Molecular Modelling PRINCIPLES AND APPLICATIONS Second edition Andrew R. Leach Glaxo Wellcome Research and Development An imprint of Pearson Education Harlow, England London New York Reading, Massachusetts San Francisco Toronto Don Mill, Ontario Sydney Tolyo Singapore Hong Kong Seoul Taipet Cape Town Madrid Mexico City » Amsterdam - Munich Paris MilanPearson Education Limited Edinburgh Gate Harlow Essex CM20 2JE England and Associated Companies around the world Visit us on the World Wide Web at ww penrsoneduc com First published under the Longman imprint 1996 Second edition 2001 © Pearson Education Limited 1996, 2001 The right of Andrew R Leach to be identified as the author of this Work has been asserted by him in accordance with the Copyright, Designs and Patents Act 1988 All rights reserved, No part of this publication may be reproduced, stored ina retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise without either the prior written permission of the publisher or a licence permitting restricted copying in the United Kingdom issued by the Copyright Licensing Agency Ltd, 90 Tottenham Court Road, London W1P OLP. ISBN 0-582-38210-6 British Library Cataloguing-in-Publication Data ‘A catalogue record for this book can be obtained from the British Library Library of Congress Cataloging-in-Publication Data Leach, Andrew R. Molecular modelling principles and applications / Andrew R. Leach. - 2nd ed p.cm. Includes bibliographical references and index ISBN 0-582-38210-6 1 Molecular structure~Computer simulation 2 Molecules-Models-Computer simulation I. Title. (QD480.L43 2001 541 2'20113-de21 00-046480 987654321 05 04 03 02 01 Top right-hand cover image © American Institute of Physics ‘Typeset by 60 Printed in Great Britain by Henry Ling Ltd, at the Dorset Press, Dorchester, DorsetContents Preface to the Second Edition Preface to the First Edition Symbols and Physical Constants Acknowledgements 1 Useful Concepts in Molecular Modelling 11 Introduction 12 — Coordinate Systems 13. Potential Energy Surfaces 14 Molecular Graphics 15 Surfaces 1.6 Computer Hardware and Software 17 — Units of Length and Energy 18 The Molecular Modelling Literature 1.9 The Internet 110 Mathematical Concepts Further Reading References 2 An Introduction to Computational Quantum Mechanics Introduction One-electron Atoms Polyelectronic Atoms and Molecules Molecular Orbital Calculations ‘The Hartree-Fock Equations Basis Sets Calculating Molecular Properties Using ab initio Quantum Mechanics Approximate Molecular Orbital Theories Semi-empirical Methods Hiickel Theory Performance of Semi-empirical Methods Appendix 2.1 Some Common Acronyms Used in Computational Quantum Chemistry Further Reading References xiii xvii Soocwormauenen 24 24vill 3 Advanced ab initio Methods, Density Functional Theory and Solid-state Quantum Mechanics 3.1 Introduction 3.2 Open-shell Systems 3.3. Electron Correlation 3.4 Practical Considerations When Performing ab initio Calculations 3.5 Energy Component Analysis 3.6 Valence Bond Theories 3.7 Density Functional Theory 38 Quantum Mechanical Methods for Studying the Solid State 3.9 The Future Role of Quantum Mechanics: Theory and Experiment Working Together Appendix 3.1 Alternative Expression for a Wavefunction Satisfying Bloch’s Function Further Reading References 4 Empirical Force Field Models: Molecular Mechanics 423 Introduction Some General Features of Molecular Mechanics Force Fields Bond Stretching Angle Bending Torsional Terms Improper Torsions and Out-of-plane Bending Motions Cross Terms: Class 1, 2 and 3 Force Fields Introduction to Non-bonded Interactions Electrostatic Interactions Van der Waals Interactions Many-body Effects in Empirical Potentials Effective Pair Potentials Hydrogen Bonding in Molecular Mechanics Force Field Models for the Simulation of Liquid Water United Atom Force Fields and Reduced Representations Derivatives of the Molecular Mechanics Energy Function Calculating Thermodynamic Properties Using a Force Field Force Field Parametrisation Transferability of Force Field Parameters The Treatment of Delocalised Systems Force Fields for Inorganic Molecules Force Fields for Solid-state Systems Empirical Potentials for Metals and Semiconductors Appendix 4.1. The Interaction Between Two Drude Molecules Further Reading References Contents 160 161 161 162Contents 5 Energy Minimisation and Related Methods for Exploring the Energy Surface 5.10 Introduction Non-derivative Minimisation Methods Introduction to Derivative Minimisation Methods First-order Minimisation Methods Second Derivative Methods: The Newton-Raphson Method Quasi-Newton Methods Which Minimisation Method Should I Use? Applications of Energy Minimisation Determination of Transition Structures and Reaction Pathways Solid-state Systems: Lattice Statics and Lattice Dynamics Further Reading References 6 Computer Simulation Methods 6.9 Introduction Calculation of Simple Thermodynamic Properties Phase Space Practical Aspects of Computer Simulation Boundaries Monitoring the Equilibration ‘Truncating the Potential and the Minimum Image Convention Long-range Forces Analysing the Results of a Simulation and Estimating Errors Appendix 6.1 Basic Statistical Mechanics Appendix 6.2 Heat Capacity and Energy Fluctuations Appendix 6.3. The Real Gas Contribution to the Virial Appendix 6.4 Translating Particle Back into Central Box for Three Box Shapes Further Reading References 7 Molecular Dynamics Simulation Methods rae 72 73 74 Introduction Molecular Dynamics Using Simple Models Molecular Dynamics with Continuous Potentials Setting up and Running a Molecular Dynamics Simulation Constraint Dynamics Time-dependent Properties Molecular Dynamics at Constant Temperature and Pressure Incorporating Solvent Effects into Molecular Dynamics: Potentials of Mean Force and Stochastic Dynamics Conformational Changes from Molecular Dynamics Simulations Molecular Dynamics Simulations of Chain Amphiphiles 253 253 258 261 262 267 270 273 279 295 301 382 387 392 394Contents Appendix 7.1 Energy Conservation in Molecular Dynamics Further Reading References Monte Carlo Simulation Methods 813 Introduction Calculating Properties by Integration Some Theoretical Background to the Metropolis Method Implementation of the Metropolis Monte Carlo Method Monte Carlo Simulation of Molecules Models Used in Monte Carlo Simulations of Polymers ‘Biased’ Monte Carlo Methods Tackling the Problem of Quasi-ergodicity: J-walking and Multicanonical Monte Carlo Monte Carlo Sampling from Different Ensembles Calculating the Chemical Potential The Configurational Bias Monte Carlo Method Simulating Phase Equilibria by the Gibbs Ensemble Monte Carlo Method. Monte Carlo or Molecular Dynamics? Appendix 8.1. The Marsaglia Random Number Generator Further Reading References Conformational Analysis 91 Introduction 9.2. Systematic Methods for Exploring Conformational Space 9.3. Model-building Approaches 9.4 Random Search Methods 95 Distance Geometry 9.6 Exploring Conformational Space Using Simulation Methods 97 Which Conformational Search Method Should I Use? A Comparison of Different Approaches 98 Variations on the Standard Methods 99 — Finding the Global Energy Minimum: Evolutionary Algorithms and Simulated Annealing 910 Solving Protein Structures Using Restrained Molecular Dynamics and Simulated Annealing 9.11 Structural Databases 9.12 Molecular Fitting 913 Clustering Algorithms and Pattern Recognition Techniques 9.14 Reducing the Dimensionality of a Data Set 9.15 Covering Conformational Space: Poling, 9.16 A ‘Classic’ Optimisation Problem: Predicting Crystal Structures 405 406 406 410 457 457 465 467 475 476 477 479 483 489 490 491 497 499 501Contents 10 i Further Reading References Protein Structure Prediction, Sequence Analysis and Protein Folding 10.1 10.2 103 10.4 10.5 10.6 10.7 10.8 10.9 Introduction Some Basic Principles of Protein Structure First-principles Methods for Predicting Protein Structure Introduction to Comparative Modelling Sequence Alignment Constructing and Evaluating a Comparative Model Predicting Protein Structures by ‘Threading’ A Comparison of Protein Structure Prediction Methods: CASP Protein Folding and Unfolding Appendix 10.1 Some Common Abbreviations and Acronyms Used in Bioinformatics Appendix 10.2 Some of the Most Common Sequence and Structural Databases Used in Bioinformatics Appendix 10.3 Mutation Probability Matrix for 1 PAM. ‘Appendix 10.4 Mutation Probability Matrix for 250 PAM Further Reading References Four Challenges in Molecular Modelling: Free Energies, Solvation, Reactions and Solid-state Defects ae 11.2 11.3 114 115 116 ae neg rai 11.10 11.11 11.12 11.13 11.14 Free Energy Calculations The Calculation of Free Energy Differences Applications of Methods for Calculating Free Energy Differences The Calculation of Enthalpy and Entropy Differences Partitioning the Free Energy Potential Pitfalls with Free Energy Calculations Potentials of Mean Force Approximate/‘Rapid’ Free Energy Methods Continuum Representations of the Solvent The Electrostatic Contribution to the Free Energy of Solvation: The Born and Onsager Models Non-electrostatic Contributions to the Solvation Free Energy Very Simple Solvation Models Modelling Chemical Reactions Modelling Solid-state Defects Appendix 11.1 Calculating Free Energy Differences Using Thermodynamic Integration Appendix 11.2 Using the Slow Growth Method for Calculating Free Energy Differences xi 505 506 509 ae 517 522 522 539 545 547 549 553 555 557 557 558 563 563, 564 569 574 574 Bees: 580 585 592 593, 608 609 610 622 630 631xi Appendix 11.3 Expansion of Zwanzig Expression for the Free Energy Difference for the Linear Response Method Further Reading References 42. The Use of Molecular Modelling and Chemoinformatics to Discover and Design New Molecules 12.1 Molecular Modelling in Drug Discovery 12.2 Computer Representations of Molecules, Chemical Databases and 2D Substructure Searching 12.3 3D Database Searching 12.4 Deriving and Using Three-dimensional Pharmacophores 12.5 Sources of Data for 3D Databases 12.6 Molecular Docking 12.7 Applications of 3D Database Searching and Docking 128 Molecular Similarity and Similarity Searching 12.9 Molecular Descriptors 12.10 Selecting ‘Diverse’ Sets of Compounds 12.11 Structure-based De Novo Ligand Design 12.12 Quantitative Structure~Activity Relationships 12.13. Partial Least Squares 12.14 Combinatorial Libraries Further Reading References Index Contents 631 632 633Preface to the Second Edition The impetus for this second edition is a desire to include some of the new techniques that have emerged in recent years and also extend the scope of the book to cover certain areas that were under-represented (even neglected) in the first edition. In this second volume there are three topics that fall into the first category (density functional theory, bioinformatics/ protein structure analysis and chemoinformatics) and one main area in the second category (modelling of the solid state). In addition, of course, a new edition provides an opportunity to take a critical view of the text and to re-organise and update the material, Thus whilst much remains from the first edition, and this second book follows much the same path through the subject, readers familiar with the first edition will find some changes which I hope they will agree are for the better. As with the first edition we initially consider quantum mechanics, but this is now split into two chapters. Thus Chapter 2 provides an introduction to the ab initio and semi-empirical approaches together with some examples of the uses of quantum mechanics. Chapter 3 covers more advanced aspects of the ab initio approach, density functional theory and the particular problems of the solid state. Molecular mechanics is the subject of Chapter 4 and then in Chapter 5 we consider energy minimisation and other ‘static’ techniques. Chapters 6, 7 and 8 deal with the two main simulation methods (molecular dynamics and Monte Carlo). Chapter 9 is devoted to the conformational analysis of ‘small’ molecules but also includes some topics (e.g. cluster analysis, principal components analysis) that are widely used in informatics. In Chapter 10 the problems of protein structure prediction and protein folding are considered; this chapter also contains an introduction to some of the more widely used methods in bioinformatics. In Chapter 11 we draw upon material from the previous chapters in a discussion of free energy calculations, continuum solvent models, and methods for simulating chemical reactions and defects in solids. Finally, Chapter 12 is concerned with modelling and chemoinformatics techniques for discovering and designing new molecules, including database searching, docking, de novo design, quantitative structure-activity relationships and combinatorial library design. As in the first edition, the inexorable pace of change means that what is currently considered ‘cutting edge’ will soon become routine. The examples are thus chosen primarily because they illuminate the underlying theory rather than because they are the first application of a particular technique or are the most recent available. In a similar vein, it is impossible ina volume such as this to even attempt to cover everything and so there are undoubtedly areas which are under-represented. This is not intended to be a definitive historical account or a review of the current state-of-the-art. Thus, whilst I have tried to include many literature references it is possible that the invention of some technique may appear to be incorrectly attributed or a ‘classic’ application may be missing. A general guiding principle has beenxiv Preface to the Second Edition to focus on those techniques that are in widespread use rather than those which are the province of one particular research group. Despite these caveats I hope that the coverage is sufficient to provide a solid introduction to the main areas and also that those readers who are ‘experts’ will find something new to interest them. A Companion Web Site accompanies Molecular Modelling: Principles and Applications, Second Edition by Andrew Leach Visit the Molecular Modelling Companion Web Site at www.booksites. netfleach The website contains general information about the book, up-to-date hyperlinks to related chemistry sources on the web, reference copies of appendices of relevant acronyms, and twenty-six full screen, full-colour graphical representations of molecular structures.Preface to the First Edition Molecular modelling used to be restricted to a small number of scientists who had access to the necessary computer hardware and software. Its practitioners wrote their own programs, managed their own computer systems and mended them when they broke down. Today's computer workstations are much more powerful than the mainframe computers of even a few years ago and can be purchased relatively cheaply. It is no longer necessary for the modeller to write computer programs as software can be obtained from commercial soft- ware companies and academic laboratories, Molecular modelling can now be performed in any laboratory or classroom. This book is intended to provide an introduction to some of the techniques used in molecular modelling and computational chemistry, and to illustrate how these techniques can be used to study physical, chemical and biological phenomena. A major objective is to provide, in one volume, some of the theoretical background to the vast array of methods available to the molecular modeller. I also hope that the book will help the reader to select the most appropriate method for a problem and so make the most of his or her modelling hardware and software. Many modelling programs are extremely simple to use and are often supplied with seductive graphical interfaces, which obviously helps to make modelling techniques more accessible, but it can also be very easy to select a wholly inappropriate technique or method. Most molecular modelling studies involve three stages. In the first stage a model is selected to describe the intra- and inter-molecular interactions in the system. The two most common models that are used in molecular modelling are quantum mechanics and molecular mechanics. These models enable the energy of any arrangement of the atoms and molecules in the system to be calculated, and allow the modeller to determine how the energy of the system varies as the positions of the atoms and molecules change. The second stage of a molecular modelling study is the calculation itself, such as an energy minimisation, a molecular dynamics or Monte Carlo simulation, or a conformational search. Finally, the calculation must be analysed, not only to calculate properties but also to check that it has been performed properly. The book is organised so that some of the techniques discussed in later chapters refer to material discussed earlier, though I have tried to make each chapter as independent of the others as possible. Some readers may therefore be pleased to know that it is not essential to completely digest the chapters on quantum mechanics and molecular mechanics in order to read about methods for searching conformational space! Readers with experience in one or more areas may, of course, wish to be more selective. Ihave tried to provide as much of the underlying theory as seems appropriate to enable the reader to understand the fundamentals of each method. In doing so I have assumed some background knowledge of quantum mechanics, statistical mechanics, conformational analysis and mathematics. A reader with an undergraduate degree in chemistry shouldxvi Preface to the First Edition have covered this material, which should also be familiar to many undergraduates in the final year of their degree course. Full discussion can be found in the suggestions for further reading at the end of each chapter. | have also attempted to provide a reasonable selection of original references, though in a book of this scope it is obviously impossible to provide a comprehensive coverage of the literature. In this context, I apologise in advance if any tech- nique is inappropriately attributed. ‘The range of systems that can be considered in molecular modelling is extremely broad, from isolated molecules through simple atomic and molecular liquids to polymers, bio- logical macromolecules such as proteins and DNA and solids. Many of the techniques are illustrated with examples chosen to reflect the breadth of applications. It is inevitable that, for reasons of space, some techniques must be dealt with in a rudimentary fashion (or not at all), and that many interesting and important applications cannot be described. Molecular modelling is a rapidly developing discipline and has benefited from the dramatic improve- ments in computer hardware and software of recent years Calculations that were major undertakings only a few years ago can now be performed using personal computing facilities. Thus, examples used to indicate the ‘state of the art’ at the time of writing will invariably. be routine within a short time.Symbols and Physical Constants This list contains the most frequently used symbols and physical constants ordered according to approximate appearance in the text. » 146 ij,k 99,0 (x) or Lagrange multiplier spherical polar coordinates orthogonal unit vectors along x, y, z axes Euler angles arithmetic mean value of x unit matrix square root of —1 unit vector exponent in Gaussian function (normal distribution) standard deviation variance Planck’s constant (6.626 18 x 10“ Js) h/2m (1.05459 x 10-*Js) particle mass molecular wavefunction, 0 [0x + 8 /ay + &/82 (‘del-squared’) Hamiltonian spatial orbital spin functions (‘spin up’ and ‘spin down’) spin orbital (product of spatial orbital and a spin function) basis function/atomic orbital (usually labelled $,, 4, $x, 0) indicates an integral over all spatial coordinates indicates an integral over all spin coordinates indicates an integral over all spatial and spin coordinates distances between two particles i and j (usually electrons in quantum mechanics) distance between two nuclei A and B Kronecker delta (6; = 1 if jy = Oifi x)) exchange operator Coulomb operator core Hamiltonian operator Fock matrix overlap matrix overlap integral between orbitals i and j Fock operator matrix of basis function coefficientsSymbols and Physical Con: xx G metric matrix (in distance geometry) Pi ith principal component Zz variance-covariance matrix x coupling parameter (used in free energy calculations) We’) weighting function used in umbrella sampling We number density (= N/V) Sap similarity coefficient between two molecules A and B Dap ‘distance’ between two molecules A and B a Hammett substitution constant fe partition coefficient of solute between two solvents © log(P,/Pu:) for a substituent X relative to a hydrogen substituent r squared correlation coefficient R squared correlation coefficient in multiple linear regression cross-validated R?Acknowledgements For this second edition I would like to thank Drs Neil Allan, Paul Bamborough, Gianpaolo Bravi, Richard Bryce, Julian Gale, Richard Green, Mike Hann and Alan Lewis, who commen- ted on various parts of the new text. Julian Gale’s suggestions were particularly useful for refining the sections concerning materials science and solid-state applications. I would also like to record my thanks once more to those who gave their time to read and comment on draft copies of various chapters of the first edition, upon which this second edition is based (in alphabetical order): Dr D B Adolf, Dr J M Blaney, Professor A V Chadwick, Dr PS Charifson, Dr C-W Chung, Dr A Cleasby, Dr A Emerson, Dr J W Essex, Dr D V S Green, Dr I R Gould, Dr M M Hann, Dr C A Leach, Dr M Pass, Dr D A Pearlman, Dr C A Reynolds, Dr D W Salt, Dr M Sagi, Professor J I Siepmann, Dr W C Swope, Dr N R Taylor, Dr P J Thomas, Professor D J Tildesley and Mr O Warschkow. Assistance with the illustrations for the second edition was provided by Drs R Groot, S McGrother and V Milman. Many of the figures from the first edition are also included here and so I would like to thank again Dr S E Greasley, Dr M_M Hann, Dr H Jhoti, Dr S N Jordan, Professor G R Luckhurst, Dr P M McMeekin, Dr A Nicholls, Dr P Popelier, Dr A Robinson and Dr T E Klein. Alexandra Seabrook, Pauline Gillet and Julie Knight at Pearson Education provided the foundation of the publishing team, coping with a steady stream of questions and keeping everyone to schedule. Especial thanks are due to Julie, who did a splendid job as editor. Any errors that remain are of course my own responsibility. If you do find any, I would like to know! I will also be pleased to receive any constructive suggestions, comments or criticisms. We plan to set up a web site that will provide access to various material from the book (such as electronic versions of the colour images) together with email contacts. This can be accessed via www. booksites.net. Molecular modelling would not be what it is today without the efforts of those who develop computer hardware and software and I would like to acknowledge the authors of the follow- ing computer programs which were used to generate figures and/or data described in the text. All calculations were performed using Silicon Graphics computers. AMBER DA Pearlman, D A Case, J C Caldwell, GL Seibel, U CSingh, P Weiner and P A Kollman 1991 Amber 3 0, University of California, San Francisco. Cambridge Structural Database: F H Allen, § A Bellard, M D Brice, B A Cartwright, A Doubleday, H Higgs, T Hummelink, B G Hummelink-Peters, O Kennard, W D $ Motherwell, J R Rodgers and D G Watson 1979 The Cambridge Crystallographic Data Centre’ Computer-Based Search, Retrieval, Analysis and Display of Information. Acta Crystallographica B35:2331-2339. Cambridge Crystallographic Data Centre, Cambridge, United Kingdom. CASTEP, Molecular Simulations Inc,, 9685 Scranton Road, San Diego, California, USA. Catalyst. Molecular Simulations Inc , 9685 Scranton Road, San Diego, California, USA. Cerius2, Molecular Simulations Inc., 9685 Scranton Road, San Diego, California, USA.xxii Acknowledgements COSMIC: J G Vinter, A Davis, MR Saunders 1987, Strategic approaches to drug design. I. An integrated software framework for molecular modeling. Journal of Computer-Aided Molecular Design. 1(1):31-51. Dials and Windows. G Ravishanker, § Swaminathan, D L Beveridge, R Lavery and H Sklenar 1989, Journal of Biomolecular Structure and Dynamics 6:669-699. Wesleyan University, USA Gaussian 92: M J Frisch, G W Trucks, M Head-Gordon, P M W Gill, M W Wong, J B Foresman, B G Johnson, H B Schlegel, M A Robb, E S Replogle, R Gomperts, J L Andres, K Raghavachari, J S Binkley, C Gonzalez, R L Martin, D J Fox, D J DeFrees, J Baker, J J P Stewart and J A Pople. Gaussian Inc, Pittsburgh, Pennsylvania, USA GCG: Genetics Computer Group, Inc., University Research Park, 575 Science Drive, Suite B, Madison, Wisconsin 53711, USA. GRASP (Graphical Representation and Analysis of Surface Properties): A Nicholls, Columbia University, New York, USA GRID: P J Goodford 1985 A Computational Procedure for Determining Energetically Favorable Binding Sites on Biologically Important Macromolecules. Journal of Medicinal Chemistry 28:849- 857. Molecular Discovery Ltd, Oxford, United Kingdom. InsightII: Molecular Simulations Inc., 9685 Scranton Road, San Diego, California, USA. IsoStar: | J Bruno, ) C Cole, J PM Lommerse, R § Rowland, R Taylor and M L Verdonk 1997. Isostar: a library of information about nonbonded interactions, Journal of Computer-Aided Molecular Design 11:525-537. Cambridge Crystallographic Data Centre, Cambridge, United Kingdom. Micromol: S M Colwell, A R Marshall, R D Amos and N C Handy 1985. Quantum chemistry on microcomputers, Chemistry in Britain 21:655~659 Molscript: Pj Kraulis 1991, Molscript - A program to produce both detailed and schematic plots of protein structures. Journal of Applied Crystallography 24:946-950. PROCHECK. R Laskowski, M W MacArthur, D S Moss and J M Thornton 1993. Procheck - A program to check the stereochemical quality of protein structures. Journal of Applied Crystallography 26:283~ 291. Quanta: Molecular Simulations Inc., 9685 Scranton Road, San Diego, California, USA. Spartan: Wavefunction Inc., 18401 Von Karman, Suite 370, Irvine, California, USA. SPASMS (San Francisco Package of Applications for the Simulation of Molecular Systems). D A Spellmeyer, W C Swope, E-R Evensen, T Cheatham, D M Ferguson and P A Kollman. University of California, San Francisco, USA. Sybyl: Tripos Inc,, 1699 South Hanley Road, St. Louis, Missouri, USA. The following programs were used to produce draft copies of the manuscript and diagrams: Microsoft Word (Microsoft Corp.), Gnuplot (T Williams and C Kelley), Kaleidagraph (Abelbeck Software), Chem3D (CambridgeSoft Corp ) and Microsoft Excel (Microsoft Corp.). We are grateful to the following for permission to reproduce copyright material: Figure 1.11 from Mathematical Methods in the Physical Sciences, 2nd edn, Boas ML, ©1983. Reprinted by permission of John Wiley & Sons, Inc. Figure 1.14 from The FFT Fundamentals and Concepts by Ramirez ©1985, Reprinted by permission of Prentice-Hall, Inc., Upper Saddle River, NJ. Figures 27 and 33 from Ab initio Molecular Orbital Theory, Hehre W J, L Radom, P v R Schleyer, J A Hehre ©1986 Reprinted with permission by John Wiley & Sons, Inc, Figure 35 from Gerratt J, D 1. Cooper, P B Karadakov and M Raimondi 1997. Modern valence bond theory. Chemical Society Reviews 87-100. Reproduced by permission of The Royal Society of Chemistry Figure 3.22 from Needs RJ and Mujica 1995. First-principles pseudopotential study of the structural phases of silicon. Physical Review B 51:9652-9660Acknowledgements xxill Figure 4.18 from Buckingham A D 1959. Molecular Quadrupole Moments Quarterly Reviews of the Chemical Society 13:183-214. Reproduced by permission of The Royal Society of Chemistry. Figure 4.29 from Computer Simulation in Chemical Physics, edited by Allen M P and D J Tildesley, 1993. Effective Pair Potentials and Beyond, Sprik M, with kind permission from Kluwer Academic Publishers Figure 4 49 reprinted with permission from Pranata J and W L Jorgensen. Computational Studies on FK506: Conformational Search and Molecular Dynamics Simulations in Water, The Journal of the American Chemical Society 113:9483-9493 ©1991 American Chemucal Society. Figure 450 from Molecular Parameters for Organosilicon Compounds Calculated from Ab Initio Computations, Grigoras $ and T H Lane, Journal of Computational Chemistry 9:25-39, ©1988 Reprinted by permission of John Wiley & Sons, Inc. Figures 5 4 and 58 Press W H, B P Flannery, S A Teukolsky and W T Vetterling, Numerical Recipes in Fortran. 1992, Cambridge University Press. Figure 5 21 reprinted with permission from Chandrasekhar J, S F Smith and W L Jorgensen. Theoretical Examination of the Sx? Reaction Involving Chloride Ion and Methyl Chloride in the Gas Phase and Aqueous Solution. The Journal of the American Chemical Society 107:154~163. ©1985 American Chemical Society. Figure 523 reprinted with permission from Doubleday C, J McIver, M Page and T Zielinski ‘Temperature Dependence of the Transition-State Structure for the Disproportionation of Hydrogen Atom with Ethyl Radical. The Journal of the American Chemical Society 107:5800-5801 ©1985 American Chemical Society. Figure 529 from Gonzalez C and H B Schlegel 1988 An Improved Algorithm for Reaction Path Following, The Journal of Chemical Physics 902154-2161. Figure 530 reprinted from Chemical Physical Letters, 194, Fischer S and M Karplus. Conjugate Peak Refinement: An Algorithm for Finding Reaction Paths and Accurate Transition States in Systems with Many Degrees of Freedom. 252-261, ©1992, with permission from Elsevier Science Figure 5 35 reprinted with permission from Houk K N, J Gonzdlez and Y Li. Pericyclic Reaction Transition States. Passions and Punctilios 1935-1995. Accounts of Chemical Research 28:81-90. (©1995 American Chenucal Society. Figure 6.25 reprinted from Chemical Physics Letters, 196, Ding H-Q, N Karasawa and W A Goddard III, The Reduced Cell Multipole Method for Coulomb Interactions in Periodic Systems with Million- Atom Unit Cells, 6-10, ©1992, with permission of Elsevier Science Figure 7 2 from Alder B J and TE Wainwright 1959, Studies in Molecular Dynamics. I. General Method. ‘The Journal of Chemical Physics 31:459-466, Figure 7 11 from Alder B J and T E Wainwright 1970. Decay of the Velocity Autocorrelation Function. Physical Review A 1:18-21. Figure 7 12 from Guillot B 1991 A Molecular Dynamics Study of the Infrared Spectrum of Water. The Journal of Chemical Physics 95:1543~1551 Figure 7 13 reprinted with permission from Jorgensen W L, R C Binning Jr and B Bigot Structures and Properties of Organic Liquids: n-Butane and 1,2-Dichloroethane and Their Conformational Equilibria. The Journal of the American Chemical Society 103:4393-4399, ©1981 American Chemical Society. Figure 7.24 (and on cover) from Groot R D and T J Madden 1998, Dynamic simulation of diblock copolymer microphase separation The Journal of Chemical Physics 108:8713-8724. © American Institute of Physics Figure 8 16 from Frantz, D D, D L Freeman and J D Doll 1990. Reducing quasi-ergodic behavior in Monte Carlo simulations by J-walking: applications to atomic clusters. The Journal of Chemical Physics 93:2769-2784xxiv Acknowledgements Figure 8.17 from Cracknell RE, D Nicholson and N Quirke 1994. A Grand Canomucal Monte Carlo Study of Lennard-Jones Mixtures in Slit Pores; 2: Mixtures of Two-Centre Ethane with Methane. Molecular Simulation 13:161-175. ©1994 OPA (Overseas Publishers Association) N.V. Permussion to reproduce granted by Gordon and Breach Publishers. Figure 8.22 reprinted with permission from Smit B and J I Siepmann. Simulating the Adsorption of Alkanes in Zeolites Science 264 1118-1120 ©1994 American Association for the Advancement of Science. Figure 9,34 from Poling Promoting Conformational Variation, Smellie A S, § L Teig and P Towbin Journal of Computational Chemistry 16:171-187, ©1995. Reprinted by permission of John Wiley & Sons, Inc. Figure 10.18 from Pearson W R and D J Lipman 1988 Improved tools for biological sequence comparison Proceedings of the National Academy of Sciences USA 85 2444-2448. Figure 10.21 reprinted from Current Opinion in Structural Biology, 6, Eddy § R, Hidden Markov Models, 361-365 ©1996, with permission from Elsevier Science Figure 10 26 Reprinted with permission from Liithy R, J U Bowie and D Eisenberg Assessment of Protein Models with Three-Dimensional Profiles Nature 356:83-85. ©1992 Macmillan Magazines Limuted. Figure 11.6 from Lybrand T P, J A McCammon and G Wipff 1986 Theoretical Calculation of Relative Binding Affinity in Host-Guest Systems. Proceedings of the National Academy of Sciences USA 83:833- 835 Figure 11.18 reprinted with permission from Guo Z and C L Brooks III Rapid Screening of Binding Affinities. Application of the Dynamics Method to a Trypsin-Inhibitor System. The Journal of the American Chemical Society 120:1920-1921. ©1998 American Chemucal Society. Figure 11 24 reprinted with permission from Still W C, A Tempczyrk, R C Hawley and T Hendrickson Semianalytical Treatment of Solvation for Molecular Mechanics and Dynamucs. The Journal of the American Chemical Society 112 6127-6129 ©1990 American Chemical Society. Figure 11 35 reprinted with permission from Chandrasekhar J and W L Jorgensen 1985, Energy Profile for a Nonconcerted $x2 Reaction in Solution. The Journal of the American Chemical Society 107.2974- 2973. ©1985 American Chemical Society Figure 11.37 reprinted with permussion from Aqvist J, M Fothergill and A Warshel. Computer Simulation of the CO,/HCO; Interconversion Step in Human Carbonic Anhydrase I. The Journal of the American Chemical Society 115:631-635 ©1993 American Chemical Society Figure 11.40 reprinted with permussion from Saitta A M, P D Sooper, E Wasserman and M L Klein 1999. Influence of a knot on the strength of a polymer strand. Nature 399:46-48. ©1999 Macmillan Magazines Limited. Figure 11 42 from NATO ASI Series C 498 (New Trends in Materials Chemistry), 1997, 285-318, Defects and Matter Transport in Solid Materials, Chadwick A V and J Corish, with kind permission of Kluwer Academic Publishers Whilst every effort has been made to trace the owners of copyright material, we take this opportunity to offer our apologies to any copyright holders whose rights we may have unwittingly infringed.Useful Concepts in Molecular Modelling 1.1 Introduction What is molecular modelling? ‘Molecular’ clearly implies some connection with molecules, The Oxford English Dictionary defines ‘model’ as ‘a simplified or idealised description of a system or process, often in mathematical terms, devised to facilitate calculations and predic- tions’. Molecular modelling would therefore appear to be concerned with ways to mimic the behaviour of molecules and molecular systems. Today, molecular modelling is invariably associated with computer modelling, but it is quite feasible to perform some simple molecular modelling studies using mechanical models or a pencil, paper and hand calcula- tor. Nevertheless, computational techniques have revolutionised molecular modelling to the extent that most calculations could not be performed without the use of a computer. This is not to imply that a more sophisticated model is necessarily any better than a simple one, but computers have certainly extended the range of models that can be considered and the systems to which they can be applied. The ‘models’ that most chemists first encounter are molecular models such as the ‘stick’ models devised by Dreiding or the ‘space filling’ models of Corey, Pauling and Koltun (commonly referred to as CPK models). These models enable three-dimensional represen- tations of the structures of molecules to be constructed. An important advantage of these models js that they are interactive, enabling the user to pose ‘what if ...’ or ‘is it possible to .- " questions. These structural models continue to play an important role both in teaching and in research, but molecular modelling is also concerned with more abstract models, many of which have a distinguished history. An obvious example is quantum mechanics, the foundations of which were laid many years before the first computers were constructed. There is a lot of confusion over the meaning of the terms ‘theoretical chemistry’, ‘computa- tional chemistry’ and ‘molecular modelling’. Indeed, many practitioners use all three labels to describe aspects of their research, as the occasion demands! ‘Theoretical chemistry’ is often considered synonymous with quantum mechanics, whereas computational chemistry encompasses not only quantum mechanics but also molecular mechanics, minimisation, simulations, conformational analysis and other computer-based methods for understanding and predicting the behaviour of molecular systems. Molecular modellers use all of these methods and so we shall not concern ourselves with semantics but rather shall consider any theoretical or computational technique that provides insight into the behaviour of molecular systems to be an example of molecular modelling. If a distinction has to be2 Chapter 1 made, it is in the emphasis that molecular modelling places on the representation and manipulation of the structures of molecules, and properties that are dependent upon those three-dimensional structures. The prominent part that computer graphics has played in molecular modelling has led some scientists to consider molecular modelling as little more than a method for generating ‘pretty pictures’, but the technique is now firmly established, widely used and accepted as a discipline in its own right. A closely related subject is molecular informatics. This is a rather new term, making a precise definition difficult, but it is usually considered to encompass two disciplines: chemoinfor- matics and bioinformatics. Of these two areas, chemoinformatics (also written cheminfor- matics) is the newer name but the older discipline; chemists have been using computers to store, retrieve and manipulate information about molecules almost since computers were invented. Both chemoinformatics and bioinformatics have risen to prominence primar- ily as a consequence of the introduction of new experimental techniques. For the chemist these experimental techniques are combinatorial library synthesis and high-throughput screening, which enable very large numbers of molecules to be synthesised and tested; for the biologist they are the automated sequencing machines that are being used to determine the human genome. A characteristic feature of molecular informatics is that it is concerned with information about large numbers of molecules, much larger than is typically the case for a traditional molecular modelling study. For this reason, informatics was initially more concerned with less complex representations of molecules that did not fully represent their three-dimensional properties. However, even this distinction is now being eroded and there is increasing use made of more traditional molecular modelling techniques within informatics. In the rest of this chapter we shall discuss a number of concepts and techniques that are relevant to many areas of molecular modelling and so do not sit comfortably in any individual chapter. We will also define some of the terms that will be used throughout the book. 1.2 Coordinate Systems It is obviously important to be able to specify the positions of the atoms and/or molecules in the system to a modelling program’. There are two common ways in which this can be done. The most straightforward approach is to specify the Cartesian (x, y,z) coordinates of all the atoms present. The alternative is to use internal coordinates, in which the position of each atom is described relative to other atoms in the system. Internal coordinates are usually written as a Z-matrix. The Z-matrix contains one line for each atom in the system. A sample Z-matrix for the staggered conformation of ethane (see Figure 1-1) is “For a system containing a large number of independent molecules it is common to use the term ‘configuration’ to refer to each arrangement; this use of the word ‘configuration’ is not to be con- fused with its standard chemical meaning as a different bonding arrangement of the atoms in a moleculeUseful Concepts in Molecular Modelling 3 Fig. 11 The staggered conformation of ethane as follows: al Le 2 Cc 154061 3 H 1.0 7 109.5 7 4 H 1.0 2 109.5 * 180.0 3 5 H 10 * 109.5 2 60.0 4 6 H 1.0 2 109.5 2 60.0 5 7 H 1.0 x 109.5 io 180.0 6 8 H pe 2 109.5 ' 60.0 7 In the first line of the Z-matrix we define atom 1, which is a carbon atom. Atom number 2 is also a carbon atom that is a distance of 1.54A from atom 1 (columns 3 and 4). Atom 3 is a hydrogen atom that is bonded to atom 1 with a bond length of 1.0A. The angle formed by atoms 2-1-3 is 109.5°, information that is specified in columns 5 and 6. The fourth atom is a hydrogen, a distance of 1.0 A from atom 2, the angle 4-2-1 is 109.5°, and the torsion angle (defined in Figure 1.2) for atoms 4-2-1-3 is 180°. Thus for all except the first three atoms, each atom has three internal coordinates: the distance of the atom from one of the atoms previously defined, the angle formed by the atom and two of the previous atoms, and the torsion angle defined by the atom and three of the previous atoms. Fewer internal coordinates are required for the first three atoms because the first atom can be placed anywhere in space (and so it has no internal coordinates); for the second atom it is only necessary to specify its distance from the first atom and then for the third atom only a distance and an angle are required. It is always possible to convert internal to Cartesian coordinates and vice versa. However, one coordinate system is usually preferred for a given application. Internal coordinates can usefully describe the relationship between the atoms in a single molecule, but Cartesian coordinates may be more appropriate when describing a collection of discrete molecules. Internal coordinates are commonly used as input to quantum mechanics programs, whereas calculations using molecular mechanics are usually done in Cartesian coordinates. The total number of coordinates that must be specified in the internal coordinate system is six fewer4 Chapter 1 Fig. 1.2 A torsion angle A-B-C-D is defied as the angle between the planes A, B, Cand B, C, D A torsion angle can vary through 360° although the range ~180° to +180" is most commonly used We shall adopt the IUPAC definition of a torsion angle in tohich an eclipsed conformation corresponds to a torsion angle of 0” and a trans or anti conformation to a torsion angle of 180°. The reader should note that this may not correspond to some of the definitions used in the literature, where the trans arrangement is defied as a torsion angle of 0° {fone looks along the bond B-C, then the torsion angle is the angle through which it is necessary 10 rotate the bond AB in a clockwise sense in order to superimpose the two planes, as shown than the number of Cartesian coordinates for a non-linear molecule. This is because we are at liberty to arbitrarily translate and rotate the system within Cartesian space without changing the relative positions of the atoms 41.3 Potential Energy Surfaces In molecular modelling the Born-Oppenheimer approximation is invariably assumed to operate. This enables the electronic and nuclear motions to be separated; the much smaller mass of the electrons means that they can rapidly adjust to any change in the nuclear posi- tions. Consequently, the energy of a molecule in its ground electronic state can be. considered a function of the nuclear coordinates only. If some or all of the nuclei move then the energy will usually change. The new nuclear positions could be the result of a simple process such as a single bond rotation or it could arise from the concerted movement of a large number of atoms, The magnitude of the accompanying rise or fall in the energy will depend upon the type of change involved. For example, about 3 kcal/mol is required to change the covalent carbon-carbon bond length in ethane by 0.1A away from its equilibrium value, but only about 0.1 kcal/mol is required to increase the non-covalent separation between two argon atoms by 1A from their minimum energy separation. For small isolated molecules, rotation about single bonds usually involves the smallest changes in energy. For example, if we rotate the carbon-carbon bond in ethane, keeping all of the bond lengths and angles fixed in value, then the energy varies in an approximately sinusoidal fashion as shown in Figure 1.3, with minima at the three staggered conformations. The energy in this case can be considered a function of a single coordinate only (ie. the torsion angle of the carbon-carbon bond), and as such can be displayed graphically, with energy along one axis and the value of the coordinate along the other.Useful Concepts In Molecular Modelling 5 30 25 20) 10 Energy (kcal/mol) 05 ol 0 60 120 190 240 300 360 C-C torsion angle Fig. 13. Variation in energy with rotation of the carbon-carbon bond in ethane Changes in the energy of a system can be considered as movements on a multidimensional ‘surface’ called the energy surface. We shall be particularly interested in stationary points on the energy surface, where the first derivative of the energy is zero with respect to the internal or Cartesian coordinates. Ata stationary point the forces on all the atoms are zero, Minimum points are one type of stationary point; these correspond to stable structures. Methods for locating stationary points will be discussed in more detail in Chapter 5, together with a more detailed consideration of the concept of the energy surface. 1.4 Molecular Graphics Computer graphics has had a dramatic impact upon molecular modelling, It should always be remembered, however, that there is much more to molecular modelling than computer graphics. It is the interaction between molecular graphics and the underlying theoretical methods that has enhanced the accessibility of molecular modelling methods and assisted the analysis and interpretation of such calculations. Molecular graphics systems have evolved from delicate and temperamental pieces of equip- ment that cost hundreds of thousands of pounds and occupied entire rooms, to today’s inexpensive workstations that fit on or under a desk and yet are hundreds of times more powerful Over the years, two different types of molecular graphics display have been used in molecular modelling. First to be developed were vector devices, which construct pictures using an electron gun to draw lines (or dots) on the screen, in a manner similar to an oscilloscope. Vector devices were the mainstay of molecular modelling for almost two decades but have now been largely superseded by raster devices. These divide the screen into a large number of small ‘dots’, called pixels. Each pixel can be set to any of a large number of colours, and so by setting each pixel to the appropriate colour it is possible to generate the desired image. Molecules are most commonly represented on a computer graphics screen using ‘stick’ or ‘space-filling’ representations, which are analogous to the Dreiding and Corey-Pauling- Koltun (CPK) mechanical models. Sophisticated variations on these two basic types have6 Chapter 1 been developed, such as the ability to colour molecules by atomic number and the inclusion of shading and lighting effects, which give ‘solid’ models a more realistic appearance. Some of the commonly used molecular representations are shown in Figure 1.4 (colour plate section). Computer-generated models do have some advantages when compared with their mechanical counterparts. Of particular importance is the fact that a computer model can be very easily interrogated to provide quantitative information, from simple geometrical measures such as the distance between two atoms to more complex quantities such as the energy or surface area. Quantitative information such as this can be very difficult if not impossible to obtain from a mechanical model. Nevertheless, mechanical models may still be preferred in certain types of situation due to the ease with which they can be manipulated and viewed in three dimensions. A computer screen is inherently two-dimensional, whereas molecules are three-dimensional objects. Nevertheless, some impression of the three-dimensional nature of an object can be represented on a computer screen using techniques such as depth cueing (in which those parts of the object that are further away from the viewer are made less bright) and through the use of perspective. Specialised hardware enables more realistic three-dimensional stereo images to be viewed. In the future ‘virtual reality’ systems may enable a scientist to interact with a computer-generated molecular model in much the same way that a mechanical model can be manipulated. Even the most basic computer graphics program provides some standard facilities for the manipulation of models, including the ability to translate, rotate and ‘zoom’ the model towards and away from the viewer. More sophisticated packages can provide the scientist with quantitative feedback on the effect of altering the structure. For example, as a bond is rotated then the energy of each structure could be calculated and displayed interactively. For large molecular systems it may not always be desirable to include every single atom in the computer image; the sheer number of atoms can result in a very confusing and cluttered picture. A clearer picture may be achieved by omitting certain atoms (e.g. hydrogen atoms) or by representing groups of atoms as single ‘pseudo-atoms’ The techniques that have been developed for displaying protein structures nicely illustrate the range of computer graphics representation possible (the use of computational techniques to investigate the structures of proteins is considered in Chapter 10). Proteins are polymers constructed from amino acids, and even a small protein may contain several thousand atoms. One way to produce a clearer picture is to dispense with the explicit representation of any atoms and to represent the protein using a ‘ribbon’. Proteins are also commonly represented using the cartoon drawings developed by J Richardson, an example of which is shown in Figure 1.5 (colour plate section). The cylinders in this figure represent an arrangement of amino acids called an o-helix, and the flat arrows an alternative type of regular structure called a 3-strand. The regions between the cylinders and the strands have no such regular structure and are represented as ‘tubes’. 1.5 Surfaces Many of the problems that are studied using molecular modelling involve the non-covalent interaction between two or more molecules. The study of such interactions is often facilitatedUseful Concepts in Molecular Modelling 7 Solvent-accessible surface Probe sphere ‘dw sutace ~Re-entrant surface Molecular Contact surface surface Fig 16° The van der Wauls (vdw) surface of a molecule corresponds to the outward-facing surfaces of the van der Waals spheres of the atoms The molecular surface is generated by rolling a spherical probe (usually of radius 1.4.A to representa water raolecule) on the vant der Waals surface The molecular surface is constructed from contact and re-entrant surface elements "The centre of the probe traces out the accessible surface by examining the van der Waals, molecular or accessible surfaces of the molecule. The van der Waals surface is simply constructed from the overlapping van der Waals spheres of the atoms, Figure 1.6. It corresponds toa CPK or space-filling model. Let us now consider the approach of a small ‘probe’ molecule, represented as a single van der Waals sphere, up to the van der Waals surface of a larger molecule. The finite size of the probe sphere means that there will be regions of ‘dead space’, crevices that are not accessible to the probe as it rolls about on the larger molecule. This is illustrated in Figure 1.6. The amount of dead space increases with the size of the probe; conversely, a probe of zero size would be able to access all of the crevices. The molecular surface [Richards 1977] is traced out by the inward-facing part of the probe sphere as it rolls on the van der Waals surface of the molecule. The molecular surface contains two different types of surface element. The contact surface corresponds to those regions where the probe is actually in contact with the van der Waals surface of the ‘target’ ‘The re-entrant surface regions occur where there are crevices that are too narrow for the probe molecule to penetrate. The molecular surface is usually defined using a water molecule as the probe, represented as a sphere of radius 14 A. The accessible surface is also widely used. As originally defined by Lee and Richards [Lee and Richards 1971] this is the surface that is traced by the centre of the probe molecule as it rolls on the van der Waals surface of the molecule (Figure 1.6). The centre of the probe molecule can thus be placed at any point on the accessible surface and not penetrate the van der Waals spheres of any of the atoms in the molecule. Widely used algorithms for calculating the molecular and accessible surfaces were devel- oped by Connolly [Connolly 1983a, b], and others [e.g. Richmond 1984] have described formulae for the calculation of exact or approximate values of the surface area. There are many ways to represent surfaces, some of which are illustrated in Figure 1.7 (colour plate Section). As shown, it may also be possible to endow a surface with a translucent quality, which enables the molecule inside the surface to be displayed. Clipping can also be used8 Chapter 1 to cut through the surface to enable the ‘inside’ to be viewed. In addition, properties such as the electrostatic potential can be calculated on the surface and represented using an appro- priate colour scheme. Useful though these representations are, it is important to remember that the electronic distribution in a molecule formally extends to infinity. The ‘hard sphere’ representation is often very convenient and has certainly proved very valuable, but it may not be appropriate in all cases [Rouvray 1997, 1999, 2000]. 1.6 Computer Hardware and Software One cannot fail to be amazed at the pace of development in the computer industry, where the ratio of performance-to-price has increased by an order of magnitude every five years or so. The workstations that are commonplace in many laboratories now offer a real alternative to centrally maintained ‘supercomputers’ for molecular modelling calculations, especially as a workstation or even a personal computer can be dedicated toa single task, whereas the super- computer has to be shared with many other users. Nevertheless, in the immediate future there will always be some calculations that require the power that only a supercomputer can offer, The speed of any computer system is ultimately constrained by the speed at which electrical signals can be transmitted. This means that there will come a time when no further enhancements can be made using machines with ‘traditional’ single-processor serial architectures, and parallel computers will play an ever more important role. A parallel computer couples processors together in such a way that a calculation is divided into small pieces with the results being combined at the end. Some calculations are more amenable to parallel processing than others, and a significant amount of effort is being spent converting existing algorithms to run efficiently on parallel architectures. In some cases completely new methods have been developed to take maximum advantage of the opportunities of parallel processing. The low cost of personal computer chips means that large ‘farms’ of processors can be constructed to give significant computing power for relatively small outlay. To perform molecular modelling calculations one also requires appropriate programs (the software). The software used by molecular modellers ranges from simple programs that perform just a single task to highly complex packages that integrate many different methods. There is also an extremely wide variation in the price of software! Some programs have been so widely used and tested that they can be considered to have reached the status of a ‘gold standard’ against which similar programs are compared. One hesitates to specify such programs in print, but three items of software have been so widely used and cited that they can safely be afforded the accolade. These are the Gaussian series of programs for per- forming ab initio quantum mechanics, the MOPAC/AMPAC programs for semi-empirical quantum mechanics and the MM2 program for molecular mechanics. Various pieces of software were used to generate the data for the examples and illustrations throughout this book. Some of these were written specifically for the task; some were freely available programs; others were commercial packages. I have decided not to describe specific programs in any detail, as such descriptions rapidly become outdated. Nevertheless,Useful Concepts in Molecular Modelling 7 all items of software are accredited where appropriate. Please note that the use of any particular piece of software does not imply any recommendation! 1.7 Units of Length and Energy Itwill be noted that our Z-matrix for ethane has been defined using the angstrom as the unit of length (1 A= 10-” m = 100 pm). The angstrém is a non-SI unit but is a very convenient one to use, as most bond lengths are of the order of 1-2 A. One other very common non-SI unit found in the molecular modelling literature is the kilocalorie (1 kcal = 4.1840 K)). Other systems of units are employed in other types of calculation, such as the atomic units used in quantum mechanics (discussed in Chapter 2). It is important to be aware of, and familiar with, these non-standard units as they are widely used in the literature and throughout this book. 1.8 The Molecular Modelling Literature The number of scientific papers concerned with molecular modelling methods is rising rapidly, as is the number of journals in which such papers are published. This reflects the tremendous diversity of problems to which molecular modelling can be applied and the ever-increasing availability of molecular modelling methods. It does, however, mean that it can be very difficult to remain up to date with the field. A number of specialist journals are devoted to theoretical chemistry, computational chemistry and molecular modelling, each with their own particular emphasis. Relevant papers are also published in the more ‘general’ journals, and there are now a number of books covering aspects of molecular modelling, some aimed at the specialist reader, others at the beginner. Many scientists are now fortunate to have access to electronic catalogues of publications which can be searched to find relevant papers. As many journals are now available over the internet it is possible to perform a literature search and obtain copies of the relevant papers without even having to leave the office. Some of the journals which are devoted to short reviews of recent develop- ments often include molecular modelling sections (such as the ‘Current Opinion’ series); in others, useful review articles appear on an occasional basis, One particularly valuable source of information on molecular modelling methods is the Reviews in Computational Chemistry, edited by Lipkowitz and Boyd, beginning in 1990 (see Further Reading). Each of these volumes contains chapters on a variety of subjects, each written by an appropriate expert. A recent addition is the Encyclopaedia of Computational Chemistry by Schleyer et al. (1998) (see Further Reading), which contains many chapters that cover a wide range of topics. 1.9 The Internet In the first edition of this book I wrote, ‘A major use of the Internet is for electronic mail, but extremely rapid growth is being observed in other areas, particularly the “World-Wide Web” (WWW). ’. Such a phrase seems an understatement; despite the ‘hype’, the Internet has certainly made a dramatic impact, not least on the scientific community, where its10 Chapter 1 origins lie. Anything written about the Internet is almost certain to become obsolete more rapidly than any other topic in this book and so this section will be brief. I will assume that all readers of this book will be familiar with the use of a web browser and the concept of a hyperlink, which enables documents to be linked together. The URL (Uniform Resource Locator) is the currency of the WWW, being the ‘electronic address’ which enables the particular item to be identified. Most documents are still written using HTML (HyperText Markup Language) but increasingly incorporate more sophisticated features. Given the tremendous growth in the Web it is important to be able to locate relevant information. This is the role of the Internet search engines, which can be used to identify relevant sites of interest via some form of keyword search. Within the molecular modelling context, several trends can be noted Whilst the Web was initially used to distribute mostly textual information, it is increasingly used for much more sophisticated applications. Interactive molecular graphics are a feature of many sites. Some sites enable calculations or database searches to be performed via the Web, with the results being delivered interactively or via email. This is particularly true for ‘intranets’ within an organisation. XML (eXtensible Markup Language) is likely to play an increasingly important role in the ‘intelligent’ exchange of information over the Web, especially in specialist areas such as chemistry [Murray-Rust and Rzepa 1999]. Several ‘electronic conferences’ have been held with partici- pants from many different countries. Perhaps the only prediction that one can safely make about the Web is that it is here to stay and its use will continue to grow, 1.10 Mathematical Concepts A full appreciation of all of the techniques of molecular modelling would require a mathematical treatment beyond that appropriate to a book of this size and scope. However, a proper understanding does benefit from some knowledge of mathematical concepts such as vectors, matrices, differential equations, complex numbers, series expansions and Lagran- gian multipliers, and some very elementary statistical concepts. There is only space in this book for a cursory introduction to these mathematical concepts and ideas, with very brief descriptions and some key results. The suggestions for further reading provide detailed background information on all of the mathematical topics required. 1.10.1 Series Expansions There are various series expansions that are useful for approximating functions. Particularly important is the Taylor series: if f(x) is a continuous, single-valued function of x with contin- uous derivatives f(x), f"(x),..., then we can expand the function about a point xp as follows: foro +2) =fleo) +2 fo) +5 feo) +E FM) 4 AMO) 12) Taylor series are often truncated after the term involving the second derivative, which makes the function vary in a quadratic fashion This is a common assumption in many of the minimisation algorithms that we will discuss in Chapter 5Useful Concepts In Molecular Modelling 1 ‘A Maclausin series is a specific form of the Taylor series for which xy = 0. Some standard expansions in Taylor series form are: 7 228 altxtytyt gt (12) sinx =x 3 (13) In(a sex Ze (14) The binomial expansion is used for functions of the form (1+ x)*: = me ca (ta) =1texta(e—1)F + alae) P+ (15) All these series must have |x| < 1 to be convergent. 1.10.2 Vectors A vector is a quantity with both magnitude and direction. For example, the velocity of a moving body is a vector quantity as it defines both the direction in which the body is travel- ling and the speed at which it is moving, In Cartesian coordinates a vector such as the velocity will have three components, indicating the contribution to the overall motion from the component motions along the x, y and z directions. The addition and subtraction of vectors can be understood using geometrical constructions, as shown in Figure 18. Thus, if we want to calculate the force on an atom due to its interactions with all other atoms in the system (as required in molecular dynamics calculations, see Chapter 7), we would perform a vector sum of all the individual forces. Some of the common manipulations that are performed with vectors include the scalar product, vector product and scalar triple product, which we will illustrate using vectors 11, T) and r, that are defined in a rectangular Cartesian coordinate system: nauitynj+uk t= i+ yj + 2k 15 = xsi + yaj + Zk (1.6) n+r A : Ti n-n = . " Addition Subtraction Fig 1.8. The addition and subtraction of vectors12 Chapter 1 T "1 Scalar product Vector product ee 3 (0X2) " Scalar triple product Fig 1.9. The soalar product, vector product and scalar triple product. i,j and k are orthogonal unit vectors along the x, y and z axes. The scalar product is defined as: 11°%) = |n| |r| cos 0 (1.7) [ry] and |r2| are the magmtudes of the two vectors (|r)| = \/x} + ¥} + 2}) and @ is the angle between them (Figure 1.9). The angle can be calculated as follows: (18) ‘The scalar product of two vectors is thus a scalar. The vecior product of two vectors r, x t (sometimes written r; A 12) is a new vector (v), ina direction perpendicular to the plane containing the two original vectors (Figure 1.9). The direction of this new vector is such that rj, r2 and the new vector form a right-handed system. If r, and ry are three-component vectors then the components of v are given by: V = (yrz2— Zaya)i + (Za%2 — ¥122)j + (1Y2 — YrX2)k (1.9) Note that the vector product r; x1; is not the same as the vector product rj xr, as it corresponds to a vector in the opposite direction. The vector product is thus not commutative. The scalar triple product x5 (t2 x 13) equals the scalar product of r, with the vector product of ry and ry. The result is a scalar. The scalar triple product has a useful geometrical interpreta- tion; it is the volume of the parallelepiped whose sides correspond to the three vectors (Figure 1.9) 1.10.3 Matrices, Eigenvectors and Eigenvalues A matrix is a set of quantities arranged in a rectangular array. An m x n matrix has mt rows and n columns. A vector can thus be considered to be a one-column matrix. Matrix additionUseful Concepts in Molecular Modelling 13 and subtraction can only be performed with matrices of the same order. For example: 4 cee 4 3 ea Lt and B= 5 2 8 2 5 3 0 10 8 4 Then A+B=|2 7]; A-B=|-8 3 (1.10) a1 R -5 ‘Multiplication of two matrices (AB) is only possible if the number of columns in A is equal to the number of rows in B. If A is an m x n matrix and B is ann x 0 matrix then the product AB is an m x 0 matrix. Each element (i, j) in the matrix AB is obtained by taking each of the n values in the ith row of A and multiplying by the corresponding value in the jth column of B. To illustrate with a simple example: eee eeaecsaee 2 if a=( . a) and B=] -2 4 seer ecer emcee 16 Then pa (OXOF(-2%-D+GxI) Bx3)+(-2x4) +6 x6) -(Ceotaentacn (-3 x3) + (4x4) + (1x6) _(9 31 ua Nee 4) cee We shall often encounter square matrices, which have the same number of rows and columns. A diagonal matrix is a square matrix in which all the elements are zero except for those on the diagonal. The unit or identity matrix I is a special type of diagonal matrix in which all the non-zero elements are 1; thus the 3 x 3 unit matrix is: fee I=/010 (1.12) 001 A matrix is symmetric if it is a square matrix with elements such that the elements above and below the diagonal are mirror images; Ay = Ay. Multiplication of a matrix by its inverse gives the unit matrix: ATA=I (BE) To compute the inverse of a square matrix it is necessary to first calculate its determinant, |A|. The determinants of 2 x 2 and 3 x 3 matrices are calculated as follows: ab alamo (1.14)14 : Chapter 1 oe : | e de fl= tal ‘4d Sala ile flee Ff ghi = alei — hf) — b(di — fa) + e(dh — eg) (1.15) For example: 3 6 a - 3|=2 25 0/=28 (1.16) 20 3 Ascan be seen, the determinant of a 3 x 3 matrix can be written as a sum of determinants of 2.x 2 matrices, obtained by first selecting one of the rows or columns in the matrix (the top row was chosen in our example). For each element A, in this row, the row and column in which that number appears are deleted (ie. the ith row and the jth column). This leaves a 2.x 2 matrix whose determinant is calculated and then multiplied by (—1)'*/, The result of this calculation is called the cofactor of the element Aj. For example, the cofactor of the element Aj in the 3 x 3 matrix 42 -2 A 25 0 peo Oa is -6. When calculating the determinant the cofactor is multiplied by the element Ay. The determinants of larger matrices can be obtained by extensions of the scheme illustrated above; thus the determinant of a 4 x 4 matrix is initially written in terms of 3 x 3 matrices, which in tun can be expressed in terms of 2 x 2 matrices. Determinants have many useful and interesting properties. The determinant of a matrix is zero if any two of its rows or columns are identical. The sign of the determinant is reversed by exchanging any pair of rows or any pair of columns. If all elements of a row (or column) are multiplied by the same number, then the value of the determinant is multiplied by that number. The value of a determinant is unaffected if equal multiples of the values in any row (or column) are added to another row (or column). The vector product and the scalar triple product can be conveniently written as matrix determinants. Thus: iy k nxn=|m wy 4 (1.17) Rn 2 nym (2x0) =|%2 2 2 (1.18) % Ys %Useful Concepts in Molecular Modelling _ ‘The transpose of a matrix, A’, is the matrix obtained by exchanging its rows and columns. ‘Thus the transpose of an m x n matrix is ann x m matrix: 47 -3 8 if A=|}-3 5 at=( ) (1.19) 7b 2 8 -2 ‘The transpose of a square matrix is, of course, another square matrix. The transpose of a symmetric matrix is itself. One particularly important transpose matrix is the adjoint matrix, adjA, which is the transpose matrix of cofactors. For example, the matrix of cofactors of the 3 x 3 matrix 42 -2 1 -6 10 A={ 25 0] is |-6 8 4 (1.20) eeaeeeae = 10 -4 16 In this case the adjoint matrix is the same as the matrix of cofactors (as A is a symmetric matrix). The inverse of a matrix is obtained by dividing the elements of the adjoint matrix by the determinant: -1_adjA ea (1.21) Thus the inverse of our 3 x 3 matrix is 15/28 -3/14 5/14 At=|-3/14 9 2/7 -1/7 (1.22) Cia 47 One of the most common matrix calculations involves finding its eigenvalues and eigenvectors. An eigenvector is a column matrix x such that Ax =x (1.23) Ais the associated eigenvalue. The eigenvector problem can be reformulated as follows: Ax = dxI > Ax— xd =0 5 (A-ADx =0 (1.24) A trivial solution to this equation is x = 0. For a non-trivial solution, we require that the determinant |A — AIj equals zero. One way to determine the eigenvalues and their asso- ciated eigenvectors is thus to expend the determinant to give a polynomial equation in A. For our 3 x 3 symmetric matrix this gives: (ey te 5-d 0 (1.25) -2 0 3-a or: (4A) — A) — d) — 2[2(3 — d)] - 2726 — d] =0 (1.26) This can be factorised to give: (1-A)(7—A(4—-d) =0 (1.27)16 Chapter 1 ‘The eigenvalues are thus A = 1, 2 = 4, As =7. The corresponding eigenvectors are: 2/3 -1/3 2/3 M=tim =] -18] %=4:m=] 2/8] 2=7:45=] 2/3] (128) 2/3 2/3 -1/3 Here we have expressed the eigenvectors as vectors of unit length; any multiple of each eigenvector would also be a solution. A is a real, symmetric matrix. The eigenvalues of such matrices are always real and orthogonal (i.e. the scalar products of all pairs of eigen- vectors are zero). This can be easily seen in our example. ‘Ascan be readily envisaged, expanding the determinant and solving a polynomial in ) is not the most efficient way to determine the eigenvalues and eigenvectors of larger matrices. Matrix diagonalisation methods are much more common. Diagonalisation of a matrix A involves finding a matrix U such that: uTAU=D (1.29) D is the diagonal matrix of eigenvalues. When A is a real symmetric matrix, then U is the matrix of eigenvectors and U™" is the inverse matrix of eigenvectors. Thus, for our example: 2/3 -1/3 2/3 22 2/3 1/3 2/3 1/3 2/3 2/3 275) 01) 1/3) 2/3 2/3 2/3 2/3) 1/3) \ 2 0 8 2/3 2/3 1/3 10-0 =|0 40 (1.30) 007 Note that for a real symmetric matrix A, the inverse U~' is the same as the transpose, U". Many methods have been devised for diagonalising matrices; some of these are specific to certain classes of matrices such as the class of real symmetric matrices. Many modelling techniques require us to calculate the eigenvalues and eigenvectors of a matrix, including self-consistent field quantum mechanics (Section 2.5), the distance geometry method for exploring conformational space (Section 9.5) and principal components analysis (Section 9.13.1). The class of positive definite matrices is important in energy minimisation and when finding transition structures; the eigenvalues of a positive definite matrix are all posi- tive. A positive semidefinite matrix of rank m has m positive eigenvalues. 1.10.4 Complex Numbers 4 complex number has two components: a real part (a) and an imaginary part (b), as follows: xsatbi (1.31)Useful Concepts in Molecular Modelling 7 imagmary Fig. 110 The Argand diagram used to represent complex numbers iis the square root of —1 (i = V—1). Complex numbers enable certain types of equation that have no real solutions to be solved. For example, the roots of the equation x7 — 2x +3 = Oare x=1+ V2i and x =1- Vi. A complex number can be considered as a vector in a two- dimensional coordinate system. Complex numbers are commonly represented using an Argand diagram, in which the x coordinate corresponds to the real part of the complex number and the y coordinate to the imaginary part (Figure 1.10). Arithmetical operations on complex numbers are performed much as for vectors. Thus, if x=a-+biand y=c+di, then: xty=(ate)t(b+d)i (1.32) x-y=(@-0)+(b—d)i (1.33) xy = (ac ~ bd) + (ad + be)i (1.34) The complex conjugate, %, equals a ~ bi and is obtained by reflecting x in the real axis in the Argand diagram. A commonly used relationship involving complex numbers is: e” = cos0 + isind (1.35) where 9 is any real number. This relationship is used in Fourier analysis and can be derived from the expansions of the exponential, cosine and sine functions: i P @ a Pa1+i0— a atat (1.36) : & sind = 6~ 3 (1.37) coso=1-& (1.38)18 Chapter 1 Various other relationships can be defined. For example: &te® cos @ = 2 (1.39) 1.10.5 Lagrange Multipliers Lagrange multipliers can be used to find the stationary points of functions, subject to a set of constraints. Suppose we wish to find the stationary points of a function f(x,y) = 4x? 43x +2y? +6y subject to the constraint y=4x+2. In the Lagrange method the constraint is written in the form g(x,y) = 0: g(tsy) =y—4x-2=0 (1.40) To find stationary points f(x,y) subject to g(x,y) = 0 we first determine the total derivative df, which is set equal to zero: a = Pete Fy = (8x43) a + by +6) dy =0 (1.41) Without the constraint the stationary points would be determined by setting the two partial derivatives Of /Ax and df /By equal to zero, as x and y are independent. With the constraint, x and y are no longer independent but are related via the derivative of the constraint function g: 88 ay 5 8 = Bax +3 dy = —4dx + dy =0 1.42) dg at ay + dy (1.42) The derivative of the constraint function, dg, is multiplied by a parameter (the Lagrange multiplier) and added to the total derivative df: (ZorX) ac (Z+azt) dy =0 (1.43) The value of the Lagrange multiplier is obtained by setting each of the terms in parentheses to zero. Thus for our example we have: 8x+3-4\=0 (1.44) (1.48) 4yt+6+A= From these two equations we can obtain a further equation linking x and y: = 2x43/4=-6—4y or x=—27/8—2y (1.46) Combining this with the constraint equation enables us to identify the stationary point, which is at (59/72, -23/18). This simple example could, of course, have been solved by simply substituting the constraint equation into the original function, to give a function of just one of the variables. However, in many cases this is not possible. The Lagrange multiplier method provides a powerful approach which is widely applicable to problems involving constraints such as in constraint dynamics (Section 7.5) and in quantum mechanics.Useful Concepts in Molecular Modelling : 4.10.6 Multiple Integrals Many of the theories used in molecular modelling involve multiple integrals. Examples include the two-electron integrals found in Hartree-Fock theory, and the integral over the positions and momenta used to define the partition function, Q. In fact, most of the multiple integrals that have to be evaluated are double integrals, ‘A‘traditional’ or one-dimensional integral corresponds to the area under the curve between the imposed limit, as illustrated in Figure 1.11. Multiple integrals are simply extensions of these ideas to more dimensions. We shall illustrate the principles using a function of two variables, f(x,y). The double integral | faxavrey =| [ fenacay (1.47) “A A is the sum of the volume elements f(x, y)5x éy (see Figure 1.11) over the area A as 6x and 6y tend to zero Note that the ‘dx dy’ can be put either immediately after the integral sign or at the end; in this book we often use the first method for multiple integrals. Some multiple integrals can be written as a product of single integrals. This occurs when f(z,y) is itself a product of functions g(x)h(y), in which case the integral can be separated: | Jesavsceren = farce) fayncy) 4s) h yah) area =f(x)8x total area = [f (x)de FOy) total volume = [If (x, y)dxdy y -volume element = f (x, y)Sxdy Fg. 111. Single and double integrals. (Figure adapted in part from Boas M L, 1983, Mathematical Methods in the Physical Sciences. 2nd Edition. New York, Wiley )20 Chapter 1 For example: 1 pif : a oe Cn eve haf vas cosy = [ #ax[siny]773 =2( 7 y ‘ (1.49) We will use the separation of multiple integrals throughout our discussion of quantum mechanics and computer simulation methods (Chapters 2, 3, 6, 7 and 8). 1.10.7 Some Basic Elements of Statistics Statistics is concerned with the collection and interpretation of numerical data. The subject is a vast and complex one, and all we shall do here is to state some of the definitions commonly used and to explain some of the terminology. The arithmetic mean of a set of observations is the sum of the observations divided by the number of observations: aes | Vx (1.50) Nis the number of observations, The mean may also be written (x). The variance, o, indicates the extent to which the set of observations cluster around the mean value and equals the average of the squared deviations from the mean: e=42S,-2 1.51 ye) (151) The variance can also be calculated using the following formula, which may be more convenient: Pay [Se -a (3 «)| (1.52) NAM) NLA : The standard deviation, c, equals the (positive) square root of the variance: ay o=\5>Gi-0 (1.53) It is often desired to compare the distribution of observations in a population with a theore- tical distribution. The normal distribution (also called the Gaussian distribution) is a particu- larly important theoretical distribution in molecular modelling. The probability density function for a general normal distribution is: f= The factor before the exponential ensures that the integral of the function f(x) from —oo to +co equals 1. The distribution is often written in terms of a parameter a: fey= 2 2 exp[-(x — 3)°/20°] (1.54) om o ~a(e—2)? (1.55)Useful Concepts in Molecular Modelling 21 Fig 112. Three normal distributions with different values of « (Equation (155)) The functions are normalised, so the area undei each curve is the same, In Figure 1.12 we show three normal distributions that all have zero mean but different values of the variance (0). A variance larger than 1 (small a) gives a flatter function and a variance less than 1 (larger a) gives a sharper function. 1.10.8 The Fourier Series, Fourier Transform and Fast Fourier Transform Consider a periodic function x({) that repeats between t = —r/2.and ! = +7/2 (ie. has period 7). Even though x(!) may not correspond to an analytical expression it can be written as the superposition of simple sine and cosine functions or Fourier series, Figure 1.13. X(E) = ay + a1 COS wt + My COS Wwf +++» + by Sinwyt + by sin Qwot +--+ (1.56) x(1) = 9 + Y>(a, cos nut + b, sin nuit) (1.57) a1 4% is related to the period of the function by wy = 2x/r and to the frequency of the function by ui) = 271. The frequencies of the contributing harmonics are thus 111 and are separated by 1/r. The coefficients a, and b, can be obtained as follows: 1 Sal x(#) dt (1.58) = a2)" x0 x(B) cos(2nnx/r) dx (159) 2 m=2) x0 #) sin(2nmx/7) dx (1.60)22 Chapter 1 (0 25sin(2s) ~ 0.1sin(41) — 0 O3sin(61) + 0 O4sin(8) fo Fig. 1.13 In a Fourier series a periodic function is expressed as a sum of sine and cosine functions An alternative way to express a Fourier series makes use of the following relationships: sinwyt = [exp (iwot) — exp(—itvot)]/2i (1.61) cos wyt = [exp (it) + exp(—iut)]/2 (1.62) ‘The Fourier series is then written 4x x(t) = Soe, exp(inayt) (1.63) with ape = 8 inwot) dt 1 on ron )exp(inupf) at (1.64) ‘The Fourier series is used to represent a function that is periodic with period 7 in terms of frequencies nuiy — 2n/r. The Fourier transform is used when the function has no periodicity. ‘There is a close relationship between the Fourier series and the Fourier transform. One way to demonstrate the gradual change from a Fourier series to a Fourier transform is to consider how the distribution of contributing frequencies changes as the period increases. This is illustrated in Figure 1.14, where the period of a square wavefunction is gradually increased. Also shown are the frequency contributions. It can be seen that an increasing number of frequency components is needed to describe the function as the period increases, and that when the period is infinite, the frequency spectrum is continuous. ‘The Fourier transform relationship between a function x(#) and the corresponding frequency function X(v) is: x(t) = { X(v) exp(2aivt) dv (1.65)Useful Concepts in Molecular Modelling 3 h A dh > Lp — t ie A A lh - [ie <>! 7 7 i: A i ; Wy ww el, oy 3 i. A So 7 Fig. 1.14: The connection between the Fourier transform and the Fourier series can be established by gradually increasing the period of the function When the period is infinite a continuous spectrum is obtained (Figure adapted from Ramirez R W, 1985, The FFT Fundamentals and Concepts, Englewood Cliffs, NJ, Prentice Hall.) The frequency function X(v) is given by: XV) = ie x(f) exp(—2mivt) dt (1.66) In practical applications, x(f) is not a continuous function, and the data to be transformed are usually discrete values obtained by sampling at intervals. Under such circumstances, the discrete Fourier transform (DFT) is used to obtain the frequency function, Let us suppose that the time-dependent data values are obtained by sampling at regular intervals separated by 6t and that a total of M samples are obtained (starting at t= 0). From M samples, a total of M frequency coefficients can be obtained using the DFT expression24 Chapter 1 [Press et al. 1992]: Maa X(kév) = 64 S~ x(n6t) exp[-2rink/M] (167) = Here, x(nét) (n = 0,1,...,M —1) are the experimental values obtained and X(kév) is the set of Fourier coefficients (k = 0,1,...,M—1). The separation between the frequencies, 6v, depends on the number of samples and the time between samples: 6v = 1/M6t. An expres- sion for converting frequency data into the time domain is also possible: Mot x(nbt) = 4 S> X(kbv) exp[2rink/M] (1.68) k=0 To compute each Fourier coefficient X (kST) (of which there are M) it is therefore necessary to evaluate the summation 75! x(nét) exp[—2nink/M] for that value of k. There will be M terms in the summation. A simple algorithm to determine the frequency spectrum would scale with the square of the number of measurements, M. This is a severe limitation, for many problems involve an extremely large number of pieces of data. It is for this reason that the fast Fourier transform (FFT) (ascribed to Cooley and Tukey [Cooley and Tukey 1965] but, in fact, using methods developed much earlier) has made such an impact. The FFT algorithm scales as Mn M. With the FFT algorithm it is possible to derive the Fourier transforms, even with a considerable number of data points. Further Reading Bachrach S M 1996. The Internet: A Guide for Chemists Washington, D.C., American Chemical Society Boas M L 1983. Mathematical Methods in the Physical Sciences. New York, John Wiley & Sons. Grant G H and W G Richards 1995. Computational Chemistry Oxford, Oxford University Press. Goodman J M 1998. Chemical Applications of Molecular Modelling. Cambridge, Royal Society of Chemistry. Leach A R 1999, Computational Chemistry and the Virtual Laboratory. In The Age of the Molecule. Cambridge, Royal Society of Chemistry. Lipkowitz K B and D B Boyd (Editors) 1990-. Reviews in Computational Chemistry Vols 1-. New York, VCH. Ramirez R W 1985. The FFT Fundamentals and Concepts. Englewood Chifs, NJ, Prentice Hall, Schleyer, Pv R, NL Allinger ,T Clark, J Gasteiger, P A Kollman, H F Schaefer IIl and P R Schreiner 1998. The Encyclopedia of Computational Chemistry. Chichester, John Wiley & Sons. Stephenson G 1973. Mathematical Methods for Science Students. London, Longman. Winter M J, HS Rzepa and BJ Whitaker 1995, Surfing the Chemical Net. Chemistry in Britain 31: 685-689 and http://www.ch.ic.ac.uk/rzepa/cib/ References Bolin J T, D J Filman, D A Matthews, RC Hamlin and J Kraut 1982. Crystal Structures of Escherichia coli and Lactobacillus casei Dihydrofolate Reductase Refined at 1.7 Angstroms Resolution. I. Features and Binding of Methotrexate Journal of Biological Chemistry 257.13650-13662.Useful Concepts in Molecular Modelling 25 Connolly M L 1983a, Solvent-accessible Surfaces of Proteins and Nucleic Acids. Scene 221 709-713 Connolly ML 1983b Analytical Molecular Surface Calculation. journal of Applied Crystallography 16 548 558 Cooley J W and J W Tukey 1965. An Algorithm for the Machine Calculation of Complex Fourier Series ‘Mathematics of Computation 19:297-301. Lee B and F M Richards 1971, The Interpretation of Protein Structures: Estimation of Static Accessibil- ity. Journal of Molecular Biology 55:379-400. Murray-Rust P and H Rzepa 1999. Chemical Markup, XML, and the Worldwide Web. 1 Basic Principles. Journal of Chemical Information and Computer Science 39.923-942 Press W H, B P Flannery, S A Teukolsky and W T Vetterling 1992. Numerical Recipes in Fortran, Cambridge, Cambridge University Press. Richards F M 1977. Areas, Volumes, Packing and Protein Structure Annual Review in Biophysics and Bioengineering 6:151-176. Richmond T J 1984. Solvent Accessible Surface Area and Excluded Volume in Proteins Journal of ‘Molecular Biology 178:63-88. Rouvray D 1997. Do Molecular Models Accurately Reflect Reality? Chemist in Industry 15'587-590. Rouvray D 1999. Model Answers Chemistry in Britain 35:30-32. Rouyray D 2000 Atoms as Hard Spheres. Chemistry in Britain 36:25.An Introduction to Computational Quantum Mechanics 2.1 Introduction Our aim in this chapter will be to establish the basic elements of those quantum mechanical methods that are most widely used in molecular modelling. We shall assume some familiarity with the elementary concepts of quantum mechanics as found in most ‘general’ physical chemistry textbooks, but little else other than some basic mathematics (see Section 1.10). There are also many excellent introductory texts to quantum mechanics, In Chapter 3 we then build upon this chapter and consider more advanced concepts. Quantum mechanics does, of course, predate the first computers by many years, and it is a tribute to the pioneers in the field that so many of the methods in common use today are based upon their efforts, The early applications were restricted to atomic, diatomic or highly symmetri- cal systems which could be solved by hand. The development of quantum mechanical techniques that are more generally applicable and that can be implemented on a computer (thereby eliminating the need for much laborious hand calculation) means that quantum mechanics can now be used to perform calculations on molecular systems of real, practical interest. Quantum mechanics explicitly represents the electrons in a calculation, and so it is possible to derive properties that depend upon the electronic distribution and, in particular, to investigate chemical reactions in which bonds are broken and formed. These qualities, which differentiate quantum mechanics from the empirical force field methods described in Chapter 4, will be emphasised in our discussion of typical applications. There are a number of quantum theories for treating molecular systems. The first we shall examine, and the one which has been most widely used, is molecular orbital theory. However, alternative approaches have been developed, some of which we shall also describe, albeit briefly. We will be primarily concerned with the ab initio and semi-empirical approaches to quantum mechanics but will also mention techniques such as Hiickel theory and valence bond theory. An alternative approach to quantum mechanics, density functional theory, is considered in Chapter 3. Density functional theory has always enjoyed significant support from the materials science community but is increasingly used for molecular systems. Quantum mechanics is often considered to be a difficult subject, and a cursory glance at the following pages in this chapter may simply serve to reinforce that view! However, if followed carefully it is possible to see how models that are developed for very simpleAn Introduction to C omputational Quantum Mechanics 27 systems can be applied to much more complex systems. As a consequence our treatment does require some consideration of the mathematical background to the simplest and most common types of calculation. Our strategy in developing the underlying theory of molecular orbital quantum mechanical calculations is as follows. First, we revise some key features of quantum mechanics, including the hydrogen atom. We then discuss the functional form of an acceptable wavefunction for a molecular system and show how to calculate the energy of such a system from the wavefunction. This leads to the problem of determining the wavefunction itself and how this can be done using routine mathematical methods. We will then be in a position to understand how quantum mechanical calculations can be performed for ‘real’ systems and will have the background necessary to consider more advanced topics. ‘The starting point for any discussion of quantum mechanics is, of course, the Schrédinger equation. The full, time-dependent form of this equation is Pp 6 NI {-E(e+S+) +1 hu = Med (2.1) Equation (2.1) refers to a single particle (e.g, an electron) of mass m which is moving through space (given by a position vector r = xi + yj + zk) and time (1) under the influence of an external field ¥ (which might be the electrostatic potential due to the nuclei of a molecule). his Planck’s constant divided by 27 and iis the square root of —1. W is the wavefunction which characterises the particle’s motion; it is from the wavefunction that we can derive various properties of the particle. When the external potential Y” is independent of time then the wavefunction can be written as the product of a spatial part and a time part: W(x, ) = y(x)T(t). We shall only consider situations where the potential is independent of time, which enables the time-dependent Schrédinger equation to be written in the more familiar, time-independent form: Po vir = 2.2) { amv t }u EW(r) (2.2) Here, E is the energy of the particle and we have used the abbreviation V? (pronounced ‘del- squared’). eae ~ Ox2" By Ae It is usual to abbreviate the left-hand side of Equation (2.1) to #W, where # is the Hamiltonian operator: ve 23) we H= te (2.4) This reduces the Schrédinger equation to #W = EW. To solve the Schrédinger equation it is necessary to find values of E and functions such that, when the wavefunction is operated upon by the Hamiltonian, it returns the wavefunction multiplied by the energy. The Schrédinger equation falls into the category of equations known as partial differential eigen- value equations in which an operator acts on a function (the eigenfunction) and returns the28 Chapter 2 function multiplied by a scalar (the eigenvalue). A simple example of an eigenvalue equation is: wen 25) The operator here is d/dx. One eigenfunction of this equation is y = e* with the eigenvalue r being equal to a. Equation (2.5) is a first-order differential equation. The Schrédinger equation is a second-order differential equation as it involves the second derivative of W. A simple example of an equation of this type is é goo (26) The solutions of Equation (2.6) have the form y = A coskx + Bsinkx, where A, B and k are constants. In the Schrédinger equation W is the eigenfunction and E the eigenvalue. 2.1.1 Operators The concept of an operator is an important one in quantum mechanics. The expectation value (which we can consider to be the average value) of a quantity such as the energy, position or linear momentum can be determined using an appropriate operator. The most commonly used operator is that for the energy, which is the Hamiltonian operator itself, #”. The energy can be determined by calculating the following integral: _ [weve —ewdr co The two integrals in Equation (2.7) are performed over all space (i.e. from —co to +00 in the x, y and z directions). Note the use of the complex conjugate notation (% '), which reminds us that the wavefunction may be a complex number. This equation can be derived by pre- multiplying both sides of the Schrédinger equation, #'V = EW, by the complex conjugate of the wavefunction, W", and integrating both sides over all space. Thus: [erevdr= [wewar (28) Eis a scalar and so can be taken outside the integral, thus leading to Equation (2.7). If the wavefunction is normalised then the denominator in Equation (2.7) will equal 1. ‘The Hamiltonian operator is composed of two parts that reflect the contributions of kinetic and potential energies to the total energy. The kinetic energy operator is ia, Vv (29) and the operator for the potential energy simply involves multiplication by the appropriate expression for the potential energy. For an electron in an isolated atom or molecule the potential energy operator comprises the electrostatic interactions between the electron and the nucleus and the interactions between the electron and the other electrons For aAn Introduction to Computational Quantum Mechanics 29 single electron and a single nucleus with Z protons the potential energy operator is thus: Ze ~~ Gregr (210) Another operator is that for linear momentum along the x direction, which is ha > 2. i ox a The expectation value of this quantity can thus be obtained by evaluating the following integral: [vt gee py = Le (212) [ower 2.1.2 Atomic Units Quantum mechanics is primarily concerned with atomic particles: electrons, protons and neutrons. When the properties of such particles (e.g. mass, charge, etc.) are expressed in ‘macroscopic’ units then the value must usually be multiplied or divided by several powers of 10. It is preferable to use a set of units that enables the results of a calculation to be reported as ‘easily manageable’ values. One way to achieve this would be to multiply each number by an appropriate power of 10. However, further simplification can be achieved by recognising that it is often necessary to carry quantities such as the mass of the electron or electronic charge all the way through a calculation. These quantities are thus also incorporated into the atomic units. The atomic units of length, mass and energy are as follows: 1 unit of charge equals the absolute charge on an electron, |e| = 1.60219 x 10°C 1 mass unit equals the mass of the electron, m, = 9.10593 x 10 kg 1 unit of length (1 Bohr) is given by ay = 1?/4n7m,e? = 5.29177 x 10" m 1 unit of energy (1 Hartree) is given by E, = €”/4mepa = 4.35981 x 10-8] The atomic unit of length is the radius of the first orbit in Bohr’s treatment of the hydrogen atom. It also turns out to be the most probable distance of a 1s electron from the nucleus in the hydrogen atom. The atomic unit of energy corresponds to the interaction between two electronic charges separated by the Bohr radius. The total energy of the 1s electron in the hydrogen atom equals —0.5 Hartree. In atomic units Planck’s constant h = 2m and so h = 1. 2.1.3 Exact Solutions to the Schrédinger Equation ‘The Schridinger equation can be solved exactly for only a few problems, such as the particle in a box, the harmonic oscillator, the particle on a ring, the particle on a sphere and the hydrogen atom, all of which are dealt with in introductory textbooks. A common feature of these problems is that it is necessary to impose certain requirements (often called boundary30 Chapter 2 conditions) on possible solutions to the equation. Thus, for a particle in a box with infinitely high walls, the wavefunction is required to go to zero at the boundaries. For a particle on a ring the wavefunction must have a periodicity of 21 because it must repeat every traversal of the ring, An additional requirement on solutions to the Schrédinger equation is that the wavefunction at a point r, when multiplied by its complex conjugate, is the probability of finding the particle at the point (this is the Born interpretation of the wavefunction). The square of an electronic wavefunction thus gives the electron density at any given point. If we integrate the probability of finding the particle over all space, then the result must be Las the particle must be somewhere: fue dr=1 (2.13) dr indicates that the integration is over all space. Wavefunctions which satisfy this condition are said to be normalised. It is usual to require the solutions to the Schrddinger equation to be orthogonal: [unu.ar=0 (m#n) (214) A convenient way to express both the orthogonality of different wavefunctions and the normalisation conditions uses the Kronecker delta: | Wy U, dr = bin (2.15) When used in this context, the Kronecker delta can be taken to have a value of 1 if m equals n and zero otherwise. Wavefunctions that are both orthogonal and normalised are said to be orthonormal. 2.2 One-electron Atoms Inan atom that contains a single electron, the potential energy depends upon the distance between the electron and the nucleus as given by the Coulomb equation. The Hamiltonian thus takes the following form: Hees Zen H=-— Ve 21 2m” ~ 4neor oo In atomic units the Hamiltonian is: i 2 #a-5V-F (217) For the hydrogen atom, the nuclear charge, Z, equals +1. ris the distance of the electron from the nucleus. The helium cation, He", is also a one-electron atom but has a nuclear charge of +2. As atoms have spherical symmetry it is more convenient to transform the Schrédinger equation to polar coordinates r, @ and @, where r is the distance from the nucleus (located at the origin), @ is the angle to the z axis and ¢ is the angle from the x axis in the xy plane (Figure 2.1). The solutions can be written as the product of a radial function R(r), which depends only on r, and an angular function Y(6,q) called a spherical harmonic, whichAn Introduction to Computational Quantum Mechanics 31 v=1 sin @ sing x=rsin cos Fig. 21: The relationship between spherical polar and Cartesian coordinates depends on @ and ¢: Vyin = Runt) Yin(®, 9) (2.18) The wavefunctions are commonly referred to as orbitals and are characterised by three quantum numbers 1, m and I. The quantum numbers can adopt values as follows: a: principal quantum number: 0, 1, 2, .- I: azimuthal quantum number: 0, 1, ...(n — 1) m: magnetic quantum number: —1, —(I—1), ...0...(1- 1), The full radial function is: oS e fpwet P (5 JO Lust (P) (2.19) p=2Zr/nag, where ay is the Bohr radius.” The term in square brackets is a normalising factor. L'!}(p) is a special type of function called a Laguerre polynomial. We shall rarely be interested in any other than the first few members of the series; moreover, they simplify considerably if atomic units are used and we write them in terms of the orbital exponent ¢=Z/n. The first few members of the series for low values of n are given in Table 2.1 and are illustrated graphically in Figure 2.2. As can be seen, the radial part of the wave- function is a polynomial multiplied by a decaying exponential. The angular part of the wavefunction is the product of a function of @ and a function of ¢: Yin(9, 4) = Oim(8)Pm($) (2.20) ‘These functions are: (2.21) (2+ 1) C= [mbt]? Om 2 Crimi)” (cos6) (2.22) “Strictly, ay in this case is given by a = h”/x*ye, where pris the reduced mass, : = m,_M/(m, + M), M is the mass of the nucleus.32 Chapter 2 EE —————————————— a I Rut) 1 0 202 exp(—

You might also like

• Understanding Molecular Simulation - 2nd Edition - Frenkel

  

  100% (1)

  Understanding Molecular Simulation - 2nd Edition - Frenkel

  642 pages
• Molecular Modelling - Andrew R. Leach

  

  100% (1)

  Molecular Modelling - Andrew R. Leach

  785 pages
• Molecular Dynamics: Ben Leimkuhler Charles Matthews

  

  80% (5)

  Molecular Dynamics: Ben Leimkuhler Charles Matthews

  461 pages
• Lecture No. 1 - Computational Chemistry Tools & Application of Computational Methods

  

  100% (1)

  Lecture No. 1 - Computational Chemistry Tools & Application of Computational Methods

  36 pages
• Quantum Mechanics in Molecular Modeling

  

  No ratings yet

  Quantum Mechanics in Molecular Modeling

  34 pages
• Molecular Dynamics Simulations: Erik Lindahl

  

  No ratings yet

  Molecular Dynamics Simulations: Erik Lindahl

  24 pages
• Molecular Modeling of Proteins PDF

  

  100% (1)

  Molecular Modeling of Proteins PDF

  474 pages
• Molecular Modeling of Proteins PDF

  

  100% (1)

  Molecular Modeling of Proteins PDF

  474 pages
• Intro 2 Molecular Modelling & Molecular Mechanics

  

  50% (2)

  Intro 2 Molecular Modelling & Molecular Mechanics

  35 pages
• Field Quantization - Walter Greiner & Joachim Reinhardt

  

  No ratings yet

  Field Quantization - Walter Greiner & Joachim Reinhardt

  458 pages
• Comp Chem Lecture 1

  

  No ratings yet

  Comp Chem Lecture 1

  27 pages
• Introduction To Molecular Dynamics Simulation

  

  No ratings yet

  Introduction To Molecular Dynamics Simulation

  30 pages
• Quantum Electrodynamics of Strong Fields - Walter Greiner, Berndt Mueller, Johann Rafelski

  

  No ratings yet

  Quantum Electrodynamics of Strong Fields - Walter Greiner, Berndt Mueller, Johann Rafelski

  606 pages
• Modelling Molecular Structure 2nd Editio PDF

  

  No ratings yet

  Modelling Molecular Structure 2nd Editio PDF

  352 pages
• Molecular Modeling

  

  No ratings yet

  Molecular Modeling

  22 pages
• Molecular Dynamics Simulation

  

  100% (1)

  Molecular Dynamics Simulation

  49 pages
• Unit 4

  

  No ratings yet

  Unit 4

  47 pages
• Molecular Modelling

  

  No ratings yet

  Molecular Modelling

  91 pages
• MM Notes

  

  No ratings yet

  MM Notes

  75 pages
• CH1 Molecular Modeling

  

  No ratings yet

  CH1 Molecular Modeling

  68 pages
• MD MM

  

  No ratings yet

  MD MM

  82 pages
• Dicta at

  

  No ratings yet

  Dicta at

  74 pages
• Computer Aided Drug Design Unit 5

  

  No ratings yet

  Computer Aided Drug Design Unit 5

  12 pages
• Ufaz Mpcce FF - Ac

  

  No ratings yet

  Ufaz Mpcce FF - Ac

  79 pages
• Ufaz M1pcce MD - Ac

  

  No ratings yet

  Ufaz M1pcce MD - Ac

  62 pages
• Materiale Didattico:: Molecular Modelling For Beginners Alan Hinchliffe 2003-Wiley-VCH

  

  No ratings yet

  Materiale Didattico:: Molecular Modelling For Beginners Alan Hinchliffe 2003-Wiley-VCH

  45 pages
• Chem - DSE - Computational Methods and Molecular Modelling

  

  No ratings yet

  Chem - DSE - Computational Methods and Molecular Modelling

  6 pages
• BScthesis On MD

  

  No ratings yet

  BScthesis On MD

  46 pages
• HWS Computational PDF

  

  No ratings yet

  HWS Computational PDF

  34 pages
• Lecture 1 - Introduction

  

  No ratings yet

  Lecture 1 - Introduction

  38 pages
• An Introduction To Molecular Dynamics: EMBO, June 2016

  

  No ratings yet

  An Introduction To Molecular Dynamics: EMBO, June 2016

  40 pages
• Lecture 1

  

  No ratings yet

  Lecture 1

  53 pages
• Clase 5

  

  No ratings yet

  Clase 5

  52 pages
• Best Practices For Foundations in Molecular Simulations

  

  No ratings yet

  Best Practices For Foundations in Molecular Simulations

  53 pages
• Molecular Dynamics 3l5sb5ap

  

  No ratings yet

  Molecular Dynamics 3l5sb5ap

  40 pages
• Fundamentals of Atomistic Simulations Amitava Moitra: Raidighi College Raidighi

  

  No ratings yet

  Fundamentals of Atomistic Simulations Amitava Moitra: Raidighi College Raidighi

  46 pages
• Modeling Molecular Structures With Hyperchem: Computational Chemistry Laboratory

  

  No ratings yet

  Modeling Molecular Structures With Hyperchem: Computational Chemistry Laboratory

  13 pages
• Protin Prediction

  

  No ratings yet

  Protin Prediction

  36 pages
• Lecture 01y

  

  No ratings yet

  Lecture 01y

  24 pages
• Molecular Dynamics Simulations in Gromacs: Project Report

  

  No ratings yet

  Molecular Dynamics Simulations in Gromacs: Project Report

  36 pages
• Ligand Based Approaches Core Molecular Modeling

  

  No ratings yet

  Ligand Based Approaches Core Molecular Modeling

  32 pages
• MD Simu.... Theory

  

  No ratings yet

  MD Simu.... Theory

  49 pages
• Molecular Modeling and Energy Minimization Unveiling Protein Structures

  

  No ratings yet

  Molecular Modeling and Energy Minimization Unveiling Protein Structures

  30 pages
• MolecularModelling Unit 1

  

  No ratings yet

  MolecularModelling Unit 1

  20 pages
• Molecular Modelling Is A Collective Term That Refers To Theoretical Methods and Computational Techniques To

  

  No ratings yet

  Molecular Modelling Is A Collective Term That Refers To Theoretical Methods and Computational Techniques To

  18 pages
• Adv Limitation Methods

  

  No ratings yet

  Adv Limitation Methods

  14 pages
• AIChE Journal - 2009 - Maginn - From Discovery To Data What Must Happen For Molecular Simulation To Become A Mainstream

  

  No ratings yet

  AIChE Journal - 2009 - Maginn - From Discovery To Data What Must Happen For Molecular Simulation To Become A Mainstream

  7 pages
• Integration by The Method of Residues

  

  No ratings yet

  Integration by The Method of Residues

  20 pages
• Molecular Dynamics Simulations: Dr.M.Chandra Sekhar

  

  No ratings yet

  Molecular Dynamics Simulations: Dr.M.Chandra Sekhar

  24 pages
• Lecture 2 Molecular Modelling

  

  No ratings yet

  Lecture 2 Molecular Modelling

  17 pages
• Molecular Dynamics Simulation - NneomaAimsEssay

  

  No ratings yet

  Molecular Dynamics Simulation - NneomaAimsEssay

  41 pages
• CE 530 Molecular Simulation

  

  No ratings yet

  CE 530 Molecular Simulation

  20 pages
• 6TH Sem Damm and Drug Designing Notes Combined Bioinfo 2024

  

  No ratings yet

  6TH Sem Damm and Drug Designing Notes Combined Bioinfo 2024

  18 pages
• Stat Mech HW4

  

  No ratings yet

  Stat Mech HW4

  10 pages
• Molecular Simulation and Modelling: A Science Philosophical Paper

  

  No ratings yet

  Molecular Simulation and Modelling: A Science Philosophical Paper

  15 pages
• Molecular Modelling: Principles and Applications: Preface

  

  No ratings yet

  Molecular Modelling: Principles and Applications: Preface

  12 pages
• Cap 4 Pathria 1

  

  No ratings yet

  Cap 4 Pathria 1

  5 pages
• Gromacs Molecular Modeling Tutorial

  

  No ratings yet

  Gromacs Molecular Modeling Tutorial

  11 pages
• Molecular Modeling

  

  No ratings yet

  Molecular Modeling

  4 pages