values
of molecular potential energies for dissociation and charge transfer and proper-
ties like energy-band gaps in semiconductors and polarization interactions in
molecular crystals. Calculations of potential energy surfaces must be done using
240 6 Many-Electron Treatments

functionals that at least decrease the self-interaction error and preferably lead to
correct one-electron potential functions over all distances.
Another complication in the application of DFT to intermolecular forces
arises from the continuous dependence of the energy on the electron density
ρ r , which integrates to the total number of electrons Ne for two interaction
species A and B. When these interact and exchange electrons, as in the disso-
ciation of a compound AB into A+ + B–, the density must split at large distances
as ρ r ≈ ρ A r + ρ B r with each term integrating to the correct number of
A B
electrons, Ne and Ne . However, this is not assured in DFT, where a func-
tional of the AB pair density may lead to asymptotic densities integrating to
S
noninteger values Ne ± δNe of electrons and to unphysical charges in the frag-
ments S = A and B. This problem is related to the absence of discontinuities in
functionals as electron densities are changed, since the energy functionals
change continuously with electronic density variations, while physically one
must instead have sharp energy changes for varying integer number of electrons
in each fragment. This may lead to inaccurate calculations for electron ioniza-
tion energies and for proton affinities.
At a more formal level, the appearance of fractional numbers of electrons when
bonds are broken can be reinterpreted by means of an ensemble DFT (or EDFT),
where a fractional number N is considered to result from a weighted average of
energy functionals for two species with integer number of electrons N and N + 1.
The ensemble density is chosen as ρ = 1− ν ρN + νρN + 1 , with an interpolating
variable ν obtained from a variational procedure [76]. This approach provides a
foundation for chemical reactivity theory with a consistent use of chemical reac-
tivity indices [43] for species with physically correct integer number of
electrons.
The heavily parametrized functionals already mentioned in connection with
DFT calculations, including dispersion energies, have also been constructed to
minimize errors from electronic self-energy interactions and from continuous
variations of the number of electrons, and provide an alternative treatment giv-
ing more accurate dissociation energies, proton affinities, reaction barriers, and
vibrational constants [59, 77].
Two approaches have been developed to correct for self-interaction and to
impose the correct asymptotic limit for the electron potential energy function.
They are the introduction of hybrid functionals containing the exact exchange
potential energy, and treatments based on functionals of orbitals obtained from
an optimized effective potential (or OEP) with the correct asymptotic behavior.
Here they are briefly reviewed as they relate to molecular interactions.
Hybrid exchange-correlation functionals containing a mixture of DFT and
HF exchange functionals take the form

Exc = ExcLSD + a0 ExHF − ExLSD + ax ΔExGGA + ac ΔEcGGA

 6.4 The Density Functional Approach to Intermolecular Forces 241

with three coefficients a chosen to fit sets of known results, and LSD and
GGA signifying the local spin-density and generalized gradient approximations.
They were introduced and tested for thermodynamically relevant properties
like atomization energies, ionization and proton affinity energies [46]. This
approach was further developed to avoid fitting parameters [78] and to include
a dependence of the functional on the electronic kinetic energy density [48]
again free of adjustable parameters. This is potentially more useful, for the cal-
culation of intermolecular forces, than using parametrized functionals with
parameters fit to properties of molecules around equilibrium interatomic posi-
tions, insofar as in principle parameters are physically expected to change with
intermolecular distances.
HF
The calculation of Ex can be very demanding for extended systems such as
solids and biomolecules. An efficient procedure has been developed to decrease
HF
the computational demands for Ex . The electron–electron Coulomb interac-
tion is separated for this purpose into short-range and long-range e–e dis-
tances as
1 1− erf ωr erf ωr
= +
r r r
in terms of the error function erf, which goes from zero at r = 0 to one at large
distances with a slope controlled by the parameter ω, so that the first term to the
right is short-range (SR) and the the second term is long-range (LR). The semi-
local (sl) exchange energy can be separated into ωSR and ωLR terms and can be
HF , SR HF , SR sl, ωSR
combined with Ex and a sl correlation as Exc = aEx + 1 − a Ex +
sl, ωLR sl
Ex + Ec [47]. Introducing a screened electron–electron Coulomb interac-
HF , SR
tion at short distances, with a functional Ex containing the HF exchange
only at short electronic ranges (SR) where the density is larger, while keeping the
DFT exchange at long ranges (LR) leads to computational times shorter than for
other hybrid functional treatments, and provided good energy gaps for semi-
conductors and molecular crystal properties [79].
This mixture of HF and sl exchange energies can give good results for bond
energies and lengths around equilibrium interatomic distances in solids, but it
does not assure a correct − c2e r −1 asymptotic form of the KS one-electron vx r
potential. The correct asymptotic form, or electronic long-range corrected (LC)
behavior, can be recovered modifying hybrid functionals so that the long-range
exchange energy functional is equal to the Hartree–Fock expression,
with [49–51]

ExcLC − sl = ExHF , ωLR + a ExHF , ωSR + Exsl, ωSR + Ecsl

LC −sl LC −sl
from which the DFT one-electron potential energy vxc r = δExc δρ r
is found to behave correctly at large electronic distances, insofar as it involves
242 6 Many-Electron Treatments

asymptotically only the exact HF potential. This is important for treating inter-
molecular forces when electron transfer is present. The choices of exchange and
correlation functionals to be used in the above expression can be made based on
needs for accuracy and efficiency, and need not be of semilocal form, with the
above parameter a chosen to reproduce known sets of accurate values.
An alternative general treatment is to recast the theory for a system with N
electrons introducing a functional E ψ j r, ζ of a set of j = 1 to N spin-
orbitals ψ j describing electrons moving in an effective potential v r , with
the orbitals given as functionals of v r . This can be derived from the relation
δE δv r = 0, and optimized by energy minimization, to obtain an optimized
effective potential v OEP r . This was done in early work to calculate the
OEP for the Hartree–Fock functional, and from it energies for atoms [80,
81]. It has been the subject of recent developments and applications to mole-
cules [82], some of which are covered here as they can be applied to intermo-
lecular forces.
Given an accurate many-electron wavefunction, it is possible to calculate
from it ρ r and v r to compare their form with results from DFT functionals.
The deviation of v KS r from a physically correct one-electron potential v r
can be drastic. The OEP procedure instead gives a new v OEP r , which is more
physical and accurate, allowing calculation of more reliable PESs [83].
A general derivation of the equations for a possibly spin-dependent
v OEP r,ζ starts by choosing an explicit functional E OEP
ψ j r,ζ of spin
orbitals, constructed from a wavefunction treatment. A suitable choice is the HF
occ occ
functional E HF
ψ j r,ζ = j j h j + j<k jk v Coul 1 − t jk , in
the notation of Section 6.2 with the index j = (p, σ) containing both orbital
and spin quantum numbers. More accurately, an extension containing electron
correlation can be constructed from many-electron perturbation theory, so that
variation of the total energy gives a single-electron reference potential
vs r, ζ ≈ − ce 2 r − 1 , with the proper asymptotic form. Using the chain rule for
functional differentiation, with the notation x = r,ζ [84],

δE δE OEP δψ j x
= dx +c c =0
δvs x j δψ j x δvs x

and the differential in the second factor to the right can be obtained from the
equation

ℏ2 2
− ∇ + vs x ψ j x = εj ψ j x
2m
 6.5 Spin-Orbit Couplings and Relativistic Effects in Molecular Interactions 243

when adding a perturbation δvs(x), in terms of the equation resolvent (or elec-
tronic Green function) Gj(x, x ) giving δψ j(x ) = dx Gj(x , x)δvs(x) . The first
differential factor can be expressed as δE(OEP)/δψ j(x ) = ψ j(x )∗uj(x ) with the
orbital potential energy uj(x ) known by initial construction. This leads to an
OEP
integral equation for the potential vs x which can be solved when this is
expanded in a basis set, or is taken to be a functional of the electron density
ρ(x) as in the HK formalism. In this case, it can be decomposed into its Coulomb
electronic, plus external, plus exchange-correlation terms as

vsOEP x = vCoul x + vext x + vxcOEP ρ x

with the last term appearing in

δ Exc δψ j x δvs x
OEP OEP
δ Exc
vxcOEP x = = dx dx +c c
δρ x j δψ j x δvs x δρ x

to be calculated from the functional derivative χ s(x, x ) = δρ(x)/δvs(x ), which has

the physical meaning of an electronic response function or susceptibility [82, 84,
OEP
85]. This procedure has been followed to obtain OEP vs x; Q for many-
atom systems as functions of interatomic distances Q, and from them accurate
potential energy functions for molecules and dimers such as F2, He2, HeBe2+,
Ne2, and Be2, involving electronic rearrangement and long-range interaction
energies [86, 87].
Another alternative development can be based on functionals that, instead of
using the local properties of the electron gas, use the electron pair distribution
g(2)(1, 2). This readily incorporates the Fermi hole and the correlation hole and
can be made to change with atomic locations to represent an inhomogeneous
system and its changes with interatomic distances. The functionals must, how-
ever, be constructed so that they are consistent with constrains derived from
many-electron antisymmetrized wavefunctions, a nontrivial task [2, 88].

6.5 Spin-Orbit Couplings and Relativistic Effects

in Molecular Interactions
6.5.1 Spin-Orbit Couplings
Relativistic quantum mechanics requires going beyond the Coulomb interac-
tion energies among electrons and nuclei, as it takes into account the change
of mass of electrons moving with speeds close to c, the speed of light, and
the need for an electron equation of motion invariant under the Lorentz trans-
formation of relativistic mechanics. For a single electron, this is contained in the
Dirac equation [12] and leads to the appearance of electron spin terms in the
244 6 Many-Electron Treatments

Hamiltonian of a molecular system. A single electron moving in a field of nuclei

is described by electron states containing four components in a spinor. Of these,
two are large components and two are small ones, and the coupled equations for
the four amplitudes can be rearranged to display an effective equation for the
large components containing a spin-orbit coupling. In the limit of small electron
velocities compared with the speed of light c, this gives a differential equation
containing the Pauli matrices for two spatial components with up and down
spins, and the form of a one-electron Hamiltonian containing the coupling of
the electron spin with its orbital motion in the field of atomic centers. Relativ-
istic mass effects are present in the inner shells of heavy atoms where electron
speeds are substantial fractions of c [4]. As a result, the size of inner shells in
heavy atoms can shrink, which in turn affects the shape of potential energy sur-
faces and vibrational frequencies of polyatomic systems. This can be treated
with the introduction of relativistic effective core potentials for heavy atoms,
which are also needed in studies of electronic properties of solids [20, 89].
Interaction energies involving pairs of electrons in terms of two components
spinors are covered by the Breit–Pauli Hamiltonian [90–92], which bring addi-
tional terms into the Hamiltonian for many-electron treatments, and in partic-
ular the spin-other-orbit couplings between an electron spin at an atomic center
and electron orbitals at other centers, as well as coupling of the electronic spins
with the rotational angular momentum from moving nuclei. Effects of spin-
orbit couplings on intermolecular and interatomic forces were covered some
time ago [93–96], and many-electron calculations were done using Dirac–Fock
SCFs [97–99], and correlated CI, and MCSCF treatments with effective relativ-
istic atomic pseudopotentials [100] giving potential energy functions of intera-
tomic distance for diatomics containing heavy atoms, such as Au2 and Pb2.
Modern MBPT, CC, and SAPT methods have also been extended and compu-
tationally implemented to include spin-orbit couplings. Spin-orbit effects in
computed molecular properties have been reviewed in recent work [4, 101],
but much remains to be done on intermolecular forces especially when metal
atoms or heavy atoms are involved.
The modern methods described in this chapter are also applicable to molec-
ular systems including spin-orbit couplings, working with two-component spi-
nors. The Hamiltonians must, however, be extended to include spin-dependent
terms. In the Breit–Pauli treatment, spin-orbit one- and two-electron Hamilto-
nian terms are, with α = e2/(4πε0ℏc) the fine structure constant,

α2 Nel NatZu c2e c2e

H SO = u=1 r 3
r iu ×pi si − r ×pi
j r 3 ij
si + 2sj
2 i=1
iu
i
ij

for electrons i and j moving in a framework of nuclei u with position, momen-

tum, and spin r i , pi , and si , respectively, and where r iu =r i −r u and r ij =r i −r j . In
electronic structure calculations, this Hamiltonian is frequently approximated
 6.5 Spin-Orbit Couplings and Relativistic Effects in Molecular Interactions 245

eff
by a sum of one-electron terms with an effective nuclear charges Zu obtained
from a mean-field approximation and fit to atomic fine structure levels that
include atomic spin-orbit energies,
eff
α2 Nel Nat Zu c2e Nat u
H SO liu si = H SO
2 i=1 u=1 riu 3 u=1

with liu =r iu ×pi the orbital angular momentum of an electron in the field of a
nucleus, giving a sum of atomic terms. Each electronic term liu si does not com-
mute with the orbital and spin components li, ξ or si, ξ ,ξ = x,y,z, but it does com-
mute with their sum, the total electronic angular momentum components
ji, ξ = li, ξ + si, ξ for nucleus u so that its quantum numbers (j, mj) can be used
to label the q-th atomic spin-orbital located at nucleus u as q j mj u, a function
of electron position and spin variables r,ζ . For relativistic effective atomic
REP
pseudopotentials ΔUu, l r , with l an atomic orbital quantum number, chosen
to depend only on the radial distance to each nucleus u, one finds that
u REP
H SO = l ΔUu, l r l j mj l j mj u provides a useful parametrization for
potential energy calculations [100]. Reference [102] analyzes how these pseudo-
potentials are related to the detailed Breit–Pauli Hamiltonian in molecular
interactions. Insofar as the pseudopotentials are obtained from expectation
values with many-electron wavefunctions, they are found to depend on the
atomic conformation of the AB pair, with atomic positions Q = (R, Q ) appearing
REP
as parameters in ΔUu, l r; R,Q . In the limit of large R, one finds that
A B
H SO R,Q ≈ H SO QA + H SO QB as needed.

6.5.2 Spin-Orbit Effects on Interaction Energies

Given the total Hamiltonian including spin terms and dependent on nuclear
positions, it is possible to calculate potential energy surfaces using the many-
electron treatments previously described. This usually requires using multire-
ference versions applicable to degenerate or near-degenerate reference states
insofar as total spin multiplets have the same energy before allowing for
spin-orbit couplings. The introduction of spin-orbit couplings splits energy
levels of degenerate states into groups with similar energies, which must be trea-
ted with comparable accuracy. Atomic states r,ζ q j mj = ψ qjmj r, ζ can be re-
expressed in terms of factorized spin orbitals χ μη r η ζ to be used in the cal-
culation of one- and two-electron integrals. For pairs AB containing only light
atoms in the second and third rows of the periodic table, spin-orbit coupling
energies are small compared to electron–electron Coulomb energies, and
energy changes due to spin-orbit couplings can be obtained from MR-MBPT
perturbation theories as functions of interatomic distances. For molecular pairs
246 6 Many-Electron Treatments

with heavier atoms, spin-orbit energies can be comparable or larger than

electron–electron energies, and it is necessary to obtain energies from MRCI,
MRSCF, or MRCC treatments, working with determinantal wavefunctions
constructed from orbitals dependent on electron spin projection quantum
numbers.
Treatments of large many-atom systems using density functionals have also
been extensively developed and computationally implemented to incorporate
spin-orbit coupling energies, particularly for polyatomic systems and solids
containing heavy atoms [20, 96, 103, 104].
The dynamics of many-electron states are frequently affected in essential
ways by coupling of electron spins with nuclear motions, mediated by spin-orbit
couplings. Two examples are intersystem crossings where electrons in singlet
states of the pair AB rearrange into triplet states due to spin recouplings com-
pensated by orbital and nuclear motion changes, and singlet fission where the
pair AB in a singlet state is perturbed and undergoes a transition into a state
where separate A and B species are found in triplet states. States of the AB pair
changing with interatomic distances can be labeled by total electronic angular
momentum quantum numbers (J, MJ),which result from addition of all the elec-
tronic spins and orbital angular momenta. When spin-orbit coupling is small, it
is possible to label the pair’s unperturbed states by total electronic spin and
orbital quantum numbers (S, MS) and (L, ML) in an L–S coupling scheme. When
the spin-orbit coupling is large, it is instead necessary to first couple one-
electron spin and orbital quantum numbers in a j-j scheme [3].
Spin-orbit coupling has an indirect effect on intermolecular potential energies
in the AB pair. The orbital motion of electrons is affected by the electric fields of
the nuclear charges, which vary as the distance between A and B changes. For a
diatomic, this means that the electronic charge distribution has rotational sym-
metry around the internuclear axis and that electrons have axial angular
momenta. Spin-orbit coupling transfers this axial symmetry, as it changes in
strength with intermolecular distance, to the spin of the many-electron system,
and this leads to spin recoupling as the intermolecular distance changes.
In the L–S scheme, addition of electron spins in Cartesian components
2
Sξ = i si, ξ , ξ = x,y,z, provides the total electronic spin operators S and S z ,
which commute with the spin-free Hamiltonian and provide good quantum
numbers in the absence of spin-orbit couplings. When spin terms are included
in the Hamiltonian, the total spin components must be added to total orbital
momenta Lξ to obtain the electronic angular momenta J ξ = S ξ + Lξ . One must
in addition account for the coupling of the electronic angular momenta to
the rotational momenta of the nuclei, Lnu , also called R in the literature of
molecular spectroscopy, giving a total (electronic plus nuclear) angular momen-
tum J tot = L + S +Lnu . In an interaction pair A + B, the angular momenta J tot , L
and S have values along the intermolecular axis labeled, in units of ℏ, as Ω, Λ,
 References 247

and Σ, respectively, with Ω = Λ + Σ. These provide quantum numbers to identify

PESs as atomic distances vary. Whether they are good quantum numbers at a
given intermolecular distance R depends on the relative magnitude of the spin-
orbit coupling energy compared to the electronic rotational energy around the
intermolecular axis. The relative values depend on the intermolecular distance,
with the electronic axial rotational energy being larger at small distances and
disappearing at large distances.
Recouplings in polyatomic systems, changing these quantum numbers and
spin multiplicity and the shape of intermolecular PESs, can be analyzed similarly
to how it is done for diatomics, where different cases have been classified as
Hund’s cases: (i) where the electronic angular momentum rotates around the
intermolecular axis and spin-orbit coupling is large enough to force the spin
to also rotate axially, so that both Λ and Σ are good quantum numbers;
(ii) where spin-orbit coupling is weak and spin is not coupled to the axial field;
(iii) where spin-orbit coupling is very strong and J = L + S rotates axially to give
a good Ω; and (iv) when it is very weak [3, 105, 106]. Recouplings and interacting
PESs can therefore be expected as the intermolecular distance R changes.

References
1 Levine, I.N. (2000). Quantum Chemistry, 5e. New York: Prentice-Hall.
2 McWeeney, R. (1989). Methods of Molecular Quantum Mechanics, 2e. London,
England: Academic Press.
3 Atkins, P.W. and Friedman, R.S. (1997). Molecular Quantum Mechanics.
Oxford, England: Oxford University Press.
4 Jensen, F. (2001). Introduction to Computational Chemistry. New York: Wiley.
5 Szabo, A. and Ostlund, N.S. (1982). Modern Quantum Chemistry. New York:
Macmillan.
6 Woodward, R.B. and Hoffmann, R. (1970). The Conservation of Orbital
Symmetry. New York: Academic Press.
7 Pearson, R.G. (1976). Symmetry Rules for Chemical Reactions. New
York: Wiley.
8 Turro, N.J. (1978). Modern Molecular Photochemistry. Menlo Park, CA:
Benjamin-Cummings.
9 Lowdin, P.O. (1962). Normal constants of motion in quantum mechanics
treated by projection techniques. Rev. Mod. Phys. 34: 520.
10 Lowdin, P.O. (1964). Angular momentum wavefunctions constructed by
projection operators. Rev. Mod. Phys. 36: 966.
11 Lowdin, P.O. (1962). Studies in perturbation theory IV. Solution of eigenvalue
problems by a projection operator formalism. J. Math. Phys. 3: 969.
12 Messiah, A. (1962). Quantum Mechanics, vol. 2. Amsterdam: North-Holland.
248 6 Many-Electron Treatments

13 Micha, D.A. (2017). Quantum partitioning methods for few-atom and many-
atom dynamics. In: Advances in Quantum Chemistry, vol. 74, 107. London
(England): Elsevier.
14 Lowdin, P.O. (1959). Correlation problem in many-electron quantum
mechanics. I. Different approaches and some current ideas. In: Advances in
Chemical Physics, vol. II, 207. New York: Interscience.
15 Hehre, W.J., Radom, L., Schleyer, P.V., and Pople, J.A. (1986). AB INITIO
Molecular Orbital Theory. New York: Wiley.
16 Claudino, D., Gargado, R., and Bartlett, R.J. (2016). Coupled-cluster based basis
sets for valence correlation calculations (Erratum in vol. 145, p. 019901-1). J.
Chem. Phys. 144: 104106–104101.
17 Claudino, D. and Bartlett, R.J. (2018). Coupled-cluster based basis sets for
valence correlation calculations. New primitives and frozen atomic natural
orbitals. J. Chem. Phys. 149: 064105–064101.
18 Chalasinski, G. and Szszesniak, M.M. (2000). State of the art and challenges of
the ab initio theory of intermolecular interactions. Chem. Rev. 100: 4227.
19 Ashcroft, N.W. and Mermin, N.D. (1976). Solid State Physics. London, England:
Thomson.
20 Martin, R.M. (2004). Electronic Structure: Basic Theory and Practical Methods.
Cambridge, England: Cambridge University Press.
21 Schaefer, H.F. (1972). The Electronic Structure of Atoms and Molecules.
Reading, MA, USA: Addison-Wesley.
22 Schaefer, H.F. (1977). Methods of Electronic Structure Theory. Volume 3: Modern
Theoretical Chemistry (eds. W.H. Miller and H.F. Schaefer). New York: Plenum.
23 Roos, B.O. (1987). The complete active space self-consistent field method and
its applications in electronic structure calculations. In: Advances in Chemical
Physics, vol. 69 (ed. K.P. Lawley), 399. New York: Wiley.
24 Bruna, P.J. and Peyerimhoff, S.D. (1987). Excited-state potentials. In: Advances
in Chemical Physics, Part I, vol. 67 (ed. K.P. Lawley), 1. New York: Wiley.
25 Shavitt, I. and Bartlett, R.J. (2009). Many-Body Methods in Chemistry and
Physics. Cambridge, England: Cambridge University Press.
26 Bartlett, R.J. (1981). Many-body perturbation theory and coupled cluster theory
for electronic correlation in molecules. Annu. Rev. Phys. Chem. 32: 359.
27 Bartlett, R.J. and Musial, M. (2007). Coupled-cluster theory in quantum
chemistry. Rev. Mod. Phys. 79: 291.
28 Stanton, J.F. and Gauss, J. (2003). A discussion of some problems associated
with the quantum mechanical treatment of open-shell molecules. In: Advances
in Chemical Physics, vol. 125, 101. Hoboken, NJ: Wiley-Interscience.
29 Hirata, S., Fan, P.D., Auer, A.A. et al. (2004). Combined coupled-cluster and
many-body perturbation theory. J. Chem. Phys. 121: 12197.
30 Krylov, A.I. (2008). Equation-of -Motion Coupled-Cluster Methods for Open-
Shell and Electronically Excited Species (Annu. Rev. Phys. Chem, vol. 59), 433.
Palo Alto, CA: Annual Reviews Inc.
 References 249

31 Bartlett, R.J. (2012). Coupled-cluster theory and its equation-of-motion

extensions. WIREs Comput. Mol. Sci. 2: 126.
32 Jeziorski, B. and Monkhorst, H.J. (1981). Coupled cluster method for
multideterminantal reference states. Phys. Rev. A 24: 1668.
33 Goddard, W.A.I. and Harding, L.B. (1978). generalized valence bond description
of bonding in low-lyingstates of molecules. Annu. Rev. Phys. Chem. 29: 363.
34 Dunning, T.H., Xu, L.T., Takeshita, T.Y., and Lindquist, B.A. (2016). Insight
into the electronic structure of molecules from generalized valence bond
theory. J. Phys. Chem. A 120: 1763.
35 Jeziorski, B., Moszynski, R., Ratkiewitcz, A. et al. (1993). SAPT: A Program for
Many-Body Symmetry-Adapted Perturbation Theory Calculations of
Intermolecular Interaction Energies (METTEC-94, vol. B) (ed. E. Clementi), 79.
Cagliari, Italy: STEF.
36 Jeziorski, B., Moszynski, R., and Szalewicz, K. (1994). Perturbation theory
approach to intermolecular potential energy surfaces of van der Waals
complexes. Chem. Rev. 94: 1887.
37 Casimir, H.B.G. and Polder, D. (1948). The influence of retardation on the
London-van der Waals forces. Phys. Rev. 73: 360.
38 Szalewicz, K. (2012). Symmetry-adapted perturbation theoryof intermolecular
forces. WIREs Comput. Mol. Sci. 2: 254.
39 Misquitta, A.J., Podeszwa, R., Jeziorki, B., and Szalewicz, K. (2005).
Intermolecular potentials based on SAPT with dispersion energies from
time-dependent density functional calculations. J. Chem. Phys. 123:
214103–214101.
40 Lotrich, V.F. and Szalewicz, K. (1997). Symmetry-adapted perturbation theory
of three-body nonadditivity of intermolecular interaction energy. J. Chem. Phys.
106: 9668.
41 Podeszwa, R. and Szalewicz, K. (2007). Three-body symmetry-adapted
perturbation theory based on the Kohn-Sham description of the monomers.
J. Chem. Phys. 19: 194101.
42 Hohenberg, P. and Kohn, W. (1964). Inhomogeneous electron gas. Phys. Rev. B
136: 864.
43 Parr, R.G. and Yang, W. (1989). Density Functional Theory of Atoms and
Molecules. Oxford, England: Oxford University Press.
44 Trickey, S.B. (ed.) (1990). Advances in Quantum Chemistry. Volume 21: Density
Functional Theory of Many-Fermion Systems. San Diego, CA, USA:
Academic Press.
45 Lundqvist, S. and March, N.H. (eds.) (1983). Theory of the Inhomogeneous
Electron Gas. New York: Plenum Press.
46 Becke, A.D. (1993). Density-functional theormochemistry III. The role of exact
exchange. J. Chem. Phys. 98: 5648.
47 Heyd, J., Scuseria, G.E., and Ernzerhof, M. (2003). Hybrid functionals based on a
screened Coulomb potential. J. Chem. Phys. 118: 8207.
250 6 Many-Electron Treatments

48 Perdew, J.P., Tao, J., Staroverov, V.N., and Scuseria, G.E. (2004). Meta-
generalized gradient approximation: explanation of a realistic nonempirical
density functional. J. Chem. Phys. 120: 6898.
49 Iikura, H., Tsuneda, T., and Yanai, T.H.K. (2001). A long-range corrected
scheme for generalized-gradient approximation exchange functionals. J. Chem.
Phys. 115: 3540.
50 Vydrov, O.A. and Scuseria, G.E. (2006). Assessment of a long-range corrected
hybrid functional. J. Chem. Phys. 125: 234109.
51 Chai, J.-D. and Head-Gordon, M. (2008). Systematic optimization of long-
range corrected hybrid density functionals. J. Chem. Phys. 128: 084106.
52 Gunnarsson, O. and Lundqvist, B.I. (1976). Exchange and correlation in atoms,
molecules, and solids by the spin density functional formalism. Phys. Rev. B
13: 4274.
53 Jones, R.O. and Gunnarsson, O. (1989). The density functional formalism, its
applications and prospects. Rev. Mod. Phys. 61: 689.
54 Grimme, S., Hansen, A., and Brandenburg, J.G.C. (2016). Dispersion corrected
mean-field electronic structure methods. Chem. Rev. 116: 5105.
55 Hermann, J., DiStasio, R.A.J., and Tkatchenko, A. (2017). First-principles
models for van der Waals interactionsin molecules and materials: concepts,
theory, and applications. Chem. Rev. 117: 4714.
56 Sato, T. and Nakai, H. (2009). Density functional method including weak
interactions: dispersion coefficients based on the local response approximation.
J. Chem. Phys. 131: 224104.
57 Sato, T. and Nakai, H. (2010). Local response dispersion method. II.
Generalized multicenter interactions. J. Chem. Phys. 133: 194101.
58 Podeszwa, R. and Szalewicz, K. (2012). Communication: density functional
theory overcomes the failure of predicting intermolecular interaction energies.
J. Chem. Phys. 136: 161102–161101.
59 Zhao, Y. and Truhlar, D.G. (2008). The M06 suite of energy functionals
systematic testing of four M06 functionals and 12 other functionals. Theor.
Chem. Accounts 120: 215.
60 von Lilienfeld, O.A., Tavernelli, I., Rothlisberger, U., and Sebastiani, D. (2005).
Performance of atom-centered potentials for weakly bonded systems using
density functional theory. Phys. Rev. B 71: 195119.
61 Peverati, R. and Truhlar, D. (2012). Exchange-correlation functionals with good
accuracy for both structural and energetic properties while depending only on
the density and its gradient. J. Chem. Theory Comput. 8: 2310.
62 Peverati, R. and Truhlar, D.G. (2012). Screened exchange density functionals
with broad accuracy for chemistry and solid state physics. Phys. Chem. Chem.
Phys. 14: 16187.
63 Langreth, D.C., Lundqvist, B.I., Chakarova-Kaeck, S.D. et al. (2009). A density
functional for sparse matter. J. Phys. Condens. Matter 21: 084203.
 References 251

64 Bortolani, V., March, N.H., and Tosi, M.P. (eds.) (1990). Interaction of Atoms
and Molecules with Solid Surfaces. New York: Plenum Press.
65 Vydrov, O.A. and Van Voohis, T. (2010). Nonlocal van der Waals density
functional: the simpler the better. J. Chem. Phys. 133: 244103–244101.
66 Nozieres, P. and Pines, D. (1999). The Theory of Quantum Liquids. Cambridge,
MA: Perseus.
67 Dobson, J. F. and T. Gould, "Calculation of dispersion energies," J. Phys.
Condens. Matter, vol. 24, p. 073201, 2012.
68 Dobson, J.F. (2012). Dispersion (van der Waals) forces and TDDFT. In:
Fundamentals of Time-Dependent Density Functional Theory (eds. M.A.L.
Marques, N.T. Maitra, F.M.S. Nogueira, et al.), 417. Berlin: Springer-Verlag.
69 Berland, K., Cooper, V.R., Lee, K. et al. (2015). van der Waals forces in density
functional theory: a review of the vdW-DF method. Rep. Prog. Phys. 78: 066501.
70 Sun, J., Ruzsinszky, A., and Perdew, J. (2015). Strongly constrained and
appropriately normed semilocal density functional. Phys. Rev. Lett. 115: 036402.
71 Staroverov, V.N., Scuseria, G.E., Tao, J., and Perdew, J.P. (2003). Comparative
assesment of a new nonempirical density functional: molecules and hydrogen
bonded complexes. J. Chem. Phys. 119: 12129.
72 Xu, X. and Goddard, W.A. (2004). Bonding properties of the water dimer: a
comparative study of density functional theories. J. Phys. Chem. A 108: 2305.
73 Taylor, D.E., Angyan, J.G., Galli, G. et al. (2016). Blind test of density functional
based methods on intermolecular interaction energies. J. Chem. Phys. 145:
124105–124101.
74 Mermin, N.D. (1965). Thermal properties of the inhomogeneous electron gas.
Phys. Rev. 137: A1441.
75 Gross, E.K.U., Oliveira, L.N., and Kohn, W. (1988). Density functional theory
for ensembles of fractionally occupied states. I. Basic formalism. Phys. Rev. A
37: 2809.
76 Cohen, M.H. and Wassermann, A. (2007). On the foundations of chemical
reactivity theory. J. Phys. Chem. A 111: 2229.
77 Zhao, Y. and Truhlar, D.G. (2008). Density functionals with broad applicability
in chemistry. Acc. Chem. Res. 41: 157.
78 Perdew, J.P., Burke, K., and Ernzenhof, M. (1996). Generalized gradient
approximation made simple. Phys. Rev. Lett. 77: 3865.
79 Heyd, J. and Scuseria, G.E. (2004). Efficient hybrid density functional
calculations in solids: assessment of the Heyd – Scuseria – Ernzerhof screened
Coulomb hybrid functional. J. Chem. Phys. 121: 1187.
80 Talman, J.D. and Shadwick, W.F. (1976). Optimized effective atomic central
potentials. Phys. Rev. A 14: 36.
81 Krieger, J.B., Li, Y., and Iafrate, G.J. (1995). Recent developments in Kohn-Sham
theory for orbital dependent exchange-correlation energy functionals. In:
Density Functional Theory (eds. E.K.U. Gross and R.M. Dreizler), 191.
New York: Plenum Press.
252 6 Many-Electron Treatments

82 Kuemmel, S. and Kronik, L. (2008). Orbital-dependent density functionals:

theory and applications. Rev. Mod. Phys. 80: 3.
83 Bartlett, R.J. (2010). Ab initio DFT and its role in electronic structure theory.
Mol. Phys. 108: 3299.
84 Grabowski, I., Hirata, S., Ivanov, S., and Bartlett, R.J. (2002). Ab initio density
functional theory: OEP-MBPT(2). A new orbital dependent correlation
functional. J. Chem. Phys. 116: 4415.
85 Heaton-Burgess, T. and Yang, W. (2008). Optimized effective potentials from
arbitrary basis sets. J. Chem. Phys. 129: 194102–194101.
86 Bartlett, R.J., Lotrich, V.F., and Schweigert, I.V. (2005). Ab initio density
functional theory: the best of both worlds? J. Chem. Phys. 123: 062205–062201.
87 Lotrich, V.F., Bartlett, R.J., and Grabowski, I. (2005). Intermolecular potential
energy surfaces of weakly bound dimers computed from ab initio density
functional theory: the right answer for the right reason. Chem. Phys. Lett.
405: 43.
88 Ayers, P.W. and Levy, M. (2005). Using the Kohn–Sham formalism in pair
density-functional theories. Chem. Phys. Lett. 415: 211.
89 Hay, P.J. and Martin, R.L. (1998). Theoretical studies of the structures and
vibrational frequencies of actinide compounds using relativistic effective core
potentials. J. Chem. Phys. 109: 3875.
90 Bethe, H.A. and Salpeter, E.E. (1957). Quantum Mechanics of One- and Two-
electron Atoms. Berlin: Springer-verlag.
91 Hirschfelder, J.O., Curtis, C.F., and Byron Bird, R. (1954). Molecular Theory of
Gases and Liquids. New York: Wiley.
92 Langhoff, S.R. and Kern, W.C. (1977). Molecular fine structure and
spectroscopy by ab initio methods. In: Applications of Modern Theoretical
Chemistry (ed. H.F. Schaefer), 381. New York: Plenum Press.
93 Hirschfelder, J.O. and Meath, W.J. (1967). The nature of intermolecular forces.
Adv. Chem. Phys. 12: 3.
94 Chang, T.Y. (1967). Moderately long range intermolecular forces. Rev. Mod.
Phys. 39: 911.
95 Richards, W.G., Trivedi, H.P., and Cooper, D.L. (1981). Spin-Orbit Coupling in
Molecules. Oxford, England: Oxford University Press.
96 Marian, C.M. (2001). Spin-orbit coupling in molecules. In: Reviews in
Computational Chemistry, vol. 17 (eds. K.B. Lipkowitz and D.B. Boyd).
Wiley-VHC.
97 Pyykko, P. (1978). Relativistic quantum chemistry. In: Advanced Quantum
Chemistry, vol. 11, 353. New York: Academic Press.
98 Pyykko, P. (1988). Relativistic effects in structural chemistry. Chem. Rev.
88: 563.
99 Pisani, L. and Clementi, E. (1994). Relativistic Dirac-Fock calculations for
closed shell molecules. J. Comput. Chem. 15: 466.
 References 253

100 Balasubramanian, K. and Pitzer, K.S. (1987). Relativistic quantum chemistry. In:
Advances in Chemical Physics. Volume 67: Ab Initio Methods in Quantum
Chemistry Part I (ed. K.P. Lawley), 287. New York: Wiley.
101 Fedorov, D.G., Koseki, S., Schmidt, M.W., and Gordon, M.S. (2003). Spin-orbit
coupling in molecules: chemistry beyond the adiabatic approximation. Int. Rev.
Phys. Chem. 22: 551.
102 Fedorov, D.G. and Gordon, M.S. (2000). A study of the relative importance of
one and two-electron contributions to spin–orbit. J. Chem. Phys. 112: 5611.
103 Hafner, J. (2008). Ab initio simulation of materials using VASP: density
functional theory and beyond. J. Comput. Chem. 29: 2044.
104 Kleinschmidt, M., Tatchen, J., and Marian, C.M. (2002). Spin-orbit coupling of
DFT/MRCI wavefunctions: method, test calculations and applications to
thiophene. J. Comput. Chem. 23: 824.
105 Herzberg, G. (1950). Molecular Spectra and Molecular Structure I. Spectra of
Diatomic Molecules, 2e. Princeton NJ: Van Nostrand.
106 Herzberg, G. (1966). Molecular Spectra and Molecular Structure III.
Polyatomic Molecules. New York: Van Nostrand Reinhold.
 255

Interactions Between Two Many-Atom Systems

CONTENTS

7.1 Long-range Interactions of Large Molecules, 255

7.1.1 Interactions from Charge Density Operators, 255
7.1.2 Electrostatic, Induction, and Dispersion Interactions, 258
7.1.3 Population Analyses of Charge and Polarization Densities, 260
7.1.4 Long-range Interactions from Dynamical Susceptibilities, 262
7.2 Energetics of a Large Molecule in a Medium, 265
7.2.1 Solute–Solvent Interactions, 265
7.2.2 Solvation Energetics for Short Solute–Solvent Distances, 268
7.2.3 Embedding of a Molecular Fragment and the QM/MM Treatment, 270
7.3 Energies from Partitioned Charge Densities, 272
7.3.1 Partitioning of Electronic Densities, 272
7.3.2 Expansions of Electronic Density Operators, 274
7.3.3 Expansion in a Basis Set of Localized Functions, 277
7.3.4 Expansion in a Basis Set of Plane Waves, 279
7.4 Models of Hydrocarbon Chains and of Excited Dielectrics, 281
7.4.1 Two Interacting Saturated Hydrocarbon Compounds: Chains and Cyclic
Structures, 281
7.4.2 Two Interacting Conjugated Hydrocarbon Chains, 284
7.4.3 Electronic Excitations in Condensed Matter, 289
7.5 Density Functional Treatments for All Ranges, 291
7.5.1 Dispersion-Corrected Density Functional Treatments, 291
7.5.2 Long-range Interactions from Nonlocal Functionals, 294
7.5.3 Embedding of Atomic Groups with DFT, 297
7.6 Artificial Intelligence Learning Methods for Many-Atom Interaction Energies, 300
References, 303

7.1 Long-range Interactions of Large Molecules

7.1.1 Interactions from Charge Density Operators
The interactions of atoms and molecules with an extended many-atom system
such as a solid surface, atomic cluster, molecular crystal, polymer, or a biomo-
lecule can be described by means of electronic charge densities and polarization
Molecular Interactions: Concepts and Methods, First Edition. David A. Micha.
© 2020 John Wiley & Sons, Inc. Published 2020 by John Wiley & Sons, Inc.
256 7 Interactions Between Two Many-Atom Systems

densities per unit volume and Hamiltonian operators involving space integrals
over charge density operators [1–7]. This bypasses the need for expansions of
interaction energy operators in terms of electrical multipoles of high order.
Long-range interaction energies follow from perturbation theory or from
many-body treatments in terms of response functions.
The form of the interaction energy operator between two species A and B,
now considered to be extended systems containing many atoms, has been given
in Sections 2.1 and 2.4 and in Section 3.2 in terms of charge density operators
cA r and cB r , defined as functions of positions over the whole region of inter-
actions. The two species are given by the locations of the nuclei in each one,
labeled by the index a for A and by b for B. They are then defined by the col-
lection QA = r a of atomic position vectors for A and similarly QB for B, which
are parameters in the density operators. Those density expressions are used here
in applications to extended systems such as clusters, polymers, and solid sur-
faces interacting with molecules or among themselves. The expectation values
of the density operators, for given electronic states of A and B, occupy regions of
space with a variety of shapes. Some may show cavities in which a species may be
located, so that a distance R between centers of mass may not be meaningful as a
separation distance. Two examples are shown in Figure 7.1.
We consider two species A and B with density boundaries sufficiently far away
so that electron exchange is improbable and does not affect the properties of the
system and can be omitted. In this case, electrons i = 1 to NA can be assigned to
species A, and electrons j = NA + 1 to NA + NB are assigned to species B. Their
relative distance rij remains larger than some distance d and their Coulomb
interaction remains finite, insofar electrons i and j are well separated during
the interaction of the two many-electron systems. This treatment, however,
must be reconsidered when electron exchange is relevant.

(a) (b)
B(1)

) B
B(

(2
9)

B
B(8
)
A B(3) A
)
B (7 B(
4)
B(6)

B(5
)

Figure 7.1 (a) Molecule A in a solvent B containing molecules B(n); (b) A in a cavity of a large
molecule B.
 7.1 Long-range Interactions of Large Molecules 257

It is possible to avoid expansions of interaction energies in terms of multi-

poles, by instead using the charge density operator of all the charges CI,
I = a, i, in species A, given as functions of space positions r by
nu el
cA r = I
CI δ r −r I = cA r + cA r

nu el
cA r = a
Ca δ r −r a , cA r = Ce i
δ r −r i

a sum of nuclear and electronic charge densities, and similarly for species B. The
nu
nuclear charge density cA is, for fixed nuclei, simply a function of nuclear posi-
el
tions, but the electronic charge density cA is a one-electron operator in the
space of many-electron wavefunctions.
The Hamiltonian of species A and B can be written in terms of nuclear and
electronic charge densities as
e en ee
HA = K A + HA + HA
el nu
e ℏ2 en 1 c r cA r
KA = − ∇2 , H A =
i i
d3 r d3 r A
2me 4πε0 r−r
el el el
ee 1 c r cA r −δ r − r cA r
HA = d3 r d3 r A
4πε0 r−r
where the form of the electron–electron Coulomb energy accounts for the
absence of interaction of each electron with itself. A similar expression applies
to species B.
The Coulomb interaction energy operator for all charges in A interacting with
all those in B, but without electron exchange between them, is

Coul −1 −1
H AB QA ,QB = 4πε0 d 3 r d 3 s cA r; QA r − s cB s; QB

Coul nn n, e e, n ee
which can be expanded as H AB = H AB + H AB + H AB + H AB to display types
of interactions of nuclear and electronic charges. Insofar the nuclear charge dis-
tributions are dependent on nuclear positions for fixed nuclei, and the elec-
tronic distributions are operators dependent on the electronic position
operators, the second and third terms are one-electron operators dependent
on all nuclear positions while the last term is a two-electron operator.
nn
In more detail, H AB is just the Coulomb energy of interacting nuclear charge
distributions for A and B,

nn −1 nu −1 nu
HAB QA ,QB = 4πε0 d 3 r d 3 s cA r; QA r − s cB s; QB
258 7 Interactions Between Two Many-Atom Systems

which can alternatively be written as the Coulomb energy of nuclear point

charges at their positions.
The electronic-nuclear interaction Hamiltonian energy is

ne nu nu
H AB = d 3 r cA r; QA φB r + d 3 s φA s cB s; QB

el el
1 c s 1 c r
φB r = d3 s B , φA s = d3 r A
4πε0 r−s 4πε0 r−s
where φB r is the electric field created by electronic charges of B at the location
of the nuclear charges of A, and similarly for φA s . This Hamiltonian is a one-
electron operator and its matrix elements can be obtained from one-electron
1 1
density operators ρA and ρB .
The electron–electron interaction is
ee −1 el −1 el
H AB = 4πε0 d 3 r d 3 s cA r r − s cB s

and does not depend explicitly on the nuclear positions. This Hamiltonian is a
two-electron operator, and calculation of its matrix elements between many-
electron states of A and B involve the two-electron density operator
2 1 1
ρAB = ρA ρB , derived from the separate many-electron states of A and B, insofar
they do not exchange electrons. The matrix elements do change with nuclear
distances.
These one- and two-electron expressions appear in the total Hamiltonian in a
form useful for calculations involving both ground and excited electronic states
for fixed atomic positions.

7.1.2 Electrostatic, Induction, and Dispersion Interactions

Long-range electrostatic, induction, and dispersion energies are obtained as
shown in Chapter 3 from first- and second-order perturbation theory, using
basis sets JA and KB for species A and B. The electrostatic interaction energy
of A and B in their ground electronic states is
AB Coul
Eels QA ,QB = 0A ,0B H AB 0A , 0B

Writing the interaction Hamiltonian operator in terms of charge density opera-

tors shows that the electrostatic energy contains their expectation values in the
ground state
A A
c0 r = 0A cA r 0A = a
Ca δ r −r a + Ce ρ0 r
 7.1 Long-range Interactions of Large Molecules 259

A
where ρ0 r is the ground state electronic density of A, and similarly with
B
c0 s for B. Its form is

AB −1 A −1 B
Eels QA ,QB = 4πε0 d 3 r d 3 s c0 r; QA r − s c0 s; QB

and can be further decomposed into Coulomb interaction energies of nuclei and
AB nn ne ee
electrons in A and B, Eels = E0, 0 + E0, 0 + E0, 0 .
Induction and dispersion energies given by
2
Coul
0A ,0B H AB JA ,0B
AB
Eind QA ,QB = − A A
J 0
EJ − E0
2
Coul
0A ,0B H AB 0A ,KB
− K 0 B B
EK − E0
2
Coul
0A ,0B H AB JA ,KB
AB
Edsp QA ,QB = − J 0 K 0 A A B B
EJ − E0 + EK − E0
A
contain transition values c0J r = 0A cA r JA of the charge density opera-
tors for A, and similarly for B, with
A A B B
c0J r = cnuA r δ0J + Ce ρ0J r , c0K s = cnuB s δ0K + Ce ρ0K s
where contributions from nuclei and electrons have been separated. Here
A el
Ce ρ0J r = 0A cA r JA is an electronic charge excitation, and similarly
for B, obtained from their many-electron states. The second-order energies
Eind = E0, 0 AB∗ + E0, 0 A∗ B and Edsp = E0, 0 A∗ B∗ involve integrals like
AB ind ind AB dsp

Coul −1 A −1 B
0A , 0B H AB JA ,0B = 4πε0 d 3 r d 3 sCe ρ0J r r − s c0 s
A B

for the interaction of charges of B with electronic density fluctuations in A,

appearing in the induction energy, and integrals
Coul −1 A −1 B
0A , 0B H AB JA ,KB = 4πε0 d 3 r d 3 sCe ρ0J r r − s Ce ρ0K s
A B

for excitation density-excitation density couplings in the dispersion energy.

These integrals appear in summations over usually unknown electronic excited
states of A∗ and B∗, and it is of practical interest to find ways to calculate the
perturbation corrections while avoiding sums over excited states. Also, large
260 7 Interactions Between Two Many-Atom Systems

many-atom systems may involve degenerate ground states all of the same
energy, requiring modified treatments for perturbation of degenerate states.
In addition, large systems may have vanishing or very small excitation energies
A A B B
EJ − E0 and EK − E0 , which lead to large perturbation energy corrections
and require alternative treatment. These problems can sometimes be avoided
introducing static and dynamical susceptibilities functions of excitation fre-
quencies. Instead of using the molecular excited states, the summations in sec-
ond-order perturbation terms can best be calculated as integrals over their
frequencies with susceptibilities obtained from a variety of electronic basis sets.

7.1.3 Population Analyses of Charge and Polarization Densities

Given the above energy expressions, calculation of the energies must be done
introducing many-electron states ΦJA 1,2, …,NA = 1,2, …,NA JA for A and
ΦKB NA + 1,NA + 2,…,NA + NB for B. Insofar no electron exchange or transfer
needs consideration, the combined states for the AB pair are given by JA, KB =
J A ⨂ KB .
A
Transition density values cJJ r = JA cA r JA of the charge density
operators for A, and similarly for B, are
A A B B
cJJ r = cnuA r δJJ + Ce ρel, JJ r , cKK s = cnuB s δKK + Ce ρel, KK s

A el
where Ce ρel, JJ r = JA cA r JA is an electronic charge density fluctuation,
and similarly for B, obtained from their many-electron states. Transition den-
A
sities ρel, JJ can be obtained from one-electron transition density functions
1
ρJJ 1,1 , for states J and J of A with NA electrons. They are defined as an exten-
sion of the single-state one-electron density functions in Chapter 6, by means of

,NA ∗ ΦJ 1 , 2,
1
ρJJ 1; 1 = NA d 2 d NA ΦJ 1, 2, ,NA

with (1) standing for space and spin electron variables r 1 , ζ1 = x1 , and so that

A 1
ρel, JJ r = ρ
ζ JJ
r,ζ; r ,ζ

after adding over the spin variables.

Densities take simple forms when the J and J states are NA-electron determi-
nantal wavefunctions containing orthonormal molecular spin-orbitals (or MSOs)
ψ j(xn) = xn j , where the indices n and j can be taken to label rows and columns
of a NA × NA matrix. Given determinantal functions ΦJ(1, 2, … , NA)
 7.1 Long-range Interactions of Large Molecules 261

−1 2
= (NA!)−1/2 det[ xn j ] and ΦJ 1 ,2 , …, NA = NA det xn j con-
structed from a basis set of N orthonormal SOs, one finds

1 1 n = NA
ρJJ x1 ; x1 = d 2 d NA det j xn xn j
NA −1 n=1 2 = 2 , …, NA = NA

obtained recalling that the determinant of a product of matrices equals the

product of their determinants, and doing next the integrals over electron vari-
ables (2), …, (NA). This one-electron density is different from zero only if the
many-electron states J and J are equal or if they differ by only one MSO. Gen-
eral treatments of transition density functions, also for nonorthogonal basis
sets, can be found in references [8, 9].
Furthermore, when the MSOs ψ j(x) are written as linear combinations of a
basis set of functions ξμ(r ,ζ , generally nonorthogonal, the one-electron tran-
sition density functions can be expanded as
1 ∗
ρJJ r,ζ; r ,ζ = μ μ
ξμ r,ζ PJJ μ,μ ξμ r ,ζ

with expansion coefficients PJJ μ,μ determined by the form of the many-
electron states J. The basis functions ξμ(r ,ζ can be chosen as atomic spin orbi-
tals (ASOs) to form MSO-LCAO orbitals, or as delocalized functions.
Choosing a basis set of ASO μ for A, with overlaps ξμ ξμ = Sμμ , and sim-
1
ilarly for ρKK with AOs ξν(r , ζ for B, it is possible to perform a population anal-
ysis to decompose the densities into terms giving atomic populations and atom–
atom bond orders. Labeling ASOs centered at the nuclei a of A, with orbital and
spin quantum numbers (n, l, m, ms) by μ = (a, n, l, m, ms), atomic components
PJJ a of transition densities at each center a are obtained adding all terms
where μ and μ contain only a. This can be quite generally done introducing pro-
jection operators

Pa = μ a μ a
ξμ S − 1 μμ
ξμ

which satisfy Pa Pa = δaa P a and add up to the identity. Applied to the density
operator, they select its components localized at centers a in A and give
1
ρJJ = P
a JJ
a Pa + a a
P a PJJ a,a P a

where PJJ a is a transition component at location a, while bond (aa ) compo-

nents PJJ a, a of the transition density are given by the sums that contain a
1
a . There is a similar decomposition for B with ρKK containing the ASO ξν. This
in turn allows for decomposition of interaction energies for AB into atom–
atom, atom–bond, and bond–bond intermolecular components, useful when
treating the whole system as made up of molecular fragments.
262 7 Interactions Between Two Many-Atom Systems

A second choice for the basis functions ξμ(r ,ζ is provided by plane waves
times spin functions Ω − 1 2 exp −iqn r ϑ ζ with quantized wavevectors satis-
fying periodic boundary conditions in a large volume Ω. This is a convenient
choice when the many-atom systems are very large, when they display structural
periodicity, or when electrons are delocalized, and is further considered in what
follows.

7.1.4 Long-range Interactions from Dynamical Susceptibilities

Induction energies arise from the static susceptibilities of A and B, coupled
respectively to charges in B and A. The static electrical susceptibility of B in
its ground electronic states is given by
B B ∗
B c0K s c0K s
χ0 s, s = −2 K 0 B B
E0 − EK
a positive-valued tensor, in terms of which the induction energy at A due to the
polarization of B, shown as AB∗, is given by
A B A
AB∗ 1 3 c0 r χ0 s, s c0 r
Eind =− d r d3 r d3 s d3 s 2
2 4πε0 r−s r −s
with the same integration limits in A and B. This corresponds to a total (nuclear
plus electronic) static charge in a volume d3r of A at location r interacting with
B
an electronic charge fluctuation c0K s within d3s of B, followed by the reverse
interaction of a charge fluctuation in B with an element of total static charge in
A∗ B
A. An expression for the other induced energy Eind follows by interchanging A
A
and B functions and arguments and contains χ 0 r, r .
The dispersion energy resulting from charge density fluctuations in A and B,
A∗ B∗
shown here as Edsp , involves the susceptibility of the pair AB and all its pair
A A B B
excitation energies EJ −E0 + EK −E0 in the denominator of the perturba-
tion expansion of the dispersion energy. This appears to intertwine the proper-
ties of A and B. However, as done in Section 3.4 for the dispersion interaction of
two molecules due to their electric dipole fluctuations, which could be
expressed in terms of their separate dynamical dipole polarizabilities, it is pos-
sible here to transform the AB pair static susceptibility into an integral over sep-
arate dynamical susceptibilities of the two extended systems A and B, now
without a restriction to particular multipole polarizabilities. We again use the
Casimir–Polder integral (for positive a, b)
∞
2 du a b 1
=
π 0 a2 2 2
+u b +u 2 a+b
 7.1 Long-range Interactions of Large Molecules 263

A A B B
with a = EJ −E0 and b = EK − E0 and introduce the dynamical susceptibility

A A A A ∗
c0J r E0 − EJ c0J r
A
χ0 r , r ; ω = −2 J 0 2
A A 2
E0 − EJ − ℏω

which leads, with the choice ℏω = u, to the convenient form

∞ A B
A∗ B∗ ℏ χ0 r, r ; iω χ 0 s, s ; iω
Edsp =− d3 r d3 r d3 s d3 s dω 2
2π 0 4πε0 r−s r −s

written in terms of the susceptibilities for imaginary-valued frequencies.

The susceptibilities have so far been written as sums over excited many-
electron states of A and B. However, these states are seldom known for extended
many-atom systems. This difficulty can be avoided writing the susceptibility
in operator forms using resolvents, which can be calculated in a convenient
A A
basis set. The summation in χ 0 r, r ; iω over states JA with energies EJ
A
can be rewritten in a compact form using that H A JA = EJ JA and introdu-
cing the projection operator P A = I − 0A 0A = J 0 JA JA . This allows us
to write

A
A E0 − H A
χ0 r, r ; iω = −2 0A cA r P A 2 P A cA r 0A
A 2
E0 −H A + ℏω

which can now be calculated with an alternative many-electron basis set. Prop-
erties of this susceptibility and ways to calculate it and to provide upper and
lower bounds to it as a function of the variable u have been the subject of
detailed research [10–12]. The procedure is similar to the one followed in
Section 3.4 for molecular interactions. The static susceptibilities in the induc-
tion energy are found for ω = 0.
Two particularly simple forms are obtained from a closure approximation,
A
and from an interpolation of χ 0 iω between its values at ω = 0 and ω ∞.
A
In a closure approximation, the operator E0 − H A P A in the dynamical sus-
A A
ceptibility is replaced by E0 − Eexc P A, which contains an excitation energy
A
Eexc to be treated as an adjustable parameter chosen to reproduce the value of
264 7 Interactions Between Two Many-Atom Systems

the dispersion energy E0, 0 A∗ A∗ , extracted from measurements or calcula-

dsp

tions. With this approximation, the susceptibility becomes

A
χ 0, exc r, r ; iω = 2Fexc
A
ω GccA r, r
A A
GccA r, r = 0 A cA r P A cA r 0A = 0A cA r cA r 0A −c0 r c0 r

2
A A A A A
with the frequency in the factor Fexc ω = E0 − Eexc E0 −Eexc

2
+ ℏω , which multiplies the charge–charge correlation function
A
Gcc r, r for the ground state expectation value of charge densities at two posi-
tions. This contains one- and two-electron operators averaged only over a
A A
known many-electron ground state. The choice Eexc = E1 , the first excited
state of A, gives an upper bound to this approximate susceptibility. The corre-
lation function provides the static susceptibility through
A A A A A
χ 0 r, r ; 0 = 2Gcc r, r E0 − Eexc = χ 0 r, r , with the parameter
A A
ℏω A = E0 −Eexc to be extracted from calculations or measurements.
B
With a similar approximation for χ 0 iω , it is possible to estimate the induc-
tion and dispersion energies of the AB pair. Induction energies are given by
interactions between permanent charges and expectation values of charge–
charge fluctuations, however involving just the ground states of A and B, and
information on their energies for AA and BB. By analogy with the treatment
of interaction energies for two molecules in the dipole approximation, it follows
that here we can also introduce a combination rule giving the dispersion energy
of AB in terms of the values for AA and BB. By analogy with the treatment for
two interacting molecules leading to combination rules for the dispersion
energy [13], here for two large many-atom systems, it is convenient to introduce
A A A A
the limit for ω ∞, χ 0, exc r, r ; iω ≈ 2 E0 − Eexc Gcc r, r ℏω 2 =
2 A
ω A ω χ 0 r, r ; 0 . The limits at ω = 0 and ω ∞ can be combined in
an interpolated approximate form
2 −1
A A ω
β0 r, r ; ω = χ0 r, r 1+
ωA
A A
where β0 r, r ; ω = χ 0 r, r ; iω . The parameter ω(A) can be obtained from
A∗ A∗ , containing a calculated Gcc r, r
dsp A
E0, 0 and written in terms of
B∗ B∗ , these two para-
A dsp
β0 ω . Doing the same for ω (B)
obtained from E0, 0
A B
meters and the correlation functions Gcc r, r and Gcc s,s can be used
A∗ B∗ from a combination rule, here for extended systems.
dsp
to obtain E0, 0
 7.2 Energetics of a Large Molecule in a Medium 265

7.2 Energetics of a Large Molecule in a Medium

7.2.1 Solute–Solvent Interactions
AB
The total energy E0, slt in a solution (slt) of a solute molecule A in a medium B
(which can be a liquid or solid solvent [slv], a cavity, or a solid surface) must
AB
include the energy E0 QA ,QB of A interacting with B, plus the interaction
B
energy E0, slv of all solvent molecules or medium fragments in B,

AB AB B
E0, slt QA ,QB = E0 QA ,QB + E0, slv QB

The terms to the right can be separately considered and are simpler to con-
struct when electrons are not exchanged between A and B, and they are also
sufficiently separated so that short-range repulsion does not distort them. This
is first done as follows.
The electrostatic, induction, and dispersion energies for A in a given elec-
tronic state 0A interacting with a species B in a general electronic state KB
els ind dsp
can be obtained from potential energy operators V AB ,V AB , and V AB arising
from the charges in A interacting with electric potentials originating in B. The
els
electrostatic potential energy operator V AB r; QA , QB involves the charge
density operator c A r; QA of A multiplying the electric scalar potential
els
φB r; QB resulting from the stationary charge distribution of B in its ground
ind
state 0B . Additional potential energy operators V AB r, r ; QA ,QB , and
dsp
V AB r, r ; QA , QB can be constructed from the charge distributions
cA r; QA and cA r ; QA at two locations linked by reaction potential operators
ind dsp
φAB r, r ; QA ,QB and φAB r, r ; QA ,QB generated by density fluctuations
when B is excited into states KB . These reaction potentials can be obtained
as shown below, or from time-dependent response theory [6].
This treatment is convenient when describing a large solute molecule A in a
medium B composed of many solvent molecules treated as molecular frag-
ments [14], or when B is a medium involving many atoms in a complex structure
described by distributed multipoles and polarizabilities [15]. The external electric
potentials generated by B can then be considered as one- and two-electron opera-
tors to be added to the Hamiltonian H A of A in a treatment requiring a many-
electron description of A but treating B only in terms of its electric multipoles
and polarizabilities. The energy of the solute A in its ground electronic state in a
medium provided by B is then obtained to first order in the potentials as

AB els ind dsp

E0 QA ,QB = 0A H A + V AB + V AB + V AB 0A
266 7 Interactions Between Two Many-Atom Systems

which is valid here only provided the electronic charge distributions of species A
and B in their ground and excited states do not overlap. The electric potentials
can be derived comparing the above terms with the known forms of terms in
AB A AB AB AB
E0 QA ,QB = E0 QA + Eels QA ,QB + Eind QA , QB + Edsp QA , QB . The
ext
external potentials V AB , with ext = els, ind, dsp, can also be used in a pertur-
bation treatment of states of A to obtain more accurate interaction energies
when they are large compared to internal energies in A.
From the previous ground state first-order perturbation energy, one finds

AB els
Eels QA ,QB = d3 r 0A cA r; QA 0A φB, 0 r; QB

B
els 1 c s; QB
φB, 0 r; QB = d3 s 0
4πε0 r−s
els els
V AB r; QA ,QB = cA r; QA φB, 0 r; QB

which is just the classical electrostatic interaction between the ground state
A
charge distributions c0 r = 0A cA r; QA 0A and the electrical potential
els B
φB, 0 created by c0 s = 0B cB s; QB 0B .
The second-order energies lead to forms of the desired induction and disper-
sion electric potentials generated by B, after rewriting
2
Coul Coul Coul
0A ,0B H AB JA ,KB = 0A , 0B H AB JA ,KB JA , KB H AB 0A ,0B

and introducing the resolvents

0 JA JA 0 KB KB
RA = − J 0 A A
, RB = − B B
,
EJ − E0 K 0
EK − E0

to obtain a compact expression for the induction electrical potential (a reaction

ind
potential) φAB, 0 r, r ; QA ,QB , as

ind ind ind

φAB, 0 r, r = φA∗ B, 0 r, r + φAB∗ , 0 r, r
B B
ind 1 0 c0 s c0 s
φA∗ B, 0 r, r = RA d3 s d3 s
4πε0 2 r−s r −s

ind 1 cB s 0 cB s
φAB∗ , 0 r, r = 0A 0A d3 s d3 s 0B RB 0B
4πε0 2
r −s r−s
 7.2 Energetics of a Large Molecule in a Medium 267

which as expected depends on the conformations of both A and B. The term

ind
φA∗ B, 0 corresponds to permanent charges of B interacting with charge fluctua-
tions of A, which can be constructed from the charge distribution or multipoles
ind
of B, and also involves excited states of A. The term φAB∗ , 0 involves excitations
of B that interact with the permanent charges of A and can be constructed from
polarizabilities of B. This electric reaction potential is a two-electron term giving
an induction potential energy
ind ind
V AB r, r ; QA ,QB = cA r; QA φAB, 0 r, r ; QA ,QB cA r ; QA
to be added to the Hamiltonian of A. It contributes an induction energy
AB ind
Eind QA ,QB = d 3 r d 3 r 0A V AB r, r ; QA ,QB 0A

The dispersion electric potential, containing the resolvent

0 JA ,KB JA ,KB
RAB = − J 0 K 0 A A B B
EJ − E0 + EK − E0
is similarly given by the electric potential operator

dsp 1 cB s 0 cB s
φAB, 0 r, r = d3 s d3 s 0B RAB 0B
4πε0 r −s r−s

which is again a compact expression, suitable for calculations in an extended

system B from its polarizability function or from expansions in multipole com-
ponents. This leads to the potential energy operator
dsp dsp
V AB r, r ; QA ,QB = cA r; QA φAB, 0 r, r ; QA ,QB cA r ; QA

and the dispersion energy

AB dsp
Edsp QA ,QB = d 3 r d 3 r 0A V AB r, r ; QA ,QB 0A

arising from the coupling of many-electron density fluctuations in A with the

dynamical polarizabilities of B.
These expressions are especially useful in applications to solvation energetics,
when one wants to calculate the solvation energy of species A interacting with
many solvent molecules in a liquid or solid B, or when B is an extended many-
atom structure enveloping A. Introducing many-electron states of A and B, the
A B
electrostatic energy can be written in terms of the densities ρel, 0 r and ρel, 0 s ,
A
and induction and dispersion energies follow from transition densities ρel, 0J r
268 7 Interactions Between Two Many-Atom Systems

B
and ρel, 0K s . However, it is then necessary to make sure there are no contribu-
tions from overlapping electronic states of A and B, which would lead to small
values of r − s in integrands, and inaccurate energies.
For decreasing distances between solute and solvent, one can proceed some-
what as done when damping long-range interactions for small distances to avoid
unphysical superposition of forces. Here there is no single intermolecular dis-
tance R to construct a damping function, but the problem can be avoided repla-
−1
cing r − s with a damped charge–charge interaction function such as
−1
fd r − s r−s with

exp α r − s − 1
fd r − s ; d,α = ≈0 for r − s 0
exp α r − s − exp αd

where d is the range over which the function is damped as r − s decreases, and
with α giving the slope of the transition region. This function goes to 1.0 for
large r − s d and preserves the values of the electric potentials at these large
distances. The calculation of energies with this damped function requires
numerical evaluation of the integrals over r and s.

7.2.2 Solvation Energetics for Short Solute–Solvent Distances

When the distances between the boundaries of A and B are small and their elec-
tronic charges overlap, it is necessary to treat the pair AB as a single entity, to
account for two effects: (i) the combined Pauli repulsion of electronic distribu-
tions and interelectronic exchange energy, resulting from the requirement of
wavefunction antisymmetry, and (ii) electron transfer between A and B. This
can be done with additional effective one-electron potential energy operators
exc −rep ctr
V AB r; QA ,QB and V AB r; QA ,QB . They can be obtained from Har-
tree–Fock or SAPT treatments [16, 17] and from approximate treatments of
the electronic states in the AB, A+B−, and A−B+ pairs [14, 18]. The total energy
AB B
E0, slt for A in a solution must also include the energy E0, slv of the solvent includ-
ing all its interaction energies, as in
AB AB B
E0, slt QA ,QB = E0 QA ,QB + E0, slv QB
AB exc −rep ctr els ind dsp
E0 QA ,QB = 0A H A + V AB + V AB + V AB + V AB + V AB 0A
B B B B B B
E0, slv QB = E0, exc −rep QB + E0, ctr QB + E0, els QB + E0, ind QB + E0, dsp QB

The one-electron potential energy operators can be constructed to

AB exc− rep
reproduce interaction energies in E0, exc− rep QA ,QB = 0A V AB 0A and
 7.2 Energetics of a Large Molecule in a Medium 269

AB ctr
E0, ctr QA ,QB = 0A V AB 0A , with these energies obtained in suitable
approximations.
A reliable treatment of exchange-repulsion and charge transfer at short dis-
tances can be obtained from a self-consistent field (or Hartree–Fock) treatment
of the electronic structure of the pair AB, insofar the short-range region of elec-
tronic interactions is dominated by Coulomb and exchange electron energies.
In the notation of Chapter 6, the H–F energy of species A is
A occ occ
E0 HF = jA
jA h jA − ϵjA + jA < kA
jA k A v 1 − t jA k A

in terms of MSOs jA and kA of A, and similarly for B. The H–F energy of the
pair AB is given as
AB occ occ
E0 HF = j
j h j −ϵj + j<k
jk v 1 −t jk

where MSOs j = jA + jB and k = kA + kB extend over both A and B.

This contains the H–F energies of each component A and B, and also their elec-
trostatic, induction, and dispersion interaction energies to the extent they
appear in the H–F treatment. The needed exchange-repulsion energy is the
remaining energy after all these have been subtracted, so that
AB exc −rep
E0, exc − rep = 0A V AB 0A
AB A B AB AB
= E0 HF −E0 HF −E0 HF − E0, els HF − E0, ind HF
AB
− E0, dsp HF
exc − rep
which provides the form of V AB as an operator between states containing
MSOs jA and kA of A. The charge transfer energy and one-electron potential
AB ctr
energy operator in E0, ctr QA ,QB = 0A V AB 0A can be constructed in the
same way from H–F energies for charge states A B or A−B+. But here it is nec-
+ −

essary to use spin-polarized MSOs, consisting of different orbitals for different

spins, to account for the open-shell structures of the ions.
Calculation of all these energy terms is clearly a major undertaking, limited by
computational speed and digital storage. Progress is being made to obtain
results by partitioning the large systems into smaller, more accessible, compo-
nents. This has been computationally implemented within an effective fragment
potential (or EFP) treatment of the medium B and its interactions with the sol-
ute A [14], and has been generalized to include exchange-repulsion and charge
transfer in calculations of intermolecular forces [19]. It has been applied to
amino acids and to aqueous solutions of organic molecules.
Another implementation has developed force fields including polarization
effects, named X-Pol to begin with, and has been applied to water clusters
270 7 Interactions Between Two Many-Atom Systems

and polypeptide chains [20]. It considers the effect of charge distributions

within B as they polarize A and change its force field. In the context of the pre-
vious equations, it amounts to incorporation of the electrical potential
els
φB, 0 r; QB in calculations of the electronic structure of A, by means of a var-
iational procedure for the interaction energy that includes the potential energies
from charges in B, and is a functional of the orbitals in A [21].
In the present treatment, polarization effects on A have been included to sec-
els els
ond order in φB, 0 , insofar the electrostatic energy E0, 0 QA ,QB contains its
ind els 0
first-order effect and the reaction potential φA∗ B, 0 r, r = φB, 0 r RA
els ind
φB, 0 r accounts for the second-order effect within E0, 0 . Terms in the total
AB
energy E0, slt of the molecular solution include in addition the dispersion energy
dsp
E0, 0 of interaction, which arises from mutual polarization of A and B.

7.2.3 Embedding of a Molecular Fragment and the QM/MM

Treatment
Large systems can be treated selecting groups of atoms, or molecular fragment,
in a large many-atom system to be described in detail by means of many-
electron wavefunctions, while simplifying the treatment of the environment
to which the selected atoms are either bonded, or the treatment of the environ-
ment that perturbs the selected atoms through long-range forces. The selected
atoms may be described as a molecular fragment with peripheral bonds satu-
rated with chosen atoms (or links) such as hydrogen, or more generally can
be treated with embedding methods where the fragment is self-consistently
coupled to its environment. The embedding procedure has computational
advantages when the environment is treated by means of a force field (within
molecular mechanics, or MM) while the selected fragment is treated with wave-
functions in a quantum mechanical (or QM) way. In this QM/MM treatment,
the computational effort for an environment (also called the subsidiary system
env
SS or outer system O) containing NAt atoms scales linearly with this number,
while a fragment (called a primary system PS or inner system I) is described with
frg frg At
Nfcn = At nfcn electronic basis functions, which is proportional to the num-
ber of atoms in the group and a sum over the number of basis functions per
frg
atom as shown, and scales in principle as Nfcn to a power of 3 or larger depend-
ing on the quality of the many-electron treatment [22]. The large powers in scal-
ings with the number of basis functions can, however, be decreased using a
variety of computational and software procedures for calculation of electron
integrals, an active area of research [23]. This has recently allowed QM
 7.2 Energetics of a Large Molecule in a Medium 271

treatments of very large fragments, with sizes that can be increased to obtain
convergence in total energy values.
Similar aspects have been treated in the previous section on the interactions of
a molecule in a solution, as they relate to long-range forces but without bonding.
Here more generally one is dealing with strong coupling due to bonding
between a cluster of atoms and its environment. The QM/MM treatments ori-
ginated on work dealing with dynamics and reactions in biomolecular systems
[24–26]. and has been extended in several ways, for example, introducing mod-
els with a transition region between the QM region of the fragment and the MM
region of the environment [27]. The challenge here is to identify the boundaries
of the regions and to allow for changes of the electronic distribution in the QM
region and of the force field in the MM region, due to electronic rearrangements
or chemical reactions within the selected QM region. Recent developments have
been reviewed as they apply to molecular complexes and to biomolecules
[28–31] and are briefly summarized here.
To be consistent with our notation, we let A indicate the selected molecular
fragment (or PS), including a set of link atoms capping bonds, and let B indicate
the surrounding medium (or SS). We show with labels QM or MM that a many-
electron treatment or force field description is being done. The total energy can
be obtained starting with the MM treatment of the whole system and correcting
it with the detailed energy change for the subsystem A,
E AB
QM MM = E AB
MM + E A
QM − E A
MM
The calculation of the MM energies are done as described in Chapter 4, with a
sum of terms for bond distance, bond angles, dihedral angles, electrostatic, induc-
tion, and dispersion energies, parametrized from independent calculations. The
A
remaining term E(A)(QM) is obtained from an electronic wavefunction Φ0 cal-
culated for the state of interest from the Hamiltonian operator H A of A. This is
called a subtraction scheme or mechanical embedding and is straightforward to
implement numerically and to improve if needed by increasing the size of the PS
region. It assumes that the force field in the SS region is unchanged as the two
regions interact, and even if electronic rearrangement occurs in A.
However, physically one knows that all three energy terms depend on all the
atomic positions and are functionals of the total electronic density of the pair
AB. Therefore, charge rearrangement or chemical reactions in A (the PS), invol-
ving bond breaking or formation, affect B (the SS) by polarizing the force field
and require some self-consistent coupling between the two regions. This is done
with an additive scheme or electrostatic embedding writing instead that
AB A B AB
E =E QM + E MM + Eint QM MM
(B)
with E (MM ) obtained from a polarizable force field in B (show as MM ) and
AB
the interaction Eint (QM/MM ) obtained from the quantal expectation value of
272 7 Interactions Between Two Many-Atom Systems

AB
the energy coupling operator between A and B for many-electron states ΦJ .
Coul
This must be calculated from the full Hamiltonian H AB = H A + H B + H AB
in
our previous notation. Here the B (or SS) region is treated as a static but polar-
ized force field so that H B = E B MM is an energy value, and the AB coupling
Coul Coul
H AB = V AB is a potential energy operator that depends on the electronic
and nuclear position variables of A, but is only a parametric function of atomic
AB
positions and polarizable charge densities in B. Therefore, states ΦJ satisfying
Coul AB AB AB
H A + V AB +E B
MM ΦJ = EJ ΦJ

can be generated as functions of the electronic variables in A, with parameters

given by the atomic positions and charge densities in B. In turn, the charge den-
sities in B must be allowed to polarize as electronic rearrangement or chemical
reactions occur in A, and to alter the force field in B. This is considered in more
detail in what follows relating to density functional methods for a large system.

7.3 Energies from Partitioned Charge Densities

7.3.1 Partitioning of Electronic Densities
The energetics of two large interacting molecular structures, as they are present
in organic chemistry and biochemistry, can be described in terms of smaller
molecular fragments or functional groups of atoms. The energetics of materials
and interfaces can instead be treated taking advantage of periodicity of struc-
tures or of the presence of smooth charge densities due to electron delocaliza-
tion. These treatments have in common the introduction of convenient
decompositions of electronic densities for the structures, which can be done
in general by expansions of densities and polarizabilities in suitable basis sets.
The starting point in such partitioning involves the structural and electronic
properties of the free fragments. But this must be modified to account for
changes in the properties due to interactions among the fragments in each mol-
ecule. The interaction is strong when fragments are chemically bonded, and the
property changes must be obtained from knowledge of the many-electron wave-
function of the molecule.
An extensively developed and used treatment introduces expansions of den-
sities and polarizabilities in a basis set of functions χ μs r = Rlm r −r s
2
exp − ζ r −r s with the angular dependence of the (lm) = μ solid spherical
harmonic Rlm localized around selected points r s situated at atoms and bonds.
This leads to treatments of energies involving distributed density multipoles
and distributed multipolar polarizabilities [15]. The one-electron density
 7.3 Energies from Partitioned Charge Densities 273

ρ r = ζ ρ 1 r, ζ; r ,ζ from a many-electron state of molecule A can be rewrit-

ten as ρ r = sμ, tν Pμs, ,νt χ μs r χ νt r where Pμs,,νt is a density matrix in the loca-
lized basis set, which must be obtained from the known many-electron
wavefunction of A. The product of orbitals at locations s and t can be expressed
as an expansion around a new location r a as
st , a a
χ μs r χ νt r = a, λ
Cμν, λ χ λ r

st , a
with known coefficients Cμν, λ , which contain by construction information
about the molecular charge distribution. One can then derive multipole ampli-
a
tudes Qlm from expansions
a a
ρr = a, lm
Qlm χ lm r

a
where χ lm is localized at a position r a in A, and rotates as a spherical tensor.
Doing the same for another molecule B, the interaction Hamiltonian energy
a
for the AB molecular pair can be written as a bilinear expression in Qlm and
b
Qpq operators physically arising from the energy of interaction of multipoles
at r a in A with multipoles at r b in B,
a a, b
H AB = a, lm b, pq
Qlm Tlm, pq Qpqb
a, b
with coupling coefficients Tlm, pq fit to calculated energies for each separate frag-
a, b
ment pair ab and containing a damping factor so that Tlm, pq ≈ 0 as r a −r b
becomes small. This has been done and tested for many systems and requires
a
using rotational and translational properties of the functions χ lm as relative
positions and orientations of A (and B) are changed. The calculations can be
accurate provided they account for changes in values of properties going from
free fragments to fragments interacting with their neighbors, but can be labo-
rious [15].
A simpler but less general procedure decomposes the one-electron density by
means of weights generated from the electronic densities of atoms in A and B
[32]. The required two items of information are the spherically averaged elec-
tronic densities ρ a r of each free atom a appearing in the molecule A, and the
electronic density ρ A r of the whole molecule. A set of atomic weight func-
tions is defined by
a
ρfree r
wa r = a
a ρfree r
274 7 Interactions Between Two Many-Atom Systems

adding as a wa r = 1 . They provide a first version of functions to be used in

integrals over space to coarse-grain the densities as sums over atomic compo-
nents, by means of

d3 r F r = a
d 3 r F r wa r = F
a a

for a generic function F r . The free atom densities can be corrected to account
for bondings in A using knowledge of the molecular density, by defining the
a
bonded atom densities ρbnd r = wa r ρ A r . The atomic deformation density
a a a
is given by Δρbnd r = ρbnd r −ρfree r and the molecular deformation density is
a
Δρ A r = ρ A r − a ρfree r , with both functions providing insight on the
changes due to bonding.
The weights can be used to decompose molecular properties into atomic frag-
ments and to define effective values for each fragment. For example, the physical
proportionality between a fragment electronic volume and its polarizability has
led to the introduction of an effective static dipolar polarizability for atomic
fragment a as

3
a a Ad
3
r wa r ρ A r r −r a
αeff 0 = αfree 0 3
3r ρ A r r −r a
Ad

This together with interpolation over frequencies as described in Chapter 3

a
can provide effective dynamical polarizabilities αeff ω . They have been exten-
sively used to calculate van der Waals interactions in AB from sums over pairs
(ab), with encouraging results involving small and large molecules interacting
with solid surfaces [33].

7.3.2 Expansions of Electronic Density Operators

Doing expansions for density operators, instead of expanding their state-
averaged density functions of positions, allows for a general treatment valid
for both ground and excited electronic states and for couplings between states.
el
The electronic density operator cA r = Ce iδ r −r i can be re-expressed in
terms of density components obtained by expanding it in a complete and ortho-
normal basis set fα A r of functions of electronic positions, with α a triplet of
A ∗ ∗
indices, satisfying d 3 r fα A r fα r = δαα and α fα A r fα A r = δ r − r .
These functions can be chosen to partition real space into manageable regions,
for example, using a coarse-graining treatment for spatial integration, or can be
 7.3 Energies from Partitioned Charge Densities 275

chosen to partition large molecules into fragments. Expansion of the charge

density operator for A gives
∗
el el el
cA r = c f A r , cAα = Ce
α Aα α
f A ri
i α

with operator-valued coefficients, which reconstruct the electronic charge den-

sity operator. The completeness of a set is in practice only approximately sat-
isfied and must be verified by increasing the number of functions in the finite set
el
to reach convergence. Each coefficient cAα is a one-electron partial density
operator extending over a region of space selected by the shape of the density
related fα A function, which allows a breakup of the density into components
chosen to describe certain molecular fragments, or certain properties of an
extended system. The density expansions for A, and for B with a basis set
B
fβ s , transform the energy integrals over space into sums, more conven-
ient for numerical work.
The matrix elements

el A A
JA cAα JA = Ce d 3 r fα A r ρJJ r = Ce ρα, JJ

and similarly for B in its many-electron basis set { KB }, that appear in the inter-
action energies, can be obtained from the many-electron functions. The matrix
A B
elements of one-electron Hamiltonian terms contain only ρα, JJ and ρβ, KK , while
A B
matrix elements of two-electron terms contain products ρα, JJ ρβ, KK . When
A
using a density basis set of localized functions, a population analysis of ρα, JJ
in terms of atomic orbitals will show that many of its terms will be small and
negligible when the location of AOs are far removed from the location of fα A r .
Coul nn
The interaction energy operator can again be written as H AB = H AB +
ne ee
H AB + H AB , with the first term given by the Coulomb repulsion of nuclear
charges (or atomic ion cores) and the other two terms decomposed into con-
tributions from electronic charge components. In particular, the electronic–
nuclear interaction Hamiltonian energy is
ne −1 el AB AB el
H AB = 4πε0 c I
α, b Aα αb
sb Cb + β, a
Ca Iaβ r a cBβ
B
AB 3 fα A r AB 3
fβ s
Iαb sb = d r , Iaβ ra = d s
r −sb ra − s
AB AB
where the Iαb Cb and Ca Iaβ integrals are electric potentials generated by
nuclei of charges Ca and Cb interacting with electronic density distributions.
276 7 Interactions Between Two Many-Atom Systems

This Hamiltonian is a one-electron operator and its matrix elements can be

B
obtained from one-electron density components ραA and ρβ .
The electron–electron interaction is
ee −1 el AB el
H AB = 4πε0 c J
α, β Aα αβ
cBβ

AB 1 B
Jαβ = d 3 r d 3 s fα A r f s
r−s β
with the integral J in the second line giving the interaction of two densities. This
Hamiltonian is a two-electron operator and its calculation involves two-
AB B
electron density components ραβ = ραA ρβ derived from the many-electron
states of A and B separately, insofar they do not exchange electrons. These
one- and two-electron expressions appear in the total Hamiltonian in a form
useful for calculations involving both ground and excited electronic states for
fixed atomic positions. The choices of the density basis sets are made to limit
the number of (α, β) terms needed for convergence.
Interaction energies between A and B in their ground electronic states involve
the total charge densities
A A
c0 r = cnuA r + c f A
α α, 0 α
r
A el
containing the expectation values cα, 0 = 0A cAα 0A , and also involve
dynamical susceptibilities
A A A
χ0 r, r ; iω = α
f
α α
A
r χ0 α,α ; iω fα r

with a similar expression for B. Electrostatic, induction, and dispersion energies

A A
follow from densities cα, 0 and susceptibilities χ 0 α, α ; iω , and similar func-
tions for B.
The electrostatic interaction energy for A and B in their ground electronic
AB AB AB AB
states is Eels = Enn + Ene + Eee , obtained from the corresponding Hamilto-
el A
nians as averages with the operators cAα replaced by ground state values cα, 0 and
similarly for B. Induction and dispersion energies follow from related replace-
ments, so that, for example,
AB∗ 1 A AB B AB A
Eind =− 2 αα ββ
cα, 0 Jαβ χ 0 β, β Jβ α cα , 0
2 4πε0
A∗ B A∗ B∗
and corresponding expressions for Eind and Edsp .
Depending on the structure of the extended system, it is convenient to intro-
duce basis sets that allow for decomposition of the total system into physical
fragments, or allow for expansion of the density function into spatial
 7.3 Energies from Partitioned Charge Densities 277

components. Integrals over the susceptibility position variables can be con-

verted into sums over basis set indices. As mentioned in Chapter 2, two conven-
ient basis sets involve expansions in localized oscillator functions of position
variables, or expansions in Fourier (plane wave) components of the whole
system.

7.3.3 Expansion in a Basis Set of Localized Functions

Given positions r a and r b within the two species A and B, one choice for fα A r is
to use at position r a in A a selected set of Na harmonic oscillator functions

1 λ2 ξ− ξa 2
uma ξ; λ = λ2 Cm exp Hm λ ξ− ξa , m = 1 to Na
2

with Hm(x) the Hermite polynomial [34], containing displacements along direc-
tions ξ = x, y, z, at each location r a , and with optimized exponent parameters λ,
which may vary with direction to account for local anisotropy. The spatial basis
a a a a
functions umx x umy y umz z = um r with (m = mx, my, mz) can be used to
expand the charge distribution as

el
cA r = a A
c el
m a
m uma r

el
where the coefficients ca m provide a coarse-graining representation of the
charge densities, convenient for further treatment of interaction energies in a
many-atom system.
el
Oscillator functions in cA at different locations a and a are not orthonormal
a, a a a
and their overlap integrals Sm,m = um um , involving different nuclear posi-
tions, must be incorporated into calculations. These overlaps are, however, short-
ranged in (a,a ) due to the localized nature of the basis functions. The expansion
a
coefficients are obtained multiplying the above equation to the left times um and
a, a
integrating over space to obtain in matrix notation with S = Sm,m ,

a, a
S −1
a
cael m = a A m i
um ri m ,m

which is a one-electron operator localized around nucleus a. The overlap

integrals S appear in the Franck–Condon treatment of molecular transitions
and have been thoroughly analyzed using harmonic oscillator operators, to
derive recurrence relations in the m indices [35]. Interaction integrals I and
278 7 Interactions Between Two Many-Atom Systems

J within A can be obtained in compact form introducing the inverse function

representation
∞
2
r −1 = 1 dκ exp −κ 2 r 2 ,
π2 0

using the property that a product of Gaussian functions gives a new Gaussian at a
shifted position, and using translation properties of the harmonic oscillator func-
tions within A. The resulting integrals are similar to the ones extensively employed
in calculations of electronic structure with Gaussian basis sets [36]. Expansions
b b b b
for B are done similarly with a basis of functions un = unx x uny y unz z loca-
lized at r b .
Interaction energies between A and B involve oscillator functions of r and s
localized around r a and r b in each structure. Their products overlap in space
at electron positions ρ and the products of Gaussian functions for A and B con-
tain the relation

2 2
exp −α ρ −r a exp − β ρ −r b

αβ 2 2
= exp − r a −r b exp − α + β ρ −r I
α+β

where r I = αr a + βr b α + β is an intermediate location. The first exponential

to the right is very small when fragments of A and B are well separated, indicat-
ing that their interaction energies can be discarded. This can greatly decrease
the number of integrals required to achieve accuracy.
An alternative is to use distributed multipoles with the density operator for A
given as sums of terms localized around positions r a [15],
el el
cA r = a
c L
n, l, m a, nlm a, nlm
r
1 2
4π 2
La, nlm r = Anlm ral Ylm ϑa ,φa exp − αn r −r a
2l + 1
where Anlm is a normalization constant, and (ϑa, φa) are angular variables for
r −r a . The La, nlm r functions contain solid spherical functions with selected
exponential parameters αn. They have known translational and rotational prop-
erties, and can be located at nuclear positions and along bonds. This expansion
allows decomposition of large molecules into molecular fragments selected to
contain certain nuclei.
a
Returning to expansions with the oscillator functions um r , they introduce
el el
localized electronic density operators ca m and cb n into the Hamiltonian
 7.3 Energies from Partitioned Charge Densities 279

int nn ne ee
operator H AB = H AB + H AB + H AB . This leads to expressions for electrostatic,
induction, and dispersion energies in terms of ground state electronic charge
el a a b
densities 0A ca m 0A = c0 m = Ce ρ0,m for A and Ce ρ0,n for B. The
dynamical susceptibility for A is a double sum over locations a and a ,
A Na Na aa a
χ0 r, r ; iω = a A m=1 a A
ua
m =1 m
r χ0 m,m ; iω um r

with small terms when the distance r a − r a is large for molecular fragments far
away, so that many of them can be discarded in calculations. A similar expres-
b
sion applies to the susceptibility of B using un s .
The two expansions replaced in induction and dispersion energy integrals
over space variables contain products of summations over (a, m) and (b, n),
and here again many terms can be discarded when r a −r b is large.

7.3.4 Expansion in a Basis Set of Plane Waves

An expansion in a basis set of plane waves is done enclosing the whole
system in a volume Ω = LxLyLz, and introducing periodic boundary
conditions, which restrict the wavevectors in expansion functions
r qn = Ω −1 2 exp −iqn r = r + L qn to a denumerable set qn with
n = (nx, ny, nz) a triplet of integers, and qnξ = nξ 2π Lξ ,ξ = x,y, z. The choice
Lξ = Nξlξ, with lξ a distance over which densities change, and nξ ≤ Nξ generates
a grid with a desired density of points in the reciprocal space of qn . The charge
density operator for A or B is Fourier transformed from space to reciprocal
space using that qn qn = Ω −1 Ω d 3 rexp − iqn r exp iqn r = δnn , and that
within the volume Ω, this basis set satisfies n qn qn = I so that

cr = n
exp iqn r cn ,
3
d r 1
cn = exp − iqn r c r = CI exp −iqn r I
Ω Ω Ω I

and the coefficients cn are used in interactions energy operators. This expansion
is common in treatments of solid state properties, and is particularly useful for
polymers, surfaces, and solids with periodic atomic structure [37]. Products of
operators like c r c r are likely to contribute little to interactions when the
phase difference qn r −qn r is large, in which case the exponential functions
in the products show fast oscillations that add up to nearly zero in integrals.
The expansion in the plane waves Ω −1 2 exp − iqm r gives for the dynamical
susceptibility
A A
χ0 r, r ; iω = m m
exp iqm r χ 0 qm ,qm ; iω exp iqm r
280 7 Interactions Between Two Many-Atom Systems

These summations can be replaced in the space integrals for induction and
dispersion energies of interaction between A and B, containing also
B
χ 0 qn ,qn ; iω , and can be transformed as done in Section 3.2.5.
The charge density operator of all the charges I in species A distributed over
space can be re-expressed in terms of its Fourier components QA k as in

cA r = I
CI δ r −r IA = m
exp iqm r QA qm

with the inverse QA qm = Ω d 3 rexp − iqm r cA r Ω = I CI exp − iqm r I Ω

and similarly for species B.
Coul
The interaction energy operator H AB can be re-expressed in terms of Four-
−1
ier components using here that r − s = d 3 k 2π 3
exp ik r − s k −2 to
obtain, using that δ qm +qn = Ωδm, − n and q −m = −qm , the result
Coul Ω −2
H AB = m
QA qm QB −qm qm
4πε0
to be incorporated into the first- and second-order perturbation expansions.
The electrostatic interaction energy is, for the ground electronic states of A
and B,
AB Ω A B −2
Eels = m
Q0 q m Q0 −qm qm
4πε0
A
where Q0 qm = 0A QA qm 0A is the total charge density component of A.
AB∗ A∗ B A
The induction energy is Eind + Eind , containing the charge density c0 r
B
and static susceptibility χ 0 s, s
for the first term, and similarly for B in the
second term. They can be obtained from the charge density components
B
QA qm and from the transform χ 0 k, k of the susceptibility into reciprocal
space variables. Doing the integrals over space variables first, followed by inte-
−1
grals over d3k and d3k appearing in the Fourier transforms of r − s , we find
AB∗ 1 A −2 B A −2
Eind =− 2 m m
Q0 qm qm χ0 −qm , −qm Q0 qm qm
2 4πε0
B 1 B
χ0 k, k = d 3 s d 3 s exp −i k s exp − ik s χ 0 s, s =
Ω2
2
= K 0 K
ω − 1 0 QB k K K QB k 0
ℏ
This is a double sum over grid points involving the transformed static suscep-
tibility in reciprocal space, for excited states K and excitation frequencies ωK.
A∗ B
A similar expression applies for Eind reversing the A and B labels.
 7.4 Models of Hydrocarbon Chains and of Excited Dielectrics 281

Using the Casimir–Polder procedure to write the dispersion energy in terms

of dynamical susceptibilities at imaginary-valued frequencies, and with
†
QA qm = QA −qm , we find that

2 ∞
AB Ω 1 1 ℏ A B
Edsp = − m n 2 dω χ 0 qm ,qn ; iω χ 0 −qm , −qn ; iω
4πε0 qm qn 2 2π 0

A 2 ωJ
χ0 qm ,qn ; iω = 0 QA qm J J QA q n 0
ℏ J 0ω 2
J + ω2
A A
with ℏωJ = EJ −E0 , and similarly for B. This gives the dispersion energy as a
double sum over grid points in reciprocal space, and avoids an expansion in
multipolar components of the density.

7.4 Models of Hydrocarbon Chains and of Excited

Dielectrics
7.4.1 Two Interacting Saturated Hydrocarbon Compounds: Chains
and Cyclic Structures
As an example we consider the long-range interaction of two saturated hydro-
carbon chains CH3(CH2)N−2CH3, each containing a large number NC of carbon
atoms in a linear conformation. The two chain axes are taken to be parallel and
at a distance D between them. The dispersion forces can be approximately cal-
culated assuming that electronic polarizations occur at the C─C bonds, and that
the resulting bond polarizabilities are additive. This can be physically justified
and tested numerically by application of a partitioning of the chain charge den-
sity distribution into components labeled by the bond positions along the chain.
Local components of dynamical polarizabilities can be extracted from electronic
structure treatments of the whole chain. These local polarizabilities are not the
same as the polarizabilities of a single bond, because the exchange of bond elec-
trons along the chain leads to charge density interactions among bonds, so that
the local polarizabilities must be parametrized from calculations or measure-
ments. In more detail, the chain is made up of fragments or units
HC(sp3)─C(sp3)H, with each shared C atom counting as a half atom, and with
a C─C bond length lbnd = 0.127 nm [38, 39].
For each bond A(a) at location Ra along chain A we can use an expansion in
a
localized functions um r centered midway between the two C atoms in the
bond, to introduce a localized bond charge density operator
Na
cael r = c el u a
m = 1 a ,m m
r
282 7 Interactions Between Two Many-Atom Systems

el el
with the total density given by cA r = a A ca r . Neglecting bond–bond
couplings insofar the localized functions have little overlap, we can introduce
related bond susceptibilities χ a(ω), which (for now) can be taken as isotropic
averages. Similarly chain B is broken into components B(b) at Rb with their sus-
ceptibilities χ b(ω). The bond–bond induction energies are then

a b a
AB∗ 1 3 c0 r χ 0 s, s c0 r
Eind = − d r d3 r d3 s d3 s 2
a b 2 4πε0 r−s r −s
1
ab∗
ind
= E
a b 0, 0
2
a
where c0 r is the total (nuclear plus ground state electronic) charge density at
A∗ B . The integrals can be simplified consider-
ind
bond a, and similarly for E0, 0
ing that functions in the numerator are localized at Ra and Rb . The integrand can
be expanded in powers of the displacements r −Ra and s −Rb , keeping only
terms of zero and first order corresponding to total charge and dipole. To lowest
order, induction energies result from the charge C(a) at a multiplying the
−2 2
ab∗ = − 4πε0
b ind
static dipolar polarizability α1 0 at b, with E0, 0 C a

b
α1 0 2R4ab , as shown in Chapter 3. This is zero in the present example inso-
far the net charge of bond a is zero here. The same applies to E0, 0 a∗ b . Bond–
ind

bond induction energies appear, however, when there are permanent bond
dipoles and quadrupoles of a, and bond quadrupole polarizability at b.
The dispersion energy follows from a similar analysis leading to

A∗ B∗ 1
a ∗ b∗
dsp
Edsp = E
a b 0, 0
2
∞ a b
ℏ χ 0 r, r ; iω χ 0 s, s ; iω
a ∗ b∗ = −
dsp
E0, 0 d3 r d3 r d3 s d3 s dω 2
2π 0 4πε0 r−s r −s

Here again it is possible to expand the integrand around Ra and Rb and to keep
to lowest order the bond dynamical dipole polarizabilities, which give
E0, 0 a∗ b∗ = −C6 a∗ b ∗
dsp dsp dsp
4πε0 2 R6ab with the C6 coefficient expressed
a b
in terms of dipolar bond polarizabilities α1 ω and α1 ω as shown in
Chapter 3. These polarizabilities can be chosen to be isotropic averages in
the simplest case.
In the present example, the dispersion coefficients for bonds in the two chains
AB
are equal to a value C6 and one finds for chains with Nchn = NC − 1 bonds,
 7.4 Models of Hydrocarbon Chains and of Excited Dielectrics 283

AB
A∗ B∗ C6 Nbnd a
Edsp =− 2 a=1
R−6
b = 1 ab
4πε0
with R2ab = D2 + lbnd
2
a− b 2 . The two summations can be rewritten as a sum over
a = b and a sum over k = a − b , of which there are 2(Nchn − k) at distances
1 2
D2 + lbnd
2
k2 . This gives
AB
A∗ B∗ C6 Nchn Nchn −2 2 Nchn −k
Edsp =− 2 + 3
4πε0 D6 k =1 2 k2
D2 + lbnd
This allows calculation of limits when the chain length Lchn = Nchnlbnd is much
larger or much smaller than the chain-to-chain distance D. When Nchn ≥ 10, the
quotient y = k/Nchn varies almost continuously and the summation can be
replaced by the integral Nchn dy…. Doing the integration with ρ = Lchn/D as
a parameter, one finds [38]
AB
A∗ B∗ C6 Nchn ρ ρ
Edsp, chn = − + 3 atan ρ +
4πε0 2 D6 2
4 lbnd Nchn 4 1 + ρ2
When D Lchn (or ρ 1) and the chains are distant, one simply finds
dsp AB
E0, 0 = − C6 Nchn 2
4πε0 2 D6 , for pairs of bonds with energy varying as
the inverse-six distance. When instead the chains are close and D Lchn (or
ρ 1), however, the interaction of pairs does not reach far, and one finds
A∗ B∗ AB
Edsp, chn = − C6 3π Nchn 4πε0 2 8lbnd D5 , linear in the number of bonds and
AB
varying as the inverse-five distance. The value of the coefficient C6 can be
fit to theory or experiment, and has a magnitude around 0.561
kJmol−1 nm6 [38].
A similar procedure has been followed to describe the interaction of two iden-
tical cyclic compounds on parallel planes. They are taken to be circular with
aligned centers and a circumference of length Lcycl = NClbnd, at a distance D
between planes. The dispersion energy is found to be [38]
AB
A∗ B∗ C6 3π 2 ρ 2 2 1 1
Edsp, cycl = − 1+ +
4πε0 2 8 lbnd 2 D4 ρ 2 + 1
1
2 3 ρ 2 +1 ρ +1
2 2

where now ρ = L/(πD). The linear chain result is recovered at distances D small
compared with L.
More accurately, the bond can be described by anisotropic parallel and per-
a a
pendicular bond polarizabilities α ω and α⊥ ω obtained in a reference
frame attached to the bond, such that for a bond making an angle θ with a ref-
erence z-axis in a space fixed frame, the oriented bond polarizability is
284 7 Interactions Between Two Many-Atom Systems

a a a
αθ ω = cos2 θ α ω + sin2 θ α⊥ ω
a a a
with an average over orientations αAv ω = α ω + 2α⊥ ω 3. Also, a
molecular fragment can be chosen to include the C─H bonds so that the polar-
izability can be constructed along a given axis as α HC − CH =
a
α C− C + 2α C −H introducing expansions in basis sets um r located at
the midpoints of all three atom–atom bonds, and using oriented bond polariz-
abilities. Values of static polarizabilities for a large number of molecular frag-
ments have been collected [40, 41] and can be used to construct dynamical
polarizabilities with interpolation formulae as shown in Chapter 2.

7.4.2 Two Interacting Conjugated Hydrocarbon Chains

Intermolecular forces between two extended systems containing delocalized
electrons involve not just charges and dipoles, but also higher multipoles whose
interactions are not simple functions of the distance between the systems.
A general treatment can be based on charge density operators and their matrix
elements between many-electron states, which are functions of density loca-
tions. The interaction energies require integrations over two or four space vari-
ables and these can best be done expanding the density operators of A and B
with a common set of Fourier basis functions u r;qn = Ω − 1 2 exp −iqn r , to
transform space integrals into sums over Fourier components. These density
components are one-electron operators, and their matrix elements between
many-electron states, as they appear in interaction energies, can be given in
A B
basis sets of delocalized electronic orbitals φj r and φk s . The Fourier
expansion is obtained putting the interacting chains A and B within a periodic
supercell of sides Lx, Ly and Lz large enough to include the electronic densities of
both species, with wavenumbers qnξ = nξπ/Lξ, nξ = 1, … Nξ, max , and with the
total number NG = Nx, maxNy, maxNz, max of grid points as needed for accuracy.
We consider the case of electrons in two identical linear chains of conjugated
polyenes such as CH2 = [CH − CH=]kCH2 with k units of alternating single and
double C─C bonds of lengths l1 and l2, respectively, each with a frame in the
(xz)-plane and the chain axis parallel to the z-axis, with chain A at the origin
of coordinates and a parallel chain B displaced by a distance D between them
along the x-axis. A simple model separates sigma- and pi-electron orbitals on
account of their different symmetry on reflection of orbitals on the plane
of the chain frame. The electronic charge density operator is then
c el r = c σ r + c π r . The sigma-density is localized at bonds and can be
expanded as done before. The total ion and electronic charges are equal and
cancel in a neutral chain. The pi-electron density in a chain with NC carbon
 7.4 Models of Hydrocarbon Chains and of Excited Dielectrics 285

atoms and Nπ = NC pi electrons is instead delocalized and requires expansions in

a set of delocalized MOs for a chain of length Lchn = (NC + 1)(l1 + l2)/2.
π π
Expansion of the two electronic densities cA r and cB r in the same
π
u r;qn basis set provides density components QS qn , S = A,B, and its matrix
elements between pi-many-electron states JS give static densities and suscep-
tibility components χ πS qn , qn ; iω . To calculate the susceptibilities, it is nec-
essary to introduce electronic states and energies of the structures. The
model here is at the simplest level just a collection of independent pi-electrons
described by a one-electron Hamiltonian H π and by many-electron determi-
π
nantal functions containing pi-electron molecular orbitals φj r perpendicu-
lar to the chain frame, from
π π π
H π φj r = ε j φj r
The matrix elements of one-electron operators between many-electron states
π
are given by sums of one-electron integrals containing spin-orbitals φj r ϑ ζ
A
[42]. Susceptibilities involve the pi-electron transition densities ρjj r, r =
π π ∗ B
φj r φj r of A, interacting with ρkk
s,s of B, and again include all elec-
tric multipole components of the chain. The space integrals appearing in the
interaction energies can be expressed in terms of density Fourier components
as described in Section 7.3, for more efficient computational work.
The orbitals can easily be constructed in two approximations:
a) From particle-in-a-box states for a box with the dimensions of the chain, and
positive charges of the atomic ions spread and assumed to be homogeneous
inside the box on the average. The electron densities are obtained from orbi-
tals φ π r;k j of electrons in the box. The total ion and electronic charges
are equal and cancel in a neutral chain.
b) Alternatively and more accurately, the delocalized orbitals can be con-
structed as LCAOs from 2pπ atomic orbitals ξμ r orthogonal to the plane
π
of the chain frame, with φj = μ cμj ξμ as pi-MO-LCAOs.
π
Matrix elements JA QA qn JA of the one-electron density operator com-
ponents between determinantal states are different from zero only if the states
are identical or different by only one spin-orbital (the Condon-Slater rules) [42].
They are given by one-electron integrals j Ce exp −iqn r Ω j between
π
orbitals r j = φj r . Ground state expectation values of electronic density
components follow from the two locations of the chains separated by a distance
π π
D, with 0A QA qn ; 0 0A and 0B QB qn ; D 0B giving the electrostatic
286 7 Interactions Between Two Many-Atom Systems

interaction energy. The dynamical susceptibility χ πA qn ,qn ; iω of A can be cal-

π
culated from integrals 0A QA qn ; 0 JA with orbitals j initially occupied in
the ground state and excited to unoccupied j , and similarly for B. A change
of integration variables from x to x − D gives the relation

−inx πD − iqn r
j ce exp −iqn r Ω j = exp j ce exp j
B Lx Ω
A

which displays the dependence of matrix elements on the distance D.

They provide electrostatic, induction, and dispersion interaction energies.
The electronic term in the electrostatic interaction energy is

AB Ω inx πD A A −2
Eels, π D = n
exp Q0, π qn ; 0 Q0, π −qn ; 0 qn
4πε0 Lx
with wavenumbers qξ = nξπ/Lξ, nξ = 1, … Nξ, max . The total electrostatic inter-
action energy must also include interactions with nuclear charges, and the net
result is zero if the chains are neutral and far apart. Induction and dispersion
energies are obtained from susceptibilities. The dynamical susceptibility
π
χ πA qn ,qn ; iω of A can be calculated from integrals 0A QA qn ; 0 JA and
π
excitation energies Δεj j = ε π j − ε π j appearing in denominators, which
are smallest for the transition between highest occupied (HO) and lowest unoc-
cupied (LU) energy levels. Similar quantities are needed for B displaced by D
along x with the dependence on the distance D appearing in the integrals for
the Fourier components of the density.
In the box model, two chains can be located one inside a box A with bound-
aries 0 ≤ x ≤ L⊥, 0 ≤ y ≤ L⊥,and 0 ≤ z ≤ L = Lchn and the other inside a similar box
B parallel to the first with its origin displaced along the x-axis by a distance D so
that a new boundary is D ≤ x ≤ D + L⊥. In a particle-in-a-box model, the elec-
tronic states are null outside the boxes and at their edges. For A, they are
1 1 1
2 2 2 2 2 2
φπA r;k j = sin kx x sin ky y sin kz z
L⊥ L⊥ L

where kξ = jξπ/L⊥, ξ = x, y and kz = jzπ/L for integers jξ. This assumes that the
spacing of grid points is dense enough, with the total system in a box with Lξ
L⊥, L , allowing choices like Lx, y = 100L⊥ and Lz = 100L to properly describe the
interaction of the chains for varying distance D between them. The box energies are
2
ε π k j = ℏ2 2me k j , 1 ≤ jξ < ∞ , and the ground state energy has quantum
numbers j = (111). The orbitals φπB r ;k l for B are similarly obtained for the
identical displaced chain along x, with boundaries D ≤ x ≤ D + L⊥, 0 ≤ y ≤ L⊥,
1
and 0 ≤ z ≤ L and the factor (2 L⊥ 2 sin kx x− D giving the x-dependence.
 7.4 Models of Hydrocarbon Chains and of Excited Dielectrics 287

The present model assumes that the pi electrons occupy orbitals and are excited
only along the z-axis, but remain in the lowest orbital along x- and y-axes, so
that here jx = jy = 1.
The one-electron integrals j ce exp − iqn r Ω j between box orbitals
r j = φπA r ;k j are products of integrals over the x, y, and z variables. Each
factor contains products of sin and cos functions, readily integrated and giving
negligible values for very large jz. For example, the z-integrals for the density
el
component QA qz are
inz πz 2 L
π inz πz π
jz exp − jz = dz sin jz z exp − sin jz z
Lz L 0 L Lz L
jz − jz jz + jz
iL 1+ exp inz π L Lz − 1 1− exp inz π L Lz −1
=− nz +
πLz 2
jz − jz − nz 2 L Lz
2
jz + jz
2
−nz 2 L Lz
2

which can be further simplified for Lz L , showing then that the matrix ele-
ments change linearly with L /Lz. Integrals over 0 ≤ x ≤ L⊥ and 0 ≤ y ≤ L⊥ have
corresponding forms, also changing linearly with L⊥/Lξ. All integrals decrease
as jξ−2 for large orbital quantum numbers, and as nξ−1 for the density operator
Fourier components. Matrix elements k Ce exp −iqn r Ω k for B
involve the same integrals over y and z as for A, while the integral over x has
new limits ≤x ≤ D + L⊥. Integrals have the same form as given, but with the addi-
tional factor exp(−inxπD/Lx) from the integration over x − D.
Ground state expectation values of electronic density components follow
π A A A
from 0A QA qn ; 0 0A = Q0, x qn ; 0 Q0, y qn ; 0 Q0, z qn ; 0 , where the two
first factors are given by the above integral with jx = jx = 1, jy = jy = 1 and

A Ce jz = Nπ 2 inz πz
Q 0, z q n ; 0 = 2 jz exp − jz
Ω jz = 1 Lz
for an even number of carbon atoms in the chain, and orbitals with
double spin occupancy. The corresponding expectation value for B
π B A
is 0B QB qn ; D 0B = Q0, π qn ; D = exp − inx πD Lx Q0, π qn ; 0 .
The dynamical susceptibility χ πA qn ,qn ; iω of A can be calculated from inte-
π π
grals 0A QA qn ; 0 JA and excitation energies Δεj j = ε π k j −ε π k j =
2
ℏ2 2me π L jz2 −j2z and similar quantities for box B displaced by D along
x, all of which are known for the particle-in-a-box model. These susceptibilities
give the right qualitative dependences with the length of chains, but their
numerical values must be corrected introducing more accurate excitation ener-
π A A
gies Δεj j + ΔεG with a HO–LU energy gap parameter ΔεG . Related induction
288 7 Interactions Between Two Many-Atom Systems

ππ∗ π∗π π∗π∗

and dispersion energies are Eind AB∗ , Eind A∗ B and Edsp A∗ B∗ , which
depend on D and the box dimensions of the chain.
The interaction energies contain not just the dipole–dipole interaction but all
the multipolar contributions in the charge density and in the susceptibilities,
and their interactions in the electrostatic, induction, and dispersion energies
of the chains. When the distance D between A and B is much larger than the
chain dimensions, it is possible to expand in the density components as
exp(−inξπξ/Lξ) ≈ 1 − inξπξ/Lξ within the one-electron transition integrals to
obtain transition dipoles and the dipolar polarizabilities of A and B. Matrix ele-
π∗π∗
ments of H AB vary as nξ−2 and Edsp varies as nξ−6 for large nξ. Adding over
Coul

the nξ index for density components, the dispersion energy recovers in this limit
the van der Waals form corresponding to interacting dipole fluctuations, and a
D−6 dependence on the distance between the chains.
π
In the model using pi-MO-LCAOs φj r = μ cμj ξμ r , coefficients and
orbital energies can be obtained from a one-electron pi-Hamiltonian treatment
where μ = 1 to NC gives the location of a pi-AO, and orbital energies are given in
terms of Coulomb (or atom) and resonance (or bond) energy parameters α and
β, as [42]

1
2 2 jμπ
cμj = sin
NC + 1 NC + 1

jπ
εj = α + 2βcos
NC + 1
π
Matrix elements JA QA qn JA are given by sums over one-electron
atom–atom integrals μ Ce exp −iqn r Ω μ with r μ = ξμ r . These inte-
grals are large only for pi-AOs at the same or near-neighbor locations. This
2
means that we need the Fourier transform of only two densities, ξ1 r and
∗
ξ1 r ξ2 r , for a generic pair of pi AOs. Matrix elements can be obtained
from Fourier integrals μ ce exp − iqn r Ω μ = 1 Ce exp −iqn r Ω 1 =
Ce Ω Iatm qn for an atom in chain A and μ Ce exp −iqn r Ω μ + 1 = 1
Ce exp −iqn r Ω 2 = Ce Ω Ibnd qn for a bond in A, and corresponding
integrals Iatm,bnd for B which follow by displacing integrands by D along the
x-axis. These integrals have explicit forms in the Fourier variables qn for given
pi-AOs functions and their Fourier transforms.
This model can be made more accurate allowing for changes of parameters
with bond lengths. In particular, excitation energies are more accurate when
allowance is made for changes of the β parameter between short and long bond
 7.4 Models of Hydrocarbon Chains and of Excited Dielectrics 289

lengths, allowing calculation of excitation energies to within an error of 10%

[42]. Electrostatic, induction, and dispersion energies can be given in terms
of the parameters α, β, Iatm, and Ibnd, or α, β1, β2, Iatm and Ibnd, 1, Ibnd, 2 when
allowance is made for two bond lengths.
To complete the treatment of conjugated hydrocarbons, one must also calcu-
late the energies Eσπind AB∗ ,Eσπind A∗ B , and Eσπdsp A∗ B∗ from expansions of
one-electron density operators c π r and c σ r , respectively in the u r;qn
a
and um r basis sets. Furthermore, structures of the two chains other than par-
allel conformations can be considered along similar lines, for example, with the
two chains in a “T” conformation, as done already in early work [40].

7.4.3 Electronic Excitations in Condensed Matter

Light absorbed by a molecular liquid or a molecular crystal creates electronically
excited molecules in strong interactions with their neighbors. Examples are crys-
talline benzene and the simpler Ar crystal.We consider a system containing only
one species M, as an aggregate, which can be generalized to include liquid and
solid solutions of two components A and B. In a first approximation, the state
of the many-molecule system can be expressed in terms of the ground and excited
states of individual molecules M(m), m = 1 to Nmol, assuming that electron
exchange among them is not important. This allows decomposition of the Ham-
iltonian H of the system into sums of individual molecular Hamiltonians H m
int
plus their interactions H mn for each molecular pair (mn). This is a sum
int Coul rep vibr
H mn = H mn + H mn + H mn
containing the Coulomb energy, repulsion energy at short distances, and vibro-
vibr
nic interaction energy H mn resulting from couplings of electrons and nuclear
vibrations.
The Coulomb molecule–molecule interaction term is
−1 −1
Coul
H mn = 4πε0 d 3 r d 3 s c r; Qm r − s c s; Qn

in terms of molecular charge densities c r; Qm , and can be further decomposed

into contributions from electrons and nuclei (or ion cores) for each molecule.
Electronic states ΦJ ,J = J1 ,…,JNmol for the many-molecule system can be
constructed from electronic states Jm of molecule m, obtained from
m
H m Jm = EJ Jm . The ground state is Φ0 ,0 = 01 ,…,0Nmol , while singly-
excited total states corresponding to all molecules except m being in a ground
state 0m and m being in excited state Jm 0m are

ΦmJ = Jm n m
0n
290 7 Interactions Between Two Many-Atom Systems

The interaction among molecules leads to electronic energy transfer so that

states with any one molecule in the excited state J is a superposition
Φ(J) = mcm(J) ΦmJ , with coefficients and corresponding aggregate energy
Eaggr(J) obtained from the eigenstates of the total Hamiltonian of the many-
molecule system,
1 int
H aggr = Hm + H mn
m 2 m n

for J 0 as
H aggr Φ J = Eaggr J Φ J
These states and energies characterize the properties of excitons, which are
defined as electronic excitations delocalized over all the molecules. They have
been extensively studied for molecular solids and for aggregates including
organic macromolecules [35, 43–46].
The total Hamiltonian contains one- and two-molecule terms and its matrix
elements between many-molecule states have similarities with matrix elements
between many-electron states [9]. Here we provide a simplified version assum-
ing no electron exchange between molecules and using a basis set of molecular
states satisfying the approximate orthonormalization Jm Jn = δmn δJJ which is
accurate when electronic overlaps between molecules are small. Matrix ele-
ments of H m and H mn can be obtained in the basis set { Φ0 , ΦmJ } and to sim-
plify, we consider here only one excited molecular state J = 1.
Excitonic energies can be obtained from first-order perturbation theory, as
ΔE1 = Eaggr 1 − Eaggr 0 Φm1 H aggr Φm1 − Φ0 H aggr Φ0
m
Matrix elements EJJ = Jm H m Jm of H m with J = 0, 1 are nonzero only
m m m m mn
for EJJ = EJ , E0, 1 , or E1, 0 . Matrix elements of H mn are VJJ , KK =
Jm Kn H mn Jm Kn and nonzero only for J = J = 1, K = K = 0, and K = K = 1,
J = J = 0, corresponding to molecular excitation transfers 1 0 and 0 1.
Here the Coulomb interaction involves matrix elements Jm c r; Qm Jm of
the density operators, and the density–density interaction contains not only the
dipole–dipole interaction but also all higher multipoles. A formalism for inter-
acting excitons beyond the first-order perturbation treatment can be based on a
model Hamiltonian with linear, bilinear, and quartic terms in the excitation
(and de-excitation) bosonic operators B†m = 1m 0m (and Bm = 0m 1m ) [47].
More accurately, one must solve for the eigenenergies Eaggr(J), J = 0, 1. This
can be conveniently done when the aggregate is a crystal structure or periodic
chain, taking advantage of the periodicity of the molecular lattice. For a cubic
lattice with molecules at locations Rm = ξ mξaξ ,ξ = x,y, z, with −Mξ ≤ mξ ≤
Mξ an integer, and unit cells of sides aξ , a choice of periodic boundary conditions
 7.5 Density Functional Treatments for All Ranges 291

require invariance under the change Rm Rm ± M . The expansion coefficients in

the eigenstates Φ(J) can then be written as cm J = exp ik j Rm bj J with wave-
vectors kξ = jξ2π/(Mξaξ), and integers jξ giving lattice symmetry labels, which
can be used to identify energy bands of the aggregate. Periodicity allows a sim-
pler calculation of the coefficients bj(0) and bj(1) and energies Eaggr(0) and
Eaggr(1) from the eigenvalue equation for Φ(J) [43].

7.5 Density Functional Treatments for All Ranges

7.5.1 Dispersion-Corrected Density Functional Treatments
Dispersion energy functions of intermolecular and intramolecular position vari-
ables can be combined with near-range functionals obtained from DFT and
containing all energies (electrostatic, exchange, semilocal correlation, long-
range induction) except for vdW energies to be added. This is a very active area
of research where several treatments have been proposed and tested by compar-
ison with accurate results from CCSD(T) or SAPT calculations. Here the focus
is on methods applicable to large classes of chemical systems, described as
many-atom structures A and B containing sets of atoms {a} and {b} with their
centers of mass at a distance R in a body-fixed reference frame, and with internal
variables Q = Ra ,Rb referred to their C-of-M.
Although DFT is a theory that should in principle provide accurate results for
all intermolecular distances, in practice it has been implemented with semilocal
functional of the electron density and its space derivatives. As such, energy
results contain electrostatic and charge induction terms but do not include dis-
persion energies, which physically originate in the quantal coupling of charge
density fluctuations in A and in B. We have seen that the dispersion energy
can be separately treated and calculated from the dynamical susceptibilities
of the two species, or from their multipolar polarizabilities. The aim here is
to add them to the semilocal energies while avoiding double counting of
long-range interactions, and to ascertain how important these are for develop-
ment of potential energy surfaces valid at all distances.
This has led recently to treatments where the semilocal DFT energies Esl(R, Q)
at near-range distances, and dispersion (van der Waals) energies Edsp(R, Q) at
large distances are combined to generate results for all intermolecular distances,
while restricting the dispersion contribution to the long range by multiplying it
times a suitable damping function fd (R), where as before fd (R) ≈ 1, for R ∞
and fd (R) ≈ 0, for R 0, so that
Etot R,Q = Esl R,Q + Edsp R,Q fd R
This expression must be generalized when one of the species, such as B, sur-
rounds A and requires damping not only as a function of the distance R between
292 7 Interactions Between Two Many-Atom Systems

centers of mass, but also damping when atoms of A and B come too close, with a
new fd(R, Q) as a function also of interatomic distances.
The interaction of two many-atom molecules has been similarly treated add-
ing atom-pair dispersion energies with pair damping functions. This assumes
that atom-pair dispersion energies are approximately additive, which can be
shown to be correct for many-atom interactions to second order of perturbation
theory, as previously discussed. The schemes developed so far have made dif-
ferent choices for the construction of Edsp from atomic structures, as a sum over
ab
(ab) atom-pair interactions Edsp Rab ) with Rab = R +Ra −Rb multiplying
ab
instead a pair damping function fd Rab , so that

ab ab
Etot R, Q = Esl R,Q + E
a, b dsp
Rab fd Rab

where R is the relative position vector of the molecular centers of mass.

Much effort has been directed to the identification of the correct methods for
calculating the atom-pair terms. They must be constructed from atomic proper-
ties of the atoms in the medium provided by neighboring atoms, which differ
from free atomic properties. They must account for changes in size, shape,
and electrical charge of each bonded atom, for its bonding coordination, and
for changes in its dynamical response to fluctuating electric fields in the same
molecule and from the other interacting molecule. Several treatments have been
recently reviewed and provide many details of alternatives being explored [33,
48]. The above damping function is somewhat arbitrary, and can be chosen to
make the energy a smooth function of position coordinates, or instead to make
forces smoothly varying.
ab
The DFT-D approach constructs Edsp from atomic components with polar-
izabilities obtained from empirical data or more fundamentally, in the so-called
D3 version [48], from time-dependent DFT calculations of dynamical polariza-
tion of the hydrides of each atom a by an oscillating electric field of frequency ω.
This is separately done for the atom a in a medium where Na bonds to nearest
neighbors have been modeled by a corresponding number of single bondings to
Na hydrogen atoms. It provides dynamical polarizabilities for distorted atoms
with coordination numbers, called CNA for an atom a [48], given by functions
N
αa a ω in our notation, to be used in the Casimir–Polder expression for the
N N
dispersion coefficient obtained from αa a ω and αb b ω . The preferred
damping function for the long-range dispersion energy going as R−n for large
n
R is fd R = R n R n + Rd n , containing a parameter Rd to be chosen [49]. It
follows that the dispersion energy, containing R − n fd R , goes to a constant
n

at R = 0, which agrees with theoretical results [50] and provides a smoothly

 7.5 Density Functional Treatments for All Ranges 293

varying force along R. Results of DFT-D3 calculations of potential energy sur-

faces done for several functional, including PBE-D3, B3LYP-D3, and TPSS-D3,
have been compared with accurate values of equilibrium energies obtained from
the CCSD(T) treatment. The comparisons for two interacting Ar atoms,
π − π stacking energy of two coronenes in a dimer, and the lattice energy of
the benzene crystal versus unit cell volume, all shown marked improvements
when the D3 procedure is applied [48].
An alternative treatment of the atoms a in a medium introduces distributed
densities ρa r and polarizabilities that are coupled to account for medium
effects and many-body shifts of frequencies, in a new so-called many-body dis-
persion (MBD) treatment also recently applied and tested [33]. An averaged (or
0 free
scaled) static atomic polarizability αa 0 = νa αa 0 of an atom in a medium
free
is taken to be proportional to its value αa 0 when free, with a scaling factor
0 free 0 free
νa = Va Va obtained from the volumes quotient Va Va =
0 3 free 3
d 3 r ρa r r −Ra d 3 r ρa r r −Ra of bonded and free atoms with
0 free
atomic densities ρa
and ρa
, respectively. A starting partition of the total
density ρ r is obtained by means of approximate weights
free free
wa r = ρ a r a ρa r [32] calculated from free atomic densities to
begin with, and then improved to allow for changes in atomic size and charge,
reflecting electronic rearrangements in molecule A as mentioned in the
section on partitioned densities.
The static polarizabilities are used to construct dynamical polarizabilities
0
αa ω containing an effective excitation energy ℏωa as described in
Chapter 2 on molecular properties and in Chapter 3. These are then entered
in an equation for interacting induced dipoles [15] that couples the atomic
polarizability of a to nearest neighbors c through a short-range form
sr
tξη Rac = 1 − fd Rac tξη Rac of the cartesian tensor dipole–dipole coupling
factor tξη(Rac) introduced in Chapter 3. Effective atomic polarizabilities
eff N
αa ω , or αa a ω in our notation, shaped by their environment, are obtained
in tensor detail from
N 0 Nc 0 sr N
αa, ξηa ω = αa, ξη ω + c=1
α
ζ , λ a, ξζ
ω tζλ Rac αc, ληc ω

and used in a treatment of coupled charge density fluctuations.

The energy of a collection of fluctuating charges in A and B is obtained from a
Hamiltonian operator for the oscillating effective polarizabilities coupled
lr
through a long-range dipole–dipole factor tξη Rab = fd Rab tξη Rab , of
the form
294 7 Interactions Between Two Many-Atom Systems

1 1 2
H MBD = − μ = a, b
∇μ 2 − μ = a, b
ωμ 2uμ
2 2
1 Nb
1 2
lr
+ ω ω αaNa 0 αb
b a b
0 u t
ξ, η a, ξ ξη
ub, η
2 a
1
with atomic mass weighted displacements such as ua, ξ = Ma 2 ξa − Ra, ξ oscillat-
ing with frequencies ωa, ξ . Its eigenvalue equation H MBD Fp u = ℏΩp Fp u
MBD
is solved to obtain many-body dispersion (MBD) energies Edsp R,Q from a
df
sums over NAB = 3 NA + NB −3 eigenfrequency shifts Ωp(R, Q) − Ωp(∞, Q),
given that there are NA atoms in A and NB in B. This provides
MBD
Etot R, Q = Esl R,Q + Edsp R, Q

in a DFT-MBD treatment and has been applied usually combined with the PBE
density functional, to generate results for several benchmark sets of small and
intermediate size compounds (S66 and S22), crystals (X23 set), and large molec-
ular complexes (S12L). They have been compared with accurate results from the
CCSD(T) treatment with generally very good agreement [33].

7.5.2 Long-range Interactions from Nonlocal Functionals

The adiabatic connection fluctuation-dissipation (ACFD) relation gives a for-
mally exact expression for the electron–electron correlation energy in terms
of the dynamical susceptibility, which can be decomposed into semilocal (sl)
sl nl
and nonlocal (nl) terms as Ec R,Q = Ec R,Q + Ec R,Q with the first term
coming from a semilocal DFT functional and the second term given by an
ACFD nonlocal functional. This second term can be constructed for large dis-
tances between A and B from their susceptibilities, as known functionals of the
electronic density so that no parametrization from molecular properties is
needed.
2
The vdW interaction energy EAe, Be , in a notation indicating that only electrons
are being described, can be written in terms of the electronic dynamical
+
susceptibility tensors introduced in Chapter 2, with χ Se r , r ; ω ξη =
el
χ S r , r ; ω ξη ,S = A,B; ξ,η = x,y, z, in a new notation here for two separate
charge distributions. It takes the form
∞
2 ℏ c2e c2e el el
EAe, Be =− d 1 d 1 d 2 d 2 dωχ A 1, 1 ; iω χ B 2, 2 ; iω
2π r12 r1 2
A A B B 0

where integrals extend to nonoverlapping regions around A and B as shown.

Here (1) stands for r 1 and the integral over its element of volume extends only
 7.5 Density Functional Treatments for All Ranges 295

over a nonoverlapping region around A and contains the density of species A as

shown, iω is the imaginary-valued frequency that appears in the Casimir–Polder
transformation as described in Chapter 3, and an assumed isotropy of the elec-
tronic interaction has been used to replace susceptibility tensors by their
isotropic components,
el 1 el
χS r, r ; ω = χ
ξ S
r, r ; ω ξξ
3
The susceptibility function is given by the time correlation of density fluctua-
2
tions around equilibrium values at each position r, and the above EAe, Be there-
fore describes all multipolar interactions. The dipole–dipole vdW interaction
energy proportional to R−6 appears when Ce2 r12 is expanded in powers of
rA1/R and rB2/R to first order in each variable. It can be expressed in terms of
the tensor tξη in Chapter 3 and the dipolar polarizabilities as given in Chapter 2,
el el
αS r , r ;ω ξξ
= Ce2 ξS ξS χ S r , r ;ω ξξ

after using that, insofar density fluctuations integrate to zero, one finds

el el
d 1 χS 1,1 ; iω = d 1 χ S 1,1 ; iω = 0
S S

The relation with DFT is done assuming that the polarizabilities are local. The
original literature by several authors has been updated in recent reviews that
also include more general formalisms [51–54].
The present equations are simplified first choosing the local form
el el
α S r, r ; iω = δ r − r γ S r; iω , and using that for electron distributions
(1) and (2), respectively, at molecules A and B far from each other, their leading
−3
dipole–dipole interaction energy goes as r 1 −r 2 , which gives
∞
dsp 3ℏ el 1 el
EAe, Be =− dω d 1 d 2 γ A r 1 ; iω γ r 2 ; iω
π r 1 −r 2
6 B
0 A B

and applies to either A and B at a relative distance R, or to A surrounded by B. In

the first case, the dispersion energy at large R takes the form
∞
dsp C 6 6 3ℏ el el
EAe, Be R = − AB6 ,CAB = dωΓA iω ΓB iω
R π
0

el el
ΓS iω = d 1 γS 1; iω
S
296 7 Interactions Between Two Many-Atom Systems

When in addition, the local polarizability function is given a form suggested

by plasmon theory [55] as
el c2e ρe r
γS r; iω =
me ω0 r 2 + ω2

dependent on the local electronic density ρe r and a local excitation frequency

ω0 r , the dispersion energy is found to be
2
dsp 3ℏ c2e 1 ρe r 1 1 ρe r 2
EAe, Be = − d 1 d 2
2 me 6
r 1 −r 2 ω0 r 1
2
ω0 r 1 + ω0 r 2 ω0 r 2
2
A B

and can be obtained from local properties of an inhomogeneous many-electron

system in a variety of conformations for interacting A and B.
Local polarizabilities can be obtained from models of the species electronic
structure, such as from the inhomogeneous electron gas, or from a spherical
electronic distribution with an electronic density of states displaying an energy
el
gap ℏω0. One of the proposed forms in the literature is a polarizability γ DF
−1
obtained from the inhomogeneous electron gas, with ω0, DF = 3me ℏkF2
2 1 3
1 + 0 22 s2 , where kF r = 3π 2 ρe r is a local Fermi wavenumber, and
s r = ∇ρe 2kF ρe r is a dimensionless function of the electronic density gra-
6
dient [56]. Using this for species A and B, the coefficient CAB was calculated to
6
reproduce a set of 34 CAA vdW values to an accuracy of about 10%, for pairs
containing identical closed-shell atoms, diatomics, and polyatomic molecules.
This long-range dispersion energy can be combined with results from semi-
local DFT treatments to obtain an exchange-correlation functional for all dis-
tances, using
dsp
Exc R = Excsl ρe r ,…; R 1 −fd R + EAe, Be ρe r ; R fd R
where fd(R) is a damping function going to zero at small distances. The damping
factors can be dropped if the functionals have been chosen to avoid double
counting of vdW forces, but must otherwise be included. Recent calculations
sl
[57, 58] have used a hybrid version of DFT for Exc and has included higher
order multipolar dispersion interactions to compare results with accurate
CCSD(T) values for a set of compounds, showing excellent agreement.
Alternatively, a long-range correlation (lrc) energy, derived from the ACFD
formula given in the previous chapter, can be constructed so that it is valid
for all distances and does not lead to double counting of dispersion energies.
It must also be constructed so that for large interatomic distances, it reproduces
dsp
the form of EAe, Be given above within the plasmon model. A form that satisfies
these conditions is the nonlocal functional [56]
 7.5 Density Functional Treatments for All Ranges 297

nl ℏ 3 3 c4e
Elrc ρe r ,ρe r = d r d 3 r ρe r ρe r
2 2 m2e g r g r g r + g r
2
g r; r − r = ω0,VV r r − r +κ r
obtained from a model of a spherical inhomogeneous electron distribution with
an energy band gap given by
1
ω2p ℏ2 kF4 4
2

ω0, VV r = +c s
3 m2e

where ω2p = 4πρe c2e me = 4c2e 3πme kF3 is a local plasmon frequency squared,
and where κ = b ℏkF 2 m2e ωp with an adjustable parameter b, which controls
the damping of g at short distances r − r . The term containing s prevents a
el
divergence in γ S when both ω 0 and ρe 0.This choice, named VV09/
6
10, has given values for 58 CAA vdW coefficients to an accuracy within 15%,
obtained for a test set of pairs including identical closed- and open-shell atoms,
diatomics, and polyatomic molecules, with the value of the numerical parameter
c = 0.149 multiplying the s4 term adjusted to minimize standard deviations from
6
accurate values of CAA [56]. The above volume integrals over d3r and d3r
extend over the whole space containing both A and B species and have been
constructed so that the dispersion energy vanishes at short distances for a
homogeneous electron gas. An example of results obtained this way is shown
for the methane dimer in Figure 7.2 from [56].
The good accuracy of the described nonlocal density functional
nl 6
Elrc ρe r ,ρe r for CAB coefficients, and its physical behavior for all dis-
tances, allows a treatment combining the more standard semilocal exchange-
sl
correlation functionals Exc ρe r ,… , already tested for interaction energies
nl
at short and intermediate distances, with the nonlocal form Elrc to construct
intermolecular energies valid over all distances R as
nl
Exc R = Excsl ρe r ,…; R + Elrc ρe r , ρe r ; R

now without a damping factor.

7.5.3 Embedding of Atomic Groups with DFT

Large systems can be treated isolating groups of atoms to be described as clus-
ters with suitably saturated peripheral bonds, or can be more generally treated
with embedding methods where a cluster is self-consistently coupled to its envi-
ronment. This is also related to the QM/MM approaches as described in
298 7 Interactions Between Two Many-Atom Systems

0.1
(CH4)2
0.0

–0.1

–0.2
ΔE (kcal mol−1)

–0.3

–0.4

–0.5 Reference
vdW-DF2
–0.6 VV10

–0.7
3.2 3.6 4.0 4.4 4.8 5.2 5.6
R(Å)

Figure 7.2 Methane dimer interaction energy versus distance R. Source: from Ref. [56].
Reproduced with permission of American Institute of Physics. Here R is the distance between
the centers of mass of the monomers. VV10 and vdW-DF2 results were obtained self-
consistently with the aug-cc-pVTZ basis set.

Section 7.2, but can be more generally done using efficient methods from DFT
for both cluster and environment [59–61]. Embedding DFT methods start with
a partition of the physical system into molecular fragments K containing
at
selected atomic cores κ = 1 to NK and assigned NK electrons, with this number
allowed to change as interfragment electronic rearrangement occurs. This num-
ber of electrons can be fractional and interpreted as resulting from a statistical
average over an ensemble of fragments with varying integer number of electrons
in each one. In a simple version, the total system is partitioned into two: a cluster
of selected atoms with K = A and an environment with K = B.
The total energy Etot can be obtained to begin with by means of a density func-
tional embedding theory (or DFET), which describes the whole system, to be
corrected by accurate calculation of the embedded cluster (or emb cls) fragment,
for example, with a correlated many-electron wavefunction (or CW), by adding
CW DFT DFT
a correction term ΔEemb cls = Eemb cls −Eemb cls to Etot . This is similar to the men-
tioned subtraction scheme in the QM/MM approach, and assumes that the
environment is not changed by atomic rearrangements in the cluster.
 7.5 Density Functional Treatments for All Ranges 299

A better description follows from the alternative decomposition of the total

electron density and total (ionic plus electronic) energy into cluster, environ-
ment, and interaction terms, as [61]
Etot = Ecls ρcls + Eenv ρenv + Eint ρcls ,ρenv
with the densities ρcls and ρenv constructed so that the total electronic density is
ρtot = ρcls + ρenv. The functional for the interaction energy changes with densities
of cluster and environment, providing embedding potential energy functions
through the functional derivatives in
δEint cls δEint env
= Vemb r , = Vemb r
δρcls r δρenv r
Equilibration between the two fragments requires that the two embedding
cls env
potentials should be equal and gives the constraint Vemb r = Vemb r =
Vemb r , which is a unique embedding potential for the total system. This is also
the potential that has been introduced in a general density functional formalism
for chemical reactivity [62].
Given a choice of fragment atomic cores and of number of electrons for A and
B, and a constructed density functional at any of the many available levels of
accuracy, it is possible to variationally optimize the energy as a function of
atomic positions, to obtain the potential energy function Etot(QA, QB) . This
can be done writing the total energy as a functional of the densities ρA r
and ρB r , and of the embedding potential Vemb r as
Etot = EA ρA ,Vemb + EB ρB, Vemb + Eint ρA ,ρB ,Vemb
with a straightforward generalization to several fragments K. A potential func-
tional embedding theory (or PFET) has been developed and applied to molec-
ular and materials systems [61]. The computational procedure starts with a
guess for Vemb r , NA, and NB. The fragment functionals EK ρK r ,Vemb r ;
NK , QK ,K = A, B, are constructed and the ρK[Vemb; NA, NB ] are found
for fixed atom positions. The energy is minimized to find the embedding poten-
tial, calculating first δEtot δVemb r = 0 to search for a minimum for fixed {NK},
followed by the calculation of the fragment number of electrons from ∂EK/
∂NK = μ, a chemical potential common to A and B. An iteration of this sequence
leads to the final potential energy surface Etot(QA, QB) for the total system. Some
applications are covered in the following chapter on interactions at surfaces.
An alternative that avoids the issue of possibly finding unphysical fractional
electronic charges in the fragments, as they separate into A + B at large distances
in a dissociation, can be based on the introduction of an energy Etot written as a
functional of electron orbitals as explained in Section 6.4, for the total system, to
be obtained now within an optimized effective potential procedure generalized
300 7 Interactions Between Two Many-Atom Systems

K
for a collection of interacting fragments. The functional Etot ψj r,ζ of a
K
set of j = 1 to NK spin-orbitals ψ j for each fragment K can be constructed to
describe electrons moving in an effective potential Vemb r , with the orbitals
given as functionals of Vemb r . This can be derived from the relation
δE δVemb r = 0, and optimized by energy minimization, to obtain an optimized
OEEP
embedding effective potential (OEEP) Vemb r and the corresponding PES
satisfying Etot(QA, QB) ≈ EA(QA) + EB(QA) with integer number of electrons in
each dissociation species [60].

7.6 Artificial Intelligence Learning Methods

for Many-Atom Interaction Energies
The energy of two interacting structures containing very many atoms, such as
those involving quantum dots, solids, proteins, or biological membranes, pre-
sents special challenges for modeling with computational methods. Potential
energy functions depend on very many atomic position variables and even when
decomposed into fragments, these are too numerous for standard computa-
tional methods. This can be a barrier to discovery of new compounds with desir-
able properties. However, as large numbers of compounds are studied,
providing their conformations and energies, the accumulated data can be used
to identify new compounds with data mining and machine learning (ML) pro-
cedures applied to molecular structures and their interactions.
New methods are being developed making use of artificial intelligence to train
neural networks in AI-NN algorithms to provide desired results. This has been
done as briefly explained in Chapter 4 for a many-atom system in its ground
electronic state, to generate potential energy surfaces for a given structure, as
emb
V Q = a Ea G Q , a sum over atomic components with each one
dependent on a set of global variables G(Q) containing neighbor atomic loca-
tions. The decomposition is applicable to both bonding and nonbonding inter-
actions and can be validated comparing interpolated PES values with results
from selected previous sets of calculations. It requires specifying for each atom
what neighboring atoms are present within a certain range, which defines its
embedding environment. The embedded atom energies contain parameters
found by fitting procedures and can be obtained training an AI neural network
with an AI-NN procedure as described in Chapter 4.
That treatment, designed to calculate details of a potential energy surface for a
given compound, can be extended to allow for calculation of properties of new
compounds and materials from known properties of their fragments [63, 64].
The extended treatments can be applied to calculation of the bonding energy
 7.6 Artificial Intelligence Learning Methods for Many-Atom Interaction Energies 301

of new structures from a large supply of data on related structures already

known, by decomposing the energies of the many compounds and materials
in the data sets into molecular fragments, each with embedded-atom compo-
nents. The data sets contain not only the atomic environment in a single com-
pound but also its variations for the same molecular fragment as present in
many different compounds.
The applicability of the AI-NN method for calculations of the energetics of
new components depends on the extent of the data sets available for the NN
training. A large set of compounds including first and second row elements
of the periodic table is sufficient to generate the energetics of many organic
compounds, utilizing, for example, the ANI-1 potential data base [64]. As the
available data sets of compounds are expanded to include atoms of the third
and higher rows, new organic and inorganic compounds can be investigated.
The general procedure is illustrated in Figure 7.3 in the very simple test case
where the potential energy surface of the water molecule for varying bond variables
is obtained from the known energetics of other compounds containing hydrogen
and oxygen atoms bonding to each other. Here the three atoms making the water
molecule have a conformation given by internal coordinates Q = (q1, q2, q3). A set of
values of global symmetry variables GA(Q)for each atom A are inputs to atomic
NNs, which are trained using data for many molecular species containing H and
O atoms, with its output giving the embedded atom energy. The atomic energies
are added to obtain the total energy, in a procedure detailed in reference [64].
The available AI methods for the energetics of new compounds and for
molecular interactions are being extended to make them both more efficient
and more encompassing. The structure of neural networks can be optimized
with respect to the number of layers and nodes, exploring their best combina-
tion for accuracy and efficiency, insofar the NNs do not necessarily give better
results as those numbers increase. Also the NN training procedures can be
made faster utilizing alternative forms for the symmetry variables GA(Q)
describing radial and angular atomic distributions, with fewer numerical opera-
tions required for training. A by-product of the AI-NN procedure for energetics
is the availability of calculated interaction forces needed in classical molecular
dynamics treatments, so that an AI-NN procedure can be combined with time
integration of equations of motion for atomic positions, where forces are needed
at each of many time steps along trajectories and for many initial conditions.
This has been applied to the unimolecular decomposition of vinyl bromide,
and later on to the reaction dynamics of H + HBr [65].
The accuracy of derived energetics for new compounds depends on the accu-
racy of the values in the data sets. Starting with a DFT-trained network, using
lots of DFT data, and then correcting the network training with a much smaller
data set from accurate CCSD(T)/CBS calculations of input compounds, has
been found to give more accurate energetics than the original results using only
the less accurate DFT calculations [66].
302 7 Interactions Between Two Many-Atom Systems

Figure 7.3 Combined atomic neural

HD-atomic NNP (H2O) network (NN) architectures used to
obtain high-definition NN potentials in a
HD-Atomic NNP procedure, here for the
Q q1 q2 q3 three atoms making the water molecule
with its conformation given by internal
coordinates Q = (q1, q2, q3). A set of
G1O G1H G2H values of global symmetry variables
GA(Q) for each atom A are inputs to
atomic NNs, which are trained using data
for many molecular species containing
H and O atoms, with its output giving the
embedded atom energy. The atomic
energies are added to obtain the total
energy. Source: adapted from Ref. [64].
Atomic Reproduced with permission of Royal
E1O + E1H + E2H
energies Society of Chemistry.

Total
energy ET

Another methodological improvement consists of finding an alternative way

to select fewer components of a data set with active learning (or AL). It goes
beyond ML to incorporate knowledge gained in the NN training about which
subsets of data are more reliable and are sufficient for sampling the space of
new compounds [67]. This is done by discarding redundant data points with
small standard deviation of their calculated atomic energies relative to reference
energy values for those data points, while keeping data points that contain rel-
evant new information.
Extensions of the AI-NN procedure to include many-atom systems such as
solid surfaces, complex materials, and biomolecules are being done, with the
applicability to new structures dependent on how many data points are available
in related data sets for NN training. Some of the applications have already been
mentioned in Chapter 4. In more detail and as an example, intermolecular inter-
actions involved in bond breaking at a solid surface can be evaluated in test cases
where an analytical PES, constructed from hundreds of energy points from DFT
calculations, is known to begin with and is reproduced by NN training to verify
the learning procedure.
This has been done for H2 adsorbed on the (2 × 2)S/Pd(100) adsorbate surface
and with the H─H bond stretched and breaking in dissociation. The original
location of the diatomic was chosen to be at the fourfold hollow, bridge, and
on-top locations of the diatomic center of mass. The AI-NN procedure involved
nine global variables accounting for the periodicity of the surface, with their
values as input for a three-layer NN, and training done with hundreds of epochs
 References 303

(each a cycle of data inputs). The results have shown accuracy errors smaller
than 100 meVs in comparison with the original PES, and have shown smooth
behavior, so that molecular dynamics simulations of dissociation could be accu-
rately done [68]. This can also be done for many other PESs of interactions at
solid surfaces, relevant to a wide range of applications to the discovery of new
materials.
The majority of applications of AI-NN learning have been done to generate
the energies of a many-atom system in its ground electronic state. However, the
ML procedure can also be used for nonadiabatic processes to generate energies,
PES forces and momentum couplings for a set of coupled ground and excited
states, as needed for nonadiabatic dynamics calculations, provided enough
information is available to construct data sets to be used for training and ver-
ification. This has recently been shown to be doable for conical intersections
appearing in the photoisomerization of CH3NH [69] and for coupled ground
and excited states of the reactive triatomic LiFH [70].
Another related application of AI-NN learning of relevance to the develop-
ment of new medical drugs involves the calculation of receptor–ligand scoring
indices for small molecular ligands interacting with large biomolecular recep-
tors. Neural network scoring functions have been found to reproduce results
from state-of-the-art docking programs based on atomistic models, with NN
computational speeds up to two orders of magnitude larger than speeds for
the available scoring procedures [71].
Development of new methods and applications for AI-NN machine learning
is very active at present. These AI methods are complementary to the ones
based on atomistic models and detailed calculations of electronic structure,
because AI methods can be used only when large numbers of data points can
be collected and organized in data sets to be used for NN training and verifica-
tion. Organization of the data sets require the introduction of molecular struc-
ture descriptors based on physical and chemical considerations, a subject that
needs much attention. A good choice of descriptors can lead to the desired
results in a short time, while bad descriptor choices may yet lead to needed
values but after a much longer computational time. When the data sets are avail-
able, the AI-NN procedures are computationally very efficient. Many improve-
ments in accuracy and efficiency of AI-NN machine learning methods can be
expected in the near future.

References
1 Longuet-Higgins, H.C. (1956). The electronic states of composite systems. Proc.
R. Soc. A 235: 537.
2 Dzyaloshinskii, I.E., Lifshitz, E.M., and Pitaevskii, L.P. (1961). The general theory
of van der Waals forces. Adv. Phys. 10: 165.
304 7 Interactions Between Two Many-Atom Systems

3 Longuet-Higgins, H.C. and Salem, L. (1961). The forces between polyatomic

molecules. I. Long-range forces. Proc. R. Soc. Lond. A259: 433.
4 McLahlan, A.D., Gregory, R.D., and Ball, M.A. (1964). Molecular interactions by
the time-dependent Hartree method. Mol. Phys. 7: 119.
5 Linderberg, J. (1964). Dispersion energy and electronic correlation in molecular
crystals. Arkiv f. Fysik 26: 323.
6 Linder, B. (1967). Reaction-field techniques and their applications to
intermolecular forces. Adv. Chem. Phys. 12: 225.
7 Linder, B. and Rabenold, D.A. (1972). Unified treatment of van der Waals forces
between two molecules of arbitrary sizes and electron delocalization. Adv.
Quantum Chem. 6: 203.
8 Lowdin, P.O. (1955). Quantum theory of many-particle systems. II. Study of the
ordinary Hartree-Fock approximation. Phys. Rev. 97: 1490.
9 McWeeney, R. (1989). Methods of Molecular Quantum Mechanics, 2e. London,
England: Academic Press.
10 Hirschfelder, J.O., Brown, W.B., and Epstein, S.T. (1964). Recent developments
in perturbation theory. Adv. Quantum Chem. 1: 255.
11 Goscinski, O. (1968). Upper and lower bounds to polarizabilities and van der
Waals forces. I General theory. Int. J. Quantum Chem. 2: 761.
12 Langhoff, P., Gordon, R.G., and Karplus, M. (1971). Comparison of dispersion
force bounding methods with applications to anisotropic interactions. J. Chem.
Phys. 55: 2126.
13 Mavroyannis, C. and Stephen, M.J. (1962). Dispersion forces. Mol. Phys. 5: 629.
14 Gordon, M.S., Freitag, M.A., Bandyopadhyay, P. et al. (2001). The effective
fragment potential method: a QM-based MM approach to modeling
environmental effects in Chemistry. J. Phys. Chem. A 105: 293.
15 Stone, A.J. (2013). The Theory of Intermolecular Forces, 2e. Oxford, England:
Oxford University Press.
16 Jeziorski, B., Moszynski, R., and Szalewicz, K. (1994). Perturbation theory
approach to intermolecular potential energy surfaces of van der Waals
complexes. Chem. Rev. 94: 1887.
17 Szalewicz, K. (2012). Symmetry-adapted perturbation theoryof intermolecular
forces. WIREs Comput. Mol. Sci. 2: 254.
18 Gordon, M.S., Fedorov, D.G., Pruitt, S.R., and Slipchenko, L.V. (2011).
Fragmentation methods: A route to accurate calculations on large systems.
Chem. Rev. 112: 632.
19 Gordon, M.S., Smith, Q.A., Xu, P., and Slipchenko, L.V. (2013). Accurate first
principles model potentials for intermolecular interactions. Annu. Rev. Phys.
Chem. 64, Palo Alot, CA, USA, Annual Reviews Inc.: 553.
20 Gao, J., Truhlar, D.G., Wang, Y. et al. (2014). Explicit polarization: a quantum
mechanical framework for developing next generation force fields. Acc. Chem.
Res. 47: 2837.
 References 305

21 Xie, W., Song, L., Truhlar, D.G., and Gao, J. (2008). Variational explicit
polarization potential and analytical first derivative of energy: towards a new
generation force field. J. Chem. Phys. 128: 234108.
22 Jensen, F. (1999). Introduction to Computational Chemistry. New York: Wiley.
23 Ufimtsev, I.S. and Martinez, T.J. (2008). Quantum chemistry on graphical
processing units. 1. Strategies for two-electron integral evaluation. J. Chem.
Theory Comp. 4: 222.
24 Warshel, A. and Levitt, M. (1976). QM/MM. J. Mol. Biol. 103: 227.
25 Bash, P.A., Field, M.J., and Karplus, M. (1987). Free energy perturbation method
for chemical reactions in the condensed phase: a dynamic approach based on a
combined quantum and molecular mechanics potential. J. Am. Chem. Soc.
109: 8092.
26 Field, M.J., Bash, P.A., and Karplus, M. (1990). A combined quantum mechanical
and molecular mechanical potential for molecular dynamics simulations.
J. Comput. Chem. 11: 700.
27 Maseras, F. and Morokuma, K.I. (1995). A new integrated Ab Initio+ molecular
mechanics geometry optimization scheme of equilibrium structures and
transition states. J. Comput. Chem. 16: 1170–1179.
28 Friesner, R.A. and Guallar, V. (2005). Ab initio quantum chemical and mixed
quantum mechanics/molecular mechanics (QM/MM) methods for enzymatic
catalysis. Annu. Rev. Phys. Chem. 56: 389.
29 Lin, H. and Truhlar, D.G. (2007). QM/MM: what have we learned, where are we,
and where do we go from here? Theor. Chem. Accounts 117: 185.
30 Senn, H.M. and Thiel, W. (2009). QM/MM methods for biomolecular systems.
Angew. Chem. Int. Ed. 48: 1198.
31 Brunk, E. and Rothlisberger, U. (2015). Mixed quantum mechanical/molecular
mechanical molecular dynamics simulations of biological systems in ground and
electronically excited states. Chem. Rev. 115: 6217.
32 Hirschfeld, F.L. (1977). Bonded-atom fragments for describing molecular charge
densities. Theor. Chim. Acta 44: 129.
33 Hermann, J., DiStasio, R.A.J., and Tkatchenko, A. (2017). First-principles models
for van der Waals interactions in molecules and materials: concepts, theory, and
applications. Chem. Rev. 117: 4714.
34 Cohen-Tanoudji, C., Diu, B., and Laloe, F. (1977). Quantum Mechanics, vol. 1,
Chap. IV. New York: Wiley.
35 May, V. and Kuhn, O. (2000). Charge and Energy Transfer Dynamics in
Molecular Systems. Berlin: Wiley-VCH.
36 Shavitt, I. (1963). The Gaussian function in calculations of statistical mechanics
and quantum mechanics. In: Methods of Computational Physics, vol. 2 (eds. B.
Alder, S. Fernbach and M. Rotenberg), 1. New York: Academic Press.
37 Martin, R.M. (2004). Electronic Structure: Basic Theory and Practical Methods.
Cambridge, England: Cambridge University Press.
306 7 Interactions Between Two Many-Atom Systems

38 Salem, L. (1962). Attractive forces between long saturated chains at short

distances. J. Chem. Phys. 37: 2100.
39 Zwanzig, R. (1963). Two assumptions in the theory of atttractive forces between
long saturated chains. J. Chem. Phys. 39: 2251.
40 Hirschfelder, J.O., Curtis, C.F., and Bird, R.B. (1954). Molecular Theory of Gases
and Liquids. New York: Wiley.
41 Israelachvili, J. (1992). Intermolecular and Surface Forces. San Diego, CA:
Academic Press.
42 Levine, I.N. (2000). Quantum Chemistry, 5e. Upper Saddle River: Prentice-Hall.
43 Davydov, A.S. (1971). Theory of Molecular Excitons. New York: Plenum.
44 Nitzan, A. (2006). Chemical Dynamics in Condensed Phases. Oxford, England:
Oxford University Press.
45 Kittel, C. (2005). Introduction to Solid State Physics, 8e. Hoboken, NJ: Wiley.
46 Mukamel, S. (1995). Principles of Nonlinear Optical Spectroscopy. Oxford,
England: Oxford University Press.
47 Mukamel, S. and Abramovicius, D. (2004). Many-body approaches for
simulating coherent non-linear spectroscopies of electronic and vibronic
excitons. Chem. Rev. 104: 2073.
48 Grimme, S., Hansen, A., and Brandenburg, J.G.C. (2016). Dispersion corrected
mean-field electronic structure methods. Chem. Rev. 116: 5105.
49 Johnson, E.R. and Becke, A.D. (2005). A post-Hartree-Fock model of
intermolecular interactions. J. Chem. Phys. 123: 024101.
50 Koide, A. (1976). A new expansion for dispersion forces and its applications.
J. Phys. B Atomic Mol. Phys. 9: 3173.
51 Langreth, D.C., Lundqvist, B.I., Chakarova-Kaeck, S.D. et al. (2009). A density
functional for sparse matter. J. Phys. Condens. Matter 21: 084203.
52 Dobson, J.F. and Gould, T. (2012). Calculation of dispersion energies. J. Phys.
Condens. Matter 24: 073201.
53 Dobson, J.F. (2012). Dispersion (van der Waals) forces and TDDFT. In:
Fundamentals of Time-Dependent Density Functional Theory (eds. M.A.L.
Marques, N.T. Maitra, F.M.S. Nogueira, et al.), 417. Berlin: Springer-Verlag.
54 Berland, K., Cooper, V.R., Lee, K. et al. (2015). van der Waals forces in density
functional theory: A review of the vdW-DF method. Rep. Prog. Phys. 78: 066501.
55 Ashcroft, N.W. and Mermin, N.D. (1976). Solid State Physics. London, England:
Thomson.
56 Vydrov, O.A. and Van Voohis, T. (2010). Nonlocal van der Waals density
functional: the simpler the better. J. Chem. Phys. 133: 244103–244101.
57 Sato, T. and Nakai, H. (2009). Density functional method including weak
interactions: Dispersion coefficients based on the local response approximation.
J. Chem. Phys. 131: 224104.
58 Sato, T. and Nakai, H. (2010). Local response dispersion method. II. Generalized
multicenter interactions. J. Chem. Phys. 133: 194101.
 References 307

59 Huang, P. and Carter, E.A. (2008). Advances in correlated electronic structure

methods for solids, surfaces, and nanostructures. Annu. Rev. Phys. Chem.
59: 261.
60 Huang, C. and Carter, E.A. (2011). Potential-functional embedding theory for
molecules and materials. J. Chem. Phys. 135: 194104–194101.
61 Libisch, F., Huang, C., and Carter, E.A. (2014). Embedded correlated
wavefunction schemes: theory and applications. Acc. Chem. Res. 47: 2768.
62 Cohen, M.H. and Wassermann, A. (2007). On the foundations of chemical
reactivity theory. J. Phys. Chem. A 111: 2229.
63 Ramakrishnan, R., Dral, P.O., Rupp, M., and von Lilienfeld, O.A. (2015). Big data
meets quantum chemistry approximations: the delta-machine learning
approach. J. Chem. Theory Comput. 11: 2087.
64 Smith, J.S., Isayev, O., and Roitberg, A.E. (2017). ANI-1: an extensible neural
network potential with DFT accuracy at force-field computational cost. Chem.
Sci. 8: 3192.
65 Raff, L.M., Malshe, M., Hagan, M. et al. (2005). Ab initio potential energy
surfaces for complex, multichannel systems using modified novelty sampling
and feedforward neural networks. J. Chem. Phys. 122: 084104–084101.
66 Smith, J.S., Nebgen, B.T., Zubatyuk, R. et al. (2018). Outsmarting quantum
chemistry through transfer learning. ChemRxiv https://doi.org/10.26434/
chemrxiv.6744440.v1.
67 Smith, J.S., Nebgen, B., Lubbers, N. et al. (2018). Less is more: sampling chemical
space with active learning. J. Chem. Phys. 148: 241733.
68 Lorenz, S., Scheffler, M., and Gross, A. (2006). Description of surface chemical
reactions using a neural network representation of the potential-energy surface.
Phys. Rev. B 73: 115431.
69 Chen, W.-K., Liu, X.-Y., Fang, W.-H. et al. (2018). Deep learning for nonadiabatic
excited-state dynamics. J. Phys. Chem. Lett. 9: 6702.
70 Guan, Y., Zhang, D.H., and Yarkony, D.R. (2018). representation of coupled
adiabatic potential energy surfaces using neural network based quasi-adiabatic
Hamiltonians: 1,2 2A’ states of LiFH. Phys. Chem. Chem. Phys. https://doi.org/
10.1039/c8cp06598e.
71 Durrant, J.D. and McCammon, J.A. (2011). NNScore 2.0: a neural network
receptor-ligand scoring function. J. Chem. Inf. Model. 51: 2897.
 309

Interaction of Molecules with Surfaces

CONTENTS

8.1 Interaction of a Molecule with a Solid Surface, 309

8.1.1 Interaction Potential Energies at Surfaces, 309
8.1.2 Electronic States at Surfaces, 314
8.1.3 Electronic Susceptibilities at Surfaces, 319
8.1.4 Electronic Susceptibilities for Metals and Semiconductors, 321
8.2 Interactions with a Dielectric Surface, 324
8.2.1 Long-range Interactions, 324
8.2.2 Short and Intermediate Ranges, 329
8.3 Continuum Models, 332
8.3.1 Summations Over Lattice Cell Units, 332
8.3.2 Surface Electric Dipole Layers, 333
8.3.3 Adsorbate Monolayers, 335
8.4 Nonbonding Interactions at a Metal Surface, 337
8.4.1 Electronic Energies for Varying Molecule–Surface Distances, 337
8.4.2 Potential Energy Functions and Physisorption Energies, 341
8.4.3 Embedding Models for Physisorption, 347
8.5 Chemisorption, 349
8.5.1 Models of Chemisorption, 349
8.5.2 Charge Transfer at a Metal Surface, 354
8.5.3 Dissociation and Reactions at a Metal Surface from Density Functionals, 359
8.6 Interactions with Biomolecular Surfaces, 363
References, 367

8.1 Interaction of a Molecule with a Solid Surface

8.1.1 Interaction Potential Energies at Surfaces
The interaction of a molecule with a solid surface depends on the atomic struc-
ture of the solid, which is usually classified as: (i) a covalent structure, (ii) an
ionic structure, (iii) a weak-bonded (van der Waals) structure, (iv) a hydro-
gen-bonded structure or, (v) a metal. The first four involve localized electrons
Molecular Interactions: Concepts and Methods, First Edition. David A. Micha.
© 2020 John Wiley & Sons, Inc. Published 2020 by John Wiley & Sons, Inc.
310 8 Interaction of Molecules with Surfaces

and can be treated as dielectrics, while metals (and the related semiconductors)
contain delocalized electrons and require a different treatment. Surfaces of
these solids have corresponding electronic properties, which affect the way
the molecule interacts with the surface. Furthermore, the solid can be regular
with a periodic lattice structure or amorphous. Rearrangement of atomic posi-
tions and of electronic distributions occur at a surface when this is formed in a
solid, which further affects its interaction energy with a molecule.
The potential energy of interaction of a molecule near a solid surface can be
constructed as a sum of terms for the molecule interacting with unit cells for a
periodic lattice or with atom groups for amorphous surfaces. This provides a
starting description of more complicated situations, where many molecules
may be present as the solid surface is in contact with a gas, liquid, or another
solid. At short distances between an adsorbed molecule and a surface, the inter-
action leads to physisorption if it is weak and nonbonding, while a strong inter-
action involving electronic rearrangement and bond breaking or formation
leads to chemisorption. The interaction between two adsorbed molecules is both
direct and indirect through surface effects, and their interaction strength
depends on whether they are physisorbed or chemisorbed.
A clean and regular surface of a crystal can be defined by a cut through its
three-dimensional lattice, giving a plane with atomic or molecular units in a
periodic structure. An introduction to the nomenclature and definitions for lat-
tices can be found in book chapters on the physics and chemistry of solids [1, 2]
and in more detailed treatments of clean surfaces [3] and surfaces with adsor-
bates [4]. In brief, a plane through a crystal lattice formed by periodic translation
of a unit cell with side vectors (primitive translations) a, b , c is defined by
Miller indices (h, k, l), which are the inverse of intersection distances along
the three axes, in units of primitive side lengths. Examples for a cubic lattice
are the (001) plane, cutting the c axis and parallel to the a, b plane, (011) cut-
ting b and c axes, and (111) cutting all three.
For a given planar lattice, the primitive translations a, b define five Bravais
lattice types (oblique, square, hexagonal, rectangular, and centered rectangular),
which combine with 10 crystalline point group symmetries to give 17 two-
dimensional space groups. Translation vectors between cell units on a plane
are of the form T s = ma + nb with m and n integers. Lines on the plane are
defined by two Miller indices (h, k) and the distance between two lines, when
a b = 0, is d(h, k) = [(h/a)2 + (l/b)2]−1/2.
Electronic densities in the planar lattice must satisfy periodicity conditions,
and they determine the periodic potential energies of interaction with a nearby
molecule. Periodicity can be imposed introducing the primitive reciprocal lat-
tice vectors A, B , which define a Brillouin lattice in reciprocal space, and
 8.1 Interaction of a Molecule with a Solid Surface 311

are given by A a = 2π, A b = 0, and B a = 0, B b = 2π. Translations in the

reciprocal lattice are given by the reciprocal lattice vectors G = hA + k B.
Considering structures where the c = cnz primitive vector is perpendicular to
the surface plane and along a z-axis with a unit vector nz pointing into the vac-
uum, an origin of coordinates can be chosen at one of the lattice units. The
interaction potential energy V R between an atom or molecule with its center
of mass at location R =Rs + Znz , where Rs has components (X, Y) on the surface,
with Z ≥ 0, and a planar regular lattice must satisfy the periodicity condition
V R +T s = V R . This can be imposed with the choice

V R = V Z exp iG Rs
G G

1
VG Z = d 2 Rs V R exp − iG Rs
As As

where As is the area of a unit surface cell and the VG are Fourier components of
the potential energy at each distance Z. Periodicity follows from the fact that
G T s = 2π hm + kn with the factor in the parenthesis being an integer.
The potential energy V R is composed of long-range, intermediate-, and
short-range interactions of the molecule with the surface and depends also
on their internal atomic positions (QA, QS) for the molecule and surface.
Long-range interactions can be obtained by analogy with the procedure for mol-
ecule–molecule systems, with the difference that molecule–surface interactions
may extend over a large region at the solid surface and also inside it, so that the
overall interaction results from sums of local interactions over the surface and
volume of the solid.
This strongly modifies the dependence of the potential energy on the distance
Z between a molecular center-of-mass and a surface plane. The van der Waals
interaction energy, which goes as R−6 for two molecules, becomes a function
going as Z−3 after addition over the solid’s atomic components, with a coeffi-
cient expressed in terms of the molecular and solid dynamical susceptibilities.
The induction energy between a charged molecule and induced anisotropic
dipoles extending over the solid surface similarly changes from R−4 to Z−1, with
a coefficient given in terms of the solid polarizability per unit volume. And the
interaction energy between a surface dipole layer and a charge at the molecule
becomes a constant independent of Z. This will be shown in more detail in what
follows.
Intermediate- and short-range interaction energies can be obtained from
functionals of the electronic densities of molecule and solid surface, similarly
to what is done for molecule–molecule interactions. The treatment, however,
312 8 Interaction of Molecules with Surfaces

must be adapted to the degree of delocalization of electrons in the solid surface.

When the electronic charge density is well represented by a sum over atomic
components, as is the case for dielectrics, intermediate- and short-range inter-
actions follow from the treatments in previous chapters for interactions in
many-atom systems, applied to the interaction of the molecule with atomic
groups at the surface. In the case of metals, however, electrons are delocalized
and energies depend on the electronic charge distribution over the whole solid
surface, which must be calculated in the presence of the molecule. These two
situations are separately described in what follows.
The combination of long-, intermediate- and short-range forces lead to the
formation of attraction wells in potential energy functions and to adsorption
of the molecule to the surface. Depending on the strength of the binding energy
at the adsorbate site and on the extent of electronic rearrangement, the attrac-
tion leads to physisorption or chemisorption. Physisorption involves small elec-
tronic rearrangement and weak binding of the species, with van der Waals
binding energies of the order of 20 kJ mol−1 (about 200 meV), typical distances
between adsorbate and substrate around 350 pm, where the adsorbed species
maintain their properties so that the total system electronic density can be con-
structed from the densities of separate molecule and solid surface, possibly
weakly perturbed by their interaction. Chemisorption involves larger binding
energies, of the order 200 kJ mol−1 (about 2 eV) corresponding to formation
or breaking of chemical bonds with extensive electronic charge redistribution,
or to electron transfer, with adsorbate–surface bond distances around 150 pm,
and with rearrangement of atomic components of the molecule and surface.
This can be the result of electronic charge transfer at the surface, molecular dis-
sociation, or bonding between molecular and surface atoms. Physisorption and
chemisorption can both be present at a surface as the distance between A and S
changes, with a transition (or activation) energy linking them as shown in
Figure 8.1a. Physisorption and chemisorption will be further considered, with
their treatment emphasizing general concepts and computational methods.
Figure 8.1 shows typical potential energy surfaces for a species A physisorbed
on a surface S, for chemisorption of A on S due to electronic charge transfer,
and for a species AB undergoing dissociation into A + B due to its interaction
with the surface S.
Electron transfer involves two or more potential energy surfaces and their
crossings, as treated in the previous chapter on intermolecular interactions.
The same arguments apply here about adiabatic or nonadiabatic potential
functions and about potential couplings due to atomic displacements leading
to electronic rearrangement. Dissociation at a surface is a simple case of a
chemical reaction and can be treated introducing potential energy functions
of several variable distances to describe dissociation energy barriers, similarly
to what is done for atom–diatom interactions leading to bond breaking and
formation.
 8.1 Interaction of a Molecule with a Solid Surface 313

Figure 8.1 Potential energy V(Z) functions (a)

of the distance Z between a molecular
V
center-of-mass and a surface plane of a
solid S for fixed internal atomic positions in
the molecule and solid. (a) Physisorption of Ze Zm Z
A on S at large distances around Zm and 0
chemisorption at short distances around Ze, A+S
linked by an activation barrier; (b) electron-
transfer chemisorption of A; (c) dissociative
chemisorption of AB on S leading to A + BS,
for variables Z and distance d(A − B) (b) A++S–
between A and B, with a potential energy AS
V
surface saddle point and reaction energy
barrier shown as ABS .
Ze Z
0
A+S

(c)
A+BS
V ABS

Zm Z

AB+S
Ze
d(A–B) A+B+S
ABS≠

The treatment in this chapter is focused on a single molecule interacting with

the surface, but the same concepts and methods can be used to describe the
interaction of two molecules at the surface, or the interaction of the surface with
islands or layers of adsorbed molecules [4, 5].
In some detail, the calculation of the interaction potential energy between
molecule A and solid surface S can be done by analogy with the developments
in Chapter 7 on the interaction of molecules A and B, with B now replaced by
the solid surface S. The interaction energy can be obtained from energy func-
tionals of the densities of AS, A, and S, within a subtraction scheme of the form

VA, S R,QA ,QS = E AS

R, QA ,QS ; ρAS r −E A
QA ; ρA r
−E S
QS ; ρS r
This expression can be used in calculations to describe a situation where the
molecule A and surface S undergo electronic rearrangement as the relative
314 8 Interaction of Molecules with Surfaces

position R changes while the internal positions (QA, QS) remain close to their
values at large relative distances.
However, a more general approach is needed when there is extensive atomic
rearrangement at shorter distances, for example due to molecular dissociation
of A, which affect not only A but also the substrate S. This situation can be
described introducing new fragments A and S and their self-consistent cou-
pling, with new densities ρA and ρS , and are functions of new atomic internal
positions QA ,QS , and the relative coordinates. The fragment A includes A
and neighboring surface atoms, while S is the remaining indented surface struc-
ture. Their densities can be constructed so that the total density is ρAS = ρA + ρS ,
and the total energy of the AS system can be obtained from an additive scheme
(or electronic embedding) writing instead that the total energy for A at relative
location R over the surface is
AS
E AS
=E A
R; ρA r +E S
R; ρS r + Eint R; ρA r + ρS r

AS
which contains the interaction energy Eint between the two coupled subsys-
tems. This is given by the quantal expectation value of the energy coupling oper-
ator between charges of A and S obtained from a many-electron state Φ(AS),
which allows for atomic rearrangement, calculated from the full Hamiltonian
Coul Coul
H AS = H A + H S + H AS = H A + H S + H A S . Alternatively E(AS) can be con-
structed as an approximate functional within DFT. The interaction energy
can be given as
AS
VA, S R = ΔE A
R; ρA r + ΔE S
R; ρS r + Eint R; ρA r + ρS r

where ΔE A
=E A
−E A
and ΔE S
=E S
−E S
are fragment adsorption and
indentation energies changing with R, in an embedding treatment.
The following sections provide details on electronic states, densities, and sus-
ceptibilities for a solid surface, as needed to consider its interaction with
molecules.

8.1.2 Electronic States at Surfaces

The electronic structure of a solid with a surface separating it from a vacuum
can be obtained from a model of atomic ion cores, containing localized elec-
trons, interacting with delocalized electrons. The number Ne of delocalized
electrons moves in the field of positive ion charges with the same but opposite
total charge ce Ne. They can be treated as independent electrons in spin orbi-
tals ψ νσ (x), where ν and σ are orbital and spin quantum numbers and x = r,ζ
 8.1 Interaction of a Molecule with a Solid Surface 315

includes position and spin variables, obtained from a one-electron effective

Hamiltonian. Total energies follow from a sum of one-electron energies ενσ
of occupied orbitals, corrected by electron–electron correlation energies.
More accurately, energies follow from a many-electron treatment where Ne
electrons are described by a many-electron Hamiltonian and wavefunctions
Ψ J x1 , …,xNe accounting for electrostatic, electronic exchange, and correlation
effects. In this case, a practical approach involves the introduction of a set of Ne
Kohn–Sham spin orbitals, the solid electronic density ρ(x), and an energy den-
sity functional.
For an electron moving in a lattice potential with a translational symmetry so
that v r + T = v r , its orbitals φk , j r for each band j and wavevector k must
satisfy the Bloch theorem as φk , j r + T = exp ik T φk , j r , with
k = k1 A + k2 B + k3 C, the solutions of a one-electron Schrodinger equation with
eigenenergies giving bands of energy ϵk , j [2]. Periodicity means that if ϵk , j is
an eigenenergy, so is ϵk +G, j with G a reciprocal lattice vector. The presence of
a lattice potential leads to couplings of orbitals φk , j and φk +G, j and to splitting
of energy bands at edges in reciprocal space corresponding to the elementary
reciprocal vector boundaries of a Brillouin zone [2].
The orbitals of dielectrics and semiconductors, including transition metal
oxides, can be accurately treated within a tight-binding approach that combines
atomic orbitals χ μ r to construct Bloch orbitals as

φk , μ r = T
exp ik T χ μ r − T

Alternatively it is possible to first construct atomic orbitals as combinations of

localized and plane wave functions and then form from them Bloch orbitals of
the correct translational symmetry. For the example of solid Si, its atomic
ground state configuration 3s23p2 allows formation of four hybrids χ j =
spxpypz j, j = 1 to 4, along tetragonal directions, and gives Bloch orbitals
φk , j r with the correct electronic distribution for the Si solid bulk and surface
atoms.
More detailed treatments accounting for electrostatic interactions, exchange,
and correlation can be done with the methods described in the chapter on
many-electron approaches. Among these, a preferred one for solids and their
surfaces is again DFT [6], which has been developed using a variety of func-
tionals and basis sets. Introduction of atomic core pseudopotentials and a basis
set of plane waves has allowed calculations of many solid structures with surface
boundaries [7].
Many of the features of the surface electronic structure of metals and semi-
conductors, and related interaction with nearby atoms and molecules, can
be qualitatively treated within a simple model where the lattice of positive
316 8 Interaction of Molecules with Surfaces

ions is averaged into a homogeneous positive background of density

ρ + Z = ρ + , Z ≤ 0, ρ + Z = 0, Z > 0 per unit volume extending from a planar
surface going through the uppermost layer of ions and perpendicular to the
z-axis down into − ∞ ≥ Z ≥ 0. A set of independent delocalized electrons of
density ρ r are bound to the substrate by its positive charge and form what
is usually called the jellium model. They spill beyond the positive surface
boundary to form an electronic surface dipole.
To allow for the electronic spread into the vacuum and to impose simple
boundary conditions, the electrons are assumed to move in a well of potential
energy V r of relative value zero inside a semi-infinite region with − ∞ ≤ z ≤ D,
with D to be chosen so that the total charge of electrons plus lattice ions
averages to zero, and with V infinitely repulsive for z ≥ D so that the delocalized
electrons stay within − ∞ ≤ z ≤ D. By symmetry, the electron density ρ r = ρ z
and electron states φk r can be obtained from the solution for a particle with
momentum ℏk in a semi-infinite box as

φk r = Ak exp ik r sin k⊥ z −D , z ≤ D, φk r = 0, z > D

where k and k ⊥ are parallel and perpendicular vector components relative to

the surface plane, such that k 2 + k⊥2 = k2. At absolute-zero temperature, the
electronic states satisfy the Fermi–Dirac distribution and are doubly occupied
(with up and down spins) if their energies are below and up to the Fermi value
EF = ℏ2kF2/(2me) [2].
The number of electronic states per unit volume of k-space in a homogeneous
medium of large volume Ω is obtained from periodic boundary conditions as
Ω/(2π)3, and the total number of occupied spin orbitals is Nst = 2 Ω 2π 3
k
4π 0 F dk k 2 = kF 3 Ω 3π 2 . With one electron per occupied spin orbital, so that
Nst = Nel, kF is related to the electronic density ρel = Nel/Ω = kF3/(3π 2) or equiv-
alently to the inverse of the electron spread radius rs defined by the electronic
volume Ω/Nel = 4πrs3/3 = 3π 2/kF3. Values of the spread radius for metals are
within 2.0 ≤ rs/a0 ≤ 6.0, with rs/a0 equal to 2.07, 2.67, 3.02, and 3.93, respectively,
for the common metals Al, Cu, Ag, and Na.
2
The total local density of electrons is ρ r = k ≤ kF dk
3
φk r = ρ z and the
orbital amplitudes Ak can be chosen all equal to A0, to satisfy the neutrality con-
2
dition ρ ≈ ρ+ for z − ∞ far inside the solid. Writing ρ z = A0 k ≤ kF d
2
k
dk⊥ sin2 k⊥ z , z = z − D, and integrating gives

2 3cos 2kF z 3sin 2kF z

ρ z = A0 1+ 2 − 3
2kF z 2kF z
 8.1 Interaction of a Molecule with a Solid Surface 317

which has the limits ρ ≈ |A0|2 = ρ+ for z − ∞ and ρ ≈ 0 for z D−, and is
zero for z > D. This shows that the density rises from z = D inwards and oscil-
lates along decreasing z away from the surface plane, with a length period
zF = kF/π which increases with the density of electrons, and with an amplitude,
which decreases as the density increases. The parameter D can now be fixed
requiring that deviations of the electronic density ρ(z; D) from the value − ρ + ,
which neutralizes the positive background, should add up to zero,
D
or − ∞ dz ρ z; D + ρ + = 0.
More details about the surface electronic structure can be obtained from a
model of a solid with a surface perpendicular to the z-axis at location z = 0, with
an electronic potential energy V(z) of constant value Vin < 0 inside a semi-
infinite region with − ∞ ≤ z ≤ 0, and value Vout = 0 for z ≥ 0 in a vacuum, so that
now electrons stay mostly within − ∞ ≤ z ≤ 0 but spill out into the vacuum. As
before, the electron density ρ r = ρ z and electron states φk r can be obtained
from the solution for a particle with momentum ℏk, but now they decay expo-
nentially into the vacuum when their energy is εk = ℏ2 k 2 2me + Vin < Vout so that

φk r = Ak z exp ik r sin k⊥ z + γ ,z ≤ 0,

φk r = Bk exp ik r exp − κz z > 0

with ℏ2 κ2 2me = Vout − εk . Imposing continuity of φk and ∂φk ∂z at z = 0 and sol-

ving the Schrodinger equation for φk r provide values for the constants γ and Bk
and for the function Ak z . This function can be obtained from boundary con-
ditions, for a delocalized state extending all the way inside − ∞ ≤ z ≤ 0, or as a
bound state localized at the surface and normalizable. Surface states are found
for Ak z = A0 exp k⊥, p z , and the k⊥, p , with p an integer, can be interpreted
as the discretized imaginary part of a complex-valued k⊥ wavevector [2, 8].
These features are maintained in more accurate treatments accounting for
electronic exchange and correlation such as done within the density functional
approach. Here we summarize results from the theory as given, for example, in
Refs. [6, 9]. The functional form of E[ρ] in Chapter 5 can be adapted to the
present jellium model with the uniform positive background charge
ρ + z = ρ + , z ≤ 0, ρ + z = 0, z > 0 adding also a related lattice potential energy
per electron v + r due to the ion cores in the solid, giving the electron–nuclei
interaction energy d 3 r ρ r v + r . Adding to this the electron–electron Cou-
lomb energy and exchange and correlation energy functionals, one finds the
effective potential energy functional per electron veff r; ρ r appearing in the
KS
Kohn–Sham equations for the orbitals φ r . This effective potential energy
k
veff = v + vxc , containing an exchange–correlation term, is quite different from
318 8 Interaction of Molecules with Surfaces

the electrostatic potential v r = v + r + vCoul r , and leads to larger attraction

for electrons inside the metal.
Calculation of orbital energies, with the zero of energy at the bottom of the veff
value as z − ∞, gives a corrected Fermi energy εF and a work function
(the energy needed to remove an electron from the bulk inside the metal)
Wblk = veff(z 0) − εF, if one ignores the distortion of the ion lattice and elec-
tronic charges at the surface. However, the surface forces created by the surface
dipole require an additional energy Wsrf giving a corrected work function
W = Wblk + Wsrf, which can be obtained from density functional calculations.
To proceed further and obtain properties dependent on the lattice structure
of the surface, such as work functions and surface densities as functions of the
surface lattice Miller indices (h, k, l), it is necessary to introduce the ion lattice
structure of the metal and its surfaces and to calculate electronic properties for
them. A convenient model involves a periodic solid slab between two surface
planes of Miller indices (h, k, l) perpendicular to the z-axis, with a group of ions
in each unit cell at the positions T m, n, p , and electronic density ρ r in each cell.
The cell electrostatic potential at a location r is ϕ r − d 0 +T m, n, p ; ρ r , a func-
tional of the density, and the total electrostatic potential is the sum over
cell units,
h, k , l
ϕh, k , l r = m, n, p
ϕ r − d 0 +T m, n, p

This summation extends over all cell units contained within the two surface
planes (h, k, l) defining the slabs and is different for each set of plane indices.
Computational treatments and results of DFT for surfaces can be found, for
example, in Ref. [7].
The potential energy of an electron in the solid can be calculated for this elec-
trostatic potential and provides work functions W(h, k, l), which depend on the
surface indices. Taking the surface to be perpendicular to the z-axis, with an
averaged (over x- and y-variables) electronic potential energy v z , W can be
obtained as the difference of electronic energies
W = v ∞ + EN − 1 −EN = v ∞ −μ
where μ is the chemical potential of the electrons. For example, values for fcc Cu
surfaces are 4.59, 4.48, and 4.98 eV (or 442.9, 432.3 and 480.5 kJ mol−1) for sur-
face planes (100), (110), and (111), respectively. These values are obtained from
photoemission measurements, and DFT calculations show the same trends.
Accurate calculations must account for distortion of the solid lattice near the
surfaces. The work function for Si(111) is calculated to be about 4.15 eV, but
that surface shows reconstruction into Si(111) (2 × 1) and (7 × 7) at varying tem-
peratures, with a related small change in the work function for the (2 × 1) struc-
ture, but giving a work function lower by about 0.2 eV (19.3 kJ mol−1) for (7 × 7).
 8.1 Interaction of a Molecule with a Solid Surface 319

8.1.3 Electronic Susceptibilities at Surfaces

A many-electron system bounded by a surface and driven by an external electric
field, created by an interacting molecule or adsorbate, responds by showing
electronic excitations, which differ for dielectrics or conductors, and contains
effects of surface inhomogeneity. A general treatment of the response can be
formally done for both dielectrics and conductors, starting from a model of
the solid with its electrons moving in the field of its atomic ion cores. Ions
vibrating around equilibrium positions are slow compared to electronic
motions and can be assumed to stay fixed at given ionic conformations while
calculating electronic response. The dynamics of electronic motions is
described as done in Chapter 2 on properties of an extended molecule, in terms
of dynamical susceptibilities.
A time-dependent driving electric field may vary over the lengths of a
bounded solid, and it is necessary to consider a position- and time-dependent
electric potential ϕ r,t . For an extended system of charges I of electrons and
nuclei (or ionic cores) at locations r I , the total charge density operator per unit
volume at space location r is c r = I CI δ r −r I for charges CI and operator-
valued positions r I . The treatment for an extended inhomogeneous system is
conveniently done in terms of the expectation value of the total charge density,
with the potential energy of coupling between the solid and field written as
V SF = dr 3 c r ϕ r,t , and by working directly with the charge density operator.
For a time-dependent electric potential ϕ r,t , the average charge density in a
0
many-electron state a is a sum ca r,t = ca r + ca r, t of a permanent static
0
value ca plus an induced charge density changing over time, with the latter
obtained from the density operators c r,t with a many-electron state Φa(t).
It is convenient to proceed within a formal operator approach to calculate sus-
ceptibilities without expanding in the usually unknown excited states for the
extended system. The average induced charge density ca r, t arises in response
to the applied potential ϕ r,t present from time t = 0 and at all earlier times
0 ≤ t ≤ t, and can be expressed in terms of a delayed response function
1
χ a r,t;r 1 ,t1 for phenomena linear in the applied field, written here using
response functions of time and positions [10–12]. For a system initially in a time
independent unperturbed state a, it is a function of the difference t = t − t1 and
+
it is convenient to introduce the retarded susceptibility χ a r,r 1 ; t = θ t
1
χ a r,t ;r 1 ,0 where θ(t ) is the step function null for negative arguments. This
provides the useful expression
∞
ca r,t = dt d 3 r1 χ a+ r,r 1 ; t ϕ r 1 ,t − t
−∞
320 8 Interaction of Molecules with Surfaces

which is a convolution form with a Fourier transform from time to frequency

+
giving c a r,ω = η d 3 r1 χ a r,r 1 ; ω ϕ r 1 ,ω in terms of the transforms of the
+
three expressions. The dynamical susceptibility χ a ω can be obtained from a
perturbation calculation of the average polarization. As shown in Chapter 2, the
charge polarization is found to be
∞
ca r,t = dτ d 3 r χ cc+, a r, r , τ ϕ r ,t − τ
−∞
i
χ cc+, a r, r ,τ = − Ψa0 0 c r,τ c r ,0 −c r ,0 c r, τ Ψa0 0
ℏ
+
for τ > 0 and χ cc τ = 0 for τ < 0. Here the time-dependent density operator
0 † 0 0
c r,t = U t c r U t , with U t the time-evolution operator of the
unperturbed system, appears at two different times in an expectation value
with the unperturbed state, and the bracket is shown as a function of only
the difference τ = t − t . The two sides can be Fourier transformed into
+
c a ω = χ cc, a ω ϕ ω , which are complex-valued functions of frequency. The
dynamical response can be alternatively treated in terms of the dielectric func-
+
tion εa ω = 1 + χ cc, a ω , which also depends on the electronic positions r , r .
The response function of a bounded solid can be calculated in a variety of
ways, using expansions of operators in a known basis set, or numerically gen-
erating a solution over time. It can also be constructed semiempirically to incor-
porate known features of the response. The response functions can be
calculated introducing a convenient basis set r μ of states to expand the
charge operators, given a reference unperturbed state a, in which case responses
can be expressed in terms of the amplitudes μ c t μ = cμμ t and suscepti-
+
bility matrices χ μμ , a ω .
Excitations in dielectrics can be described as transitions between many-
electron states, similarly to what is done for molecular electronic excitations.
They may be electron–hole pairs localized at ionic or molecular sites, or delo-
calized electron–hole pairs in periodic lattices, forming excitons [13, 14]. Total
electrical susceptibilities arise from the distortion of valence electron shells and
from the displacement of atomic ion cores [15]. For ionic dielectrics such as
NaCl, or dielectrics composed of noble gas atoms or molecules like CO, a simple
model of the susceptibility involves electronic polarizabilities as well as atomic
displacement polarizabilities for each atomic ion core n, and sums over all the
species in a solid bounded by a surface perpendicular to the z-axis at position
Z = 0. This gives a polarization P r,t in the solid, which can be averaged over
variables in r to obtain P t inside the solid and P t = 0 outside.
The electric field int t inside the solid is related to an external field ext t
outside it by int t = ext t + ϵ0 − 1 P t . Insofar the polarization is a delayed
 8.1 Interaction of a Molecule with a Solid Surface 321

response to the applied external field, its value is P t = dτχ + τ ext t −τ ,

with χ (+)(t) a susceptibility in the solid, null outside it. The dielectric
displacement function t = dτε + τ ext t − τ defines the solid dielectric
+
function ε + t = ϵ0 εr t and equals t = ϵ0 int t inside the solid. Given
an external oscillating field of frequency ω, taking Fourier transforms over
times leads to P ω = χ ω ext ω , ω = ε + ω ext ω = ϵ0 int ω , and
+
to ε r ω = εr ω = 1 + χ ω .
A local field loc t at a chosen location is obtained in more detail as a sum
adding to the external field a depolarization field from physical surface charges,
a field from the surface charges of a model cavity around the location, and the
field of the all the species J inside the cavity, with average density ρJ . Letting
αJ(ω) be the total dynamical polarizability of each species J, the dielectric func-
tion in the solid is given for the many-electron state a by [2]

εr ω −1 1
= ρ αJ ω
εr ω + 2 3ϵ0 J J

This expression can also be used to account for changes across a surface intro-
ducing a molecular density ρJ Z , which varies when averaging its value over
a surface perpendicular to the Z-axis, and gives a varying dielectric function
εr(Z, ω). It can further be simplified by introducing parametrized forms with
frequencies and intensities obtained from independent calculations or measure-
ments. Letting αJ(ω) = αJ,el(ω) + αJ,ion(ω) to include electronic and displaced ion
core terms, these polarizabilities can be written as [2]

c2e me c2J MJ
αJ , el ω = , α J , ion ω =
ωJ , 0 2 − ω2 ωJ , T 2 −ω2

where ℏωJ, 0 is the lowest electronic excitation energy of species J, while cJ, MJ,
and ℏωJ, T are the charge, mass of ion J, and the transversal optical vibrational
energy of the lattice of J ions. Typical energy values for dielectrics are ℏωJ, 0
10 eV (or 103 kJ mol−1) and ℏωJ, T 10−1 to 10−2 eV (or 10 to 1.0 kJ mol−1)
so that for photons in the visible, near-IR and near-UV spectral regions, with ω2
ωJ, 02, the electronic polarizability can frequently be taken to be a constant inde-
pendent of frequencies.

8.1.4 Electronic Susceptibilities for Metals and Semiconductors

The dielectric function ε(Z, ω) of a metal or semiconductor driven by a homo-
geneous external oscillating field ext t; ω must instead be derived keeping in
mind that electrons are delocalized in these solids, so that it is convenient to
322 8 Interaction of Molecules with Surfaces

work with its spatial Fourier transform from Z to the wavenumber q within a
slab of thickness L,

1
εr q,ω = dZexp −iqZ ε r Z, ω
L L

a function of the wavenumber q, which accounts for the components of the

induced electric field with wavelengths 2π/q. Two relevant limits are εr(0, ω),
for phenomena over very long wavelengths such as occur for molecules inter-
acting from large distances to the surface, and the static limit εr(q, 0), which can
be obtained from static external and induced charges [2].
A simple treatment of the electronic dielectric function ϵr, el ω = ε el 0,ω ϵ0 of
metals and semiconductors driven by an external oscillating field ext t; ω is
based on a classical treatment of a homogeneous electronic density ρe dis-
placed a distance u(t; ω) by the field with respect to a positive homogeneous
charge background representing the averaged lattice ions. It is also assumed that
the wavelength of the external field is large compared to the thickness of the
solid, so that the field is homogeneous inside the solid. The displacement creates
surface charges σ = ± uρece perpendicular to the field and a resulting induced
field ind t; ω = − u t; ω ρe ce ϵ0 in between them [2]. The equation of
motion for the displacement of electrons is me ρe d 2 u dt 2 = ρe ce ext , which
can readily be solved for a harmonic field ext t = 0 cos ωt with the choice
u(t) = u0 cos(ωt) giving u t = ce ext t me ω2 . The induced field is then
ind = − ωp ext ω , with ωp = ρece /(ϵ0me) the square of the frequency of
2 2 2 2

the collective oscillation of the electrons, forming the plasmon. The electronic
dielectric function follows from

ext + ind ωp 2
ϵr, el ω = = 1−
ext ω2

which gives propagating waves for positive values, as found from the Maxwell
equations, when ω > ωp.
In more detail, the susceptibilities of conductive materials, whether semicon-
ductors like Si or TiO2, or metals like Cu or Ag, can be obtained from their
atomic structure expanding charge density operators in a basis set of stationary
many-electron states ΨK x1 ,…,xNe and deriving the dynamical susceptibility
+
χ cc, a ω , with K = a, from its expression given above and involving the time-
correlation of charge density operators, as described in Chapter 2 for dynamical
polarizabilities. A useful but more approximate treatment derives the suscepti-
bility from self-consistent field or DFT one-electron treatments with a basis
+
set of Bloch orbitals φk , j r = φλ r giving matrices χ λλ , a ω , a susceptibility
+ +
χa ω = χ a ω and ε r, a ω = εr , a ω = 1 + χ a ω .
 8.1 Interaction of a Molecule with a Solid Surface 323

As explained in the treatment of solids [2, 15], the electronic dielectric func-
tion ϵ el 0, ω = ϵ0 ϵr , el ω of an isotropic metal with electronic density ρe can be
written as a sum over electronic state-to-state transitions Λ = (λλ ), for fixed
nuclei, so that

ρe ce 2 fΛ
ϵr, el ω = 1 +
ϵ0 me Λ ωΛ 2 − ω2 − iωΓΛ

where ωΛ, fΛ, and ΓΛ are transition frequencies, strengths, and rates. In more
detail, the summation must be written as a double integral over densities of band
state λ = k,j and λ = k ,j per unit energy. The quotient ρece2/(ϵ0me) = ωp2 is
the square of the plasmon frequency for the free electron gas of the given den-
sity, with typical plasmon energies ℏωp of the order of 10 eV (about 103 kJ
mol−1) for metals. At zero frequency, the static dielectric constant is then
ϵr,el(0) = 1 + ωp2 ΛfΛ/ωΛ2 = 1 + ωp2/ω02 with ω0 a frequency parameter. At large
frequencies, provided transitions have small strengths fΛ for ω2 ωΛ2, it is
1
ωp 2 ρ e ce 2 2
ϵr, el ω = 1 − , for ω > ωp =
ω2 ϵ0 me

as already found with the simple displacement model. The total dielectric func-
tion including ionic contributions can be obtained within a model of oscillating
ion charges with a restoring (transverse optical) frequency ωT and ionic plas-
mon frequency ωp, ion from ωp, ion2 = ρioncion2/(ϵ0Mion) giving a dielectric func-
tion for each type of ion in the solid metal as

ϵion ω = 1 + ω2 p, ion ωT 2 −ω2

at low frequencies ω < ωT, and ϵion(ω) = ϵion(∞) at the high frequencies when the
ions do not respond to a rapidly oscillating field. The constant parameter ωT2
can be re-expressed in terms of a measurable ϵion(0). The total relative dielectric
function can be written as
2
ϵr ω = ϵr , el ω ϵion ω = ϵion ω − ωp ω ω2

with ωp 2 = ϵion ω ωp 2 [15], at frequencies ω > ωp. Typical values for the static
dielectric function ϵr(0) are about 12 for Si, 11 for Al, and 6 for Cu.
The presence of a surface in a conductor alters its collective electronic oscilla-
tions, and leads to the appearance of slower collective oscillations tangential to
the surface with surface plasmon frequencies ωs = ωp/ √ 2. This follows from
consideration of the total fields inside and outside the surface created by an
oscillating charge localized at the surface, and by imposition of continuity of
the perpendicular electric displacement fields inside and out, at the surface
324 8 Interaction of Molecules with Surfaces

[2]. These displacement fields are z ω = ε ω z ω , with the electric field

z ω of surface charges equal but opposite inside and outside, so that
ε r, ins ω z ω = −ε r , out ω z ω gives 1 − ωp2/ω2 = − 1 from which ωp2/ωs2 = 2.
Some typical surface plasmon energies are 2.25 eV (217.1 kJ mol−1) for Cs and
10.6 eV (1022.7 kJ mol−1) for Al and Si.
In more detail, the dependence of the dielectric function on locations Z near
the surface can be extracted from the Maxwell equations and external and
induced electronic densities ρext r,t and ρind r,t . To simplify, consider only
functions of distances along the z-direction. The Maxwell divergence
relations give ∂ ∂Z = ρext + ρind ϵ0 and ∂ ∂Z = ρext . Taking Fourier trans-
forms to q and ω from Z and t give q q, ω = ρ ext ω + ρ ind q,ω ϵ0
and q q, ω = ρ ext ω and from these q,ω ϵ0 q,ω = ε r q,ω =
1− ρ ind q,ω ρ ext ω + ρ ind q,ω , which can be expanded around q = 0. For
an isotropic medium and absence of the term linear in q, this gives the dispersive
dielectric function
ρ ind 0, ω
ε r q,ω = 1 − − A ω q2
ρ ext ω + ρ ind 0,ω
where A(ω) depends on ρ ext ω and the second term equals ωp2/ω2 for ω > ωp.
An even more detailed treatment of surface effects on dielectric functions can
be derived introducing an inhomogeneous perturbing potential energy Vs r,t
for the surface, and its Fourier transform V s q, ω , and calculating the linear
response to this perturbation by the electrons in the solid. This provides a more
general dielectric function εr q,ω , which accounts for spatial as well as tempo-
ral effects [8].

8.2 Interactions with a Dielectric Surface

8.2.1 Long-range Interactions
Basic insight on the interaction energy of a molecule at or near the surface of a
solid with a dielectric function of frequencies, ε(ω), can be obtained from a
model with point charges outside a semi-infinite solid and with their image
charges created inside the solid by electronic rearrangement. The dielectric
function is related to the total dynamical susceptibility χ ω =
P ω ε0 ω = χ el ω + χ ion ω = ε ω − 1 of the bounded solid, containing
the polarizability P(ω) response of its electrons and atomic cores (ions) to an
oscillating electric field ω as presented in the previous section. A simple
treatment locates a charge C at a distance d above the surface of a semi-infinite
solid with a static dielectric constant (or permittivity) ε = εrε0 as shown in
 8.2 Interactions with a Dielectric Surface 325

(a) z (b) z
–e
→
rH C
+e
d+zH r d
d
0 rD 0
x x
– – – – – – – – – –
L + + + + + + + + + +
d
–eimg d+zH Dv
–d0

+eimg

Figure 8.2 (a) Atomic nucleus and electron point charges ±e above a dielectric and their
images ±eimg inside it; (b) A charge C above a surface dipole layer of a dielectric, with a dipole
density DV per unit volume.

Figure 8.2a, leading to rearrangement of the electronic density near the surface
and to appearance of an image charge Cimg and an electric potential ϕ R , which
can be derived from C and its image charge Cimg at distance −d inside the solid.
Its magnitude is determined by continuity of the gradient of the electric poten-
tial at the surface.
A single positive point charge C at a distance d from a slab surface perpen-
dicular to the z-axis attracts electrons in the slab toward the surface, and this
creates the same effect as a charge image Cimg of opposite sign inside the slab
at a distance d from its surface [16]. The electrical potential ϕ R at a point
R =Rs + Znz in the vacuum due to the charge and its image is

1 C Cimg
ϕ vac R = +
4πϵ0 R + dnz R −dnz

The force between the charge at distance Z from the surface and its image
is Fimg = CCimg/(4πε02Z)2, which gives by integration over d ≤ Z ≤ ∞ a
charge-image potential energy
vac
Eimg = CCimg 4πϵ0 4d
Inside the solid, the electric potential ϕ(sol) has the same form as ϕ(vac) but with
ε0 replaced by εrε0.
326 8 Interaction of Molecules with Surfaces

The distribution σ Rs of surface charges per unit area at Rs induced by the

point charge C outside the solid follows from the potential gradient at the surface
as σ Rs = − 1 4π ∂φ vac ∂Z Rs
. A special case is that of a metal with perfect
conductivity, in which case one finds that Cimg = − C, Eimg = − C2/(4πε04d),
and
C d
σ Rs = −
8π 2 ε0 3
Rs + dnz

which integrates over the surface to a charge −C/(4πε0).

More generally, continuity of the electric field Rs , given by the gradients of
the potentials inside and outside the solid at the surface location, provides the
relation between the charges as [17]
Cimg = −C εr − 1 εr + 1
with the two charges equal and opposite in the special case of a perfectly con-
ducting metal where εr ∞.
As a simple application, consider a H atom near the surface with its nucleus of
charge Cp = e at a distance d and its electron of charge Ce = − e at a position r H
relative to the nucleus. Assuming that these two charges instantly interact with
their opposite images ±e (εr − 1)/(εr + 1), the interaction potential energy oper-
ator is, for proton-(image proton) and electron-(image electron) attractions plus
twice the electron-(other proton) repulsions, from Figure 8.2b,

e2 εr −1 1 1 1
H H , img r H = − − +2
4πε0 εr + 1 4d 4 d + zH 2 2d +r H

where zH is the z-component of r H . This can be expanded in powers of the

electron coordinates in r H . The zeroth-order term is zero for the neutral atom,
the first order gives forces that add up to zero, and the second-order term pro-
vides the leading average electronic energy. In detail,
e2 εr −1 2
H H , img r H = − r H + zH 2 +
4πε0 16d 3 εr + 1
and upon averaging over the H atom ground state g H, with g ξH 2 g H
=
2
g rH g 3 = a0 for ξ = x, y, z, this gives a long-range electrostatic potential
H
energy

els e2 εr − 1 2
VH , img d = H H , img =− g rH g
g 4πε0 12d3 εr + 1 H
 8.2 Interactions with a Dielectric Surface 327

that shows its distance dependence and trend with changing dielectric permittiv-
ity. The interaction is attractive insofar εr > 1, it decreases as εr 1 and reaches
an attraction maximum when εr ∞, in pure metals.
Along the same lines, another application of the images model can be made
for a polar molecule A with a permanent dipole DA at an angle ϑ with the surface
normal nz . From a model of two opposite charges making the dipole, and their
images one finds
els 1 εr −1 2
Vdip, img d,ϑ = − DA 1 + cos2 ϑ
4πε0 12d 3 εr + 1
for the electrostatic interaction energy. Here again the energy varies as d−3 with
distance, and reaches a maximum for a perfectly metallic solid.
A more general treatment for electrostatic, induction, and dispersion interac-
tions of a molecule near a surface can be done constructing an interaction Ham-
iltonian from the collection of all charges and their images, and calculating first-
and second-order perturbation energies of interaction. Results can be obtained
from charge distributions and the dynamical polarizabilities of the molecule
and solid. The dispersion interaction energy appears in a second-order perturba-
tion treatment and involves, similarly to the expression for two interacting mole-
cules, the polarizability αA(ω) of the species A, which is a 3 × 3 tensor, and a
dielectric function of frequency, or relative permittivity εr(ω) = εS(ω) of the
solid (or a slab). Insofar a transient dipole DA(ω) creates an image dipole
DA,img(ω) = DA(ω)(εS − 1)/(εS + 1), one expects (and finds) that such quotient
appears in the molecule–surface dispersion energy. More accurately, the treat-
ment also must account for retardation effects in the very long-range interaction
of adsorbate charges with surface charges very far away. This can be done within a
quantal dielectric response treatment to second order in the strength of the inter-
action Hamiltonian, alternatively using quantum field theory [18], the theory of
fluctuating electromagnetic fields [16], response theory [19], and a quantal per-
turbation treatment [20], all of which lead to the same results, with the response
treatment being simpler. The main relations are summarized as follows.
The atomic structure of the whole system can be described by the distance d
between the center-of-mass of the molecule and the surface plane and by the
internal atomic positions QA of the molecule and QS of the surface. The polar-
izability αA and dielectric function ϵS depend on these internal positions. The
result for the molecule–surface dispersion potential energy, omitting fixed
internal positions and with αA given by its the tensor zz-component, is [19]

∞
dsp 1 ℏ ϵS iω − 1
VA, S d =− dω αA iω
4πε0 2π 0 ϵS iω + 1
4 4ω 2ω2 dω
× 3 + 3 + 3 exp − 2
2d c 2d c2 2d c
328 8 Interaction of Molecules with Surfaces

where c is the speed of light. This expression contains retardation effects that
arise for very long-range interactions. For relatively small distances d c/ω,
compared to radiation wavelengths, the dispersion energy becomes
∞
dsp 1 ℏ ϵS iω − 1
VA, S d =− dω αA iω
4πε0 4πd3 0 ϵS iω + 1

which is usually employed to calculate the energy from parametrized polariz-

abilities and dielectric functions. This shows a d−3 dependence on the distance
to the surface, which we’ll find below to also result from the van der Waals R−6
dependence on interatomic distances, after averaging over atomic positions in
the solid lattice. The solid dielectric permittivity function ϵS(iω) can be obtained
as shown in Section 8.1.3 in terms of the sum F ω = J ρJ αJ ω 3ϵ0 , which
leads to

ϵS iω −1 3F iω
=
ϵS iω + 1 2 + F iω
to be obtained from the polarizabilities of species in the dielectric solid.
The quotient containing the solid dielectric function in the above integral
over frequencies is the one that arises in the response of a dielectric to pertur-
bation by an oscillating dipole far from the surface. It can be reinterpreted in
terms of an effective susceptibility of the solid. Introducing the susceptibility
dsp
χ S(ω) = ϵS(ω) − 1, one finds that the quotient in the integral for VA, S is equal
to χ S ω = χ S ω 2 + χ S ω a renormalized solid susceptibility, which accounts
for medium effects and gives a χ S iω smaller than the original χ S(iω). In this
notation,
∞
dsp CA, S 1 ℏ
VA, S d =− 3
,CA, S = dω αA iω χ S iω
d 4πε0 4π 0

with the dispersion coefficient CA,S a function of internal coordinates QA and

QS. This coefficient can be evaluated from the molecular polarizability αA(ω)
as described in Chapter 2 and from models of the solid dielectric susceptibility
χ S(ω) = 3F(ω)/[1 − F(ω)] as given in the previous section. As done for molecules,
we can introduce the function αA(iω) = β(A)(ω). As before, we have for ω ∞
A A∗ A∗
the asymptotic value ω2 β0 ≈ c2e ℏ2 me Nel , with Nel the number of active
electrons of A in its ground electronic state. This together with the static iso-
A A
tropic average value β0 0 = α0 0 at zero frequency gives the previous inter-
A A 2 −1
polation function β0 ω = α0 0 1+ ω ω A with the frequency
A 2 A∗
parameter ω A satisfying α0 0 ωA = c2e ℏ2 me Nel and of the order
0 02 ≤ ω A
≤ 2 00 atomic units [21]. Similar expressions can be used for the
 8.2 Interactions with a Dielectric Surface 329

dsp
species J in the solid dielectric, to parametrize VA, S using static polarizabilities
J
α0 0 of atoms and molecules and their frequency constants ω J .
A more detailed analysis of the coupling of electronic charge density fluctua-
tions at the surface with fluctuations in a molecule at position R, with coordi-
nates (X, Y, Z) relative to the solid surface, leads to a correction needed at short
distances. It replaces Z−3 with the form (Z − Z0)−3 where Z0 is the position of a
reference plane and depends on the surface electronic density [20]. Without
going into details, the procedure is to express the second-order interaction
energy E(2) as a sum over excited states, with state-to-state transition integrals
for the Coulomb electronic interaction Hamiltonian containing 1/r functions
written in cylindrical coordinates ρ,z there, doing a two-dimensional Fourier
transform from ρ to a wavevector q , and re-expressing the sum over excited
states as an integral over frequencies containing the dynamical susceptibilities
of molecule and solid.
∞
This leads to an expression E 2
= −2 π 0 dω αA iω FS iω;q ,Z where Z
is the distance from A to the surface location defined by the first layer of
atomic positions. Derivation of FS ω;q ,Z and its expansion in powers of
dsp
q for large Z leads to the expansion of E(2) as VA, S Z = − A Z 3
1 + 3Z0 Z+ ≈ − A Z −Z0 3 with an explicit form for Z0 in terms of sus-
ceptibilities [20]. This correction to the Z−3 dependence is relevant at short dis-
tances, but a more detailed treatment is needed to account for the transition
from long distances d to short ones. For a molecule interacting with a surface
lattice, the coefficient A depends also on the coordinates (X, Y) of molecule
A above the surface, with the form
dsp 3
VA, S R = − A X, Y Z −Z0

and the coefficient A dependent also on internal coordinates QA of the molecule

and QS of the solid, appearing in the molecular and solid susceptibility functions.

8.2.2 Short and Intermediate Ranges

The interaction energies between a dielectric surface and a molecule at inter-
mediate and short distances can be treated with the same methods used for mol-
ecule–molecule interactions, insofar the electrons in the dielectric are localized.
This means that many of the concepts and tools already introduced in previous
chapters would also help here. They rely, in semiempirical treatments, on the
parametrization of a near-range (short and intermediate range) potential energy
function of atomic positions arising from the overlap of electronic charges
of molecule and surface atomic groups and on smooth fitting to long-range
330 8 Interaction of Molecules with Surfaces

energies. A more fundamental treatment involves calculations using density

functional theory and also more accurate many-electron methods [22–24].
These treatments provide interaction potentials VA, S R for a molecular spe-
cies A containing a group of atoms {a} interacting with a solid surface S contain-
ing lattice groups of atoms {b}. These atomic groups may be in a cell periodically
repeated forming a crystal, or may form an amorphous lattice.
Dielectric surfaces resulting from cuts through dielectric lattices, such as in
Ar or in LiF crystals, have electronic density distributions approximately given
by sums of densities of their unit cell components at location T m, n, p = ma + nb
+ pc at or below the surface. We consider a molecular unit B in the cell made up
of atoms at positions d b . Potentials of interaction at short and intermediate
distances, which are functionals of electronic densities of the molecule and
the surface, can be given to good accuracy as sums of potential energies
Ab
Vm, n, p R = V Ab
R −Rβ with Rβ =d b +T m, n, p ,β = m,n,p,b , between the
molecule A and each atom b in cell component (m,n,p). The total potential
energy of A near the surface S, with both systems in their ground electronic
state, and using a no-dispersion label (nd), is given by

nd
VA, S R = β
V Ab
R −Rβ

Each term can be assumed to contain short- and intermediate-range terms

accounting for electrostatic (including charge-induced polarization), exchange
and correlation energies, as well as long-range electrostatic and induction terms,
and can be constructed as described in the chapter on model potentials. However,
parameters in these potential terms must account for medium effects on the
properties (such as charge, size, polarizability) of the cell atoms b, with values that
may be quite different from those of free atoms. The long-range interactions
describing dispersion forces must be added with a proper damping function at
short distances, fd R , going from 1.0 at large distances Z to zero at Z = 0,
giving
nd dsp
VA, S R = VA, S R + VA, S R fd R

The total potential must satisfy the condition VA, S R +T s = VA, S R , and its
Fourier components VG Z can be extracted from the sum over atomic posi-
tions β given above.
For example, the potential energy for He in empty space at location (X, Y, Z)
interacting with the surface LiF(001) of the ionic solid, with the LiF molecular
unit in a f.c.c. cell of side length a = 0.402 nm, can be constructed from
 8.2 Interactions with a Dielectric Surface 331

interactions of He with ions Li+ at location (0,0,0) and F− at (1/2,1/2,0) in the

primitive cell for the (001) surface. Its form after adding over all pair interactions
can be given as

VHe, S X,Y ,Z = V0 Z + V1 Z Q X,Y

with the function Q satisfying Q(X + a, Y) = Q(X, Y + a) = Q(X, Y). It has been
used in calculations of cross sections for low-energy elastic scattering of the
atom by the ionic surface, which requires introduction of both attractive and
repulsive potential energies along the z-direction. A useful parametrization is
given by [25]

V0 Z = Dexp α Zm −Z exp α Zm −Z − 2 0

V1 Z = −2βDexp 2α Zm − Z
2πX 2πY
Q X,Y = cos + cos
a a
with parameters a = 0.284 nm, α = 11.0 nm−1, D = 7.63 meV (0.7362 kJ/mol),
β = 0.04 to 0.10, and Zm= 0.10 nm, with the range of β values available to fit
experimental results [26]. This gives a potential well along Z, and the same form
can be used for other rare gas atoms with suitable parameters. Addition of the
nd
dispersion energy to this VHe, S R would only give a small correction, com-
pared with the attractive induction energy due to polarization of the atom by
the ions. This potential energy function has been used in calculations of reso-
nance energies and shapes, in low-energy collisions of noble gas atoms with the
ionic dielectric surface.
Better agreement with scattering experimental results were obtained for He +
LiF(001) with a semi-(ab initio) potential energy obtained from a surface-
projectile sum V He, S R of He-(ion pair p) potential energies v He, p R−R β

with each term including a long-range van der Waals attraction as well as a
ind
short-range repulsion, plus an additional term V R describing the induc-
tion attraction of a polarized He atom to the lattice of ions in the solid [27]. The
latt 2
induction term is obtained as V ind
R = − αHe 2 R , where the elec-
latt
tric field = − ∇ ϕ latt is the gradient of the electric potential ϕ latt R of
the lattice ions interacting with He, with the potential itself a sum over all lattice
charges Cb at Rβ . This lattice potential is calculated for LiF(001) as a sum over
Yukawa-type potentials for each ion, decomposed as a Fourier series, and taken
332 8 Interaction of Molecules with Surfaces

He, S
to the limit of the Coulomb potentials. The results is VHe, S R = V R +
ind
V R , or in detail

2
He, S αHe latt
VHe, S R = V
G G
Z exp iG Rs − fd Z G G
Z exp iG Rs
2

latt
with G as derived in [27] in terms of charges and their positions for the given
lattice. The induction interaction is cut-off at short distances with a damping
factor fd (Z) going from 1.0 at large distances to zero at Z = 0.
More accurately, the long-range induction and dispersion energies can be
obtained from the charge distributions and dynamical polarizabilities of the
atom and the solid surface, using methods of response theory [19].

8.3 Continuum Models

8.3.1 Summations Over Lattice Cell Units
When the molecule A near the surface interacts with many substrate lattice cell
units, the total potential energy can be approximated in a continuum model
where the positions T m, n, p of cells are considered to be changing smoothly,
and sums over cell indices (m, n, p) are approximately calculated as integrals.
To simplify, consider a surface where there is one atom B per cubic unit cell,
with its position in the primitive cell at its center d 0 = −d0 c, and its interaction
potential with A given by v(A), and omit the dependence of parameters on inter-
nal coordinates of A and of the solid. The total interaction potential, given as a
sum over all the cells, or

VAS R = m, n, p
v A R − d 0 +T m, n, p

is approximated as an integral for A located on the Z-axis, and with atoms B

forming a lattice with vector position variables Sx = ma, Sy = nb, and Sz = Z −
pc + d0 for fixed Z. The indices m, n, p can be assumed to be continuously chan-
ging as − ∞ < m, n < ∞ , − ∞ < p ≤ 0, with the summation changed into inte-
grals, giving

∞ ∞ ∞
1
VAS R dSx dSy dSz v A S
abc −∞ −∞ Z + d0
 8.3 Continuum Models 333

For a cell species B such that the interaction with A has axial symmetry around
the z-axis, with a = b, integration variables (Sx, Sy) can be replaced by the length
Ss = (Sx2 + Sy2)1/2 and an axial angle. The interaction potential becomes
∞ ∞
2π 1 2
VAS Z dSs Ss dSz v A Ss 2 + Sz 2
a2 c 0 Z + d0

which is now only a function of the distance Z from A to the surface, and can be
calculated for a variety of A–B pair interactions where the distance of A to the
species B is given by (Ss2 + Sz2)1/2.
Three useful examples, with a = b = c, for isotropic potentials and with bars
over their parameters signifying values adjusted for medium effects, are [26]:
a) An inverse power potential energy v(A) = − W(R0/R)m, giving
∞ ∞
π W R0m
VA, S Z = − d Ss2 dSz m
a3 0 Z + d0 Ss 2 + Sz 2 2

3 m −3
2πW R0 R0
=−
m −3 m −2 a Z + d0

which in particular shows that the R−6 atom–atom dispersion interaction

becomes a (Z + d0)−3 atom–surface interaction.
12 6
b) The Lennard-Jones (12,6) potential v A = ϵ Rm R − Rm R , giving

9 3
zm zm
VA, S Z = W −3
Z + d0 Z + d0

and zm Rm = 5 − 1 6 , showing how the well

1 3
with W ϵ = 52 π 9 Rm a
depth and position relate to the original L–J (12,6) potential parameters.
c) An exponential repulsion v(A) = Wexp[−α. (R − R0)] giving

4πW α Z + d0
VA, S Z = 1+ exp − α Z + d0 −R0
a3 α 3 2
which is a shifted repulsion in Z with the same relative slope.

8.3.2 Surface Electric Dipole Layers

A continuum model can also be used to describe interactions of a molecule with
a surface electric dipole layer formed by the rearrangement of atomic positions
and electronic charge at the surface. The surface dipole appears even in clean
surfaces in a vacuum due to the spilling of electronic charge from the solid into
the vacuum, which creates a layer of surface dipoles oriented along the
334 8 Interaction of Molecules with Surfaces

z-direction pointing into the solid, as shown in Figure 8.2b. A dipolar layer also
appears when a layer of polar molecules is adsorbed at the solid surface. For a
collection of charges CI at distances dI from the surface, contained in a polar mol-
ecule, the interaction energy can be obtained by first locating each single positive
charge C in the vacuum at location (0, 0, d) above the surface plane, interacting
with a layer of surface electric dipoles of thickness l divided into cells with dipole
density DV per unit volume. The dipoles are assumed to be perpendicular to the
surface with the dipolar plane centered at −d0 = − l/2, as shown in Figure 8.2b.
The interaction energy per unit volume for the charge at distance R = [SD2 +
dD2]1/2 from a cell dipole located on the surface at distance SD from the z-axis,
with dD = d + d0 the distance from C to the plane of the dipole layer is given,
including the cosine dD/R of the angle between dipole and charge position, by

DV dD
uC , S R = − C
4πε0 R2 R

which can be integrated over a surface area of radius LD with elements of volume
l 2πSDdSD. The integral over 0 ≤ SD ≤ LD gives a total electrostatic energy

els CDV l dD CDV l dD

VC , S d = − 2π 1 − 1 2
≈ − 2π 1−
4πε0 LD + dD 2
2 4πε0 LD

for small dD/LD. It provides an attractive force for a positive charge and shows a
small decrease of the interaction energy as the distance d of the charge C to the
surface increases, for a fixed radius of the dipole layer. It also shows increasing
attraction at a given distance as the radius of the dipolar layer increases. This
result can be extended to a collection of charges CI at distances dI contained
in a polar molecule, adding over charges in the presence of a surface with a dipo-
lar layer. More generally, the energy can be obtained for any charge density dis-
tribution at the solid surface, solving the Poisson equation for the resulting
electrostatic potential [17].
The previous result can be applied as an example to the calculation of the elec-
trostatic interaction of a hydrogen atom with the planar surface dipole distribu-
tion. Locating the hydrogen proton with charge Cp = |ce| = + e at a distance
ZD = Z + d0 from the dipole layer, and the electron at position r e relative to
the proton, the interaction Hamiltonian of the two atomic charges and the sur-
face dipoles is
C p DV l
H H , S Z,ze = − 2π γ ZD − γ ZD + ze
4πε0
with γ(Z) = 1 − Z/(LD2 + Z2)1/2. This can be expanded up to second order in
powers of ze insofar ze/ZD 1 away from the surface, and the electrostatic
els
potential energy can be calculated from VH , S ZD = g H H , S g H to first order
 8.3 Continuum Models 335

in this coupling Hamiltonian for the H ground state g H using that g ze g H = 0,

with the result
els C p DV l ZD
VH , S Z = 2π g re 2 g H 3 2
, ZD = Z + d0
4πε0 LD + ZD 2
2

where furthermore g ze2 g H = g re2 g H/3 = a0, which gives a repulsive

force for the chosen dipole orientation, and is seen to go as a Z−2 at large dis-
tances ZD LD between the H atom and a dipolar surface layer with a finite
radius.

8.3.3 Adsorbate Monolayers

Two molecules adsorbed on a solid surface interact directly between them and
also indirectly through the surface. When the surface is that of a dielectric or a
semiconductor solid, and the molecular adsorption is weak and does not
strongly change the surface electronic density, it is possible to describe the inter-
action energy of the pair of adsorbed molecules in terms of molecule–molecule
and molecule–surface interactions using the electronic densities and properties
of the isolated molecules and of the clean solid. This is the case for adsorbed
closed-shell molecules and atoms, such as Ar atoms adsorbed on graphite or
N2 adsorbed on NaF(s). Adsorption of many molecules or atoms leads to for-
mation of a surface adsorbate lattice or of an adsorbate island. Depending on
whether the adsorbed species are bound to locations or moving around, the
adsorbate system can be treated as a lattice or as a fluid, by analogy with the
treatments presented in the chapter on model potentials, but done here for pla-
nar structures.
A simple treatment describes an adsorbed monolayer as a continuum with a
given density. A finite adsorbate lattice or an adsorbate island of surface area A,
with its electronic density concentrated between distances d and d + l along a z-
axis perpendicular to an infinitely large substrate surface, is attracted to the sub-
strate with an energy derived from the long-range molecule–surface interaction
energy. The molecule–surface dispersion energy changes as Z−3 as shown
above. Taking the limit of a continuous distribution of adsorbate molecules with
a number density ρads = Nads/(Al), the overall adsorbate interaction energy per
unit of surface area A is found integrating over Z the inverse power Z−m + 3 in the
model potential energy given above in 8.3.1, between limits d and d + l for an
adsorbate layer of thickness l, so that
3 d+l m− 3
2πW R0 R0
V ads
d A= − ρ ads dZ
m−3 m−2 a d Z
3 m −4 m−4
2πW R0 R0 d
=− ρ ads R0 1−
m − 4 m − 3 m −2 a d d+l
336 8 Interaction of Molecules with Surfaces

using the results in Section 8.3.1.Therefore the dispersion energy, where m = 6

for the atom–atom interaction, goes here as d−2 for large l, for a layer of A spe-
cies interacting with an infinite substrate surface made up of B atoms. To this
one must add a repulsive interaction energy function of d at short distances,
including a damping factor for the long-range term.
Returning to the description of an adsorbate layer as a collection of many
interacting adsorbed molecules, a lattice model of an adsorbate layer under-
going phase transformations at a temperature T (for instance, in a bath under
an external inert gas) can be described with a Hamiltonian containing a sum of
(single molecule)–surface interactions, plus molecule–molecule interactions,
and also adding a three-molecule term if the density of adsorbates is large.
The substrate surface can be divided into M two-dimensional cells labeled by
two indices in j = (j1, j2), with cell dimensions large enough to accommodate
one molecule. For Nads adsorbed molecules A at an average distance Zads from
the substrate and with small bonding energy εads = v(A)(Zads) to its lattice sites j,
among which the molecules can jump, the Hamiltonian can be written in terms
of lattice site occupation operators cj = 1, 0 respectively for occupied or empty
lattice site j, as

H ads = − εads c−
1≤j≤M j 1≤j<k ≤M
φ j,k cj ck

where φ(j, k) = v(jk)(Zads) is the (j,k) molecule–molecule interaction energy.

Thermal states of the adsorbate can be described in terms of the temperature
and its density ρads, to obtain the statistical coverage density
θads = M j = 1 cj T M with cj T an average over a thermal distribution [28]. This
treatment is mathematically analogous to the Ising model of up- and down-
spins in two dimensions, as can be displayed with the introduction of spin vari-
ables sj = ± 1 and the correspondence cj = 1 − sj 2. Conclusions from the Ising
model about phase transitions, for example, for the heat capacity of He on
graphite and for X-ray intensity versus light wavelength profiles of inert gases
on graphite, have been confirmed by experiments [28].
When the adsorbate molecules A are mobile, it is convenient to instead
describe the properties of a monolayer or of an adsorbate island as a liquid
of molecules bound by the attraction of the B species in the substrate. The mole-
cules A move at an average distance Zads from the substrate, and are found at
adsorbate locations S = Sx ,Sy . A treatment by analogy to what was done in
the previous chapter on interaction energies in liquids constructs the
thermodynamical internal energy Uads(ρads, β) with β = 1/(kBT) from A–A
pair interaction energies v(S; Zads), where S = (Sx2 + Sy2)1/2 in terms of the
two-dimensional pair-distribution function gads(S; ρads, β), as
∞
Nads Nads
Uads ρads , β = + ρ dS 2πS v S; Zads gads S; ρads ,β − εads Mθads ρads , β
β 2 ads 0
 8.4 Nonbonding Interactions at a Metal Surface 337

Here the first term contains the thermal kinetic energy in two dimensions,
and the last term accounts for the bonding energy of the adsorbate to the sub-
strate, with coverage density θads(ρads, β).

8.4 Nonbonding Interactions at a Metal Surface

8.4.1 Electronic Energies for Varying Molecule–Surface Distances
Metals such as Na, Al, Cu, or Ni contain localized electrons bound in cores of
lattice ions, and electrons delocalized over the lattice framework. At a metal sur-
face, such as at Na(011), Al(111), Cu(001), or Ni(001), delocalized electrons
spread beyond the surface ions further into the vacuum as compared with their
distribution inside the solid. This creates local net charge densities and forma-
tion of a surface dipole. Metal atoms at the surface also have unsaturated bonds
available for adsorption of other atoms or of molecules. A full treatment of
interactions of a molecule A near or at a metal surface M requires modeling
of the interaction energy dependence on the position R of the center of mass
of A relative to the surface and on the location of atoms in the molecule and
at the metal surface, VA, M R,QA ,QM . This can be done starting from a collec-
tion of all atomic ion cores and all delocalized electrons for the metal and also
for the atoms in the molecule.
Physisorption or chemisorption are special cases where the molecule is
adsorbed at the surface at a location Req around which it vibrates. The
potential energy surface (PES) at equilibrium is then VA, M Req , QA , QM =
ads
VAM QA ,QM a function of internal atomic positions, showing small binding
energies of the order of 20 kJ mol−1 (about 200 meV) for physisorption of A and
large binding energies, of the order 200 kJ mol−1 (about 2 eV), for chemisorption
of A or its fragments. Section 8.4 deals with nonbonding interactions and phy-
sisorption due to weak attraction forces. As examples, the physisorption ener-
gies at equilibrium (not including vibrational zero-point energy) for He on
Cu(001) is about 5.6 meV (0.5403 kJ mol−1), and for H2 on Cu(001), it is in the
range of 40–80 kJ mol−1 depending on the location of the adsorbate at the sur-
face. The physisorption energies depend also on the surface Miller indices and for
H2, they are smaller at Cu(111) and larger at Cu(110) compared to Cu(001).
AM A M
A self-consistent field model of Ne = Ne + Ne delocalized electrons in
eff
A + M introduces an effective potential energy vAM x derived by a mean-field
treatment of the electron motions, such as the Hartree–Fock or Kohn–Sham
treatments with an effective Hamiltonian F = m f m for independent electrons
m = 1 to Ne, with f m = hm + um a one-electron effective Hamiltonian containing
338 8 Interaction of Molecules with Surfaces

an averaged electron–electron interaction um . The delocalized electrons move

in the field of pseudopotentials for positive ions in a structure with a total charge
AM
ce Ne , that add to the electron kinetic energy in hm . To this one must also
res
add a residual mean-field correlation energy term U = m < n vmn − m um to
res
obtain the full Hamiltonian H AM = F AM + U AM for A + M. In some detail, elec-
tron spin orbitals are solutions of
eff
− ℏ2 2me ∇2 + vAM x ψ νσ x = ενσ ψ νσ x

and total energies are obtained adding all one-electron energies for spin orbitals
with occupation number nνσ and subtracting an electron–electron residual cor-
AM
relation energy Eres after accounting for electronic correlation, to avoid dou-
ble counting of correlation energy, so that the total energy is [29, 30]
AM
Etot = n ε −EresAM
νσ νσ νσ

AM A M
from which one can obtain VAM R = Etot R −Etot R − Etot with subtrac-
A M
tion of similar expressions for Etot and Etot . This is also a function of all the
internal coordinates of atoms in A and M.
Simple models for molecules near metal surfaces help to extract physical
insight into the many aspects of surface interactions. A one-electron treatment
relies on a piecewise potential energy along the z-direction and expands the
expressions introduced in Section 8.1 for a pure metal M to include the poten-
tial energy at the location of an adsorbed or nearby molecule A, and the inter-
action of each electron with the atomic ions in A and M in the field of all the
other Ne − 1 electrons screening the ion charges. The electronic structure of
A + M can be obtained from a one-electron jellium model of the metal with
a surface perpendicular to the z-axis at location Z = 0, with an electronic poten-
tial energy vM(z) of constant value vin < 0 inside a semi-infinite region with
− ∞ ≤ z ≤ 0, and value vout = 0 for z ≥ 0 in a vacuum, plus an attractive molecular
potential vA r for an isolated A at location Z so that now orbitals of electrons
staying mostly within − ∞ ≤ z ≤ 0 spill out to overlap the orbitals of A. One must
int
add an interaction potential energy vAM r,Z to account for electron delocal-
ization between A and M.
As A approaches M to a distance Z, each screened core ionic charge Ca in A
creates an image charge Ca, img = − Ca (εr − 1)/(εr + 1) in M, and each electron in
A interacts with all the ion charges and their images, and also with its own image
charge. This creates a long-range electronic potential energy operator
a, img
vAM r,Z,QA for each charge a, which add up to give an electrostatic poten-
img
tial energy vAM r,Z, QA between the electron and all other charges. By analogy
 8.4 Nonbonding Interactions at a Metal Surface 339

with the treatment of a hydrogen atom near a surface in Section 8.2.1, it is found
that the leading interaction term at large distances between a neutral molecule
A and the surface comes from the electron at location r A = r − Zuz interacting
with all charges, so that

img Cel2 εr − 1 2
vAM r,Z = − r A + zA 2
4πϵ0 16Z 3 εr + 1
AM
plus higher order terms in 1/Z, with Cel = eNe =
a Ca . Further addition of a
nr
near-range repulsive electronic potential energy vAM gives an one-electron
int img nr
interaction potential energy vAM r,Z = vAM r,Z + vAM r,Z , and a total
one-electron potential energy operator
eff int
vAM r,Z = vM z + vA r,Z + vAM r,Z
which also depends on internal positions QA.
A one-electron description of the interaction valid for all distances can be based
eff
on orbitals for electrons moving in the vAM r,Z potential energy, con-
structed from linear combinations of molecular and metal orbitals, φμ r and
φk r with energies εμ and εk , respectively, and normalizations φμ φμ = δμμ
and φk φ = δ k − k , as
k

φλ r; Z = c
μ μλ
Z φμ r + d 3 k ckλ Z φk r

where λ is a continuous orbital label, and the dependence on QA has been omit-
ted. These orbitals are obtained from solutions of the one-electron Schrodinger
equation for fixed atomic positions, with a normalization φλ φλ = δ(λ − λ )
suitable for delocalized orbitals. The one-electron energies go asymptotically
int
as Z−3, as follows from the state averages φλ vAM Z φλ of the interaction
potential. Total energies are given by the sum of all one-electron energies of
occupied orbitals corrected by subtracting the residual electronic correlation
AM
energy Eres Z so that

E AM
Z = dε ε nAM ε,Z gAM ε,Z − EresAM Z

with nAM(ε, Z) and gAM(ε, Z) the population and density of levels.

The coupling of localized molecular orbitals with delocalized metal orbitals
leads to broadening and shifting of the molecular energy levels with values
dependent on the distance Z, and requires special treatment. The density of
electronic states per unit energy gAM(ε; Z) = dλ δ[ε − ελ(Z)] changes with
the distance Z. To simplify, we consider only the highest occupied and lowest
340 8 Interaction of Molecules with Surfaces

unoccupied molecular orbitals and metal orbitals filled up to the Fermi level. At
large distances with no interaction between A and M, it is given for each electron
spin state by

gAM ε = d 3 kδ ε− εk + δ ε −εHO + δ ε− εLU

k ≤ kF

where the first term is gM(ε) with kF the Fermi wavenumber, and the last two
terms give gA(ε). At a shorter distance Z, the interaction shifts and broadens
εHO by amounts ΔεHO(Z) and γHO(Z), and εLU by amounts ΔεLU(Z) and γHO(Z).
The density of states is then of the form
γHO γLU
gAM ε; Z = d 3 kδ ε −εk + 2 + 2
k ≤ kF π ε − εHO + γ2HO π ε − εLU + γ2LU

with a new Fermi wavenumber kF Z = kF + ΔkF , εHO = εHO + ΔεHO , and

εLU = εLU + ΔεLU and level widths γHO and γLU. In cases of physisorption, where
the interaction of molecule and surface is weak, the level shifts and widths can
be obtained from perturbation theory and are found to be of second order in the
int
coupling vAM .
The perturbed density of states can be used to account for the population
NAM(T; Z) of states at a given temperature T introducing the Fermi–Dirac
energy distribution nFD(ε, T), with NAM(T; Z) = dε gAM(ε; Z)nFD(ε, T) [15].
Corresponding combinations of orbitals can be bonding between A and M,
and delocalized, or antibonding and localized near the surface. Energies of
interaction VAM(Z, QA) follow from independent electron models for A + M,
AM A M
A, and M under the assumption that Eres Eres + Eres for the weak interac-
tions typical of physisorption, as

VAM Z, QA = dεε nAM ε; Z, QA gAM ε; Z,QA − nA ε gA ε −nM ε gM ε

where nAM(ε; Z, QA) is the electronic energy level population in AM. It shows
that the interaction energy dependence on (Z, QA) comes from the density of
one-electron energies of A + M and also from the broadening and shift of molec-
ular levels.
This simple treatment provides some insight on the dependence of interac-
tion energies as the distance Z between A and M is varied. In particular, insofar
vAM r,Z ≈ − Q r Z −3 , the sum of one-electron energies gives an attractive
img

potential energy changing as 1/Z3, and inasmuch level shifts and broadenings
int
involve the square of the coupling matrix elements φμ vAM Z φk , it is
found that at large distances, Δελ and γλ contribute a repulsive energy changing
as 1/Z6.
 8.4 Nonbonding Interactions at a Metal Surface 341

A more detailed treatment can be done introducing a Hartree–Fock or Kohn–

Sham effective Hamiltonian operator F and its orbitals in an approximation
adding electron correlation effects, to obtain energy shifts and widths from
energy and overlap matrix elements [31]. Such treatment provides insight on
trends of adsorption energies for several adsorbate atoms on a given metal sur-
face, and has been extended to account for the atomic structure of the substrate
and adsorbate site [8], in which case VAM is obtained as a function of all atomic
positions in R, QA ,QM . It gives density of states and energies that can be ana-
lyzed in weak A–M coupling and also in strong coupling cases, and provides
variations of physisorption energies for different transition metal surfaces.
Results show the right trends but are usually inaccurate for values of adsorption
energies.
Details of physisorption energies and of interaction potential energies versus
atomic structure, and accurate parametrizations, require many-electron treat-
ments such as those provided by many-atom cluster models or by generalized
density functional treatments. A variety of theoretical treatments of energies of
physisorption have been extensively compared to experimental results [32]. We
show in what follows some of the related concepts and methods.

8.4.2 Potential Energy Functions and Physisorption Energies

In a treatment of molecule–metal dispersion energies using polarizabilities,
valid for light of wavelengths large compared to surface interatomic structures,
we set the wavenumbers introduced in Section 8.2.1 to q = 0, so we can write for
the metal permittivity εr(0, ω) = εM(ω), and for the dispersion interaction energy
between A and M at the relative distance Z between A and the surface plane, the
potential energy

∞ dsp
1 ℏ ϵM iω − 1 C
= − A,3M
dsp
VA, M Z =− dω αA iω
4πε0 4πZ 3 0 ϵM iω + 1 Z
dsp
with CA, M a dispersion coefficient dependent on internal coordinates of mol-
ecule and solid metal. As done for the interaction of molecules with dielectric
solids, we can introduce the function αA(iω) = β(A)(ω), and also define a new
function ϵM(iω) = ηM(ω). Similarly for the solid metal, we have ω2ηM(ω) ≈ ω2
+ ωp2 for ω ∞, ηM(ω) = 1 + ωp, ion2/(ωT2 + ω2) for ω < ωT, and the static value
ηM(0) = ϵM(0) as a parameter. A suggested interpolation formula for a metal with
a single-atom composition is

ω2p, ion ωp 2
ηM ω = 1 + 1 − gd ω + 1 − gd ω
ωT 2 + ω2 ω2
342 8 Interaction of Molecules with Surfaces

with gd(ω) a damping function equal to one at large frequencies and decaying to
zero over a transition region ωT < ω < ωp. This can be used in the approximate
dispersion coefficient

∞
dsp 1 A ℏ 1 ηM ω − 1
CA, M = α0 0 dω
4πε0 4π 0 1+ ω ω A
2 ηM ω + 1

which can be calculated from the static polarizability and the known number
of valence electrons of A contained in ω A , and from the static dielectric con-
stant and plasmon frequency of the solid. This shows that the solid dielectric
function contributes to the integral mostly when ω < 2ω A , and should give rea-
sonable results for noble gases on metal surfaces provided transitions Λ = (λ λ)
there, as introduced in Section 8.1.3, have small oscillator strengths fΛ where
ω2 ωΛ2 [18].
The effect of variations of the surface electronic charge distribution
ρe r ; R ,QA ,QM , for fixed atomic positions, on the dispersion energy can be
approximately calculated introducing a plasmon frequency dependent on the
2
location r across the surface by means of ωp r = ρe r ce 2 ϵ0 me and integrat-
dsp
ing over r the functional VA, S R; ρe r to average it across the metal surface.
This can provide trends for the interaction energy dependence on the distances
over which the metal electron density ρe decays outside the metal surface. An
alternative is to introduce the dispersion dependence of ϵr on the wavevector q
of light, derived from a Boltzmann-equation treatment of the electronic fluid
and parametrized as ϵr , el q,ω = 1 −ωp 2 ω ω− iΓ − αq 2 [33], to be com-
bined with a parametrized form for the polarizability of the molecule to obtain
dsp
an energy density vA, S R, q , and to integrate this over a distribution of q
values.
More generally, the energy of physisorption of a molecule A at all distances Z
from the surface shows a distance dependence with a minimum at a well posi-
tion Zm, which results from combined long-range electrostatic, induction, and
dispersion attraction and near-range electrostatic, exchange, and correlation
dsp
forces giving repulsion. Long-range dispersion energies VA, M R,QA ,QM are
obtained from polarizabilities of A and M and go as Z−3. Near-range potential
nr
energies VA, M R,QA ,QM , containing also electrostatic and induction long-
range forces for ground electronic states, must be obtained from the combined
molecule–surface electronic charge distributions where they overlap. They are
repulsive at short range due to the Pauli exclusion effect between electrons at
the molecule and surface, imposed by the antisymmetry of joint electronic
 8.4 Nonbonding Interactions at a Metal Surface 343

wavefunctions, and typically decay exponentially at short distances, while elec-

trostatic and induction components go as powers of Z−1 away from the surface.
Near-range (nr) potential energies can be obtained in a variety of approxima-
tions, many of them based on DFT. Constructing a model of the adsorbate A and
substrate M and choosing a density functional E ρ r , the interaction energy at
position R of the molecule relative to the surface is obtained from the difference
between energies of the complex and its components as
nr
VA, M R = EA, M R; ρA, M r −EA R; ρA r − EM ρM r

where care must be taken in calculations to subtract large numbers in an accu-

rate way, by using the same computational treatment for all terms, and by care-
fully subtracting energies calculated at the same interatomic distances. The full
interaction potential energy is
nr dsp
VA, M R = VA, M R + fd R VA, M R

with fd a damping function in the Z variable and showing surface lattice peri-
odicity in the (X, Y) variables. The two terms also depend on the internal atomic
coordinates (QA, QM).
Results of near-range DFT calculations containing electrostatic, exchange,
correlation, and polarization terms can be fitted to parametrized forms, and
in the simplest case have been fitted in the literature to functions decreasing
exponentially with Z. However, DFT can provide some of the attraction forces
through polarization and a more general potential function fit should include a
potential well. Combined near-range and long-range potential energies can be
parametrized for A at position R outside the solid S, with functions such as a
near-range Morse potential with a minimum coming from the induction attrac-
tion, as

VA, M R = W X,Y exp − α X,Y Z − Zm exp − α X,Y Z −Zm − 2 0

dsp −3
−fd Z CA, M X,Y Z −Z0
where the parameters W, α, and CAM are shown to depend on the location of
species A along the surface, to account for changes of its interaction energies
along atoms in the surface lattice, and fd(Z) is a damping function at short dis-
tances. The parameters are periodic functions of lattice cell distances a and b for
crystalline surfaces. This dependence can be averaged integrating over the sur-
face positions when the distance Z is much larger than interatomic distances at
the surface.
Insofar physisorption does not involve electronic rearrangement at the sur-
nr
face, insight can be extracted about VA, M R even when using in functionals
344 8 Interaction of Molecules with Surfaces

a simple additive form of total electron density as ρA, M r; R = ρA r; R + ρM r ,

with mutually consistent functionals for components and total system invol-
ving kinetic energy, exchange, and correlation, and with an energy subtraction
procedure similar to what has been used for nonbonded molecule–molecule
interactions [34].
An alternative based on physical considerations corrects for density overlaps
and introduces instead an effective medium treatment where the physisorption
is modeled calculating first the energy of the adsorbate immersed in a medium
of electrons with a density ρM of the substrate metal, obtained averaging over
angles around the adsorbate. This is followed by a perturbation treatment to
account for the energy resulting from the distortion charge density
ΔρAM r = ρAM r − ρM , due to the addition of the adsorbate, in the presence
of the electric potential change Δϕ r due to the substrate density change
ρM r −ρM [22, 35]. Features of physisorption binding and of interaction poten-
tial energy of a closed-shell species can be described in many cases introducing
such a model, which has also provided insight on chemisorption of atomic spe-
cies such as H and O on metals.
Early results for noble gas atoms on metals were obtained from DFT varia-
nr
tional calculations of VA, M R with the jellium model of the metals [9]. These
treatments appear to provide correct trends as structures change with different
adsorbates at a given metal surface, or with different substrate surfaces for a
given adsorbate.
Calculations of the electronic density for physisorption systems, for example,
Ar adsorbed on jellium with rs = 3 au (near the value for the density of Ag), show
that some electronic charge moves from the surface toward the atom at a dis-
tance d from the surface, to provide an attraction force and an interaction
energy minimum for varying distance d [36]. Results are shown in
Figure 8.3a for the potential energy curve versus distance, and in Figure 8.3b
as isocontours of electronic density changes on a plane perpendicular to the sur-
face containing the atom, with solid lines for positive changes and dashed lines
for negative changes. Changes are more pronounced at shorter distances, and
the increase of electronic density is noticeable at equilibrium and in repulsive
regions. This leads to long-range attraction due to induced polarization, and
to repulsion at short distances due to the Pauli electronic exclusion.
Understanding of more subtle aspects, such as the change of the interaction
energy of the adsorbate with the lattice Miller indices (h, k, l) of a given metal
surface, or of the adsorbate location of lowest energy (on top of a surface atom
or at a bridge or hollow), must be done introducing the atomic structure of the
whole system, with more accurate treatments of the electronic distribution and
long-range interaction energies. One such accurate treatment is based on sym-
metry-adapted perturbation theory (SAPT) as described in Chapter 6. This is
 8.4 Nonbonding Interactions at a Metal Surface 345

(a)

–0.09 Ar

Atomic binding energy (eV)

(rs = 3)
–0.06

–0.03

0.03

0.06

3 4 5 6 7 8
d (Bohr)
(b)
Ar Ar Ar

+ + +

d = 3 bohr d = 4.25 bohr d = 6.5 bohr

(repulsive) (equilibrium) (attractive)

Figure 8.3 (a) Potential energy versus distance for Ar interacting with the surface of a metal
with electronic density corresponding to a spread radius rs = 3 au (near the value for the
density of Ag). (b) Electronic density changes for three distances, with full and dashed lines
for positive and negative changes, respectively. Source: from Ref. [36]. Reproduced with
permission of American Physical Society.

very demanding of computational times, but has nevertheless been applied to

several physisorption cases such as a noble gas atom interacting with a surface
of TiO2(110) [37] and to several noble gas atoms interacting with a coronene,
graphene, or graphite surface [38] modeled both by a cluster embedded in a
medium and by an adsorbate on a surface with periodic boundary conditions.
These calculations provide accurate standards to compare with a variety of
approximations based on density functional methods and including dispersion
energies in different models.
Early treatments using the jellium model and the effective medium approx-
imation have been superseded by detailed DFT treatments with a variety of
semilocal and hybrid density functionals. Work was done, for example, using
346 8 Interaction of Molecules with Surfaces

semilocal density functionals for systems such as Na adsorbed on the Al(001)

and Al(111) surfaces, and Xe adsorbed on Cu(111) and other metal surfaces,
to obtain binding energies and bond distances at equilibrium conformations,
generally showing correct trends as compared with experimental results [24,
39, 40]. Results using a generalized gradient density functional show that noble
gas atoms have their largest physisorption binding energy at locations on top of
metal atoms instead of locating at bridges or hollows, because the repulsion
energies at the top positions are relatively smaller while the long-range attrac-
tion energies are very similar over the surface [40]. This somewhat surprising
conclusion has been found to be in agreement with reinterpreted experimental
results, and has also been reached in more recent theoretical work with density
functionals including van der Waals forces [41, 42]. The introduction of disper-
sion energies allows more accurate calculation of binding energies, which vary
between different site locations, and give more accurate bond distances between
adsorbed atoms and surface atoms.
Atomistic models combined with DFT treatments are useful for calculations
of structures and binding energies at surfaces, provided the long-range van der
Waals interactions are added. Addition of the dispersion energy changes bind-
ing energies and bond distances and can be done with recent treatments of the
type vdW-DFT [41, 43, 44] and DFT-D [45, 46]. The DFT-D treatments start
with a calculation of energies using a selected semilocal functional, and these
are then corrected by addition of dispersion energies that multiply damping
functions as described in the Chapter 7. This appears to improve agreement
with experimental values of properties at equilibrium, and gives potential
energy surfaces, which can be used to treat the molecular dynamics of atomic
scattering at the surfaces. Earlier treatments have been replaced in recent years
by more accurate treatments with semilocal and nonlocal density functionals,
and hybrid functionals of Kohn–Sham electronic orbitals, which provide short-
and intermediate-range interaction energies [7, 24]. They can be combined with
long-range van der Waals functions to cover all distances, such as done with
the DFT-D3 treatment [45] and the DFT-MBD treatment [46].
The alternative vdW-DFT approach starts from the formally exact expression
of the exchange-correlation energy in terms of the two-electron dynamical sus-
ceptibility, obtained from DFT and implemented so that it correctly reproduces
the long-range dispersion energy [44]. As explained in Chapters 6 and 7, the
adiabatic connection fluctuation–dissipation (ACFD) relation gives a formally
exact expression for the electron–electron correlation energy in terms of the
dynamical susceptibility, which can be decomposed into semilocal (sl) and non-
local (nl) terms. For species A at a relative position R from a surface plane, the
c slc nlc
correlation energy is EA, M R,Q = EA, M R, Q + EA, M R, Q with the first term
coming from a semilocal DFT correlation functional and the second term given
by an ACFD nonlocal correlation functional. This second term can be
 8.4 Nonbonding Interactions at a Metal Surface 347

constructed for large distances between A and M from their susceptibilities

given as known functionals of the electronic density, so that no parametrization
from molecular properties is needed [47–49]. This approach has been exten-
sively applied to the interaction of the noble gases with metal surfaces, such
as Xe on Cu(111) and Cu(110), and to organic molecules with metal surfaces,
such as butane (C4H10) adsorbed on Cu(111) and benzene on graphite [43, 44].
Accurate values for short-range and intermediate-range energies can be
obtained from the coupled-cluster method or from symmetry-adapted pertur-
bation theory as described in the chapter on many-electron treatments, with
models using clusters or periodic boundary conditions for the substrate. Insight
can also be extracted from extended atomic cluster models, which include the
adjacent substrate atoms near the adsorbate location, with atoms added to sat-
urate the peripheral (dangling) cluster bonds. These models have been recently
extended in embedding treatments as briefly presented in the next subsection.

8.4.3 Embedding Models for Physisorption

The embedding treatment described in Chapter 7 and Section 8.1.1 is suitable
for studies of physisorption, choosing the atomic group A’ to be a cluster (or cls)
including the adsorbate species A and neighboring atoms in the substrate, and
choosing B = M’ to be the indented substrate or environment (env), usually
modeled as a slab sufficiently thick to avoid confinement effects on the physi-
sorption properties. The total electron density is constructed so that ρtot = ρcls
+ ρenv and the total (ionic plus electronic) energy decomposes into cluster, envi-
ronment, and interaction terms, as [50] Etot = Ecls[ρcls] + Eenv[ρenv] + Eint[ρcls,
ρenv ]. The functionals and densities depend on the atomic positions in
R,QA ,QM , and in the case of physisorption, the relative position is chosen
to be around an equilibrium value Req while the remaining positions are varied,
ads
so that Etot QA ,QM = Ecls Req ,QA ,QM gives the adsorption energy as the
molecular structure is varied. The embedding procedure can be implemented
for physisorption so that the number of electrons Ncls and Nenv are fixed and
do not change if the adsorbate conformations are varied, while the indented
substrate can be constructed so that it has the same atomic positions as before
adsorption.
Insofar the environment does not change its atomic structure or electronic
charge distribution, it is possible to obtain the total energy using the subtraction
procedure where Etot is described within a density functional treatment for the
DFT
whole system, giving Etot . Some electronic rearrangement, however, occurs
near the adsorption location, and it can be accounted for by accurate calculation
of the energy of the embedded cluster (or emb cls) fragment, for example with
a correlated many-electron wavefunction (or CW) Φcls, by adding a term
CW DFT DFT
ΔEemb cls = Eemb cls −Eemb cls to Etot . This is similar to the previously mentioned
348 8 Interaction of Molecules with Surfaces

subtraction scheme in the QM/MM approach. The assumption to begin with,

that the environment is not changed by atomic displacements in the cluster, can
be tested by enlarging the cluster size and verifying that the environment is
unchanged.
The functional for the interaction energy varies with the densities of cluster
cls
and environment, providing embedding potential energy functions Vemb r and
env
Vemb r through the functional derivatives in
δEint cls δEint env
= Vemb r , = Vemb r
δρcls r δρenv r
Equilibration between the cluster and substrate requires that the two embed-
cls env
ding potentials should be equal and gives the constraint Vemb r = Vemb r =
Vemb r , which is a unique embedding potential for the total system, and is illus-
trated in Figure 8.4, adapted from Ref. [51].
Choosing the cluster to contain all the atoms likely to move due to interac-
tions between adsorbate and surface, the internal atomic coordinates of the
environment can be fixed at their equilibrium values, as Qenv = Qenv eq
, while
the atomic positions of the cluster change with the position R of the center
of mass of the adsorbate with respect to the surface, with variables R,Qcls .
DFT
The potential energy function of the variable positions is Etot R,Qcls and
can be written as a functional of the densities ρcls r and ρenv r , and also of
the embedding potential Vemb r in
DFT DFT DFT
Etot = Ecls ρcls , Vemb + Eenv
DFT
ρenv, Vemb + Eint ρcls ,ρenv ,Vemb

The computational procedure starts with a guess for ρcls, ρenv, and Vemb r .
The fragment functionals EK ρK r ,Vemb r ,K = cls,env, are constructed for
fixed atom positions. The energy is minimized to find the densities ρK r for

Vemb
+
Vemb

Figure 8.4 Embedding components for physisorption of an adsorbate A on a substrate (left),

decomposed into its environment (center) and a cluster (right). The embedding potential
Vemb constrains surface atoms to their positions, near equilibrium values for physisorption.
Source: adapted from Ref. [51]. Reproduced with permission of American Physical Society.
 8.5 Chemisorption 349

the given embedding potential, and this is updated by calculating δEtot δVemb r
to search for a minimum for fixed ρK r [52]. An iteration of this sequence
DFT
leads to the final potential energy surface Etot for the total system. This must
CW DFT DFT
be corrected with the addition of ΔEemb cls = Eemb cls − Eemb cls , where Eemb cls differs
DFT
from Ecls insofar the first term accounts for the boundary effect of Vemb r
present in the Hamiltonian for the cluster. The equations for the combined
DFT and CW treatment including the embedding potential are
DFT
δ Ecls
+ Vemb r = μemb cls
δρcls r
Ncls CW
H cls + j=1
Vemb r j Φcls = Eemb cls Φcls

with H cls the many-electron Hamiltonian (excluding the embedding potential)

DFT
of the cluster for a chosen structure. Similarly, Ecls is the energy functional for
the cluster excluding the embedding potential, and solving the first equation
provides the optimized density ρemb cls r to be used in

DFT DFT cls

Eemb cls = Ecls ρemb cls + d 3 r Vemb r ρemb cls r

The embedding treatment with the subtraction scheme has been applied to
physisorption of atoms and molecules.
For example, embedding has been used to obtain the conformation and vibra-
tional spectra of CO adsorbed on a Cu surface. Binding sites and energies and
vibrational spectra of this adsorbate have been obtained for the Cu(001) [53]
and Cu(111) surfaces. In the case of CO adsorbed on Cu(111), cluster and slab
models left doubts as to whether the CO would have larger binding energy on
top of a Cu atom or in a hollow of the surface, with different conclusions
depending on the quality of the DFT functional. It was reconsidered with the
embedded cluster treatment [54], using DFT for the environment and a multi-
configuration wavefunction with single and double excitations for the cluster,
which lead to the conclusion that CO is perpendicular to the surface and sits
on top of a Cu(111) atom, as known from experiments.

8.5 Chemisorption
8.5.1 Models of Chemisorption
Chemisorption is the result of extensive electronic rearrangement when a mol-
ecule or atom interacts with a surface, leading to electron transfer or to bond
breaking and formation [4, 8, 22, 28, 55]. Treatments of chemisorption energies
350 8 Interaction of Molecules with Surfaces

depend on whether the solid surface refers to a dielectric surface S with localized
electrons, to a semiconductor surface, or to a metal (using then S = M) with
delocalized electrons.
Chemisorption at a dielectric or semiconductor surface can be accurately
described with a cluster model, which includes the adsorbate and nearby atoms
in the substrate, with boundary bonds atomically saturated. The whole many-
atom system can be treated as a supermolecule with many-electron wavefunc-
tions including electron correlation as described in Chapter 6, to allow for both
short-range and long-range interaction energies. A cluster model can also be
treated within DFT-D formulations [45, 46], allowing for dispersion energies
as done for two interacting many-atom systems in Chapter 7. This is less
demanding of computing times and can be applied to quite large clusters.
The DFT-D treatments can also be applied to models of crystalline solid sur-
faces such as slabs with extended periodic lattices. This avoids problems with
structural and bonding properties dependent on the size of the cluster models
that may converge only for very large cluster sizes.
Chemisorption on metal surfaces brings in new aspects due to the delocali-
zation and high susceptibility of electrons in the metal, and must be treated dif-
ferently. This requires models with extended lattice structures for the metal
surfaces, with their electronic properties usually described within DFT. It can
be done including dispersion energies with a version of DFT-D that, however,
must go beyond sums over atom-pair interactions to account for many-atom
effects. Some of these extensions have been covered in Chapter 7 and in reviews
[45, 46].
An alternative to the DFT-D approaches has been provided by the vdW-DF
treatment [43, 44, 49] derived from the adiabatic connection theorem described
in Chapters 6 and 7, which also includes dispersion energies in a seamless way
between short and long ranges. Its introduction of a nonlocal expression for the
electronic correlation energy has been implemented for both localized and delo-
calized electron distributions and therefore is suitable for calculations of chem-
isorption energies for atoms and molecules on dielectric, semiconductor, and
metal surfaces.
Reviews of published calculations have compared results for chemisorption
with a variety of DFT functional also adding dispersion interaction energies,
and the general conclusion is that dispersion contributions are important
and needed to obtain correct chemisorptions binding energies and structural
bond distances [44–46]. Some examples are mentioned in what follows.
Physical insight can be obtained from one-electron treatments that account
for strong coupling of orbitals in the molecule and metal and that incorporate
electronic correlation through density functionals [8, 31]. The coupling leads to
shifts of molecular levels and to their broadening, depending on the relations
between the ionization and affinity energies of the molecule, and the work func-
tion and energy band shapes of the metal. This section considers features of
 8.5 Chemisorption 351

chemisorption common to all solid surfaces, whether dielectric, semiconduct-

ing, or metallic, which can be described in terms of energy functionals. The
following sections deal with chemisorption on metal surfaces, due to electronic
charge transfer or to bond breaking and formation.
Electronic charge transfer can happen when an open-shell atomic or molec-
ular adsorbate A interacts with surface atoms, whether the surface is a dielectric
or metal, as electronic charge of magnitude δce is transferred between the
adsorbate and the atoms of the surface lattice S, going from A + S to A± + S ,
with the direction of charge transfer determined by relative values of the ioni-
zation and affinity energies of A compared to the electronic work function of S.
The charge transfer can be described with different models depending on
whether the surface contains localized electrons or delocalized ones. Using
energy functionals, the interaction potential energy is
el −tr A± S
VA, S R, QA ,QS = E R,QA ,QS ; ρA ± S r −E A
QA ; ρA r
−E S
QS ; ρS r,QS

where R,QA , QS are the relative position of A with respect to S, the internal
atomic positions of A, and those of S. Electronic charge transfer is likely to affect
the atomic structure of the solid surface near the adsorbate.
As described in Section 8.1, these adsorbates can be treated calculating the
±
functional E A S as a sum over energies of the charged fragments with internal
atomic positions QA ± and QS and with densities ρA ± and ρS , plus their inter-
action energy in an embedding treatment, letting

A± S A±
E R,QA , QS ; ρA ± S r =E R,QA ± ; ρA ± r +E S
R,QS ; ρS r
±
A S
+ Eint R,QA ± ,QS ; ρA ± S r

with ρA ± S = ρA ± + ρS by construction. Calculations can proceed as previously

described for physisorption, with a chosen indented surface and cluster, and
their atoms held in place by an embedding potential, but allowing here for
new locations of atomic cores in the ionic fragments resulting from displace-
ments in the original A and S structures. Unlike the case of physisorption,
the indented surface atomic structure must be allowed to change from the
el − tr
original structure, and the total potential energy function VA, S R,QA ,QS
A± A±
can be obtained from the charge transfer energies ΔE =E −E A and
ΔE S = E S −E S plus the interaction energy of the fragments.
An example is provided by the chemisorption of Ag2 on the TiO2(011) sur-
face. The diatomic can be located on top of a Ti atom, or at a bridge oxygen, or in
a hollow of the surface. Calculations of interaction potential energies have been
352 8 Interaction of Molecules with Surfaces

200

Energy (meV)
Z
–200 DLPNO-CCSD(T) Ag2/Ti9O25H14
PBE-D3(BJ) Ag2/Ti9O25H14
PBE Ag2/Ti9O25H14
–400
PBE-D3(BJ) Ag2/TiO2(110) periodic

–600
2 4 6 8 10 12 14 16 18 20
Z (Å)

Figure 8.5 (a) To the left, cluster model Ag2/Ti9O25H14 of the perpendicular adsorbate and
nearby surface atoms, with boundary bonds saturated with hydrogen atoms. (b) To the right,
potential energy versus distance Z from a Ag atom to a nearby Ti surface atom, for the cluster
model calculated from CCSD, DFT and DFT-D, and also for a periodic slab with four layers, from
DFT-D. Source: from Ref. [56]. Reproduced with permission of Royal Society of Chemistry.

done for a cluster model Ag2/Ti9O25H14 [56] as shown in Figure 8.5a for the
adsorbate in a perpendicular orientation including nearby surface atoms (with
boundary bonds saturated with hydrogen atoms) with accurate many-electron
CCSD(T) wavefunctions, and also with a DFT-D treatment including long-
range dispersion energies for the same cluster. The addition of the dispersion
energy has been done as explained in Chapter 7, with a damping function.
The DFT-D treatment has also been implemented within a model of Ag2
adsorbed on a slab of the solid, with a periodic lattice. Results for potential ener-
gies are shown in Figure 8.5b for the on-top Ti location. They indicate excellent
agreement for the binding energies between CCSD(T) and PBE-D3 results, but
large errors if the dispersion energy correction is ignored.
Chemisorption also happens for closed-shell molecules such as H2, CO, N2,
H2O, CO2, and C6H6 on solid surfaces, when electronic rearrangement near the
surfaces weakens the molecular bonds of a species AB and results in dissociation
into fragments A and B. These can both bond to surface atoms to form adsor-
bates AS and BS, or one or both fragments can go free away from the surface,
depending on the bond strength or repulsion force between each fragment and
the surface atoms.
The dissociation of a molecule AB as it approaches a surface S and breaks into
AS + BS occurs on a potential energy function E ABS
R,QA , QB ,QS ; ρABS r ,
with the relative position of the center of mass of AB near the surface reaching
a value Rads where the dissociation of AB occurs. The fragments then move on a
 8.5 Chemisorption 353

diss
PES VAB, S Rads ,QA ,QB ,QS = VAB, S QA ,QB ,QS , a dissociation potential
energy function obtained from
diss
VAB, S Q = E ABS
Rads ,Q; ρABS r −E AB
QA , QB ; ρAB r − E S
QS ; ρS r
with Q = (QA, QB, QS). The functional E(ABS) can be obtained for different atomic
arrangements along the dissociation path from embedding treatments, with
clusters A , B , and AB including, respectively, A, B, and AB and also neighbor
atoms at the surface, and for a surface S with an indentation large enough so
that all three clusters can be accommodated in the surface hole. At the short
distance Rads between the adsorbate AB and the surface, the energy isocontours
of E(ABS) show a minimum for the equilibrium structure of AB , and a dissoci-
ation path leading to A + B .
The energy function E ABS
R,QA , QB , QS describes more generally dissoci-
ation for varying approach distances and orientation of AB with respect to the
surface, with energy isocontours showing a valley as AB approaches the surface
along a variable distance ZAB − S and as AB breaks into A + B , which separate
along RAS − BS with an activation energy along a path into an energy valley for AS
ads
+ BS. The potential energy function VAB, S ZAB− S ,RAS −BS can be fit to a Lon-
don–Eyring–Polanyi–Sato (or LEPS) function of the two position variables
for a surface rearrangement, as derived from a valence-bond description of bond
breaking and forming and described in Chapter 4. However, one must keep in
mind that here pair bonding and repulsion-potential functions are affected by
the presence of the remaining surface atoms.
An example of molecular dissociative adsorption on a dielectric surface is
found in the calculation of the adsorption fragmentation of a benzene C6H6
molecule on a Si(001)-(2 × 1) surface obtained within the vdW-DF approach,
which shows two possible fragmentations. The dispersion energy contributions
appear to be relevant to clarification of which dissociation structures are the
most stable [43].
The rearrangement reaction A + BC AB + C on the surface S, for all species
already adsorbed on the surface, involves the reaction potential energy function
rct ABC , S A + BC , S
VABC , S Q = E Q; ρABC , S r −E Q; ρA + BC , S r
with Q = (QA, QB, QC, QS) the collection of all atomic positions in a reference
frame attached to the surface. The energies to the right must be obtained as
functions of the positions of A, B, and C on the surface S. In a treatment
using embedding in an indented surface S , the energies of fragments A , B ,
and C containing the species A, B, and C and neighbor surface atoms can
be obtained along reaction paths leading from A + BC to AB + C . Surface
atoms in the fragments must be allowed to rearrange too, but the atomic
354 8 Interaction of Molecules with Surfaces

structure of the indented surface S can be kept fixed if the fragments and
indentation are large enough.

8.5.2 Charge Transfer at a Metal Surface

Electron transfer between a molecule and the surface of an insulator or semi-
conductor involves localized surface spin orbitals ψ s and spin orbitals ψ μ of the
molecule to form localized ionic bonds at the surface, similarly to what is found
in intermolecular bonding. The interaction with delocalized electrons at a metal
surface is fundamentally different in that the metal provides a large number of
orbitals to be combined with adsorbate orbitals. This involves the coupling of
molecular spin orbitals ψ μ of well-defined atomic energies ϵμ labeled by space
and spin quantum numbers μ, and metal energy band spin orbitals ψ j with a
continuum of energy levels ϵj, where j = b, k,σ stands for the indices of a band
b, electronic wavevector k, and electron spin quantum number σ. For solid sur-
faces of transition metals such as Ni or W, an intermediate situation arises invol-
ving narrow energy bands from states of both localized d- or f-orbitals and broad
energy bands of delocalized s-orbitals, competing for bonding with the adsorb-
ate atom orbitals. Bonding here shows some of the features (bond distances
and binding energies) of ligand-field theory [1, 29] for transition element
compounds.
For an atom adsorbed on a metal surface, the interaction of an atomic orbital
with metal orbitals leads to shifts and broadening of the atomic energy levels and
to charge rearrangement. A one-electron picture provides insight on the charge
transfer process. At large distances, the electron transfer can be understood
comparing the ionization I = − Eion and affinity A = − Eafn energies of an atomic
electron with the solid’s work function W, leading to an electron moving from
the atom to the solid if W > I or to electron attraction from the solid to the atom
if A > W. At shorter distances, as described in Section 8.2, the electron in the atom
interacts with its opposite-sign image in the solid and this creates an attractive
potential energy CeCe,img/(4πε04Z), with CeCe,img < 0, on the electron toward
the solid, which decreases as the electron is removed away from the surface.
Therefore, the effective ionization energy for the atom at distance Z from the sur-
face is given by Ieff (Z) = I + CeCe,img/(4πε04Z), a smaller value than I. Bringing
instead an electron to the molecule from far away and considering its positive
image charge gives an effective affinity Aeff (Z) = A − CeCe,img/(4πε04Z), a larger
value. Comparing these to the work function W of the solid, it follows that when
W > Ieff (Z), it is energetically favorable for an electron to move from the adsorbate
to the solid to gain a binding energy W − Ieff (Z), while when Aeff (Z) > W, the
electron moves from the solid to the adsorbate to form a negative ion. These
arguments also apply to a molecule interacting with the metal surface, with its
center of mass at a distance Z from the surface.
 8.5 Chemisorption 355

Within a many-electron description, the ground potential energy as a func-

tion of the distance Z between A and M shows a rapid change at a crossing point
Zc between a rapidly changing Coulomb attraction potential energy of the two
charged species and a slowly changing potential function for the neutral species,
shaped by electronic exchange and correlation, as shown in Figure 8.1b. The
first case, of an electron moving into the solid to form A+ + M− at intermediate
distances where W > Ieff(Z), is usual for the interaction of an alkali atom with the
surface. As the distance Z increases, the Coulomb potential energy between the
+ −
two ionic species goes as V A M Z = − Ce 2 4πε0 Z + W −Ieff Z , which
increases towards W − I at large distances. However, it is likely to intersect a
potential energy V (AM)(Z) for the two neutral species at a crossing distance
+ −
Zc where V A M Zc = V AM Zc and the electron moves back to the mole-
cule. Taking V (AM)
(Zc) ≈ 0, this gives

Ce 2 − Ce Ce, img 4 Ce 2 3 εr + 2 1
Zc = =
4πε0 W −I 4πε0 4 εr + 1 W − I

where as before, εr is the static dielectric constant of the solid. At short dis-
tances, repulsion forces take over, while at distances larger than Zc, the potential
energy is made up of induction and dispersion terms. A potential can be con-
structed over all distances using a crossing (or switching) function around Zc, as
done in the chapter on model potentials.
The reverse charge transfer situation involves A− + M+ present at intermedi-
ate distances, going into A + M at large distances, for example for halide atoms
interacting with the surface. Transfer happens for increasing distances at the
− +
crossing value determined instead by the difference A − W in V A M Zc =
− Ce 2 4πε0 Zc + Aeff Zc −W ≈0, past which the electron moves back to
the metal.
A one-electron model extends the treatment described in Section 8.4 for phy-
sisorption, to allow here for strong coupling of molecular and metal orbitals.
The strong coupling leads to large changes in molecular orbital energy shift
and broadening. These can be obtained using a partitioning treatment, which
leads to energy-dependent level shifts and widths, and several different cases
for chemisorption due to electron transfer [8, 31]. The molecular and metal
orbitals φμ and φj, j = b, k, σ , for given electron spin quantum number, can
be taken to be orthonormalized to begin with, so that φμ φμ = δμμ and
φj φj = δjj . Molecular and metal orbitals are nonorthogonal with overlap
φμ φj = Sμj. A transformation of molecular orbitals from the set {φμ} to a
new basis set {φν} can, however, be done by symmetrical orthogonalization
[57], to obtain new molecular and metal orbitals so that φν φj = δνj, and to sim-
plify a partitioning treatment. In practice, the original overlap integrals are small
356 8 Interaction of Molecules with Surfaces

and the new functions φν can be physically related to the old φμ and assigned to
atomic positions. Partitioning relies on the introduction of projection operators

A = ν
φν φν , M = j
φj φj

satisfying A + M = I, 2A = A , 2M = M , and A M = 0. The density of elec-

tronic states per unit energy gAM(ε) follows from the resolvent operator G ε in
the equation for electron m,

ε + iη− f m Gm ε = I

with f m = hm + um the effective Hamiltonian containing a mean-field averaged

electron–electron interaction um , Gm is the resolvent operator for the equation,
and the limiting parameter η 0 + . For a given electron, projecting the above
equation with A to the right, then with A to the left and also with M to
the left,

ε + iη− f AA GAA ε − f AM GMA ε = A

− f MA GAA ε + ε + iη− f MM GMA ε = 0

where, for example, f AM = A f M . Solving formally for GMA from the second
line and substituting in the first line, one obtains
−1
1
GAA ε = ε + iη −f AA − f AM f MA
ε + iη− f MM

Then in the limit η 0+, we can use the relation

1 1
= − iπδ E
E + iη E

with indicating the principal value of an integral, to obtain a reduced resolvent

displaying energy shift ΔA and width ΓA operators in
−1
GAA ε = ε + iη− f AA − ΔA ε + iΓA ε

ΔA ε = f AM f MA
ε −f MM

ΓA ε = πf AM δ E −f MM f MA
 8.5 Chemisorption 357

The density of states at A follows from

1
gA, M ε = φν δ ε −f φν = − Im φν GAA ε φν
ν π ν

−1
φν GAA ε φν = ε + iη− fνν − Δν ε + iΓν ε
and its imaginary part as

1 Γν ε
gA, M ε =
π ν ε −fνν − Δν ε 2
+ Γν ε 2

2
Γν ε = π f
j νj
δ ε − fjj ,
2 ∞
fνj Γν ε
Δν ε = j ε−f
= dε
jj −∞ ε −ε

where the summation over j = b, k involves also integration over values of the
wavevector k. This density of states can be considered for each separate state φν
of energy fνν to find how the molecular energy levels have changed due to inter-
actions with the metal. The density of states function gA,M(ε) can be used to cal-
culate a potential energy function VAM R,QA ,QM similar to the one given in
Section 8.4.1 within a one-electron treatment of interaction energies at surfaces,
possibly including now the change in residual correlation energies, by means of

VAM = dεε nAM ε gAM ε − nA ε gA ε −nM ε gM ε + EresAM −EresA −EresM

calculated for varying atomic positions. The partitioning treatment has also
been developed for time-dependent equations that arise in the description of
electron transfer dynamics and has been applied to calculations of their prob-
abilities in collisions of ions with metal surfaces [58, 59].
Given the forms of chemisorption shifts and widths Δν(ε) and Γν(ε) as they
follow from the location and widths of energy bands in the metal and from
the strength of the molecule–metal coupling energies fνj, it is possible to classify
the changes of A energy levels due to chemisorption as giving either sharp iso-
lated levels or shifted and broadened ones that are overlapping [8, 31]. For iso-
lated levels, it is convenient to find the roots εν of the equation ε − fνν − Δν(ε) = 0,
and to expand the density of states around each root value. Letting ε1 and ε2 be
the lower and higher limits of the range of Γν(ε), and considering first weak cou-
pling and only one root εν near fνν, the new level εν = fνν + Δν εν has no width if
εν < ε1 or εν > ε2 , and has a small width Γν(εν) if ε1 ≤ εν ≤ ε2 .
When the coupling is strong and level widths overlap, an alternative treatment
is to first combine an adsorbate orbital with one or more substrate orbitals, such
358 8 Interaction of Molecules with Surfaces

as a 3s- or 3p- orbital of Si and a d-orbital of Ni(001) for Si adsorbed on Ni, to form
surface (bonding and anti-bonding) orbitals φs, s = a, b and to calculate their ener-
gies εs. Allowing for their interaction with the remaining energy band orbitals of
the metal substrate, the partitioning procedure provides their shifts Δs and widths
Γs. This is a generalization of tight-binding models, valid for adsorbates and their
energies near metal surfaces [7, 8].
In applications, the matrix elements of f m can be parametrized at the Har-
tree–Fock level using ionization and affinity energies for the molecule and
energy bandwidths for the metal, to reach qualitative conclusions about chem-
isorption energies. Parametrizations of chemisorption energy expressions con-
taining the coupling energy between adsorbate and substrate orbitals have also
been developed in semiempirical models and have been used in treatments of
catalysis, at surfaces of simple and transition metals [23].
Some of the considered systems have been hydrogen, oxygen, and alkali atoms
adsorbed on light metals and transition metals and CO adsorbed on transition
metals [8, 60]. An example is provided by the chemisorption of O atoms on the
Ni(111) and Ni(001) surfaces. The coordination number of O with a Ni surface
atom changes from three to four between the surfaces, respectively, with a
change of the amount of electronic charge transfer between O and Ni, in a
one-electron description. This can qualitatively explain the corresponding
changes in binding energy and bond distance for the two surfaces [8, 55]. Cor-
rect qualitative conclusions on chemisorption energy trends can be obtained
with one-electron treatments, but accurate values of chemisorption energies
and potential energy changes with the relative position of the molecule near
the surface require a many-electron treatment, or a DFT treatment including
dispersion energies.
At intermediate and shorter distances, the potential energy is shaped by elec-
tron exchange and correlation, while large distances involve electrostatic,
induction, and dispersion energies. Many-electron treatments can be done
within DFT extended to include long-range dispersion energies, or more accu-
rately with atomic lattice models for the metal slab and adsorbate. The roles of
ionization and affinity electron energies were found in early DFT calculations,
with electronic density distributions for Li, Si, and Cl adsorbed on a metal with
the density of Al corresponding to a spread radius rs = 2.0 au, and the jellium
model for the metal, and are shown in Figure 8.6 [61]. Electronic rearrangement
increases that electronic density for Li and Si in the region between atom and
surface, and decreases it between Cl and the surface. Potential energies of inter-
action follow from the subtraction of energy density functionals, as previously
given, and display bonding minima and equilibrium chemisorption distances
due to the attraction effect of long-range induction forces.
The main features of the DFT results for chemisorption can be reproduced by
an effective-medium treatment of chemical binding of an atom A to a surface M
with a realistic inhomogeneous electronic distribution ρAM r [35], embedding
 8.5 Chemisorption 359

Lithium Silicon Chlorine

Total
1a.u.

superposition
Total minus
XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

XX

metal
Bare
Figure 8.6 Electron density contours for chemisorption of Li, Si, and Cl on a substrate with the
electron density of Al, corresponding to a spread radius rs = 2.0 au (or 0.03 electrons/bohr3). The
middle panels (the second row) show the total densities minus the superposition of the atom
and metal densities and the extent of electronic charge transfer, with full lines indicating a net
increase in density and dashed lines a decrease. The bottom panels show the bare metal
electron density near its surface. Source: from Ref. [61]. Reproduced with permission of Elsevier.

the atom in a homogeneous electron gas of a density ρS equal to the average

density of the host surface, to calculate its energy. Simple first-order and
second-order corrections to the energy account for binding changes due to
the host charge inhomogeneity and related distortion charge density
ΔρAM r = ρAM r − ρM . More recently, density functionals including gradient
corrections, hybrid forms with partial exact exchange energy, and with added
dispersion energies, have been used to generate accurate PESs not only for elec-
tron transfer, but also for adsorbate dissociation and reactions.

8.5.3 Dissociation and Reactions at a Metal Surface from Density

Functionals
The PES for a diatomic molecule AB brought to the metal surface at a relative
position Rads from which it dissociates into A + B is given by
diss AB, M
VAB, M Q = E ABM
Q; ρABM r −E Q; ρAB, M r −E M
QM ; ρM r
360 8 Interaction of Molecules with Surfaces

with Q = (QA, QB, QM) containing atomic positions in a reference frame

attached to the surface. The first term to the right contains all active atoms
of AB and of the substrate metal, the second term is the energy of the AB mol-
ecule initially adsorbed at the metal surface, and the third term is the energy of
the pure metal. The set of all metal and adsorbate atomic positions can be sepa-
rated into a small subset for the active atoms that undergo large displacements,
and a larger subset for the remaining atoms, which stay close to their original
equilibrium positions.
Chemisorption of a closed-shell adsorbate on a metal surface usually involves
bond breaking in the adsorbate, with interaction energies, which depend on its
orientation and the surface structure as defined by its Miller indices [39]. The
location and orientation of a diatomic molecule near a surface are fully given
by six position variables, which specify the surface location (X, Y) under the dia-
tomic center of mass (or CM), the distance Z between a surface plane and the CM,
the distance R between the diatomic atoms, and two angles (ϑ, φ) for the orien-
tation of the diatomic axis. A detailed treatment of interaction energies involves
consideration of the six variables and also their changes with the Miller indices of
the surface, a very extensive effort that has been undertaken in only a few cases.
Some insight can be extracted from a simple example, for H2 on the Ni(001)
surface leading to its dissociation and to adsorption of the two hydrogen atoms
[55, 62]. In a simple description, isocontours of the potential energy V(Z, R) ver-
sus the distance Z of the center of mass of H2 to the surface, oriented parallel to
it and to a row of surface atoms, and versus the distance R between the two
hydrogen atoms, shows an energy valley as Z decreases and R increases from
the equilibrium RHH distance, with an activation energy along a path going into
a products valley where R is large and Z goes to a constant at the equilibrium
ZHNi distance. This is also found for N2 dissociating on W. Adsorption of frag-
ments following dissociation frequently occurs at transition metal substrate
solids such as Ni or W with their localized d- and f-electron surface orbitals
in open shells available for bonding with the fragments. In contrast, H2 oriented
parallel to a Cu surface does not show dissociation upon adsorption because the
Cu atom d-shell is closed and the H + Cu interaction is weaker, but H2 is phy-
sisorbed there.
However, chemisorption is found in more detailed treatments involving cal-
culations of interaction energies as functions of all the six coordinates of the
adsorbed H2 and as the Cu surfaces change with their Miller indices. They have
been done for H2 on Cu(001) with a cluster model and ab initio configuration
interaction wavefunctions [63], and also with a slab model and a GGA DFT
functional [64], and have been done for H2 on Cu(111) with DFT functionals
[65]. These calculations show reaction paths with activation barriers toward dis-
sociation of H2 into two adsorbed H atoms, with barrier energies that can be
overcome in collisional events by the molecular translational energy or by its
vibrational energy. The addition of dispersion energies to these results, not
 8.5 Chemisorption 361

included in the cited work, may, however, be needed to obtain accurate values
for activation barriers.
Another example is given by the diatomic dissociation and atomic adsorption
of O2 on an aluminum surface, for which one wants to establish whether there is
an activation energy and if so what is its origin, insofar it involves in principle
electronic charge and spin rearrangements. The orientation of O2 is again fully
specified by six atomic coordinates, and the aluminum surface has different
shapes for different Miller indices. A full treatment of the energetics of the dis-
sociation would be very laborious, but insight can be extracted considering only
approaches of O2 with its axis parallel or perpendicular to the surface of Al(111)
and at locations with chosen surface symmetries such as on-top, fcc, hcp, or
bridge of nearby surface Al atoms.
Figure 8.7 shows isocontours of energy versus O – O and O – Al distances
obtained by an embedding treatment [51, 52], with a cluster including O2
and several Al atoms embedded in an environment provided by an indented
Al surface. Results are presented in Figure 8.7a for an O2 perpendicular to
the surface and for the fcc symmetry, described with the PBE density functional,
while Figure 8.7b shows the results for the cluster and a CW treatment imple-
mented with CASSCF + PT2 calculations [50], which include long-range inter-
actions. The isocontours differ, with the CW results showing an activation

(a) (b)
DFT fcc ⊥ Embedded CW fcc ⊥
5 5 5 5
2 2
E(eV)

E(eV)

4 4 4 4
LA1–O2 (Å)

LA1–O2 (Å)

–1.3 –1.3
3 3 3 3

2 2 2 2

1 1 1 1
1.2 1.6 2.0 2.4 1.2 1.6 2.0 2.4
LO–O (Å) LO–O (Å)

Figure 8.7 Isocontours of energy versus O – O and O – Al distances obtained by an

embedding treatment with a cluster including O2 and several Al atoms embedded in an
environment consisting of an indented Al surface. Results in Figure 8.7a are for an O2
perpendicular to the surface and for the fcc symmetry, described with the PBE density
functional, while Figure 8.7b shows the results for a correlated wavefunction treatment
implemented with CASSCF + PT2 calculations. Source: From Ref. [50].
362 8 Interaction of Molecules with Surfaces

energy due to electronic charge rearrangement, which varies with the location
on the surface and is around 600 meV (57.891 kJ mol−1) for the on-top position
with R(O – Al) 280 pm, in accordance with experimental results. Additional
calculations of the PES for O2 on Al(111) have been used to parametrize a flex-
ible periodic LEPS potential energy function, suitable for reaction dynamics on a
periodic surface structure. It is constructed as in the LEPS formulation from
valence-bond Coulomb and exchange energies using Morse functions and
parameters, which are, however, expanded in Fourier series with the periodicity
of two-dimensional lattice distances corresponding to the surfaces’ Miller indi-
ces [66].
A surface chemical reaction such as A + BC AB + C, with A, B, and C here
taken to be atoms to simplify, starts with the two reactant species adsorbed at
positions RA, ads and RBC , ads relative to the surface, and proceeds on a PES

Q; ρABCM r − E A, M QA ,QM ; ρA, M r

rct
VABC , M Q = E ABCM

BC , M
−E QB ,QC ,QM ; ρBC , M r
with Q = (QA, QB, QC, QM). The first term to the right depends on positions of all
participating atoms including metal atoms, the second term is the energy of an
adsorbed atom A, and the last term is the energy of an adsorbed BC diatomic. As
in the case of dissociation, the set of all metal atom positions can be separated
into a small subset for active atoms that undergo large displacements during
reactions, and a larger subset for the remaining atoms that stay close to their
original equilibrium positions. These two subsets define a cluster and its envi-
ronment, respectively, and can be treated within embedding methods as
described.
One of the concerns about using DFT for chemisorption energies is that the
treatment can lead to fractional electronic charge transfer for each reaction frag-
ment, with a related unphysical energy dependence on a continuously varying
number of electrons, instead of physically meaningful fragments with integer
numbers of electrons. As described in Chapters 6 and 7, an alternative treatment
can be based on the introduction of an energy Etot written as a functional of elec-
tron orbitals as explained in Section 6.4, for the total system, to be obtained
within an optimized effective potential procedure for interacting adsorbate
K
and substrate fragments K. The functional Etot ψj r,ζ of a set of j = 1 to NK
K
spin orbitals ψ j for each fragment K can be constructed to describe electrons
moving in an effective potential Vemb r , optimized by energy minimization, to
OEEP
derive an optimized embedding effective potential (OEEP) Vemb r suitable
diss
for chemisorption phenomena, with potential energy functions VAB, M Q
rct
and VABC , M Q obtained as sums of energies for fragments with integer num-
bers of electrons in each one [52, 67].
 8.6 Interactions with Biomolecular Surfaces 363

8.6 Interactions with Biomolecular Surfaces

The interaction of molecules with biomolecular surfaces is a very extensive and
very active area of research, where the same physical and chemical concepts can
be used as done so far for interactions at inorganic surfaces. But the complex
atomic composition of biological materials, plus the fact that they are soft mat-
ter presenting multiple structural shapes, pushes the available methodologies to
their limits and requires novel approaches for theory and computational work.
Progress has nevertheless been made using force fields for biosurfaces interact-
ing with both small molecules and large molecules such as proteins, and using
structural multiscale methods. Ongoing research involves development of new
polarized force fields, and many-atom modeling of cohesive energies and of
interaction energies between molecules and biosurfaces.
Biological surfaces can take spherical shape as in micelles, or can display
extended surfaces as in membranes of biocells, in the presence of a water sol-
vent. They are usually formed by assemblies of amphiphilic organic molecules,
with hydrophilic electrically charged heads assembled to form the boundaries of
the micelles or membranes in contact with water molecules, and long hydro-
phobic tails pointing away from the aqueous region. Molecules with short tails
can form micelles while molecules with long tails form layers or bilayers [68].
For example, phospholipid molecules with a phosphoric acid group as a head
and two hydrocarbon tails assemble to form a bilayer membrane, of thickness
between 5 and 8 nm, with the heads containing phosphate groups arrayed as a
surface, shown in Figure 8.8.
Molecules can interact with micelles and membranes to form biological com-
plexes. Biological membranes are found to contain extrinsic proteins at their
surfaces, or intrinsic proteins inside the membranes. Small molecules composed
of a few atoms can cross biomembranes through pores moving along channels
created by tubular proteins [5].
Because biomolecular surfaces display large displacements of their atoms,
their fluid nature is more accurately described by treatments similar to those
for liquids, in terms of atom pair distribution functions and force fields, as
described in Sections 4.5 and 4.6.
The interaction energies between a molecule and a biosurface, in the presence
of a solvent such as water, are similar to the ones found between a molecule and
an extended many-atom solids, and they can be described in terms of force fields
provided these are modified to account for extended polarization effects, which
introduce dielectric constants in force fields, and must also incorporate solvent
effects as treated in Chapter 7. Polarization effects can be included by parame-
trizing potential energies between atomic groups containing embedded charges
and localized polarizabilities, as described in Section 4.6.
Force-field developments for modeling of lipid membranes have been recently
reviewed [69–71]. They are based on parametrized potential energy functions
364 8 Interaction of Molecules with Surfaces

Water

Hydrophilic
head

Hydrophobic
tail

Water

Figure 8.8 Biomembrane bilayer, to the left, formed by self-assembly of phospholipid

molecules as seen to the right, with hydrophilic heads surrounded by water molecules.
Source: Adapted from public domain figures.

containing bonding V (B) and nonbonding V (NB) terms as described in Section

4.6.2, with
B
V = Vstr + Vbnd + Vtor + Vbck + VH − bnd
where Vstr is the bond stretch component involving sums over two atoms, Vbnd
is the angle bending components for sets of three atoms, Vtor the torsion angle
component with sums over four atoms, Vbck is the buckling component also
for four atoms, and the last term is a hydrogen-bonding energy dependent
on the positions of a hydrogen and two other atoms. The nonbonding potential
energy is
12 6
σ IJ σ IJ Coul ind
V NB
= ϵ
I , J IJ
−2 + vIJ Rij + vIJ Rij
RIJ RIJ
 8.6 Interactions with Biomolecular Surfaces 365

Coul Ci Cj
vIJ Rij = fd Rij
i I, j J 4πε0 εr Rij

ind Ci αj + αi Cj
vIJ Rij =− fd Rij
i I, j J 4πε0 εr 2 Rij 4

where εr is the dielectric constant of the medium. The first term in the summa-
tion describes dispersion energies, and the Coulomb and induction terms have
been given as sums of pair interactions between point charges Ck and polariz-
abilities αk in fragments = , located at the relative distance Rij. They con-
tain a damping function fd(Rij) ≈ 0 for Rij 0 at short distances, as described in
Chapters 4 and 7. These point charges and polarizabilities are approximate
values for the actual spatial distributions of charges and polarizabilities, which
furthermore change as conformations vary. With reference to citations in [69],
well-tested force fields (FFs) are available from GROMOS, where CHn, n = 1−3,
groups of atoms are treated as a single particle, as recently improved for lipids
[72], and from the all-atoms CHARMM FF as recently improved for lipids [73],
AMBER with improvements incorporated for lipids, MARTINI with a coarse-
grained model [74], and other ongoing developments for polarizable FFs suit-
able for lipids.
The very large number of atoms to be considered in the calculation of struc-
tural and dynamical properties of a biomembrane interacting with a large
molecule presents a challenge in the computational modeling of the complex.
This can be dealt with grouping several atoms into chemically meaningful
fragments, which become basic constituents, reducing the total number of
degrees of freedom in the whole system. It requires parametrization of frag-
ment properties independently calculated with many-electron treatments.
Consistent ways to do this employ multiscale techniques based on a full
atomic description and proceed from the ground-up by coarse-graining
many-atom systems into fragments (or new particles), to describe condensed
matter and biomolecular systems with fewer interacting particles [75–79].
Optimal ways to extract parameters for the fragment structures and their
interaction, from extensive atomistic calculations, can be found using
machine learning of artificial intelligence rules from big data collected for
biomolecules [80–82].
A quantitative coarse-graining (CG) treatment of the modeling of a large
many-atom system by a smaller collection of particles can be done starting from
a mapping of the set RA = r 3NA of all the 3NA atomic position coordinates into
a set of 3NP model particle position coordinates RP = r 3NP plus NL model
parameters λ, such as point charges and polarizabilities, collected in the set
of particle variables (Λ = RP,λ). The mapping is of form

Λ = M RA
366 8 Interaction of Molecules with Surfaces

where Λ and RA are column matrices and M is a matrix of order 3NA × (3NP +
NL). This describes a general case where the parameters λ are functions of the
original atomic positions and change with the conformation of molecules.
A simpler treatment is usually developed where the particle parameters do
not change, in which case a simpler mapping is done between atomic positions
and particle positions of the form RP = MPA RA ; λ .
The construction of particles as groups of atoms must be done so that their
interactions reproduce statistical properties of the original many-atom system.
This can be enforced using the probability distribution pA RA of the atomic
ensemble, to obtain the probability distribution of particles by means of

pP Λ = dRA pA RA δ M RA − Λ

in a compact notation signifying an integral over 3NA variables. Alternatively,

the model system can be described by a potential of mean force UP(Λ, β) when
the original atomic system has been kept at a temperature T = 1/(kBβ), defined
by a Boltzmann-type distribution pP Λ,β = exp − βUP Λ ZP β , with ZP a
partition function for the system of particles. This gives for the potential of
mean force

UP Λ, β = −β − 1 ln ZP β dRA pA RA ,β δ M RA − Λ

where pA RA ,β is the original Boltzmann distribution for the atomic system. It

simplifies to give a function UP(RP, β) when the parameters can be chosen to be
independent of conformations. Construction of this potential energy starts with
a calculation of pA RA , β possibly for a small atomic subsystem, to identify par-
ticles’ structure and properties to be incorporated in the full system. A variety of
ways to do this by iteration has been described in the reviews already cited, and
also in a proposed treatment based on the minimization of a relative entropy
function Srel = I pP I ln pP I pA I where the summation is over configura-
tions in the CG procedure [83].
An example of application of the CG procedure is found in a calculation
where the potential of mean force is decomposed into sums of pairs of particle
interactions, within a multiscale-CG (or MS-CG) procedure, which does the
force-matching by linearization of equations with the original pair forces
[84]. It has been applied to a model lipid bilayer composed of dimyristoilpho-
sphatidilcholine (or DMPC) molecules, containing two alkane chains. This mol-
ecule, with nearly 50 atoms, was modeled by 13 particles associated with the
choline (CH), phosphate (PH), glycerol (GL), ester groups (E1 and E2), six
alkane particles (SM), and two tail ends (ST). Results from the MS-CG simula-
tion reproduced the structural properties of the lipid bilayer as obtained from
the atomistic simulation, quite accurately [84].
 References 367

As more applications of coarse-graining treatments are done, and provide big

data collections, it becomes possible to use machine learning of rules for choos-
ing optimal mappings of the original many-atom system into a smaller system of
particles, while reproducing relevant properties, such as total energies of chan-
ging conformations. Artificial intelligence tools are being developed and applied
for that purpose [80–82], and are available for treatments of interactions at
complex biomolecular surfaces, including the related important determination
of the optimal location and binding of a small bioactive molecule on a mem-
brane from among a very large number of possible sites.

References
1 Atkins, P. and De Paula, J. (2010). Physical Chemistry, 9e. New York: W. H.
Freeman and Co.
2 Kittel, C. (2005). Introduction to Solid State Physics, 8e. Hoboken, NJ: Wiley.
3 Somorjai, G.A. (1972). Principles of Surface Chemistry. Englewood Cliffs, NJ:
Prentice-Hall.
4 Somorjai, G.A. (1994). Introduction to Surface Chemistry and Catalysis. New
York: Wiley-Interscience Publication.
5 Israelachvili, J. (2011). Intermolecular and Surface Forces, 3e. New York: Elsevier.
6 Lang, N.D. (1973). The density-functional formalism and the electronic
structure of metal surfaces. In: Solid State Physics, vol. 28 (eds. H. Ehrenreich, F.
Seitz and D. Turnbull), 225. New York: Academic Press.
7 Martin, R.M. (2004). Electronic Structure: Basic Theory and Practical Methods.
Cambridge, England: Cambridge University Press.
8 Desjonqueres, M.C. and Spanjaard, D. (1995). Concepts in Surface Physics, 2e.
Berlin, Germany: Springer-verlag.
9 Lang, N.D. (1983). Density functional approach to the electronic structure of
metal surfaces and metal-adsorbate systems. In: Theory of the Inhomogeneous
Electron Gas (eds. S. Lundqvist and N.H. March), 309. New York: Plenum Press.
10 May, V. and Kuhn, O. (2000). Charge and Energy Transfer Dynamics in
Molecular Systems. Berlin: Wiley-VCH.
11 Mukamel, S. (1995). Principles of Nonlinear Optical Spectroscopy. Oxford,
England: Oxford University Press.
12 Micha, D.A. (2015). Generalized response theory for a photoexcited many-atom
system. In: Adv. Quantum Chemistry, vol. 71, Chapter 8 (eds. J.R. Sabin and
R. Cabrera-Trujillo), 195. New York: Elsevier.
13 Yu, P. and Cardona, M. (2005). Fundamentals of Semiconductors: Physics and
Materials Properties, 3e. Berlin, Germany: Springer-Verlag.
14 Haug, H. and Koch, S.W. (2004). Quantum Theory of the Optical and Electronic
Properties of Semiconductors. World Scientific.
368 8 Interaction of Molecules with Surfaces

15 Ashcroft, N.W. and Mermin, N.D. (1976). Solid State Physics. London, England:
Thomson.
16 Landau, L.D. and Lifshitz, E.M. (1960). Electrodynamics of Continuous Media.
London: Pergamon Press.
17 Jackson, J.D. (1975). Classical Electrodynamics. New York: Wiley.
18 Mavroyanis, C. (1963). The interaction of neutral molecules with dielectric
surfaces. Mol. Phys. 6: 593.
19 McLachlan, A.D. (1964). Van der Waals forces between an atom and a surface.
Mol. Phys. 7: 381.
20 Zaremba, E. and Kohn, W. (1976). van der Waals interaction between an atom
and a solid surface. Phys. Rev. B 13: 2270.
21 Kramer, H.L. and Herschbach, D.R. (1970). Combination rules for van der Waals
force constants. J. Chem. Phys. 53: 2792.
22 Bortolani, V., March, N.H., and Tosi, M.P. (eds.) (1990). Interaction of Atoms and
Molecules with Solid Surfaces. New York: Plenum Press.
23 Greeley, J., Noerskov, J.K., and Mavrikakis, M. (2002). Electronic structure and
catalysis on metal surfaces. Annu. Rev. Phys. Chem. 53: 319.
24 Huang, P. and Carter, E.A. (2008). Advances in correlated electronic
structure methods for solids, surfaces, and nanostructures. Annu. Rev. Phys.
Chem. 59: 261.
25 Wolken, G.J. (1976). Scattering of atoms and molecules from solid surfaces. In:
Dynamics of Molecular Collisions Part A (ed. W.H. Miller), 211. New York:
Plenum Press.
26 Goodman, F.O. and Wachman, H.Y. (1976). Dynamics of Gas-Surface Scattering.
New York: Academic Press.
27 Celli, V., Eichenauer, E., Kaufhold, A., and Toennies, J.P. (1985). Pairwise
additive semi ab initio potential for the elastic scattering of He atoms from the
LiF(001) crystal surface. J. Chem. Phys. 83: 2504.
28 Zangwill, A. (1988). Physics at Surfaces. Cambridge, England: Cambridge
University Press.
29 Levine, I.N. (2000). Quantum Chemistry, 5e. Upper Saddle River: Prentice-Hall.
30 Parr, R.G. and Yang, W. (1989). Density Functional Theory of Atoms and
Molecules. Oxford, England: Oxford University Press.
31 Newns, D.M. (1969). Self-consistent model of hydrogen chemisorption. Phys.
Rev. 178: 1123.
32 Maurer, R.J., Ruiz, V.G., Camarillo-Cisneros, J. et al. (2016). Adsorption
structures and energetics of molecules on metal surfaces: bridging experiment
and theory. Prog. Surf. Sci. 91: 72.
33 Landman, U. and Kleiman, G.G. (1977). Microscopic approaches to
physisorption. In: Surface and Defect Properties of Solids, vol. 6 (eds. M.W.
Roberts and J.M. Thomas), 1. London: Chemical Society.
34 Gordon, R.G. and Kim, Y.S. (1972). Theory for the forces between closed shell
atoms and molecules. J. Chem. Phys. 56: 3122.
 References 369

35 Norskov, J.K. and Lang, N.D. (1980). Effective-medium theory of chemical

binding: application to chemisorption. Phys. Rev. B 21: 2131.
36 Lang, N.D. (1981). Interaction between closed-shell systems and metal surfaces.
Phys. Rev. Lett. 46: 842.
37 deLara-Castells, M.P., Stoll, H., and Mitrushchenkov, A.O. (2014). Assessing the
performance of dispersionless and dispersion-accounting methods: Helium
interaction with cluster models of the TiO2 Surface. J. Phys. Chem. A 118: 6367.
38 deLara-Castells, M.P., Bartolomei, M., and Mitrushchenkov, A.O. (2015).
Transferability and accuracy by combining dispersionless density functional and
incremental post-Hartree-Fock theories: Noble gases adsorption on coronene/
graphene/graphite surfaces. J. Chem. Phys. 143: 194701–194701.
39 Scheffler, M. and Stampfl, C. (2000). Theory of adsorption on metal surfaces.
In: Handbook of Surface Science: Vol. 2 Electronic Structure (eds. K. Horn and
M. Scheffler), 286. Amsterdam: Elsevier.
40 DaSilva, J.L.F., Stampfl, C., and Scheffler, M. (2005). Xe adsorption on metal
surfaces: first principles investigations. Phys. Rev. B 72: 075424–075421.
41 Chen, D.-L., AlSaidi, W.A., and Johnson, J.K. (2012). The role of van der Waals
interactions in the adsorption of noble gases on metal surfaces. J. Phys. Condens.
Matter 24: 424211.
42 Maurer, R.J., Ruiz, V.G., and Tkatchenko, A. (2015). Many-body dispersion
effects in the binding of adsorbates on metal surfaces. J. Chem. Phys. 143:
102808–102801.
43 Langreth, D.C., Lundqvist, B.I., Chakarova-Kaeck, S.D. et al. (2009). A density
functional for sparse matter. J. Phys. Condens. Matter 21: 084203.
44 Berland, K., Cooper, V.R., Lee, K. et al. (2015). van der Waals forces in
density functional theory: a review of the vdW-DF method. Rep. Prog. Phys. 78:
066501.
45 Grimme, S., Hansen, A., and Brandenburg, J.G.C. (2016). Dispersion corrected
mean-field electronic structure methods. Chem. Rev. 116: 5105.
46 Hermann, J., DiStasio, R.A.J., and Tkatchenko, A. (2017). First-principles models
for van der Waals interactionsin molecules and materials: concepts, theory, and
applications. Chem. Rev. 117: 4714.
47 Dobson, J.F. (2012). Dispersion (van der Waals) forces and TDDFT. In:
Fundamentals of Time-Dependent Density Functional Theory (eds. M.A.L.
Marques, N.T. Maitra, F.M.S. Nogueira, et al.), 417. Berlin: Springer-Verlag.
48 Dobson, J.F. and Gould, T. (2012). Calculation of dispersion energies. J. Phys.
Condens. Matter 24: 073201.
49 Vydrov, O.A. and Van Voohis, T. (2010). Nonlocal van der Waals density
functional: the simpler the better. J. Chem. Phys. 133: 244103–244101.
50 Libisch, F., Huang, C., Liao, P. et al. (2012). Origin of the energy barrier to
chemical reactions of O2 on Al(111). Phys. Rev. Lett. 109: 198303–198301.
51 Libisch, F., Huang, C., and Carter, E.A. (2014). Embedded correlation function
schemes: theory and applications. Acc. Chem. Res. 47: 2768.
370 8 Interaction of Molecules with Surfaces

52 Huang, C. and Carter, E.A. (2011). Potential-functional embedding theory for

molecules and materials. J. Chem. Phys. 135: 194104–194101.
53 Sudhyadhom, A. and Micha, D.A. (2006). Bonding and excitation in CO/Cu(001)
from a cluster model and density functional treatments. J. Chem. Phys. 124:
101102.
54 Sharifzadeh, S., Huang, P., and Carter, E. (2008). Embedded configuration
interaction description of CO on Cu(111): resolution of the site preference
conundrum. J. Phys. Chem. C 112: 4649.
55 Norskov, J.K. (1990). Chemisorption at metal surfaces. Rep. Prog. Phys. 53: 1253.
56 de Lara Castells, M.P., Cabrillo, C., Micha, D.A. et al. (2018). Ab initio design of
light absorption through silver atomic cluster decoration of TiO2. Phys. Chem.
Chem. Phys. 20: 19110.
57 Jensen, F. (2001). Introduction to Computational Chemistry. New York: Wiley.
58 Feng, E.Q., Micha, D.A., and Runge, K. (1991). A time-dependent molecular
orbital approach to electron transfer in ion-metal surface collisions. Int. J.
Quantum Chem. 40: 545.
59 Micha, D.A. and Feng, E.Q. (1994). The calculation of electron transfer
probabilities in slow ion-metal surface collisions. Comput. Phys.
Commun. 90: 242.
60 Muscat, J.P. and Newns, D.M. (1978). Chemisorption on metals. Prog. Surf. Sci. 9: 1.
61 Lang, N.D. and Williams, A.R. (1978). Theory of atomic chemisorption on
simple metals. Phys. Rev. B 18: 616.
62 Kresse, G. (2000). Dissociation and sticking of H2 on the Ni(111),(100), and (110)
substrate. Phys. Rev. B 62: 8295.
63 Madhavan, P. and Whitten, J.L. (1982). Theoretical studies of chemisorption of
hydrogen on copper. J. Chem. Phys. 77: 2673.
64 Wiesenekker, G., Kroes, G.J., and Baerends, E.J. (1996). An analytical six-
dimensional potential energy surface for dissociation of molecular hydrogen on
Cu(001). J. Chem. Phys. 18: 7344.
65 Hammer, B., Scheffler, M., Jacobsen, K.W., and Norskov, J.K. (1994).
Multidimensional potential energy surface for H2 dissociation over Cu(111).
Phys. Rev. Lett. 73: 1400.
66 Yin, R., Zhang, Y., Libisch, F. et al. (2018). Dissociative chemisorption of O2 on
Al(111): Dynamics on a correlated wavefunction-based potential energy surface.
J. Phys. Chem. Lett. 9: 3271.
67 Kummel, S. and Kronik, L. (2008). Orbital dependent density functionals: theory
and applications. Rev. Mod. Phys. 80: 3.
68 Lehninger, A.L., Nelson, D.L., and Cox, M.M. (1992). Principles of Biochemistry,
2e (Chapter 10). New York: Worth.
69 Lyubartsev, A.P. and Rabinovich, A.L. (2016). Force field developments for lipid
membrane simulations. Biochim. Biophys. Acta 1858: 2483.
70 Pluhackova, K., Kirsch, S.A., Han, J. et al. (2016). A critical comparison of
biomembrane force fields: structure and dynamics of model DMPC, POPC, and
POPE bilayers. J. Phys. Chem. B 120: 3888.
 References 371

71 Sandoval-Perez, A., Pluhackova, K., and Boeckmann, R.A. (2017). Critical

comparison of biomembrane force fields: Protein-Lipid interactions at the
membrane interface. J. Chem. Theory Comput. 13: 2310.
72 Poger, D., VanGunsteren, W., and Mark, A.E. (2010). A new force field for
simulating phosphatidylcholine bilayers. J. Comput. Chem. 31: 1117.
73 Klauda, J.B., O’Connor, J.W., Venable, R.M. et al. (2010). Update of the
CHARMM all-atom additive force field for lipids: validation on six lipid types.
J. Phys. Chem. B 114 (23): 7830–7843.
74 Marrink, S.J., Risselada, H.J., Yefimov, S. et al. (2007). The MARTINI force field:
cgrained model for biomolecular simulations. J. Phys. Chem. B 111: 7812.
75 Voth, G.A. (2009). Coarse-Graining of Condensed Phase and Biomolecular
Systems. Boca Raton, FL, USA: CRC Press.
76 Nielsen, S.O., Lopez, C.F., Srinivas, G., and Klein, M.L. (2004). Coarse grain
models and the computer simulation of soft materials. J. Phys. Condens. Matter
15: R481.
77 Shih, A.Y., Arkhipov, A., Freddolino, P.L. et al. (2007). Assembly of lipids and
proteins into lipoprotein particles. J. Phys. Chem. B 111: 11095.
78 Kamerlin, S.C.L., Vicatos, S., Dryga, A., and Warshel, A. (2011). Coarse-grained
(multiscale) simulations in studies of biophysical and chemical systems. Annu.
Rev. Phys. Chem. 62: 41.
79 Noid, W.G. (2013). Perspective: coarse-grained models for biomolecular
systems. J. Chem. Phys. 139: 090901–090901.
80 Rupp, M. (2015). Machine learning for quantum mechanics in n Nutshell. Int. J.
Quantum Chem. 115: 1058.
81 von Lilienfeld, O.A., Lins, R.D., and Rothlisberger, U. (2005). Variational particle
number approach for rational compound design. Phys. Rev. Lett. 95: 153002.
82 Brandt, S., Sittel, F., Ernst, M., and Stock, G. (2018). Machine learning of
biomolecular reaction coordinates. J. Phys. Chem. Lett. 9: 2144.
83 Shell, M.S. (2008). The relative entropy is fundamental to multiscale and inverse
thermodynamical problems. J. Chem. Phys. 129: 144108.
84 Izvekov, S. and Voth, G.A. (2005). A multiscale coarse-graining method for
biomolecular systems. J. Phys. Chem. B 109: 2469.
 373

Index

a adsorbate embedding potential 348

acceptance probability 150 adsorbate Hamiltonian 336
ACFD dispersion energy 294 adsorbate internal energy 336
activation barrier 14, 18, 120, 121, 126, adsorbate layer energy 335
127, 313, 360, 361 adsorbate level broadening and
activation energy 14, 149, 312, 313, shifting 339
353, 360, 361 adsorbate phase transformations 336
active learning 302 adsorbate vdW-DFT 346
active orbitals 213 adsorbed monolayer 335
additive adsorbate densities 344 adsorption potential 140
additive scheme 271, 314 Ag2 on TiO2(011) 351
adiabatic approximation 178, 189 Ag2/Ti9O25H14 352
adiabatic coefficient gradient 173 Aharonov–Bohm phase 189
adiabatic connection (ACFD) 232 AI-NN procedure 149–152, 300–303
adiabatic energies 174, 176, 182, 183 alkali–halide interactions 118
adiabatic expansion 173 all-ranges potential energy 93
adiabatic potential energy 10, 11, 173, AMBER 365
182, 183, 187, 188, 192 angle bending 145, 364
adiabatic representation 172–175, angle torsion 15, 16, 114, 145, 147,
178, 182, 184, 192 158, 364
adsorbate charge transfer 351 angular momentum projection
adsorbate charge transfer from operator 200
DFT 358 ANI-1 potential data 301
adsorbate density of states 340 anisotropic bond 283
adsorbate DFT-D 346 antibonding region 164
adsorbate DFT energies 343 anti-Morse function 125
adsorbate DFT-MBD 346 antisymmetrizer 199, 225
adsorbate dispersion coefficient 342 antisymmetry projection operator 199
adsorbate dispersion energy 341 Ar + Ar 91
adsorbate embedding 347 ArCH4 29

Molecular Interactions: Concepts and Methods, First Edition. David A. Micha.

© 2020 John Wiley & Sons, Inc. Published 2020 by John Wiley & Sons, Inc.
374 Index

Ar crystal 289 biomolecular surfaces 363–367

ArF 198 Bloch states 315
Ar-graphite interaction 335 body-fixed (CMN-BF) frame 52, 79,
Ar + H2 212 80, 158, 159
Ar+ + H2 187 bond-bond dispersion energy 282
Ar + H2+ 187, 188 bond–bond induction energy 282
Ar+H2O 227 bonded atom density 274
Ar on jellium 344 bonding region 164
artificial intelligence (AI) 149, bond stretching 145
300–303, 365, 367 Born–Mayer function 91
asymptotic consistency 112, 196, 210, Born–Mayer potential
215, 217 116, 119
asymptotic convergence 73 Born-Oppenheimer 5, 10
atom-atom bond orders 261 Born–Oppenheimer
atom forces 152 approximation 38, 178
atomic correlation function 131 bound nuclei 3, 5
atomic-fragment polarizability 274 Bravais lattice 310
atomic natural orbitals (ANOs) 207 Breit–Pauli treatment 244
atomic populations 261 Brillouin lattice 310
atomic pseudopotentials 208, 244, 245 Brillouin theorem 217
atomic weight function 273 Buckingham potential 115
atom-pair dispersion 292 buckling angles 145
atom-surface model potentials 333
atom-surface scattering potential 331 c
Au2 244 Cartesian tensor 71, 293
augmented basis functions 207 Casimir–Polder integral 77, 81,
avoided crossing 11, 17, 119, 176, 178, 238, 262
181–183, 186–187 cavity shape 256
avoided intersection 20–21 CCSD(T) treatment 219
avoiding double counting 228, center of mass of the nuclei
230, 291 (CMN) 52, 64, 65, 111–112, 122,
axial rotational energy 247 138, 158, 169
center-of-mass of the nuclei, space-
b fixed (CMN-SF) 64, 69, 112, 158,
basis set superposition error 208 159, 169
Be2 243 C2H∗2 17
benzene crystal 293 C5H5 189
benzene dimers 227, 237 charge-and bond-order 39
benzene on graphite 347 charge density 38, 39, 41, 50, 64, 65,
Berry phase 189 76–78, 88, 104, 106, 117, 122, 143,
bias parameter 150 144, 222–223, 230, 235, 238,
biomembranes 363–365 255–260, 264, 272–282, 288, 319,
biomolecular machine-learning 367 322, 334
 Index 375

charge density fluctuation 260, 262, computing times 209, 210, 350
291, 329 Condon approximation 176–180
charge-image potential energy 325 Condon states 179
CHARMM FF 365 conductive materials 322
CH2=[CH–CH=]kCH2 284 configuration interaction (CI) 99, 100,
chemical reactivity 2, 240, 299 113, 122, 165, 166, 198, 201,
chemisorption 310 209–215, 360
cluster model 350 configuration state function
density of states 357 (CSF) 210–215
on metals 350, 351 conical intersections 15, 17, 19, 181,
shifts 357 184–189, 192, 303
widths 357 conjugated polyenes 284
CH3(CH2)N-2CH3 281 consistent basis sets 206
CH3NH 303 continuum model 39, 144, 332–337
C4H10 on Cu(111) 347 contracted gaussians 206
C6H6 on Si(001)-(2×1) 353 CO on Cu(001), Cu(111) 349
chronological Table 2 coordinates scaling 167
CISD treatment 210 core point 115
Cl + H2 149 coronene dimer 293
closure approximation 58–59, 263 correlated embedding
CO2 22, 59, 60, 352 wavefunction 347
(CO2)2 227 correlation consistent orbitals 207
coarse-grain entropy 366 correlation diagram 95, 96
coarse-graining 144, 274, 277, correlation potential 216
365, 367 Coulomb forces 3, 4, 164,
coarse-grain mapping 365 166, 230
coherent states 68 Coulomb functional 89
cohesive energy 30, 106, 127–130, Coulomb integral 97, 99, 125, 211
132, 363 counterpoise (CP) correction 208
collisional cross sections 27–28 coupled cluster treatment 198,
combination rules 82–83, 91, 94, 201, 218
116–117, 264 coupled stretching bonds 146
short distances 91 covalent-ionic function 99
compensating phase functions 191 covalent-ionic states 223
complementary damping 118 crossing point 176, 182, 355
complete active space 214 crystalline benzene 289
complete active space self-consistent crystal structure 127, 290
field (CASSCF) 210, 214 CsCl(s) 129
complex-valued crossing 183 Cu(s) 128, 316, 322, 323, 337,
complimentary 349, 360
antisymmetrizer 199, 225 cubic crystalline lattice 141
complimentary operator 199 curvature 115
compound AB 196, 240 cyclic compounds 283
376 Index

d dielectric properties 3, 130

damped charge-charge 268 dielectric relaxation 57
damping function 93, 117–119, 147, different orbitals for different
230, 232, 268, 291, 292, 296, 330, spins 209, 215, 219, 269
342, 343, 346, 352, 365 dipolar hyperpolarizability 42
Darwin-like contact term 103 dipole-dielectric surface
data mining 300 interaction 327
data storage 209, 219 dipole polarizability 42, 262, 282
Debye equation 26 dipole rotation 80
Debye model 57 Dirac–Fock SCF 244
Debye unit 37 direct (or “pol”) energy 226
decay rate 48, 49, 59 direct product 56, 183, 187, 201
degenerate reference state 215, 220 dispersion energy 8, 24, 67, 74, 76, 77,
degenerate state 68, 187–189, 202, 79–87, 91, 93, 104, 105, 146, 151,
245, 260 226, 228, 230–232, 236, 240, 259,
degrees of freedom 13, 112, 119, 122, 262, 264, 265, 267, 270, 271,
169, 179, 184, 186, 365 279–282, 286, 288, 289, 291,
delayed response function 46, 292, 296, 331, 332, 336, 342, 345,
50, 319 346, 350, 352, 353, 358,
delocalized electrons 22, 128, 140, 360–361, 365
205, 284, 310, 314, 316, 337, 338, dispersive dielectric function 324
350, 354 dissociation 14–16, 90, 185, 221, 229,
density decomposition 38 239, 240, 299, 302, 303, 352,
density normalization 203 359–362
density operator expansion 275 dissociation at a surface 312, 352,
depolarization field 130, 321 359–362
derived properties 2 distorted atoms 292
DFT approach 90 distortion force 184
DFT-CW combined embedding 349 from MO, AO, CI 165
DFT-D approach 292, 350 distortion multipoles 7, 74, 117
DFT-D3 calculations 293 distributed multipoles 39, 122, 124,
DFT-D treatments 230, 231, 346, 265, 272, 278
350, 352 distributed polarizabilities 272
DFT-MBD treatment 294, 346 DMPC bilayer 366
DFT-plus-vdW 230 double point groups 183
D + H2 HD + H 191 double PT 227
diatomic-diatomic 122 dynamical polarizability 8, 23, 43–49,
diatomics on metals 360 59, 82, 106, 226, 227, 232, 267, 274,
dielectric displacement 321 281, 284, 292, 293, 322, 327, 332
dielectric function 320–324, 327, dynamical susceptibility 51, 77, 130,
328, 342 230, 232, 234, 236, 260, 262–264,
dielectric polarizability 324 276, 279, 281, 286, 287, 291, 294,
dielectric polarization 320 319, 320, 322, 329, 346
 Index 377

e electrostatic embedding 271

effective atomic polarizabilities 293 electrostatic energy 8, 64, 71, 73, 258,
effective charge 97, 99 267, 270, 334
effective fragment potential 269 electrostatic force theorem 163–164
effective Hamiltonian 217, 221, 315, electrostatic potential 68, 70–72, 76,
337, 341, 356 265, 318, 326, 334, 338
effective-medium treatment 344, 358 embedded cluster 298, 347, 349
effective nuclear charges 97, 99, 100, embedding DFT 298
117, 166, 244, 245 embedding effective potential 300
electric/electrical multipole 8, 10, embedding environment 300
21–22, 35–40, 60, 80, 148, 180, 203, embedding equilibration 299
239, 262, 265 embedding potential 299, 348,
electric field 22, 23, 25, 35–37, 40–43, 349, 351
46, 47, 51, 53, 246, 258, 275, 292, energy components 118, 150
319, 320, 324, 326, 331 atom-atom, atom-bond,
electric potential 36, 37, 50, bond-bond 261
52, 70, 234, 265–268, 325, energy functional 7, 87, 89, 91, 97, 161,
331, 344 165, 231, 240, 241, 313, 317,
electron charge 5, 22, 36, 58, 84 349, 351
electron exchange 8, 10, 15, 38, 42, energy isocontours 14, 127, 147, 353
87–101, 104, 196, 198, 199, equation-of-motion CC
226–228, 256, 257, 260, 289, (EOM-CC) 219
290, 358 ethylidene 17, 19
electronic charge density 38, 64, 88, Euler angles 53, 76, 79, 80, 122,
144, 255–257, 260, 275, 279, 284, 134, 158
312, 329 exchange antisymmetry 198
electronic charges overlap 268 exchange-correlation 93, 163, 231,
electronic density matrix 203–205 234, 236, 243, 296, 317–318
electronic embedding 314 exchange-correlation energy 89, 162,
electronic exchange 10, 66, 98–101, 232, 346
143, 162, 195–197, 230–238, 315, exchange energy 7, 10, 226–229, 239,
317, 355 241, 268, 362
electronic kinetic energy 5, 89, 144, exchange integral 99, 101, 125,
160, 162, 169, 241 126, 211
electronic representations 170, exchange repulsion 268, 269
172–181 excitation operator 218, 219, 221
electronic surface dipole 316 excited state 5, 11, 12, 14, 15, 23, 38,
electron self-interaction 229, 239 41, 59, 67, 68, 73, 76, 79, 80, 83, 84,
electron transfer 11, 15, 17, 18, 29, 87, 105, 106, 148, 185, 189, 200, 210,
99, 196, 197, 222–224, 242, 268, 214, 217, 219, 220, 226, 235, 259,
312, 349, 354, 355, 357, 359 260, 264, 266, 267, 289, 290, 303,
at a metal 354 319, 329
at surfaces 312 excitonic energies 290
378 Index

excitons 290, 320 g

expansion divergence 78 Gaussian distribution 57
expansion parameter 66, 102 Gaussian orbital 206
exponential-6(α) 115 generalized gradient functionals 229,
exponential potential 7 237, 346
exponential-6 potential 9 generalized valence-bond
extended molecule 49–52, 319 (GVB) 222–224
extended valence-bond (EVB) 99, 198 geometric phase 189, 191, 192
Gibbs free energies 19–21, 24, 25
f glancing parabolas 183
F2 243 glancing paraboloids 186
Fe2+/Fe3+ 20 global symmetry variable 301, 302
F + H2 FH + H 15, 126 global variables 148–152, 300, 302
finite elements 39, 40 Gordon–Kim procedure 162, 163
first-order energy 67, 105 Gordon–Kim treatment 90, 228
fixed-nuclei approximation 178 GROMOS 365
flux divergence 172 ground state 11, 14, 43, 56, 57, 67, 73,
flux function 172 74, 78, 80, 81, 83–86, 88, 98, 101,
Fock operator 211, 215, 217 104–106, 161, 162, 212, 214, 219,
force boundary 164 220, 228, 229, 235, 258–260,
force constants 5, 146 264–266, 276, 279, 285–287, 289,
force densities 164 315, 326, 335
force fields 114, 145–148, 269–272, group characters 56
363, 365 group symmetry 2 11, 56, 180, 183,
formally exact expression 232, 236, 185, 196, 201, 310
294, 346
Fourier components 76–78, 133, 280, h
284–287, 311, 330 H2 55, 59, 165, 170, 179,
Fourier density component 76 198, 302
Fourier transform 40, 44, 47, 51, 76, H2+ 10
78, 138–140, 142, 235, 280, 288, (H2O)2 29, 227
320, 321, 324, 329 (H2O)N 227
fractional electron transfer 362 Hamiltonian operator 9, 10, 28, 36, 40,
fragment adsorption energy 314 64, 68, 89, 102, 103, 144, 148, 159,
fragment charge 146 160, 168, 169, 181, 187, 234, 256,
fragment-fragment interactions 144 258, 271, 293, 341
fragment functionals 299, 348 H2+ + Ar 15
fragment polarizability 146 harmonic field 47
frame change 53 H atom-dielectric inetraction 326
Frank–Condon overlaps 180 H atom-dipole layer interaction 334
free energy 21, 24–25 HCN 120, 121
full-CI expansion 213 HCN H + CN 15
full-CI wavefunction 210 HCN HNC 15
 Index 379

HD 165 H2 on Ni(001) 360

H2 + D2 126 H2/Pd(100) 149, 151
He+ 11 H2/(2×2)S/Pd(100) 302
He2 198, 243 Hund’s cases 247
(He)2 227 hybrid exchange-correlation 240
heavy atoms 102, 244, 246 hybrid functionals 229, 231, 240,
HeBe2+ 243 241, 346
He + H+ 100 hydrocarbon chains 281–291
He + H2 14, 122 hydrogen atom continuum 86
He + He 86, 198, 212 hydrogen bonding 124, 146, 364
Heitler–London–Wang hydrogenic molecules 165–166
wavefunction 99 hydrogenic orbitals 96, 206
He + LiF(001) 17 hyperpolarizability 22, 42, 43, 48
He-LiF(001) 330
Hellmann–Feynman theorem i
3, 166–168, 232 image charges 324–326, 338, 354
He+Ne 79 image dipole 327
He on Cu(001) 337 imaginary-value frequencies 82
HF 59, 80, 81, 211, 212, 214, 217, 221, inactive orbitals 212
227, 239–242 indentation energy 314
H + H 86, 98–100, 198, 212 indented surface structure 314, 351
H + H+ 11, 12, 96, 98, 99 induced multipole 24, 68–78
H+ + H− 99 induction energy 8, 24, 67, 74, 79, 85,
H(1s) + H(1s) 11, 215 93, 104, 231, 259, 262–264, 267,
H+ + H2 15, 187 280, 282, 311, 331
H + H2+ 187 inflection point 8, 115, 116, 118, 119
H + HBr 301 inhomogeneous fields 35
H + He 87, 93 inhomogeneous gas gap 296
HHe+ 165 integral Hellmann–Feynman
H2 H + H 214, 221 theorem 168
H + H2 H2 + H 14, 120, 186 interacting excitons 290
hidden layer 150, 151 intermediate normalization 66, 202,
H + Li 212 216, 218, 225, 226
HNO(1Δ) 185 intermediate-range 6, 90, 93, 228,
H(2S) + NO(2Π) 185 329–332, 346, 347
H2O 14, 22, 55–57, 59, 60, 81, 120, internal energy 19, 24–26, 127, 130,
123, 135, 352 132, 134, 135, 148, 266, 336
Hohenberg–Kohn decomposition 162 internal variables 137, 139, 145, 179,
Hohenberg–Kohn theorem 88 181, 184, 187, 230, 291
H2O–H2O 123 interpolation polarizability 328
homogeneous structure 142 intra-monomer corrections 226
H2 on Cu(001) 337, 360 intruder states 198
H2 on Cu(001), Cu(111) 358 invariant form 191
380 Index

inverse function representation 278 local electric field 25, 130

inverse powers expansion 69, 73, 77 local excitation frequency 296
irreducible representation 55, 56, 183, local gauge transformation 191
201, 224 localized harmonic functions 277
isomerization 15, 17, 19, 148, 185, 189 local parametrization 117–119
iterative equations 46 local plasmon frequency 297, 342
local polarizability(ies) 237, 281, 296
j London–Eyring–Polanyi–Sato (LEPS)
Jahn–Teller theorem 187 function 126, 186, 353, 362
jellium model 316, 317, 338, 344, long-range 6
345, 358 long-range correction 241
Lorentz field 130
k Lorentzian distribution 57
Keesom formula 27 lower bounds 59–60, 83–86, 263
Kohn–Sham orbitals 90, 315
Kohn–Sham procedure 161, 162 m
Kohn–Sham spin-orbitals 163 machine learning (ML) 148–152, 300,
Kuhn–Thomas sum rule 82 365, 367
Madelung constant 129
l M06 and MN12 231
laboratory frame 52 many-atom decomposition 114
large spin components 244 many-atom method 196, 222–228
lattice energy 127, 293 many-body dispersion (MBD)
lattice sum 128 energies 293, 294
layer nodes 150, 151 many-body energies 106
Lennard Jones (n,6) 115 many-body perturbation 198
Lennard–Jones n-6 potential 9 Many-body perturbation theory
Li+ 17, 18, 20, 59, 60, 78, 331 (MBPT) 215–218, 221, 244
LiF 11, 14, 182, 330 MARTINI 365
Li + F 11 mass polarization 52, 102, 103
LiFH 303, 308 mass weighted momentum 159
LiH 165 Maxwell field 130
Li++HF 80–81 mechanical embedding 271
Li+-Ne 78 metal dielectric function 321
linear combinations of atomic orbitals metal surface charges 322
(LCAOs) 96, 100, 113, 165, 198, methane dimer 297, 298
205, 285 micelles 363
linked diagrams 101, 216–218 Miller indices 310, 318, 337, 344,
lipid membranes 363 360–362
liquid cohesive energy 132 minimum point 115
liquid polarization 135 2m multipole 72, 78
liquid solution energy 136 mobile adsorbate 336
local dielectric function 324 molecular adsorption 312
 Index 381

molecular beams 2, 27, 30–31, 113 Na(nl) + H(1s) 176

molecular crystal 128, 231, 239, 241, Na + Na 91
255, 289 Na on Al(001) 346
molecular fragment 38, 39, 114, natural spin-orbitals (NSOs) 205
143–144, 146, 213, 261, 265, Ne 59, 127
270–272, 275, 278, 279, 284, 298 Ne2 243
molecular orbital formed as a linear (Ne)2 227
combination of atomic orbitals Ne–Ar 92
(MO-LCAO) 94, 99, 166 Ne + Na 91
molecule-dielectric dispersion Ne + Ne 90
coefficient 328 network architecture 150
molecule-field interaction 36, 41, 43 network training 149, 301
molecule-metal one-electron neural networks (NN) 149, 151,
states 339 301–303
molecule-metal physisorption 337 training 300
molecule–surface dispersion N2-NaF(s) interaction 335
energy 327, 335 noble gas atom on graphite 345
Moller-Plesset PT (MPPT) 217, 219 noble gas on TiO2(110) 345
momentum coupling 173–176, 178, no-dispersion molecule-surface
180, 183, 187–190, 303 interaction 330
momentum-coupling matrix 171, non-adiabatic processes 303
173, 175 non-adiabatic
Morse function 125, 362 representations 174–175
Morse(α) potential 8, 115, 116, 343 non-crossing rule 182
moving nuclei 169–172, 191, 244 non-local correlation 232, 346
MR-CC, 221 nuclear charge density 257
MRD-CI procedure 215 nuclear kinetic energy 5, 6, 9, 10
multiconfiguration SCF (MCSCF) 214 null phase function 190
multipolar tensor 72
multipole components 37, 53, 56, 57, o
80, 267, 285 O atoms on Ni(111), Ni(001) 358
multipole–multipole occupied orbitals 87, 88, 211, 212, 239,
interaction 23, 122 315, 339
multireference projection one-electron integral 94, 285, 287
operator 202, 221, 225 one-electron-metal interaction 339
multiscale-CG (MS-CG) 366 one-to-one correspondence 161
O2 on Al(111) 360
n optimized effective potential
N2 60, 219–221, 352, 360 (OEP) 240, 242, 299, 362
NaCl 119, 127, 128, 151, 320 optimized embedding effective potential
Na+Cl− 116 (OEEP) 300, 362
NaH 176, 178, 182 orbital functional 11, 242
Na+ + H− 176 orbital hybridization 97
382 Index

orbital polarization 97 tensor 49

orthogonality property 55, 200 volume 22
oscillator strength 28, 58, 82, 86, 342 polarizable force fields 148, 271
overlap integral 94, 125, 126, 223, polarization 24–26, 39, 45–48, 50–56,
277, 355 59, 74, 81, 88, 90, 93–98, 100, 101,
130, 134–137, 143, 144, 146, 148,
p 206, 207, 222–224, 231, 239, 255,
pair correlation function 26, 138 260–262, 269, 270, 281, 292, 320,
pair distribution function 131–139, 321, 330, 331, 343, 344, 363
141, 363 polarization orbitals 207
pair potential energies 6, 132, 136, 140 polarized force fields 272, 363
particle-in-a-box states 285 polypeptide chain 15, 16, 270
particles distribution 366 post-symmetrization 201
partitioned densities 272, 293 potential of mean force 26–27, 134,
partition function 25, 366 137, 139, 366
partitioning treatment 355, 357 primitive gaussians 206
path geometric phase 191 products of Gaussians 278
Pauli exclusion 87, 88, 117, 342 projection operator 55, 59, 85, 98,
Pauli pressure 10 199–202, 210, 212, 214, 215, 221,
Pb2 244 224, 225, 261, 263, 356
periodic boundary conditions 40, 262,
279, 290, 316, 345, 347 q
periodic lattice potential 315 quadrupole distributions 70
perturbation theory 42, 64–68, 86, 98, quadrupole tensor 37
101–104, 132, 134, 139, 144, 164, quantal dielectric response 327
180, 187, 219, 256, 292, 340 quantum mechanics/molecular
perturbation treatment 7, 41, 47, 50, mechanics (QM/MM)
68, 73, 102, 103, 217, 226, 344 treatment 143, 270–272
phase function factor 190
phospholipid molecules 363, 364 r
photoisomerization 303 reaction potential 265–267, 270, 353
physical continuity 112 receptor–ligand scoring 303
physisorption 310 reciprocal lattice vectors 149, 310,
jellium model 338 311, 315
subtraction procedure 347 reciprocal space 39, 40, 142, 206, 208,
piecewise potential 119, 338 279–281, 310, 315
pi electrons 284, 285, 287 reciprocal space grid 279
pi-MO-LCAOs 285, 288 recursion procedure 66
plane-wave expansion 279 reduced atomic density 131
plasmon 296, 322, 324 reduced density functions 203
plasmon frequency 297, 323, 342 reduced density matrix (RDM) 205
point symmetry groups 54, 183 reduced resolvent 356
polarizability 22 reduced resolvent operator 202
 Index 383

reduced variables 115 228, 241, 265, 269, 277, 293, 311,
reference frame 6, 13, 43, 52–54, 64, 312, 331, 347, 350
76, 79, 80, 94, 111, 122, 131, 139, Si(s) 127
158–160, 169, 230, 283, 291, Si structures 151
353, 360 size consistency 101, 112, 196, 222
refractive index 26 size extensivity 196, 198, 217, 218
relativistic Hamiltonian 103 Slater determinant 99
renormalized susceptibility 328 Slater-type orbitals (STOs) 206
representation transformation 174 smooth coupling 175
resolvent operator 41, 50, 83, 104, smooth diabatic potentials 176
202, 356 solid Si 315
resonance energy 68, 74 solute pair correlation function 138
resonance interaction 68, 98 solute pair distribution function 137
resonance peak 49 solute–solvent interactions 265–268
response function 44, 46, 50, 51, 57, space-fixed (CMN-SF) frame 79, 158
235, 236, 243, 256, 319, 320 space groups 54, 310
restricted Hartree–Fock (RHF) 100, space symmetry 200
198, 211–213, 219 space symmetry projection 201
restricted open shell HF (ROHF) 212 spatial and spin symmetry 112
retardation effects 8, 102–103, spatial correlation function 234
327, 328 special points 18, 115, 181, 182
retarded susceptibility 46, 47, 51, 319 spectroscopy 28–30, 113, 116, 145,
rotation matrices 53 170, 175, 176, 179, 189, 246
rotation matrix 80, 158 spherical components 37, 38, 43, 53,
Rydberg potential 116 57, 70, 75–76, 78, 79
spherical multipoles 37, 54, 57, 69, 75
s spherical waves 77
saddle point 14, 15, 18, 21, 121, spin couplings 9
126, 313 spin density matrix 229
scalar potential 40, 191, 265 spin-dependent potential 126
scaled atomic polarizability 293 spin-orbit coupling 10, 78, 102–103,
scaled dynamical susceptibility 235 183, 196, 243–247
scaled response function 235 spin-other-orbit couplings 244
scaling in CCSD(T) 221, 222 spin-projection rising and
screened electron–electron lowering 200
interaction 241 spin recoupling 246
second-order energy 67, 83, 217 spin-space projection operator 200
selection rules 54, 55, 57, 68, 74, 76, spin states 90, 101, 126, 166, 183,
78–80, 84 196, 340
SH2 185 spline function 119
short-range (SR) 6, 7, 27, 69, 91, 101, spontaneous emission 49
103, 114, 117, 118, 120, 122, 124, spread radius rs 316, 345, 358, 359
129, 132, 133, 136, 143, 144, 146, static susceptibility 262, 264, 280
384 Index

stationary molecular state 10, 41 symmetry-adapted projector 202

steady state 2, 7, 44, 47 symmetry elements 54
stimulated absorption-emission 49 symmetry movements 54–56, 201
strength parameter 42 symmetry operators 54, 55
strictly diabatic
representation 174–176 t
structure descriptors 303 technological applications 31
structure factor 133, 134, 138, 139 Tersoff function 129
subtraction scheme 271, 298, 313, thermodynamics properties 30
348, 349 third-order dispersion energy 106
sum over atom pairs 292 three atom terms 114
sum over cells 318, 332 three-body dispersion energy 104
sum rule 58–59 three-body operators 104
supermolecule AB 196 three-body systems 103–106, 227
supermolecule method 196, 198, three identical nuclei 190
209–222 tight-binding approach 315
surface charges 130, 321, 322, 324, time-dependent state 44, 45
326, 327 time-evolution operator 44, 320
surface crossing point 355 TIP5P model 122
surface dielectric function 320 torsion angles 15, 16, 114, 145, 147,
surface dipole energy 318 158, 364
surface dipole layer 311, 325, 333 transformation angle 175, 176
surface electrical potential 325 transition density 260, 261
surface electronic structure 315, 317 transition density matrices 203
surface location 311, 326, 329, 360 transition force 180
surface periodicity 20 transition multipole 56, 75
surface plasmon frequency 323 transport rates 28
surface-projectile sum 331 transposition operator 211, 216, 225
surface properties 318 2D-conical intersection 184
surface reactions 353 bound-to-bound, bound-to-
surface response 319 unbound, unbound-to-
surface states 317 unbound 185
surface susceptibilities 319 2×2 transformation 175
surface translations 310
surface van der Waals potential 142 u
surface vectors 142 unbound nuclei 4–6
susceptibility upper bound 264 unique functional theorem 160
symmetric anisotropy 120 unrestricted HF (UHF) 198, 212, 227
symmetric group 224 upper and lower bounds 59–60,
symmetry-adapted perturbation theory 83–86, 263
(SAPT) 101, 198, 225–228, 230, second order energy 85
231, 244, 268, 291, 344 upper bounds 5, 6, 59, 86, 161, 162,
HF, UHF, DFT 227 201, 264
 Index 385

v vinyl bromide 151, 301

valence-bond (VB) 101, 125, 170, 223, virial theorem 166–169
353, 362 virtual orbitals 212
valence-bond (VB) function 98, 186 VV09/10 treatment 297
validation set 150
Van der Waals complex 28–29, 115 w
van der Waals equation 27, 134 wave operator 202, 210, 213, 216–218,
variable parameter 121, 124 220, 221, 225
variational functional 86, 99, 166 weak anisotropy 120
variational parameters 94, 97, 100, Wigner rotational matrix 53, 76, 79
165, 166 work function 318, 350, 351, 354
variational treatment 7 work functions W(h, k, l) 318
vdW-DF, 232, 237, 298, 346, 350, 353
vector potential 40 x
vibrational frequencies 5, 238, Xe on Cu(111) 346
239, 244 X-Pol force field 269
vibronic coupling 180, 187
vibronic description 179 z
vibronic displacement 180 zero-point energy 128, 337
 WILEY END USER LICENSE AGREEMENT
Go to www.wiley.com/go/eula to access Wiley’s ebook EULA.