Molecular Electrostatic Potentials - Read Online

Book Preview

Molecular Electrostatic Potentials

You've reached the end of this preview. Sign up to read more!
Page 1 of 1


Molecular Electrostatic Potentials

Concepts and Applications

Jane S. Murray

Department of Chemistry, University of New Orleans, New Orleans, LA 70148, USA

Kalidas Sen

School of Chemistry, University of Hyderabad, Hyderabad 500 046, India

ISSN  1380-7323

Volume 3 • Number (C) • 1996


Table of Contents

Cover image

Title page

Theoretical and Computational Chemistry

Series Editors

Volume 1

Volume 2

Volume 3

Front Matter

Molecular Electrostatic Potentials

Copyright page



MEP: a tool for interpretation and prediction. From molecular structure to solvation effects

1 Introduction

2 Thirty Years Ago: The Evolution of Chemical Quantum Theory

3 The Molecular Electrostatic Potential as an Interpretative Tool for Intermolecular Interactions

4 Intermolecular Energy: A Full Decomposition At HF Level

5 Molecular Electrostatics And Semiclassical Approximation In Solvation Effects


Molecular Electrostatic Potentials from Density Functional Theory


2 Calculation of Electrostatic Observables

3 Simplified Analytic Expressions for the Molecular Electrostatic Potential

4 Critical Points of the Molecular Electrostatic Potential

5 Evolution of the Molecular Electrostatic Potential During Chemical Reactions


7 Acknowledgement

The Use of Electrostatic Potential Fields in QSAR and QSPR

1 Introduction


3 EP-based 3D QSAR/QSPR approaches


Generalization of the Molecular Electrostatic Potential for the Study of Noncovalent interactions

1 Introduction

2 Introduction Of Environment Effects in the Mep

3 Introduction of Non-Electrostatic Energy Terms in the MEP

4 Future Directions

4 Acknowledgements

Molecular Recognition via Electrostatic Potential Topography


2 Models for Weak Intermolecular Interactions

3 Topography of Molecular Scalar Fields

3.1 Topography of Molecular electron density and Electrostatic potential

4 Topography–Based Molecular Interaction Model



Molecular electrostatic potentials and fields: hydrogen bonding, recognition, reactivity and modelling



3 Charge Distribution

4 Representation of MEP and MEF

5 Reactivity, Hydrogen Bonding and Other Properties

6 Recognition and Modelling


Molecular electrostatic potentials for large systems


2 Reactivity Concepts

3 Calculation of Cumulative Atomic Multipole Moments in Sindo1

4 Calculation of the MESP With Sindo1

5 The Molecular Surface

6 Silicon Clusters

7 Solid Silicon



Protein electrostatics

1 Introduction

2 Methodology

3 Applications



The Lorentz-Debye-Sack theory and dielectric screening of electrostatic effects in proteins and nucleic acids

1 Introduction

2 Lorentz-Debye-Sack Theory of Polar Molecules and Radial Permittivity Profiles

3 Electrostatic Screening in Macromolecular Systems

4 Application of Electrostatic Screening to the Calculation of Equilibrium Properties

5 Electrostatic screening in molecular dynamics and monte carlo simulations

6 Conclusions


Modelling Intrinsic Basicities: The Use of the Electrostatic Potentials and the Atoms-in-Molecules Theory



3 Carbonyl vs. Thiocarbonyl Compounds

4 Three-Membered Rings

5 Concluding Remarks


Computed electrostatic potentials in molecules, clusters, solids and biosystems containing transition metals



3 Electrostatic Potentials as a Reactivity Index for Complex Ions and Molecules

4 Polar Molecules in Solution

5 Experimental Electrostatic Potentials

6 Conclusion

Studies on the molecular electrostatic potential inside the microporous material and its relevance to their catalytic activity

1 Introduction


3 Applications

4 Conclusions

X-ray diffraction and the potential distribution in crystals



3 Multipole Analysis

4 Partitioning

5 Data Collection


7 Comparison with Theory


Molecular Electrostatic Potentials vs. DFT descriptors of reactivity


2 DFT Based Reactivity Descriptors: Conceptual and Methodological Issues

3 Results and Discussion

4 Conclusions

5 Acknowledgements

Electrostatic Potential, Bond Density and Bond Order in Molecules and Clusters

1 Introduction

2 Electrostatic Potential at the Nucleus of a Neutral Atom Related to Electronic Correlation Energies of Atomic Ions

3 Chemical Potential and Bond Mid-Point Properties

4 Cluster Properties

5 Dissociation of Doubly-Charged Clusters: Study of Supermolecular Ions AND

6 Bond Density and Chemical Network Model

7 Correlation Energy and Electron Density


Relationships of Electrostatic Potentials to Intrinsic Molecular Properties

1 Introduction

2 Atomic and Molecular Energies as Functions of Electrostatic Potentials at Nuclei

3 Electrostatic Potentials and Chemical Potentials

4 Topographical Analyses

5 Lattice Energies and Ionic Radii

6 Covalent Radii and Bond Dissociation Energies

7 Electronic Densities and Electrostatic Potentials

























Theoretical and Computational Chemistry

Series Editors

Professor P. Politzer

Department of Chemistry

University of New Orleans

New Orleans, LA 70418, U.S.A.

Professor Z.B. Maksi

Ruder Boskovi Institute

P.O. Box 1016,

10001 Zagreb, Croatia

Volume 1

Quantative Treatments of Solute/Solvent Interactions

P. Politzer and J.S. Murray (Editors)

Volume 2

Modern Density Functional Theory: A Tool for Chemistry

J.M. Seminari0 and P. Politzer (Editors)

Volume 3

Molecular Electrostatic Potentials: Concepts and Applications

J.S. Murray and K. Sen (Editors)

Front Matter

Molecular Electrostatic Potentials

Concepts and Applications

Edited by

Jane S. Murray

Department of Chemistry

University of New Orleans

New Orleans, LA 70148, USA

Kalidas Sen

School of Chemistry

University of Hyderabad

Hyderabad 500 046, India



Amsterdam – Lausanne – New York – Oxford – Shannon – Tokyo

Copyright page


Sara Burgerhartstraat 25

P.O. Box 211, 1000 AE Amsterdam, The Netherlands

ISBN: 0 444 82353 0

© 1996 Elsevier Science B.V. 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, without the prior written permission of the publisher, Elsevier Science B.V., Copyright & Permissions Department, P.O. Box 521, 1000 AM Amsterdam, The Netherlands.

Special regulations for readers in the U.S.A. – This publication has been registered with the Copyright Clearance Center Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923. Information can be obtained from the CCC about conditions under which photocopies of parts of this publication may be made in the U.S.A. All other copyright questions, including photocopying outside of the U.S.A., should be referred to the copyright owner, Elsevier Science B.V., unless otherwise specified.

No responsibility is assumed by the publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein.

This book is printed on acid-free paper.

Printed in The Netherlands.


Over the past twenty–five years, the molecular electrostatic potential has become firmly established as an effective guide to molecular interactions. With the recent advances in computational technology, it is currently being applied to a variety of important chemical and biological systems. Its range of applicability has expanded from primarily a focus on sites for electrophilic and nucleophilic attack to now include solvent effects, studies of zeolite, molecular cluster and crystal behavior, and the correlation and prediction of a wide range of macroscopic properties. In addition, the growing prominence of density functional theory has brought the molecular electrostatic potential to a new prominence on a more fundamental conceptual level. It is rigorously defined in terms of the electron density, and has very interesting topological characteristics since it explicitly reflects opposing contributions from the nuclei and the electrons.

This volume begins with a survey chapter by one of the original pioneers of the use of the electrostatic potential in studies of chemical reactivity, Jacopo Tomasi. Though the flow of the chapters that follow is not stringently defined, the overall trend is that the emphasis changes gradually from methodology to applications. Chapters discussing more theoretical topics are placed near the end. Our hope is that the reader will find the wide variety of topics provided by an international group of authors to be useful and satisfying.


We greatly appreciate the very efficient assistance of Ms. Elizabeth Blankenship and Ms. Anita Buckel in putting together this volume, and especially in the tedious task of preparing the index.

MEP: a tool for interpretation and prediction. From molecular structure to solvation effects

J. Tomasia and B. Mennuccia,     aDipartimento di Chimica e Chimica Industriale, University of Pisa, Via Risorgimento 35, 56100 Pisa, Italy

R. Cammib,     bDipartimento di Chimica, University of Parma, Viale delle Scienze 1, 43100 Parma, Italy

1 Introduction

This paper aims at giving a partial view of the evolution of the research in the definition, and use, of semiclassical descriptions performed at Pisa over the years. We stress the emphasis on the adjective partial as, in writing this paper, we realized that it would have been too long to consider also the part of the research addressed to apply the semiclassical approach to the description of the internal structure of molecules.

We have thus limited our attention to molecular interactions. Even with this limitation, the exposition is quite long, as it seemed us convenient to insert what we have done, and we are still doing, in a broader framework, including also critical elaboration of methods and proposals suggested by other groups.

Our effort has thus assumed the aspect of an essay, where space has been given to methodological considerations, to the exposition of the motivations which led us, and other people, to do what has been done, and to the identification of the mutual interplay in the research programmes of various groups.

We hope to have been able to give a living picture of the evolution of scientific research on a specific theme. We also hope that this presentation will encourage young people to contribute further to this evolution. The three authors are of very different age, and the historical perspective given at some points, mainly reflects the experience of the elder (J.T.). However the paper has been collectively written, and reflects the working style (if it is possible to use this word) in our group, where both methodological considerations and critical appraisal of the evolution of the methodologies are subject to continuous discussions.

The paper is organized in four main sections:

1. Thirty years ago: the evolution of chemical quantum theory.

2. The molecular electrostatic potential as an interpretative tool for intermolecular interactions.

3. Intermolecular energy: a full decomposition at HF level.

4. Molecular electrostatics and semiclassical approximation in solvation effects.

In the first section we shall briefly consider the problems which Quantum Chemistry had to face at the beginning of the computational era (1959-1969) to put in the most appropriate contest the proposal of using the molecular electrostatic potential (MEP), which is the topic motivating this essay.

In the second section we shall analyze the properties and characteristics of the MEP as an interpretative tool for intermolecular interactions, in order to find a rationale of this function’s shape, and to analyze computationally convenient formulations to be used on larger molecular systems and for systematic applications. To this purpose we shall introduce a very important approach in the analysis of intermolecular interactions, namely to extract from the model the interactions which can be treated classically and to describe them at the most accurate level.

In the last two sections we shall describe some results of a systematic examination of the performances of this semiclassical model in dimers and in a specific class of many– molecules systems, i.e. the solutions, with the aim of putting in evidence the usefulness and the limits of this inherently approximate representation of chemical interactions.

2 Thirty Years Ago: The Evolution of Chemical Quantum Theory

In the early sixties the newly formed group of Theoretical Chemistry of the University of Pisa dedicated its main effort to elaborate methods and computer codes for ab initio quantum mechanical calculations, from the basic integrals (over Slater-type orbitals) to CI methods of various nature and complexity. The effort of several years of hard work, we think, was well rewarded: the set of combined computer programs computed there was one among the firsts, to be able to treat relatively complex molecules (say, with more than four atoms) within a large span of applications, from open and closed shell electronic structures at various levels of approximations, and with basis sets also including d functions, to a variety of properties, such as electric and magnetic response functions, electronic excitations, etc.

One of the fields considered for applications was the study of structure and properties of molecules of relatively large dimensions in their ground state. When the first extensive report, regarding several three-membered cyclic molecules [1], was submitted to the Journal of Chemical Physics, both the referees expressed, with different words, favourable comments about the problem of ‘what to do with wavefunctions computed with such a considerable effort’ and ‘the ingenuity of applying the results of quantum mechanical calculations to reactivity problems in a novel way’.

The problem of ‘what to do’ with molecular calculations is a basic problem in theoretical chemistry for which different solutions can be given, and in fact they have been given. It is a problem still alive nowadays, but of crucial importance at the beginning of the computer era. Coulson’s after-dinner speech at the banquet closing an important conference on molecular calculations (Boulder, 1959) is a very significant document still deserving consideration, 36 years after its publication [2]. With a plain and relaxing style Coulson raised serious questions about ‘the very nature of quantum chemistry: what relation it has with experiments, what functions we expect it to fulfil, what kind of questions we would like it to answer’ to quote Coulson’s words.

Coulson was worried by the perspective of a splitting of quantum chemistry into three separate domains, characterized by the ‘depth’ of the theory in use in each domain, and with little mutual interaction and cross-fertilization. People belonging to the first domain (group I) would have been addressed to the very accurate calculations of the wavefunction and of some properties of a given (small at that time) molecule, abandoning, to reach this goal, all ‘the normal natural conventional concepts of chemistry, such as bonds, orbitals, and overlapping hybrids’. Group II, on the contrary, would have been linked to these conventional and elementary concepts, and have considered not so important to compute a property accurately, nor that these concepts would have been expressed in a rigorous manner. The third domain (group III) would have regarded the ‘spreading’ of quantum chemistry into biology. ‘Biological systems are much more perverse than any laboratory chemical system’ Coulson says, and to reach the immense prizes this field promises people belonging to the latter group would have been willing to accept simplifications in the theory wholly mistrusted, or considered with horror, by people belonging to the first two groups.

This is the scenario which expresses Coulson’s worries, actually shared by many others, among which E. Scrocco who was just building up in Pisa the first Italian quantum chemistry laboratory.

Fortunately, the evolution of quantum chemistry has not followed this scenario. The impressive growth in this field, stimulated by the continuous and fast advances in computer technology, has not introduced a separation of the quantum chemistry community into different bodies, each with its own collection of concepts and methods, and with little communication among them.

The evolution of computational methods, which have exploited mathematical formulations not used in the early stages of quantum chemistry, has been accompanied both by a deepening of the formal foundations of the discipline and, at the same time, by a continuous reformulation of the ‘chemical concepts’ of which we have given few examples quoting Coulson’s words. These reformulations, and the elaboration of further concepts of similar nature, have been done taking into account both the deepening of the formal foundations and the advances in computational techniques.

At present, our scientific community can rely on an impressive number of tools, both of conceptual nature and of computational character, allowing to give very different answers to the problem we mentioned before: what to do with molecular calculations. Direct accurate property calculations of molecular systems of increasing complexity are now possible, often in competition with experimental methods. New Interpretations of phenomena of chemical and physical interest, based on models and ideas having a sound conceptual basis and on constitutive elements computed at the necessary level of accuracy, are currently proposed. Predictions of phenomena based on these interpretations, and not relying on complete direct calculations on the single molecular system have proved their validity in many cases and their use is increasing (also for the ‘perverse’ biological phenomena). The use of molecular calculations as an aid to visualize complex molecular systems and their properties is a current practice in many laboratories not directly addressed to theoretical studies.

We are drawing here a positive picture of the present state of theoretical and computational chemistry, not giving emphasis to limitations, which will be progressively reduced, we are confident, by the future evolution of the discipline. This evolution will be the result of the combined effort of the members of our community. It is important to remark, in this context, that the past evolution has not been ruled by chance and serendipity but by the conscious effort of a more restricted number of scientists aware of Coulson’s worries.

The paper we quoted at the beginning [1] represents the contribution given by Pisa’s group to this effort (to be more precise, one of the contributions given). In that paper we presented and used the molecular electrostatic potential (MEP), which is the topic motivating this essay, and the characterization of the molecular subunit properties in terms of localized orbitals (LO). The motivations giving origin to this selection of topics to supplement a set of ab initio molecular calculations (very hard to compute at that time, in fact they required us a couple of years) were not explicitly expressed in that paper. Some are evident if one looks at papers published by our group in the immediately following years or to a set of short notes we published and distributed in 1970 as technical reports where other applications of MEP and other quantities drawn from molecular calculations were considered [3].

Other points, mainly of methodological nature, were not purposely published, since in our group the prevailing opinion was that our main task was to study specific themes of quantum chemistry, and not to shift to methodological proposals not yet accompanied by evidence of their utility. That methodological framework has been widely tested in the following years and still continues to guide our efforts in the research. It has been partially sketched, or resumed, in more recent papers [4–9] but it is worthy resuming it here again, in order to give a better understanding of our attempts (motivations and results) and, perhaps, to stimulate colleagues to improve it and to use it to make the evolution of theoretical molecular methods more efficient.

We shall keep this short overview of methodological considerations at a low and colloquial level, without any pretention of formal presentation or completeness. No attempts will be done to merge in a unified view approaches and considerations suggested by other researchers, or to assess priorities for each single item.

In selecting and planning a specific research project it is convenient to evaluate if it is able to give a significant contribution to the understanding and interpretation of a specific chemical or physical phenomenon. To get this evaluation one has to explicitly express the model chosen for this study (in theoretical chemistry we are always going on working on models) and to ponder on quality and performance of the tools necessary, and available, to perform the research.

To give an example we may come back to the late sixties again. In those years the scientific community was stirred by news indicating that it was experimentally found that water could assume a state of aggregation unknown until then, with very peculiar properties, the so-called polywater [10]. At that time we were developing models and tools to study non-covalent molecular associations, and assemblies of water molecules were an important case to be considered. We examined the problem with care, and we decided that the available tools were not sufficient to give a meaningful answer to the question, neither to confirm or to refuse (they are different problems) modellistic descriptions of polywater; hence we did nothing on that subject. Later, it was shown with experimental techniques that polywater was a completely false concept, perhaps the most unfortunate incident in the history of molecular science of the last decades, and theoretical chemistry did nothing to dispel credence on this fictitious state of aggregation of water. The incident of anomalous water came too early: now theoretical models and tools would be adequate to properly treat this problem.

To state in an explicit way the characteristics of the model can be a difficult task unless appropriate guide-lines are used. One guide-line is to adopt a general dissection of models.

We shall not attempt here a general classification of models of possible use in chemistry (some remarks can be found in ref.[9], similar views have been expressed by Maksič [11] and before by Trindle [12] as we discovered later, but the literature on this subject is larger; we shall confine ourselves to quote a book [13]). For theoretical studies on molecular systems it is convenient to divide the model into three parts collecting the material, the physical, and the mathematical components, respectively (later we shall add a fourth component).

The material model (let us use this shortened denomination) is the portion of matter described by the model. It may correspond to a molecule, an assembly of molecules of a larger portion of matter described in a ‘realistic’ way by theoretical chemistry (e.g. by a wavefunction or a density function) or by means of a simplified formulation (e.g. by a set of local charges, or of coupled oscillators).

The physical model regards the physical interactions taken into account. They may be limited to the material model (for example, those enclosed in the usual time-dependent Hamiltonian, supplemented, or not, by spin-orbit coupling and other small terms) or include interactions with the exterior (for example an electro-magnetic interaction or some electrostatic interactions with other bodies). In some models quantum mechanical (QM) formulations are replaced by classical descriptions.

The mathematical model collects all the methods and approximations used to describe the physical interactions which are active in the model. The prescriptions of quantum calculations are to be considered here (QM theory level, basis set, evaluation of matrix elements, etc.). We shall include here the methods used in the manipulation of the results obtained when the model is applied.

Raw results obtained by putting into action the model machinery always need some manipulation. We may distinguish three aspects, or three phases, of this further elaboration; i.e. report, interpretation and explanation.

To shorten this digression we shall omit considerations on the report. It will suffice to remark that all our models are of computational nature, and that a task left to the report is to extract from the huge amount of numerical information provided by the computer the relatively small portion necessary for the following step, the interpretation. As a matter of fact an interpretation is always necessary.

Let us consider first the simplest case, namely the use of the model to get the value of some molecular observables. The latter are computed without any bar errors, since by repeating the calculation with the same prescriptions one obtains the same result. However, there are some bar errors in the calculation of the observables. To get them, one has to use the model in another way, by examining the critical points of the calculations, and by introducing some suitable changes (in the mathematical model first, as basis set or the QM theory level, but also, when necessary, in the physical and material components of the model) to quantify sensitivity of results with respect to the parameters one has chosen. This is a quite specific type of interpretation, often requiring reports conceived to this end.

The main motivation of using computational models is not the calculation of some observables in competition with experiment. What theoreticians are searching is to understand the phenomenon under examination, or in other words, to find an interpretation (or description) of it. This is the second phase in the manipulation of the numerical outcome provided by the model.

By the word interpretation we mean the search for relations between the aspects of the phenomenon brought into evidence by the report and a set of ‘chemical concepts’ which introduce a rationale in the numerical empirical evidence. This is the realm of ‘chemical theories’, which are less accurate than the underlying quantum mechanical general theory. In fact, the latter is too general to directly give the interpretation we are looking for, and we have to rely on chemical theories even if they are uncomplete and often in competition with each other. Interpretation in terms of chemical theories and concepts is strictly related to the definition and use of appropriate mathematical tools. We have introduced these tools in the mathematical part of the model, but it may be convenient to consider them separately as a fourth component of the model. This further decomposition can be useful when we are confident enough on the reliability (or pertinence, to use a different point of view) of the other components of the model.

For example, we know that a model which considers an isolated molecule with an appropriate QM formalism within the Born-Oppenheimer approximation, gives a reliable description of the static properties of the corresponding real molecule in low-pressure gas phase. In this case we may focus our attention to elaborate models in order to understand, and to predict, for example, chemical substitution effects. Here we have mentioned a possible outcome of interpretation, i.e. prediction. We may distinguish between two different meanings of the word prediction. The first one can be expressed in the following way: when a given physical and mathematical model has been applied with success to compute a property for some material systems for which the property is known, we may apply with some confidence the same model to other material systems for which the property is unknown. In the second meaning, the word prediction allows to extract from the model some key elements ruling the property, and the prediction is based on some suitable changes of these key factors.

A good interpretation will allow the elaboration of models which are simpler than those used at the initial stage of the study, but in the same time they still preserve the features of the ‘object’ (see ref.[9] for its definition) found to be essential to describe the property or the phenomenon according to the interpretation.

In the previous sentence we have added a proviso, to stress that an interpretation is not the final stage of a complete study.

Many interpretations of the same phenomenon can be given, based on different models (the differences may regard all the four parts of the model, or just one), and the different interpretations must be compared, contrasted and submitted to trials. Here falsification techniques can play an important role. The finding of counter examples spurs the search for more complete and satisfactory interpretations also via the elaboration of new models.

When we arrive at a satisfactory end on this critical comparison of models and interpretations, we shall be able to formulate an explanation of the phenomenon. This should take into account all pertaining known facts, predict hitherto unknown aspects, allow connections with other phenomena. In short it should be able, to add a new piece to the ‘growing building’ of our knowledge of the material world.

It turns out from these considerations that both the definition and elaboration of tools for the analysis of the numerical outcome of the model play a very important role. Progress in this field has allowed to fill the gap between the two approaches to quantum chemistry pointed out by Coulson [2]. At the same time, the above reported considerations should indicate that definitions and strategies for the use of these mathematical and conceptual tools must be chosen with care. Now we can come back and examine the reasons for our efforts in this field.

3 The Molecular Electrostatic Potential as an Interpretative Tool for Intermolecular Interactions

When the possibility of computing ab initio wavefunctions of relatively complex molecules passed from the realm of projects to that of reality, it was clear that the first step to take was to consider methods of analysis of the diagonal element of the first-order density matrix. A reduction of the information available from the moleculear QM model was necessary, but at that time, as many other people working in those years on this kind of models, we were quite generous in the report, and we stored additional information to be used later. QM molecular calculations were in fact terribly expensive.

In selecting and planning research projects we chose several themes (partly considered in ref. [3]) which can be collected into two groups, i.e. description and interpretation of the molecular charge distribution, and interactions with external agents such as electric fields, photons and other molecules. We cannot consider the evolution of all these themes, and in this essay we shall confine ourselves to intermolecular interactions, using, where needed, some elements drawn from the other themes, which have been pursued in parallel.

The layout of the theory for intermolecular interactions was set in the first years of quantum mechanics, completed and resumed by Margenau in a basic paper [14]. In the sixties much more material was available regarding both the theoretical and experimental aspects. The state-of-the-art at that time is described by some collections of papers [15, 16] and by some books [17, 18] which are still useful reference books.

However, the theoretical approach used in those years, essentially based on the perturbation theory and on multipole expansion of the interaction potential, was not convenient for our purposes. Our wish was to exploit to the maximum the effort made to get detailed information from QM ab initio variational methods on one, or a pair of, weak interacting molecules.

This wish was in line with a general trend, which was not so clear at that time although already in act, leading to an evolution from physical approaches of selected molecular problems to chemical approaches paying more attention to the fine details of the molecular model. Some comments on this general evolution, and some different examples, can be found in a recent review [19].

The solution we devised has some methodological interest. It consists in formulating a set of ‘working hypotheses’ to be checked point-by-point, and encompassing a wide range of subjects, among which intra- and inter-molecular interactions.

Later we defined the outcome of this set of hypotheses as ‘semiclassical approximation’.

The semiclassical approximation has been set out with some detail and numerous examples in previous occasions [5–8,20].

The basic idea was to extract from the model the interactions which can be treated classically, to describe them at the most accurate level, and to compare the results obtained by fully applying of the QM model.

The goals were to separate intrinsically quantum aspects from the classical ones, in order to see whether the latter were sufficient to describe (or to interpret) the phenomenon and then, if necessary, to find a description of the quantum effect, or to develop simple predictive models based on the classical terms. This meant to focus our attention on the interpretative step we have discussed in the previous section, to dedicate efforts towards more precise ‘chemical concepts’ based on classical interactions, and to develop a hierarchy of models in which both physical and mathematical components have an ascending, or descending, order of complexity.

In the case of intramolecular interactions to accept the semiclassical hypothesis means to consider classical electrostatic terms only. It is of course convenient to separate electrostatic interactions into rigid charge distributions and mutual polarization effects, and to consider rigid Coulombic interactions first.

When QM charge distribution functions became available, the only possible computational and interpretative tool which could allow the description of the details of the interaction was the Mulliken’s population analysis [21] and, in particular, atomic charges.

It is hard to compress in a couple of sentences a motivated appreciation of the very important role played by Mulliken’s population analysis in the development of theoretical chemistry. Suffice it to say that this clever and computationally simple method has been one of the most important tools for the interpretation of chemical facts, as described by QM models. It is a very nice example of reduction of the information coming from computational models we have stressed as an important point in the use of such models. However, Mulliken’s charges are not the most convenient tool to examine Coulombic intermolecular interactions. The reasons are well known. These charges are numerical indices defined in terms of some specific features of the mathematical model (expansion of molecular orbitals over atomic basis sets), and they suffer from limitations inherent in their definition. These aspects have been discussed many times, and there is no need to repeat them here. It is possible to modify these molecular indices; some modifications were already proposed in the sixties, others followed in the ensuing years. We have considered it convenient to discard such indices, passing to the direct use of the representation of the Coulombic interaction between molecules A and B given by the QM model.

and of the nuclear charge distribution function,

, the Coulombic (or electrostatic, ES) interaction energy term can be written as


where RAB indicates the parametric dependence of EES upon the mutual distance and orientation of A and B.

Equation (1) may also be written in the following form:




is the molecular electrostatic potential (MEP) of molecule A.

It was easy, even with the computer facilities available in the sixties, to encode this formal elaboration in QM models with a dimer A·B as material component. As a matter of fact the mathematical model requires only a small portion of the two-electron integrals–the real bottleneck at that time– necessary for the full computation of the dimer energy. Therefore it was possible to check with a minimal additional effort the validity of the approximation by reducing the interaction energy ΔE (A·B; RAB) = E (A·B; RAB)−EA EB to the electrostatic component alone, EES (A·B; RAB), coupling both ΔE and EES at the same level of accuracy.

does not depend on the position of B, but it is an intrinsic molecular property. It was quite easy to compute it with the codes we elaborated to compute two-electron integrals for the isolated A :


here χμ and χv belong to the basis set {χA } used to compute the wavefunction of A. Being


we have:




has as elements the Vμv given in eq. (4). Here it has been introduced the convention of using boldface characters to indicate matrices.

The expression of EES (A·B), which, we recall, is just one of the elements we were planning to include in our analysis, when given in terms of the MEP of one of the two parts, presents other advantages, some of which will be commented here, others later.

Assuming that the electrostatic hypothesis is giving acceptable predictions on some aspects of the intermolecular interactions (now, we know both its merits and its limitations), it would be easy to compute EES (A·B) with less expensive methods than the direct evaluation of eq. (1).


the calculation of EES (A·Bvalues, each with a weight qkB :


Moreover, the simmetry of EES allows to use the alternative expression:


hence we can have an estimate of the errors introduced with this approximation by comparing the two results.

With the use of approximation . Mulliken’s charges represent a solution for this problem, but we selected another solution which will be discussed later.

The use of approximations (9) and (10) makes easier the testing of the first step in our ‘working hypotheses’, which now we may state more precisely and give it the name we used when we expressed this concept for the first time: the electrostatic assumption [22]. This hypothesis assumes that EES alone is able to give sufficiently correct information about the relative energies of the most stable conformations of A·B dimer, about its geometry, and about the shape and deepness of the favoured incoming channel leading to the formulation of the different dimer conformations. The information we have thus defined can be drawn by the examination of a 6-dimension potential energy surface (PES), if the monomers are kept at a fixed internal geometry, to be computed point-by-point using the supermolecule E (A·B; RAB) energy.

It was, and still is, a quite demanding computational task. It is by far simpler to provisionally assume the hypothesis as right, to draw the 6-D PES within this approximation, and then to check the validity by selecting appropriate points in which the supermolecule energy E (A·B; RAB) has to be computed. This is another methodological point of general interest, namely to use information coming from the simplified model so as to have a general view of the phenomenon, and then select appropriate points for checking. Application of this method is possible when the simplified model is transparent, directly related to the original model, and ‘robust’ (for the definition of ‘robustness’, see Primas is a one-electron observable. The computational recipe we used in the first papers, based on the results of the first integration step of two-electron integrals, is not compulsory. Vμv (eq. 4) can be computed as a one-electron penetration integral. This fact made it possible, at a time in which ab initio is a scalar one-electron observable defined in the 3D space. This is a acceptable level of reduction of the information. For the analyses, the information should be neither too abundant nor too scarce. A good compromise is necessary.

Another electrostatic property we considered in the initial phase of the elaboration of models for the semiclassical hypothesis is the molecular electric field (MEF). This is another one-electron observable, which offers some advantages, or interesting alternatives, with respect to MEP; the first one is that it directly leads to the consideration of forces. However, MEF is a vectorial one-electron observable defined in the 3D space, and that makes it more difficult to visualize the function and to interpret its role in describing molecular interactions.

of eq. (3) is not expressed in terms of multipole expansions. This avoids all troubles about convergence and asimptoticity of these mathematical objects; since these requirements are met at distant point in the space and not in the proximity to the molecule itself. A straightforward use of multipolar expansion to describe A·B interactions of real chemical interest often leads, especially when the multipolar expansion is ‘accurate’, to disastrous results as we shall see later.

Once the MEP has been defined and found to be, by formal consideration and empirical evidence, a tool which promised a better understanding of intramolecular interaction, our research program on the semiclassical hypothesis branched off. One way was to use MEP as here defined and to continue by adding other classical terms in the description of the intermolecular interaction, while the other way was to analyze MEP more deeply in order to find a rationale of this function’s shape, and to search computationally less expensive formulations to be used on larger molecular systems and for systematic applications.

Similar bifurcations often appear in complex research programmes. In our case both aspects were important and the study was pursued on both directions in parallel. In setting out our research, as we are doing here, a sequential report is compulsory. We chose to treat the analysis of MEP first, leaving for a further section the report on a better assessment of the semiclassical intermolecular interaction model.

3.1 Simplified expressions for MEP from the global molecular wavefunction

In writing in terms of a set of point charges so as to give an accurate estimate of EES (A·B; RAB) It is possible to proceed a step further and to derive for both charge distributions, γA and γB, a point charge expansion which meets this requirement. If the two sets of point charges, {qlA } and {qkB }, are available, EES can be obtained from the sum of a finite number of products:


The most direct and convenient way to get {qlA } and {qkB , respectively (or on other mathematical techniques all having MEP as a reference function), the reason being given by the simmetry of the Coulomb energy term.

The first example we know of MEP fitting with point charges has been done in our laboratory [24]. A series of fittings, with different number of points, and with different additional constraints was performed by Alagona, when still an undergraduate student, and subsequently used in several studies, of which we shall quote here those using both eq. (9) and eq. (10) to determine EES (A·B; RABwith point charges was then extended to other molecules, often of sizeable dimensions, as the nucleic acid bases [28].

What we not used in these studies (all belonging to the early seventies) was the reduction of {qkX was suggested by Momany [29], and it has been a fortunate suggestion. The history of ‘potential derived’ atomic charges (PD–AC is one of the several acronyms in use) is well known and it will not be reported in detail here. A review on this topic is available [30], other reviews on charge distributions pay a considerable attention to this subject [20, 31].

The introduction of PD charges can be considered as a step of some importance in various approaches which involve mutual interactions among molecules. It would be too long to give a list here; we shall only quote their use in the development of the semiclassical interaction potentials used to assess molecular mechanics (MM) force fields for intramolecular interactions, and for molecule–molecule interaction potentials to be used in computer simulations on condensed state systems of various nature.

The introduction of atomic charges in MM force fields encountered some resistance. This is due in part to historical reasons. The first force fields were mostly based on experimental data and, in the sets of data to be fitted, the largest part regarded non-polar compounds (see the very readable and authoritative monograph by Burkert and Allinger [32], still recommended as an introduction to a fast-growing field, which depicts the situation since at the very first developments). Other reasons were more practical. Atomic charges were difficult to be derived and not very reliable. The charges obtained by fitting experimental data of molecules with heteroatoms were found to have a limited transferability (we can say that the model to get them was too raw, corresponding to the effort of condensing into an atomic index, i.e. the charge, effects of various nature, among which polarization effects). The charges derived from QM calculations were of limited reliability because of the low computational level in use at that time and of the method followed to get charges from the wavefunction (semiempirical charges [33] have been used where strictly necessary [34]).

Both historical and practical reasons merged in the formulation of a modellistic strategy which minimized the use of charges. This position had some consequences on the elaboration of two–body potentials for small molecules, for which, at that moment, the first relatively accurate point-by-point calculations of ΔE (A·B; RAB) were available. The derivation of the analytical expression of ΔE (A·B; RAB) was thus often based on purely numerical considerations on the fitting, aimed at obtaining the best fitting without any attempt to relate the terms of the analytical expansion to some physical models. The introduction of PD atomic charges allowed a large confidence about the description of Coulombic interactions, and now MM force fields and pair potentials for simulations are merged into a unified model, in which more space is given to the dissection of the interaction into physically acceptable contributions, and which also allows to combine both MM and computer simulations into a unique computational tool. Large perspectives have been thus opened to the computational description of systems in which large molecules are combined with small ones, such as the solutions of large molecules. One example of these perspectives is the use of combined QM/MM methods in which the effects of surrounding molecules (generally of semiclassical origin) are inserted in the Hamiltonian of the target molecule [35].

on these charges ranges from good to fair and even poor, according to the molecule. For some applications more detailed descriptions are thus necessary. They can be searched by resorting to multipole expansions or by increasing the number of charges. It is convenient to consider both approaches.

and its exact multipole expansion is called the penetration term:




, θ, ϕ refer to spherical coordinates centered on the expansion center).

has a correct asymptotic behaviour, that is when the number of terms in the expansion is truncated at a given lmax. This property could be considered as granted, but it is advisable to stress it here, as it is not granted for analogous multipole expansions regarding other terms of the molecular interaction, such as polarization and dispersion.

. In these cases the use of expansions with large lmax may give disastrous descriptions.

near chemical groups placed in the crevices of a corrugated molecule.

, can be obtained in many ways. One possible way, which is only an extension of the procedure followed to have PD charges, is to set the expansion centers on the nuclei, to fix the largest value of l coefficients. This has been done by several authors [given by atomic charges is not significantly corrected by atomic multipoles. To go further one is compelled to consider other expansion centers and to define a strategy to select their most efficient positioning.

To do it, let us consider, for simplicity’s sake, the electronic wavefunctions expressed as a combination of Gaussian functions which are the basis sets of larger use, and let us exploit some mathematical properties of theirs.

where the χi are Gaussian functions:


Rμv is the element of the one-electron density matrix, corresponds to two Gaussian functions of s type, having as exponent α μ and αv respectively, the distribution is reduced to a single s Gaussian function having α m = αμ + α v as exponent, to be centered at a point determined by the vector:


It is immediate to derive from these single Gaussian functions an exact multipole expansion composed by a finite number of terms, the upper of which is the sum of the angular momentum quantum number associated with the two original functions χ μ and χ v.

, given in terms of elementary Gaussian charge distributions can be thus replaced by a multicenter expansion which contains a finite number of elements.

when it was expanded on the basis set {χ }, with another analytical expression involving other functions, placed on other centers. This critical remark is apparently strenghtened by a closer examination of the expansion of elementary charge distribution. Let us confine to the analysis of wavefunctions expanded over atomic Gaussian basis sets. The situation is worse for other basis sets, also containing ‘bond’ functions, i.e. basis functions centered on the middle of chemical bonds, or ‘shifted’ functions, slightly displaced with respect to the nucleus, which, however, are of limited use. We may remark that given two atoms, A and B, the set of biatomic elementary distributions χμ (A)χv (B) will give origin to Gaussian functions centered in differents points, according to the possible different combinations of the orbital exponents αμ and αv. Moreover, the number of elementary distributions, and hence of the centers, is not limited to the number of contracted functions, but extended to the larger number of primitive functions. Therefore, to give a few examples, for H2O described by the STO-3G basis set there are 46 expansion centers, and 134 for the same molecule described by the 6-31G** basis set. During the same year of publication of the first paper on MEP we analyzed the charge distribution of adenine described with the minimal BS given by Clementi et al. [38]: the number of expansion centers was 2125.

Actually, things are not so bad as this analysis indicates. All the expansion centers lie on the nuclei (this is the center for χμ (A)χv (A) distributions) or on straight lines connecting atoms. If there are N atoms, there will be N (N − 1)/2 joining segments, on which available number of expansion centers are placed, and N vertices, also acting as expansion centers. Expansion centers connecting too distant atoms can be eliminated, the coefficients Rμv between two centers with a different position, using appropriate weights. These further expansions are no longer bound to some given l must not be too large, otherwise the expansion will contain some important contributions from high l centers and the maximum acceptable l centers would require a higher l max to have an accuracy comparable with the expressions related to a larger number of centers.

This analysis has been used by several groups. Vigné-Maeder and Claverie in Paris [39] presented the most complete analysis we know, examining 17 different contraction schemes. Among these schemes we found some which had had a previous formulation and many applications. We shall quote here the OMPT (overlap multipole truncated potential) a procedure elaborated by the same group in Paris, the first application of which dates back to 1973 [40]. The expansion terms are limited to one per segment, with lmax =2, plus the nuclei. The OMPT expansion was later simplified by limiting its expansion terms to segments corresponding to true bonds and to a point on the segment connecting two atoms A and C chemically linked to the same central atom B. This expansion has been considered by Vigné-Maeder and Claverie to be the best compromise between simplicity and accuracy. It has been used by Pullman’s group and by others in many studies on molecules of biological interest.

This approach has been explicitly used by Hall in a set of papers, starting from 1973 [42–49]. Among them we shall quote here both the analysis of Vpen (see eq. 12) and the proposal of adding some diffuse gaussian functions to point charges or to local multipoles, which have been reconsidered by other authors more recently.

has been carried out by Stone with his distributed multipole analysis (DMA). In the first formulation [50] the expansion centers are reduced to nuclei and midbound points; lmax is 2, i.e. local charges, dipoles and quadrupoles are considered. In another paper Stone and Alderton [51] somewhat tempered the picture given in terms of shrinked χμχv values (in previous pages we made a similar remark for Mulliken’s population).

It would be better to use a ‘physical space’ related to some one-electron properties, like MEP, since the available basis functions are combined in the most suitable way to maintain the observable almost constant. This is a judicious remark, supported by the checks made by Stone and Alderton. The problem lies in the definition of the partition of the ‘physical space’. It may happen that the selected partition gives origin to local multipole moments larger by orders of magnitude than those given by the usual DMA. This is a point which will be re-considered in the following pages.

Stone’s DMA has widely passed the stage of initial checks and now can be considered as a tool of permanent use in the study of molecular interactions. It has been implemented in several molecular computational packages of large use [51, 52]. A similar method has been implemented by Fernandez-Rico et al. [53].

The number of local multipole expansions available in literature is quite large and we cannot give here a critical and exhaustive report of all pertinent papers. We shall confine ourselves to recall here a couple of approaches, both widely supported by applications to which, for different reasons, we have paid particular attention.

First, let us quote Rein’s work: his atomic multipole expansion for semiempirical wavefunctions was the first to attract our attention in the years when we were developing our approach. The first stages of his research are documented in a review of 1973 [54], and later applied with several improvements to the study of numerous systems of biological interest [55]. Secondly, let us consider the CAMM (cumulative atomic molecular moments) procedure elaborated by Sokalski in 1983 [56], and later extended to correlated wavefunctions (CCAMM) [57], and reworked by introducing further expansion centers (the bond centers) and preserving at all orders the value of the global molecular moment [58]. We recall that multipole moment values depend on the choice of the expansion center position, a problem to which Sokalski has given his solution. Sokalski’s CAMM/CCAMM procedures have been adapted to a set of molecular computation packages of large use. A detailed description of these methods and results can be found in a recent review paper [59].

, is quite flexible and presents many points of interest. It has been recently adopted by Jug and coworkers [60] in an approach (called asymptotic density model, ADM) which aims at fitting exact MEP values both at large distance and at the nuclei. The ADM starts from the consideration of isolated atoms, for which an expansion into a spherical (l = 0) and a ‘polarization’ part (l = 1, m =− 1,0,1) is deduced. By inserting this partition into the Poisson’s equation:


in terms of CAMMs; the examples given in this first paper are quite convincing.

As said before, we should continue by mentioning other formulations. Their large number indicates the interest of this approach and at the same time how difficult it is to find a balance between accuracy and computational efficiency; i.e. a balance that shifts to different positions by changing the chemical problem.

This said, we shall pass now to consider the second option we have introduced, namely the use of a number of point charges higher than the number of atoms.

We have already introduced our earlier attempts to do it [22,24–27], but many other formulations have been given. We shall quote here, taking examples from an abundant literature, some models in which additional charges are addressed to describe local anisotropies of the charge distribution, putting them either in positions which intuitively correspond to the charge centers of the lone pairs, or along the bonds.

Good point charge models have been recently proposed by Luque and Orozco’s group [61]. They tested a very large set of models, for a series of small and medium-size molecules, selecting then a few models for each molecule: to give an example, 50 models for H2CO have been reduced to 4 with 4,5,7, and 8 point charges, respectively. From this study one may derive some indications about the strategy to be followed in