Molecular processes in nature affect human health, the availability of resources and the Earth's climate. Molecular

*202*
*69*
*19MB*

*English*
*Pages 440
[439]*
*Year 2016*

- Author / Uploaded
- Kubicki
- James D(Editor)

*Table of contents : Title Page......Page 5Copyright......Page 6Contents......Page 7List of Contributors......Page 13Preface......Page 15 1.1 Introduction......Page 17 1.2 Essentials of Quantum Mechanics......Page 18 1.2.2 Fundamental Examples......Page 20 1.3.1 The Hartree and Hartree–Fock Approximations......Page 23 1.3.2 Density Functional Theory......Page 29 1.4.1 The Born–Oppenheimer Approximation......Page 33 1.4.2 Basis Sets and the Linear Combination of Atomic Orbital Approximation......Page 34 1.4.3 Periodic Boundary Conditions......Page 36 1.4.4 Nuclear Motions and Vibrational Modes......Page 37 1.5 From Quantum Chemistry to Thermodynamics......Page 38 1.5.1 Molecular Dynamics......Page 40 1.6 Available Quantum Chemistry Codes and Their Applications......Page 43 References......Page 44 2.1 Introduction......Page 49 2.2.1 The Non-bonded Interactions......Page 51 2.2.3 Polarisation Effects......Page 53 2.2.4 Reactivity......Page 55 2.2.5 Fundamentals of Coarse Graining......Page 56 2.3.1 Combining Rules Between Unlike Species......Page 58 2.3.2 Optimisation Procedures for All-Atom Force Fields......Page 59 2.3.3 Deriving CG Force Fields......Page 61 2.3.4 Accuracy and Limitations of the Fitting......Page 63 2.4.1 General Force Fields......Page 64 2.4.2 Force Field Libraries for Organics: Biomolecules with Minerals......Page 65 2.4.3 Potentials for the Aqueous Environment......Page 66 2.4.4 Current CGFF Potentials......Page 67 2.4.5 Multi-scale Methodologies......Page 69 2.5.1 Calcium Carbonate......Page 70 2.5.2 Clay Minerals......Page 72 2.5.4 Silica and Silicates......Page 76 2.5.5 Iron-Based Minerals......Page 77 2.6 Concluding Remarks......Page 79 References......Page 80 3.1 Introduction......Page 93 3.2.1 Translation Invariance and Periodic Boundary Conditions......Page 95 3.2.2 HF and KS Methods......Page 96 3.2.3 Bloch Functions and Local BS......Page 97 3.3 Structural Properties......Page 98 3.3.1 P–V Relation Through Analytical Stress Tensor......Page 99 3.3.2 P–V Relation Through Equation of State......Page 101 3.4.1 Evaluation of the Elastic Tensor......Page 102 3.4.3 Directional Seismic Wave Velocities and Elastic Anisotropy......Page 105 3.5 Vibrational and Thermodynamic Properties......Page 107 3.5.1 Solid-State Thermodynamics......Page 109 3.6 Modeling Solid Solutions......Page 111 3.7 Future Challenges......Page 114 References......Page 115 4.1 Introduction......Page 123 4.2 Overview of the Theoretical Methods and Approximations Needed to Perform AIMD Calculations......Page 125 4.3.1 Bulk Structural Properties......Page 129 4.3.2 Bulk Electronic Structure Properties......Page 134 4.4.1 Surface Structural Properties......Page 139 4.4.2 Electronic Structure in the Surface Region......Page 143 4.4.3 Water Adsorption on Surface......Page 145 4.5.1 CPMD Simulations of the Vibrational Structure of the Hematite (012)–Water Interface......Page 146 4.5.2 CPMD Simulations of Fe2+ Species at the Mineral–Water Interface......Page 148 Appendix......Page 150 A.1 Short Introduction to Pseudopotentials......Page 151 A.1.1 The Spin Penalty Pseudopotential......Page 153 A.2 Hubbard-Like Coulomb and Exchange (DFT+U)......Page 154 A.3 Overview of the PAW Method......Page 155 References......Page 159Chapter 5 Computational Isotope Geochemistry......Page 167 5.1 A Brief Statement of Electronic Structure Theory and the Electronic Problem......Page 168 5.2 The Vibrational Eigenvalue Problem......Page 170 5.3 Isotope Exchange Equilibria......Page 172 5.4 Qualitative Insights......Page 175 5.5 Quantitative Estimates......Page 176 5.6 Relationship to Empirical Estimates......Page 185 5.7 Beyond the Harmonic Approximation......Page 187 5.9 Summary and Prognosis......Page 188 References......Page 189 6.1 Introduction......Page 193 6.1.1 Review Examples of Molecular Modeling Applications in Organic and Contaminant Geochemistry......Page 195 6.2.1 Molecular Mechanics: Brief Summary......Page 200 6.2.2 Quantum Mechanics: Overview......Page 203 6.2.3 Molecular Modeling Techniques: Summary......Page 208 6.2.4 Models: Clusters, Periodic Systems, and Environmental Effects......Page 211 6.3 Applications......Page 212 6.3.1 Modeling of Surface Complexes of Polar Phenoxyacetic Acid-Based Herbicides with Iron Oxyhydroxides and Clay Minerals......Page 213 6.3.2 Modeling of Adsorption Processes of Polycyclic Aromatic Hydrocarbons on Iron Oxyhydroxides......Page 233 6.3.3 Modeling of Interactions of Polar and Nonpolar Contaminants in Organic Geochemical Environment......Page 236 6.4 Perspectives and Future Challenges......Page 243 Glossary......Page 245 References......Page 247 7.1 Introduction: Petroleum Geochemistry and Basin Modeling......Page 261 7.2.1 Thermal Maturity and Vitrinite Reflectance......Page 262 7.2.2 Rock-Eval Pyrolysis......Page 263 7.2.3 Kerogen Pyrolysis and Gas Chromatography Analysis......Page 264 7.2.4 Kinetic Modeling of Kerogen Pyrolysis......Page 265 7.3.1 Ab Initio Calculations of the Unimolecular C–C Bond Rapture......Page 269 7.3.2 Quantum Mechanical Calculations on Natural Gas 13C Isotopic Fractionation......Page 272 7.3.3 Deuterium Isotope Fractionations of Natural Gas......Page 274 7.3.4 Molecular Modeling of the 13C and D Doubly Substituted Methane Isotope......Page 276 References......Page 278 8.1 Introduction......Page 287 8.2.1 Ab Initio Molecular Dynamics and Density Functional Theory......Page 291 8.3 Calculation of the Surface Acidity from Reversible Proton Insertion/Deletion......Page 296 8.4 Theoretical Methodology for Vibrational Spectroscopy and Mode Assignments......Page 298 8.5 Property Calculations from AIMD: Dipoles and Polarisabilities......Page 300 8.6.1 Organisation of Water at Silica–Water Interfaces: (0001) α-Quartz Versus Amorphous Silica......Page 302 8.6.2 Organisation of Water at Alumina–Water Interface: (0001) α-Alumina Versus (101) Boehmite......Page 307 8.6.3 How Surface Acidities Dictate the Interfacial Water Structural Arrangement......Page 309 8.6.4 Vibrational Spectroscopy at Oxide–Liquid Water Interfaces......Page 311 8.6.5 Clay–Water Interface: Pyrophyllite and Calcium Silicate......Page 315 8.7 Some Perspectives for Future Works......Page 318 References......Page 320 9.1 Introduction......Page 327 9.1.2 Mineral–Organic Interactions......Page 329 9.2.1 Biominerals: Structure, Nucleation, and Growth......Page 330 9.2.2 Conformational Sampling in Modeling Biomineralization......Page 333 9.2.3 Force Field Benchmarking......Page 340 9.2.4 Ab Initio MD and Hybrid QM/MM Approaches......Page 341 9.3 Case Studies......Page 342 9.3.1 Apatite......Page 343 9.3.2 Calcite......Page 347 9.4 Concluding Remarks and Future Perspectives......Page 350 References......Page 351 10.1 Introduction......Page 357 10.2 Theoretical Background and Methods......Page 358 10.2.1 Calculation of Vibrational Frequencies......Page 360 10.2.2 Splitting of the Longitudinal Optical (LO) and Transverse Optical (TO) Modes......Page 362 10.2.3 Calculation of Infrared (IR) and Raman Peak Intensities and of the IR Dielectric Function......Page 363 10.2.4 Estimation of the Anharmonic Constant for X–H Stretching Modes......Page 365 10.2.5 Accuracy of Basis Set and Hamiltonian......Page 366 10.3 Examples and Applications......Page 368 10.3.1 Vibrational Properties of Calcium and Magnesium Carbonates......Page 369 10.3.2 A Complex Mineral: The IR Spectra of Ortho-enstatite......Page 375 10.3.3 Treatment of the OH Stretching Modes: The Vibrational Spectra of Brucite and Diaspore......Page 376 10.4 Simulation of Vibrational Properties for Crystal Structure Determination......Page 379 10.4.1 Proton Disorder in γ-AlOOH Boehmite......Page 380 References......Page 384 11.1 Introduction......Page 391 11.2.1 Potential Energy Surfaces......Page 395 11.2.2 Choice of Solvation Methods......Page 400 11.2.3 Activation Energies and Volumes......Page 402 11.2.4 Transition States and Imaginary Frequencies......Page 406 11.2.5 Rate Constants......Page 407 11.2.6 Types of Reaction Mechanisms......Page 409 11.3.1 Diffusion......Page 410 11.3.2 Ligand Exchange Aqueous Complexes......Page 411 11.3.4 Dissolution......Page 412 11.3.5 Nucleation......Page 414 11.4.1 Femtosecond Spectroscopy......Page 415 11.4.3 Roaming......Page 416 11.4.5 Reactive Force Fields......Page 417 References......Page 419Index......Page 431EULA......Page 439*

Molecular Modeling of Geochemical Reactions

Dedication

To see a World in a grain of sand… —William Blake To my wife, Doris, and son, Cody, who bring much joy to my life.

Molecular Modeling of Geochemical Reactions An Introduction

Edited by JAMES D. KUBICKI University of Texas at El Paso, USA

This edition first published 2016 © 2016 John Wiley & Sons, Ltd. Registered Office John Wiley & Sons, Ltd, The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, United Kingdom For details of our global editorial offices, for customer services and for information about how to apply for permission to reuse the copyright material in this book please see our website at www.wiley.com. The right of the author to be identified as the author of this work has been asserted in accordance with the Copyright, Designs and Patents Act 1988. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by the UK Copyright, Designs and Patents Act 1988, without the prior permission of the publisher. Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic books. Designations used by companies to distinguish their products are often claimed as trademarks. All brand names and product names used in this book are trade names, service marks, trademarks or registered trademarks of their respective owners. The publisher is not associated with any product or vendor mentioned in this book. Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best efforts in preparing this book, they make no representations or warranties with respect to the accuracy or completeness of the contents of this book and specifically disclaim any implied warranties of merchantability or fitness for a particular purpose. It is sold on the understanding that the publisher is not engaged in rendering professional services and neither the publisher nor the author shall be liable for damages arising herefrom. If professional advice or other expert assistance is required, the services of a competent professional should be sought. The advice and strategies contained herein may not be suitable for every situation. In view of ongoing research, equipment modifications, changes in governmental regulations, and the constant flow of information relating to the use of experimental reagents, equipment, and devices, the reader is urged to review and evaluate the information provided in the package insert or instructions for each chemical, piece of equipment, reagent, or device for, among other things, any changes in the instructions or indication of usage and for added warnings and precautions. The fact that an organization or Website is referred to in this work as a citation and/or a potential source of further information does not mean that the author or the publisher endorses the information the organization or Website may provide or recommendations it may make. Further, readers should be aware that Internet Websites listed in this work may have changed or disappeared between when this work was written and when it is read. No warranty may be created or extended by any promotional statements for this work. Neither the publisher nor the author shall be liable for any damages arising herefrom. Library of Congress Cataloging-in-Publication data applied for ISBN: 9781118845080 A catalogue record for this book is available from the British Library. Set in 10/12pt Times New Roman by SPi Global, Pondicherry, India

1 2016

Contents List of Contributors Preface 1

2

Introduction to the Theory and Methods of Computational Chemistry David M. Sherman

xi xiii 1

1.1 Introduction 1.2 Essentials of Quantum Mechanics 1.2.1 The Schrödinger Equation 1.2.2 Fundamental Examples 1.3 Multielectronic Atoms 1.3.1 The Hartree and Hartree–Fock Approximations 1.3.2 Density Functional Theory 1.4 Bonding in Molecules and Solids 1.4.1 The Born–Oppenheimer Approximation 1.4.2 Basis Sets and the Linear Combination of Atomic Orbital Approximation 1.4.3 Periodic Boundary Conditions 1.4.4 Nuclear Motions and Vibrational Modes 1.5 From Quantum Chemistry to Thermodynamics 1.5.1 Molecular Dynamics 1.6 Available Quantum Chemistry Codes and Their Applications References

1 2 4 4 7 7 13 17 17 18 20 21 22 24 27 28

Force Field Application and Development Marco Molinari, Andrey V. Brukhno, Stephen C. Parker, and Dino Spagnoli

33

2.1 Introduction 2.2 Potential Forms 2.2.1 The Non-bonded Interactions 2.2.2 The Bonded Interactions 2.2.3 Polarisation Effects 2.2.4 Reactivity 2.2.5 Fundamentals of Coarse Graining 2.3 Fitting Procedure 2.3.1 Combining Rules Between Unlike Species 2.3.2 Optimisation Procedures for All-Atom Force Fields 2.3.3 Deriving CG Force Fields 2.3.4 Accuracy and Limitations of the Fitting 2.3.5 Transferability

33 35 35 37 37 39 40 42 42 43 45 47 48

vi

Contents

2.4 Force Field Libraries 2.4.1 General Force Fields 2.4.2 Force Field Libraries for Organics: Biomolecules with Minerals 2.4.3 Potentials for the Aqueous Environment 2.4.4 Current CGFF Potentials 2.4.5 Multi-scale Methodologies 2.5 Evolution of Force Fields for Selected Classes of Minerals 2.5.1 Calcium Carbonate 2.5.2 Clay Minerals 2.5.3 Hydroxides and Hydrates 2.5.4 Silica and Silicates 2.5.5 Iron-Based Minerals 2.6 Concluding Remarks References 3 Quantum-Mechanical Modeling of Minerals Alessandro Erba and Roberto Dovesi 3.1 Introduction 3.2 Theoretical Framework 3.2.1 Translation Invariance and Periodic Boundary Conditions 3.2.2 HF and KS Methods 3.2.3 Bloch Functions and Local BS 3.3 Structural Properties 3.3.1 P–V Relation Through Analytical Stress Tensor 3.3.2 P–V Relation Through Equation of State 3.4 Elastic Properties 3.4.1 Evaluation of the Elastic Tensor 3.4.2 Elastic Tensor-Related Properties 3.4.3 Directional Seismic Wave Velocities and Elastic Anisotropy 3.5 Vibrational and Thermodynamic Properties 3.5.1 Solid-State Thermodynamics 3.6 Modeling Solid Solutions 3.7 Future Challenges References 4 First Principles Estimation of Geochemically Important Transition Metal Oxide Properties: Structure and Dynamics of the Bulk, Surface, and Mineral/Aqueous Fluid Interface Ying Chen, Eric Bylaska, and John Weare 4.1 Introduction 4.2 Overview of the Theoretical Methods and Approximations Needed to Perform AIMD Calculations 4.3 Accuracy of Calculations for Observable Bulk Properties 4.3.1 Bulk Structural Properties 4.3.2 Bulk Electronic Structure Properties 4.4 Calculation of Surface Properties 4.4.1 Surface Structural Properties

48 48 49 50 51 53 54 54 56 60 60 61 63 64 77 77 79 79 80 81 82 83 85 86 86 89 89 91 93 95 98 99

107 107 109 113 113 118 123 123

Contents

4.4.2 Electronic Structure in the Surface Region 4.4.3 Water Adsorption on Surface 4.5 Simulations of the Mineral–Water Interface 4.5.1 CPMD Simulations of the Vibrational Structure of the Hematite (012)–Water Interface 4.5.2 CPMD Simulations of Fe2+ Species at the Mineral–Water Interface 4.6 Future Perspectives Acknowledgments Appendix A.1 Short Introduction to Pseudopotentials A.1.1 The Spin Penalty Pseudopotential A.1.2 Projected Density of States from Pseudo-Atomic Orbitals A.2 Hubbard-Like Coulomb and Exchange (DFT+U) A.3 Overview of the PAW Method References 5

6

vii

127 129 130 130 132 134 134 134 135 137 138 138 139 143

Computational Isotope Geochemistry James R. Rustad

151

5.1 A Brief Statement of Electronic Structure Theory and the Electronic Problem 5.2 The Vibrational Eigenvalue Problem 5.3 Isotope Exchange Equilibria 5.4 Qualitative Insights 5.5 Quantitative Estimates 5.6 Relationship to Empirical Estimates 5.7 Beyond the Harmonic Approximation 5.8 Kinetic Isotope Effects 5.9 Summary and Prognosis Acknowledgments References

152 154 156 159 160 169 171 172 172 173 173

Organic and Contaminant Geochemistry Daniel Tunega, Martin H. Gerzabek, Georg Haberhauer, Hans Lischka, and Adelia J. A. Aquino

177

6.1 Introduction 6.1.1 Review Examples of Molecular Modeling Applications in Organic and Contaminant Geochemistry 6.2 Molecular Modeling Methods 6.2.1 Molecular Mechanics: Brief Summary 6.2.2 Quantum Mechanics: Overview 6.2.3 Molecular Modeling Techniques: Summary 6.2.4 Models: Clusters, Periodic Systems, and Environmental Effects 6.3 Applications 6.3.1 Modeling of Surface Complexes of Polar Phenoxyacetic Acid-Based Herbicides with Iron Oxyhydroxides and Clay Minerals 6.3.2 Modeling of Adsorption Processes of Polycyclic Aromatic Hydrocarbons on Iron Oxyhydroxides

177 179 184 184 187 192 195 196 197 217

viii

Contents

6.3.3 Modeling of Interactions of Polar and Nonpolar Contaminants in Organic Geochemical Environment 6.4 Perspectives and Future Challenges Glossary References 7 Petroleum Geochemistry Qisheng Ma and Yongchun Tang 7.1 Introduction: Petroleum Geochemistry and Basin Modeling 7.2 Technology Development of the Petroleum Geochemistry 7.2.1 Thermal Maturity and Vitrinite Reflectance 7.2.2 Rock-Eval Pyrolysis 7.2.3 Kerogen Pyrolysis and Gas Chromatography Analysis 7.2.4 Kinetic Modeling of Kerogen Pyrolysis 7.2.5 Natural Gases and C/H Isotopes 7.3 Computational Simulations in Petroleum Geochemistry 7.3.1 Ab Initio Calculations of the Unimolecular C–C Bond Rapture 7.3.2 Quantum Mechanical Calculations on Natural Gas 13C Isotopic Fractionation 7.3.3 Deuterium Isotope Fractionations of Natural Gas 7.3.4 Molecular Modeling of the 13C and D Doubly Substituted Methane Isotope 7.4 Summary References 8 Mineral–Water Interaction Marie-Pierre Gaigeot and Marialore Sulpizi 8.1 Introduction 8.2 Brief Review of AIMD Simulation Method 8.2.1 Ab Initio Molecular Dynamics and Density Functional Theory 8.3 Calculation of the Surface Acidity from Reversible Proton Insertion/Deletion 8.4 Theoretical Methodology for Vibrational Spectroscopy and Mode Assignments 8.5 Property Calculations from AIMD: Dipoles and Polarisabilities 8.6 Illustrations from Our Recent Works 8.6.1 Organisation of Water at Silica–Water Interfaces: (0001) α-Quartz Versus Amorphous Silica 8.6.2 Organisation of Water at Alumina–Water Interface: (0001) α-Alumina Versus (101) Boehmite 8.6.3 How Surface Acidities Dictate the Interfacial Water Structural Arrangement 8.6.4 Vibrational Spectroscopy at Oxide–Liquid Water Interfaces 8.6.5 Clay–Water Interface: Pyrophyllite and Calcium Silicate 8.7 Some Perspectives for Future Works References

220 227 229 231 245 245 246 246 247 248 249 253 253 253 256 258 260 262 262 271 271 275 275 280 282 284 286 286 291 293 295 299 302 304

9 Biogeochemistry Weilong Zhao, Zhijun Xu, and Nita Sahai

311

9.1 Introduction 9.1.1 Mineral–Water Interactions 9.1.2 Mineral–Organic Interactions

311 313 313

Contents

9.2

Challenges and Approaches to Computational Modeling of Biomineralization 9.2.1 Biominerals: Structure, Nucleation, and Growth 9.2.2 Conformational Sampling in Modeling Biomineralization 9.2.3 Force Field Benchmarking 9.2.4 Ab Initio MD and Hybrid QM/MM Approaches 9.3 Case Studies 9.3.1 Apatite 9.3.2 Calcite 9.4 Concluding Remarks and Future Perspectives Acknowledgments References

10 Vibrational Spectroscopy of Minerals Through Ab Initio Methods Marco De La Pierre, Raffaella Demichelis, and Roberto Dovesi 10.1 Introduction 10.2 Theoretical Background and Methods 10.2.1 Calculation of Vibrational Frequencies 10.2.2 Splitting of the Longitudinal Optical (LO) and Transverse Optical (TO) Modes 10.2.3 Calculation of Infrared (IR) and Raman Peak Intensities and of the IR Dielectric Function 10.2.4 Estimation of the Anharmonic Constant for X–H Stretching Modes 10.2.5 Accuracy of Basis Set and Hamiltonian 10.3 Examples and Applications 10.3.1 Vibrational Properties of Calcium and Magnesium Carbonates 10.3.2 A Complex Mineral: The IR Spectra of Ortho-enstatite 10.3.3 Treatment of the O─H Stretching Modes: The Vibrational Spectra of Brucite and Diaspore 10.4 Simulation of Vibrational Properties for Crystal Structure Determination 10.4.1 Proton Disorder in γ-AlOOH Boehmite 10.5 Future Challenges Acknowledgements References 11 Geochemical Kinetics via Computational Chemistry James D. Kubicki and Kevin M. Rosso 11.1 Introduction 11.2 Methods 11.2.1 Potential Energy Surfaces 11.2.2 Choice of Solvation Methods 11.2.3 Activation Energies and Volumes 11.2.4 Transition States and Imaginary Frequencies 11.2.5 Rate Constants 11.2.6 Types of Reaction Mechanisms 11.3 Applications 11.3.1 Diffusion 11.3.2 Ligand Exchange Aqueous Complexes 11.3.3 Adsorption

ix

314 314 317 324 325 326 327 331 334 335 335 341 341 342 344 346 347 349 350 352 353 359 360 363 364 368 368 368 375 375 379 379 384 386 390 391 393 394 394 395 396

x

Contents

11.3.4 Dissolution 11.3.5 Nucleation 11.4 Future Challenges 11.4.1 Femtosecond Spectroscopy 11.4.2 H-Bonding 11.4.3 Roaming 11.4.4 Large-Scale Quantum Molecular Dynamics 11.4.5 Reactive Force Fields References Index

396 398 399 399 400 400 401 401 403 415

List of Contributors Adelia J. A. Aquino Institute for Soil Research, University of Natural Resources and Life Sciences, Vienna, Austria and Department of Chemistry and Biochemistry, Texas Tech University, Lubbock, TX, USA Andrey V. Brukhno Department of Chemistry, University of Bath, Bath, UK Eric Bylaska Environmental Molecular Sciences Laboratory, Pacific Northwest National Laboratory, Richland, WA, USA Ying Chen Chemistry and Biochemistry Department, University of California, San Diego, La Jolla, CA, USA Marco De La Pierre Nanochemistry Research Institute, Curtin Institute for Computation, and Department of Chemistry, Curtin University, Perth, Western Australia, Australia Raffaella Demichelis Nanochemistry Research Institute, Curtin Institute for Computation, and Department of Chemistry, Curtin University, Perth, Western Australia, Australia Roberto Dovesi Dipartimento di Chimica, Università degli Studi di Torino and NIS Centre of Excellence “Nanostructured Interfaces and Surfaces”, Torino, Italy Alessandro Erba Dipartimento di Chimica, Università degli Studi di Torino, Torino, Italy Marie-Pierre Gaigeot LAMBE CNRS UMR 8587, Université d’Evry val d’Essonne, Evry, France and Institut Universitaire de France, Paris, France Martin H. Gerzabek Institute for Soil Research, University of Natural Resources and Life Sciences, Vienna, Austria Georg Haberhauer Institute for Soil Research, University of Natural Resources and Life Sciences, Vienna, Austria James D. Kubicki Department of Geological Sciences, University of Texas at El Paso, El Paso, TX, USA Hans Lischka Department of Chemistry and Biochemistry, Texas Tech University, Lubbock, TX, USA and Institute for Theoretical Chemistry, University of Vienna, Vienna, Austria Qisheng Ma Department of Computational and Molecular Simulation, GeoIsoChem Corporation, Covina, CA, USA Marco Molinari Department of Chemistry, University of Bath, Bath, UK Stephen C. Parker Department of Chemistry, University of Bath, Bath, UK Kevin M. Rosso Physical Sciences Division, Pacific Northwest National Laboratory, Richland, WA, USA

xii

List of Contributors

James R. Rustad Corning Incorporated, Corning, NY, USA Nita Sahai Department of Polymer Science, Department of Geology, and Integrated Bioscience Program, University of Akron, Akron, OH, USA David M. Sherman School of Earth Sciences, University of Bristol, Bristol, UK Dino Spagnoli School of Chemistry and Biochemistry, University of Western Australia, Crawley, Western Australia, Australia Marialore Sulpizi Department of Physics, Johannes Gutenberg Universitat, Mainz, Germany Yongchun Tang Geochemistry Division, Power Environmental Energy Research Institute, Covina, CA, USA Daniel Tunega Institute for Soil Research, University of Natural Resources and Life Sciences, Vienna, Austria John Weare Chemistry and Biochemistry Department, University of California, San Diego, La Jolla, CA, USA Zhijun Xu Department of Polymer Science, University of Akron, Akron, OH, USA and Department of Chemical Engineering, Nanjing University, Nanjing, China Weilong Zhao Department of Polymer Science, University of Akron, Akron, OH, USA

Preface Humility is an underrated scientific personality characteristic. When I think of William Blake’s famous lithograph of Sir Isaac Newton toiling away at the bottom of a dark ocean, I am always reminded of how much we do not know. Science is a humbling enterprise because even our most notable achievements will likely be replaced by greater understanding at some date in the future. I am allowing myself an exception in the case of publication of this volume, however. I am proud of this book because so many leaders in the field of computational geochemistry have agreed to be a part of it. We all know that the best people are so busy with projects that it is difficult to take time away from writing papers and proposals to dedicate time to a chapter. The authors who have contributed to this volume deserve a great deal of appreciation for taking the time to help explain computational geochemistry to those who are considering using these techniques in their research or trying to gain a better understanding of the field in order to apply its results to a given problem. I am proud to be associated with this group of scientists. When my scientific career began in 1983, computational geochemistry was just getting a toehold in the effort to explain geochemical reactions at an atomic level. People such as Gerry V. Gibbs and John (Jack) A. Tossell were applying quantum chemistry to model geologic materials, and C. Austen Angell and coworkers were simulating melts with classical molecular dynamics. As an undergraduate, I had become interested in magmatic processes, especially the generation of magmas in subduction zones and the nucleation of crystals from melts. Organic chemistry exposed me to the world of reaction mechanisms which were not being studied extensively at the time in geochemistry. When the opportunity arose in graduate school to use MD simulations to model melt and glass behavior, I jumped at the chance to combine these interests in melts and mechanisms naïve to the challenges that lie ahead. Fortunately, through the guidance of people such as Russell J. Hemley, Ron E. Cohen, Anne M. Hofmeister, Greg E. Muncill, and Bjorn O. Mysen at the Geophysical Laboratory, I was able to complement the computational approach with experimental data on diffusion rates and vibrational spectra. This approach helped benchmark the simulations and provide insights into the problems at hand that were difficult to attain with computation alone. This strategy has worked throughout my career and has led to numerous fascinating collaborations. A key step in this process occurred while I was working as a postdoc at Caltech under Geoffrey A. Blake and Edward M. Stolper. I met another postdoc, Dan G. Sykes, who also shared a passion for melt and glass structure. As I was learning how to apply quantum mechanics to geochemistry, Dan and I discussed his models for explaining the vibrational spectra of silica and aluminosilicate glasses. Dan’s model differed from the prevailing interpretations of IR and Raman spectra, but his hypotheses were testable via construction of the three- and four-membered ring structures he thought gave rise to the observed trends in vibrational frequencies with composition. We argued constantly over the details of his model and came up with several tests to disprove it, but, in the end, the calculations and observed spectra agreed well enough that we were able to publish a series of papers over the objections of reviewers who were skeptical of the views of two young postdocs. Among these papers, a key study was published with the help of George R. Rossman whose patience and insight inspired more confidence in me that the path we were following would be fruitful. This simple paper comparing calculated versus observed H-bond frequencies ended up being more

xiv

Preface

significant than I had known at the time because this connection is critical in model mineral–water interactions that became a theme later in my career. When I could not find work any longer doing igneous-related research, I turned to a friend from undergraduate chemistry at Cal State Fullerton, Sabine E. Apitz, to employ me as a postdoc working on environmental chemistry. Fortunately, the techniques I had learned were transferable to studying organic–mineral interactions. This research involving mineral surfaces eventually led to contacts with Susan L. Brantley and Carlo G. Pantano who were instrumental in landing a job for me at Penn State. Numerous collaborations blossomed during my tenure in the Department of Geosciences, and all these interdisciplinary projects kept me constantly excited about learning new disciplines in science. Recently, I made the decision to move to the University of Texas at El Paso to join a team of people who are creating an interdisciplinary research environment while simultaneously providing access to excellent education and social mobility. The rapid developments in hardware, software, and theory that have occurred since 1983 have propelled research in computational geochemistry. All of us appreciate the efforts of all those developing new architectures and algorithms that make our research possible. We offer this book as a stepping stone for those interested in learning these techniques to get started in their endeavors, and we hope the reviews of literature and future directions offered will help guide many new exciting discoveries to come. James D. Kubicki Department of Geological Sciences The University of Texas at El Paso October 3, 2015

1 Introduction to the Theory and Methods of Computational Chemistry David M. Sherman School of Earth Sciences, University of Bristol, Bristol, UK

1.1

Introduction

The goal of geochemistry is to understand how the Earth formed and how it has chemically differentiated among the different reservoirs (e.g., core, mantle, crust, hydrosphere, atmosphere, and biosphere) that make up our planet. In the early years of geochemistry, the primary concern was the chemical analysis of geological materials to assess the overall composition of the Earth and to identify processes that control the Earth’s chemical differentiation. The theoretical underpinning of geochemistry was very primitive: elements were classified as chalcophile, lithophile, and siderophile (Goldschmidt, 1937), and the chemistry of the lithophile elements was explained in terms of simple models of ionic bonding (Pauling, 1929). It was not possible to develop a predictive quantitative theory of how elements partition among different phases. In the 1950s, experimental studies began to measure how elements are partitioned between coexisting phases (e.g., solid, melt, and fluid) as a function of pressure and temperature. This motivated the use of thermodynamics so that experimental results could be extrapolated from one system to another. Equations of state were developed that were based on simple atomistic (hard-sphere) or continuum models (Born model) of liquids (e.g., Helgeson and Kirkham, 1974). This work continued on into the 1980s. By this time, computers had become sufficiently fast that atomistic simulations of geologically interesting materials were possible. However, the computational atomistic simulations were based on classical or ionic models of interatomic interactions. Minerals were modeled as being composed of ions that interact via empirical or ab initio-derived interatomic potential functions (e.g., Catlow et al., 1982; Bukowinski, 1985). Aqueous solutions were composed of ions solvated by (usually) rigid water molecules modeled as point charges (Berendsen et al., 1987). Many of these simulations have been very successful and classical models of minerals and aqueous Molecular Modeling of Geochemical Reactions: An Introduction, First Edition. Edited by James D. Kubicki. © 2016 John Wiley & Sons, Ltd. Published 2016 by John Wiley & Sons, Ltd.

2

Molecular Modeling of Geochemical Reactions

solutions are still in use today. However, ultimately, these models will be limited in application insofar as they are not based on the real physics of the problem. The physics underlying geochemistry is quantum mechanics. As early as the 1970s, approximate quantum mechanical calculations were starting to be used to investigate bonding and electronic structure in minerals (e.g., Tossell et al., 1973; Tossell and Gibbs, 1977). This continued into the 1980s with an emphasis on understanding how chemical bonds dictate mineral structures (e.g., Gibbs, 1982) and how the pressures of the deep earth might change chemical bonding and electronic structure (Sherman, 1991). Early work also applied quantum chemistry to understand geochemical reaction mechanisms by predicting the structures and energetics of reactive intermediates (Lasaga and Gibbs, 1990). By the 1990s, it became possible to predict the equations of state of simple minerals and the structures and vibrational spectra of gas-phase metal complexes (Sherman, 2001). As computers have become faster, it now possible to simulate liquids, such as silicate melts or aqueous solutions, using ab initio molecular dynamics. We are now at the point where computational quantum chemistry can be used to provide a great deal on insight on the mechanisms and thermodynamics of chemical reactions of interest in geochemistry. We can predict the structures and stabilities of metal complexes on mineral surfaces (Sherman and Randall, 2003; Kwon et al., 2009) that control the fate of pollutants and micronutrients in the environment. We can predict the complexation of metals in hydrothermal fluids that determine the solubility and transport of metals leading to hydrothermal ore deposits (Sherman, 2007; Mei et al., 2013, 2015). We can predict the phase transitions of minerals that may occur in the Earth’s deep interior (Oganov and Ono, 2004; Oganov and Price, 2005). Computational quantum chemistry is now becoming a mainstream activity among geochemists, and investigations using computational quantum chemistry are now a significant contribution to work presented at major conferences on geochemistry. Many geochemists want to use these tools, but may have come from a traditional Earth science background. The goal of this chapter is to give the reader an outline of the essential concepts that must be understood before using computational quantum chemistry codes to solve problems in geochemistry. Geochemical systems are usually very complex and many of the high-level methods (e.g., configuration interaction) that might be applied to small molecules are not practical. In this chapter, I will focus on those methods that can be usefully applied to earth materials. I will avoid being too formal and will emphasize what equations are being solved rather than how they are solved. (This has largely been done for us!) It is crucial, however, that those who use this technology be aware of the approximations and limitations. To this end, there are some deep fundamental concepts that must be faced, and it is worth starting at fundamental ideas of quantum mechanics.

1.2 Essentials of Quantum Mechanics By the late nineteenth and early twentieth centuries, it was established that matter comprised atoms which, in turn, were made up of protons, neutrons, and electrons. The differences among chemical elements and their isotopes were beginning to be understood and systematized. Why different chemical elements combined together to form compounds, however, was still a mystery. Theories of the role of electrons in chemical bonding were put forth (e.g., Lewis, 1923), but these models had no obvious physical basis. At the same time, physicists were discovering that classical physics of Newton and Maxwell failed to explain the interaction of light and electrons with matter. The energy of thermal radiation emitted from black bodies could only be explained in terms of the frequency of light and not its intensity (Planck, 1900). Moreover, light (viewed as a wave since Young’s experiment in 1801) was found to have the properties of particles with discrete energies and momenta (Einstein, 1905). This suggests that light was both a particle and a wave. Whereas a classical particle could have any value for its kinetic and potential energies, the electrons bound to atoms were found

Introduction to the Theory and Methods of Computational Chemistry

3

to only have discrete (quantized) energies (Bohr, 1913). It was then hypothesized that particles such as electrons could also be viewed as waves (de Broglie, 1925); this was experimentally verified by the discovery of electron diffraction (Davisson and Germer, 1927). Readers can find an accessible account of the early experiments and ideas that led to quantum mechanics in Feynman et al. (2011). The experimentally observed wave–particle duality and quantization of energy were explained by the quantum mechanics formalism developed by Heisenberg (1925), Dirac (1925), and Schrodinger (1926). The implication of quantum mechanics for understanding chemical bonding was almost immediately demonstrated when Heitler and London (1927) developed a quantum mechanical model of bonding in the H2 molecule. However, the real beginning of computational quantum chemistry occurred at the University of Bristol in 1929 when Lennard-Jones presented a molecular orbital theory of bonding in diatomic molecules (Lennard-Jones, 1929). The mathematical structure of quantum mechanics is based on set of postulates: Postulate 1: A system (e.g., an atom, molecule or, really, anything) is described by a wavefunction Ψ(r1, r2, …, rN, t) over the coordinates rN , the N-particles of the system, and time t. The physical meaning of this wavefunction is that the probability of finding the system at a set of values for the coordinates r1, r2, …, rN at a time t is |Ψ(r1, r2, …, rN, t)|2. Postulate 2: For every observable (measurable) property λ of the system, there corresponds a mathematical operator L that acts on the wavefunction. Mathematically, this is expressed as follows: LΨ = λΨ

11

Ψ is an eigenfunction of the operator L with eigenvalue λ. An eigenfunction is a function associated with an operator such that if the function is operated on by the operator, the function is unchanged except for being multiplied by a scalar quantity λ. This is very abstract, but it leads to the idea of the states of a system (the eigenfunctions) that have defined observable properties (the eigenvalues). Observable properties are quantities such as energy, momentum, or position. For example, the operator for the momentum of a particle moving in the x-direction is p = iℏ

∂ i ∂x

12

where i is −1, ℏ is Planck’s constant divided by 2π, and i is the unit vector in the x-direction. Since the kinetic energy of a particle with mass m and momentum p is T=

p2 2m

the operator for the kinetic energy of a particle of mass m that is free to move in three directions (x, y, z) is T=

−ℏ2 ∂2 ∂2 ∂2 − ℏ2 2 ∇ = + + 2m ∂x2 ∂y2 ∂z2 2m

13

In general, the operator for the potential energy V of a system is a scalar operator such that V = V . That is, we multiply the wavefunction by the function that defines the potential energy. The operator Ê for the total energy E of a system is

4

Molecular Modeling of Geochemical Reactions

E = iℏ

∂ ∂t

14

It is important to recognize whether or not a quantity is a “quantum mechanical observable.” Chemists (and geochemists) often invoke quantities such as “ionicity,” “bond valence,” “ionic radius,” etc., that are not observables. These quantities are not real; they exist only as theoretical constructs. They cannot be measured.

1.2.1

The Schrödinger Equation

In classical mechanics, we express the concept of conservation of energy in terms of the Hamiltonian H of the system: H =E =T +V

15

In quantum mechanics, we express the Hamiltonian in terms of the operators corresponding to E, T, and V: HΨ = T + V Ψ = EΨ

16

or HΨ = T + V Ψ = iℏ

∂Ψ ∂t

17

This is the time-dependent Schrödinger equation. If the kinetic T and potential V energies of the system are not varying with time, then we can write: Ψ r1 , r2 ,…,rN ,t = Ψ r1 ,r2 ,…,rN e −iEt

ℏ

18

Substituting this into the Hamiltonian gives: HΨ = T + V Ψ = EΨ

19

This is the time-independent Schrödinger equation, and it is what we usually seek to solve in order to obtain a quantum mechanical description of the system in terms of the wavefunction and energy of each state.

1.2.2

Fundamental Examples

At this point, it is worthwhile to briefly explore several fundamental examples that illustrate the key aspects of quantum mechanics.

1.2.2.1

Particle in a Box

This is, perhaps the simplest problem yet it illustrates some of the fundamental features of quantum reality. Consider a particle of mass m inside a one-dimensional box of length L (Figure 1.1). The potential energy V of the system is 0 inside the box but infinite outside the box. Therefore, inside the box, the Schrödinger equation is

Introduction to the Theory and Methods of Computational Chemistry V(x) = infinity

V(x) = 0

n=4

V(x) = infinity

Ψ4

n=3

5

E = 16h2/(8 mL2)

Ψ3

Energy

E = 9h2/(8 mL2)

n=2

Ψ2

E = 4h2/(8 mL2) n=1

Ψ1

E = h2/(8 mL2) 0

Figure 1.1

x

L

E=0

Wavefunctions and energy levels for a particle in a one-dimensional box.

− ℏ2 d 2 Ψ x 2m dx2

= EΨ x

1 10

The solution to this differential equation is of the form: Ψ x = A sin kx + Bcos kx

1 11

Since the potential energy is infinite outside the box, the particle cannot be at x = 0 or at x = L. That is, we have Ψ 0 = Ψ L = 0. Hence, Ψ 0 = Asin 0 + Bcos 0 = 0 which implies that B = 0. However, since Ψ L = A sin kL = 0 we find that kL = nπ, where n = 1, 2, 3, …

1 12

6

Molecular Modeling of Geochemical Reactions

If we substitute Ψ(x) back into the Schrödinger equation, we find that E=

ℏ2 k 2 ℏ2 n 2 π 2 = 2m 2m L2

1 13

That is, the energy is quantized to have only specific allowed values because n can only take on integer values. The quantization results from putting the particle in a potential energy well (the box). However, the quantization is only significant if the dimensions of the box and the mass of the particle are on the order of Planck’s constant (h = 6.6262 × 10−34 J/s, i.e., if the box is angstroms to nanometers in size). The formalism of quantum mechanics certainly applies to our macroscopic world, but the quantum spacing of a 1 g object in a box of, say, 10 cm in length is too infinitesimal to measure. 1.2.2.2

The Hydrogen Atom

Now, let’s consider the hydrogen atom consisting of one electron and one proton as solved by Schrödinger (1926). We will consider only the motion of the electron relative to the position of the proton and not consider the motion of the hydrogen atom as a whole. Hence, our wavefunction for the system is Ψ(r) where r is the position of the electron (in three dimensions) relative to the proton (located at the origin). The Schrödinger equation for this problem is then −ℏ2 2 e2 ∇ Ψr − Ψ r = EΨ r 2m 4πε0 r

1 14

where e is the charge of the electron and ε0 is the permittivity of free space. To avoid having to write all the physical constants, it is convenient to adopt “atomic units,” where m, ℏ2, and e2/4πε0 are all set to unity. The unit of energy is now the Hartree (1 H = 27.21 eV = 4.360 × 10−18 J) and the unit of distance is the Bohr (10−10 m = 1 Å = 0.529177 Bohr). Computer programs in quantum chemistry often express their results in hartrees and bohrs; it is important that the user be aware of these units and know how to convert to more conventional units such as kJ/mol and angstrom (Å). In atomic units, the hydrogen Schrödinger equation becomes 1 1 − ∇2 Ψ r − Ψ r = EΨ r 2 r

1 15

Since the problem has spherical symmetry, it is more convenient to use spherical coordinates Ψ(r) = Ψ(r, θ, ϕ) rather than Cartesian coordinates (Figure 1.2). Since the coordinates are independent of each other, we can write Ψ r,θ, ϕ = Rnl r Ylm θ,ϕ

1 16

where Ylm (θ, ϕ) are the spherical harmonic functions; they give the angular shape of the wavefunctions. The radial wavefunction is

Rnl r =

2 n

3

n −l −1 − rln 2r l 2l + 1 2r e Ln −l − 1 2n n + l n n

1 17

2l + 1 where Ln− l − 1 are special functions called Laguerre polynomials. The important result that emerges from this solution is that the wavefunction can be specified by three quantum numbers n, l, and m with values

Introduction to the Theory and Methods of Computational Chemistry z

7

–e r

θ

ϕ +e

y

x

Figure 1.2 Spherical coordinates used for solution of the hydrogen atom.

n = 1,2, 3, … l = 0, 1, 2, …, n −1 m = −l, −l + 1,…,0, …,l −1, l The energy of the electron is quantized with values En = −

1 1 2 n2

1 18

This predicted quantization of the hydrogen electron energies triumphantly explains the empirical model proposed by Bohr for the energies of the lines observed hydrogen atom spectrum. It is standard practice to denote orbitals with l = 0 as “s” (not to be confused with the spin-quantum number described later), orbitals with l = 1 as “p,” orbitals with l = 2 as “d,” and orbitals with l = 3 as “f.” The three m quantum numbers for the p-orbitals are denoted as px , py , and pz . For the d-orbitals, we take linear combinations of the orbitals corresponding to the different m quantum numbers to recast them as dxy , dxz , dyz , d3z2 −r2 , and dx2 − y2 . The schematic energy-level diagram and the shapes of the hydrogenic orbitals are shown in Figure 1.3. Unfortunately, we cannot find an exact analytical solution for the Schrödinger equation for an atom with more than one electron. However, the hydrogenic orbitals and their quantum numbers enable us to rationalize the electronic structures of the multielectronic elements and the structure of the periodic table. As will be shown later, we will use the hydrogenic orbitals as building blocks to approximate the wavefunctions for multielectronic atoms.

1.3 1.3.1

Multielectronic Atoms The Hartree and Hartree–Fock Approximations

Consider the helium atom with two electrons and a nucleus of charge +2. The coordinate of electron 1 is r1 and the coordinate of electron 2 is r2. We will assume that the nucleus is fixed at the origin. The Schrödinger equation for the system is then

8

Molecular Modeling of Geochemical Reactions +

– –

+

+

+

–

3z2–r2

x2–y2

+

– x

–

+

–

+

–

+

+

–

+

–

+

–

xy

xz

+

+

–

–

y

z

yz

3d

3p

Energy

+ 3s +

+

+ –

–

–

x

y

z

2p

+ 2s

+ 1s

Figure 1.3

Schematic energy levels and orbital shapes for the hydrogen atom.

1 1 2 2 1 − ∇21 − ∇22 − − + 2 2 r1 r2 r1 −r2

Ψ r1 ,r2 = EΨ r1 ,r2

1 19

A reasonable approach to solving this might be to assume that Ψ r1 ,r2 = ψ a r1 ψ b r2

1 20

This is known as the Hartree approximation; it provides a very important conceptual reference point because it introduces the idea of expressing our many-body problem in terms of single-particle functions (“one-electron orbitals”). However, because of the interelectronic repulsion, described by the term 1 r1 −r2

1 21

the Hartree approximation is too crude to be quantitatively useful; we cannot really separate out the motions of the electrons. We say that the electrons are correlated. In spite of this shortcoming, we will still express the wavefunction for a multielectronic system in terms of single-particle wavefunctions. However, we must go beyond the pure Hartree approximation because, in a multielectronic system, the electrons must be indistinguishable from each other. This is more fundamental than stating that the electrons have identical mass, charge, etc. It means that observing one electron in a

Introduction to the Theory and Methods of Computational Chemistry

9

system is the same as observing any other electron. Hence, even if we ignore the interelectronic interaction, the wavefunction for a two-electron system must have the following form: Ψ r1 ,r2 = ψ a r1 ψ b r2 ± ψ a r2 ψ b r1

1 22

Here, we must digress: All fundamental particles have a property called “spin”; this is an intrinsic angular momentum with magnitude S = ℏ s s + 1 . For an electron, s = 1 2. The classical analogy is that an electron is a like a little sphere spinning on its axis; however, this is not what is really happening. Spin is a purely quantum mechanical phenomenon. Nevertheless, since spin is a type of angular momentum, it has a z-axis component, sz . However, sz can take on only quantized values of ms ℏ where ms = ± 1 2. If ms = + 1 2, we say the spin is “up” or α-spin; if ms = − 1 2, we say the spin is “down” or β-spin. Fundamental particles in the universe are either Fermions (with halfinteger values of s = 1 2, 3 2, …) or Bosons (with integer values of s = 0,1, 2, …). Fermions must have “antisymmetric” wavefunctions: Ψ r1 , r2 = ψ a r1 ψ b r2 −ψ a r2 ψ b r1

1 23

Electrons are Fermions and the electronic wavefunctions must have this antisymmetry. This is a formal and abstract way of stating the Pauli exclusion principle. The antisymmetry requirement means that no two electrons can be in the same quantum state or have the same single-particle orbital. An antisymmetric wavefunction obeys the Pauli exclusion principle since, if r1 = r2, then Ψ(r1,r2) = 0 ψ b r1 . If we build a multielectronic atom in terms of single-particle hydrogenic orbiunless ψ a r1 tals, the Pauli exclusion principle means that no two electrons can have the same four quantum numbers n,l, ml ,ms . From now on, instead of using ms , we will designate the spin of an electron by α or β. The spin coordinate of an electron will be accounted for by using separate wavefunctions for α- and β-spin electrons designated as ψ 1α , ψ 1β . Using separate wavefunctions for α- and β-spin electrons is referred to as a spin-unrestricted formalism. As we will see later, the two wavefunctions ψ 1α , ψ 1β will be numerically different if the number of α-spin electrons differs from the number of β-spin electrons because the interelectronic repulsion and electron experiences will depend on its spin. The construction of antisymmetric wavefunctions from the single-particle (hydrogenic) orbitals is much easier if we use the algebraic trick of expressing the wavefunction as the determinant of a matrix of the one-electron orbitals: Ψ r1 , r2 =

ψ 1 r1

ψ 2 r1

ψ 1 r2

ψ 2 r2

≡ ψ 1ψ 2

1 24

(on the right-hand side is a shorthand notation for the determinant). Or, for an N-electron atom,

Ψ r1 ,r2 ,…, rN =

ψ 1 r1

ψ 1 r2

ψ 1 rN

ψ 2 r1

ψ 2 r2

ψ 2 rN

ψ N r1

ψ N r2

≡ ψ 1 ψ 2 …ψ N

1 25

ψ N rN

These are called Slater determinants. If any two columns of a matrix are identical, the determinant of the matrix is zero. Hence, if any two electrons occupy the same orbital, we will have two columns to be the same and the determinant (and, hence, the wavefunction) will be zero. The Hartree–Fock approximation is that we can express our multielectronic wavefunction using a single-Slater determinant. This is an important starting point as it gives us a conceptual framework to

10

Molecular Modeling of Geochemical Reactions

understand electronic structure and the powerful concept of electron configuration. We use the Hartree–Fock approximation and construct a wavefunction for the multielectronic atom using the hydrogenic orbitals (1s, 2s, 2p, etc.) and populate those orbitals according to the Pauli exclusion principle. Hence, using the shorthand notation for a determinant, the Hartree–Fock wavefunction for Mg is β α β α β α ϕ1s ϕ2s ϕ2s ϕ2px ϕ2p ϕα ϕβ ϕα ϕβ ϕα ϕβ Ψ = ϕ1s x 2py 2py 2pz 2pz 3s 3s

1 26

which, for simplicity, is written as the electron configuration: 1s

2

2s

2

2p

6

3s

2

Although this is not a quantitative solution, the concept of electronic configuration is an immensely powerful tool in predicting the chemical behavior of the elements. The Hartree–Fock approximation also gives us a starting point in calculating the energies and wavefunctions of a multielectronic system. 1.3.1.1

The Variational Principle and the Hartree–Fock Equations

Suppose that our system is described by a Hamiltonian Ĥ and a wavefunction Ψ. The expectation value (the mean value of an observable quantity) of the total energy is given by E =

Ψ∗ HΨ∗ dr Ψ∗ Ψ∗ dr

where the asterisk (∗) means the complex conjugate (i.e., replace i by –i). Suppose that we did not know Ψ for a given Ĥ but had a trial guess for it (of course, this is true for nearly all of our problems). The variational principle states that the expectation value of the total energy we obtain from our trial wavefunction will always be greater than the true total energy. This is extremely useful because the problem becomes one of minimizing the total energy with respect to the trial wavefunction, and we will obtain the best approximation we can. Formally, we require that δE =0 δΨ The δ symbol refers to the functional derivative; a functional is a function of a function. The functional derivative comes from the calculus of variations; we will not go into the details here, but a discussion is given in Parr and Yang (1989). Now, suppose our unknown wavefunction Ψ is that of a multielectronic atom with N electrons and nuclear charge Z. If we use the Hartree–Fock approximation and express the wavefunction as a singleSlater determinant over N single-particle orbitals, then the expectation value of the total energy will be N

Ψ∗j r

E = j N

+ j Rb + > Cs + ; divalent cations usually decrease the swelling properties of the clay reducing the affinity to water. Most clay minerals have swelling properties and are able to uptake water and inorganic cations and organic molecules; hence they are used in applications where cation exchange is of particular importance. Clay minerals are widely used in subsurface disposal systems for spent nuclear fuel (Arcos et al., 2008; Dixon et al., 1985), drilling fluids (Anderson et al., 2010), cosmetics, detergents and, more recently, in nanotechnology and biotechnology (Aguzzi et al., 2007; Lin et al., 2002; Paul and Robeson, 2008; Ray and Bousmina, 2005). In terms of pollutant remediation, clays are able to remove both heavy metal ions (Celis et al., 2000) and, importantly, organic pollutants. It is clear that the success of such applications will depend on the morphologies and properties of the clay

Force Field Application and Development

57

minerals at the nanometre scale. As the experimental description of local interfaces is still limited, simulations become useful for gaining insights in describing local environment and computing dynamical properties at the nanoscale. Molecular computer simulations have been widely exploited to provide the atomistic descriptions of the structure and behaviour of clay minerals. Most of these have focussed on the swelling behaviour of smectite clays using either Monte Carlo or classical molecular dynamics. More rigorous QM methods have been applied to clay minerals, but they have largely been limited to simple clay structures (Alexandrov and Rosso, 2013; Austen et al., 2008; Zhang et al., 2012). Alternative strategies include fully ionic shell model potentials derived for the component oxides that have been applied to mica (Collins and Catlow, 1992) and pyrophyllite (Austen et al., 2008). The first models derived specifically for clay minerals were proposed by Boek et al. (1995) and Skipper et al. (1995) who applied them to understand the swelling of clay minerals using MC techniques. Later, Teppen et al. (1997) successfully derived a new model to reproduce a wider range of clay minerals including pyrophyllite, kaolinite and beidellite. The advantage of this model was the improved description of aluminium by the introduction of a bending O–Al–O for both the tetrahedral and octahedral coordinated aluminium atoms; however, the bonds between all the species must be identified and specified prior to the simulation. Highly parameterised models are generally available for phyllosilicate including the FF of Sainz-Diaz et al. for highly charged structures (Sainz-Diaz et al., 2001) and of Heinz et al. (2005). The latter is called phyllosilicate FF (PFF) and was originally developed for mica, montmorillonite and pyrophyllite. Although all the bonds between the atoms need to be specified, it has the great advantage to be fully compatible with organic PCFF, CVFF, CHARMM and GROMACS. The development of this FF was carried out since the first release, and improvements have led to the development of a large number of potential parameters into the INTERFACE FF (Heinz et al., 2013) for many inorganic compounds including clay minerals, silicates, aluminates, metals, sulfates and apatites. Cygan et al. took another approach to FF development; they developed CLAYFF. The FF is based on an ionic non-bonded description of the metal–oxygen interactions associated with hydrated phases, which gives full flexibility and transferability across different structures including layered hydroxides boehmite, portlandite and brucite and clay kaolinite, pyrophyllite, montmorillonite and hydrocalcite. CLAYFF has been successfully used to tackle different intriguing topics, such as the water transport properties in interparticle porosity of MTM, which depends on the pore water composition and the surface charge density (Churakov et al., 2014); the rotational disorder of distorted MTM, which leads to energetically favourable configurations at temperature and pressure relevant to geological carbon storage (Myshakin et al., 2014); and the behaviour of supercritical CO2 and aqueous fluids on both the hydrophilic and hydrophobic basal surfaces of kaolinite showing that the CO2 droplets do not interact directly with the gibbsite-like surface but do through a mixture of adsorbed CO2 and H2O molecules on the siliceous surface (Cygan et al., 2012; Tenney and Cygan, 2014). The list of simulation studies on the remediative properties of clays towards pollutants and heavy metals increases every year, and it drives the development of the FF, by adding compatible parameters or by modifying the current ones. For example, Greathouse et al. studied the adsorption of uranyl onto MTM, beidellite and pyrophyllite (Greathouse and Cygan, 2005, 2006; Greathouse et al., 2005; Zaidan et al., 2003), as clay minerals are used as sealant in nuclear waste repositories, using CLAYFF and a mix of potential parameters for the uranyl (Guilbaud and Wipff, 1996) and the carbonate (Greathouse et al., 2002). Adsorption of molecular species is not limited to inorganic complexes but can also concern organic nanoparticles and hazardous molecules. Zhu et al. studied the adsorption of C60, a carbon nanoparticle, onto pyrophyllite (Zhu et al., 2013) (Figure 2.8) and phenol molecules into

58

Molecular Modeling of Geochemical Reactions

Figure 2.8 Time-averaged image of water (blue) and C60 (yellow) densities above the pyrophyllite (001) surface (O in red, Si in blue, Al in light blue, H in white and C in gray). Adapted from Zhu et al. (2013). Reproduced with permission of American Chemical Society.

the nano-sized aggregates formed by CTMA alkyl chains in an organo-MTM (Zhu et al., 2011), while Shapley et al. studied the sorption of dioxins at clay–water interfaces (Shapley et al., 2013) (Figure 2.9) showing that the adsorption was due to the hydrophobic environment of the surface created by the organic cations. All these studies used a combination of potential-based and quantum methods; this has the great advantage of validating the potential-based results with higher accuracy simulations but more importantly of validating the performance and reliability of the FF. However, as DFT calculations are still limited to much smaller systems compared to the MM counterpart, the DFT is applied to a small subset of calculations while the broader picture of the dynamics is given by potential-based MD. Potential-based simulations are indeed utilised for studies on the structure and dynamics of intercalated molecules in clay minerals, which are important phenomena when dispersions in nanocomposites and soil properties are considered. As the systems can include higher level of structural and compositional complexity compared to the ab initio counterpart, potential-based simulations can be used to explore more complex phenomena. Heinz et al. studied alkylammonium-modified MTMs with different cation exchange capacity (CEC) finding that low CEC leads to stepwise increases of the basal plane spacing contrary to high CEC that leads to a continuous increase in basal plane spacing with increasing chain length (Heinz et al., 2006). Fu et al. studied the cleavage properties of 50 surfactant-modified clay minerals with different CECs (Fu and Heinz, 2010). The knowledge of mechanical properties of clays is of crucial importance for designing next-generation materials such as clay– polymer filler nanocomposites. Suter et al. estimated the bending modulus corresponding to the in-plane Young’s modulus of MTM (Suter et al., 2007), while Carrier et al. computed the elastic properties of hydrated MTM finding that the major factors controlling the stiffness of the

Force Field Application and Development

59

Figure 2.9 Adsorption of dioxin at the water–(001) surface of pyrophyllite interface (O in red, Si in blue, Al in light blue and H in white). Adapted from Shapley et al. (2013). Reproduced with permission of American Chemical Society.

material were the water content and the temperature (Carrier et al., 2014). More complex hybrid systems have been considered in the design of materials with tailored strength. Duque-Redondo et al. using CLAYFF and CHARMM found that the intercalation of organic dyes into the interlayer of laponite reduces the mechanical strength compared to the simple clay–water system and related this effect to the disruption of the hydrogen bonding network present in the interlayer space (Duque-Redondo et al., 2014). One of the areas that is ripe for exploitation in the future is the interface with biological molecules. There have already been some progress in the area of bio-composites (see Chapter 11). Bio-composites composed of the polysaccharide xyloglucan (XG) and MTM clay, which have potential as a ‘green’ replacement of conventional petroleum-derived polymers in the packaging industry, have been studied using CLAYFF and GLYCAM06. Wang et al. studied the molecular interaction between XG and MTM, which are responsible for the tensile properties (Wang et al., 2014). Bio-composites are also an important area of study in the ‘origin of life’ where simulations are performed to understand the possible chemical pathways to the formation of biomolecules pertaining to the relative adsorption of bio-macromolecules on mineral surfaces (see Chapter 12). Folding of RNA on MTM in the presence of charge balancing cations was studied by Swalding et al. using the CLAYFF and the AMBER FFs (Swadling et al., 2010). The RNA sequences fold to characteristic secondary structural motifs on the mineral surface that were not seen in the corresponding bulk water simulations. It is clear that all these examples have been made feasibly by the ease of combining different FFs.

60

2.5.3

Molecular Modeling of Geochemical Reactions

Hydroxides and Hydrates

Hydroxide minerals are another important class found in many soils, but they are also relevant in technological applications such as the cement industry. The interest in the latter has increased the need for developing transferable potential parameters for hydroxides and hydrates, particularly for Ca-silicates, aluminates and sulphates and, because of the water, a complex combination of hydrated phases (e.g. ettringite, tobermorites, calcium hydroxide and C-S-H). These mineral phases are not just important for the cement industry, but they represent a large group of naturally occurring minerals geochemically relevant. The C-S-H system is a poorly understood phase. Much of the work is based on modelling the mineral–water interfaces of minerals related to cement phase. Kalinichev et al. applied first CLAYFF (Kalinichev et al., 2007) to tobermorite, showing that on the (001) surface water shows strong structuring above the surface and in the channels between the drietkette silicate chains due to the development of an integrated H-bond network involving the water and the surface sites. The calculated diffusion coefficients for the surface-associated water were found to be in good agreement with published experimental results. CLAYFF was also applied to study the water structure in nanopores of brucite finding again highly structured water at the surface (Wang et al., 2004). One of the recent improvements in the CLAYFF is related to the vibrational modes of brucite. Zeitler et al. (2014) have modified the Mg–O–H interaction to gain a better comparison with the quantum mechanical counterpart, which also provided an improved description of the edge surfaces of the mineral and again demonstrates the application of DFT for refining potential models. Durability is an important property of cement-based material that determines the long-term behaviour of the material. The transport of water and ions in the nanopores of the calcium silicate hydrate (C-S-H) gel influences the durability of the cement. Hou et al. used jennite as a model to investigate the structural and dynamical properties of water and Na and Cl ions in a C-S-H system finding that the uptake of Cl is due to the formation of aggregates with Na and providing nanoscale interpretation of the 35Cl NMR and isotherm adsorption experimental studies (Hou and Li, 2014). Hydrated oxides are another class of minerals that has been often modelled. Shahsavari et al. (2011) compared the performances of a core–shell model (Pellenq et al., 2009) and CSH-FF, a modification of the CLAYFF, to simulate hydrated calcio-silicate materials; both gave good representation of the two modifications of tobermorite, but the rigid ion approach resulted clearly considerably less computational intensive than the core–shell model. The INTERFACE FF was, for example, used to further the understanding of the initial hydration and cohesive properties of cement particles made of calcium silicate, the major mineral phase in cements (Mishra et al., 2013). In this study, the authors quantified the cleavage energies, mechanical properties, the adsorption of organic additives and the agglomeration of calcium silicate cement particles, which are all related causes of the durability of cement-based materials. Finally, it is worth noting that the UFF, described earlier, has been used to simulate the interaction between ettringite and phosphate retarders on hydrating cements (Coveney and Humphries, 1996). Here again the challenge was the large amount of species involved in the system. 2.5.4

Silica and Silicates

Silica, SiO2, is the second most abundant mineral on the crust after feldspar, another silicate, and has a structure consisting of oxygen sharing SiO4 tetrahedra. Quartz, α and β, cristobalite and tridymite are important polymorphs of SiO2. They are topologically identical but geometrically distinct; β quartz and cristobalite are high-temperature polymorphs. With this variety of naturally occurring polymorphs, phase transitions in silica are therefore extensively studied using simulations. The amorphisation of α quartz has been also explained by the presence of elastic instabilities (Tse

Force Field Application and Development

61

and Klug, 1991; Watson and Parker, 1995a, 1995b) using the partial charge model of van Beest et al. (van Beest et al., 1990). Two related potential models, a rigid ion derived by Kramer et al. (1991) and a shell model presented by Sanders et al. (1984), were later used to study the transition between α cristobalite and β cristobalite and the disorder present in the β phase (Bourova et al., 2000). The contraction of the unit cell at high temperature was attributed to the decrease in the Si–Si distances because of the large thermal motion of oxygen atoms, which causes the neighbouring tetrahedral to come closer. The dynamical disorder in β cristobalite at high temperatures was also related to the structure to three possible modifications of the low-temperature α phase. Another example of a prediction from potential-based simulations is that zeolite-structured silicas show a negative thermal expansivity (Couves et al., 1993). On the other hand, as the interaction of water with silica is relevant in geological (Koretsky et al., 1997; Newton and Manning, 2008), biological (Tamerler et al., 2007) and technological (Pinto et al., 2009) contexts, the application of FFs to this material has been widely investigated. Quartz, the most abundant SiO2 polymorph in soil, is a challenging material to be studied in the presence of water. Quartz has a significant solubility in water, and with a point-of-zero charge reported between 2 and 4 (Kosmulski, 2002), it shows deprotonation of the surface silanol groups at pH above 2, leading to negatively charged surfaces. Different FFs have been used to model surfaces of quartz and silica and their interaction with water, including both shell model (de Leeuw et al., 1999), rigid ion models (Leung et al., 2006), potential parameters compatible with the CHARMM empirical FF by Lopes et al. (2006), the CHARMM water contact angle (CWCA) FF (Cruz-Chu et al., 2006; Lorenz et al., 2008) and reactive FF (Fogarty et al., 2010; Lockwood and Garofalini, 2009). However, one of the major drawbacks of is how to define the accuracy of the FF to describe the water structure above the silica surfaces. Comparison between results from FF and experiments and ab initio calculations is therefore crucial to assess the performances and ultimately improve the FF parameters if needed. Skelton et al. (2011) have performed a comparison between the water structure on the quartz (101-1) surface using three different FFs (CLAYFF, Lopes et al. and CWCA), ab initio technique and X-ray reflectivity. Overall, better agreement was found between the CLAYFF and the experimental results. Attention has also focused on silicates, which are the most abundant mineral in the subsurface and upper mantle. Peridotite is a rock mostly consisting of olivine (Mg, Fe)2SiO4 and pyroxene (Mg, Fe, Ca)2(Si, Al)2O6. Forsterite and fayalite are the Mg and Fe endmembers of olivine. In olivine, the silica tetrahedra are connected by divalent cations in contrast to silica polymorphs where the tetrahedral are polymerised. This structural feature causes weak mechanical resistance to weathering. The high cation content and the ease of breaking the oxygen–metal bond make this class of minerals highly reactive and particularly attractive to the carbon sequestration industry (Matter and Kelemen, 2009). The reactivity is not limited to CO2. Water is another relevant and topical molecule, and its transport through porous rocks has been investigated. For example, recently the structure and dynamics of forsterite –scCO2/H2O interfaces as a function of water content were studied by Kerisit et al. (2012). Their MD simulations suggested that, in the presence of sufficient water, Hx CO3 2− x − formation occurs in the water films and not via direct reaction of CO2 with the forsterite surface. 2.5.5

Iron-Based Minerals

Amongst many varieties of iron oxides, magnetite (Fe3O4) and hematite (α-Fe2O3) are the most important. Hematite contains only Fe3 +, unlike magnetite that comprises a mixture of Fe2 + and Fe3 +. While hematite forms in oxygen-rich environments, magnetite forms in anoxic environments, one amongst the other is magnetotactic bacteria, and alters to hematite in oxic environments.

62

Molecular Modeling of Geochemical Reactions (a)

(b)

Figure 2.10 Hematite (Fe2O3) nanoparticles comprising (a) the {01.2} (sides) and an oxygen terminated (00.1) (truncated corner) surfaces and (b) the {10.0} (sides) and iron-terminated (00.1) (top) surfaces.

Weathering of this Fe-rich mineral usually enhances oxidation and produces iron oxyhydroxide minerals goethite and lepidocrocite amongst others (see Chapter 7). Magnetite is also ferromagnetic while hematite is antiferromagnetic leading to many studies on the magnetism of these minerals. Bulk properties of the minerals are of extreme importance not only from a geochemical but also from a technological prospective. Chicot et al. (2011) have applied potential-based molecular dynamics and instrumented indentation to calculate the hardness and elastic properties of pure and complex (corrosion products) iron oxides, magnetite (Fe3O4), hematite (α-Fe2O3) and goethite (α - FeOOH) indicating good agreement between the calculated and the measured properties. Using the shell models of Lewis and Catlow (1985) and Woodley et al. (1999), a modification of these potential parameters was used also to provide the calculated infrared and Raman spectra of the minerals (Chamritski and Burns, 2005). Of particular note is the study of Kerisit and Rosso, who performed a comparative study of the rate exchange of charge transfer, applying Marcus theory (Marcus and Sutin, 1985) to wüstite (Kerisit and Rosso, 2005). Wüstite is FeO, and it forms in highly reducing conditions. Accounting for the polarisability of the system via shell models was essential in reproducing the ab initio charge transfer rates, compared to RIM counterpart (Bush et al., 1994). Surfaces have also been extensively studied. For example, Kundu et al. (2006) studied the inverse spinel magnetite and its bare, hydroxylated and hydrated surfaces, to determine the predominant faces in the crystal morphology, while Spagnoli et al. have studied pure-water and salt solutions on the hematite (001) surface (Spagnoli et al., 2006) and nanoparticles (Spagnoli et al., 2009) (Figure 2.10). There are also an increasing number of studies using computer simulations in combination with experiments; adsorption of water vapour on magnetite nanoparticles was simulated, using a combination of UFF and TIP4P, to be a nucleation-like process and forming welldistinguished hydration layers (Tombácz et al., 2009). Static FFs have been widely used, but with the advent of reactive FFs, the dynamical evolution of the mineral–water interfaces can be explored, for instance, to calculate the populations of the surface functional groups, the distribution of electrolyte in the solution and the surface hydrogen bonding on the magnetite (001) surface in contact with a pure and a NaClO4 water solutions (Rustad et al., 2003).

Force Field Application and Development

63

Fe-based minerals in the Earth’s subsurface are also involved in the ‘mineral iron respiration’, a process that a diversity of microorganisms use to obtain energy from a substrate (Richardson et al., 2013) (see Chapter 11). This anaerobic process involves electron transfer (see Chapter 7) between the organic metabolism and the inorganic Fe metabolite, which is reduced from the ferric to the ferrous state; however, the mechanisms underpinning the transfer are not yet fully explained. Magnetosome membrane proteins are responsible for the direct interaction with the mineral substrate; therefore the interaction between amino acids and magnetite has attracted much attention (Buerger et al., 2013). Iron oxyhydroxides are found in soils and are major sorption substrates for toxins, contaminants and heavy metals (Bargar et al., 1997; Waychunas et al., 2005). Perhaps goethite is the most abundant Fe oxyhydroxide, and its solubility depends on the pH and the oxidation state of Fe (Schwertmann, 1991). Explicit treatment of the solvent molecules in capturing the effect of the surface on the liquid phase and the sorption and dynamics of monovalent ions in modelling goethite– water interactions was highlighted by using a combination of static potential models (Kerisit et al., 2005b, 2006). Reactive FFs have been also heavily applied to goethite mineral surfaces to understand the proton equilibria at the surfaces (Aryanpour et al., 2010; Rustad et al., 1996). The work by Rustad and co-workers on the goethite–water interface used a water potential of Halley et al. (1993). This water potential is a polarisable and dissociating water model based on the work of Stillinger and David (1980). The advantage of using a water model that has the ability to dissociate gives the possibility of describing the acid–base chemistry at the goethite–water interface. Recently, Aryanpour and co-workers (2010) have developed a reactive FF, within the ReaxFF methodology, to describe the goethite–water interface. The structure of the solution on the goethite surface shows very good agreement with experimental data and provides the framework for future applications in other iron oxide systems. The examples given earlier highlight the fact that FFs can reproduce very complex systems with high degrees of accuracy. As the level of understanding from experimental evidence increases on processes such as crystal growth, adsorption and dissolution, the advancement of complexity in the FFs developed has had to occur to mirror that understanding. These examples also show that FFs can be used as a complement to experiment to aid in the understanding very complex geochemical systems.

2.6

Concluding Remarks

We have shown how the development of FFs has been successful for important minerals and their properties. With the availability of high-quality electronic structure simulations, there is much more data available to derive more sophisticated potentials and hence increased interest in deriving manybody potentials, such as the multipoles and EAM-based potentials (Cooper et al., 2014). Furthermore, it is worth mentioning the ‘on the fly’ potentials, where the potential parameters are evaluated and updated between quantum-derived structures during the course of the simulation. However, challenges remain and can be divided in terms of models and methodologies. The widespread usage of FF-based techniques is intrinsically limited by the lack of a UFF that can be used to simulate the entire periodic table with high accuracy. The FF should be able to simulate all phases, materials, minerals, organic and biological molecules without any problem related to transferability. It should capture the differences in the interactions between species with different hybridisation, oxidation state and environment. It should deal with polarisability with a smart inclusion of all multipole interactions. It should be intrinsically reactive to deal with the change of the pH and ionic strength. Finally, each level of complexity should switchable; the FF should include parameters to turn on and off the different level of complexity. The development of the models has to be followed by the development of methodologies associated to each level of complexity in the models,

64

Molecular Modeling of Geochemical Reactions

such as more efficient ways to include the multipoles and their interactions and reliable ways of accounting for reactivity, without the drawback of over-parameterised FFs. Furthermore, novel methods of deriving and testing FFs need to be developed into more automatic procedures including using machine learning approaches such as artificial neural networks and multilayer perceptrons that could help determining and representing the potential energy surface (Handley and Popelier, 2010). Another important area to exploit is the development of tools that link the different levels of theory to be able to span between different length and time scales, such as the hybrid methodologies, described earlier, but as discussed, it is still in its infancy. The challenge of modelling still larger systems for longer times will ensure that the testing and development of reliable FFs will continue to be required for the foreseeable future.

References Abascal, J.L.F., et al., A potential model for the study of ices and amorphous water: TIP4P/ice. Journal of Chemical Physics, 2005. 122(23): p. 9. Abrams, C.F. and K. Kremer, Combined coarse-grained and atomistic simulation of liquid bisphenol A-polycarbonate: liquid packing and intramolecular structure. Macromolecules, 2003. 36(1): p. 260–267. Aguado, A., et al., Multipoles and interaction potentials in ionic materials from planewave-DFT calculations. Faraday Discussions, 2003. 124: p. 171–184. Aguzzi, C., et al., Use of clays as drug delivery systems: possibilities and limitations. Applied Clay Science, 2007. 36(1–3): p. 22–36. Aicken, A.M., et al., Simulation of layered double hydroxide intercalates. Advanced Materials, 1997. 9(6): p. 496–500. Alexandrov, V. and K.M. Rosso, Insights into the mechanism of Fe(II) adsorption and oxidation at Fe–clay mineral surfaces from first-principles calculations. The Journal of Physical Chemistry C, 2013. 117(44): p. 22880–22886. Allen, M.P. and D.J. Tildesley, Computer Simulation of Liquids. 1989, Oxford: Clarendon Press. p. 385. Allinger, N.L., Conformational-analysis. 130. MM2: hydrocarbon force-field utilizing V1 and V2 torsional terms. Journal of the American Chemical Society, 1977. 99(25): p. 8127–8134. Allinger, N.L., Y.H. Yuh and J.H. Lii, Molecular mechanics: the MM3 force-field for hydrocarbons. 1. Journal of the American Chemical Society, 1989. 111(23): p. 8551–8566. Allinger, N.L., K. Chen and J.-H. Lii, An improved force field (MM4) for saturated hydrocarbons. Journal of Computational Chemistry, 1996. 17(5–6): p. 642–668. Anderson, R.L., et al., Clay swelling: a challenge in the oilfield. Earth-Science Reviews, 2010. 98(3–4): p. 201–216. Applequist, J., A multipole interaction theory of electric polarization of atomic and molecular assemblies. Journal of Chemical Physics, 1985. 83(2): p. 809–826. Arcos, D., et al., Long-term geochemical evolution of the near field repository: Insights from reactive transport modelling and experimental evidences. Journal of Contaminant Hydrology, 2008. 102(3–4): p. 196–209. Aryanpour, M., A.C.T. van Duin and J.D. Kubicki, Development of a reactive force field for ironoxyhydroxide systems. The Journal of Physical Chemistry A, 2010. 114(21): p. 6298–6307. Attard, P., Electrolytes and the Electric Double Layer, in Advances in Chemical Physics, ed. I. Prigogine and S.A. Rice. 2007, Chichester: John Wiley & Sons, Inc. p. 1–159. Austen, K.F., et al., Electrostatic versus polarization effects in the adsorption of aromatic molecules of varied polarity on an insulating hydrophobic surface. Journal of Physics. Condensed Matter, 2008. 20(3): p. 035215. Bader, R.F.W., Atoms in molecules. Accounts of Chemical Research, 1985. 18(1): p. 9–15. Bains, A.S., et al., Molecular modelling of the mechanism of action of organic clay-swelling inhibitors. Molecular Simulation, 2001. 26(2): p. 101−145. Bargar, J.R., G.E. Brown and G.A. Parks, Surface complexation of Pb(II) at oxide-water interfaces. 2. XAFS and bond-valence determination of mononuclear Pb(II) sorption products and surface functional groups on iron oxides. Geochimica et Cosmochimica Acta, 1997. 61(13): p. 2639–2652. Barlow, S., et al., Molecular mechanics study of oligomeric models for poly(ferrocenylsilanes) using the extensible systematic forcefield (ESFF). Journal of the American Chemical Society, 1996. 118(32): p. 7578–7592.

Force Field Application and Development

65

Bayly, C.I., et al., A well-behaved electrostatic potential based method using charge restraints for deriving atomic charges: the RESP model. Journal of Physical Chemistry, 1993. 97(40): p. 10269–10280. Berendsen, H.J.C., J.R. Grigera and T.P. Straatsma, The missing term in effective pair potentials. Journal of Physical Chemistry, 1987. 91(24): p. 6269–6271. Berthelot, D., Sur le Mélange des Gaz. Comptes rendus de l’Academie des Sciences, 1889. 126: p. 1703–1706. Bhowmik, R., K.S. Katti and D. Katti, Molecular dynamics simulation of hydroxyapatite-polyacrylic acid interfaces. Polymer, 2007. 48(2): p. 664–674. Bickmore, B.R., et al., Bond-valence constraints on liquid water structure. The Journal of Physical Chemistry A, 2009. 113(9): p. 1847–1857. Boek, E.S., P.V. Coveney and N.T. Skipper, Monte Carlo molecular modeling studies of hydrated Li-, Na-, and K-smectites: understanding the role of potassium as a clay swelling inhibitor. Journal of the American Chemical Society, 1995. 117(50): p. 12608–12617. Bourova, E., S.C. Parker and P. Richet, Atomistic simulation of cristobalite at high temperature. Physical Review B, 2000. 62(18): p. 12052–12061. Brancato, G. and M.E. Tuckerman, A polarizable multistate empirical valence bond model for proton transport in aqueous solution. Journal of Chemical Physics, 2005. 122(22): p. 224507. Brenner, D.W., Empirical potential for hydrocarbons for use in simulating the chemical vapor-deposition of diamond films. Physical Review B, 1990. 42(15): p. 9458–9471. Brenner, D.W., et al., A second-generation reactive empirical bond order (REBO) potential energy expression for hydrocarbons. Journal of Physics. Condensed Matter, 2002. 14(4): p. 783–802. Bresme, F., Equilibrium and nonequilibrium molecular-dynamics simulations of the central force model of water. The Journal of Chemical Physics, 2001. 115(16): p. 7564–7574. Brini, E., V. Marcon and N.F.A. van der Vegt, Conditional reversible work method for molecular coarse graining applications. Physical Chemistry Chemical Physics, 2011. 13(22): p. 10468–10474. Brini, E., et al., Thermodynamic transferability of coarse-grained potentials for polymer-additive systems. Physical Chemistry Chemical Physics, 2012. 14(34): p. 11896–11903. Brini, E., et al., Systematic coarse-graining methods for soft matter simulations: a review. Soft Matter, 2013. 9(7): p. 2108–2119. Brooks, B.R., et al., CHARMM: a program for macromolecular energy, minimization, and dynamics calculations. Journal of Computational Chemistry, 1983. 4(2): p. 187–217. Broukhno, A., et al., Depletion and bridging forces in polymer systems: Monte Carlo simulations at constant chemical potential. Journal of Chemical Physics, 2000. 113(13): p. 5493–5501. Broukhno, A., et al., Polyampholyte-induced repulsion between charged surfaces: Monte Carlo simulation studies. Langmuir, 2002. 18(16): p. 6429–6436. Brukhno, A.V., T. Akesson and B. Jonsson, Phase behavior in suspensions of highly charged colloids. The Journal of Physical Chemistry B, 2009. 113(19): p. 6766–6774. Buerger, A., U. Magdans and H. Gies, Adsorption of amino acids on the magnetite-(111)-surface: a force field study. Journal of Molecular Modeling, 2013. 19(2): p. 851–857. Bulo, R.E., et al., Toward a practical method for adaptive QM/MM simulations. Journal of Chemical Theory and Computation, 2009. 5(9): p. 2212–2221. Bush, T.S., et al., Self-consistent interatomic potentials for the simulation of binary and ternary oxides. Journal of Materials Chemistry, 1994. 4(6): p. 831–837. Carlson, W.D., The polymorphs of CaCO3 and the aragonite-calcite transformation Reviews in Mineralogy and Geochemistry, 1983. 11: p. 191–225. Carof, A., et al., Coarse graining the dynamics of nano-confined solutes: the case of ions in clays. Molecular Simulation, 2014. 40(1–3): p. 237–244. Carrier, B., et al., Elastic properties of swelling clay particles at finite temperature upon hydration. The Journal of Physical Chemistry C, 2014. 118(17): p. 8933–8943. Catlow, C.R.A., et al., Simulating silicate structures and the structural chemistry of pyroxenoids. Nature, 1982. 295(5851): p. 658–662. Celis, R., M.C. Hermosin and J. Cornejo, Heavy metal adsorption by functionalized clays. Environmental Science & Technology, 2000. 34(21): p. 4593–4599. Chamritski, I. and G. Burns, Infrared- and Raman-active phonons of magnetite, maghemite, and hematite: a computer simulation and spectroscopic study. The Journal of Physical Chemistry B, 2005. 109(11): p. 4965–4968. Chenoweth, K., A.C.T. van Duin and W.A. Goddard, ReaxFF reactive force field for molecular dynamics simulations of hydrocarbon oxidation. Journal of Physical Chemistry A, 2008. 112(5): p. 1040–1053.

66

Molecular Modeling of Geochemical Reactions

Chicot, D., et al., Mechanical properties of magnetite (Fe3O4), hematite (α-Fe2O3) and goethite (α-FeO OH) by instrumented indentation and molecular dynamics analysis. Materials Chemistry and Physics, 2011. 129(3): p. 862–870. Churakov, S.V., et al., Resolving diffusion in clay minerals at different time scales: combination of experimental and modeling approaches. Applied Clay Science, 2014. 96(0): p. 36–44. Collins, D.R. and C.R.A. Catlow, Computer-simulation of structures and cohesive properties of MICAS. American Mineralogist, 1992. 77(11–12): p. 1172–1181. Cooke, D.J. and J.A. Elliott, Atomistic simulations of calcite nanoparticles and their interaction with water. Journal of Chemical Physics, 2007. 127(10): p. 104706. Cooper, M.W.D., M.J.D. Rushton and R.W. Grimes, A many-body potential approach to modelling the thermomechanical properties of actinide oxides. Journal of Physics. Condensed Matter, 2014. 26(10): 105401. Couves, J.W., et al., Experimental-verification of a predicted negative thermal expansivity of crystalline zeolites. Journal of Physics. Condensed Matter, 1993. 5(27): p. L329–L332. Coveney, P.V. and W. Humphries, Molecular modelling of the mechanism of action of phosphonate retarders on hydrating cements. Journal of the Chemical Society, Faraday Transactions, 1996. 92(5): p. 831–841. Cox, S.R. and D.E. Williams, Representation of the molecular electrostatic potential by a net atomic charge model. Journal of Computational Chemistry, 1981. 2(3): p. 304–323. Cruz-Chu, E.R., A. Aksimentiev and K. Schulten, Water−silica force field for simulating nanodevices. The Journal of Physical Chemistry B, 2006. 110(43): p. 21497–21508. Cygan, R.T., et al., Atomistic models of carbonate minerals: bulk and surface structures, defects, and diffusion. Molecular Simulation, 2002. 28(6–7): p. 475–495. Cygan, R.T., J.J. Liang and A.G. Kalinichev, Molecular models of hydroxide, oxyhydroxide, and clay phases and the development of a general force field. Journal of Physical Chemistry B, 2004. 108(4): p. 1255–1266. Cygan, R.T., V.N. Romanov and E.M. Myshakin, Molecular simulation of carbon dioxide capture by montmorillonite using an accurate and flexible force field. The Journal of Physical Chemistry C, 2012. 116(24): p. 13079–13091. Dauberosguthorpe, P., et al., Structure and energetics of ligand-binding to proteins: Escherichia coli dihydrofolate reductase trimethoprim, a drug-receptor system. Proteins: Structure, Function, and Genetics, 1988. 4(1): p. 31–47. de Leeuw, N.H., F.M. Higgins and S.C. Parker, Modeling the surface structure and stability of α-quartz. The Journal of Physical Chemistry B, 1999. 103(8): p. 1270–1277. Deer, W.A., An Introduction to the Rock-Forming Minerals/W.A. Deer, R.A. Howie, J. Zussman, ed. J. Zussman and R.A. Howie. 1992, Harlow, England/New York, NY: Longman Scientific & Technical/Wiley. Delgado-Buscalioni, R., K. Kremer and M. Praprotnik, Concurrent triple-scale simulation of molecular liquids. Journal of Chemical Physics, 2008. 128(11): p. 114110. Delhommelle, J. and P. Millie, Inadequacy of the Lorentz-Berthelot combining rules for accurate predictions of equilibrium properties by molecular simulation. Molecular Physics, 2001. 99(8): p. 619–625. Demichelis, R., et al., Stable prenucleation mineral clusters are liquid-like ionic polymers. Nature Communications, 2011. 2: p. 1–8. Demiralp, E., T. Cagin and W.A. Goddard, Morse stretch potential charge equilibrium force field for ceramics: application to the quartz-stishovite phase transition and to silica glass. Physical Review Letters, 1999. 82(8): p. 1708–1711. Derjaguin, B. and L. Landau, Theory of the stability of strongly charged lyophobic sols and of the adhesion of strongly charged-particles in solutions of electrolytes. Progress in Surface Science, 1993. 43(1–4): p. 30–59. di Pasquale, N., D. Marchisio and P. Carbone, Mixing atoms and coarse-grained beads in modelling polymer melts. Journal of Chemical Physics, 2012. 137(16): p. 164111–164119. Dick, B.G. and A.W. Overhauser, Theory of the dielectric constants of alkali halide crystals. Physical Review, 1958. 112(1): p. 90. Dixon, D.A., M.N. Gray and A.W. Thomas, A study of the compaction properties of potential clay-sand buffer mixtures for use in nuclear-fuel waste-disposal. Engineering Geology, 1985. 21(3–4): p. 247–255. Downing, C.A., A.A. Sokol and C.R.A. Catlow, The reactivity of CO2 on the MgO(100) surface. Physical Chemistry Chemical Physics, 2014. 16(1): p. 184–195. Duque-Redondo, E., et al., Molecular forces governing shear and tensile failure in clay-dye hybrid materials. Chemistry of Materials, 2014. 26(15): 4338–4345. Dykstra, C.E., Electrostatic interaction potentials in molecular-force fields. Chemical Reviews, 1993. 93(7): p. 2339–2353.

Force Field Application and Development

67

Ercolessi, F. and J.B. Adams, Interatomic potentials from 1st-principles calculations: the force-matching method. Europhysics Letters, 1994. 26(8): p. 583–588. Ewald, P.P., Die Berechnung optischer und elektrostatischer Gitterpotentiale. Annalen der Physik, 1921. 369(3): p. 253–287. Faux, I.D., Polarization catastrophe in defect calculations in ionic crystals. Journal of Physics Part C: Solid State Physics, 1971. 4(10): p. L211–L216. Fenter, P., et al., Is the calcite–water interface understood? Direct comparisons of molecular dynamics simulations with specular X-ray reflectivity data. The Journal of Physical Chemistry C, 2013. 117(10): p. 5028–5042. Ferguson, D.M., Parameterization and evaluation of a flexible water model. Journal of Computational Chemistry, 1995. 16(4): p. 501–511. Finnis, M., Interatomic Forces in Condensed Matter. 2003, Oxford: Oxford University Press. Fischer, J., et al., Modeling of aqueous poly(oxyethylene) solutions. 2. Mesoscale simulations. Journal of Physical Chemistry B, 2008. 112(43): p. 13561–13571. Flekkoy, E.G., P.V. Coveney and G. De Fabritiis, Foundations of dissipative particle dynamics. Physical Review E, 2000. 62(2): p. 2140–2157. Fogarty, J.C., et al., A reactive molecular dynamics simulation of the silica-water interface. The Journal of Chemical Physics, 2010. 132(17): p. 174704. Freeman, C.L., et al., A new potential model for barium titanate and its implications for rare-earth doping. Journal of Materials Chemistry, 2011. 21(13): p. 4861–4868. Frenkel, D. and B. Smith, Understanding Molecular Simulation: From Algorithms to Applications. 1996, San Diego: Academic Press, Inc. p. 443. Fu, Y.-T. and H. Heinz, Structure and cleavage energy of surfactant-modified clay minerals: influence of CEC, head group and chain length. Philosophical Magazine, 2010. 90(17–18): p. 2415–2424. Fu, C.-C., et al., A test of systematic coarse-graining of molecular dynamics simulations: thermodynamic properties. Journal of Chemical Physics, 2012. 137(16): p. 164106. Gale, J.D., Empirical potential derivation for ionic materials. Philosophical Magazine B: Physics of Condensed Matter: Statistical Mechanics, Electronic, Optical and Magnetic Properties, 1996. 73(1): p. 3–19. Gale, J.D. and A.L. Rohl, The general utility lattice program (GULP). Molecular Simulation, 2003. 29(5): p. 291–341. Gale, J.D., P. Raiteri and A.C.T. van Duin, A reactive force field for aqueous-calcium carbonate systems. Physical Chemistry Chemical Physics, 2011. 13(37): p. 16666–16679. Gebauer, D., A. Voelkel and H. Coelfen, Stable prenucleation calcium carbonate clusters. Science, 2008. 322(5909): p. 1819–1822. Gilbert, T.L., Soft-sphere model for closed-shell atoms and ions. Journal of Chemical Physics, 1968. 49(6): p. 2640–2642. Greathouse, J.A. and R.T. Cygan, Molecular dynamics simulation of uranyl(VI) adsorption equilibria onto an external montmorillonite surface. Physical Chemistry Chemical Physics, 2005. 7(20): p. 3580–3586. Greathouse, J.A. and R.T. Cygan, Water structure and aqueous uranyl(VI) adsorption equilibria onto external surfaces of beidellite, montmorillonite, and pyrophyllite: results from molecular simulations. Environmental Science & Technology, 2006. 40(12): p. 3865–3871. Greathouse, J.A., et al., Molecular dynamics study of aqueous uranyl interactions with quartz (010). Journal of Physical Chemistry B, 2002. 106(7): p. 1646–1655. Greathouse, J.A., et al., Uranyl surface complexes in a mixed-charge montmorillonite: Monte Carlo computer simulation and polarized XAFS results. Clays and Clay Minerals, 2005. 53(3): p. 278–286. Guilbaud, P. and G. Wipff, Force field representation of the UO22+ cation from free energy MD simulations in water. Tests on its 18-crown-6 and NO3- adducts, and on its calix 6 arene(6-) and CMPO complexes. Journal of Molecular Structure. (THEOCHEM), 1996. 366(1–2): p. 55–63. Guillot, B., A reappraisal of what we have learnt during three decades of computer simulations on water. Journal of Molecular Liquids, 2002. 101(1–3): p. 219–260. Guldbrand, L., et al., Electrical double-layer forces: a Monte-Carlo study. Journal of Chemical Physics, 1984. 80(5): p. 2221–2228. Halgren, T.A., Representation of van der Waals (vdW) interactions in molecular mechanics force-fields: potential form, combination rules, and VDW parameters. Journal of the American Chemical Society, 1992. 114(20): p. 7827–7843. Halgren, T.A., Merck molecular force field .1. Basis, form, scope, parameterization, and performance of MMFF94. Journal of Computational Chemistry, 1996. 17(5–6): p. 490–519.

68

Molecular Modeling of Geochemical Reactions

Halley, J.W., J.R. Rustad and A. Rahman, A polarizable, dissociating molecular-dynamics model for liquid water. Journal of Chemical Physics, 1993. 98(5): p. 4110–4119. Handley, C.M. and P.L.A. Popelier, Potential energy surfaces fitted by artificial neural networks. The Journal of Physical Chemistry. A, 2010. 114(10): p. 3371–3383. Harmandaris, V.A., et al., Hierarchical modeling of polystyrene: from atomistic to coarse-grained simulations. Macromolecules, 2006. 39(19): p. 6708–6719. Hauptmann, S., et al., Potential energy function for apatites. Physical Chemistry Chemical Physics, 2003. 5(3): p. 635–639. Heinz, H., et al., Force field for mica-type silicates and dynamics of octadecylammonium chains grafted to montmorillonite. Chemistry of Materials, 2005. 17(23): p. 5658–5669. Heinz, H., et al., Self-assembly of alkylammonium chains on montmorillonite: effect of chain length, head group structure, and cation exchange capacity. Chemistry of Materials, 2006. 19(1): p. 59–68. Heinz, H., et al., Thermodynamically consistent force fields for the assembly of inorganic, organic, and biological nanostructures: the INTERFACE force field. Langmuir, 2013. 29(6): p. 1754–1765. Henderson, R.L., Uniqueness theorem for fluid pair correlation-functions. Physics Letters A, 1974. 49(3): p. 197–198. Hermans, J., et al., A consistent empirical potential for water-protein interactions. Biopolymers, 1984. 23(8): p. 1513–1518. Hess, B., C. Holm and N. van der Vegt, Modeling multibody effects in ionic solutions with a concentration dependent dielectric permittivity. Physical Review Letters, 2006. 96(14): p. 147801. Hofmann, D.W.M., L. Kuleshova and B. D’Aguanno, A new reactive potential for the molecular dynamics simulation of liquid water. Chemical Physics Letters, 2007. 448(1–3): p. 138–143. Holesova, S., et al., Antibacterial kaolinite/urea/chlorhexidine nanocomposites: experiment and molecular modelling. Applied Surface Science, 2014. 305: p. 783–791. Hou, D. and Z. Li, Molecular dynamics study of water and ions transported during the nanopore calcium silicate phase: case study of jennite. Journal of Materials in Civil Engineering, 2014. 26(5): p. 930–940. Huang, D.M., et al., Coarse-grained computer simulations of polymer/fullerene bulk heterojunctions for organic photovoltaic applications. Journal of Chemical Theory and Computation, 2010. 6(2): p. 526–537. Hwang, S., et al., The MS-Q force field for clay minerals: application to oil production. The Journal of Physical Chemistry B, 2001. 105(19): p. 4122–4127. Hynninen, A.P. and A.Z. Panagiotopoulos, Phase diagrams of charged colloids from thermodynamic integration. Journal of Physics. Condensed Matter, 2009. 21(46): p. 465104. Israelachvili, J.N., Intermolecular and Surface Forces (3rd Edition). 2011, Burlington: Academic Press. p. 1–674. Izvekov, S. and G.A. Voth, Multiscale coarse graining of liquid-state systems. Journal of Chemical Physics, 2005. 123(13): p. 134105. Jones, J.E., On the determination of molecular fields. I. From the variation of the viscosity of a gas with temperature. Proceedings of the Royal Society of London, Series A, 1924a. 106(738): p. 441–462. Jones, J.E., On the determination of molecular fields. II. From the equation of state of a gas. Proceedings of the Royal Society of London, Series A, 1924b. 106(738): p. 463–477. Jonsson, B., et al., Depletion and structural forces in confined polyelectrolyte solutions. Langmuir, 2003. 19(23): p. 9914–9922. Jorgensen, W.L. and C.J. Swenson, Optimized intermolecular potential functions for amides and peptides: hydration of amides. Journal of the American Chemical Society, 1985. 107(6): p. 1489–1496. Jorgensen, W.L., et al., Comparison of simple potential functions for simulating liquid water. Journal of Chemical Physics, 1983. 79(2): p. 926–935. Jorgensen, W.L., D.S. Maxwell and J. TiradoRives, Development and testing of the OPLS all-atom force field on conformational energetics and properties of organic liquids. Journal of the American Chemical Society, 1996. 118(45): p. 11225–11236. Kalinichev, A.G., J. Wang and R.J. Kirkpatrick, Molecular dynamics modeling of the structure, dynamics and energetics of mineral–water interfaces: Application to cement materials. Cement and Concrete Research, 2007. 37(3): p. 337–347. Kamerlin, S.C.L., et al., On unjustifiably misrepresenting the EVB approach while simultaneously adopting it. Journal of Physical Chemistry B, 2009. 113(31): p. 10905–10915. Kerisit, S. and S.C. Parker, Free energy of adsorption of water and metal ions on the {1014} calcite surface. Journal of the American Chemical Society, 2004. 126(32): p. 10152–10161.

Force Field Application and Development

69

Kerisit, S. and K.M. Rosso, Charge transfer in FeO: a combined molecular-dynamics and ab initio study. The Journal of Chemical Physics, 2005. 123(22): p. 224712. Kerisit, S., et al., Molecular dynamics simulations of the interactions between water and inorganic solids. Journal of Materials Chemistry, 2005a. 15(14): p. 1454–1462. Kerisit, S., et al., Atomistic simulation of charged iron oxyhydroxide surfaces in contact with aqueous solution. Chemical Communications, 2005b. (24): p. 3027–3029. Kerisit, S., E.S. Ilton and S.C. Parker, Molecular dynamics simulations of electrolyte solutions at the (100) goethite surface. The Journal of Physical Chemistry B, 2006. 110(41): p. 20491–20501. Kerisit, S., J.H. Weare and A.R. Felmy, Structure and dynamics of forsterite–scCO2/H2O interfaces as a function of water content. Geochimica et Cosmochimica Acta, 2012. 84(0): p. 137–151. Khan, A., K. Fontell and B. Lindman, Phase-equilibria of some ionic surfactant systems with divalent counterions. Colloids and Surfaces, 1984a. 11(3–4): p. 401–408. Khan, A., K. Fontell and B. Lindman, Liquid crystallinity in systems of magnesium and calcium surfactants: phase-diagrams and phase structures in binary aqueous systems of magnesium and calcium di-2-ethylhexylsulfosuccinate. Journal of Colloid and Interface Science, 1984b. 101(1): p. 193–200. Kong, C.L., Combining rules for intermolecular potential parameters. 2. Rules for Lennard-Jones (12–6) potential and Morse potential. Journal of Chemical Physics, 1973. 59(5): p. 2464–2467. Konowalow, D.D., Interatomic potentials for HeNe, HeAr, and NeAr. Journal of Chemical Physics, 1969. 50(1): p. 12–16. Koretsky, C.M., et al., Detection of surface hydroxyl species on quartz, γ-alumina, and feldspars using diffuse reflectance infrared spectroscopy. Geochimica et Cosmochimica Acta, 1997. 61(11): p. 2193–2210. Korkut, A. and W.A. Hendrickson, A force field for virtual atom molecular mechanics of proteins. Proceedings of the National Academy of Sciences of the United States of America, 2009. 106(37): p. 15667–15672. Kosmulski, M., The pH-dependent surface charging and the points of zero charge. Journal of Colloid and Interface Science, 2002. 253(1): p. 77–87. Kramer, G.J., et al., Interatomic force fields for silicas, aluminophosphates, and zeolites: derivation based on ab initio calculations. Physical Review B, 1991. 43(6): p. 5068–5080. Kundu, T.K., K.H. Rao and S.C. Parker, Atomistic simulation studies of magnetite surface structures and adsorption behavior in the presence of molecular and dissociated water and formic acid. Journal of Colloid and Interface Science, 2006. 295(2): p. 364–373. Labbez, C., et al., Surface charge density and electrokinetic potential of highly charged minerals: experiments and Monte Carlo simulations on calcium silicate hydrate. Journal of Physical Chemistry B, 2006. 110(18): p. 9219–9230. Leach, A., Molecular Modelling: Principles and Applications (2nd Edition). 2001, Harlow, England/New York: Prentice Hall. Lee, H., et al., A coarse-grained model for polyethylene oxide and polyethylene glycol: conformation and hydrodynamics. Journal of Physical Chemistry B, 2009. 1613(40): p. 13186–13194. Lenart, P.J., A. Jusufi and A.Z. Panagiotopoulos, Effective potentials for 1: 1 electrolyte solutions incorporating dielectric saturation and repulsive hydration. Journal of Chemical Physics, 2007. 126(4): p. 044509. Leung, K., S.B. Rempe and C.D. Lorenz, Salt permeation and exclusion in hydroxylated and functionalized silica pores. Physical Review Letters, 2006. 96(9): p. 095504. Lewis, G.V. and C.R.A. Catlow, Potential models for ionic oxides. Journal of Physics C: Solid State Physics, 1985. 18(6): p. 1149–1161. Liang, T., et al., Reactive potentials for advanced atomistic simulations. Annual Review of Materials Research, 2013. 43: p. 109–129. Lifson, S., A.T. Hagler and P. Dauber, Consistent force-field studies of inter-molecular forces in hydrogenbonded crystals. 1. Carboxylic-acids, amides, and the C=O…H-hydrogen-bonds. Journal of the American Chemical Society, 1979. 101(18): p. 5111–5121. Lin, F.H., et al., A study of purified montmorillonite intercalated with 5-fluorouracil as drug carrier. Biomaterials, 2002. 23(9): p. 1981–1987. Liu, P. and G.A. Voth, Smart resolution replica exchange: an efficient algorithm for exploring complex energy landscapes. Journal of Chemical Physics, 2007. 126(4): p. 045106. Lobaugh, J. and G.A. Voth, The quantum dynamics of an excess proton in water. Journal of Chemical Physics, 1996. 104(5): p. 2056–2069. Lockwood, G.K. and S.H. Garofalini, Bridging oxygen as a site for proton adsorption on the vitreous silica surface. Journal of Chemical Physics, 2009. 131(7): p. 074703.

70

Molecular Modeling of Geochemical Reactions

Lopes, P.E.M., et al., Development of an empirical force field for silica. Application to the quartz−water interface. The Journal of Physical Chemistry B, 2006. 110(6): p. 2782–2792. Lopes, P.E.M., B. Roux and A.D. MacKerell, Jr., Molecular modeling and dynamics studies with explicit inclusion of electronic polarizability: theory and applications. Theoretical Chemistry Accounts, 2009. 124 (1–2): p. 11–28. Lorentz, H.A., Ueber die Anwendung des Satzes vom Virial in der kinetischen Theorie der Gase. Annalen der Physik, 1881. 248(1): p. 127–136. Lorenz, C.D., et al., Molecular dynamics of ionic transport and electrokinetic effects in realistic silica channels. The Journal of Physical Chemistry C, 2008. 112(27): p. 10222–10232. Lyman, E., F.M. Ytreberg and D.M. Zuckerman, Resolution exchange simulation. Physical Review Letters, 2006. 96(2): p. 028105. Lyubartsev, A.P., Multiscale modeling of lipids and lipid bilayers. European Biophysics Journal with Biophysics Letters, 2005. 35(1): p. 53–61. Lyubartsev, A.P. and A. Laaksonen, Calculation of effective interaction potentials from radial-distribution functions: a reverse Monte-Carlo approach. Physical Review E, 1995. 52(4): p. 3730–3737. Lyubartsev, A.P., et al., On coarse-graining by the inverse Monte Carlo method: dissipative particle dynamics simulations made to a precise tool in soft matter modeling. Soft Materials, 2003. 1(1): p. 121–137. MacKerell, A.D., et al., All-atom empirical potential for molecular modeling and dynamics studies of proteins. Journal of Physical Chemistry B, 1998. 102(18): p. 3586–3616. Madden, P.A., et al., From first-principles to material properties. Journal of Molecular Structure (THEOCHEM), 2006. 771(1–3): p. 9–18. Mahadevan, T.S. and S.H. Garofalini, Dissociative water potential for molecular dynamics simulations. Journal of Physical Chemistry B, 2007. 111(30): p. 8919–8927. Mahoney, M. and W. Jorgensen, A five-site model for liquid water and the reproduction of the density anomaly by rigid, nonpolarizable potential functions. The Journal of Chemical Physics, 2000. 112(20): p. 8910–8922. Marcus, R.A. and N. Sutin, Electron transfers in chemistry and biology. Biochimica et Biophysica Acta (BBA) – Reviews on Bioenergetics, 1985. 811(3): p. 265–322. Markutsya, S., et al., Evaluation of coarse-grained mapping schemes for polysaccharide chains in cellulose. Journal of Chemical Physics, 2013. 138(21): p. 214108. Marra, J., Direct measurement of the interaction between phosphatidylglycerol bilayers in aqueous-electrolyte solutions. Biophysical Journal, 1986. 50(5): p. 815–825. Marrink, S.J., A.H. de Vries and A.E. Mark, Coarse grained model for semiquantitative lipid simulations. Journal of Physical Chemistry B, 2004. 108(2): p. 750–760. Marrink, S.J., et al., The MARTINI force field: coarse grained model for biomolecular simulations. Journal of Physical Chemistry B, 2007. 111(27): p. 7812–7824. Martin, P., et al., Application of molecular dynamics DL_POLY codes to interfaces of inorganic materials. Molecular Simulation, 2006. 32(12–13): p. 1079–1093. Mason, E.A., Forces between unlike molecules and the properties of gaseous mixtures. Journal of Chemical Physics, 1955. 23(1): p. 49–56. Matter, J.M. and P.B. Kelemen, Permanent storage of carbon dioxide in geological reservoirs by mineral carbonation. Nature Geoscience, 2009. 2(12): p. 837–841. Matysiak, S., et al., Modeling diffusive dynamics in adaptive resolution simulation of liquid water. Journal of Chemical Physics, 2008. 128(2): p. 024503. Mayo, S.L., B.D. Olafson and W.A. Goddard, Dreiding: a generic force-field for molecular simulations. Journal of Physical Chemistry, 1990. 94(26): p. 8897–8909. Meakin, P. and Z. Xu, Dissipative Particle Dynamics and other particle methods for multiphase fluid flow in fractured and porous media. Progress in Computational Fluid Dynamics, 2009. 9(6–7): p. 399–408. Metropolis, N., et al., Equation of state calculations by fast computing machines. Journal of Chemical Physics, 1953. 21(6): p. 1087–1092. Milani, A., et al., Coarse-grained simulations of model polymer nanofibres. Macromolecular Theory and Simulations, 2011. 20(5): p. 305–319. Mirzoev, A. and A.P. Lyubartsev, Effective solvent mediated potentials of Na+ and Cl- ions in aqueous solution: temperature dependence. Physical Chemistry Chemical Physics, 2011. 13(13): p. 5722–5727. Mirzoev, A. and A.P. Lyubartsev, Systematic implicit solvent coarse graining of dimyristoylphosphatidylcholine lipids. Journal of Computational Chemistry, 2014. 35(16): p. 1208–1218.

Force Field Application and Development

71

Mishra, R.K., R.J. Flatt and H. Heinz, Force field for tricalcium silicate and insight into nanoscale properties: cleavage, initial hydration, and adsorption of organic molecules. The Journal of Physical Chemistry C, 2013. 117(20): p. 10417–10432. Moreno, N., P. Vignal, J. Li and V. M. Calo, Multiscale modeling of blood flow: coupling finite elements with smoothed dissipative particle dynamics. Procedia Computer Science, 2013. 18: p. 2565–2574. Mueller, J.E., A.C.T. van Duin and W.A. Goddard, Development and validation of ReaxFF reactive force field for hydrocarbon chemistry catalyzed by nickel. Journal of Physical Chemistry C, 2010. 114(11): p. 4939–4949. Mulliken, R.S., Electronic population analysis on LCAO-MO molecular wave functions .1. Journal of Chemical Physics, 1955. 23(10): p. 1833–1840. Myshakin, E.M., et al., Molecular dynamics simulations of turbostratic dry and hydrated montmorillonite with intercalated carbon dioxide. The Journal of Physical Chemistry A, 2014. 118(35): 7454–7468. Neri, M., et al., Coarse-grained model of proteins incorporating atomistic detail of the active site. Physical Review Letters, 2005. 95(21): p. 218102. Newman, S.P., et al., Interlayer arrangement of hydrated MgAl layered double hydroxides containing guest terephthalate anions: comparison of simulation and measurement. The Journal of Physical Chemistry B, 1998. 102(35): p. 6710–6719. Newton, R.C. and C.E. Manning, Thermodynamics of SiO2–H2O fluid near the upper critical end point from quartz solubility measurements at 10 kbar. Earth and Planetary Science Letters, 2008. 274(1–2): p. 241–249. Nielsen, S.O., et al., Recent progress in adaptive multiscale molecular dynamics simulations of soft matter. Physical Chemistry Chemical Physics, 2010. 12(39): p. 12401–12414. Nielsen, M.H., S. Aloni and J.J. De Yoreo, In situ TEM imaging of CaCO3 nucleation reveals coexistence of direct and indirect pathways. Science, 2014. 345(6201): p. 1158–1162. Noid, W.G., Perspective: coarse-grained models for biomolecular systems. Journal of Chemical Physics, 2013. 139(9): p. 090901. Noid, W.G., et al., The multiscale coarse-graining method. I. A rigorous bridge between atomistic and coarsegrained models. Journal of Chemical Physics, 2008. 128(24): p. 244114. Parry, D.E., Electrostatic potential in surface region of an ionic-crystal. Surface Science, 1975. 49(2): p. 433–440. Parry, D.E., Correction. Surface Science, 1976. 54(1): p. 195–195. Paul, D.R. and L.M. Robeson, Polymer nanotechnology: nanocomposites. Polymer, 2008. 49(15): p. 3187–3204. Pavese, A., et al., Modelling of the thermal dependence of structural and elastic properties of calcite, CaCO3. Physics and Chemistry of Minerals, 1996. 23(2): p. 89–93. Pearlman, D.A., et al., AMBER, a package of computer-programs for applying molecular mechanics, normalmode analysis, molecular-dynamics and free-energy calculations to simulate the structural and energetic properties of molecules. Computer Physics Communications, 1995. 91(1–3): p. 1–41. Pedone, A., et al., A new self-consistent empirical interatomic potential model for oxides, silicates, and silicabased glasses. Journal of Physical Chemistry B, 2006. 110(24): p. 11780–11795. Pellenq, R.J.-M., et al., A realistic molecular model of cement hydrates. Proceedings of the National Academy of Sciences, 2009. 106(38): p. 16102–16107. Pinto, E.M., et al., Electrochemical and surface characterisation of carbon-film-coated piezoelectric quartz crystals. Applied Surface Science, 2009. 255(18): p. 8084–8090. Plassard, C., et al., Investigation of the surface structure and elastic properties of calcium silicate hydrates at the nanoscale. Ultramicroscopy, 2004. 100(3–4): p. 331–338. Plassard, C., et al., Nanoscale experimental investigation of particle interactions at the origin of the cohesion of cement. Langmuir, 2005. 21(16): p. 7263–7270. Ponder, J.W., et al., Current status of the AMOEBA polarizable force field. The Journal of Physical Chemistry B, 2010. 114(8): p. 2549–2564. Pouget, E.M., et al., The initial stages of template-controlled CaCO3 formation revealed by cryo-TEM. Science, 2009. 323(5920): p. 1455–1458. Praprotnik, M., L. Delle Site and K. Kremer, A macromolecule in a solvent: adaptive resolution molecular dynamics simulation. Journal of Chemical Physics, 2007. 126(13): p. 134902. Praprotnik, M., L. Delle Site and K. Kremer, Multiscale simulation of soft matter: from scale bridging to adaptive resolution, Annual Review of Physical Chemistry, 2008. 59: p. 545–571. Prasitnok, K. and M.R. Wilson, A coarse-grained model for polyethylene glycol in bulk water and at a water/air interface. Physical Chemistry Chemical Physics, 2013. 15(40): p. 17093–17104.

72

Molecular Modeling of Geochemical Reactions

Quigley, D., et al., Sampling the structure of calcium carbonate nanoparticles with metadynamics. Journal of Chemical Physics, 2011. 134(4): p. 044703. Rai, B., et al., A molecular dynamics study of the interaction of oleate and dodecylammonium chloride surfactants with complex aluminosilicate minerals. Journal of Colloid and Interface Science, 2011. 362(2): p. 510–516. Raiteri, P. and J.D. Gale, Water is the key to nonclassical nucleation of amorphous calcium carbonate. Journal of the American Chemical Society, 2010. 132(49): p. 17623–17634. Raiteri, P., et al., Derivation of an accurate force-field for simulating the growth of calcium carbonate from aqueous solution: a new model for the calcite-water interface. Journal of Physical Chemistry C, 2010. 114(13): p. 5997–6010. Ramirez-Hernandez, A., et al., Dynamical simulations of coarse grain polymeric systems: rouse and entangled dynamics. Macromolecules, 2013. 46(15): p. 6287–6299. Rappe, A.K. and W.A. Goddard, Charge equilibration for molecular-dynamics simulations. Journal of Physical Chemistry, 1991. 95(8): p. 3358–3363. Rappe, A.K., et al., UFF, a full periodic-table force-field for molecular mechanics and molecular-dynamics simulations. Journal of the American Chemical Society, 1992. 114(25): p. 10024–10035. Ray, S.S. and M. Bousmina, Biodegradable polymers and their layered silicate nano composites: in greening the 21st century materials world. Progress in Materials Science, 2005. 50(8): p. 962–1079. Raymand, D., et al., A reactive force field (ReaxFF) for zinc oxide. Surface Science, 2008. 602(5): p. 1020–1031. Reith, D., M. Putz and F. Muller-Plathe, Deriving effective mesoscale potentials from atomistic simulations. Journal of Computational Chemistry, 2003. 24(13): p. 1624–1636. Richardson, D.J., J.N. Butt and T.A. Clarke, Controlling electron transfer at the microbe–mineral interface. Proceedings of the National Academy of Sciences, 2013. 110(19): p. 7537–7538. Rosenbluth, M.N. and A.W. Rosenbluth, Further results on Monte-Carlo equations of state. Journal of Chemical Physics, 1954. 22(5): p. 881–884. Roterman, I.K., et al., A comparison of the CHARMM, AMBER and ECEPP potentials for peptides. II. φ-ψ Maps for N-acetyl alanine N -methyl amide: comparisons, contrasts and simple experimental tests. Journal of Biomolecular Structure and Dynamics, 1989. 7(3): p. 421–453. Ruehle, V., et al., Versatile object-oriented toolkit for coarse-graining applications. Journal of Chemical Theory and Computation, 2009. 5(12): p. 3211–3223. Rustad, J.R., A.R. Felmy and B.P. Hay, Molecular statics calculations of proton binding to goethite surfaces: a new approach to estimation of stability constants for multisite surface complexation models. Geochimica et Cosmochimica Acta, 1996. 60(9): p. 1563–1576. Rustad, J.R., A.R. Felmy and E.J. Bylaska, Molecular simulation of the magnetite-water interface. Geochimica et Cosmochimica Acta, 2003. 67(5): p. 1001–1016. Rzepiela, A.J., et al., Hybrid simulations: combining atomistic and coarse-grained force fields using virtual sites. Physical Chemistry Chemical Physics, 2011. 13(22): p. 10437–10448. Sainz-Diaz, C.I., A. Hernandez-Laguna and M.T. Dove, Modeling of dioctahedral 2: 1 phyllosilicates by means of transferable empirical potentials. Physics and Chemistry of Minerals, 2001. 28(2): p. 130–141. Sanders, M.J., M. Leslie and C.R.A. Catlow, Interatomic potentials for SiO2. Journal of the Chemical Society, Chemical Communications, 1984. 19: p. 1271–1273. Santiso, E.E. and K.E. Gubbins, Multi-scale molecular modeling of chemical reactivity. Molecular Simulation, 2004. 30(11–12): p. 699–748. Scanlon, D.O., et al., Band alignment of rutile and anatase TiO2. Nature Materials, 2013. 12(9): p. 798–801. Schmitt, U.W. and G.A. Voth, The computer simulation of proton transport in water. Journal of Chemical Physics, 1999. 111(20): p. 9361–9381. Schwertmann, U., Solubility and dissolution of iron-oxides. Plant and Soil, 1991. 130(1–2): p. 1–25. Shahsavari, R., R.J.M. Pellenq and F.-J. Ulm, Empirical force fields for complex hydrated calcio-silicate layered materials. Physical Chemistry Chemical Physics, 2011. 13(3): p. 1002–1011. Shan, T.-R., et al., Second-generation charge-optimized many-body potential for Si/SiO2 and amorphous silica. Physical Review B, 2010a. 82(23): p. 235302. Shan, T.-R., et al., Charge-optimized many-body potential for the hafnium/hafnium oxide system. Physical Review B, 2010b. 81(12): p. 125328. Shapley, T.V., et al., Atomistic modeling of the sorption free energy of dioxins at clay-water interfaces. Journal of Physical Chemistry C, 2013. 117(47): p. 24975–24984. Shell, M.S., Systematic coarse-graining of potential energy landscapes and dynamics in liquids. Journal of Chemical Physics, 2012. 137(8): p. 084503.

Force Field Application and Development

73

Shi, S.H., et al., An extensible and systematic force field, ESFF, for molecular modeling of organic, inorganic, and organometallic systems. Journal of Computational Chemistry, 2003. 24(9): p. 1059–1076. Singh, U.C. and P.A. Kollman, An approach to computing electrostatic charges for molecules. Journal of Computational Chemistry, 1984. 5(2): p. 129–145. Sirk, T.W., S. Moore and E.F. Brown, Characteristics of thermal conductivity in classical water models. The Journal of Chemical Physics, 2013. 138(6): p. 064505. Skelton, A.A., et al., Simulations of the quartz(1011)/water interface: a comparison of classical force fields, ab initio molecular dynamics, and X-ray reflectivity experiments. The Journal of Physical Chemistry C, 2011. 115(5): p. 2076–2088. Skipper, N.T., G. Sposito and F.R.C. Chang, Monte-Carlo simulation of interlayer molecular-structure in swelling clay-minerals .2. Monolayer hydrates. Clays and Clay Minerals, 1995. 43(3): p. 294–303. Spagnoli, D. and J.D. Gale, Atomistic theory and simulation of the morphology and structure of ionic nanoparticles. Nanoscale, 2012. 4(4): p. 1051–1067. Spagnoli, D., et al., Molecular dynamics simulations of the interaction between the surfaces of polar solids and aqueous solutions. Journal of Materials Chemistry, 2006. 16(20): p. 1997–2006. Spagnoli, D., et al., Prediction of the effects of size and morphology on the structure of water around hematite nanoparticles. Geochimica et Cosmochimica Acta, 2009. 73(14): p. 4023–4033. Srivastava, K.P., Unlike molecular interactions and properties of gas mixtures. Journal of Chemical Physics, 1958. 28(4): p. 543–549. Srivastava, A. and G.A. Voth, Hybrid approach for highly coarse-grained lipid bilayer models. Journal of Chemical Theory and Computation, 2013. 9(1): p. 750–765. Stillinger, F.H. and C.W. David, Study of the water octamer using the polarization model of molecularinteractions. Journal of Chemical Physics, 1980. 73(7): p. 3384–3389. Stone, A.J., Distributed multipole analysis, or how to describe a molecular charge-distribution. Chemical Physics Letters, 1981. 83(2): p. 233–239. Stone, A.J. and M. Alderton, Distributed multipole analysis: methods and applications. Molecular Physics, 1985. 56(5): p. 1047–1064. Sun, H., COMPASS: an ab initio force-field optimized for condensed-phase applications: overview with details on alkane and benzene compounds. Journal of Physical Chemistry B, 1998. 102(38): p. 7338–7364. Sun, H., et al., An ab-initio CFF93 all-atom force-field for polycarbonates. Journal of the American Chemical Society, 1994. 116(7): p. 2978–2987. Suter, J.L., et al., Large-scale molecular dynamics study of montmorillonite clay: emergence of undulatory fluctuations and determination of material properties. The Journal of Physical Chemistry C, 2007. 111 (23): p. 8248–8259. Suter, J.L., et al., Recent advances in large-scale atomistic and coarse-grained molecular dynamics simulation of clay minerals. Journal of Materials Chemistry, 2009. 19(17): p. 2482–2493. Swadling, J.B., P.V. Coveney and H.C. Greenwell, Clay minerals mediate folding and regioselective interactions of RNA: a large-scale atomistic simulation study. Journal of the American Chemical Society, 2010. 132 (39): p. 13750–13764. Tamerler, C., et al., Genetically engineered polypeptides for inorganics: a utility in biological materials science and engineering. Materials Science and Engineering: C, 2007. 27(3): p. 558–564. Taylor, C.D. and M. Neurock, Theoretical insights into the structure and reactivity of the aqueous/metal interface. Current Opinion in Solid State & Materials Science, 2005. 9(1–2): p. 49–65. Tenney, C.M. and R.T. Cygan, Molecular simulation of carbon dioxide, brine, and clay mineral interactions and determination of contact angles. Environmental Science & Technology, 2014. 48(3): p. 2035–2042. Teppen, B.J., et al., Molecular dynamics modeling of clay minerals .1. Gibbsite, kaolinite, pyrophyllite, and beidellite. Journal of Physical Chemistry B, 1997. 101(9): p. 1579–1587. Thuresson, A., et al., Monte Carlo simulations of parallel charged platelets as an approach to tactoid formation in clay. Langmuir, 2013. 29(29): p. 9216–9223. Tokarsky, J., P. Capkova and J.V. Burda, Structure and stability of kaolinite/TiO2 nanocomposite: DFT and MM computations. Journal of Molecular Modeling, 2012. 18(6): p. 2689–2698. Tombácz, E., et al., Water in contact with magnetite nanoparticles, as seen from experiments and computer simulations. Langmuir, 2009. 25(22): p. 13007–13014. Tosenberger, A., et al., Particle dynamics methods of blood flow simulations. Mathematical Modelling of Natural Phenomena, 2011. 6(5): p. 320–332. Tröster, P., et al., Polarizable water models from mixed computational and empirical optimization. The Journal of Physical Chemistry B, 2013. 117(32): p. 9486–9500.

74

Molecular Modeling of Geochemical Reactions

Tse, J.S. and D.D. Klug, Mechanical instability of α-quartz: a molecular dynamics study. Physical Review Letters, 1991. 67(25): p. 3559–3562. Ufimtsev, I.S., et al., A multistate empirical valence bond model for solvation and transport simulations of OH- in aqueous solutions. Physical Chemistry Chemical Physics, 2009. 11(41): p. 9420–9430. van Beest, B.W.H., G.J. Kramer and R.A. van Santen, Force fields for silicas and aluminophosphates based on ab initio calculations. Physical Review Letters, 1990. 64(16): p. 1955–1958. van Duin, A.C.T., et al., ReaxFF: a reactive force field for hydrocarbons. Journal of Physical Chemistry A, 2001. 105(41): p. 9396–9409. Valero, R., et al., Perspective on diabatic models of chemical reactivity as illustrated by the gas-phase S(N)2 reaction of acetate ion with 1,2-dichloroethane. Journal of Chemical Theory and Computation, 2009. 5(1): p. 1–22. Vega, C. and J.L.F. Abascal, Simulating water with rigid non-polarizable models: a general perspective. Physical Chemistry Chemical Physics, 2011. 13(44): p. 19663–19688. Vega, C., et al., What ice can teach us about water interactions: a critical comparison of the performance of different water models. Faraday Discussions, 2009. 141: p. 251–276. Verlet, L., Computer experiments on classical fluids. I. Thermodynamical properties of Lennard-Jones molecules. Physical Review, 1967. 159(1): p. 98–103. Verlet, L., Computer experiments on classical fluids. 2. Equilibrium correlation functions. Physical Review, 1968. 165(1): p. 201. Verwey, E.J.W., J.Th.G. Overbeek and K. van Nes, Theory of the Stability of Lyophobic Colloids. 1948, Amsterdam: Elsevier Publishing Company Inc. Villà, J. and A. Warshel, Energetics and dynamics of enzymatic reactions. The Journal of Physical Chemistry B, 2001. 105(33): p. 7887–7907. Vuilleumier, R. and D. Borgis, Transport and spectroscopy of the hydrated proton: a molecular dynamics study. Journal of Chemical Physics, 1999. 111(9): p. 4251–4266. Waldman, M. and A.T. Hagler, New combining rules for rare-gas van-der-Waals parameters. Journal of Computational Chemistry, 1993. 14(9): p. 1077–1084. Wallace, A.F., et al., Microscopic evidence for liquid-liquid separation in supersaturated CaCO3 solutions. Science, 2013. 341(6148): p. 885–889. Wang, Z.-J. and M. Deserno, A systematically coarse-grained solvent-free model for quantitative phospholipid bilayer simulations. Journal of Physical Chemistry B, 2010. 114(34): p. 11207–11220. Wang, J., A.G. Kalinichev and R.J. Kirkpatrick, Molecular modeling of water structure in nano-pores between brucite (001) surfaces. Geochimica et Cosmochimica Acta, 2004. 68(16): p. 3351–3365. Wang, Q., D.J. Keffer and D.M. Nicholson, A coarse-grained model for polyethylene glycol polymer. Journal of Chemical Physics, 2011. 135(21): p. 214903. Wang, Q., et al., Multi-scale models for cross-linked sulfonated poly (1, 3-cyclohexadiene) polymer. Polymer, 2012a. 53(7): p. 1517–1528. Wang, Q., et al., Atomistic and coarse-grained molecular dynamics simulation of a cross-linked sulfonated poly (1,3-cyclohexadiene)-based proton exchange membrane. Macromolecules, 2012b. 45(16): p. 6669–6685. Wang, Y., et al., Molecular dynamics simulation of strong interaction mechanisms at wet interfaces in claypolysaccharide nanocomposites. Journal of Materials Chemistry A, 2014. 2(25): p. 9541–9547. Warshel, A. and R.M. Weiss, An empirical valence bond approach for comparing reactions in solutions and in enzymes. Journal of the American Chemical Society, 1980. 102(20): p. 6218–6226. Watson, G.W. and S.C. Parker, Quartz amorphization: a dynamical instability. Philosophical Magazine Letters, 1995a. 71(1): p. 59–64. Watson, G.W. and S.C. Parker, Dynamical instabilities in alpha-quartz and alpha-berlinite: a mechanism for amorphization. Physical Review B, 1995b. 52(18): p. 13306–13309. Waychunas, G.A., C.S. Kim and J.F. Banfield, Nanoparticulate iron oxide minerals in soils and sediments: unique properties and contaminant scavenging mechanisms. Journal of Nanoparticle Research, 2005. 7(4–5): p. 409–433. Weiner, S.J., et al., An all atom force-field for simulations of proteins and nucleic-acids. Journal of Computational Chemistry, 1986. 7(2): p. 230–252. Wilson, M., et al., Molecular dynamics simulations of compressible ions. Journal of Chemical Physics, 1996a. 104(20): p. 8068–8081. Wilson, M., U. Schonberger and M.W. Finnis, Transferable atomistic model to describe the energetics of zirconia. Physical Review B, 1996b. 54(13): p. 9147–9161.

Force Field Application and Development

75

Wilson, M., S. Jahn and P.A. Madden, The construction and application of a fully flexible computer simulation model for lithium oxide. Journal of Physics. Condensed Matter, 2004. 16(27): p. S2795–S2810. Wood, W.W. and F.R. Parker, Monte Carlo equation of state of molecules interacting with the Lennard-Jones potential. 1. Supercritical isotherm at about twice the critical temperature. Journal of Chemical Physics, 1957. 27(3): p. 720–733. Woodley, S.M., et al., The prediction of inorganic crystal structures using a genetic algorithm and energy minimisation. Physical Chemistry Chemical Physics, 1999. 1(10): p. 2535–2542. Woods, A.D.B., W. Cochran and B.N. Brockhouse, Lattice dynamics of alkali halide crystals. Physical Review, 1960. 119(3): p. 980–999. Wu, Y.J., H.L. Tepper and G.A. Voth, Flexible simple point-charge water model with improved liquid-state properties. Journal of Chemical Physics, 2006. 124(2): p. 024503. Wu, Y., et al., An improved multistate empirical valence bond model for aqueous proton solvation and transport. Journal of Physical Chemistry B, 2008. 112(2): p. 467–482. Wu, Z., Q. Cui and A. Yethiraj, A new coarse-grained force field for membrane-peptide simulations. Journal of Chemical Theory and Computation, 2011. 7(11): p. 3793–3802. Xie, W., et al., Development of a polarizable intermolecular potential function (PIPF) for liquid amides and alkanes. Journal of Chemical Theory and Computation, 2007. 3(6): p. 1878–1889. Zaidan, O.F., J.A. Greathouse and R.T. Pabalan, Monte Carlo and molecular dynamics simulation of uranyl adsorption on montmorillonite clay. Clays and Clay Minerals, 2003. 51(4): p. 372–381. Zeitler, T.R., et al., Vibrational analysis of brucite surfaces and the development of an improved force field for molecular simulation of interfaces. Journal of Physical Chemistry C, 2014. 118(15): p. 7946–7953. Zelko, J., et al., Effects of counterion size on the attraction between similarly charged surfaces. Journal of Chemical Physics, 2010. 133(20): p. 204901. Zeng, Q.H., et al., Molecular dynamics simulation of organic–inorganic nanocomposites: layering behavior and interlayer structure of organoclays. Chemistry of Materials, 2003. 15(25): p. 4732–4738. Zhang, G., et al., Dispersion-corrected density functional theory and classical force field calculations of water loading on a pyrophyllite(001) surface. Journal of Physical Chemistry C, 2012. 116(32): p. 17134–17141. Zhao, T. and X. Wang, Distortion and flow of nematics simulated by dissipative particle dynamics. Journal of Chemical Physics, 2014. 140(18): p. 184902. Zhu, R., et al., Sorptive characteristics of organomontmorillonite toward organic compounds: a combined LFERs and molecular dynamics simulation study. Environmental Science & Technology, 2011. 45(15): p. 6504–6510. Zhu, R., et al., Modeling the interaction of nanoparticles with mineral surfaces: adsorbed C60 on pyrophyllite. The Journal of Physical Chemistry A, 2013. 117(30): p. 6602–6611.

3 Quantum-Mechanical Modeling of Minerals Alessandro Erba1 and Roberto Dovesi1,2 1

Dipartimento di Chimica, Università degli Studi di Torino, Torino, Italy NIS Centre of Excellence “Nanostructured Interfaces and Surfaces”, Torino, Italy

2

3.1

Introduction

The combination of growing computing resources and success of quantum-chemical methods based on the density functional theory (DFT) is rapidly widening the applicability range of first-principles techniques in the solid state to the study of minerals of geochemical interest, materials for the electronics and hydrogen storage, biomaterials, etc. (Albanese et al. 2012; Delle Piane et al. 2013; Erba et al. 2014b; Ugliengo et al. 2008). A variety of properties of materials, such as structural, electronic, vibrational, optical, elastic, magnetic, etc., can now be routinely computed with high accuracy with several periodic quantum-chemical programs. Such methods, however, in their standard formulation, describe the ground state of the system (a time-dependent formulation of the DFT would be required for describing excited states (Marques and Gross 2004)) at zero temperature and pressure, which is of course a severe limitation to their applicability in geochemical studies. The exact physicochemical composition of the Earth’s deep interior, for instance, is still debated. Different compositional models have been proposed (Anderson and Bass 1986; Bass and Anderson 1984; Ringwood 1975) which can be mainly validated on the grounds of seismological data collected during earthquakes, provided that the individual elastic response of all possible constituents of the Earth’s upper mantle, transition zone, and lower mantle is known. This characterization is an extremely difficult task from an experimental point of view (Anderson and Isaak 1995; Mainprice et al. 2013) in that high temperatures and pressures have to be simultaneously considered (typically, pressures up to 140 GPa and temperatures in the range 800–1200 K). In this respect, the predictive power of first-principles techniques could prove decisive. Nevertheless, the theoretical description of structural, thermodynamic, and elastic properties of minerals at high temperature and pressure also represents a challenge for state-of-the-art quantum-mechanical techniques for the following reasons: (i) complex algorithms have to be developed for including the effects of pressure and temperature on Molecular Modeling of Geochemical Reactions: An Introduction, First Edition. Edited by James D. Kubicki. © 2016 John Wiley & Sons, Ltd. Published 2016 by John Wiley & Sons, Ltd.

78

Molecular Modeling of Geochemical Reactions

structural and elastic properties; if not fully automated, they would require too many separate program modules, interfaces, external units for data storage, postprocessing scripts, etc., for being routinely used; (ii) all the required algorithms have to be implemented in the same program, within the same formal framework and numerical conditions; (iii) all the algorithms have to be efficiently implemented as regards massive-parallel scalability and memory use in that the resulting calculations can become computationally rather demanding; and (iv) all the algorithms should be as general as possible in terms of dimensionality of the system under investigation (from 1-D to 2-D and 3-D), chemical composition, symmetry, electronic configuration, etc. From a more fundamental point of view, all kinds of chemical interactions must be reliably described. The DFT, in its standard formulation, is known to neglect or describe in a spurious way dispersive, London-type, interactions. In this respect, in recent years, simple a posteriori semi-empirical energy corrections have become extremely popular in a molecular context (Grimme 2006, 2011; Grimme et al. 2010). Even if a reparametrization of such schemes has been attempted for molecular crystals (Civalleri et al. 2008), further effort has to be put in this direction before these schemes could be fruitfully used in the simulation of several properties of the solid state. In this chapter, we will review some of the most recent developments in this respect, as implemented in the CRYSTAL program for quantum-chemical simulations of the solid state, as we are among its developers (Dovesi et al. 2014a,b). In the last years, many efforts have been devoted to the optimization of the core algorithms of the program (calculation of the self-consistent-field (SCF) procedure and computation of atomic and cell gradients) in terms of reduced use of memory and increased efficiency in the parallel scalability (Bush et al. 2011; Dovesi et al. 2014a; Orlando et al. 2012). Several tensorial properties can now be computed automatically, such as the fourth-rank elastic tensor (Erba et al. 2014b; Perger et al. 2009), the third-rank direct and converse piezoelectric tensors (Erba et al. 2013a; Mahmoud et al. 2014b), the fourth-rank photoelastic Pockels’ tensor (Erba and Dovesi 2013), the second-rank dielectric (or polarizability) tensor, and third- and fourth-rank hyperpolarizabilities (Ferrero et al. 2008a,b,c,d; Orlando et al. 2009, 2010). Generalization of the elastic and piezoelectric tensor calculation to 1-D and 2-D systems has also been performed (Baima et al. 2013; Erba et al. 2013b; Lacivita et al. 2013b). At variance with most DFT-based programs where numerical approaches are implemented, analytical infrared (IR) and Raman intensities are now available as well as the automated computation of IR and Raman spectra (Carteret et al. 2013; De La Pierre et al. 2011). More details about these spectroscopic features are given in Chapter 10. New algorithms have also been developed for the study of solid solutions and, more generally, disordered systems (D’Arco et al. 2013; Mustapha et al. 2013). Some of these recent developments will be illustrated in this chapter, along with some application to the study of the family of silicate garnets. Garnets constitute a large class of materials of great technological and geochemical interest; they can be used as components of lasers, computer memories, and microwave optical devices and, due to high hardness and recyclability, as abrasives and filtration media (Novak and Gibbs 1971). Silicate garnets are among the most important rockforming minerals and represent the main constituents of the Earth’s lower crust, upper mantle, and transition zone. They are characterized by a cubic structure with space group Ia3d and formula X3Y2(SiO4)3, where the X site hosts divalent cations such as Ca2 + , Mg2 +, Fe2 +, and Mn2 + and the Y site is occupied by trivalent cations such as Al3 +, Fe3 +, and Cr3 +. At least 12 end-members of this family of minerals have been identified (Rickwood et al. 1968). The primitive cell contains four formula units (80 atoms), and the structure consists in alternating SiO4 tetrahedra and YO6 octahedra sharing corners to form a three-dimensional network. The most common end-members of the family are pyrope Mg3Al2(SiO4)3; almandine Fe3Al2(SiO4)3; spessartine Mn3Al2(SiO4)3; grossular Ca3Al2(SiO4)3; uvarovite Ca3Cr2(SiO4)3; and andradite Ca3Fe2(SiO4)3. Natural silicate garnets can be found in a wide range of chemical compositions since they form solid solutions.

Quantum-Mechanical Modeling of Minerals

79

The structure of this chapter is as follows: in Section 3.2 the formal framework is briefly defined within which the CRYSTAL program is developed, in terms of periodic boundary conditions and localized Gaussian-type orbital (GTO) basis sets (BSs); Section 3.3 is devoted to the discussion of a couple of techniques for the inclusion of the effect of pressure on structural properties of minerals; Section 3.4 deals with the evaluation of the elastic tensor (both at zero and nonzero pressure) and related elastic properties of minerals; a brief review on the first-principles studies that have been performed on spectroscopic properties of silicate garnets is given in Section 3.5, along with the description of the approach implemented in the CRYSTAL program for computing thermodynamic properties of solids; a brief description, along with some examples of application, of some of the techniques implemented for the study of solid solutions or disordered systems is given in Section 3.6; finally, some remarks on current developments and future challenges are given in Section 3.7, in particular as regards the inclusion of temperature on computed structural and elastic properties of minerals.

3.2

Theoretical Framework

This section is devoted to a very brief illustration of the main formal aspects of standard quantumchemical techniques, as applied to solid-state systems. For a more detailed presentation, the reader may refer to Chapter 1. The aim is here to find the electronic ground-state solution of the static Schrodinger equation at fixed nuclei and in the absence of external fields: H el Ψ0 = E0 Ψ0

31

Despite the simplification introduced by translational invariance, the intrinsic periodic nature of crystals makes the solution of Equation 3.1 practically impossible unless “average-field” approximations to the electrostatic Hamiltonian are introduced, such as those represented by Hartree–Fock (HF) or Kohn–Sham (KS) approaches (Hohenberg and Kohn 1964; Kohn and Sham 1965). The nonrelativistic electrostatic Hamiltonian, Ĥel, in Equation 3.1, for the N-electron system in the field of M nuclei of charge ZA, fixed at position RA, is given by the following expression: N

H el =

N M − ∇2n − ZA 1 N 1 1 M ZA ZB + + + 2 r 2 n, m = 1 rnm 2 A, B = 1 rAB n=1 n = 1 A = 1 nA

32

Atomic units (au) are here used; the primed double sums exclude “diagonal” terms n = m, A = B . Ĥel depends parametrically on the sets {R}, {Z} of positions and charges of the M nuclei; the same holds true for ground-state energy E0 and wave function Ψ0 which is an antisymmetric function of the space–spin coordinates xn ≡ rn ,σ n of the N electrons. 3.2.1

Translation Invariance and Periodic Boundary Conditions

Let us consider a regular lattice of vectors Tm generated from D primitive linearly independent basis D

m a , where mi are integers and D is the number of periodic vectors ai of ordinary space, Tm = i=1 i i directions (three for bulk crystals, two for slabs, one for polymeric structures). The ai vectors define (though not univocally) the unit cell of the crystal. The coordinates and charges of all nuclei in the infinite crystal can be generated from those of a translationally irreducible finite set {RA,0; ZA,0} as follows: RA, m = RA, 0 + Tm and ZA, m = ZA, 0 . For solving Equation 3.1, Born–von Kàrmàn (BvK)

80

Molecular Modeling of Geochemical Reactions

periodic boundary conditions are adopted. They impose that Ψ0 is cyclically periodic with respect to a superlattice of vectors W m defined as Tm, but starting from D superbasis vectors A i = wi ai : xn = rn ,σ n , xn = rn + W m ,σ n ,

if

33

Ψ0 …,xn ,… = Ψ0 …, xn ,…

then

The integers wi define the effective number of electrons in the system: N eff = W N0 , with W = i wi , and N0 = ΣA ZA, 0 the number of electrons per cell (only neutral systems are here considered). The setting of the wi’s (i.e., the so-called shrinking factor) is one of the main computational parameters of a periodic calculation. 3.2.2

HF and KS Methods

In their spin-unrestricted formulation, both HF and KS methods are intended to obtain a set of N oneelectron functions, the molecular spin orbitals (MSO) (or crystalline spin orbitals (CSO) in the periodic F, σ case), ψ jF x = ϕj r ω σ (here F stands for HF or KS and σ for α or β), which satisfy the equation: h

F, σ

F, σ

ϕj

r = −

∇2 + 2

A

F, σ −ZA ρF r F, σ + dr + V ϕj r = RA − r r −r

F, σ j

F, σ

ϕj

r

34

The effective Hamiltonian ĥF,σ which acts on the individual MSO contains, apart from the kinetic, nuclear attraction and Hartree operators (the last one expressing the Coulomb repulsion with all the electrons in the system), a corrective potential V F,σ , which differs in the two schemes. A singledeterminant N-electron function, Ψ0F , can be defined, after assigning the N electrons to the N MSOs F, σ corresponding to the lowest eigenvalues j of Equation 3.4 and antisymmetrizing their product. In the rest of this section we shall assume, for simplicity, that we are describing a closed-shell system so that the spin index can be dropped from the effective Hamiltonian and from the corrective potential. The position density matrix (DM) and the electron density (ED) associated with Ψ0F are then simply N 2

ϕnF r ϕnF r

P F r;r = 2

∗

N 2

; ρ F r = P F r; r = 2

n=1

ϕnF r

2

35

n=1

In the HF scheme, the corrective potential V HF is defined by imposing that the HF energy E0HF is a minimum with respect to any other single-determinant N-electron wave function. To achieve this goal, V HF must take the form of the exact-exchange operator. We have, correspondingly, HF =− E0HF ≡ ΨHF 0 H el Ψ0

+

∇2 HF P r;r 2

1 ρ HF r ρ HF r 1 drdr − 2 r −r 4

dr− r =r

P HF r; r r −r

ZA A

2

dr dr +

ρ HF r dr RA − r

1 M ZA ZB ≥ E0 2 A, B = 1 rAB

The KS scheme, formulated in the frame of DFT (Hohenberg and Kohn 1964; Kohn and Sham 1965), introduces, for any given N-electron ED, ρ(r), two universal functionals: εxc(r; [ρ]) and its functional derivative Vxc(r; [ρ]). When the exchange-correlation potential Vxc(r; [ρ]KS) is used for V KS as a

Quantum-Mechanical Modeling of Minerals

81

multiplicative operator in Equation 3.4, the density from Equation 3.5 coincides with the exact ground-state ED: ρ KS r = ρ r . The functional εxc(r; [ρ]) allows the exact ground-state energy to be calculated, again with reference to the occupied KS manifold: E0KS = −

+

∇2 KS P r; r 2

dr − r =r

ZA A

ρr dr RA −r

1 ρrρr 1 M ZA ZB = E0 drdr + ρ r εxc r; ρ dr + 2 r− r 2 A, B = 1 rAB

Equation 3.4 must be solved self-consistently in both cases, because the Hartree and the corrective potential are defined in terms of the occupied manifold. Let us note that in the HF case, the nonlocal corrective potential is perfectly defined. On the contrary, no exact formula exists for the local exchange-correlation potential Vxc(r; [ρ]); an enormous amount of proposals has been formulated so far. The two schemes, HF and KS, can be combined together in so-called hybrid functionals which introduce a fraction of exact HF exchange into the exchange-correlation DFT functional. Hybrid functionals correct for some of the main deficiencies of pure LDA and GGA functionals, that is, they reduce the self-interaction error, and they generally widen the electronic band gap of solids (Corá et al. 2004). Among others, popular hybrid functionals are B3LYP (Becke 1993), PBE0 (Adamo and Barone 1999), and HSE06 (Heyd and Scuseria 2004).

3.2.3

Bloch Functions and Local BS

Given that the one-electron effective Hamiltonian ĥF commutes with all operations of the space group , in particular of the subgroup of pure translations, its eigenfunctions, the crystalline orbitals (CO), can be classified according to the irreducible representations (irrep) of that group. As is shown in standard textbooks (Tinkham 1964), they are then characterized by an index κ, a vector of reciprocal space, such that the corresponding COs are Bloch functions (BF), ϕnF r; κ , which satisfy the property ϕnF r + Tm ; κ = ϕnF r;κ exp ικ Tm

36

Clearly, κ’s differing by a reciprocal lattice vector G define the same irreducible representation. Among all equivalent κ’s one can choose the one closest to the origin of the reciprocal space; this “minimal-length” set fills the so-called (first) Brillouin zone (BZ). The COs must also satisfy BvK conditions ϕnF r + W m ; κ = ϕnF r;κ , which means that exp ικ W m = 1. In practical calculations, the orbitals must be expanded into some suitably chosen BS. In periodic systems, it is convenient to use a BS of BFs fμ(r; κ) so that determining the COs ϕnF r; κ reduces to a secular problem that involves only basis functions of that given κ. The choice of the BS is crucial and determines the algorithms and numerical methods used in the actual solution. Most calculations use either plane waves (PWs) or atom-centered local functions, AOs. PWs are the traditional choice in solid-state physics, reflecting the delocalized nature of valence and conduction electron states in crystals. Localized BSs formed by AOs, χ μ(r), are the traditional choice in quantum chemistry, reflecting the atomic composition of matter. In periodic systems, one uses Bloch sums of AOs: fμ r; κ =

1 W

exp ικ T χ μ r −T T

37

82

Molecular Modeling of Geochemical Reactions

The CRYSTAL program, on which we focus our attention, shares with standard molecular codes the use of GTOs as local basis functions. Each atom A carries pA GTOs, each resulting from a “contraction” of MiA Gaussian “primitives” of angular momentum components ℓ, m centered in RA: MiA

χ iA rA =

ciA, j N ℓ, m αiA, j X ℓ, m rA exp − αiA, j r2A

j=1

Here rA = r −RA , Xℓ,m are real solid harmonics and Nℓ,m normalization coefficients; ciA,j are known as “coefficients” αiA,j as “exponents” of the GTO. See Appendix A of Erba and Pisani (2012) for more details. The evaluation of GTO integrals in CRYSTAL entails problems related to the periodically infinite character of the system. Sophisticated techniques have been implemented which permit the truncation or the accurate approximation of lattice sums: Ewald techniques, multipolar treatment of nonoverlapping distributions, bipolar expansion, etc. (see Pisani et al. (1988) for a detailed description of such techniques). An attractive feature related to the local character of the basis functions is that not only 3-dimensional crystals but also structures periodic in 2 (slabs), 1 (polymers), and 0 (molecules) dimensions are treated by CRYSTAL with the same basic technology without any need of artificial replication of the subunits. Another major advantage of AOs is that point symmetry of the crystal can be fully exploited at any stage of the calculation, which, given the usual high symmetry of crystals, results in large savings of computational resources (Orlando et al. 2014; Zicovich-Wilson and Dovesi 1998).

3.3 Structural Properties A good description of the structure of minerals generally relies on the effectiveness of minimization algorithms in exploring the potential energy surface (PES) of the system. Depending on the specific structural feature to be explored, one might be interested in characterizing global or relative minima and saddle points (for details on the implementation of transition-state search techniques in the CRYSTAL program, see Rimola et al. (2010)) of the PES. The energy of a crystal can be minimized both in terms of atomic coordinates within the cell and in terms of the lattice parameters of the cell; the corresponding energy gradients are implemented analytically in the CRYSTAL program (Doll 2001; Doll et al. 2001). A quasi-Newton technique combined with the Broyden–Fletcher–Goldfarb–Shanno algorithm for Hessian updating is used (Broyden 1970; Fletcher 1970; Goldfarb 1970; Shanno 1970) in the automated implementation of the geometry optimizer and convergence checked on both gradient components and nuclear displacements (Civalleri et al. 2001). As an example, in Table 3.1, we report selected structural parameters of three X3Y2(SiO4)3 silicate garnets, as obtained by fully optimizing their structure at the B3LYP hybrid level of theory and as experimentally determined with accurate low-temperature X-ray diffraction experiments (ZicovichWilson et al. 2008). The lattice parameter, a; the fractional coordinates of the oxygen atom, Oi; and selected bond lengths are reported for pyrope, Mg3Al2(SiO4)3; grossular, Ca3Al2(SiO4)3; and andradite, Ca3Fe2(SiO4)3. The overall agreement between computed and measured structural features is quite remarkable, the lattice parameter being systematically overestimated by about 1% and, correspondingly, the YO and SiO distances being overestimated by about 0.01 – 0.02 Å. When structural properties of minerals at geophysical conditions have to be determined, constant-volume and constant-pressure optimizations prove extremely useful. In the next subsections,

Quantum-Mechanical Modeling of Minerals

83

Table 3.1 Lattice parameter, a (in Å), fractional coordinates of the oxygen atom, Oi, and selected distances (in Å) of three X3Y2(SiO4)3 silicate garnets: pyrope (X = Mg and Y = Al), grossular (X = Ca and Y = Al), and andradite (X = Ca and Y = Fe). Pyrope

a Ox Oy Oz X1 O X2 O YO SiO

Grossular

Andradite

Calc.

Exp.

Calc.

Exp.

Calc.

Exp.

11.5447 0.03214 0.04971 0.65344 2.2052 2.3648 1.8987 1.6496

11.4390 0.03291 0.05069 0.65331 2.1959 2.3335 1.8850 1.6337

11.9368 0.03740 0.04515 0.65156 2.3301 2.5058 1.9398 1.6627

11.8450 0.03823 0.04528 0.65137 2.3218 2.4865 1.9255 1.6459

12.1960 0.03893 0.04838 0.65617 2.3780 2.5331 2.0497 1.6612

12.0510 0.03914 0.04895 0.65534 2.3584 2.4953 2.0186 1.6492

Zicovich-Wilson et al. (2008). Reproduced with permission of Wiley. Experimental data are from low-temperature X-ray diffraction experiments; see Zicovich-Wilson et al. (2008) for further details.

a brief account will be given of these two alternative approaches for including the effect of pressure on computed structural properties of minerals, with particular reference to the case of silicate garnets. The inclusion of the effect of temperature on such properties will be briefly discussed in Section 3.7.

3.3.1

P–V Relation Through Analytical Stress Tensor

The stress tensor σ is a symmetric second-rank tensor that can be computed analytically from the total energy density derivatives with respect to strain: σ ij =

1 ∂E 1 3 ∂E = aκj , V ∂ϵij V κ = 1 ∂aκi

38

with ϵ second-rank symmetric pure strain tensor and i,j, k = x,y,z. In the second equality, ∂E ∂ϵij has been expressed in terms of analytical energy gradients with respect to lattice parameters, with aij elements of a 3 × 3 matrix, A, where Cartesian components of the three lattice vectors a1, a2, and a3 are inserted by rows [V = a1 a2 × a3 is the cell volume]; when a distortion is applied to the cell, the lattice parameters transform as 3

δjk + ϵjk aik ,

aij =

39

k=1

where δjk is the Kronecker delta. The difficult part of the calculation of the stress tensor in Equation 3.8 is the evaluation of the analytical energy gradients with respect to the cell parameters, which have been implemented in the CRYSTAL program about 10 years ago by Doll et al. for 1-D, 2-D, and 3-D periodic systems (Doll et al. 2004, 2006). An external “prestress” in the form of a hydrostatic pressure P, σ pre ij = Pδij ,

3 10

84

Molecular Modeling of Geochemical Reactions

can be added to that of Equation 3.8. Given that the optimizer works in terms of analytical cell gradients, in order to perform a pressure-constrained geometry optimization, the total stress tensor has to be back-transformed to obtain the corresponding constrained gradients: ∂H ∂E = + PV A −1 ∂aij ∂aij

3 11

ji

Let us note that, with the inclusion of a hydrostatic pressure, the function to be minimized becomes the enthalpy H = E + PV (Souza and Martins 1997). The geometry optimizer under an external hydrostatic pressure has been implemented by Doll (2010) in the CRYSTAL program: the optimized volume V of any crystal at a given hydrostatic pressure P can then be computed analytically. In a couple of recent studies (Erba et al. 2014a; Mahmoud et al. 2014a), the P–V relation of six silicate garnet end-members has been computed using B3LYP hybrid first-principles simulations; results are summarized in Figure 3.1 and compared with available experimental data. For synthetic pyrope, a single-crystal X-ray diffraction study up to 33 GPa by Zhang et al. (1998) was available (solid circles); a subsequent study by Zhang et al. (1999) on synthetic single-crystal grossular, andradite, and almandine up to 12, 14, and 22 GPa, respectively, was taken as a reference (solid triangles) for these three systems. Computed values for almandine are reported up to 6 GPa only because for higher pressures the SCF electronic structure calculation of the system is not converging within the numerical accuracy required by such calculations. For spessartine and uvarovite, the high-pressure

1.00 Pyrope

Almandine

Spessartine

Grossular

Andradite

Uvarovite

V/V0

0.95

0.90

0.85

1.00

V/V0

0.95

0.90

0.85 0

10

20

30

40 0

10

20

30

40 0

10

20

30

P (GPa)

Figure 3.1 Pressure–volume relation of six silicate garnet end-members as computed with B3LYP first-principles simulations (continuous lines) and measured experimentally (symbols); see text for details on the experiments.

40

Quantum-Mechanical Modeling of Minerals

85

X-ray synchrotron diffraction study by Diella et al. (2004) up to 25 and 35 GPa, respectively, is taken as a reference (solid pentagons). We also report the values obtained by Leger et al. (1990) in their powder X-ray diffraction study as a function of pressure for spessartine and uvarovite up to 25 GPa (open circles). For spessartine, other two compressional experiments are considered: a recent pressure–volume–temperature study by Gréaux and Yamada (2012) up to 13 GPa (open triangles) and the single-crystal X-ray diffraction study by Zhang et al. (1999) up to 15 GPa (solid triangles). The agreement between the computed P–V relation and the experimentally determined one is overall very satisfactory, as can be inferred from insight of Figure 3.1. As we will discuss in Section 3.4, the reliable description of the P–V relation constitutes an essential prerequisite to the study of elastic properties of minerals under geophysical pressures. 3.3.2

P–V Relation Through Equation of State

An alternative and commonly adopted approach for establishing the P–V relation of a mineral is using so-called equations of state (EOSs). “Cold” EOSs are energy–volume or pressure–volume relations which describe the behavior of a solid under compression and expansion, at T = 0K, that is the case of standard quantum-mechanical simulations. Here we are interested in universal EOSs (i.e., not specific of particular materials) that are generally expressed as analytical functions of a limited set of parameters (equilibrium energy E0, equilibrium volume V0, equilibrium bulk modulus K0 = −V ∂P ∂V , and pressure derivative of equilibrium bulk modulus K0 = ∂K0 ∂P) for ease of interpolation, extrapolation, and differentiation and are quite used in solid-state physics and geophysics (Alchagirov et al. 2001; Cohen et al. 2000). EOSs have experienced a large success in theoretical simulations, in that, in principle, they would allow for passing from (few) energy–volume data in the vicinity of the equilibrium volume to the P–V relation and, possibly, to high-pressure properties. To do so, energy–volume data are numerically fitted to the analytical E(V) functional form of the EOS. From P = − ∂E ∂V , the P–V connection is established. Let us stress, however, that the analytical expression of the E(V) relation is generally obtained as a series which is truncated to some order. By taking derivatives of increasing order of this expression (for computing pressure P, bulk modulus K, and its pressure derivative K ), the error introduced by that truncation increases. Many universal EOSs have been proposed so far (Alchagirov et al. 2001; Birch 1947, 1978; Holzapfel 1996; Murnaghan 1944; Poirier and Tarantola 1998; Vinet et al. 1986). They are all phenomenological and can behave quite differently from each other as regards extrapolation at high pressures. Comprehensive reviews and comparisons of different EOSs are available in the literature (Anderson 1995; Angel 2000; Duffy and Wang 1998; Hama and Suito 1996; Stacey et al. 1981). Four EOSs are currently implemented in the CRYSTAL program: the original third-order Murnaghan’s (1944), the third-order Birch’s (1947, 1978), the logarithmic Poirier–Tarantola’s (1998), and the exponential Vinet’s (1986). The analytical stress tensor approach illustrated in Section 3.3.1 provides an extremely satisfactory description of the P–V relation of the six silicate garnets here considered (see Figure 3.1). One may wonder about the accuracy that could be reached in computing the same P–V relation by following the alternative scheme based on the EOSs. When extrapolating to high pressure, the four considered EOSs slightly deviated from each other, still remaining relatively close to the stress tensor reference (with maximum differences of 0.4% for pyrope, at 60 GPa, with the Poirier–Tarantola logarithmic EOS; 0.6% for grossular, at 60 GPa, with the Murnaghan EOS; 0.5% for andradite, at 40 GPa, with the Murnaghan EOS, for instance). Since differences among the four EOSs are very small, it is difficult to tell which one is providing the best description as regards the P–V relation of this family of garnets: Birch–Murnaghan for pyrope, Vinet for grossular, and Poirier–Tarantola for andradite, for

86

Molecular Modeling of Geochemical Reactions

instance. All of them are essentially providing an acceptable description of the compressibility of these minerals. On the contrary, large deviations from the analytical reference have been observed as regards the pressure dependence of the bulk modulus, the third-order Birch–Murnaghan one providing the best description among them (Erba et al. 2014a; Mahmoud et al. 2014a).

3.4 Elastic Properties The physicochemical composition of the Earth’s deep interior still has to be properly determined. In this respect, seismological data collected during earthquakes constitute the main source of information. In order to correctly interpret these outcomes and, possibly, to distinguish among different compositional models, to understand how seismic waves propagate during an earthquake and to trace plate dynamics, the individual elastic response of all possible constituents of the Earth’s upper mantle, transition zone, and lower mantle should be fully characterized (i.e., the corresponding elastic tensors determined). This kind of characterization is particularly challenging from an experimental point of view in that simultaneous high temperatures and pressures have to be considered (Anderson and Isaak 1995; Mainprice et al. 2013). Experimentally, compression/expansion studies at relatively low pressures/temperatures have extensively been used to fit data to various EOSs and to extrapolate at geophysical conditions (see Section 3.3.2); first derivatives of the bulk modulus K with respect to pressure and temperature could be determined which, however, showed large discrepancies among each other. As regards pressure, for instance, the comprehensive work by Knittle (1995) documents how, at ambient pressure, different experimental determinations of K agree relatively well to each other, while at high pressures disagreements up to 50% are commonly reported on K . High pressure or high-temperature single-crystal experimental elastic studies are now becoming feasible, but still, measurements of the elastic constants of a mineral single crystal at simultaneous high pressure and temperature are rare. Computational material science does represent, in principle, a powerful alternative, especially so if first-principles simulations are considered. DFT constitutes, indeed, an accurate theoretical framework for simulating elastic properties of minerals (Karki et al. 2001). Pressure and temperature are very different thermodynamic variables to be properly accounted for within quantum-mechanical calculations. The effect of pressure can be introduced in a relatively simple way into the picture by an EOS approach or by an analytical stress tensor approach (see Section 3.3). In both cases, the equilibrium volume (and, more generally, structure) of the crystal at any given pressure can be computed quite accurately and at relatively low computational cost: one essentially needs to optimize, under given constraints, the crystal structure at different pressures/volumes. According to this scheme, high-pressure elastic properties of many minerals of the Earth’s mantle have been computed from first principles by Karki and collaborators (Karki et al. 2001). The inclusion of temperature effects on computed elastic properties is by far more complex and requires to go somehow beyond the harmonic approximation to the lattice potential according to which the lattice thermal expansion would, indeed, be zero. We will briefly address this issue in Section 3.7. 3.4.1

Evaluation of the Elastic Tensor

In the absence of any finite prestress, elastic constants can be defined as second energy density derivatives with respect to pairs of infinitesimal Eulerian strains: Cijkl =

1 ∂2 E V0 ∂ϵij ∂ϵkl

, ϵ=0

3 12

Quantum-Mechanical Modeling of Minerals

87

where V0 is the equilibrium volume. These constants do represent the link between stress and strain via Hooke’s law. In the limit of zero temperature, typical of quantum-mechanical simulations, they are also referred to as athermal elastic constants. The elastic tensor ℂ is a fourth-rank tensor which, in principle, should be characterized by 81 components. Given that the pure strain tensor ϵ is symmetric, ℂ, in general, exhibits only 21 independent elements due to the following possible permutations among its indices: i j , k l , and ij kl coming from the invariance with respect to the order at which the two derivatives in Equation 3.12 are performed. The point symmetry of the different lattices can further reduce the number of independent constants down to 3 (for cubic crystals). A fully automated implementation of the elastic tensor calculation is available in the CRYSTAL program (Erba et al. 2014b; Perger et al. 2009). If a finite prestress σ pre is applied in the form of a hydrostatic pressure P, as in Equation 3.10, within the frame of finite Eulerian strain, the corresponding elastic stiffness constants read (Karki et al. 1997, 2001; Wallace 1965, 1972; Wang et al. 1995): Bijkl = Cijkl +

P 2δij δkl − δil δjk − δik δjl , 2

3 13

provided that V0 in Equation 3.12 becomes the equilibrium volume V(P) at pressure P. In the fully automated implementation in the CRYSTAL program of the calculation of the stiffness tensor B (and of S = B − 1 , the compliance tensor) under pressure (Erba et al. 2014a), V(P) is obtained from the analytical stress tensor described in Section 3.3.1. An option exists for using the V(P) relation obtained from a given EOS, as discussed in Section 3.3.2. Since both ϵ and δ are symmetric tensors, we can rewrite equality (3.13) as 0 P P

0

0

0

P 0 P

0

0

0

0

0

0

0

P P 0 Bυu = Cυu +

0 −P 0 0 0 2 0 0 0

0

−P 2

0

0 0 0

0

0

−P 2

,

3 14

where Voigt’s notation (Nye 1957) has been used, according to which v, u = 1, …, 6 (1 = xx, 2 = yy, 3 = zz, 4 = yz, 5 = xz, 6 = xy). Following the procedure described previously, we have computed the elastic stiffness constants Bvu of pyrope and grossular up to 60 GPa and those of andradite up to 40 GPa at B3LYP level of theory using all-electron BSs and the CRYSTAL program (Erba et al. 2014a). The three symmetry-independent constants, B11, B12, and B44, are reported in Figure 3.2 as black lines; the corresponding dependence on pressure is rather similar for the three garnets, and it is quasilinear, B12 showing a slightly more linear behavior than B11 and B44. Available experimental data, as obtained from Brillouin scattering measurements, are also reported in the figure. For pyrope, two measurements are reported: one by Sinogeikin and Bass (2000) who reported values at ambient pressure and at P = 14 GPa (filled symbols) and one by Conrad et al. (1999) who reported values at three different pressures (empty symbols). From inspection of the left panel of Figure 3.2, one can clearly observe that (i) the two experimental datasets agree relatively well with

88

Molecular Modeling of Geochemical Reactions

600

B11

Pyrope

Grossular

B11

Andradite

B11

Bvu (GPa)

500

400

300

200

B12

B12

B12

B44

B44

B44

100

0

0

10 20 30 40 50 60 P (GPa)

0

10 20 30 40 50 60 P (GPa)

0

10 20 30 40 50 60 P (GPa)

Figure 3.2 Elastic stiffness constants Bvu of pyrope (left panel), grossular (central panel), and andradite (right panel), as a function of pressure P. Black lines represent computed values. All experimental values are obtained from Brillouin scattering measurements (see text for details). Erba et al. (2014a). Reproduced with permission of AIP.

each other, (ii) absolute computed values of Bvu at zero pressure are extremely close to their measured counterparts (Erba et al. 2014b,c), and (iii) the computed pressure dependence of the elastic stiffness constants satisfactorily matches available experimental data in the low-pressure regime and is thus expected to be rather reliable in the high-pressure predictions. Also for grossular, two experiments are available to compare with: results by Jiang et al. (2004a) that refer to a 87% grossular-rich garnet (at eight different pressures up to 11 GPa) are reported as filled symbols; values obtained by Conrad et al. (1999) at five different pressures up to 10 GPa for the B11 constant are also reported as empty symbols. We observe that (i) the absolute values of the elastic constants at ambient pressure agree with the experiments as regards B12 and B44, while B11 is slightly overestimated in this case (let us recall that the experiment by Jiang et al. was performed on a 87% grossular-rich garnet with 9% of andradite which exhibits lower elastic constants than grossular) and (ii) the pressure dependence of all elastic constants nicely compares with the experimental behavior; in particular, the low-pressure crossing of B12 and B44 is perfectly reproduced. For andradite, two experimental datasets are available: Jiang et al. (2004b) reported values at nine pressures up to 11 GPa (filled symbols). Again, for B11 we also report, as empty symbols, the less accurate results by Conrad et al. (1999) (two crystal directions were considered at each pressure with respect to 36 in Jiang et al. (2004b)) who measured the elastic constants at five pressures up to 10 GPa. Some considerations are as follows: (i) the two experiments describe a very different pressure dependence of B11, (ii) both the absolute values at ambient pressure and the pressure dependence of computed constants are in good agreement with data by Jiang et al. (2004b), and (iii) given the reliable description of the pressure dependence of computed elastic constants of pyrope and grossular, our results for andradite confirm the higher accuracy of the measurements by Jiang et al. (2004b) with respect to those by Conrad et al. (1999).

Quantum-Mechanical Modeling of Minerals

3.4.2

89

Elastic Tensor-Related Properties

Several elastic properties of isotropic polycrystalline aggregates can be computed from the elastic stiffness and compliance constants defined previously via the Voigt–Reuss–Hill averaging scheme (Hill 1963). In particular, for cubic crystals (as in the case of silicate garnets that we will discuss in the following), the adiabatic bulk modulus K is simply defined as K=

1 1 B11 + 2B12 ≡ S11 + 2S12 3 3

−1

3 15

The average shear modulus G can be expressed as G=

1 5 B11 −B12 + 3B44 + 4 S11 −S12 + 3S44 10 2

−1

3 16

From the bulk modulus and the average shear modulus defined earlier, Young’s modulus E and Poisson’s ratio σ can be defined as well: E=

9K0 G 3K0 − 2G and σ = 3K0 + G 2 3K0 + G

3 17

General expressions for crystals of any symmetry of all these quantities in terms of elastic and compliance constants can be found, for instance, in Ottonello et al. (2010). The spatial anisotropy of Young’s modulus, linear compressibility, shear modulus, and Poisson’s ratio can also be evaluated from the computed elastic tensor and fully characterized (Marmier et al. 2010; Nye 1957). 3.4.3

Directional Seismic Wave Velocities and Elastic Anisotropy

Single-crystal Brillouin scattering experiments allow for the accurate determination of directional seismic wave velocities and elastic anisotropy of a crystal. The manifestations of elastic anisotropy might be not so evident from the sole inspection of the elastic tensor and include (i) shear-wave birefringence, that is, the two polarizations of transverse waves travel with different velocities, and (ii) azimuthal anisotropy, that is, the seismic wave velocities depend on propagation direction. If not properly recognized, anisotropic effects are generally interpreted as due to inhomogeneities such as layering or gradients and can lead to validation of wrong compositional models (Anderson 1989). Despite their relevance to the compositional analysis of the Earth’s interior, only few directional studies have been performed so far for silicate garnets (Jiang et al. 2004a,b; Sinogeikin and Bass 2000), while most studies report average seismic wave velocities (Chen et al. 1999; Gwanmesia et al. 2014, 2006; Jiang et al. 2004a,b; Kono et al. 2010; Sinogeikin and Bass 2000). According to the elastic continuum model, the three acoustic wave velocities, along any general crystallographic direction represented by unit wave vector q, can be related to the elastic constants by Christoffel’s equation which can be given an eigenvalue/eigenvector form as follows (Auld 1973; Musgrave 1970): Aq U = V2 U with Aqjk =

1 ρ

qi Cijkl ql ,

3 18

il

where Aqjk is Christoffel’s matrix, ρ the crystal density, i,j, k, l = x,y, z represent Cartesian directions, qi is the ith element of the unit vector q, V is a 3 × 3 diagonal matrix whose three elements give the acoustic velocities, and U = u1 ,u2 , u3 is the eigenvector 3 × 3 matrix where each column represents

90

Molecular Modeling of Geochemical Reactions

the polarization û of the corresponding eigenvalue. The three acoustic wave velocities, also referred to as seismic velocities, can be labeled as quasilongitudinal υp, slow quasitransverse υs1, and fast quasitransverse υs2, depending on the polarization direction û with respect to wave vector q (Karki et al. 2001). As anticipated in the preceding, the elastic anisotropy of a crystal can be fully characterized from directional seismic wave velocities. The azimuthal anisotropy for quasilongitudinal and quasitransverse seismic wave velocities can be defined as follows (Karki et al. 2001): AX =

υX max −υX min , υX

3 19

where X = p, s labels longitudinal and shear waves, and υX is the polycrystalline isotropic average velocity obtained from the Voigt–Reuss–Hill scheme (Hill 1963). Elastic anisotropy would be zero for an ideal isotropic material; even cubic crystals, such as silicate garnets, however, show a nonzero elastic anisotropy (Karki et al. 2001). For cubic crystals, the elastic anisotropy can be given a simple expression in terms of a single anisotropy index computed from the elastic constants (Authier and Zarembowitch 2006): A=

2B44 + B12 − 1 × 100 B11

3 20

In Figure 3.3, computed directional seismic wave velocities are reported for grossular, uvarovite, spessartine, pyrope, andradite, and almandine along an azimuthal angle θ which spans the (110) crystallographic plane of the lattice by exploring all the high-symmetry crystallographic directions: θ = 0∘ corresponds to the crystallographic direction [110], θ = 45∘ to [111] direction, θ = 90∘ to [001] direction, etc. Computed velocities are reported as continuous lines of increasing thickness as a function of pressure. Available directional experimental data are also reported in the figure: for andradite, data from an accurate single-crystal Brillouin scattering experiment by Jiang et al. (2004b) are reported at ambient pressure (full squares) and at 8.7 GPa (full circles); a subsequent study by Jiang et al. (2004a) on a single-crystal grossular-rich garnet at 4.3 GPa (full circles) is also taken as a reference; for pyrope, data from the study by Sinogeikin and Bass (2000) are reported. From inspection of the figure, the accuracy of the theoretical description of angular dependence, oscillation amplitudes, and pressure shift of the seismic wave velocities can be clearly seen. From Figure 3.3, the six end-members can be sorted according to increasing propagation velocity, at zero pressure, as follows: Alm < And < Spe < Uva < Pyr < Gro; this sequence does not change under increasing pressure. Almandine shows the slowest υp, while pyrope and grossular allow for the fastest propagation. This behavior can be rationalized in terms of the elemental composition of the end-members taking into account that seismic wave velocities are inversely proportional to the density of the material. Fe-bearing phases such as andradite and almandine are the most dense, followed by the Mn- and Cr-bearing phases, such as spessartine and uvarovite, whereas pyrope and grossular contain the lightest elements (Mg, Ca) thus being the least dense. Seismic wave velocities increase as pressure increases in all cases. In the absence of external pressure, the six silicate garnet end-members can be sorted in terms of increasing elastic anisotropy, as follows: Spe < Pyr < Alm < Gro < And Uva. Spessartine and pyrope show very low anisotropy, while uvarovite is by far the most anisotropic among them. The elastic anisotropy of grossular, uvarovite, spessartine, and andradite increases as a function of pressure. In particular, grossular and spessartine are the most affected by pressure, with anisotropies varying from −5.6 to −13.2% and from −2.5 to −5.1%, respectively, when passing from 0 to 40 GPa; the elastic anisotropy of andradite increases from −10 to −14.5% in the same pressure range, while the anisotropy of

Quantum-Mechanical Modeling of Minerals

91

11

Seismic velocity (km/s)

10 9 8 Grossular

Uvarovite

Spessartine

Pyrope

Andradite

Almandine

7 6 5

11

Seismic velocity (km/s)

10 9 8 7 6 5 0

45

90 θ (°)

135

180

0

45

90 θ (°)

135

180

0

45

90

135

180

θ (°)

Figure 3.3 Directional quasitransverse and quasilongitudinal seismic wave velocities of single-crystal grossular, uvarovite, spessartine, pyrope, andradite, and almandine along an azimuthal angle θ (defined in the text). Computed data at different pressures (0 GPa, 4 GPa, 8 GPa, 12 GPa, 20 GPa, 30 GPa) are reported as continuous lines of increasing thickness. Experimental data are reported when available (see text for details). Mahmoud et al. (2014a). Reproduced with permission of AIP.

uvarovite only slightly increases from −15.8 to −16.5% passing from 0 up to 30 GPa (Mahmoud et al. 2014a). Pyrope and almandine show a different behavior under pressure: their elastic anisotropy decreases. If at ambient pressure pyrope shows a larger anisotropy than spessartine, as soon as pressure increases, the anisotropy of spessartine becomes larger than pyrope.

3.5

Vibrational and Thermodynamic Properties

A wealth of information about the chemical composition, structure, and thermodynamics of minerals can be inferred from vibrational spectroscopic measurements. In this respect, quantummechanical computational spectroscopy has proven to be an extremely useful complementary tool

Molecular Modeling of Geochemical Reactions 800

500

700 X1 X2 X4 X8 X16

6

200

4

100

2

0

0

200

400 600 T (K)

800

0 1000

Γ

600 S [J/(K*mol)]

300

8 ΔCv [J/(K*mol)]

Cv [J/(K*mol)]

400

b2

b1

21

P H

18

N b3

500

15

400

12

300

9

200

6

100

3

0

0

200

400

600

800

ΔS [J/(K*mol)]

92

0 1000

T (K)

Figure 3.4 Constant-volume specific heat CV and entropy S of pyrope as a function of temperature, as computed at B3LYP level of theory (thick continuous line) with the largest SC considered (viz., X27) and compared with experimental data (full circles) from Haselton and Westrum (1980) and Tequi et al. (1991). On the right scale of the two panels, ΔCV = CVX27 − CVXn and ΔS = S X27 − S Xn are reported that show the convergence of computed thermodynamic properties on the size of the adopted SC (n = 1, 2, 4, 8, 16). The inset of the right panel shows the shape of the first Brillouin zone of silicate garnets. (a) Haselton and Westrum (1980). Reproduced with permission of Elsevier. (b) Tequi et al. (1991). Reproduced with permission of Elsevier.

for achieving full characterization of IR and Raman spectra (see Chapter 10) in terms of peak positions, intensities, and band classifications. In the CRYSTAL program, the calculation of vibration frequencies at the Γ point (k = 0, at the center of the first Brillouin zone (FBZ) in reciprocal space; for a graphical representation of the shape of the FBZ of silicate garnets, see the inset of Figure 3.4), within the harmonic approximation, is available since 2003 (Pascale et al. 2004b; Zicovich-Wilson et al. 2004). The vibration frequencies at the center of the FBZ (directly comparable with the outcomes of IR and Raman measurements) are obtained from the diagonalization of the mass-weighted Hessian matrix of the second derivatives of the total energy per cell with respect to atomic displacements u: WaiΓ, bj =

Hai0 , bj Ma Mb

∂2 E , with H0ai, bj = ∂u0ai ∂u0bj

3 21

where atoms a and b (with atomic masses Ma and Mb) in the reference cell are displaced along the ith and jth Cartesian directions. The first derivatives of the total energy per cell gia = ∂E ∂uai with respect to atomic displacements from the equilibrium configuration eq are computed analytically, whereas second derivatives numerically, using a two-point formula: ga ∂2 E ≈ i ∂uai ∂ubj

eq

, ubj = + u − gia 2u

eq

,ubj = −u

,

where u = 0 003 Å, a value 10–50 times smaller than that usually used in other solid-state programs (Kresse and Furthmüller 1996a,b; Soler et al. 2002).

Quantum-Mechanical Modeling of Minerals

93

Chapter 10 is entirely devoted to the presentation of the many tools that have been developed and implemented into the public CRYSTAL program as regards the analysis of spectroscopical properties of mineral calculation of IR and Raman analytical intensities, simulation of IR and Raman spectra, isotopic substitution effect, longitudinal optical (LO)/transverse optical (TO) splitting, anharmonic corrections to hydrogen stretching modes, graphical analysis of the normal modes of vibration, scanning of energy along normal mode coordinates, etc. Several examples of applications of these techniques to minerals of geochemical interest are also given there. In the next part of this section, we will briefly review the main studies about first-principles investigations of vibrational spectroscopic properties of silicate garnets and then move to discussing how thermodynamic properties of minerals can be simulated with standard quantum-mechanical techniques. The IR and Raman vibration frequencies of pyrope and andradite have been first computed quantum mechanically, using the B3LYP hybrid functional, in 2005 (Pascale et al. 2005a,b); all vibration modes have been characterized by evaluating the effect of the isotopic substitution. In the following years, the same methodology has been applied to other members of the silicate garnet family: grossular (Dovesi et al. 2009), spessartine (Valenzano et al. 2009), uvarovite (Valenzano et al. 2010), and almandine (Ferrari et al. 2009). In 2008, the LO/TO splitting, requiring the information about the dielectric response of the system, of pyrope, grossular, and andradite has been computed (ZicovichWilson et al. 2008). The complete IR reflectance spectra of the six end-members have been simulated by Dovesi et al. (2011). The 17 IR-active F1u TO and LO frequencies, the corresponding oscillator strengths, the high-frequency and static dielectric constants, and the reflectance spectra have been computed. The agreement with experiments for the TO and LO peaks has been documented to be always rather satisfactory, the mean absolute difference for the whole set of data (178 peaks in total) being 5 cm–1. The reflectance spectra, simulated through the classical dispersion relation, had reproduced the experimental curves extremely well. For this class of systems, the B3LYP hybrid functional has been shown to significantly improve over LDA or GGA ones (Maschio et al. 2011). Only recently, the implementation of analytical Raman intensities in the CRYSTAL14 version of the program has made the quantum-mechanical simulation of Raman spectra of crystalline materials possible (Dovesi et al. 2014a). A couple of recent studies have reported about these kinds of simulations for pyrope and grossular (Maschio et al. 2014, 2013). An example will be briefly illustrated in Section 3.6 as regards the simulation of the IR spectroscopical properties of the grossular–andradite solid solution series Ca3Fe2–2xAl2x(SiO4)3 over the whole chemical composition range 0 ≤ x ≤ 1 (De La Pierre et al. 2013). 3.5.1

Solid-State Thermodynamics

The calculation of the thermodynamic properties of crystals is generally more demanding with respect to the sole spectroscopic characterization as it requires the knowledge of phonon modes over the complete FBZ; phonons at points different from Γ can be obtained by building a supercell (SC) of the original unit cell, following a direct-space approach (Parlinski et al. 1997; Togo et al. 2008). l g a identify the general crystal cell where {at} are the direct lattice basis The lattice vectors g = t t t vectors, with t = 1,…, D (where D is the dimensionality of the system: 1, 2, 3 for 1-D, 2-D, 3-D periodic systems): within periodic boundary conditions the integers ltg run from 0 to Lt–1. The parameters {Lt} define size and shape of the SC in direct space. Let us label with G the general superlattice (i.e., whose reference cell is the SC) vector, and let us introduce the L = Πt Lt Hessian matrices {Hg} whose elements are Haig , bj = ∂2 E

g ∂u0ai ∂ubj where, at variance with Equation 3.21,

atom b is displaced in cell g, along with all its periodic images in the crystal (i.e., in cells g + G). The set of L Hessian matrices {Hg} can be Fourier transformed into a set of dynamical matrices {Wk},

94

Molecular Modeling of Geochemical Reactions

each one associated with a wave vector k = and the integers kt run from 0 to Lt–1: L− 1

Wk =

M−

t

1 2

kt Lt bt where {bt} are the reciprocal lattice vectors

HgM −

1 2

exp ιk g ,

3 22

g=0

where M is the diagonal matrix with the masses of the nuclei associated with the 3M atomic coordinates where M is the number of atoms per cell. The solution is then obtained through the diagonalization of the L matrices {Wk}: †

U k W k U k = Λk with

†

Uk Uk = I

3 23

The elements of the diagonal Λk matrix provide the vibrational frequencies, vik =

λik (au are

adopted), while the columns of the Uk matrix contain the corresponding normal coordinates. To each k-point in the FBZ, 3M harmonic oscillators (i.e., phonons) are associated which are labeled by a phonon band index i (i = 1,…,3M) and whose energy levels are given by the usual harmonic expression: εim, k = m +

1 2πvik , 2

3 24

where m is an integer. The overall vibrational canonical partition function, Qvib (T), at a given temperature T, can be expressed as follows: L− 1 3M

∞

Qvib T = k=0 i=1 m=0

exp −

εim, k , kB T

3 25

where kB is Boltzmann’s constant. According to standard statistical mechanics, thermodynamic properties of crystalline materials such as entropy S and thermal contribution to the internal energy U can be expressed as S T = kB T

∂log Qvib ∂T

+ kB log Qvib

and U T = kB T 2

∂ log Qvib ∂T

3 26

From the previous expression for U, the constant-volume specific heat CV can also be computed according to CV = ∂U ∂T. In the first-principles simulation of the thermodynamic properties of minerals, convergence of computed properties has to be carefully checked with respect to the number of k-points considered for the phonon dispersion (or, equivalently in the direct-space approach, with respect to the size of the adopted SC). Here, we consider the theoretical description of the evolution with temperature of the constant-volume specific heat, CV, and of the entropy, S, of the most abundant among silicate garnets: pyrope, Mg3Al2(SiO4)3. We compute phonon frequencies and, via Equations 3.24–3.26, thermodynamic properties with SC of increasing size (corresponding to an increasing number of k-points in the previous formalism). We start from the primitive cell of pyrope, containing 80 atoms and corresponding to 1 k-point, that we label X1; then, five other SCs are built that we label Xn as they are n times larger than the primitive one, where n = 2, 4, 8, 16, and 27 (i.e., containing up to 2160 atoms for X27 and corresponding to 27 k-points in the FBZ). All the considered SCs are cubic so that

Quantum-Mechanical Modeling of Minerals

95

their high symmetry can be exploited for reducing the computational cost of all the calculations (in all cases only nine SCF calculations, plus computation of the atomic gradients, are required). In the two panels of Figure 3.4, we report the constant-volume specific heat CV and entropy S of pyrope as computed at B3LYP level of theory (thick continuous line) with the largest SC considered (viz., X 27), and we compare with available experimental data (full circles) from Haselton and Westrum (1980) and Tequi et al. (1991). On the right scale of each of the two panels, ΔCV = CVX27 −CVXn and ΔS = S X27 −S Xn are reported that show the convergence of computed thermodynamic properties with respect to the size of the adopted SC (n = 1, 2, 4, 8, 16). It turns out that, already with an X8 SC, results are converged within 0.2% to the X27 ones for both specific heat and entropy. The agreement with experimental measurements is rather satisfactory and better than previously reported (Hofmeister and Chopelas 1991).

3.6

Modeling Solid Solutions

Solid solutions of different chemical compositions are of crucial relevance to the study of geochemical properties of minerals of the Earth’s mantle. Natural silicate garnets, for instance, can be found in a wide range of chemical compositions since they form solid solutions. Solid solutions between pyralspites (pyrope–almandine–spessartine) and ugrandites (uvarovite–grossular–andradite) seldom occur in natural garnets (Hensen 1976), whereas aluminosilicate garnets with different X cations show complete solid solubility among them at high pressure (O’Neill et al. 1989). Linear composition-bulk modulus trends have been observed for garnets in the pyralspite series (Duffy and Anderson 1989; Isaak and Graham 1976; Yeganeh-Haeri et al. 1990). On the contrary, by combining the few available experimental bulk moduli of grandite (grossular–andradite) solid solutions (Babuška et al. 1978; Bass 1986; O’Neill et al. 1989), a significant deviation from linearity seems to turn out, as discussed by O’Neill et al. (1989). However, uncertainty and nonhomogeneity of the measurements still leave room for further insights (see later). New algorithms have recently been developed and implemented into the CRYSTAL14 program for the study of solid solutions and, more generally, disordered systems (D’Arco et al. 2013; Mustapha et al. 2013). As far as solid-state solutions are concerned, the computational scheme is as follows: (i) the periodic nature of the system is preserved; (ii) in general, an SC is built in order to increase the number of atomic sites involved in the substitution; and (iii) for any chemical composition within a given series, the program finds the total number of atomic configurations and determines the symmetry-irreducible ones to be explicitly considered in order to define a statistical average. In this section, we present the results of a first-principles theoretical study of the elastic properties of the grandite solid solution, Ca3Fe2–2xAl2x(SiO4)3, as a function of its chemical composition x. Reference is made to the primitive unit cell of the end-members (cubic space group ≡ Ia3d), which counts = 48 symmetry operators and four formula units Ca3Y2(SiO4)3. There are eight Y sites involved for substitution. Solid solutions are obtained from andradite by progressively replacing Fe3+ with Al3+ cations. Apart from the two end-members, andradite x = 0 and grossular x = 1 , other seven compositions are explicitly considered: x = 0.125, 0.25, 0.375, 0.5, 0.625, 0.75, and 0.875. For each composition x, nAl = 8x aluminum atoms are present that correspond to 8 !/[nAl ! (8 – nAl) !] different substitutional configurations, that is, cation distributions among the Y sites. There is a total of 256 possible atomic configurations over the whole range of compositions. Following the symmetry analysis recently proposed by Mustapha et al. (2013), these configurations can be partitioned into 23 distinct symmetry-independent classes (SIC). Each class L consists of L = | | ∕ | L| configurations that belong to a symmetry subgroup of order | L| of

96

Molecular Modeling of Geochemical Reactions

175

K (GPa)

170 165 160 155 150 145 Andradite 0

Grossular 0.2

0.4

0.6

0.8

1

x

Figure 3.5 Bulk modulus K of the grandite solid solution, Ca3Fe2–2xAl2x(SiO4)3, as a function of its chemical composition x. Experimental data are reported as full and empty symbols (see text for details). When available, error bars are also shown. The solid line shows the quasilinear trend of our calculated values, whereas the thick dashed curve is drawn to provide an approximate fit to the experimental data, as suggested by Bass (1986) and O’Neill et al. (1989). Reproduced with permission of Wiley.

the aristotype symmetry. Since all the configurations of a given SIC are equivalent to each other, the number of calculations to be actually performed reduces to one per SIC, that is, to a total of 23. L can then be interpreted as the multiplicity of class L (see Lacivita et al. (2014) for details on the symmetry properties of the 23 configurations). Only the highest spin ferromagnetic configurations will be considered, as the difference between ferromagnetic and antiferromagnetic energies was shown to be extremely small (Meyer et al. 2010). In Figure 3.5, available experimental determinations of the bulk modulus K of grandite as a function of its chemical composition x are reported as full and empty symbols (Babuška et al. 1978; Bass 1986, 1989; Halleck 1973; Isaak and Graham 1976; Jiang et al. 2004a,b). For the grossular endmember, two values are reported (at x ≈ 1, natural single-crystal samples pure to 97 and 99%, respectively): an ultrasonic measurement by Halleck (1973) (triangle) and a Brillouin scattering experiment performed by Bass (1989) (square). For andradite, two values are given as determined by Brillouin spectroscopy by Bass (1986) (circle) and Jiang et al. (2004b) (inverted triangle). From inspection of Figure 3.5, where error bars are also reported, we can clearly see how different experimental determinations of the bulk moduli of the two end-members show finite discrepancies between each other. Experimental uncertainties become even larger for most of the intermediate compositions, as we shall discuss in the following. The main experimental investigation of the elastic properties of intermediate compositions of the grandite solid solution has been performed by Babuška et al. (1978). Four specimens of three different compositions were analyzed (rhombi in the figure). More recently, Jiang et al. (2004a) performed Brillouin spectroscopy on a grossular-rich garnet (pentagon). The thick dashed line in Figure 3.5 represents the bulk modulus trend as a function of composition x as proposed by Bass (1986) and O’Neill et al. (1989) on the grounds of available experimental data at that time. More recently, an in situ X-ray diffraction experiment has been performed by Fan et al. (2011) on three specimens of intermediate chemical compositions in the grandite binary. The corresponding bulk moduli have been evaluated and, though significantly overestimated

Quantum-Mechanical Modeling of Minerals

97

with respect to all previous determinations, seem to exhibit a linear behavior (see empty circles in Figure 3.5). The results of theoretical simulations explicitly accounting for the effect of configurational disorder are reported in Figure 3.5 as a solid line (Lacivita et al. 2014). They clearly show a quasilinear behavior of the elastic response of the grandite solid solution as a function of its chemical composition. For each composition x and each SIC, the EOS of a representative atomic configuration was computed, according to the procedure described in Section 3.3.2. As regards the two end-members, computed values are found to be in agreement with experimental data within 1% for grossular and 3% for andradite. An approximately linear dependence of volume on chemical composition is observed in the computed data (De La Pierre et al. 2013), that is, no significant excess volume of mixing is found at any intermediate composition, which confirms the ideal character of the grossular–andradite solid solution, as already suggested by several studies (Bird and Helgeson 1980; Ganguly 1976; Holdaway 1972; McAloon and Hofmeister 1995; Perchuk and Aranovich 1979) and in contrast to other investigations (Liou 1973; Meagher 1975). These findings contribute to the definition of a homogeneous frame according to which all solid solutions of the most abundant silicate garnets (pyralspite and grandite) exhibit linear elastic properties as a function of their chemical composition. Linearity of pyralspite is well known since long, whereas the presumed nonlinearity of the grandite system has been demonstrated to be an artifact due to scarcity and heterogeneity of the available experimental measurements. Optical (Lacivita et al. 2013a) and spectroscopical (De La Pierre et al. 2013) properties of the grossular–andradite, Ca3Fe2–2xAl2x(SiO4)3, solid solution series have recently been investigated at the quantum-mechanical level of theory. Refractive indices vary quite regularly between the andradite (1.860) and grossular (1.671) end-members. The high-frequency SiO stretching modes also show a rather linear behavior of both frequencies and IR intensities as a function of x. The frequencies of the low-energy bands show an almost linear dependence on composition as well; on the contrary, the behavior of the corresponding intensities is less linear. When considering different possible atomic configurations at fixed composition, some spectral features display a clear dependence upon shortrange Al/Fe cation ordering. Garnets are among the most promising candidates as possible hydrogen storage media for the Earth’s mantle, given their abundance and stability (Milman et al. 2000; Nobes et al. 2000a). A lot of attention is being devoted to the understanding of the incorporation of hydrogen into nominally anhydrous minerals (NAM) because of the remarkable effect it has on their technological and geophysical properties (Aines and Rossman 1984a,b; Beran and Putnis 1983; Freund and Oberheuser 1986; Griggs 1967; Rossman 1988; Wilkins and Sabine 1973). In particular, NAMs are of great geological interest in that they may potentially introduce large amount of “water” in the Earth mantle thus significantly modifying its elastic properties (Knittle et al. 1992; Mackwell et al. 1985; O’Neill et al. 1993). The hydrogarnet substitution SiO4 O4 H4 in grossular has received special attention in that it represents an effective mechanism for including hydrogen into silicate garnets (Lager et al. 1987, 2002, 2005; Lager and Von Dreele 1996; Nobes et al. 2000a,b; Olijnyk et al. 1991; Orlando et al. 2006; Pascale et al. 2004a; Pertlik 2003). Hydrogrossular can be represented by the general formula Ca3Al2(SiO4)3–x(OH)4x; when 0 < x < 1.5 it is called hibschite, and when 1.5 < x < 3 it is called katoite (Pertlik 2003). At low temperature there is complete solid solubility of the two end-members (grossular and silicon-free katoite). Hydrogarnets are known to be stable over the whole Earth’s mantle pressure range (Knittle et al. 1992); natural hydrogarnets equilibrated at 180 km depth have been characterized (O’Neill et al. 1993). In a couple of recent studies, first-principles simulations

98

Molecular Modeling of Geochemical Reactions

have been applied to the investigation of the behavior of katoite under pressure (Erba et al. 2015c) and of the explicit effect of chemical composition x on structural and energetic properties of hydrogrossular (Lacivita et al. 2015).

3.7 Future Challenges Most of the techniques that we are currently developing and implementing in the CRYSTAL program, relevant to the quantum-mechanical simulation of geochemical properties of minerals, are meant to include the effect of temperature on computed properties. Standard quantum-chemical methods based on the DFT represent, indeed, a powerful tool for the accurate determination of a variety of properties of materials, such as structural, electronic, vibrational, optical, elastic, magnetic, etc. (Dovesi et al. 2005; Dronskowski 2005; Grimme et al. 2010; Pisani 1996). The growing parallel computing resources are rapidly widening the range of applicability of such schemes which can now be routinely used for studying minerals of geochemical interest. If the inclusion of pressure can be modeled in a relatively simple way (with techniques like those presented in Section 3.3), that of temperature is a much more difficult task for the solid state. The most effective technique for taking into account temperature effects (including anharmonic terms) on computed properties of materials, in particular as regards thermal nuclear motion, would be ab initio molecular dynamics which, however, remains rather computationally demanding (Buda et al. 1990; Car and Parrinello 1985; Vila et al. 2012). Within the frame of standard quantum-chemical methods, temperature can be modeled by explicitly treating the lattice dynamics. A number of techniques have already been implemented in the CRYSTAL program for including the effect of harmonic thermal nuclear motion on one-electron properties such as electron charge and momentum densities (Erba et al. 2013c; Madsen et al. 2013; Pisani et al. 2012). When the lattice dynamics of a crystal is solved within the purely harmonic approximation, however, vibration frequencies are described as independent of interatomic distances, and the corresponding vibrational contribution to the internal energy of the crystal turns out to be independent of volume. It follows that, within such an assumption, a variety of physical properties would be wrongly described: thermal expansion would be null, elastic constants would not depend on temperature, constant-pressure and constant-volume specific heats would coincide with each other, thermal conductivity would be infinite as well as phonon lifetimes, etc. (Ashcroft and Mermin 1976; Baroni et al. 2010). An explicit account of anharmonic effects would require the calculation of phonon–phonon interaction coefficients with techniques such as vibrational configuration interaction (VCI), vibrational self-consistent field (VSCF), vibrational perturbation theory (VPT), transition-optimized shifted Hermite (TOSH), etc. (Jung and Gerber 1996; Lin et al. 2008; Neff and Rauhut 2009). An alternative and much simpler approach for correcting most of the aforementioned deficiencies of the harmonic approximation is the so-called quasiharmonic approximation (QHA) (Allen and De Wette 1969); according to which, the equilibrium volume, V(T), at any temperature T can be deduced by minimizing Helmholtz’s free energy F (V; T). This approach also allows for the natural combination of temperature and pressure effects. A fully automated implementation of the QHA has recently been developed in the CRYSTAL program, which relies on computing and fitting harmonic vibration frequencies at different volumes after having performed volume-constrained geometry optimizations (Erba 2014; Erba et al. 2015d). This strategy has already been successfully applied to the investigation of thermal structural and average elastic properties of diamond, periclase, lime, Al2O3 corundum, and Mg2SiO4 forsterite (Erba 2014; Erba et al. 2015a,b,d), and investigations on molecular crystals and alkali halides are currently in progress. The quasiharmonic approximation is also being generalized to the calculation of the elastic tensor of crystals at finite temperatures.

Quantum-Mechanical Modeling of Minerals

99

References Adamo C and Barone V 1999 Toward reliable density functional methods without adjustable parameters: the PBE0 model. J. Chem. Phys. 110, 6158. Aines RD and Rossman GR 1984a The hydrous component in garnets: pyralspites. Am. Mineral. 69(11–12), 1116–1126. Aines RD and Rossman GR 1984b Water in minerals? A peak in the infrared. J. Geophys. Res. Solid Earth 89(B6), 4059–4071. Albanese E, Civalleri B, Ferrabone M, Bonino F, Galli S, Maspero A, and Pettinari C 2012 Theoretical and experimental characterization of pyrazolato-based Ni(II) metal-organic frameworks. J. Mater. Chem. 22, 22592–22602. Alchagirov AB, Perdew JP, Boettger JC, Albers RC, and Fiolhais C 2001 Energy and pressure versus volume: equations of state motivated by the stabilized jellium model. Phys. Rev. B 63, 224115. Allen RE and De Wette FW 1969 Calculation of dynamical surface properties of noble-gas crystals. I. The quasiharmonic approximation. Phys. Rev. 179, 873–886. Anderson DL 1989 Theory of the Earth. Blackwell Scientific Publications, Boston. Anderson OL 1995 Equations of State of Solids for Geophysicists and Ceramic Science. Oxford University Press, New York. Anderson OL and Bass JD 1986 Transition region of the earth’s upper mantle. Nature 320, 321–328. Anderson OL and Isaak DG 1995 Elastic Constants of Mantle Minerals at High Temperature. American Geophysical Union, Washington, D.C. pp. 64–97. Angel RJ 2000 Equations of state In High Temperature and High Pressure Crystal Chemistry, Rev. Mineral. Geochem. 41 (ed. Hazen R and Downs R) Mineralogical Society of America, Washington, D.C. pp. 35–59. Ashcroft NW and Mermin ND 1976 Solid State Physics. Saunders College, Philadelphia, PA. Auld BA 1973 Acoustic Fields and Waves in Solids. Krieger Publishing Company, Malabar, FL. Authier A and Zarembowitch A 2006 Elastic properties In International Tables for Crystallography, Vol. D (ed. Authier A) John Wiley & Sons, Inc. p. 72. Babuška V, Fiala J, Kumazawa M, Ohno I, and Sumino Y 1978 Elastic properties of garnet solid-solution series. Phys. Earth Planet. In. 16(2), 157–176. Baima J, Erba A, Orlando R, Rérat M, and Dovesi R 2013 Beryllium oxide nanotubes and their connection to the flat monolayer. J. Phys. Chem. C 117, 12864–12872. Baroni S, Giannozzi P, and Isaev E 2010 Density-functional perturbation theory for quasi-harmonic calculations. Rev. Mineral. Geochem. 71(1), 39–57. Bass JD 1986 Elasticity of uvarovite and andradite garnets. J. Geophys. Res. 91(B7), 7505–7516. Bass JD 1989 Elasticity of grossular and spessartite garnets by Brillouin spectroscopy. J. Geophys. Res. 94, 7621–7628. Bass JD and Anderson OL 1984 Composition of the upper mantle: geophysical tests of two petrological models. Geophys. Res. Lett. 11, 229–232. Becke AD 1993 Density-functional thermochemistry. III. The role of exact exchange. J. Chem. Phys. 98, 5648. Beran A and Putnis A 1983 A model of the OH positions in olivine, derived from infrared-spectroscopic investigations. Phys. Chem. Miner. 9(2), 57–60. Birch F 1947 Finite elastic strain of cubic crystals. Phys. Rev. 71, 809–824. Birch F 1978 Finite strain isotherm and velocities for single-crystal and polycrystalline NaCl at high pressures and 300 K. J. Geophys. Res. 83(B3), 1257–1268. Bird DK and Helgeson HC 1980 Chemical interaction of aqueous solutions with epidote-feldspar mineral assemblages in geologic systems; 1, thermodynamic analysis of phase relations in the system CaO-FeOFe2O3-Al2O3-SiO2-H2O-CO2. Am. J. Sci. 280, 907–941. Broyden CG 1970 The convergence of a class of double-rank minimization algorithms 1. General considerations. IMA J. Appl. Math. 6(1), 76. Buda F, Car R, and Parrinello M 1990 Thermal expansion of c-Si via ab initio molecular dynamics. Phys. Rev. B 41, 1680. Bush IJ, Tomic S, Searle BG, Mallia G, Bailey CL, Montanari B, Bernasconi L, Carr JM, and Harrison NM 2011 Parallel implementation of the ab initio CRYSTAL program: electronic structure calculations for periodic systems. Proc. R. Soc. A: Math. Phys. Eng. Sci 467, 2112. Car R and Parrinello M 1985 Unified approach for molecular dynamics and density-functional theory. Phys. Rev. Lett. 55, 2471.

100

Molecular Modeling of Geochemical Reactions

Carteret C, De La Pierre M, Dossot M, Pascale F, Erba A, and Dovesi R 2013 The vibrational spectrum of CaCO3 aragonite: a combined experimental and quantum-mechanical investigation. J. Chem. Phys. 138, 014201. Chen G, Cooke JA, Jr., Gwanmesia GD, and Liebermann RC 1999 Elastic wave velocities of Mg3Al2Si3O12-pyrope garnet to 10 GPa. Am. Mineral. 84, 384–388. Civalleri B, D’Arco P, Orlando R, Saunders VR, and Dovesi R 2001 Hartree-Fock geometry optimization of periodic system with the CRYSTAL code. Chem. Phys. Lett. 348, 131. Civalleri B, Zicovich-Wilson C, Valenzano L, and Ugliengo P 2008 B3LYP augmented with an empirical dispersion term (B3LYP-D∗) as applied to molecular crystals. CrystEngComm 10, 405. Cohen RE, Gülseren O, and Hemley RJ 2000 Accuracy of equation-of-state formulations. Am. Mineral. 85, 338–344. Conrad PG, Zha CS, Mao HK, and Hemley RJ 1999 The high-pressure, single-crystal elasticity of pyrope, grossular, and andradite. Am. Mineral. 84, 374. Corá F, Alfredsson M, Mallia G, Middlemiss D, Mackrodt W, Dovesi R, and Orlando R 2004 The performance of hybrid density functionals in solid state chemistry In Principles and Applications of Density Functional Theory in Inorganic Chemistry II vol. 113 of Structure and Bonding Springer, Berlin Heidelberg. pp. 171–232. D’Arco Ph, Mustapha S, Ferrabone M, Noël Y, De La Pierre M, and Dovesi R 2013 Symmetry and random sampling of symmetry independent configurations for the simulation of disordered solids. J. Phys. Condens. Matter 25, 355401. De La Pierre M, Noel Y, Mustapha S, D’Arco P, and Dovesi R 2013 The infrared vibrational spectrum of andradite-grossular solid solutions: a quantum mechanical simulation. Am. Mineral. 98, 966–976. De La Pierre M, Orlando R, Maschio L, Doll K, Ugliengo P, and Dovesi R 2011 Performance of six functionals (LDA, PBE, PBESOL, B3LYP, PBE0 and WC1LYP) in the simulation of vibrational and dielectric properties of crystalline compounds. The case of forsterite Mg2SiO4. J. Comp. Chem. 32, 1775–1784. Delle Piane M, Corno M, and Ugliengo P 2013 Does dispersion dominate over H-bonds in drug-surface interactions? The case of silica-based materials as excipients and drug-delivery agents. J. Chem. Theory Comput. 9(5), 2404–2415. Diella V, Sani A, Levy D, and Pavese A 2004 High-pressure synchrotron x-ray diffraction study of spessartine and uvarovite: a comparison between different equation of state models. Am. Mineral. 89, 371–376. Doll K 2001 Implementation of analytical Hartree-Fock gradients for periodic systems. Comput. Phys. Commun. 137, 74. Doll K 2010 Analytical stress tensor and pressure calculations with the CRYSTAL code. Mol. Phys. 108(3–4), 223–227. Doll K, Dovesi R, and Orlando R 2004 Analytical Hartree-Fock gradients with respect to the cell parameter for systems periodic in three dimensions. Theor. Chem. Acc. 112(5–6), 394–402. Doll K, Dovesi R, and Orlando R 2006 Analytical Hartree-Fock gradients with respect to the cell parameter: systems periodic in one and two dimensions. Theor. Chem. Acc. 115(5), 354–360. Doll K, Harrison NM, and Saunders VR 2001 Analytical Hartree-Fock gradients for periodic systems. Int. J. Quantum Chem. 82, 1. Dovesi R, Civalleri B, Roetti C, Saunders VR, and Orlando R 2005 Ab initio quantum simulation in solid state chemistry. Rev. Comput. Chem. 21, 1. Dovesi R, De La Pierre M, Ferrari AM, Pascale F, Maschio L, and Zicovich-Wilson CM 2011 The IR vibrational properties of six members of the garnet family: a quantum mechanical ab initio study. Am. Mineral. 96, 1787–1798. Dovesi R, Orlando R, Erba A, Zicovich-Wilson CM, Civalleri B, Casassa S, Maschio L, Ferrabone M, De la Pierre M, D’Arco Ph, Noël Y, Causá M, Rérat M, and Kirtman B 2014a CRYSTAL14: a program for the ab initio investigation of crystalline solids. Int. J. Quantum Chem. 114, 1287–1317. Dovesi R, Saunders VR, Roetti C, Orlando R, Zicovich-Wilson CM, Pascale F, Doll K, Harrison NM, Civalleri B, Bush IJ, D’Arco Ph, Llunell M, Causá M, and Noël Y 2014b CRYSTAL14 User’s Manual Università di Torino, Torino. http://www.crystal.unito.it. Accessed December 9, 2015. Dovesi R, Valenzano L, Pascale F, Zicovich-Wilson CM, and Orlando R 2009 Ab initio quantum-mechanical simulation of the Raman spectrum of grossular. J. Raman Spectrosc. 40(4), 416–418. Dronskowski R 2005 Computational Chemistry of Solid State Materials. John Wiley & Sons, Inc, Weinheim. Duffy TS and Anderson DL 1989 Seismic velocities in mantle minerals and the mineralogy of the upper mantle. J. Geophys. Res. 94, 1895–1912.

Quantum-Mechanical Modeling of Minerals

101

Duffy TS and Wang Y 1998 Pressure-volume-temperature equations of state. Mineral. Soc. Am. Rev. Mineral. 37, 425–458. Erba A 2014 On combining temperature and pressure effects on structural properties of crystals with standard ab initio techniques. J. Chem. Phys. 141, 124115. Erba A and Dovesi R 2013 Photoelasticity of crystals from theoretical simulations. Phys. Rev. B 88, 045121. Erba A and Pisani C 2012 Evaluation of the electron momentum density of crystalline systems from ab initio linear combination of atomic orbitals calculations. J. Comput. Chem. 33, 822. Erba A, El-Kelany KE, Ferrero M, Baraille I, and Rérat M 2013a Piezoelectricity of SrTiO3: an ab initio description. Phys. Rev. B 88, 035102. Erba A, Ferrabone M, Baima J, Orlando R, Rérat M, and Dovesi R 2013b The vibration properties of the (n, 0) boron nitride nanotubes from ab initio quantum chemical simulations. J. Chem. Phys. 138, 054906. Erba A, Ferrabone M, Orlando R, and Dovesi R 2013c Accurate dynamical structure factors from ab initio lattice dynamics: the case of crystalline silicon. J. Comput. Chem. 34, 346. Erba A, Mahmoud A, Belmonte D, and Dovesi R 2014a High pressure elastic properties of minerals from ab initio simulations: the case of pyrope, grossular and andradite silicate garnets. J. Chem. Phys. 140, 124703. Erba A, Mahmoud A, Orlando R, and Dovesi R 2014b Elastic properties of six silicate garnet end-members from accurate ab initio simulations. Phys. Chem. Miner. 41, 151–160. Erba A, Mahmoud A, Orlando R, and Dovesi R 2014c Erratum to: elastic properties of six silicate garnet endmembers from accurate ab initio simulations. Phys. Chem. Miner. 41, 161–162. Erba A, Maul J, De La Pierre M, and Dovesi R 2015a Structural and elastic anisotropy of crystals at high pressures and temperatures from quantum mechanical methods: the case of Mg2SiO4 forsterite. J. Chem. Phys. 142, 204502. Erba A, Maul J, Demichelis R, and Dovesi R 2015b Assessing thermochemical properties of materials through ab initio quantum-mechanical methods: the case of α-Al2O3. Phys. Chem. Chem. Phys. 17(17), 11670–11677. Erba A, Montserrat Navarrete López A, Zicovich-Wilson C, and Dovesi R 2015c Katoite under pressure: an ab initio investigation of its structural, elastic and vibrational properties sheds light on the phase transition. Phys. Chem. Chem. Phys. 17, 2660–2669. Erba A, Shahrokhi M, Moradian R, and Dovesi R 2015d On how differently the quasi-harmonic approximation works for two isostructural crystals: thermal properties of MgO and CaO. J. Chem. Phys. 142, 044114. Fan D, Wei S, Liu J, Li Y, and Xie H 2011 High pressure X-ray diffraction study of a grossular–andradite solid solution and the bulk modulus variation along this solid solution. Chin. Phys. Lett. 28(7), 076101. Ferrari AM, Valenzano L, Meyer A, Orlando R, and Dovesi R 2009 Quantum-mechanical ab initio simulation of the Raman and IR spectra of Fe3Al2Si3O12 almandine. J. Phys. Chem. A 113(42), 11289–11294. Ferrero M, Rérat M, Kirtman B, and Dovesi R 2008a Calculation of first and second static hyperpolarizabilities of one- to three-dimensional periodic compounds. Implementation in the CRYSTAL code. J. Chem. Phys. 129, 244110. Ferrero M, Rérat M, Orlando R, and Dovesi R 2008b Coupled perturbed Hartree-Fock for periodic systems: the role of symmetry and related computational aspects. J. Chem. Phys. 128, 014110. Ferrero M, Rérat M, Orlando R, and Dovesi R 2008c The calculation of static polarizabilities of periodic compounds. The implementation in the CRYSTAL code for 1D, 2D and 3D systems. J. Comput. Chem. 29, 1450. Ferrero M, Rérat M, Orlando R, Dovesi R, and Bush I 2008d Coupled Perturbed Kohn-Sham calculation of static polarizabilities of periodic compound. J. Phys. Chem. Solids 117, 12016. Fletcher R 1970 A new approach to variable metric algorithms. Comput. J. 13, 317. Freund F and Oberheuser G 1986 Water dissolved in olivine: a single-crystal infrared study. J. Geophys. Res. Solid Earth 91(B1), 745–761. Ganguly J 1976 The energetics of natural garnet solid solution. Contrib. Mineral. Petrol. 55(1), 81–90. Goldfarb D 1970 A family of variable-metric methods derived by variational means. Math. Comput. 24, 23. Gréaux S and Yamada A 2012 PVT equation of state of Mn3Al2Si3O12 spessartine garnet. Phys. Chem. Miner. 41, 141–149. Griggs D 1967 Hydrolytic weakening of quartz and other silicates. Geophys. J. Roy. Astron. Soc. 14(1–4), 19–31. Grimme S 2006 Semiempirical GGA-type density functional constructed with a long-range dispersion correction. J. Comput. Chem. 27(15), 1787–1799. Grimme S 2011 Density functional theory with London dispersion corrections. Wiley Interdiscip. Rev. Comput. Mol. Sci. 1(2), 211–228. Grimme S, Antony J, Ehrlich S, and Krieg H 2010 A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu. J. Chem. Phys. 132, 154104.

102

Molecular Modeling of Geochemical Reactions

Gwanmesia GD, Wang L, Heady A, and Liebermann RC 2014 Elasticity and sound velocities of polycrystalline grossular garnet (Ca3Al2Si3O12) at simultaneous high pressures and high temperatures. Phys. Earth Planet. In. 228, 80–87. Gwanmesia GD, Zhang J, Darling K, Kung J, Li B, Wang L, Neuville D, and Liebermann RC 2006 Elasticity of polycrystalline pyrope (Mg3Al2Si3O12) to 9 GPa and 1000 C. Phys. Earth Planet. In. 155, 179–190. Halleck P 1973 The Compression and Compressibility of Grossular Garnet: A Comparison of X-Ray and Ultrasonic Methods. Ph.D. thesis, University of Chicago, Department of the Geophysical Sciences, Chicago. Hama J and Suito K 1996 The search for a universal equation of state correct up to very high pressures. J. Phys. Condens. Matter 8(1), 67. Haselton HT and Westrum EF 1980 Low-temperature heat capacities of synthetic pyrope, grossular, and pyrope60 grossular40. Geochim. Cosmochim. Acta 44, 701–709. Hensen BJ 1976 The stability of pyrope-grossular garnet with excess silica. Contrib. Mineral. Petrol. 55, 279–292. Heyd J and Scuseria GE 2004 Efficient hybrid density functional calculations in solids: assessment of the HeydScuseria-Ernzerhof screened Coulomb hybrid functional. J. Chem. Phys. 121(3), 1187–1192. Hill R 1963 Elastic properties of reinforced solids: some theoretical principles. J. Mech. Phys. Solids 11, 357–372. Hofmeister AM and Chopelas A 1991 Thermodynamic properties of pyrope and grossular from vibrational spectroscopy. Am. Mineral. 76, 880–891. Hohenberg P and Kohn W 1964 Inhomogeneous electron gas. Phys. Rev. 136, B864–B871. Holdaway M 1972 Thermal stability of Al-Fe epidote as a function of fO2 and Fe content. Contrib. Mineral. Petrol. 37(4), 307–340. Holzapfel WB 1996 Physics of solids under strong compression. Rep. Prog. Phys. 59(1), 29. Isaak DG and Graham EK 1976 The elastic properties of an almandine-spessartine garnet and elasticity in the garnet solid solution series. J. Geophys. Res. 81, 2483–2489. Jiang F, Speziale S, and Duffy TS 2004a Single-crystal elasticity of grossular- and almandine-rich garnets to 11 GPa by Brillouin scattering. J. Geophys. Res. 109, B10210. Jiang F, Speziale S, Shieh SR, and Duffy TS 2004b Single-crystal elasticity of andradite garnet to 11 GPa. J. Phys. Condens. Matter 16, S1041. Jung JO and Gerber RB 1996 Vibrational wave functions and energy levels of large anharmonic clusters: a vibrational SCF study of (Ar)13. J. Chem. Phys. 105(24), 10682–10690. Karki BB, Ackland GJ, and Crain J 1997 Elastic instabilities in crystals from ab initio stress—strain relations. J. Phys. Condens. Matter 9(41), 8579. Karki BB, Stixrude L, and Wentzcovitch RM 2001 High-pressure elastic properties of major materials of earth’s mantle from first principles. Rev. Geophys. 39, 507–534. Knittle E 1995 Static compression measurements of equations of state In A Handbook of Physical Constants: Mineral Physics and Crystallography (ed. Ahrens TJ) vol. 2, AGU, Washington D.C. pp. 98–142. Knittle E, Hathorne A, Davis M, and Williams Q 1992 A spectroscopic study of the high-pressure behavior of the O4H4 substitution in garnet In High-Pressure Research: Application to Earth and Planetary Sciences (ed. Syono Y and Manghnani MH) American Geophysical Union, Washington D.C. pp. 297–304. Kohn W and Sham LJ 1965 Self-consistent equations including exchange and correlation effects. Phys. Rev. 140, A1133. Kono Y, Gréaux S, Higo Y, Ohfuji H, and Irifune T 2010 Pressure and temperature dependences of elastic properties of grossular garnet up to 17 GPa and 1650 K. J. Earth Sci. 21, 782–791. Kresse G and Furthmüller J 1996a Efficiency of ab initio total energy calculations for metals and semiconductors using a plane-wave basis set. Comput. Mater. Sci. 6, 15. Kresse G and Furthmüller J 1996b Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys. Rev. B 54, 11169. Lacivita V, D’Arco Ph, Orlando R, Dovesi R, and Meyer A 2013a Anomalous birefringence in andraditegrossular solid solutions: a quantum-mechanical approach. Phys. Chem. Miner. 40, 781–788. Lacivita V, Erba A, Dovesi R, and D’Arco P 2014 Elasticity of grossular-andradite solid solution: an ab initio investigation. Phys. Chem. Chem. Phys. 16, 15331–15338. Lacivita V, Erba A, Noël Y, Orlando R, D’Arco Ph, and Dovesi R 2013b Zinc oxide nanotubes: an ab initio investigation of their structural, vibrational, elastic, and dielectric properties. J. Chem. Phys. 138, 214706. Lacivita V, Mahmoud A, Erba A, D’Arco P, and Mustapha S 2015 Hydrogrossular, Ca3Al2(SiO4)3–x(H4O4)x: an ab initio investigation of its structural and energetic properties. Am. Mineral. 100(11–12), 2637–2649.

Quantum-Mechanical Modeling of Minerals

103

Lager G, Armbruster T, and Faber J 1987 Neutron and X-ray diffraction study of hydrogarnet Ca3Al2(O4H4)3. Am. Mineral. 72(7–8), 756–765. Lager GA and Von Dreele RB 1996 Neutron powder diffraction study of hydrogarnet to 9.0 GPa. Am. Mineral. 81(9–10), 1097–1104. Lager GA, Downs RT, Origlieri M, and Garoutte R 2002 High-pressure single-crystal X-ray diffraction study of katoite hydrogarnet: evidence for a phase transition from Ia3d to I 43d symmetry at 5 GPa. Am. Mineral. 87(5–6), 642–647. Lager GA, Marshall WG, Liu Z, and Downs RT 2005 Re-examination of the hydrogarnet structure at high pressure using neutron powder diffraction and infrared spectroscopy. Am. Mineral. 90(4), 639–644. Leger JM, Redon AM, and Chateau C 1990 Compressions of synthetic pyrope, spessartine and uvarovite garnets up to 25 GPa. Phys. Chem. Miner. 17, 161–167. Lin C, Gilbert AT, and Gill PM 2008 Calculating molecular vibrational spectra beyond the harmonic approximation. Theor. Chem. Acc. 120(1–3), 23–35. Liou JG 1973 Synthesis and stability relations of epidote, Ca2Al2FeSi3O12(OH). J. Petrol. 14, 381–413. Mackwell S, Kohlstedt D, and Paterson M 1985 The role of water in the deformation of olivine single crystals. J. Geophys. Res. Solid Earth (1978–2012) 90(B13), 11319–11333. Madsen AO, Civalleri B, Ferrabone M, Pascale F, and Erba A 2013 Anisotropic displacement parameters for molecular crystals from periodic Hartree-Fock and density functional theory calculations. Acta Crystallogr. Sec. A 69, 309. Mahmoud A, Erba A, Doll K, and Dovesi R 2014a Pressure effect on elastic anisotropy of crystals from ab initio simulations: the case of silicate garnets. J. Chem. Phys. 140, 234703. Mahmoud A, Erba A, El-Kelany KE, Rérat M, and Orlando R 2014b Low-temperature phase of BaTiO3: piezoelectric, dielectric, elastic, and photo elastic properties from ab initio simulations. Phys. Rev. B 89, 045103. Mainprice D, Barruol G, and Ben Ismail W 2013 The Seismic Anisotropy of the Earth’s Mantle: From Single Crystal to Polycrystal. American Geophysical Union, Washington, D.C. Marmier A, Lethbridge ZA, Walton RI, Smith CW, Parker SC, and Evans KE 2010 Elam: a computer program for the analysis and representation of anisotropic elastic properties. Comput. Phys. Commun. 181(12), 2102–2115. Marques M and Gross E 2004 Time-dependent density functional theory. Annu. Rev. Phys. Chem. 55(1), 427–455. Maschio L, Demichelis R, Orlando R, De La Pierre M, Mahmoud A, and Dovesi R 2014 The Raman spectrum of grossular garnet: a quantum mechanical simulation of wavenumbers and intensities. J. Raman Spectrosc. 45, 710–715. Maschio L, Ferrabone M, Meyer A, Garza J, and Dovesi R 2011 The infrared spectrum of spessartine: an ab initio all electron simulation with five different functionals (LDA, PBE, PBEsol, B3LYP and PBE0). Chem. Phys. Lett. 501, 612–618. Maschio L, Kirtman B, Salustro S, Zicovich-Wilson CM, Orlando R, and Dovesi R 2013 Raman spectrum of pyrope garnet: a quantum mechanical simulation of frequencies, intensities, and isotope shifts. J. Phys. Chem. A 117(45), 11464–11471. McAloon BP and Hofmeister AM 1995 Single-crystal IR spectroscopy of grossular-andradite garnets. Am. Mineral. 80, 1145–1156. Meagher EP 1975 Crystal-structures of pyrope and grossularite at elevated-temperatures. Am. Mineral. 60, 218–228. Meyer A, Pascale F, Zicovich-Wilson CM, and Dovesi R 2010 Magnetic interactions and electronic structure of uvarovite and andradite garnets. An ab initio all-electron simulation with the CRYSTAL06 program. Int. J. Quantum Chem. 110, 338. Milman V, Winkler B, Nobes R, Akhmatskaya E, Pickard C, and White J 2000 Garnets: structure, compressibility, dynamics, and disorder. JOM 52(7), 22–25. Murnaghan FD 1944 The compressibility of media under extreme pressures. Proc. Natl. Acad. Sci. U. S. A. 30, 244. Musgrave MJP 1970 Crystal Acoustics. Holden-Day, San Francisco, CA. Mustapha S, D’Arco Ph, De La Pierre M, Noël Y, Ferrabone M, and Dovesi R 2013 On the use of symmetry in configurational analysis for the simulation of disordered solids. J. Phys. Condens. Matter 25, 105401. Neff M and Rauhut G 2009 Toward large scale vibrational configuration interaction calculations. J. Chem. Phys. 131(12), 124129. Nobes R, Akhmatskaya E, Milman V, White J, Winkler B, and Pickard C 2000a An ab initio study of hydrogarnets. Am. Mineral. 85(11–12), 1706–1715.

104

Molecular Modeling of Geochemical Reactions

Nobes R, Akhmatskaya E, Milman V, Winkler B, and Pickard C 2000b Structure and properties of aluminosilicate garnets and katoite: an ab initio study. Comput. Mat. Sci. 17(2), 141–145. Novak GA and Gibbs GV 1971 The crystal chemistry of the silicate garnets. Am. Mineral. 56, 791–825. Nye JF 1957 Physical Properties of Crystals. Oxford University Press, Oxford. Olijnyk H, Paris E, Geiger C, and Lager G 1991 Compressional study of katoite [Ca3Al2 (O4H4)3] and grossular garnet. J. Geophys. Res. Solid Earth 96(B9), 14313–14318. O’Neill B, Bass JD, and Rossman GR 1993 Elastic properties of hydrogrossular garnet and implications for water in the upper mantle. J. Geophys. Res. Solid Earth 98(B11), 20031–20037. O’Neill B, Bass JD, Smyth JR, and Vaughan MT 1989 Elasticity of a grossular-pyrope-almandine garnet. J. Geophys. Res. 94, 17819–17824. Orlando R, De La Pierre M, Zicovich-Wilson CM, Erba A, and Dovesi R 2014 On the full exploitation of symmetry in periodic (as well as molecular) self-consistent-field ab initio calculations. J. Chem. Phys. 141, 104108. Orlando R, Delle Piane M, Bush IJ, Ugliengo P, Ferrabone M, and Dovesi R 2012 A new massively parallel version of crystal for large systems on high performance computing architectures. J. Comput. Chem. 33, 2276–2284. Orlando R, Ferrero M, Rérat M, Kirtman B, and Dovesi R 2009 Calculation of the static electronic second hyperpolarizability or χ(3) tensor of three-dimensional periodic compounds with a local basis set. J. Chem. Phys. 131, 184105. Orlando R, Lacivita V, Bast R, and Ruud K 2010 Calculation of the first static hyperpolarizability tensor of three-dimensional periodic compounds with a local basis set: a comparison of LDA, PBE, PBE0, B3LYP, and HF results. J. Chem. Phys. 132, 244106. Orlando R, Torres F, Pascale F, Ugliengo P, Zicovich-Wilson C, and Dovesi R 2006 Vibrational spectrum of katoite Ca3Al2[(OH)4]3: a periodic ab initio study. J. Phys. Chem. B 110(2), 692–701. Ottonello G, Civalleri B, Ganguly J, Perger WF, Belmonte D, and Vetuschi Zuccolini M 2010 Thermochemical and thermo-physical properties of the high-pressure phase anhydrous B (Mg14Si5O24): an ab-initio all-electron investigation. Am. Mineral. 95, 563–573. Parlinski K, Li ZQ, and Kawazoe Y 1997 First-principles determination of the soft mode in cubic ZrO2. Phys. Rev. Lett. 78, 4063–4066. Pascale F, Catti M, Damin A, Orlando R, Saunders VR, and Dovesi R 2005a Vibration frequencies of Ca3Fe2Si3O12 andradite: an ab initio study with the CRYSTAL code. J. Phys. Chem. B 109(39), 18522–18527. Pascale F, Ugliengo P, Civalleri B, Orlando R, D’Arco P, and Dovesi R 2004a The katoite hydrogarnet Si-free Ca3Al2([OH]4)3: a periodic Hartree-Fock and B3LYP study. J. Chem. Phys. 121(2), 1005–1013. Pascale F, Zicovich-Wilson CM, Gejo FL, Civalleri B, Orlando R, and Dovesi R 2004b The calculation of the vibrational frequencies of the crystalline compounds and its implementation in the crystal code. J. Comp. Chem. 25, 888–897. Pascale F, Zicovich-Wilson CM, Orlando R, Roetti C, Ugliengo P, and Dovesi R 2005b Vibration frequencies of Mg3Al2Si3O12 pyrope. An ab initio study with the CRYSTAL code. J. Phys. Chem. B 109(13), 6146–6152. Perchuk L and Aranovich L 1979 Thermodynamics of minerals of variable composition: andradite-grossularite and pistacite-clinozoisite solid solutions. Phys. Chem. Miner. 5(1), 1–14. Perger WF, Criswell J, Civalleri B, and Dovesi R 2009 Ab-initio calculation of elastic constants of crystalline systems with the CRYSTAL code. Comput. Phys. Commun. 180, 1753–1759. Pertlik F 2003 Bibliography of hibschite, a hydrogarnet of grossular type. GeoLines 15, 113–119. Pisani C 1996 Quantum-Mechanical Ab-Initio Calculation of the Properties of Crystalline Materials vol. 67 of Lecture Notes in Chemistry Series. Springer Verlag, Berlin. Pisani C, Dovesi R, and Roetti C 1988 Hartree-Fock Ab Initio Treatment of Crystalline solids vol. 48 of Lecture Notes in Chemistry Series. Springer Verlag, Berlin. Pisani C, Erba A, Ferrabone M, and Dovesi R 2012 Nuclear motion effects on the density matrix of crystals: an ab initio Monte Carlo harmonic approach. J. Chem. Phys. 137, 044114. Poirier JP and Tarantola A 1998 A logarithmic equation of state. Phys. Earth Planet. In. 109, 1–8. Rickwood PC, Mathias M, and Siebert JC 1968 A study of garnets from eclogite and peridotite xenoliths found in kimberlite. Contrib. Mineral. Petrol. 19, 271–301. Rimola A, Zicovich-Wilson CM, Dovesi R, and Ugliengo P 2010 Search and characterization of transition state structures in crystalline systems using valence coordinates. J. Chem. Theory Comput. 6(4), 1341–1350. Ringwood AE 1975 Composition and Petrology of the Earth’s Mantle. McGraw-Hill, New York. Rossman GR 1988 Vibrational spectroscopy of hydrous components. Rev. Mineral. 18, 193–206. Shanno DF 1970 Conditioning of quasi-Newton methods for function minimization. Math. Comput. 24, 647.

Quantum-Mechanical Modeling of Minerals

105

Sinogeikin SV and Bass JD 2000 Single-crystal elasticity of pyrope and MgO to 20 GPa by Brillouin scattering in the diamond cell. Phys. Earth Planet. Inter. 120, 43–62. Soler JM, Artacho E, Gale JD, García A, Junquera J, Ordejón P, and Sánchez-Portal D 2002 The siesta method for ab initio order-n materials simulation. J. Phys. Condens. Matter 14, 2745–2779. Souza I and Martins J 1997 Metric tensor as the dynamical variable for variable-cell-shape molecular dynamics. Phys. Rev. B 55, 8733–8742. Stacey F, Brennan B, and Irvine R 1981 Finite strain theories and comparisons with seismological data. Geophys. Surv. 4(3), 189–232. Tequi C, Robie RA, Hemingway BS, Neuville D, and Richet P 1991 Melting and thermodynamic properties of pyrope (Mg3Al2Si3O12). Geochim. Cosmochim. Acta 55, 10051010. Tinkham M 1964 Group Theory and Quantum Mechanics. McGraw-Hill, New York. Togo A, Oba F, and Tanaka I 2008 First-principles calculations of the ferroelastic transition between rutile-type and CaCl2-type SiO2 at high pressures. Phys. Rev. B 78, 134106. Ugliengo P, Sodupe M, Musso F, Bush IJ, Orlando R, and Dovesi R 2008 Realistic models of hydroxylated amorphous silica surfaces and MCM-41 mesoporous material simulated by large-scale periodic B3LYP calculations. Adv. Mater. 20(23), 4579–4583. Valenzano L, Meyer A, Demichelis R, Civalleri B, and Dovesi R 2009 Quantum-mechanical ab initio simulation of the Raman and IR spectra of Mn3Al2Si3O12 spessartine. Phys. Chem. Miner. 36(7), 415–420. Valenzano L, Pascale F, Ferrero M, and Dovesi R 2010 Ab initio quantum-mechanical prediction of the IR and Raman spectra of Ca3Cr2Si3O12 uvarovite garnet. Int. J. Quantum Chem. 110(2), 416–421. Vila FD, Lindahl VE, and Rehr JJ 2012 X-ray absorption Debye-Waller factors from ab initio molecular dynamics. Phys. Rev. B 85, 024303. Vinet P, Ferrante J, Smith JR, and Rose JH 1986 A universal equation of state for solids. J. Phys. C. 19(20), 467. Wallace DC 1965 Lattice dynamics and elasticity of stressed crystals. Rev. Mod. Phys. 37, 57–67. Wallace DC 1972 Thermodynamics of Crystals. John Wiley & Sons, Inc, New York. Wang J, Li J, Yip S, Phillpot S, and Wolf D 1995 Mechanical instabilities of homogeneous crystals. Phys. Rev. B 52, 12627–12635. Wilkins R and Sabine W 1973 Water-content of some nominally anhydrous silicates. Am. Mineral. 58(5–6), 508–516. Yeganeh-Haeri A, Weidner DJ, and Ito E 1990 Elastic properties of the pyrope-majorite solid solution series. Geophys. Res. Lett. 17, 2453–2456. Zhang L, Ahsbahs H, and Kutoglu A 1998 Hydrostatic compression and crystal structure of pyrope to 33 GPa. Phys. Chem. Miner. 25(4), 301–307. Zhang L, Ahsbahs H, Kutoglu A, and Geiger CA 1999 Single-crystal hydrostatic compression of synthetic pyrope, almandine, spessartine, grossular and andradite garnets at high pressures. Phys. Chem. Miner. 27(1), 52–58. Zicovich-Wilson C and Dovesi R 1998 On the use of symmetry-adapted crystalline orbitals in SCF-LCAO periodic calculations. I. The construction of the symmetrized orbitals. Int. J. Quantum Chem. 67, 299–309. Zicovich-Wilson CM, Pascale F, Roetti C, Saunders VR, Orlando R, and Dovesi R 2004 The calculation of the vibration frequencies of α-quartz: the effect of Hamiltonian and basis set. J. Comput. Chem. 25, 1873–1881. Zicovich-Wilson CM, Torres FJ, Pascale F, Valenzano L, Orlando R, and Dovesi R 2008 Ab initio simulation of the IR spectra of pyrope, grossular, and andradite. J. Comput. Chem. 29(13), 2268–2278.

4 First Principles Estimation of Geochemically Important Transition Metal Oxide Properties Structure and Dynamics of the Bulk, Surface, and Mineral/Aqueous Fluid Interface

Ying Chen,1 Eric Bylaska,2 and John Weare1 1

2

Chemistry and Biochemistry Department, University of California, San Diego, La Jolla, CA, USA Environmental Molecular Sciences Laboratory, Pacific Northwest National Laboratory, Richland, WA, USA

4.1

Introduction

Reactions in the mineral surface/reservoir fluid interface control many geochemical processes such as the dissolution and growth of minerals (Yanina and Rosso 2008), heterogeneous oxidation/reduction (Hochella 1990, Brown 2001, Hochella et al. 2008, Navrotsky et al. 2008), and inorganic respiration (Newman 2010). Key minerals involved in these processes are the transition metal oxides and oxyhydroxides (e.g., hematite, Fe2O3, and goethite, FeOOH) (Brown et al. 1999, Brown 2001, Hochella et al. 2008, Navrotsky et al. 2008). To interpret and predict these processes, it is necessary to have a high level of understanding of the interactions between the formations containing these minerals and their reservoir fluids. However, these are complicated chemical events occurring under a wide range of T, P, and X conditions, and the interpretation is complicated by the highly heterogeneous nature of natural environments (Hochella 1990, Hochella et al. 2008, Navrotsky et al. 2008) and the electronic and structural complexity of the oxide materials involved (Cox 1992, Kotliar and Vollhardt 2004, Navrotsky et al. 2008). In addition, also because of the complexity of the minerals involved and the heterogeneous nature of natural systems, the direct observation

Molecular Modeling of Geochemical Reactions: An Introduction, First Edition. Edited by James D. Kubicki. © 2016 John Wiley & Sons, Ltd. Published 2016 by John Wiley & Sons, Ltd.

108

Molecular Modeling of Geochemical Reactions

of these reactions at the atomic level is experimentally extremely difficult. Theoretical simulations will provide important support for analysis of the geochemistry of the mineral surface/fluid region as well as provide essential tools to extrapolate laboratory measurements to the field environment. The atomic-level interpretation of the chemical events occurring in the surface and fluid regions has recently been supported by the rapid development of new high-resolution spectroscopic measurements of highly ordered systems (e.g., clean well-ordered surfaces with “best possible” ordered layers) using synchrotron light sources (Renaud 1998, Fenter and Sturchio 2004, Park et al. 2005, 2010, Lo et al. 2007, Tanwar et al. 2007, Catalano et al. 2010a, 2010b, Fenter et al. 2010a, 2010b, Ghose et al. 2010, Fulton et al. 2012, Huang et al. 2014, Massey et al. 2014, Stubbs et al. 2015, Yuan et al. 2015) and high-resolution NMR (Anovitz et al. 2009, 2010). As well as providing new insights into the surface structure of these complex materials, these observations provide a rich and challenging database for quantification of theoretical predictions. The objective of this chapter is to describe our efforts to use first principles dynamical simulation methods (based on direct solution to the electronic structure Schrödinger equations (Car and Parrinello 1985, Remler and Madden 1990, Marx and Hutter 2012)) to analyze and interpret these data (atom and electronic structure). With further development, these methods will provide a means to extrapolate observations from highly ordered materials to much more heterogeneous natural environments. The target of our recent calculations is the structure of the surface interface and the development of methods to simulate this dynamic region. Because of the weak interactions of the species (H2O molecules, solutes, etc.) in the interface at finite temperature, atoms are in constant motion. The measurements that are available (e.g., X-ray diffraction (Brown and Sturchio 2002, Fenter 2002)) are of the average (equilibrium) structure of these fluctuating species. In order to interpret the data, a dynamical theory of the atomic motion in the interface region must be used. The need to seamlessly model the transition metal bulk, solid surface, fluid interface, and fluid bulk poses a difficult problem for simulation because the bonding character changes from ionic and covalent bonding in the solid material to polarized closed-shell hydrogen bonding (H-bonding) in the interface region and to closed-shell/H-bonding water–water and water–solute interaction in the bulk fluid region. The simulation methods discussed in this article efficiently carry out dynamical simulations with forces calculated directly from first principles. There have been several previous efforts to calculate the properties of transition metal oxide surface properties for geochemical applications using first principles methods (Becker et al. 1996, Rosso and Rustad 2001, Kubicki et al. 2008), but most of this work has focused on the static structure of the mineral–water interface. These methods avoid the problem of defining empirical force fields for conventional molecular dynamics (CMD) simulations by calculating the interactions between atoms on the fly directly using various levels of approximation to the electronic Schrödinger equation (herein called ab initio molecular dynamics (AIMD) (Car and Parrinello 1985, Remler and Madden 1990, Marx and Hutter 2012)) in the three regions (mineral, interface, and solution bulk). There have been a number of efforts to model the dynamical behavior of these systems using CMD (Rustad et al. 1996, 2003, Shroll and Straatsma 2003, Kerisit 2011, Kerisit et al. 2012). Although these calculations have led to important insights, detailed interpretation of data is limited by the difficulty of defining empirical force fields that reflect changes in bonding and electronic structure in the three regions (Kerisit et al. 2012). On the other hand, although AIMD methods provide a more reliable description of the changing interactions in the system, the calculation of the forces for these complex systems is much more computationally demanding. Therefore, the limitations as to the equilibration of the system and the number of atomic species that can be considered are more restrictive for AIMD methods than CMD methods. The results presented here are close to the limit of what can be practically calculated with presently available computational platforms (Bylaska et al. 2011b).

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

109

The transition metal oxide and oxyhydroxide minerals considered in this chapter and their interface properties are important to other technological applications such as solar hydrogen production, heterogeneous catalysis, and magnetic materials applications (Renaud 1998, Valdes et al. 2012). First principles calculations for these systems have been reported (Huda et al. 2010, Pozun and Henkelman 2011, Valdes et al. 2012). The calculations reported here use the NWChem software, which can be downloaded from http://www.nwchem-sw.org/index.php/Download. These application packages have been designed for implementation on highly parallel computers (up to 100 K cores; http://www.nwchem-sw.org/index.php/Benchmarks) and with special emphasis on providing tools for interpreting chemical environments.

4.2

Overview of the Theoretical Methods and Approximations Needed to Perform AIMD Calculations

In order to model interactions of transition metal oxide minerals with solution interfaces, it is necessary to have an accurate and realistic representation of the bulk/surface (10–15 Å), the interface water–solute, and the bulk solution regions. For the transition metal oxides, the computational complexity is greatly increased by the highly correlated electronic behavior of these materials and the complexity of the unit cells. The possibility of electron localization in the unit cell due to local character of the 3d atomic orbitals in these materials leads to local spin ordering within the unit cell further complicating the electronic structure calculation. These effects are extremely difficult to capture theoretically and remain a current topic of theoretical research in condensed matter physics (Kotliar and Vollhardt 2004). As will be discussed in more detail in the following, there are significant problems with the application of orbital density functional theory (DFT) (Hohenberg and Kohn 1964, Kohn and Sham 1965, Parr and Yang 1995) to these problems. However, a minimum requirement to achieve spin ordering is to enforce single electron orbital occupation (e.g., the number of orbitals equals the number of electrons, roughly 1000 valence electrons in the calculations reported in the following). For surfaces interacting with a loosely bound fluid layer, the observed structure (e.g., CTR measurements (Fenter and Sturchio 2004)) represents the equilibrium average of the positions of the solution species (H2O and dissolved solutes). The strength of interaction and the nature of the bonding of the species (e.g., highly polarized and H-bonding for H2O) vary with the position of the molecule in the mineral/fluid interface. This effect is incorporated in the AIMD simulation, but the number of particles required to capture changes in bonding due to local and long-range interactions greatly affects the time required for simulation. Typically, dynamical simulation time scales that can be practically reached are up to few hundred picoseconds (Atta-Fynn et al. 2013), but most simulations are significantly shorter. The calculation of the electronic structure of a highly correlated system is a problem of current interest in condensed matter physics (Kotliar and Vollhardt 2004). Fortunately, there is an approximate approach, the DFT of Hohenberg, Kohn, and Sham (Hohenberg and Kohn 1964, Kohn and Sham 1965, Parr and Yang 1995) that provides estimates of many properties at a practical computational cost. Though not expected to provide more than qualitative accuracy for spin-dependent properties, essentially all AIMD methods implement this approach. To clarify the following discussion, it will be useful to briefly outline some of the aspects of this theory that affect the accuracy possible in dynamical simulations and its application to highly correlated systems. (For more detail on DFT, refer to Chapter 1.)

110

Molecular Modeling of Geochemical Reactions

Hohenberg and Kohn demonstrated that the total electronic energy of a many-electron system can be written as an orbital-based functional of the electron density (Hohenberg and Kohn 1964, Kohn and Sham 1965, Parr and Yang 1995). That is, occupied

E ρ, RI

= i=1

nion 1 2 KS 1 + Vext r,RI ρ r dr + ∇ ψ KS − ψ i i 2 2 I =1

ρrρr drdr + Exc ρ r −r 41

In Equation 4.1 E[ρ, {RI}] is the total electronic energy and is a function of the electron density, occupied

ψ KS r i

ρr =

2

42

i=1

and the atomic positions RI. Exc[ρ] is the exchange–correlation energy defined in Hohenberg and Kohn (Hohenberg and Kohn 1964, Parr and Yang 1995). The existence of the functional Exc[ρ] is demonstrated by the Hohenberg Kohn theorem, but the form as a function of density is still a topic of much research (Burke 2012, Becke 2014). In Equation 4.1 ψ KS r are the Kohn–Sham orbital wave functions (Kohn and Sham 1965) found from i the constrained variation of the total energy as (Parr and Yang 1995) nion 1 − ∇2 + Vext r, RI + 2 I =1

ρr dr + vxc ρ r r −r

ψ KS r = εi ψ KS r i i

43

The orbital Kohn–Sham functions must be constrained to be orthonormal: ψ KS ψ KS = δi , j i j

44

vxc(ρ(r)) is the exchange–correlation function defied by functional variation of Equation 4.1. Our objective is to calculate the dynamics of the system from the solution to the Kohn–Sham equations (Eq. 4.3). Given the forces FI RI = −

∂E ∂RI

45 δE δρ r

=0

we can write Newton’s equations of motions as MI R I = F I

46

Time integration of Equation 4.6 provides the first principles dynamics of the system of particles (AIMD) by the Kohn–Sham equations (Eq. 4.3). Car and Parrinello (CP) (Car and Parrinello 1985) have introduced a modification of Equations 4.3–4.6 that can lead to a more efficient algorithm. These equations are used in most of the dynamical calculations reported in the following. In this algorithm the dynamical variables are expanded to include the Kohn–Sham wave functions and fictitious dynamical equations. Lagrange multipliers, Λij, maintain orthogonality between the Kohn–Sham orbitals. The dynamical equations for this approach are modified to be

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties nion 1 μψ KS r = − ∇2 + Vext r, RI + i 2 I =1

ρr dr + vxc ρ r r− r

111

occupied

ψ KS r − i

Λij ψ KS r i

47

j=1

r is the fictitious acceleration of the Kohn–Sham wave function (as defined In Equation 4.6, ψ KS i in the Car–Parrinello approach (Car and Parrinello 1985, Remler and Madden 1990)). A corresponding fictitious kinetic energy, KE(ψ), of the wave function degree of freedom may also be computed. This propagation must be carried out with the orthonormality constraints, Equation 4.4, held tight (Car and Parrinello 1985, Remler and Madden 1990, Marx and Hutter 2012). The magnitude of KE(ψ) is controlled by the fictitious mass μ and must be kept small to obtain realistic dynamics for the real position variables, RI(t). To further define the application of Equations 4.1–4.6 to model systems, choices must be made that affect the accuracy and efficiency of the calculation. These are discussed in the following. More detailed discussions of some of the more complex of these issues are given in the appendices. The first consideration that must be made is the number of atoms (number of electrons, N) to be included in the simulation. The numerical work in the calculation scales as N3. For the problems we will discuss in the following, the target systems are periodic bulk materials (3-D periodicity, test calculations; see following text) or perfect surface terminations with disordered fluid interfaces (2-D periodicity). For the bulk phases the periodic unit cell defines the minimum size of the calculation with appropriate Brillouin zone sampling (Ziman 1972, Ashcroft and Mermin 1976). (In some cases we use a large unit cell (e.g., X × Y × Z Å3) to avoid Brillouin sampling (Ziman 1972, Remler and Madden 1990).) For interface problems, we use a slab construction (limited number of bulk layers). For some problems where required in order to model changes in the surface fluid structure, we will use an enlarged surface unit cell. This will be indicated in the calculations discussed in the application sections in the following. The nonlinear partial differential equations (Eqs. 4.3 and 4.6) are not defined until the external potential, Vext(r, RI), the exchange–correlation function, vxc(ρ(r)), and the fictitious mass, μ, are defined. μ is a parameter that controls how tightly the Kohn–Sham DFT equation constraint, Equation 4.6, is maintained. As μ 0, the Kohn–Sham equations are closely constrained, but the fictitious dynamics of the wave functions are very fast, implying that the time integration step must be kept very short (e.g., 0.12 fs) or the system will not stay close to the Born–Oppenheimer surface defined by the solution to Equation 4.3. Also as μ increases, the rate of transfer of energy into the fictitious kinetic energy of the electronic degrees of freedom (Remler and Madden 1990) increases. As this occurs, the total energy in real structural coordinates will not be conserved. Generally a compromise must be made between the efficiency of the CP algorithm and the retention of the kinetic energy in the particle degrees of freedom. The exchange–correlation potential, vxc(ρ(r)), is an unknown function and must be considered as part of the phenomenology of DFT. There have been many efforts to develop functions that will represent accurately a large number of systems (Zhao and Truhlar 2008, Burke 2012, Becke 2014, Luo et al. 2014). These include purely local functions (LDA) (Kohn and Sham 1965), semilocal functions (e.g., PBE (Perdew et al. 1997)), and many others (Zhao and Truhlar 2008). In systems that contain localized d states, exchange is particularly important and exact exchange corrected functionals (hybrid functionals, e.g., PBE0 (Adamo and Barone 1999)) and DFT+U corrections (Anisimov et al. 1993, Liechtenstein et al. 1995, Dudarev et al. 1998, Rollmann et al. 2004, Zhou et al. 2004) are frequently used. In the following, we will present computational results that evaluate the use of several of the most popular versions of exchange correlation (e.g., PBE (Perdew et al. 1997) and PBE0 (Adamo and Barone 1999)) for calculations applicable to the transition metal oxide fluid interface.

112

Molecular Modeling of Geochemical Reactions

To solve Equations 4.3 and 4.6, the Kohn–Sham functions must be expanded in a basis. Two common choices are local functions (atom-like orbital functions typically used in molecular calculations) (Szabo and Ostlund 1996) and plane waves (typically used in condensed matter calculations) (Ashcroft and Mermin 1976, Car and Parrinello 1985, Marx and Hutter 2012). To achieve the efficiency necessary for dynamical simulations, the calculations that we report here utilize plane waves. There are efficient local basis dynamical methods as well (Marx and Hutter 2012). In either choice, the basis must be large enough so that the solutions to the KS PDEs (e.g., Eqs. 4.3 and 4.6) are close to the intrinsic accuracy of the DFT approximations. The external potential, Vext(r, RI), representing the attractive interaction between the electrons and the atomic nuclei centers is the remaining function in Equations 4.3 and 4.7 that must be specified before calculation. In an all-electron calculation, this potential is defined in terms of the nuclear charge and position, ZI and RI. The core electrons and other valence electron selfconsistently screen these potentials. However, the effective atom–electron potentials experienced by the valence electrons are still fast varying near the nuclear centers. In addition, the orthogonality of the valence solutions to the core atom-like states in heavy atoms creates nodes in the valence orbital functions also resulting in their fast variation. To expand these rapid changes in plane waves would require an impractically large basis. (There are similar problems even when local basis functions are used.) In addition, the magnitude of the computational problem is a function of the number of Kohn–Sham orbital functions retained in the calculation (leading to an approximately N3 scaling). With the expectation that bond formation between atoms in a condensed system is accounted for by the valence electrons (wave functions), it is common to develop potentials (pseudopotentials) that operate only on the valence wave functions. The idea behind these potentials is that a much smoother potential may be developed for the valence electrons while still providing accurate bond energies and wave function variations in the bonding region. In addition, the introduction of these potentials removes the need to include core orbitals. There is an extensive literature in this area (see, e.g., Pickett 1989). The general theory is discussed in more detail in Appendix A.1. Pseudopotentials are developed from atomic calculations performed at the same level as proposed for the full many-atom condensed phase calculation. Therefore, the use of pseudopotentials does not represent a parameterization of the target condensed phase problem. However, there are issues that must be addressed that affect the transferability of the pseudopotential from the atomic problem to the condensed phase problem. Possibly the most important decision that has to be made is how many valence orbitals on a particular atom will contribute to the structure of the valence band. As we have mentioned, the time to solution for any DFT solver increases dramatically as the number of valence orbitals increases. Hence, the number of orbitals retained in the calculation is a critical decision. For the second row elements (see the α-Al2O3 (corundum) calculations in the following), this is fairly straightforward. The np = 1 (np is the principal quantum number) 1s orbital function in the core is well separated from the np = 2, 2s, and 2p orbital functions. On the other hand, for second row elements, the separation between the 2s and the 2p functions in the np = 2 shell is not sufficient (even on the right-hand side of the periodic table) to suppress hybridization in condensed materials (Kawai and Weare 1990). For these materials, the 2s and 2p orbital pseudopotentials are needed. For the third row and higher periods, this decision is tricky. For example, for the 3d Fe atom (an important component of the transition metals of interest to this work), the 3s, 3p, 3d, and 4s are all candidates for hybridization. For efficiency, the five 3d orbitals and the fact that there may be highspin and low-spin configurations create a difficult problem. For these atoms, we have found that the 3s and 3p orbitals that are filled in the isolated atom may play a role in the bulk bonding problem (Fulton et al. 2010, 2012). Including these orbitals is straightforward but does increase the cost of the

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

113

calculation. In the next section, we will discuss this issue by direct comparison of calculated results to observed bulk properties. These calculations will illustrate the level of DFT theory that is necessary to describe (even qualitatively) transition metal oxide minerals and also set the stage for the discussion of the effects of adding a fluid interface in the section following. The projector augmented plane wave method (PAW) of Blochl removes many of the problems of the somewhat ad hoc nature of the pseudopotential approach (Blochl 1994, Holzwarth et al. 1997, Kresse and Joubert 1999, Valiev and Weare 1999, Bylaska et al. 2001, 2002, Valiev et al. 2003). Similar to pseudopotential methods, PAW retains the use of a plane wave basis and all the advantages associated with it. However, in the PAW approach instead of discarding the rapidly varying parts of the electronic functions, these are projected onto a local basis set (e.g., a basis of atomic functions). By carefully choosing the local basis, the convergence of the plane wave problem can be systematically improved. The PAW method is an all-electron method, and no part of the electron density is removed from the problem. In addition, the norm-conservation condition can be relaxed in PAW resulting in smaller plane wave cutoffs. Both of these features can offer a significant advantage over the pseudopotentials, in particular for systems containing elements with hard pseudopotentials (e.g., oxygen, fluorine, first row transition metals, and lanthanides).

4.3

Accuracy of Calculations for Observable Bulk Properties

A first consideration in the development of an atomic-level model is to establish the expected accuracy of the methods of calculation that you intend to use. In this section, we will evaluate the accuracy of the application of the approaches that we have discussed when used to calculate bulk properties (a relatively inexpensive calculation). Similar calculations and comparisons to solution properties can be found in our recent efforts to simulate X-ray structures of aqueous solutions (Fulton et al. 2010, 2012). In the following sections, we will discuss the application to surface (Section 4.4) and mineral/fluid interface problems (Section 4.5). The transition metal oxide hematite (α-Fe2O3) and the oxyhydroxide goethite (α-FeOOH) are central to many geochemical applications and are a significant challenge to the computational methods that we are using. The structures of the unit cells of these solids are given in Figure 4.1. The d electrons in these materials are localized about the Fe atoms and exhibit strong correlation. Because of the complicated physics associated with modeling the correlation of five d electrons in confined orbits, the development of a computationally tractable model for their electronic structure leads to some uncertainty (difficult choices) in the development of pseudopotentials and in the level of exchange used. In order to compare the accuracy of these calculations to a very well-determined mineral, we include similar calculations of corundum (α-Al2O3). This mineral has a structure similar to hematite, but the pseudopotential is more straightforward to develop and the electronic structure calculation is relatively easy because it is closed shell. 4.3.1

Bulk Structural Properties

Tables 4.1 and 4.3 summarize our calculations of the bulk lattice structures given in Figure 4.1 using the various levels of approximation discussed in Section 4.2. The estimates of the bond lengths illustrated in Figure 4.1 are given in Tables 4.2 and 4.4. 4.3.1.1

Calculated Structures of Corundum

The conventional unit cell lattice parameters for this mineral are reported at various levels of electronic structure calculation (LDA, PBE, and PBE0) as well as observed values in Table 4.1. The

114

Molecular Modeling of Geochemical Reactions (b)

(a)

b

c

a

c

a

Figure 4.1 Unit cells for bulk goethite structure (a) and corundum structure (b) (left, 30-atom cell; right, 10-atom cell). Hematite and corundum have same crystal structure (left, 30-atom cell; right, 10-atom cell). Fe and Al, blue; O, red; H, white. The “a,” “b,” and “c” are the lattice vectors. Table 4.1 Lattice parameters of the conventional cell of corundum (Å for a, b, c; for α, β, γ) calculated using LDA, PBE, and PBE0 plane wave DFT calculations. LDA

PBE

PBE0

Experimenta

1×1×1 2×2×1 4.675 4.675 12.736 90.0 90.0 120.0

1×1×1 2×2×1 4.767 4.767 12.999 90.0 90.0 120.0

1×1×1 1×1×1 4.681 4.681 12.909 90.0 90.0 119.9

4.758 4.758 12.990 90 90 120

Lattice parameters Supercell MP a b c α β γ

100 Ry (2721 eV) and 200 Ry (5442 eV) were used for the wave function and density cutoff energies. The corundum conventional cell contains 12 Al and 18 O atoms. a Kirfel and Eichhorn (1990).

Table 4.2 Bond lengths and atom center distances (in Å) of corundum calculated using LDA, PBE, and PBE0 plane wave DFT calculations. Distances

LDA

PBE

PBE0

Experimenta

Supercell MP Al─Al-a Al─Al-b Al─O1 Al─O2

1×1×1 2×2×1 2.611 3.757 1.817 1.941

1×1×1 2×2×1 2.666 3.833 1.853 1.980

1×1×1 1×1×1 2.638 3.816 1.818 1.962

2.654 3.840 1.854 1.971

100 Ry (2721 eV) and 200 Ry (5442 eV) were used for the wave function and density cutoff energies. a Kirfel and Eichhorn (1990).

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

115

Table 4.3 Lattice parameters for the conventional cells of hematite and goethite (Å for a, b, c; for α, β, γ) calculated using LDA, PBE, PBE+U, and PBE0 plane wave DFT calculations. Lattice parameters Hematite Supercell MP a b c α β γ Goethite Supercell MP a b c α β γ

LDA PSP1

LDA PSP2

PBE PSP1

PBE PSP2

PBE+U PSP1

PBE0 PSP1

Experiment

1×1×1 2×2×1 4.683 4.681 13.515 90.68 89.27 119.1

1×1×1 2×2×1 4.498 4.499 13.228 90.65 89.37 119.3

1×1×1 2×2×1 5.154 5.155 13.820 89.98 90.02 119.9

1×1×1 2×2×1 4.947 4.947 13.602 89.93 90.07 120.0

1×3×2 1×1×1

1×1×1 1×1×1 5.122 5.122 13.63 87.84 92.16 119.6

5.036a 5.036a 13.747a 90a 90a 120a

1×1×1 1×3×2 9.662 2.586 4.383 90.0 90.0 90.0

1×1×1 1×3×2 9.285 2.644 4.220 90.1 90.0 89.9

1×1×1 1×3×2 10.107 3.024 4.611 90.0 90.0 90.0

1×1×1 1×3×2 9.858 2.976 4.515 90.0 90.0 90.0

2×2×1 1×1×1

1×1×1 1×1×1 9.812 3.168 4.240 90.0 90.0 90.0

9.951–9.956b 3.018–3.025b 4.598–4.616b 90b 90b 90b

100 Ry (2721 eV) and 200 Ry (5442 eV) were used for the wave function and density cutoff energies. The hematite conventional cell contains 12 Fe and 18 O atoms, and the goethite conventional cell contains 4 Fe, 8 O, and 4 H atoms. U = 4 eV for all PBE+U calculations in this paper. The PSP1 and PSP2 pseudopotentials for Fe atom have [Ar]3s3p and [Ar] as core. The largest calculation here is PBE+U calculation using 2 × 2 × 1 cell that contains 408 spin-up and 408 spin-down electrons. a Maslen et al. (1994). Reproduced with permission from Wiley. b Alvarez et al. (2008).

relatively simple LDA calculations provide reasonable results comparing to experiments, with a difference around 2.0%. Using this level of approximation, the bonds are usually shorter than experimental observations. The semilocal GGA calculations, PBE, which can be done for roughly the same computational cost as LDA give much better agreement with experiment, within about 0.2%. The higher-level electronic structure calculation, PBE0, provides slightly less accuracy of 1.6%. PBE0 is about the highest-level calculation that can practically be performed for these minerals. It is discouraging that the accuracy of geometry decreases slightly in going from PBE to PBE0. However, the excellent results from the more efficient PBE calculations are quite positive. Problems of this sort are common in structural DFT calculations (Khein et al. 1995). There is much more support for the PBE0-type calculations in the electronic structure calculations for the spin-ordered systems hematite and goethite. The bond lengths for corundum given in Table 4.3 show the same trends. Again, the LDA bond lengths are shorter than observed. PBE calculations provide predictions that are in better agreement with the structural data. In this case as well, PBE0 calculations are not as accurate as the PBE calculations. Although changes can be made to the hybrid functional (Pozun and Henkelman 2011) or other functionalities introduced (Zhao and Truhlar 2008), this is probably the best level of agreement experiment one can expect from a present-day DFT calculation. (You might want to read and refer to DeMichelis et al. (2010). They use a variety of methods to model silicates which may be an even more tractable system.)

116

Molecular Modeling of Geochemical Reactions

Table 4.4 Bond lengths and atom center distances (in Å) of hematite and goethite calculated using LDA, PBE, PBE+U, and PBE0 plane wave DFT calculations. Distances Hematite Supercell MP Fe─O1 Fe─O2 Fe─Fe-a Fe─Fe-b Goethite Supercell MP O1─H O2─H Fe─O1 Fe─O2 Fe─Fe-a Fe─Fe-b

LDA PSP1

LDA PSP2

PBE PSP1

PBE PSP2

PBE+U PSP1

PBE0 PSP1

Experiment

1×1×1 2×2×1 1.878 1.971 2.646

1×1×1 2×2×1 1.795 1.930 2.595

1×1×1 2×2×1 1.994 2.132 2.906

1×1×1 2×2×1 1.887 2.122 2.933

1×3×2 1×1×1 1.996 2.187 2.922

1×1×1 1×1×1 1.964 2.137 2.904

1.946a 2.116a 2.900a

1×1×1 1×3×2 1.093 1.302 1.915 1.906 2.890 3.415

1×1×1 1×3×2 1.094 1.306 1.867 1.846 2.802 3.316

1×1×1 1×3×2 0.973 1.743 2.134 1.973 3.363 3.456

1×1×1 1×3×2 0.968 1.808 2.100 1.911 3.317 3.352

2×2×1 1×1×1 0.982 1.651 2.123 1.992 3.323 3.493

1×1×1 1×1×1 0.963 1.637 2.079 1.949 3.245 3.381

2.077–2.103b 1.937–1.954b 3.283–3.311b 3.432–3.465b

100 and 200 Ry were used for the wave function and density cutoff energies. a Maslen et al. (1994). Reproduced with permission from Wiley. b Alvarez et al. (2008).

4.3.1.2

Calculated Structures for Hematite and Goethite

In Tables 4.3 and 4.4 structure calculations at the various levels are reported for hematite and goethite. The LDA results are significantly worse than the PBE results. The average agreement for unit cell parameters is ~0.1 Å (~1.5%) for the PBE and PBE0, whereas the LDA result is ~0.3 Å (~3.0%). Due to the strongly correlated nature of d electrons for these systems, GGA calculations are expected to be less accurate than the hybrid calculations (Luo et al. 2014). From Table 4.4, the difference from experiment of the bond lengths (Fe─O bond) within the unit cell is 0 are spin-up band functions, and y < 0 are spin-down). Most of the Fe-d component in valence bands is spin-down and the Fe-3d component in the conduction bands is primarily spin-up. The most important difference

(a)

pbe96_fe pbe96_o

PDOS per atom

4

2

0

–2

–4 –10

–8

–6

–4

–2 Energy (eV)

0

2

4

6

(b) pbe96_u_fe pbe96_u_o

4

PDOS per atom

2

0

–2

–4 –10

–8

–6

–4

–2 0 Energy (eV)

2

4

6

(c) pbe0_fe pbe0_o

4

PDOS per atom

2

0

–2

–4 –10

–5

0 Energy (eV)

5

10

15

Figure 4.3 Projected density of states for bulk hematite using different DFT levels. The pink line represents PDOS for the Fe atoms, and the red line represents PDOS for the O atoms. The line above the middle line (y = 0) is spin-up density line and below y = 0 is spin-down density line. The upper, middle, and lower figures are from (a) PBE, (b) PBE+U, and (c) PBE0, respectively.

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

123

between the various levels of calculations is that in the DFT+U and the PBE0 calculation, the occupied d band is much narrower and moved to lower energy than in the PBE calculation reflecting the expected localization in a more correlated theory. This means that the states just below the band gap in the PBE+U and PBE0 calculation are primarily O-2sp states, suggesting that these materials are O-2p Fe-3d charge transfer insulators (near ion lattices, which is consistent with X-ray absorption and emission spectra measurements) (Fujimori et al. 1986, Lad and Henrich 1989, Ciccacci et al. 1991, Cox 1992, Dräger et al. 1992). When using the PBE functional, more Fe-3d and O-2p hybridization is obtained leading to the high Fe-3d Fe-3d character in states at the top of the band in hematite. The widths of the O-2sp bands in all the PBE0 and DFT+U calculations are similar. These calculations again show that some level of exchange beyond PBE is necessary to provide a reasonable interpretation for these minerals. However, in the surface and interface calculations that we report in the following section, the PBE0 calculation is extremely expensive, so we do most of these calculations at the DFT+U level.

4.4

Calculation of Surface Properties

4.4.1 4.4.1.1

Surface Structural Properties Surface Termination

O-

ter mi n Oter ated mi _0 na 1 Fe ted 2_# -te _ rm 01 1 2_ ina #2 ted _0 12

To initiate calculations of the surface and interface regions, unreconstructed surfaces cut from the bulk (see Figures 4.4 and 4.5) are taken as starting structures to be optimized. Because of the loss of three-dimensional symmetry, the calculations can become expensive, and it is not feasible to search all possible configurations. Electronic structure calculations have not been applied directly to more

O-terminated_001 Fe-terminated_001_#1 Fe-terminated_001_#2

c a

Figure 4.4 Different possible terminations of hematite and corundum (oxygen, red; iron and aluminum, blue). The “a” and “c” are the lattice vectors.

124

Molecular Modeling of Geochemical Reactions (a)

b c c

a

a

(b)

a

c

Figure 4.5 Al2O3 (001) surfaces: (a) Al-terminated corundum (001) surface (top Al layer moves down to O layer). (b) O-terminated corundum (or hematite) (001) protonated surface. The “a,” “b,” and “c” are the lattice vectors.

complicated models (such as roughness, steps) that have been considered in experiments but will need many more atoms computationally. As illustrated in Figures 4.4 and 4.5, selecting different surface cleavage planes (terminations) results in different surface structures and compositions. In forming the cleaved surface structure, bonds are broken, leaving the surface atoms in the newly terminated surface with unsaturated bonds. These dangling bonds may be passivated (terminated) by surface reconstruction, by reaction with surrounding phases, or by protonation. These changes affect the local electronic structure and nature of bond formation in the surface region. Different sample preparations and observational environments can result different termination structures with the same surface planes and atomic compositions. For the transition metal oxide and oxyhydroxide minerals, determining this surface structure and the composition in a particular experimental setup may be difficult (Kerisit 2011). Additional information may be obtained from computational estimates of the relative stability of various surfaces (ab initio thermodynamics (Wang et al. 2000, Reuter and Scheffler 2002, Lo et al. 2007)). In principle, temperature corrections to the surface free energy may be calculated (Esler et al. 2010). However, for minerals at low temperatures and pressures, it is reasonable to use enthalpy differences, neglecting the zero-point and temperature effects in the free energy. One needs to keep in mind, however, that neglect of entropic effects can make

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

125

determination of the equilibrium configuration problematic when the enthalpy differences between configurations are relatively small (Wesolowski et al. 2012). 4.4.1.2

Corundum Terminations

The closed-shell mineral corundum (α-Al2O3) has the same structure as one of our principal mineral targets, hematite (α-Fe2O3), but is an easier computational problem allowing us to explore the accuracy of further choices that affect the accuracy of our calculations, for example, the number of mineral layers in the calculation required for reasonable accuracy and the loss of 3-D symmetry in the surface region (Rustad et al. 2003, Liu et al. 2010, Kumar et al. 2013). The changes in structure created by expanding the surface unit shell from (1 × 1) to (2 × 2) or (2 × 3) surfaces are shown in Tables 4.1–4.3. These results suggest that at a unit cell least around 10 × 10 Å is required to retain the accuracy of the DFT calculation. In our plane wave basis calculations, we use a unit with cell with a large dimension perpendicular to the surface to include at least four layers of metal atoms (roughly 10–15 Å) to accurately represent the transition from the bulk to the surface region. This means that at least a few hundred of atoms are needed to model the surface structure accurately. 4.4.1.3

Surface Relaxation Corundum (001)

For this mineral, the (001) surface termination obtained in an experimental setup is dependent on the method generating of the surface and the measurement environment (e.g., for corundum, high vacuum (Renaud 1998), aqueous overlayers (Catalano 2011)). For the (001) surface in vacuum, the Al termination (Figure 4.5a) appears to be the most stable (Renaud 1998), whereas for surfaces in contact with absorbed water layer the O-termination (Figure 4.5b) agrees well with interpretation (Catalano 2011) and the Al-terminated surface has not been reported. Considerable surface reconstruction has been observed for the Al-terminated (001) surface shown in Figure 4.5A (Guenard et al. 1998, Renaud 1998). Our calculations show that the unsaturated Al atoms in the terminated top layer move down into the surface forming shorter bonds with the surface O atoms. These PBE results agree with prior DFT (Manassidis et al. 1993) and qualitatively with empirical potential results (Catlow et al. 1982, Mackrodt et al. 1987). Our locally optimized calculations show the top Al layer has been moved down by 84.8%; this is in qualitative agreement with experimental estimates of 51.0% (Guenard et al. 1996). The Al atoms in the top layer are bonded with the three surface Al atoms, while the bulk Al atoms are sixfold coordinated. The Al─O bond length in the reconstructed surface is 1.678 Å (Al─O bond lengths in the solid are 1.853 and 1.980 Å). The surface termination shown in Figure 4.5b agrees most closely with experiment (Catalano 2011). In the calculations for this surface, the excess negative charge created in the termination is passivated by protonation of the three dangling O bonds. As opposed to the Al-terminated surface, the protonation of the O surface bonds reduces the excess valence of the exposed O atoms and reduces the need to form shortened bonds (Brown and Shannon 1973, Brown 2014). Therefore, there is much less reconstruction in this surface than in the Al termination (see Figure 4.3). For the calculated O-terminated hydrated surface of corundum (001), the O layer on top moves up by 3.5%. The two types of bond lengths of Al─O are 1.87 and 2.01 Å versus the bulk bond lengths of 1.85 and 1.98 Å. 4.4.1.4

Protonation of the Goethite (100) Surfaces

In Figure 4.6, we show results of the calculation of the cleavage of goethite to produce the two possible (100) surfaces with different surface protonation schemes (see also Kubicki et al. (2008)).

126

Molecular Modeling of Geochemical Reactions

Termination l

Termination Il

a

b

c

Figure 4.6 Bulk structure and two types of (100) surface terminations of goethite. The “a,” “b,” and “c” are the lattice vectors.

We estimated the surface energies of these two terminations with full reconstruction after terminating the unsaturated O atoms with protons to produce a neutral surface. In this calculation, the termination I produced a surface energy of 156.7 meV/Å2 versus 130.57 meV/Å2 for termination II. This predicts termination II is more stable which agrees with the interpretation of the experimental observation of Ghose et al. (2010). 4.4.1.5

The Protonation of Hematite Surfaces

For the O-terminated hematite (012) surface, Catalano and Fenter (Catalano et al. 2007) proposed O-terminated_012_#1, Figure 4.4, after investigating this surface using X-ray reflectivity (crystal truncation rod (CTR) experiments (Fenter and Sturchio 2004)). On the other hand, Tanwar et al. (2007) proposed the O-terminated_012_#2 in Figure 4.4 to interpret X-ray CTR diffraction data. We used Catalano’s termination for this surface in the calculations reported here. In these calculations, the O atoms are capped with protons. Tanwar et al. (2009) proposed both single-domain and double-domain models for hematite (001) in vacuum. Trainor et al. (2004) studied hydrated hematite (001) surface and suggested two hydroxyl moieties on surface. For the O─O─Fe─O─Fe─O─O─Fe─R (012) cleavage of hematite, there are two possible protonation models (Kerisit 2011), producing a neutral surface (OH)─(OH)─Fe─O─Fe─O─O─Fe─R (Figure 4.7a, Model 1) (Yin and Ellis 2009) and (OH2)─O─Fe─O─Fe─O─O─R (Figure 4.7b, Model 2). In both models, there are singly Fecoordinated (OII) and triply Fe-coordinated (OI) oxygen atoms. The OII is deeper in the surface layer. (In the bulk O atoms are tetrahedrally coordinated with Fe atoms (Figure 4.1b).) Both these models require the expansion of the 2d surface unit cell to 2 × 2. To determine the most stable surface protonation scheme, we optimized the structures of the two different (2 × 2) protonated surface models at the PBE+U level. (PBE+U is used for all hematite surface and interface calculations in the following.) The Model 1 protonation scheme (OH)─(OH)─Fe─O─Fe─O─O─Fe─R was found to be 11.7 meV/A2 lower in surface energy than Model 2. After structural optimization the deeper OIIH hydroxyl bridges to the OIH further stabilize this surface termination by forming an H-bond (2.73 Å, 154.6 ) (Figure 4.4a and b, H-bond 1). The stability of the Model 1 structure may be supported by bond valence analysis (Brown and Shannon 1973, Brown 2009). H-bond 1 reduces the bond valence of (O II producing slight oversaturated

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties Model 1

(a)

(c)

Model 2 H-bond a

c

a

b

(b)

H-bond b

H-bond 1

Singly Fe-coordinated oxygen Ol Triply Fe-coordinated oxygen OlI c a

127

b

(d)

H-bond A H-bond B

c b

c

a

Figure 4.7 Water molecule absorbed on different hydration models for hematite (012) surface. (a) and (b) are side views of Model 1, and (c) and (d) are side views of Model 2. The “a,” “b,” and “c” are the lattice vectors.

valences. The protonation of OI contributes bond valence to the OI oxygen, which is only singly Fe coordinated. Even with this protonation, the O atoms in the terminated surface still are considerably under saturated (in the bond valence analysis) and will be sites for absorption of electron-donating species. 4.4.2 4.4.2.1

Electronic Structure in the Surface Region Corundum (001) Al Terminated

The PDOS for surface atoms of the Al-terminated (001) (projection on the Al atom) and bulk corundum (001) (also projected at the Al atom) are illustrated in Figure 4.2. While valence bands look similar in the PDOS, the biggest difference between bulk and surface corundum is the existence of surface states in the edge of conduction bands, which narrow the band gap from 7.0 to 4.0 eV. The reason for this is the large change in the position of the surface Al and the resulting surfacelocalized states created by the (001) reconstruction (see Figure 4.5). Similar band gap changes have been reported in other materials (Lazic et al. 2013, Caban-Acevedo et al. 2014) and are important to solar energy conversion. 4.4.2.2

Oxygen-Terminated Hematite (001) Surface

In Figure 4.8a, the PDOS for Fe atoms in bulk and on the (001) surface have been plotted. The PDOS widens slightly in the surface region; however, the bandwidths, band gaps, and local spin moments change very little on approaching the surface. The surface geometry of the (001) termination is illustrated in Figure 4.5b. Each Fe atom in hematite is hexagonally coordinated connecting two adjacent O layers. This coordination is similar to the bulk so the chemical environment of

128

Molecular Modeling of Geochemical Reactions (a)

PDOS per atom

4

Fe_bulk Fe_surface

2

0

–2

–4 –10

–8

–6

–4

–2 Energy (eV)

0

2

4

6

(b) 4

PDOS per atom

O_bulk O_surface 2

0

–2

–4 –10

–8

–6

–4

–2 Energy (eV)

0

2

4

6

Figure 4.8 Projected density of states for Fe atoms (a) and O atoms (b) in surface and bulk region of O-terminated hematite (001) using PBE+U.

surface Fe layer is similar. We show elsewhere (Chen 2015) that the spin ordering does not change in going from the bulk to the surface. The PDOS projected from the surface O is illustrated in Figure 4.8b. The spin-up (y > 0) and spindown (y < 0) DOS are the same for O atoms in the bulk. However, in the surface region as is illustrated in Figure 4.8b, there is a substantial difference between the spin-up and spin-down PDOS. As we have shown in solution simulations of transition metal ions (Bogatko et al. 2010), the local spin moment of transition metal atom can influence the spin polarization of connected O atom. The transition metal oxide hematite has spin moments locally on metal atoms in the bulk. However, O atoms in hematite are not spin polarized because up-spin and down-spin are connected to adjacent metal atoms with opposite spin moment directions in the symmetrical bulk structure. This suggests that the

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

129

presence of a surface termination removes metal atoms in certain spin direction and breaks this symmetry leading to the polarized spin states of the surface O atoms.

4.4.3

Water Adsorption on Surface

Even with protonation to remove excess charge, the valence of the O atoms in the surface may remain under saturated (i.e., able to accept more positive charge) (Brown and Shannon 1973, Brown 2014). This means that proton interaction between water molecules outside the surface O atoms and the H-bonds between surface atoms can stabilize the surface region. This will play a significant role in determining the water structure on surface. Recently major advances have been made in the observation of water layer structures on the surface region using synchrotron-generated X-ray diffraction. Ghose et al. (2010) employed CTR techniques to study the 3-D structure of water layers on the goethite (100) surface. Tanwar et al. (2007) also used this technique to investigate water adsorption on hematite (012) surface. Structural models identified in this work can be investigated with the computational tools. Ab initio (DFT+U) methods have been used by Kubicki et al. (2008) to study water layer structure on goethite (100) surface. More recent calculations by Chen (2015) have investigated water adsorption for the same surface. Lo et al. (2007) concluded the surface water can stabilize surface structure by hydrogen bonding when studying water adsorption on hematite (012) surface. As a starting point for adsorption optimizations of the interface structure on hematite, we studied the single water adsorption process on the protonated hematite (012) surface by placing one water molecule in contact with the (2 × 2) surfaces of Model 1 and Model 2 shown in Figure 4.7. The adsorption energy for one water molecule on both protonated models can be calculated as Eadsorption = Ewater + Esurface −Ewater − surface Our calculated water adsorption energy for Model 1 is 28.05 kJ/mol. This is slightly greater than the adsorption energy 23.15 kJ/mol calculated for Model 2. In Model 1, the absorbed water forms an H-bond with an (O I atom as a donor (2.71 Å, 160.4 , H-bond A in Figure 4.7b) and an H-bond with another (O I atom as an acceptor (2.70 Å, 168.9 , H-bond B in Figure 4.7b). In Model 2 (see Figure 4.7c and d), the absorbed water forms an H-bond with OII donating electrons (2.90 Å, 166.0 , H-bond b in Figure 4.7c). It also forms an H-bond with an OI − H hydroxyl where its proton points toward the surface (2.75 Å, 146.4 , H-bond a in Figure 4.7c). The side views of water adsorption, hydrated models, and H-bonds are shown in Figure 4.7. Combining the result from surface protonation section, Model 1 is slightly more stable energetically as a bare protonated surface. This structure also has slightly greater water adsorption energy when one water molecule is absorbed on the (2 × 2) surface. However, the energy difference between Model 1 and Model 2 is small, suggesting that the coexistence of Model 1 and Model 2 is possible. Catalano and Fenter (Catalano et al. 2007) in analyzing their CTR data on this surface proposed water layer, which is regarded forming hydrogen bonds with (O II atoms agreeing more with Model 2. In our Model 1, the OIIH group is oversaturated and forms a donor hydrogen bond with OI (Figure 4.4a), suggesting that it will not form hydrogen bonds to additional waters. On the other hand, in Model 2, OII atoms are undersaturated, and it is possible to attract water molecules, which may form a water layer closer to the surface and more in agreement with the CTR data interpretation of Catalano and Fenter.

130

Molecular Modeling of Geochemical Reactions

4.5 Simulations of the Mineral–Water Interface The interactions of the solution species in the interface region are relatively weak and at finite temperature the moles are in constant motion. For example, CTR measurements of surface regions (Fenter 2002, Fenter and Sturchio 2004, Park et al. 2005, Catalano et al. 2007, 2010a, 2010b, Fenter et al. 2010a, 2010b, 2013) are observing the thermally averaged structure of the system. For such systems the dynamics of the system must be taken into account. This may be done with CMD (Shroll and Straatsma 2003, Kerisit 2011, Boily 2012, Kerisit et al. 2012) in which classical force fields are used or with much more computer-intensive models in which the forces between atoms are calculated from first principles (Kubicki et al. 2008, Kumar et al. 2013, Huang et al. 2014). Kerisit (2011) used CMD simulations with four different force fields to determine the atomiclevel structure of three hematite–water interfaces, (001), (110), and (012). The simulation results were compared with experimental electron density profile obtained from CTR data (Catalano et al. 2007). In CMD, large-scale models containing many thousands of atoms can be used due to the simplicity of the empirical potentials. However, these interaction models cannot account for the changes in bonding conditions that occur in the interface region. These many-body effects are especially difficult to treat for the transition metal oxides and oxyhydroxides. The first principles dynamical methods we use avoid these problems by calculating the interactions directly from the solution to the electronic structure problem. This produces a parameter-free model of the interface region. The theoretical background of AIMD has been discussed in Section 4.2. Similar to the bulk and surface optimization calculation, DFT and DFT+U are the most common electronic structure methods used in AIMD. Huang et al. (2014) ran AIMD simulation, structural properties, and infrared spectra of the corundum (001)/water interface. Kumar et al. used AIMD to study H-bonds and vibrations of water on (110) rutile. In the calculations reported here, AIMD simulations using the Car–Parrinello molecular dynamics (CPMD) method were used (see Section 4.2). We have reported AIMD studies of the goethite (100)–water interface (Chen 2015), comparing types of surface bonds, types of water molecules, and dynamics process on the surface. Hematite (001) and (012) surface– water interface properties as well as Fe2+ ion adsorption processes are also in preparation for publication (Chen 2015). Some of these results for the (012) termination of hematite are briefly discussed in the following.

4.5.1

CPMD Simulations of the Vibrational Structure of the Hematite (012)–Water Interface

Similar to our slab in the surface section previously, we include four Fe layers and employ Model 1 (Figure 4.7), which can be represented by (OH)2-X-(OH)2-Fe2-O2-Fe2-O2-O2-Fe2-O2-Fe2-(OH)2X-(OH)2 in c axis. Between two surfaces of the slab in the simulation cell (periodic boundary condition employed), we add 54 water molecules (c = 25.0 Å). The density of water is set to 1.0 g/cm3. The total simulated cell is electrically neutral; it has 1104 valence electrons and 290 atoms, which include 32 Fe, 118 O, and 140 H. Equation of motions in CPMD were integrated using Verlet algorithm, with a time step of 0.12 fs and fictitious orbital mass of 600.0 a.u. All H atoms have been replaced by deuterium to allow longer time steps. For very accurate thermodynamics and time correlation functions, one needs to be careful about using different masses. This can affect the entropy and other chemical behaviors such as water dissociation constants (Kw) (Mesmer and Herting 1978). For the most part these effects are small, and in any case to obtain a very accurate thermodynamic description of H vibrations will require that they be treated using

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

131

Bulk water

Interface water Ol Oll b

a

c

Figure 4.9 Interfacial and bulk water regions of hematite (012)–water interface. The “a,” “b,” and “c” are the lattice vectors.

path integral quantum dynamic techniques (Allen and Tildesley 1987). The canonical ensemble was chosen for this simulation. Nose–Hoover thermostat was used to control the temperature for ions (300 K) and electrons (1200 K). CPMD simulations have been performed 12 ps for each system, and 1 ps equilibration was performed after which the trajectory data was collected. As discussed in the Section 4.4, the OID and OIID bonds (Figure 4.7) play significant roles in surface structural and dynamical properties. The changes in structure as the interface layer goes from the surface interface to the water bulk are illustrated in Figure 4.9 and discussed in more detail in other publications (Chen 2015). In the following, we present the power spectrum analysis using generating trajectory from CPMD. 4.5.1.1

Calculation of Power Spectrum

The power spectrum can be calculated by taking Fourier transform of the velocity autocorrelation functions (Kohanoff 1994, Thomas et al. 2013, Huang et al. 2014): P ω =m

R τ R t+τ

τ

e −iωt dt

R τ R t + τ denotes the velocity autocorrelation of the velocity R. Deuterium atom positions (all H atoms have been replaced by deuterium) associated with water molecules in the trajectories of the species in the hematite (012)–water interface are used to calculate power spectrum. The results are shown in Figure 4.10. The peaks located in the range of 0–800 cm−1 in both Figures 4.10a and b, corresponding to the intermolecular librational and translational motions, are difficult to interpret. In Figure 4.10b, there are peaks around 1185 cm−1 which are not shown in Figure 4.10a. That peak is due to the bending motion of water molecules, so that’s not observed in vibration of the OD group spectrum. The value

132

Molecular Modeling of Geochemical Reactions

(a)

(b) OD1 OD2

0

500

1000

1500

2000

2500

Water_interface Water_bulk

3000 0

500

1000

1500

2000

2500

3000

Figure 4.10 Computed power spectrum of two types of OD groups on hematite (012) surface, water on interfacial region and bulk region. (a) OID and OIID surface group (cm−1) and (b) bulk water molecules (cm−1).

1185 cm−1 is in good agreement with the experimental value 1209 cm−1 in liquid water. The peaks in the range from 1800 to 2500 cm−1 are the contribution of OD stretching motion modes. In Figure 4.8a, the direct water surface bond with OID has a blueshift compared to OIID; this shows OIID forms a relatively stronger H-bond with OI, while OID likely has a weakened or H-bond in the interfacial region. In Figure 4.10b, in higher-frequency region, there is one peak 2180 cm−1, which is closer to experimental bulk ice frequency 2190 cm−1and another peak 2287 cm−1 that is closer to experimental bulk liquid frequency 2260 cm−1. This indicates water molecules in interfacial region have both liquid-like water and icelike vibrational properties as has been identified in the corundum calculations of (Huang et al. 2014).

4.5.2

CPMD Simulations of Fe2+ Species at the Mineral–Water Interface

Fe2+─Fe3+ redox cycling associated with iron (hydr)oxides is important to many geochemical and environmental processes including CO2 sequestration (DePaolo and Orr 2008), inorganic respiration, confinement of toxic materials (Brown et al. 1999, Brown 2001, Hochella et al. 2008, Navrotsky et al. 2008), dissolution and secondary mineral precipitation (Yanina and Rosso 2008), and cycling of biological nutrients (Newman 2010). Probably the simplest reaction in Fe2+─Fe3+ cycling is the oxidation of aqueous Fe2+(aq) with Fe3+ (oxyhydr)oxides such as hematite and goethite. Recently experiments by Rosso, Scherer, and others (Williams and Scherer 2004, LareseCasanova and Scherer 2007, Cwiertny et al. 2008, Gorski and Scherer 2008, 2009, 2011, Yanina and Rosso 2008, Scherer et al. 2010, Schaefer et al. 2011, Katz et al. 2012, Latta et al. 2012) have put into question the classical view of absorption Fe2+ occurring at static surface sites (e.g., surface complexation modeling) (Stumm et al. 1976, Davis and Kent 1990, Dzombak 1990, Sverjensky 1993, Katz and Hayes 1995) suggesting a much more fundamentally based framework (Gorski and Scherer 2011) in which the adsorbed Fe2+ oxidizes to Fe3+ with the electron transferring into the conduction or polaron (Rosso et al. 2003) band of the Fe3+ oxide, followed either by electron trapping in the solid and participating in a redox reaction with an environmental contaminant near the iron oxide surface (Bylaska et al. 2011a) or by electron transport leading to another surface Fe3+ to reductively dissolve (Yanina and Rosso 2008).

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

133

Accurate modeling of these interface processes for the iron oxide minerals presents a difficult electronic structure problem because their properties are heavily dependent on strongly localized d electrons, complex hydration processes, disorder of the surface/solution interface, the interaction of the solution phase with the highly charged mineral surface, etc. A previous electronic structure study using the DFT+U method has called into question whether or not Fe2+ spontaneously oxidizes upon bonding to a Fe3+ oxide surface (Russell et al. 2009). In this study they did not find evidence for electron transfer from a Fe2+ hexaqua complex into the (021) surface of goethite. However, they did find that the U correction changed the character of the small amount of electron sharing present from O-2p orbitals to Fe-3d orbitals. To handle this apparent discrepancy, it was suggested that defects might be playing a critical role in the spontaneous oxidation of absorbed Fe2+. Because of the limitations of their approach at the time, these simulations did not contain waters of solvation and a search for lowest-energy conformations of the adsorbed Fe2+ ions was not possible at the time. For this chapter, we revisited this problem performing studies of Fe2+ + hematite–water interfaces using the PBE+U method with full solvation and limited equilibrium using CPMD. Both (012) and (001) surfaces were simulated. As shown in Figure 4.11, we also found very little electron transfer resulting from the Fe2+ bidentate bond to the (012) hematite surface. However, we did find a significant amount of electron transfer with a tridentate bond to the (001) surface, which is somewhat consistent with the assertions of Rosso and Scherer (Rosso et al. 2003, Williams and Scherer 2004, Kerisit and Rosso 2006, Larese-Casanova and Scherer 2007, Cwiertny et al. 2008, Gorski and Scherer 2008, 2009, 2011, Yanina and Rosso 2008, Catalano et al. 2010a, 2010b, Scherer et al. 2010, Schaefer et al. 2011, Katz et al. 2012, Latta et al. 2012). While the PBE+U approach can lead to nontrivial amount of charge delocalization, a full conversion of a near surface Fe3+ to Fe2+ was not observed at this level of theory for this surface.

(a)

(b)

Figure 4.11 (a) Highest occupied molecular orbital (HOMO) of Fe2+ bidentate bond to the 012 hematite surface. Note little charge transfer. (b) HOMO of Fe2+ tridentate bond to the 001 hematite surface. Note charge transfer.

134

Molecular Modeling of Geochemical Reactions

4.6 Future Perspectives The development of realistic molecular models of geochemical processes is becoming possible because of the availability of ever-increasing computational capabilities. In this chapter, we discuss key aspects associated with simulating the structure and dynamics of the mineral–water interface using plane wave DFT. To illustrate these types of simulations, we showed a variety of results for the bulk, surface, and mineral–water interface properties of the well-characterized minerals corundum, hematite, and goethite. Good results are now obtainable using this level of theory, and further improvements in accuracy and efficiency of these methods will make these methods even more accessible to geochemical researchers in the future. While density functional calculations provide remarkable structural predictions, they can be qualitatively incorrect in predicting properties such as band gaps, conductivity, and spin states. These problems are partially treated by adding exact exchange. However, the inclusion of exchange in this way is not justifiable and is an ad hoc correction. To make further progress along these lines, other higher-level methods that go beyond DFT-based methods, such as dynamical mean field theory (DMFT), GW, and variational and diffusion Monte Carlo methods, will need to be carefully tested on geochemical systems. Besides the well-known issues of modern electronic structure theory, certainly the most important unsolved problem in simulation technology is the inability of present methods to efficiently search phase space and identify reaction mechanisms. This is an especially difficult problem for first principles simulation methods because the typical time scales of these processes far exceed the time scales that can be achieve even by CMD. This means that much more efficient sampling methods for identifying relevant structure and mechanisms must be developed. Lastly, it is important to recognize that the new computers that are appearing are based on the inclusion of many processor and fast communication. The development of algorithms that will capture the performance of these emerging computers is a difficult problem. To continue to make progress in the use of first principles methods for complex geochemical systems, scalability to very large numbers of processor cores will be needed. This will not only require novel software development and performance analysis tools but also a reevaluation of mathematical and algorithmic approaches.

Acknowledgments This research was supported by the BES Geosciences program and the BES Heavy element program of the US Department of Energy, Office of Science—DE-AC06-76RLO 1830. Additional support from ASCR petascale tools program and EMSL operations. EMSL operations are supported by the DOE’s Office of Biological and Environmental Research. We wish to thank the Scientific Computing Staff, Office of Energy Research, and the US Department of Energy for a grant of computer time at the National Energy Research Scientific Computing Center (Berkeley, CA). Some of the calculations were performed on the Chinook and Cascade computing systems at the Molecular Science Computing Facility in the William R. Wiley Environmental Molecular Sciences Laboratory (EMSL) at PNNL.

Appendix In the following sections, we discuss briefly some of the details of the approximations that are necessary to define plane wave DFT. This discussion supports the discussions of accuracy that we report in Sections 4.2–4.4 in the main text.

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

A.1

135

Short Introduction to Pseudopotentials

The variation/strength of the atom center electron potential, Vext, must be reduced in order for KS wave functions to be expanded in a reasonable number of terms. In order to do this, pseudopotentials have been developed (Pickett 1989). These are widely used; nevertheless, it is typical to have to modify these functions to obtain the accuracy required for a particular calculation. A significant advance in the development of the pseudopotential method was made by Hamann, Schluter, and Chiang (HSC) (Hamann et al. 1979) with the introduction of norm-conserving pseudopotentials. While there are differences between various approaches, all popular pseudopotentials adopt the basic prescription of HSC. These methods have been highly developed in the condensed matter community and are well explained and reviewed (see, e.g., the detailed review of Pickett (1989) and the original papers cited therein). The basic idea of pseudopotential is that the core region of the atomic potential is replaced by a much slower varying function designed to specifically reproduce the behavior of the valence wave functions in regions outside the core (presumed to be the bonding region). The smoothed potential has a nodeless solution that can be expanded by a smaller plane wave basis. It can be shown that with proper care, replacing the atomic potential with a pseudopotential will produce the same solutions beyond the region of replacement while also maintaining the normalization of the orbital function. Pseudopotentials are derived from first principles single-atom DFT calculations at the same level of approximation (GGA or hybrid exchange) as used in the full many-atom condensed system simulation. These potentials are precomputed before use in the condensed matter calculation so the simulation remains parameter-free (no parameters adjusted in the simulation). On the other hand, there are issues such as the contribution of the atomic valence structure to the bonding that must be decided before the development of the pseudopotential. These choices will determine the transferability of the pseudopotential from the atomic to the condensed environment. Such issues are carefully discussed in the Picket review. In the calculations reported in the next section, the effects of various assumptions will be tested by comparison of calculations to bulk structural observations. In the HSC approach, given the selection of valence orbitals (corresponding to various l to be included in the active space), a pseudopotential for each total angular momentum is found from the direct inversion of the Schrödinger equation (with a selected DFT functional; see Hamann 1989). This produces a nonlocal pseudopotential of the form ∗ r Ylm r Vl r δ r −r Ylm

val val r + V ps r,r = VM r + V pseud = VM l, m

val r is the Coulomb and exchange potential due to the (nonactive) valence electrons, Ylm r where VM is the spherical harmonic defined by the angular momentum, l, and magnetic quantum, m, numbers, r is a unit vector in the r direction, and Vl(r) is the radial potential found from the inversion of the DFT solution to the radial Schrödinger equation for the equivalent atomic problem (see HSC (Hamann et al. 1979, Hamann 1989)). The operator V ps r,r acts on function of r as ∗ Ylm r Vl r δ r −r Ylm r ψ r dr

V ps ψ r = l, m

This potential has a semilocal form, neither just local (radial) nor fully separable (see KB (Bylander and Kleinman 1984)). In this semilocal form, the pseudopotential is computationally difficult to calculate with a plane wave basis set, because the kernel integration is not separable in r and r (see KB (Bylander and Kleinman 1984)). To produce a more efficient calculation while retaining as much of the atomic form as possible, Kleinman and Bylander approximated the form by

136

Molecular Modeling of Geochemical Reactions KB

Plm r hl P∗lm r

V ps r,r = Vlocal r + l, m

where the atom-centered projectors Plm(r) are of the form Plm r = Vl r −Vlocal r φl r Ylm r and the coefficient hl is −1

∞

hl =

φl r Vl r −Vlocal r φl r r dr 2

4π 0

where φl r are the zero radial node pseudo-wave functions of the potentials, Vl(r), calculated in the KB

atomic environment. Note that V ps φl Ylm = Vl φl Ylm , that is, that the fully nonlocal KB form preserves the form of the potential in the atomic problem. The choice of the local potential Vlocal(r) is somewhat arbitrary, but for transition metals it is often chosen to be the Vl = 0 r potential. A larger series expansion in pseudo-wave functions can be used to improve the fully local description of the semilocal form. This leads to the general form Pnlm r hnl , n P∗n lm r

V ps r,r = Vlocal r + l , m n, n

There is a large body of literature describing pseudopotential methods and illustrating the accuracy and efficiency of this approach (Pickett 1989). For use in the structural calculations, the pseudopotentials are developed entirely from fitting atomic calculations and, therefore, should not be considered as part of the data fitting process. Nevertheless, there are questions about accuracy of the pseudopotential approximation, for example, how many φnl r are required to accurately represent the valence structure of the condensed system and how much of the unscreened atomic potential is assigned as the core region (roughly speaking the region removed). This is a function of the atomic structure of the particular element (e.g., separation of the highest-filled lowest excited states, highest valence states, etc.). In this work we found that including the 3s and 3p functions in the active space of the pseudopotential of the Fe3+ and Cr3+ ions considerably improved the agreement with the scattering data. The default pseudopotential included only the 3d orbitals. Additional issues that have to be considered in the pseudopotential representation include the functional form for the Vl(r) potentials selected and the evaluation of the parameters in these potentials by comparison to atomic calculations at the same level of electronic structure calculation. The radius of the region of replacement of the external with the pseudopotential form determines to a large extent the smoothness of the pseudopotential (the larger the region, the smoother the pseudopotential). However, if this region is too large, the bond formation will be affected and the pseudopotential representation will produce incorrect bonding results. An example of the derived smooth pseudopotential and nodeless pseudo-wave functions is given in Figure A.1 for the Fe3+ ion: σ nelc nion

Epsp = σ= , i=1 I =1

ψ iσ

I Vlocal

ψ iσ

I lmax

l

I nmax n Imax

+ l = 0 m = −l n = 1 n = 1

I hIl , n, n PnI lm ψ iσ ψ iσ Pnlm

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

137

1.5 rΨspsp(r)

rΨl(r)

1

rΨppsp(r) rΨdpsp(r)

0.5

rΨsAE(r)

0

rΨpAE(r)

–0.5

rΨdAE(r)

–1

0

0.5

1

1.5

2

2.5

3

Vl(r) (Hartrees)

0 Vs(r) Vp(r) Vd(r)

–10 –20 –30 –40

0

1

0.5

1.5

2

2.5

3

r (Bohr)

Figure A.1 Comparison of the pseudo-wave functions (dashed lines) with the full-core atomic valence wave functions (solid lines) for Fe3+. The lower panel shows the corresponding pseudopotentials.

In the next section, we will illustrate the accuracy of the various choices discussed previously by direct application to the calculation of observed bulk properties of three minerals: goethite (FeOOH), hematite (Fe2O3), and corundum (Al2O3). A.1.1

The Spin Penalty Pseudopotential

As discussed in the text, in order to reliably calculate spin-ordered systems in DFT, it is often necessary to develop a spin-ordered initial state prior to full optimization or dynamical simulation. In fact there may be a number of competing spin orderings in the unit cell (see Figure 4.1). A convenient way to generate these states is to introduce terms in the external potential (e.g., a pseudopotential) that will stabilize the electronic wave function at a selected set of sites, that is, add a spin penalty function to the pseudopotential, This will break the symmetry of the wave function and create spin localization. If the state is an approximation to the DFT spin eigenfunctions, further optimization with the spin penalty turned off will lead to a spin-ordered DFT solution. The expectation of the spin penalty energy, Epsp + penalty , is σ nelc nion

I ψ iσ Vlocal ψ iσ

Epsp + penalty = σ= , i=1 I =1 I lmax

l

I nmax n Imax

+ l = 0 m = −l n = 1 n = 1

1 −δl, l σ δI , ionlist ξ σ − 1

I hIl , n, n PnI lm ψ iσ ψ iσ Pnlm

138

A.1.2

Molecular Modeling of Geochemical Reactions

Projected Density of States from Pseudo-Atomic Orbitals

The projected density of states (PDOS) are calculated from ρI ε =

2

φI , i ψ i

δ ε− εi

i, i

where I is the atom index for the PDOS, n is the index of valence states in the system, and φI , i is the ith pseudo-atomic orbital centered in the Ith atom. The pseudo-atomic orbital is generated by solving Kohn–Sham Equation 4.3 using pseudopotential approach to single atoms.

A.2

Hubbard-Like Coulomb and Exchange (DFT+U)

In DFT framework, most exchange–correlation functionals are generated from expansions around homogeneous electron gas limit. When DFT is used to model these systems that have localized electrons, the predicted electronic states can be significantly away from the localized states. DFT+U is inspired by the Hubbard model, which can be used to model strongly correlated systems. It adds a Hubbard term (on-site Coulomb and exchange) to the DFT functional for strongly correlated electrons (d, f electrons…) and uses a regular DFT functional on other valence electrons. The Hubbard term is expressed using Slater integrals and local occupation matrix: EHubbard =

1 2 +

,σ ρImn

I , σ m, n, x, y

1 2

, σ I , −σ I I ρImn ρxy χ m χ x Vee χ nI χ yI

I , σ m , n , x, y

I fk χ m φkσ

=

,σ I,σ ρImn ρxy

I I I I χm χ x Vee χ nI χ yI − χ m χ n Vee χ xI χ yI

φkσ χ nI

k

However, the double-counting problem appears, because the Hartree term and exchange– correlation term in DFT functional both include some fragments of Coulomb interactions. The double-counting term has been derived and been subtracted from the total energy functional: EDFT + U = EDFT + EHubbard 1 EDC = U 2

NσI N −I σ + I

,σ ρImm

ρI , σ

−EDC

1 U −J 2

I

σ

NσI N −I σ −1

N is the trace of local occupation matrix. U and J are the average Coulomb and exchange parameters. There are different approaches to calculate those two numbers. Dudarev et al. (1998) used those two average parameters to replace Slater integrals in equation, and DFT+U functionals have been simplified. In this approach, only (U − J is meaningful: EDFT + U = EDFT +

1 U −J 2

I,σ

ρIjj, σ − j

ρIjl, σ ρIlj, σ j, l

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

139

I There are different choices of projectors χ m , which can be used in the calculation (i.e., projectors from pseudopotential or PAW), and the number of projectors for each calculation is limited, so the cost of DFT+U method is approximately equivalent to DFT methods which is much cheaper than adding exact exchange.

A.3

Overview of the PAW Method

The main idea in the PAW method (Blochl 1994) is to project out the high-frequency components of the wave function in the atomic sphere region. Effectively this splits the original wave function into two parts: ψn r = ψn r +

ψ nI r I

The first part ψ n r is smooth and can be represented using a plane wave basis set of practical size. The second term is localized with the atomic spheres and is represented on radial grids centered on the atoms as ψ nI r =

I φαI r − φIα r cnα

α

I where the coefficients cnα are given by I cnα = pIα ψ n

This decomposition can be expressed using an invertible linear transformation, T, which relates the stiff one-electron wave functions ψ n to a set of smooth one-electron wave functions ψ n : ψ n = Tψ n ψ n = T − 1ψ n which can be represented by fairly small plane wave basis. The transformation T is defined using a local PAW basis, which consists of atomic orbitals, φαI r , smooth atomic orbitals, φIα r which coincide with the atomic orbitals outside a defined atomic sphere, and projector functions, pαI r , where I is the atomic index and α is the orbital index. The projector functions are constructed such that they are localized within the defined atomic sphere and in addition are orthonormal to the atomic orbitals. Blochl defined the invertible linear transformations by φIα − φαI

T =1+ α

I

T −1 = 1 + I

pIα =

α

pI φ I β

φαI − φIα −1 αβ

pβI

pαI pIα

140

Molecular Modeling of Geochemical Reactions

The main effect of the PAW transformation is that the fast variations of the valence wave function in the atomic sphere region are projected out using local basis set, thereby producing a smoothly varying wave function that may be expanded in a plane wave basis set of a manageable size. In order for a PAW calculation to be manageable, it is important for the PAW basis not to be too large and for the smooth atomic orbitals and smooth projector functions to be adequately described by fairly small plane wave basis. However, for the PAW calculation to be accurate, the basis must also accurately describe the regions near the atomic centers. Procedures used to determine the PAW basis can be found in several places, and for the most part these procedures are the same as the ones used to generate Vanderbilt pseudopotentials (Vanderbilt 1990), with additional Gram–Schmidt steps to enforce the orthonormality relations. The expression for the total energy in PAW method can be separated into the following 15 terms: EPAW =E kinetic − pw +E vlocal − pw +E Coulomb − pw +E xc− pw +Eion− ion +Ekinetic − atom +Elocal − atom +Exc− atom + Ecmp − vloc + EHartree −atom + Ecmp −cmp + Ecmp − pw + Evalence − core + Ekinetic −core + Eion− core The first five terms are essentially the same as for a standard pseudopotential plane wave program, minus the nonlocal pseudopotential, where E kinetic −pw = i

E Coulomb − pw = E xc −pw =

Eion −ion =

1 2Ω G −

ε π

4π G

0

2

ZI2 − I

G2

e − i 4ε

G

Ω 2G

4π G

0

Ω N1 N2 N3

ρ∗ G ρ G

2

ρ r ϵxc ρ r r

ZI e −iG∙RI ZJ e − iG∙RJ + I,J

π 2ε2 Ω

G2 ∗ ψ G ψ G 2

1 2

ZI ZJ a I , J RI − RJ + a

erf ε RI −RJ + a RI −RJ + a

2

ZI I

ρ G Vlocal G

E vlocal −pw = G

The local potential in the E vlocal − pw term is the Fourier transform of Vlocal r = −

ZI I

erf r −RI σ I I r −RI + vps r −RI

It turns out that for many atoms σ I needs to be fairly small. This results in Vlocal(r) being stiff. However, since in the aforementioned integral this function is multiplied by a smooth density ρ G , the expansion of Vlocal(G) only needs to be the same as the smooth density. The auxiliary pseudopotenI r −RI is defined to be localized within the atomic sphere and is introduced to remove ghost tial vps states due to local basis set incompleteness.

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

141

The next four terms are atomic based and they essentially take into account the difference between the true valence wave functions and the pseudo-wave functions: I ψ i pIα tatom

αβ

pIβ ψ i

I ψ i pIα uatom

αβ

pIβ ψ i

Ekinetic −atom = αβ

i

I

Elocal −atom = αβ

i

I I rcut

Exc− atom =

r2 ρ I r, θ,φ ϵxc ρ I r,θ, φ −ρI r,θ, φ ϵxc ρI r, θ,φ

wθφ θ, φ

I

dr

0

I Watom

EHartree − atom = I

1 = 2

I

i

ψ i pIα

αβ

pIβ ψ i

ψ j pIμ

μν

j

pIν ψ j

lm I τlm lα mα , lβ mβ τlμ mμ , lν mν VHeff lm

l αβμν

The next three terms are the terms containing the compensation charge densities: ρcmp G V local G + ρcmp G

Ecmp − vloc =

Vlocal G −V local G

G

ρcmp r − ρcmp r

+

4π

Ecmp − cmp = Ω G 0

+

1 2

G

2

G 0

1 ρcmp G ρcmp G − ρcmp G ρcmp G 2

ρcmp r −ρcmp r ρcmp r − ρcmp r drdr r −r 4π

Ecmp − pw = Ω

Vlocal r −V local r dr

G

2

ρcmp G ρ G

In the first two formulas, the first terms are computed using plane waves and the second terms are computed using Gaussian two-center integrals. The smooth local potential in the Ecmp−vloc term is the Fourier transform of V local r = −

ZI I

erf r −RI σ I r − RI

The stiff and smooth compensation charge densities in the aforementioned formula are I σI Qlm glm r −RI

ρcmp r = I

lm

I

lm

I σI Qlm glm r −RI

ρcmp r =

142

Molecular Modeling of Geochemical Reactions

where ψ i pIα

I = Qlm αβ

i

l

I pIβ ψ i τlm lα mα , lβ mβ qcomp

αβ

The decay parameter σ I is defined the same as aforementioned, and σ I is defined to be smooth enough in order that ρcmp r and V local r can readily be expanded in terms of plane waves. The final three terms are the energies that contain the core densities: ψ i pIα Vvalence −core Iαβ pIβ ψ i

Evalence − core = i

αβ

I

αβ

I

∞

φnI c lc r

Ekinetic − core = c

φnI c lc r

− + lc lc + 1

φnI c lc r φnI c lc r r2

dr

0

ρcI r ρcI r ρI r drdr − c ZI + ZIcore r −r r

1 2

Eion− core = I

The matrix elements contained in the aforementioned formula are

I tatom αβ

δm m δl l = α β αβ 2

I rcut

φnI α lα r

− φInα lα r

φnI β lβ r

φInβ lβ r

0

+ lα lα + 1 I uatom

αβ

=

ZI VI 4π comp

φnI α lα r φnI β lβ r − φInα lα r φInβ lβ r r2 l=0

+

αβ

2ZI I qcomp 2πσ I

I rcut

+ δmα mβ δlα lβ

l=0 αβ

−ZI r

φnI α lα r φnI β lβ r

dr

+ φInα lα r φInβ lβ r

I −vps r

0 I VHeff

l αβμν

= VHI

l −2 αβμν

l VHI αβμν

I Vcomp

4π = 2l + 1

l αβ

I I rcut rcut

0

I qcomp

rl< rl>+ 1

0

l μν

− vgI

l

I qcomp

l αβ

I qcomp

φnα lα r φnβ lβ r φnμ lμ r φnν lν r

−φnα lα r φnβ lβ r φnμ lμ r φnν lν r drdr I Vcomp

4π = αβ 2l + 1 l

I I rcut rcut

φnα lα r φnβ lβ r 0

0

rl< g I r r 2 drdr rl>+ 1 l

l μν

dr

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

I qcomp

∞

l αβ

143

r l φnα lα r φnβ lβ r − φnα lα r φnβ lβ r dr

= 0

vgI

l

=

4 2π 2l + 1 2l + 1 !!σ I2l + 1

2π π

τlm lα mα , lβ mβ

Tlm θ,ϕ Tlα mα θ, ϕ Tlβ mβ θ,ϕ sin θdθdϕ

= 0 0

References Adamo, C. and V. Barone (1999). “Toward reliable density functional methods without adjustable parameters: The PBE0 model.” Journal of Chemical Physics 110(13): 6158–6170. Allen, M. P. and D. Tildesley (1987). Computer Simulation of Liquids. Oxford: Clarendon Press. Alvarez, M., E. E. Sileo and E. H. Rueda (2008). “Structure and reactivity of synthetic Co-substituted goethites.” American Mineralogist 93: 584–590. Anisimov, V. I., I. V. Solovyev, M. A. Korotin, M. T. Czyżyk and G. A. Sawatzky (1993). “Density-functional theory and NiO photoemission spectra.” Physical Review B 48(23): 16929–16934. Anovitz, L. M., D. R. Cole, G. Rother, J. W. Valley and A. Jackson (2010). “Analysis of nano-porosity in the St. Peter Sandstone using (ultra) small angle neutron scattering.” Geochimica et Cosmochimica Acta 74(12): A26. Anovitz, L. M., G. W. Lynn, D. R. Cole, G. Rother, L. F. Allard, W. A. Hamilton, L. Porcar and M. H. Kim (2009). “A new approach to quantification of metamorphism using ultra-small and small angle neutron scattering.” Geochimica et Cosmochimica Acta 73(24): 7303–7324. Ashcroft, N. W. and N. D. Mermin (1976). Solid State Physics. New York: Holt, Rinehart and Winston. Atta-Fynn, R., E. J. Bylaska and W. A. de Jong (2013). “Importance of counteranions on the hydration structure of the curium ion.” Journal of Physical Chemistry Letters 4(13): 2166–2170. Becke, A. D. (2014). “Perspective: Fifty years of density-functional theory in chemical physics.” Journal of Chemical Physics 140(18): 18A301. Becker, U., M. F. Hochella and E. Apra (1996). “The electronic structure of hematite{001} surfaces: Applications to the interpretation of STM images and heterogeneous surface reactions.” American Mineralogist 81(11–12): 1301–1314. Blake, R. L., R. E. Hessevick, T. Zoltai and L. W. Finger (1966). “Refinement of the hematite structure.” American Mineralogist 51(1966): 123–129. Blochl, P. E. (1994). “Projector augmented-wave method.” Physical Review B 50(24): 17953–17979. Bocquet, S. and S. J. Kennedy (1992). “The Néel temperature of fine particle goethite.” Journal of Magnetism and Magnetic Materials 109: 260–264. Bogatko, S. A., E. J. Bylaska and J. H. Weare (2010). “First principles simulation of the bonding, vibrational, and electronic properties of the hydration shells of the high-spin Fe3+ ion in aqueous solutions.” The Journal of Physical Chemistry A 114(5): 2189–2200. Boily, J.-F. (2012). “Water structure and hydrogen bonding at goethite/water interfaces: Implications for proton affinities.” The Journal of Physical Chemistry C 116(7): 4714–4724. Brown, G. E. (2001). “Surface science—how minerals react with water.” Science 294(5540): 67–69. Brown, G. E., V. E. Henrich, W. H. Casey, D. L. Clark, C. Eggleston, A. Felmy, D. W. Goodman, M. Gratzel, G. Maciel, M. I. McCarthy, K. H. Nealson, D. A. Sverjensky, M. F. Toney and J. M. Zachara (1999). “Metal oxide surfaces and their interactions with aqueous solutions and microbial organisms.” Chemical Reviews 99(1): 77–174. Brown, G. E. and N. C. Sturchio (2002). “An overview of synchrotron radiation applications to low temperature geochemistry and environmental science.” In Fenter, P. A., Rivers, M. L., Sturchio, N. C. and Sutton,

144

Molecular Modeling of Geochemical Reactions

S. R. (Eds.) Applications of Synchrotron Radiation in Low-Temperature Geochemistry and Environmental Sciences, Vol. 49. Washington, DC: Mineralogical Society of America, pp. 1–115. Brown, I. D. (2009). “Recent developments in the methods and application of the bond valence.” Chemical Reviews 109: 6858–6919. Brown, I. D. (2014). “Bond valence theory.” In Brown, I. D. and Poeppelmeier, K. R. (Eds.) Bond Valences, Vol. 158. Heidelberg: Springer, pp. 11–58. Brown, I. D. and R. D. Shannon (1973). “Empirical bond-strength bond-length curves for oxides.” Acta Crystallographica. Section A 29(3): 266–282. Burke, K. (2012). “Perspective on density functional theory.” Journal of Chemical Physics 136(15): 150901. Bylander, D. M. and L. Kleinman (1984). “Outer-core electron and valence electron pseudopotential.” Physical Review B 29(4): 2274–2276. Bylaska, E. J., A. J. Salter-Blanc and P. G. Tratnyek (2011a). “One-electron reduction potentials from chemical structure theory calculations.” In Tratnyek, P. G., Grundl, T. J. and Haderlein, S. B. (Eds.) Aquatic Redox Chemistry, Vol. 1071. Washington, DC: American Chemical Society, pp. 37–64. Bylaska, E. J., K. Tsemekhman, N. Govind and M. Valiev (2011b). “Large-scale plane-wave-based density functional theory: Formalism, parallelization, and applications.” In Reimers, J. R. (Ed.) Computational Methods for Large Systems: Electronic Structure Approaches for Biotechnology and Nanotechnology. Hoboken, NJ: John Wiley & Sons, Inc., pp. 77–116. Bylaska, E. J., M. Valiev, R. Kawai and J. H. Weare (2002). “Parallel implementation of the projector augmented plane wave method for charged systems.” Computer Physics Communications 143(1): 11–28. Bylaska, E. J., M. Valiev and J. H. Weare (2001). “Comparing the efficiency and accuracy of the projector augmented wave DFT method against Gaussian based DFT methods.” Abstracts of Papers of the American Chemical Society 221: U435. Caban-Acevedo, M., N. S. Kaiser, C. R. English, D. Liang, B. J. Thompson, H. E. Chen, K. J. Czech, J. C. Wright, R. J. Hamers and S. Jin (2014). “Ionization of high-density deep donor defect states explains the low photovoltage of iron pyrite single crystals.” Journal of the American Chemical Society 136(49): 17163–17179. Car, R. and M. Parrinello (1985). “Unified approach for molecular-dynamics and density-functional theory.” Physical Review Letters 55(22): 2471–2474. Catalano, J. G. (2011). “Weak interfacial water ordering on isostructural hematite and corundum (001) surfaces.” Geochimica et Cosmochimica Acta 75(8): 2062–2071. Catalano, J. G., P. Fenter and C. Park (2007). “Interfacial water structure on the (012) surface of hematite: Ordering and reactivity in comparison with corundum.” Geochimica et Cosmochimica Acta 71(22): 5313–5324. Catalano, J. G., P. Fenter, C. Park, K. M. Rosso, A. J. Frierdich and B. T. Otemuyiwa (2010a). “Fe(II)-induced structural transformations of hematite surfaces and their impact on contaminants.” Geochimica et Cosmochimica Acta 74(12): A150. Catalano, J. G., P. Fenter, C. Park, Z. Zhang and K. M. Rosso (2010b). “Structure and oxidation state of hematite surfaces reacted with aqueous Fe(II) at acidic and neutral pH.” Geochimica et Cosmochimica Acta 74(5): 1498–1512. Catlow, C. R. A., R. James, W. C. Mackrodt and R. F. Stewart (1982). “Defect energetics in alpha-Al2O3 and rutile TiO2.” Physical Review B 25(2): 1006–1026. Chen, Y. (2015). Ab initio study of structure and dynamics of bulk, surface and the mineral/aqueous fluid interface regions. University of California, San Diego, CA. Ciccacci, F., L. Braicovich, E. Puppin and E. Vescovo (1991). “Empty electron states in Fe2O3 by ultraviolet inverse-photoemission spectroscopy.” Physical Review B 44(19): 10444–10448. Coey, Y. M. D. and G. A. Sawatzky (1971). “A study of hyperfine interactions in the system (Fe1–xRhx)2O3 using the Mossbauer effect (Bonding parameters).” Journal of Physics C: Solid State Physics 4: 2386. Cornell, R. M. S. U. (2003). The Iron Oxides: Structure, Properties, Reactions, Occurrences, and Uses. Weinheim: Wiley-VCH. Cox, P. A. (1992). Transition Metal Oxides. Oxford: Clarendon Press. Cwiertny, D. M., R. M. Handler, M. V. Schaefer, V. H. Grassian and M. M. Scherer (2008). “Interpreting nanoscale size-effects in aggregated Fe-oxide suspensions: Reaction of Fe(II) with goethite.” Geochimica et Cosmochimica Acta 72(5): 1365–1380. Davis, J. A. and D. B. Kent (1990). “Surface complexation modeling in aqueous geochemistry.” Reviews in Mineralogy and Geochemistry 23: 177–260.

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

145

DeMichelis, R., B. Civalleri, M. Ferrabone and R. Dovesi (2010). “On the performance of eleven DFT functionals in the description of the vibrational properties of aluminosilicates.” International Journal of Quantum Chemistry 110: 406–415. DePaolo, D. J. and F. M. Orr (2008). “Geoscience research for our energy future.” Physics Today 61(8): 46–51. Dräger, G., W. Czolbe and J. A. Leiro (1992). “High-energy-spectroscopy studies of a charge-transfer insulator: X-ray spectra of alpha-Fe2O3.” Physical Review B 45(15): 8283–8287. Dudarev, S. L., G. A. Botton, S. Y. Savrasov, C. J. Humphreys and A. P. Sutton (1998). “Electron-energy-loss spectra and the structural stability of nickel oxide: An LSDA+U study.” Physical Review B 57(3): 1505–1509. Dzombak, D. A. (1990). Surface Complexation Modeling: Hydrous Ferric Oxide. New York: John Wiley & Sons. Esler, K. P., R. E. Cohen, B. Militzer, J. Kim, R. J. Needs and M. D. Towler (2010). “Fundamental high-pressure calibration from all-electron quantum Monte Carlo calculations.” Physical Review Letters 104(18): 185702. Fenter, P., S. Kerisit, P. Raiteri and J. D. Gale (2013). “Is the calcite–water interface understood? Direct comparisons of molecular dynamics simulations with specular X-ray reflectivity data.” The Journal of Physical Chemistry C 117(10): 5028–5042. Fenter, P., S. S. Lee, C. Park, J. G. Catalano, Z. Zhang and N. C. Sturchio (2010a). “Imaging interfacial topography and reactivity with X-rays.” Geochimica et Cosmochimica Acta 74(12): A287. Fenter, P., S. S. Lee, C. Park, J. G. Catalano, Z. Zhang and N. C. Sturchio (2010b). “Probing interfacial reactions with X-ray reflectivity and X-ray reflection interface microscopy: Influence of NaCl on the dissolution of orthoclase at pOH 2 and 85 C.” Geochimica et Cosmochimica Acta 74(12): 3396–3411. Fenter, P. and N. C. Sturchio (2004). “Mineral-water interfacial structures revealed by synchrotron X-ray scattering.” Progress in Surface Science 77(5–8): 171–258. Fenter, P. A. (2002). “X-ray reflectivity as a probe of mineral-fluid interfaces: A user guide.” In Fenter, P. A., Rivers, M. L., Sturchio, N. C. and Sutton, S. R. (Eds.) Applications of Synchrotron Radiation in LowTemperature Geochemistry and Environmental Sciences, Vol. 49. Washington, DC: Mineralogical Society of America, pp. 149–220. Finger, L. W. and R. M. Hazen (1980). “Crystal structure and isothermal compression of Fe2O3, Cr2O3, and V2O3 to 50 kbars.” Journal of Applied Physics 51(10): 5362–5367. Fujimori, A., M. Saeki, N. Kimizuka, M. Taniguchi and S. Suga (1986). “Photoemission satellites and electronic structure of Fe2O3.” Physical Review B 34(10): 7318–7328. Fulton, J. L., E. J. Bylaska, S. Bogatko, M. Balasubramanian, E. Cauet, G. K. Schenter and J. H. Weare (2012). “Near-quantitative agreement of model-free DFT-MD predictions with XAFS observations of the hydration structure of highly charged transition-metal ions.” Journal of Physical Chemistry Letters 3(18): 2588–2593. Fulton, J. L., S. M. Kathmann, G. K. Schenter, E. J. Bylaska, S. A. Bogatko and J. H. Weare (2010). “XAFS spectroscopy and molecular dynamics: Aqueous ions and ion pairs under non-ideal conditions.” Geochimica et Cosmochimica Acta 74(12): A311. Ghose, S. K., G. A. Waychunas, T. P. Trainor and P. J. Eng (2010). “Hydrated goethite (alpha-FeOOH) (100) interface structure: Ordered water and surface functional groups.” Geochimica et Cosmochimica Acta 74(7): 1943–1953. Gorski, C. A. and M. M. Scherer (2008). “ENVR 87-Influence of magnetite stoichiometry on Fe(II) uptake and nitroaromatic reduction.” Abstracts of Papers of the American Chemical Society 236: 87-ENVR. Gorski, C. A. and M. M. Scherer (2009). “Influence of magnetite stoichiometry on Fe-II uptake and nitrobenzene reduction.” Environmental Science & Technology 43(10): 3675–3680. Gorski, C. A. and M. M. Scherer (2011). “Fe2+ sorption at the fe oxide-water interface: A revised conceptual framework.” In Tratnyek, P. G., Grundl, T. J. and Haderlein, S. B. (Eds.) Aquatic Redox Chemistry, Vol. 1071. Washington, DC: American Chemical Society, pp. 315–343. Guenard, P., G. Renaud, A. Barbier and M. Gautier-Soyer (1996). “Determination of the α-Al2O3 (0001) surface relaxation and termination by measurements of crystal truncation rods.” MRS Online Proceedings Library 437: 15–20. Guenard, P., G. Renaud, A. Barbier and M. Gautier-Soyer (1998). “Determination of the alpha-Al2O3 (0001) surface relaxation and termination by measurements of crystal truncation rods.” Surface Review and Letters 5(1): 321–324. Hamann, D. R. (1989). “Generalized norm-conserving pseudopotentials.” Physical Review B 40(5): 2980–2987. Hamann, D. R., M. Schluter and C. Chiang (1979). “Norm-conserving pseudopotentials.” Physical Review Letters 43(20): 1494–1497.

146

Molecular Modeling of Geochemical Reactions

Herbstein, F. H. (2000). “How precise are measurements of unit-cell dimensions from single crystals?” Acta Crystallographica. Section B 56(4): 547–557. Hochella, M. F. (1990). “Atomic-structure, microtopography, composition, and reactivity of mineral surfaces.” Reviews in Mineralogy 23: 87–132. Hochella, M. F., S. K. Lower, P. A. Maurice, R. L. Penn, N. Sahai, D. L. Sparks and B. S. Twining (2008). “Nanominerals, mineral nanoparticles, and earth systems.” Science 319(5870): 1631–1635. Hohenberg, P. and W. Kohn (1964). “Inhomogeneous electron gas.” Physical Review B 136(3B): B864. Holzwarth, N. A. W., G. E. Matthews, R. B. Dunning, A. R. Tackett and Y. Zeng (1997). “Comparison of the projector augmented-wave, pseudopotential, and linearized augmented-plane-wave formalisms for densityfunctional calculations of solids.” Physical Review B 55(4): 2005–2017. Huang, P., T. A. Pham, G. Galli and E. Schwegler (2014). “Alumina(0001)/water interface: Structural properties and infrared spectra from first-principles molecular dynamics simulations.” The Journal of Physical Chemistry C 118(17): 8944–8951. Huda, M. N., A. Walsh, Y. F. Yan, S. H. Wei and M. M. Al-Jassim (2010). “Electronic, structural, and magnetic effects of 3d transition metals in hematite.” Journal of Applied Physics 107(12): 123712. Katz, J. E., X. Y. Zhang, K. Attenkofer, K. W. Chapman, C. Frandsen, P. Zarzycki, K. M. Rosso, R. W. Falcone, G. A. Waychunas and B. Gilbert (2012). “Electron small polarons and their mobility in iron (oxyhydr)oxide nanoparticles.” Science 337(6099): 1200–1203. Katz, L. E. and K. F. Hayes (1995). “Surface complexation modeling. I. Strategy for modeling monomer complex formation at moderate surface coverage.” Journal of Colloid and Interface Science 170(2): 477–490. Kawai, R. and J. H. Weare (1990). “From van der Waals to metallic bonding—the growth of Be clusters.” Physical Review Letters 65(1): 80–83. Kerisit, S. (2011). “Water structure at hematite-water interfaces.” Geochimica et Cosmochimica Acta 75(8): 2043–2061. Kerisit, S. and K. M. Rosso (2006). “Computer simulation of electron transfer at hematite surfaces.” Geochimica et Cosmochimica Acta 70(8): 1888–1903. Kerisit, S., J. H. Weare and A. R. Felmy (2012). “Structure and dynamics of forsterite-scCO(2)/H2O interfaces as a function of water content.” Geochimica et Cosmochimica Acta 84: 137–151. Khein, A., D. J. Singh and C. J. Umrigar (1995). “All-electron study of gradient corrections to the local-density functional in metallic systems.” Physical Review B 51(7): 4105–4109. Kirfel, A. and K. Eichhorn (1990). “Accurate structure analysis with synchrotron radiation. The electron density in Al2O3 and Cu2O.” Acta Crystallographica Section A 46: 271–284. Kohanoff, J. (1994). “Phonon spectra from short non-thermally equilibrated molecular dynamics simulations.” Computational Materials Science 2(2): 221–232. Kohn, W. and L. J. Sham (1965). “Self-consistent equations including exchange and correlation effects.” Physical Review 140(4A): 1133. Kotliar, G. and D. Vollhardt (2004). “Strongly correlated materials: Insights from dynamical mean-field theory.” Physics Today 57(3): 53–59. Kresse, G. and D. Joubert (1999). “From ultrasoft pseudopotentials to the projector augmented-wave method.” Physical Review B 59(3): 1758–1775. Kubicki, J. D., K. W. Paul and D. L. Sparks (2008). “Periodic density functional theory calculations of bulk and the (010) surface of goethite.” Geochemical Transactions 9: 4. Kumar, N., P. R. C. Kent, D. J. Wesolowski and J. D. Kubicki (2013). “Modeling water adsorption on rutile (110) using van der Waals density functional and DFT+U methods.” The Journal of Physical Chemistry C 117(45): 23638–23644. Lad, R. J. and V. E. Henrich (1989). “Photoemission study of the valence-band electronic structure in FexO, Fe3O4, and alpha-Fe2O3 single crystals.” Physical Review B 39(18): 13478–13485. Larese-Casanova, P. and M. M. Scherer (2007). “Fe(II) sorption on hematite: New insights based on spectroscopic measurements.” Environmental Science & Technology 41(2): 471–477. Latta, D. E., C. A. Gorski, M. I. Boyanov, E. J. O’Loughlin, K. M. Kemner and M. M. Scherer (2012). “Influence of magnetite stoichiometry on U-VI reduction.” Environmental Science & Technology 46(2): 778–786. Lazic, P., R. Armiento, F. W. Herbert, R. Chakraborty, R. Sun, M. K. Y. Chan, K. Hartman, T. Buonassisi, B. Yildiz and G. Ceder (2013). “Low intensity conduction states in FeS2: Implications for absorption, opencircuit voltage and surface recombination.” Journal of Physics. Condensed Matter 25(46): 465801.

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

147

Liechtenstein, A. I., V. I. Anisimov and J. Zaanen (1995). “Density-functional theory and strong interactions: Orbital ordering in Mott-Hubbard insulators.” Physical Review B 52(8): R5467–R5470. Liu, L.-M., C. Zhang, G. Thornton and A. Michaelides (2010). “Structure and dynamics of liquid water on rutile TiO2 (110).” Physical Review B 82: 161415. Lo, C. S., K. S. Tanwar, A. M. Chaka and T. P. Trainor (2007). “Density functional theory study of the clean and hydrated hematite (1(1)over-bar02) surfaces.” Physical Review B 75(7): 075425. Luo, S. J., B. Averkiev, K. R. Yang, X. F. Xu and D. G. Truhlar (2014). “Density functional theory of openshell systems. The 3d-series transition-metal atoms and their cations.” Journal of Chemical Theory and Computation 10(1): 102–121. Mackrodt, W. C., R. J. Davey, S. N. Black and R. Docherty (1987). “The morphology of alpha-Al2O3 and alpha-Fe2O3—the importance of surface relaxation.” Journal of Crystal Growth 80(2): 441–446. Manassidis, I., A. Devita and M. J. Gillan (1993). “Structure of the (0001) surface of alpha-Al2O3 from first principles calculations.” Surface Science 285(3): L517–L521. Marx, D. and J. Hutter (2012). Ab Initio Molecular Dynamics: Basic Theory and Advanced Methods. Cambridge, UK: Cambridge University Press. Maslen, E. N., V. A. Streltsov, N. R. Streltsova and N. Ishizawa (1994). “Synchrotron X-ray study of the electron-density in alpha-Fe2O3.” Acta Crystallographica. Section B 50: 435–441. Massey, M., J. Lezama-Pacheco, M. E. Jones, E. Ilton, J. Cerrato, J. Bargar and S. Fendorf (2014). “Competing retention pathways of uranium upon reaction with Fe(II).” Geochimica et Cosmochimica Acta 142: 166–185. Mesmer, R. E. and D. L. Herting (1978). “Thermodynamics of ionization of D2O and D2PO4−.” Journal of Solution Chemistry 7: 901–912. Mochizuki, S. (1977). “Electrical conductivity of α-Fe2O3.” Physica Status Solidi A 41: 591–594. Navrotsky, A., L. Mazeina and J. Majzlan (2008). “Size-driven structural and thermodynamic complexity in iron oxides.” Geochimica et Cosmochimica Acta 72(12): A673. Newman, D. K. (2010). “Feasting on minerals.” Science 327(5967): 793–794. Park, C., P. A. Fenter, J. G. Catalano, S. S. Lee, K. L. Nagy and N. C. Sturchio (2010). “Aqueous-mineral interfaces toward extreme conditions: The potential experimental approaches with synchrotron X-ray probe.” Geochimica et Cosmochimica Acta 74(12): A794. Park, C., P. A. Fenter, N. C. Sturchio and J. R. Regalbuto (2005). “Probing outer-sphere adsorption of aqueous metal complexes at the oxide-water interface with resonant anomalous X-ray reflectivity.” Physical Review Letters 94(7): 076104. Parr, R. G. and W. T. Yang (1995). “Density-functional theory of the electronic-structure of molecules.” Annual Review of Physical Chemistry 46: 701–728. Perdew, J. P., K. Burke and M. Ernzerhof (1997). “Generalized gradient approximation made simple (vol 77, pg 3865, 1996).” Physical Review Letters 78(7): 1396. Perevalov, T. V., A. V. Shaposhnikov, V. A. Gritsenko, H. Wong, J. H. Han and C. W. Kim (2007). “Electronic structure of α-Al2O3: Ab initio simulations and comparison with experiment.” JETP Letters 85(3): 165–168. Pickett, W. E. (1989). “Pseudopotential methods in condensed matter applications.” Computer Physics Reports 9(3): 115–197. Pozun, Z. D. and G. Henkelman (2011). “Hybrid density functional theory band structure engineering in hematite.” Journal of Chemical Physics 134(22): 224706. Remler, D. K. and P. A. Madden (1990). “Molecular-dynamics without effective potentials via the Car-Parrinello approach.” Molecular Physics 70(6): 921–966. Renaud, G. (1998). “Oxide surfaces and metal/oxide interfaces studied by grazing incidence X-ray scattering.” Surface Science Reports 32(1–2): 1. Reuter, K. and M. Scheffler (2002). “Composition, structure, and stability of RuO2(110) as a function of oxygen pressure.” Physical Review B 65(3): 035406. Rollmann, G., A. Rohrbach, P. Entel and J. Hafner (2004). “First-principles calculation of the structure and magnetic phases of hematite.” Physical Review B 69(16): 165107. Rosso, K. M. and J. R. Rustad (2001). “Structures and energies of AlOOH and FeOOH polymorphs from plane wave pseudopotential calculations.” American Mineralogist 86(3): 312–317. Rosso, K. M., D. M. A. Smith and M. Dupuis (2003). “An ab initio model of electron transport in hematite (α-Fe2O3) basal planes.” Journal of Chemical Physics 118(14): 6455–6466.

148

Molecular Modeling of Geochemical Reactions

Russell, B., M. Payne and L. C. Ciacchi (2009). “Density functional theory study of Fe(II) adsorption and oxidation on goethite surfaces.” Physical Review B 79(16): 165101. Rustad, J. R., A. R. Felmy and E. J. Bylaska (2003). “Molecular simulation of the magnetite-water interface.” Geochimica et Cosmochimica Acta 67(5): 1001–1016. Rustad, J. R., A. R. Felmy and B. P. Hay (1996). “Molecular statics calculations for iron oxide and oxyhydroxide minerals: Toward a flexible model of the reactive mineral-water interface.” Geochimica et Cosmochimica Acta 60(9): 1553–1562. Schaefer, M. V., C. A. Gorski and M. M. Scherer (2011). “Spectroscopic evidence for interfacial Fe(II)-Fe(III) electron transfer in a clay mineral.” Environmental Science & Technology 45(2): 540–545. Scherer, M., C. Gorski, M. Schaefer, D. Latta, E. O’Loughlin, M. Boyanov and K. Kemner (2010). “Fe(II)-Fe (III) electron transfer in Fe oxides and clays: Implications for contaminant transformations.” Geochimica et Cosmochimica Acta 74(12): A920. Shroll, R. M. and T. P. Straatsma (2003). “Molecular dynamics simulations of the goethite-water interface.” Molecular Simulation 29(1): 1–11. Stubbs, J., A. Chaka, E. Ilton, C. Biwer, M. Engelhard, J. Bargar and P. Eng (2015). “UO2 oxidative corrosion by non-classical diffusion.” Physical Review Letters 114(24): 246103. Stumm, W., H. Hohl and F. Dalang (1976). “Interaction of metal-ions with hydrous oxide surfaces.” Croatica Chemica Acta 48(4): 491–504. Sverjensky, D. A. (1993). “Physical surface-complexation models for sorption at the mineral–water interface.” Nature 364: 776–780. Szabo, A. and N. S. Ostlund (1996). Modern Quantum Chemistry: Introduction to Advanced Electronic Structure Theory. Mineola, NY: Dover Publications. Tanwar, K. S., C. S. Lo, P. J. Eng, J. G. Catalano, D. A. Walko, G. E. Brown, G. A. Waychunas, A. M. Chaka and T. P. Trainor (2007). “Surface diffraction study of the hydrated hematite (1(1)over-bar-02) surface.” Surface Science 601(2): 460–474. Tanwar, K. S., S. C. Petitto, S. K. Ghose, P. J. Eng and T. P. Trainor (2009). “Fe(II) adsorption on hematite (0001).” Geochimica et Cosmochimica Acta 73(15): 4346–4365. Tews, W. and R. Gründler (1982). “Electron-energy-loss spectroscopy of different Al2O3 modifications. I. Energy loss function, dielectric function, oscillator strength sum rule and the quantity ϵ2E.” Physica Status Solidi B 109(1): 255–264. Thomas, M., M. Brehm, R. Fligg, P. Vöhringer and B. Kirchner (2013). “Computing vibrational spectra from ab initio molecular dynamics.” Physical Chemistry Chemical Physics 15(18): 6608–6622. Trainor, T. P., A. M. Chaka, P. J. Eng, M. Newville, G. A. Waychunas, J. G. Catalano and G. E. Brown Jr (2004). “Structure and reactivity of the hydrated hematite (0001) surface.” Surface Science 573(2): 204–224. Tunega, D. (2012). “Theoretical study of properties of goethite (α-FeOOH) at ambient and high-pressure conditions.” The Journal of Physical Chemistry C 116(11): 6703–6713. Valdes, A., J. Brillet, M. Gratzel, H. Gudmundsdottir, H. A. Hansen, H. Jonsson, P. Klupfel, G. J. Kroes, F. Le Formal, I. C. Man, R. S. Martins, J. K. Norskov, J. Rossmeisl, K. Sivula, A. Vojvodic and M. Zach (2012). “Solar hydrogen production with semiconductor metal oxides: New directions in experiment and theory.” Physical Chemistry Chemical Physics 14(1): 49–70. Valiev, M., E. J. Bylaska and J. H. Weare (2003). “Calculations of the electronic structure of 3d transition metal dimers with projector augmented plane wave method.” Journal of Chemical Physics 119(12): 5955–5964. Valiev, M. and J. H. Weare (1999). “The projector-augmented plane wave method applied to molecular bonding.” The Journal of Physical Chemistry A 103(49): 10588–10601. Vanderbilt, D. (1990). “Soft self-consistent pseudopotentials in a generalized eigenvalue formalism.” Physical Review B 41(11): 7892. Wang, X. G., A. Chaka and M. Scheffler (2000). “Effect of the environment on alpha-Al2O3 (0001) surface structures.” Physical Review Letters 84(16): 3650–3653. Wesolowski, D. J., J. O. Sofo, A. V. Bandura, Z. Zhang, E. Mamontov, M. Predota, N. Kumar, J. D. Kubicki, P. R. C. Kent, L. Vlcek, M. L. Machesky, P. A. Fenter, P. T. Cummings, L. M. Anovitz, A. Skelton and J. Rosenqvist (2012). “Comment on ‘structure and dynamics of liquid water on rutile TiO2 (110)’.” Physical Review B 85: 167401. Williams, A. G. B. and M. M. Scherer (2004). “Spectroscopic evidence for Fe(II)-Fe(III) electron transfer at the iron oxide-water interface.” Environmental Science & Technology 38(18): 4782–4790. Yanina, S. V. and K. M. Rosso (2008). “Linked reactivity at mineral-water interfaces through bulk crystal conduction.” Science 320(5873): 218–222.

First Principles Estimation of Geochemically Important Transition Metal Oxide Properties

149

Yin, S. and D. E. Ellis (2009). “DFT studies of Cr(VI) complex adsorption on hydroxylated hematite (11 02) surfaces.” Surface Science 603(4): 736–746. Yuan, K., E. S. Ilton, M. R. Antonio, Z. Li, P. Cook and U. Becker (2015). “Electrochemical and spectroscopic evidence for the one-electron reduction of U(VI) to U(V) on magnetite.” Environmental Science & Technology 49(10): 6206–6213. Zhao, Y. and D. G. Truhlar (2008). “Density functionals with broad applicability in chemistry.” Accounts of Chemical Research 41(2): 157–167. Zhou, F., M. Cococcioni, C. A. Marianetti, D. Morgan and G. Ceder (2004). “First-principles prediction of redox potentials in transition-metal compounds with LDA+U.” Physical Review B 70(23): 235121. Ziman, J. M. (1972). Principles of the Theory of Solids. Cambridge, UK: Cambridge University Press.

5 Computational Isotope Geochemistry James R. Rustad Corning Incorporated, Corning, NY, USA

This chapter is concerned with calculation of the energetics of isotope exchange reactions from electronic structure calculations (see also Chapter 7). The history of this area of research has been marked by a serendipitous simultaneous rapid evolution of analytical techniques that allowed highly precise measurement of the isotopic compositions of earth materials over a broad range of the periodic table and advances in computational chemistry software and hardware that allowed sufficiently accurate and feasible calculations of vibrational frequencies also over a broad range of the periodic table. The discussion here focuses on the thermodynamics of equilibrium isotope exchange as derived from harmonic partition functions. In this context, the overall strategy in calculating isotope exchange equilibria is: 1. Make a guess at an atomic-level model of the two environments that will exchange isotopes. As an example, these two environments could be HCO3 − aq and CaCO3 exchanging carbon isotopes. The molecular models representing the environments could be the HCO3 − 7H2 O and CCa8 O36 52 − embedded clusters in Figure 5.1. 2. Use an electronic structure code, implementing some representation of electron–electron interactions (this representation is often called the “model chemistry”), to adjust the atom positions (or a subset of these positions) in each environment until the forces associated with all allowed degrees of freedom are zero. 3. For each environment, use the electronic structure code to find the harmonic vibrational frequencies associated with these degrees of freedom. Two lists are needed for each site on which the isotope exchange will take place, one set of frequencies with the heavy isotope and one with the light isotope. 4. For each of these lists of frequencies, calculate the harmonic partition function ratio for the heavy and light isotopes. Molecular Modeling of Geochemical Reactions: An Introduction, First Edition. Edited by James D. Kubicki. © 2016 John Wiley & Sons, Ltd. Published 2016 by John Wiley & Sons, Ltd.

152

Molecular Modeling of Geochemical Reactions (a)

(b)

Figure 5.1 Molecular representation of environments for carbonate ion in (a) HCO3 − aq and (b) CaCO3. Both representations are “core” structures to be embedded in more extended structures which are not shown so that the core structures are more easily seen.

5. Take the logarithm of the ratio of the partition function ratios to find the free energy of isotope exchange. This chapter focuses on how to understand the essential problem in simple terms, how to choose representative atomic environments and model chemistries to give the best possible results, and some techniques for assessing the errors in quantum chemical calculation of isotope fractionation factors. The chapter also discusses some generalizations of the “rules of thumb” often used to qualitatively understand isotope fractionation that have been more completely understood through electronic structure calculations. It is mainly aimed at people new to such calculations, but parts of the chapter, especially those parts concerned with model construction and error estimates, could also be interesting to more experienced modelers.

5.1 A Brief Statement of Electronic Structure Theory and the Electronic Problem As reviewed in this volume and in many places elsewhere,1 an electronic structure calculation gives the total electronic energy (total potential and electron kinetic energy) of a collection of atoms as well as the forces on each of the nuclear centers in a collection of atoms. To solve this problem one generally has to specify: a. A set of nuclear positions. These are often represented simply by positive point charges corresponding to the atomic number and a given mass corresponding to the isotope of interest. For problems with heavy elements, the shapes of particular nuclei may also be specified. b. A set of basis functions for expressing the electronic wave function (in molecular orbital (MO) calculations) or the electron density (in density functional theory (DFT)). 1

A good source of information that is also a good read is C. J. Cramer’s book Essentials of Computational Chemistry (2004). An older book, published before the widespread DFT revolution in the 1990s, is A. Szabo and N. Ostlund’s Modern Quantum Chemistry (1996). Another great source of information is the slides from T. Helgaker’s talks on electronic structure theory (http://folk.uio.no/helgaker/talks). The book Molecular Electronic Structure Theory (2000) by T. Helgaker, P. Jørgensen, and J. Olsen is authoritative but challenging for someone with a geological/geochemical background.

Computational Isotope Geochemistry

153

c. A model for the electron–electron interactions. This might come from DFT or from MO methods. The choices of (b) and (c) are often coupled and sometimes collectively referred to as the “model chemistry” or “level of theory.” d. The total electron spin angular momentum (the number of spin-up—the number of spin-down electrons, nup − ndown + 1 is sometimes called the “multiplicity”). e. An initial guess for the electronic state (e.g., the electron densities of a noninteracting collection of atoms). Calculating the properties of the “gas” of electrons whizzing around the fixed nuclei is a complex many-body problem. To a first approximation, each electron can be thought of as moving in the average field of all the other electrons (the “Hartree–Fock” approximation). But since they are quantum objects, the electrons do not have “paths.” If one electron interacts with another, the two may emerge from the interaction with their identities exchanged. Of course, it cannot be said, after the interaction, which electron is which; one never knows exactly where the electrons are so there is no way to tell if they have exchanged. This effect gives rise to an interaction called the “exchange interaction.” Also, there will be large instantaneous deviations from the average field as the electron swims around in the gas, due to other electrons colliding with it. For example, the interaction of a “d” electron in a ferric iron atom with the four other d electrons will deviate strongly from any “average” field seen by any “average” electron. These deviations from the mean field picture contribute to the “correlation energy.” The two main methods used in electronic structure codes are the MO methods and DFT methods. These are discussed in many places, including elsewhere in this volume. The DFT approach is to parameterize the exchange and correlation effects into a functional, called the exchange–correlation functional, that gives the energy of the gas based on the electron density at all points in the gas (local density approximation (LDA)) and may also include terms related to the gradient of the density (generalized gradient approximation (GGA)) (Kohn and Sham, 1965). The MO methods start with the Hartree–Fock solution to the electronic problem, which has exact exchange but limited correlation, and then make improved estimates of the correlation energy beyond mean field theory. The MO methods have the advantage that they can be systematically improved, but the systematic improvement is expensive. The most efficient approach to estimate the correlation is through the use of “collision” operators to give single, double, triple, and other excitations to the reference Hartree–Fock wave function and then calculate the energies associated with these excitations. This is called coupled cluster (CC) theory (Bartlett and Musiał, 2007). Another approach, called manybody perturbation theory, can be used to approximate or augment CC. The most commonly used version of CC theory explicitly deals with single and double (SD) excitations and uses a perturbative correction for triple (T) excitations (abbreviated CCSD(T)). Second-order Møller–Plesset perturbation theory (MP2; Møller and Plesset, 1934) can often be used as a sort of approximation to CCSD (note that this does not imply that CCSD predictions are necessarily closer to experiment than MP2). These methods (MP2, CCSD, and CCSD(T)) are computationally very expensive methods, in part because they require large correlation-consistent basis sets (Dunning, 1989; Kendall et al., 1992) to do their job. Because they grow with at least the fifth power of the system size (MP2), they can be applied to only very small systems. At the time of this writing, a full vibrational calculation on an ion with six water molecules around it would be a large calculation for these kinds of methods. In most geochemical applications their main utility is in benchmarking DFT methods. Unlike DFT, they have the right physics to treat weak interactions such as the van der Waals interaction. The most basic solution to the electronic problem is to plug in a guess at the initial wave function (or electron density in the case of DFT) and run through the iterative self-consistent field (SCF) loop in the electronic structure code to converge to a stationary (i.e., time-independent) electronic state.

154

Molecular Modeling of Geochemical Reactions

In simple problems, this state would normally be the lowest energy state (or ground state), and the code would yield the total energy (electron kinetic energy, electron–electron interaction energy, electron–nuclear attraction energy, and nuclear repulsion energy) as well as the forces on each of the nuclei (gradients of the energy with respect to nuclear positions). The energy depends, in the Born–Oppenheimer approximation, parametrically on the nuclear coordinates. In the Born– Oppenheimer approximation, the time scale for nuclear motion is regarded as so slow relative to the time scale of electronic motion that the electrons completely relax around the nuclei as they move and thus provide a “potential energy surface” for nuclear motion, meaning that for a given set of nuclear coordinates, the forces on the nuclei are always the same, regardless of how the system arrived at that set of nuclear coordinates. The forces are often used to find a set of nuclear positions where all of these forces (or at least some subset of these forces) are zero. This is known as “geometry optimization” and locates a minimum on the potential energy surface. This optimization may apply to all degrees of structural freedom in the system or some subset of them (others being constrained to particular values for reasons to be discussed). Issues associated with (a–c) are discussed in more detail in Chapter 1. Concerning the more purely practical issues (d–e), it is not at all uncommon that the iterative SCF procedure for solution of the electronic problem fails to converge. For simple cases, this problem can be solved by running the calculation with a smaller basis set, saving the result, and using it as a guess for a new calculation with the larger basis set. Sometimes additional flexibility in generating guesses for the initial electronic state is required to obtain convergence. One approach that often works for molecular systems is to put the molecule in an electric field. This usually enables the system to achieve SCF convergence. Then the converged guess can be put in a weaker field and run through another SCF cycle. Gradually the field can be reduced to zero through several of these cycles. For transition metal systems with multiple centers having unpaired electrons, it helps to build up the guess with calculations on small fragments or individual atoms, assembling these into a larger system meeting the requirements in (c). This capability is sometimes called the “fragment guess.” NWChem (Valiev et al., 2010) and Gaussian (Frisch et al., 2004) codes have this capability. One should always remember the possibility of converging to excited electronic states. Excited states, of course, have different potential energy surfaces and different vibrational frequencies than ground states. Another potential problem is that it is possible to converge to a stationary state that is not even a local minimum in energy. In some codes, such as Gaussian, there are facilities to do stability analyses to check whether a true minimum has been found in the iterative SCF procedure. For systems with multiple transition metals, it is probably an excellent investment to run such an analysis.

5.2 The Vibrational Eigenvalue Problem Having found such a stationary state at the minimum in the electronic energy, with all forces associated with the relevant degrees of freedom equal to zero (within some preselected criteria), the derivatives of these forces with respect to the nuclear positions (sometimes called the force constants) are also calculated. In some cases it may be possible to calculate this matrix analytically through perturbation theory; in others it may be necessary to physically displace each atom (perhaps taking account of symmetry) and calculate the force derivatives numerically. However they are obtained, the force derivatives form a matrix which may be written as dFαi dβj √ mi mj , where dFαi is the change in the force felt by atom i in the α direction when atom j is displaced in the β direction (e.g., α and β could be either x, y, or z if Cartesian coordinates are used) and mi and mj are the masses of atoms i and j.

Computational Isotope Geochemistry

155

The eigenvectors of this matrix give the normal modes of the system of atoms (i.e., the atomic displacements associated with each vibrational frequency), and the square roots of the eigenvalues give the frequencies of oscillation associated with the normal modes (see Chapter 10). In the harmonic approximation, solving the eigenvalue problem turns the complicated motions of the system of atoms into 3N independent harmonic oscillator problems. The frequencies of these oscillators depend, in general, on the masses of all the atoms because of the √ mi mj term in dFαi dβj √ mi mj . On the other hand, if a particular vibrational mode involves no displacement of atom i, the frequency will be independent of the mass of atom i. Note that in the Born– Oppenheimer approximation, the matrix of force constants dFαi/dβj does not depend on the nuclear masses, so that if the eigenvalue problem needs to be solved for a different set of atomic masses, only the √ mi mj needs to be modified and there is no need to recompute the matrix of force constants. Relative to the time required to compute the dFαi/dβj matrix, the solution of the vibrational eigenvalue problem takes negligible computational time. So to do a problem involving isotopic substitution, the force-constant matrix dFαi/dβj is computed, saved, and then divided by the √ mi mj term according to the isotope substitution scheme of interest. The matrix dFαi/dβj is also called the “Hessian matrix,” and the dFαi dβj √ mi mj is often called the “dynamical matrix.” In the simple case of a single atom vibrating harmonically in an isotropic potential well, the dynamical matrix A is very simple: K m

0

0

0

K m

0

0

0

K m

Solving the eigenvalue problem Ax = λx, one can see by inspection that there are three orthogonal eigenvectors (100), (010), and (001) with eigenvalues (K/m) and, hence, the frequencies are simply the familiar ω = √ K m . As a practical matter, if the reader is trying out one of these calculations, the first task is to figure out how to save the Hessian (dFαi/dβj) matrix separately and then to do the mass substitution to get the dynamical matrix dFαi dβj √ mi mj to find the vibrational frequencies. Because most people doing electronic structure calculations are not isotope geochemists and are just interested in getting the frequencies for the standard atomic weights, isotope substitution is not a commonly carried out procedure, and it may take some time to figure out how to save dFαi/dβj, especially for researchers doing calculations on periodic solids.2 Generally it is best to try it out first on a simple system. Hessian matrix evaluations on large systems can be computationally demanding. If the format of the Hessian matrix can be determined, it is easy to write small code to read it in, do the mass division, and use pass it to any suitable program (like JACOBI.f in the Numerical Recipes codes (Press et al., 1986)) to find the eigenvalues and eigenvectors. As an aside, keep in mind that a numerical evaluation of the Hessian can usually be restarted in the event of a computer crash, whereas with an analytical Hessian, one has to start over completely.3

2

In one commonly used density functional code for solids, the facility to do simple mass substitutions to rebuild the dynamical matrix had not even been considered by the developers; the entire construction of the Hessian part of matrix had to be done all over again with the new mass! The author once had his computer tied up for 6 weeks evaluating an analytical Hessian for a particularly large system. On the last day, watching anxiously as a freak September California storm approached, he saw the last iteration on the coupled perturbed Hartree–Fock solution complete just as the power went out and the entire matrix was lost. If you do these calculations frequently on your own computers, it’s a good idea to invest in a backup power system.

3

156

Molecular Modeling of Geochemical Reactions

5.3 Isotope Exchange Equilibria Through solving the eigenvalue problem, we have now turned our problem into 3N independent quantum harmonic oscillator problems. The quantum mechanical solution of the harmonic oscillator is discussed in every physical chemistry textbook. The take-home message is that there is a set of energy levels n + 1 2 hω, where h = Planck’s constant, ω = frequency, and n = 0,1,2,…, with the interesting point being the existence of a zero-point energy that is not zero. Due to the uncertainty principle, a particle cannot sit right at the bottom of its potential well in quantum mechanics. But a heavier particle gets closer to zero than a lighter particle because its frequency is lower and its zeropoint energy 1 2 hω is less. By running the electronic structure code, we now have two lists of frequencies for each environment involved in the isotope exchange, one list of frequencies for the heavy isotope and one list for the light isotope. The basic physics of isotope exchange energetics is nicely reviewed in many places (e.g., by Schauble (2004) and Wolfsberg et al. (2009) provides a great discussion from a computational point of view). Only a brief discussion is provided here. Geological processes fractionate isotopes for many reasons. One important reason is the difference in vibrational free energy between products and reactants in an equilibrium involving isotope exchange reactions such as h

M A + l M B = hM B + l M A

51

where M is some element in environment A (e.g., an aqueous solution) and environment B (e.g., a mineral) and hM represents the heavy isotope of M and lM represents the light isotope. An obvious point worth emphasizing is that if the isotope is the same (i.e., h = l), then the energy change is zero. This means that the isotope exchange reaction is a particularly simple one with the electronic contributions cancelling on each side of the reaction. That said, most of the rest of the world is focused on computing energies for reactions such as 2CH4 + 2O2 = CO2 + 2H2 O

52

here the electronic contributions don’t cancel. It is good to keep in mind that most of the efforts of the quantum chemistry community are focused on reactions like (5.2) rather than reactions like (5.1). Reaction 5.1 is sort of the ultimate isodesmic reaction, as no bonds at all are broken or formed. Theoretical improvements designed to obtain better energies for reactions such as Reaction 5.2 (i.e., improved exchange–correlation functionals for reaction energies in DFT) are not necessarily going to make great improvements in the energies for Reaction 5.1. The simplest way to view Reaction 5.1 is to imagine that element M vibrates harmonically in the A and B environments with characteristic force constants KA and KB, representing the stiffness of each environment, where a higher K corresponds to a stiffer environment. It was described earlier how to evaluate this stiffness with an electronic structure code by finding the equilibrium structure, where the forces on all active atoms are zero, and then displacing the M atom by dx and calculating the force F on M which, in the harmonic approximation, is given by −Kdx where K = dF dx. At the bottom of an approximately parabolic well, K is the curvature of the energy as a function of distance away from the origin dF dx = d 2 E dx2 . Imagine that in our reaction, we take a system where KA = 1, KB = 2, h M = 2, and l M = 1. Thus on the right-hand side of Equation 5.1, we put the heavy isotope in the stiff environment and the light isotope in the soft environment. On the left-hand side of the reaction, we put the heavy isotope in the soft environment and the light isotope in the stiff environment.

Computational Isotope Geochemistry

157

What would be the equilibrium constant for Reaction 5.1 at zero temperature? In this case, the energy E of the products and reactants can be evaluated using E = 1 2 hω (n is zero at zero temperature in E = n + 1 2 hω). Remembering that ω = √ K M and using units such that h = 1, the zeropoint energy of the product side of Equation 5.1 is 1

2

√ 2 2 + 1 2√ 1 1 = 1

the zero-point energy of the reactant side is 1

2

√ 2 1 + 1 2 √ 1 2 = 3 2√2 ≈ 1 06066

and the energy change for the reaction (the energy of the products minus the energy of the reactants) is approximately −0.06066. The minus sign indicates that the product side of the reaction is more favorable; thus it is apparently better to pair the heavy isotope with the stiffer environment. This makes physical sense because the heavy isotope, having the lower-frequency spring vibration, will burrow itself lower into the vibrational potential energy well than will the light isotope 1 1 . This burrowing is more pronounced as the spring vibrational fre2 √ K Mh < 2 √ K Ml quency increases, being proportional to the square root of the stiffness K. The stiffer the environment, the lower the heavy isotope rides in the well relative to the light isotope. So, to achieve the lowest energy possible, the heavy isotope will tend to partition into the strongest bonding environments. If temperature increases beyond zero kelvin, the Boltzmann law gives the populated energy levels and the zero-point energy is replaced with the slightly more complex equation: E = hω

1

2

+1

e hω

kT

−1

53

In any real system, there will be many characteristic vibrational frequencies associated with atoms in particular environments, associated with the normal modes as described earlier. But this is straightforward because through the solution of the eigenvalue problem each of these normal modes is by definition independent of the other normal modes. Since the energy of many oscillators is the sum of the energies of the individual oscillators, and the logarithm of the partition function gives the free energy, all the normal modes are multiplied together in the partition function for a single isotope exchange (Bigeleisen and Mayer, 1947; Urey, 1947): β = Qh Ql =

1− 3N

uhi uli × exp −uhi 2

1 − exp −uhi

× 1 − exp −uli

exp − uli 2 54

where uh li = hω h l kT For gas-phase molecules there are three rotations and three translations that have zero frequency that do not contribute to the β (sometimes called the reduced partition function ratio, or RPFR). The equilibrium constant for a single isotope exchange between environments A and B, often called αAB, is given by βA/βB where βA is the RPFR for environment A. To keep all the signs straight between the heavy and light isotopes, just remember that the larger the β associated with a particular environment, the more that environment accumulates heavy isotopes. So that if the beta for 13C/12C substitution in CO2 is βCO2 and the beta for 13C/12C substitution in CH4 is βCH4 , the heavy carbon accumulates in CO2 and the equilibrium constant for the reaction 12

CO2 + 13 CH4 = 13 CO2 + 12 CH4

55

158

Molecular Modeling of Geochemical Reactions Table 5.1 Frequencies (cm − 1 ) for CO2 and CH4 calculated with DFT using the exchange–correlation functional B3PW91 and basis set aug-cc-pVTZ. CO2 β = 1.198533 Heavy 660.48 660.48 1374.27 2358.26 CH4 β = 1.118065 Heavy 1318.47 1318.47 1318.47 1551.32 1551.32 3033.69 3137.72 3137.72 3137.72 Masses:

12

C = 12.000000 amu;

Light 679.83 679.83 1374.27 2427.35 Light 1326.71 1326.71 1326.71 1551.32 1551.32 3033.69 3148.93 3148.93 3148.93 13

C = 13.0033548 amu.

with the equilibrium constant given by α = βCO2 βCH4

56

that is, the β in the numerator goes with the environment having the heavy isotope on the product side of the reaction and the one in the denominator goes with the environment having the heavy isotope on the reactant side of the equation. Generalization to periodic solids requires sums over points in the Brillouin zone. The partition function can also be generalized in terms of the vibrational density of states as authoritatively discussed by Kieffer (1982). There is a certain beauty, however, in keeping things straightforward and simple and representing environments in crystalline materials as embedded molecules (as discussed in the following) without the intellectual baggage of k-points and Brillouin zones. The molecular approach has some other practical advantages in terms of available software, which, in general, is more well developed for molecules than for periodic solids. The main point is that, in the harmonic approximation, we can associate a single number called the RPFR with each environment and a particular pair of isotopes, h and l. The ratio of the RPFRs gives the equilibrium constant for the isotope exchange reaction between the two environments, and the environment enriched in the heaviest isotope has the highest RPFR. This is often reported as 1000 ln(K) in “per mil” notation. To illustrate these points consider the isotope exchange reaction in (5.5). We need a list of frequencies for CO2 with 12C and 13C as well as one for CH4 with 12C and 13C. Choosing the standard atomic weights for O and H, we obtain the lists in Table 5.1. Fortran code for calculation of the RPFR given a list of frequencies for heavy and light masses implicit none real*8 numerator, denominator, prod real*8 u,up,h,c,a real*8 omegah(1000),omegal(1000) real*8 rt,hc

Computational Isotope Geochemistry

159

integer nfreq,i c=29979245800.0d0 a=6.022141d23 h=6.626069d-34 rt=8.3144*298.15d0 hc=h*c*a 56 format(i5,2f12.4,e20.10,f10.6) read(*,*)nfreq doi=1,nfreq read(*,*)omegah(i),omegal(i) enddo prod=1.0d0 doi=1,nfreq u=hc*omegah(i)/(rt) up=hc*omegal(i)/(rt) numerator=(u*dexp(-u/2.0d0)/(1.-dexp(-u))) denominator=(up*dexp(-up/2.0d0)/(1.-dexp(-up))) prod=prod*numerator/denominator print 56,i,omegah(i),omegal(i),(numerator/denominator)-1. &,prod enddo stop end Running the lists in Table 5.1 with the code given earlier, we obtain βCO2 = 1 198533 and βCH4 = 1 118065. The equilibrium constant for the reaction is then 1.198533/1.118065 = 1.071971, which would correspond to an enrichment of 13C in CO2 of approximately 69.5‰ at 25 C. This is an equilibrium value and says nothing about the time scale to reach equilibrium, which may be very long at 25 C. These are the main points on the subject of equilibrium isotope distributions in the harmonic approximation. Some qualitative lessons from the discussion need to be highlighted.

5.4

Qualitative Insights

Short bonds tend to have higher vibrational frequencies than long bonds. This is known as Badger’s rule (Badger, 1934). Especially when considering a specific type of bond, for example, between iron and oxygen, we expect that shorter iron–oxygen bonds will be stiffer than longer iron–oxygen bonds. Any systematic chemical environmental change that affects the bond length will likely be manifested in a change in isotope fractionation. For example, any ionic compound in which Fe is in the Fe(III) oxidation state will tend to accumulate heavy iron relative to a compound in which iron is in the Fe(II) oxidation state. This is because the Fe(III)–ligand bond lengths are shorter than the Fe(II)–ligand bond lengths and therefore their vibrational frequencies are higher, and therefore it is somewhat more favorable to put heavy isotopes in the Fe(III)–ligand bonds. Elements with a lower coordination number have shorter bonds than elements with a higher coordination number, so that we can expect, for example, that silicon isotopes in enstatite, with fourfold coordinated silicon, should tend to be isotopically heavy relative to majorite garnet, with sixfold coordinated silicon. For open-shell transition metals with multiple unpaired electrons, elements in the low-spin state will tend to have shorter, stronger bonds (if the electrons pair up, they need less room) and therefore tend to accumulate heavier isotopes than elements in high-spin states.

160

Molecular Modeling of Geochemical Reactions

Besides offering quantitative estimates of such effects, electronic structure calculations have identified some interesting insights into exceptions to these generalizations. For example, take the fractionation of iron isotopes between ferropericlase Mg 1− x Fex O and ferroperovskite Mg 1− x Fex SiO3 in the Earth’s mantle. By the rule of thumb that the heavy isotopes fractionate into the environments with the lowest coordination numbers, it would be expected that the 6-fold coordinated iron in ferropericlase would be isotopically heavier than the 12-fold coordinated iron in ferroperovskite. In fact, DFT calculations predict the opposite. Because of its simple chemistry, the cumulative contributions to the RPFR of ferropericlase are finished by approximately 800 cm−1 (i.e., there are no frequencies above 800 cm−1 involving motion of iron). For ferroperovskite, on the other hand, contributions to the RPFR from coupling of iron motion to the Si─O stretching frequencies continue to be made through 900 cm−1. These are enough to drive the RPFR for ferroperovskite above ferropericlase, despite the 6-fold versus 12-fold coordination environment. In this case, chemical composition is more important than coordination number in determining the final RPFR (see Rustad and Yin, 2009). The ferropericlase/ferroperovskite system is also interesting from the point of view of the effect of spin state on the RPFR. For ferropericlase the room-temperature RPFR between high-spin and low-spin electronic states is about 7.6‰, with the low-spin state having the shorter bond and the larger RPFR (and, hence, a stronger tendency to be enriched in heavy iron). For ferroperovskite the difference between the RPFR for the high-spin state and the RPFR for the low-spin state is much smaller. This unexpected behavior arises from the asymmetry of the Fe─O bonds in the low-spin coordination environment. While in ferropericlase all Fe─O bonds become significantly shorter after the spin transition, in ferroperovskite the low-spin ferrous iron is too small for the 12-fold coordination environment and sits asymmetrically in the 12-fold coordinated perovskite “B” site with some short bonds and some long bonds. These tend to compensate one another in their contributions to the RPFR so that the spin transition has little effect (Rustad and Yin, 2009). Along these lines, we have talked loosely about the differences in “stiffness” in the environment and made the observation that we should expect to find the heaviest isotopes preferentially in the stiffest bonding environments. But this is an ambiguous statement in some ways. One needs to be careful not to define “stiff environment” in terms of, for example, shear modulus. Consider the isotopic composition of magnesium dissolved in calcite in equilibrium with an aqueous solution. If asked about the distribution of 26Mg and 24Mg in these environments, one might, from the foregoing discussion, answer that the 26Mg should be concentrated in the calcite, as the calcite environment is obviously “stiffer” than water. But what matters here is the frequencies experienced by the magnesium in each of the environments, not the fact that the shear modulus of water is zero while the shear modulus of calcite is not. It turns out, in fact, that heavy magnesium accumulates in the aqueous solution. Analogous to the ferropericlase/ferroperovskite example given previously, this is mostly because of high-frequency vibrations caused by magnesium coupling with wagging motions of water molecules (Rustad et al., 2010a). This is predicted by first-principles calculations, but it is counterintuitive, if one uses a definition of stiffness that is too rigid.

5.5 Quantitative Estimates There is no doubt that through detailed calculations we have some better qualitative insights into the factors that influence these familiar rules of thumb in describing equilibrium isotope fractionation in different chemical environments. These types of insights are largely independent of the details of the calculations, that is, choices we make for (a–c) of Section 5.1.

Computational Isotope Geochemistry

161

If an accurate prediction is sought, then conclusions drawn will usually depend on the choices we make for (a–c) of Section 5.1. In other words, rather than asking, “Is Ca2 + aq enriched or depleted in heavy calcium relative to calcite?” one is interested in the question, “Is the 44 Ca − 40 Ca separation factor between Ca2 + aq and calcite 1‰ or is it closer to 2‰?” To address quantitative problems, at least at present, effective computational chemistry depends on one’s ability to choose problems that naturally result in a high degree of error cancellation. Consider the isotopic fractionation between Fe3 + and Fe2 + in aqueous solution. This is a great example of a problem with a large signal that does not depend very much on the basis set used, the model for the electron–electron interactions, or the particular molecular model used to represent the systems (Fe H2 O 6 3 + 2 +, that is, an iron ion with six water molecules around it, works fine). The reason is that beyond the first solvation shell, the environments of Fe2 + and Fe3 + are very similar. Although there is a significant change in the RPFR resulting from improving the molecular model from Fe H2 O 6 2 + to Fe H2 O 6 12H2 O2 + (i.e., including a partial second solvation shell), there is also a very similar change in going from Fe H2 O 6 3 + to Fe H2 O 6 12H2 O3 +, and this part of the RPFR cancels in considering fractionation between the two environments. A more difficult problem is the fractionation between, say, Fe2 + aq and hematite, α-Fe2O3. These environments are quite different from one another, so that one cannot rely on error cancellation in the representation of the environment. A basis set/environment combination that worked well for the Fe2 + Fe3 + fractionation might not work well at all for this problem; one has to get closer to the “true” RPFR characteristic of the environment of interest. In other words, if there is a significant effect on the absolute RPFR by including a partial second solvation shell for Fe2 + aq , this will not be compensated by a corresponding effect in Fe2O3, as that environment is different from the aqueous environment. In this case, it turned out that a second solvation shell, plus a continuum representation of the rest of the solvation environment, was required to get an accurate value for the individual Fe2 + aq RPFR and, hence, for the aqueous/mineral fractionation (Rustad et al., 2010b). Addressing quantitative problems requires that due attention be paid to (a–c). The basis set needs to be large enough so that the answer does not change significantly upon making it a little larger but not so large that it makes the problem computationally intractable. Generally the investigator must experiment to see how the results depend on the basis set, picking the best affordable one and then backing down a bit to a smaller set of functions and see how the results change. Or another approach is to pick a very small system, yet still meaningful, and use that system as a benchmark. There are many kinds of basis sets, including localized functions such as Gaussians (Dunning, 1989) or Slatertype functions, as well as plane waves (Troullier and Martins, 1991). To use plane waves, the core region must be taken care of separately either with an auxiliary localized basis set (the so-called linearized augmented plane wave or LAPW method as used, e.g., in WIEN2K (Schwarz and Blaha, 2003)) or through a pseudopotential representing the repulsion of the core electrons or by smoothing the plane waves near the core in a prescribed way (this is the so-called projector augmented wave (PAW) method (Bloechl, 1992)). Different environments will in general have different basis set requirements. For example, when looking at isotope effects arising from isotope variations in nuclear shape (Schauble, 2007), it might be advantageous to use Slater-type functions rather than Gaussian functions as these are a more faithful representation of the electron density at the core, which is where the action is in this type of problem. It is also a good idea to keep system-dependent basis set convergence issues in mind when trying to manage error cancellation. For example, the Mg2 + aq requires a more complete basis set than the vibrational spectrum of Mg2 + in MgCO3. This is because the aquo

162

Molecular Modeling of Geochemical Reactions

ion includes important contributions from the relatively weak interactions between water molecules, and large basis sets with diffuse functions are required to get a converged result, even if one is using DFT and not describing the interactions accurately. Consider a problem involving fractionation between an aquo ion and a mineral. The aquo ion environment is run with a small basis set A and a large basis set B. The cluster representing the mineral environment is larger, however, and can only be run with basis set A. Should you estimate the fractionation by running basis A on both the mineral cluster and the aquo ion cluster? At first, you might think that better error cancellation could be obtained this way, but what ends up happening is that the mineral RFPR would hardly change at all in going from basis A to basis B, but the aquo ion would have a relatively large change. So a better estimate is actually obtained by running the aquo ion with the large basis set and the mineral cluster with the small basis set. Of course, the best approach would be to run them both with basis set B, but the B-aquo ion/A mineral combination will be closer to the B-aquo ion/B mineral combination than will the A-aquo ion/A mineral combination (Rustad et al., 2010b). Concerning the model for the electron–electron interactions: for most problems of interest in geochemistry, we live in the world of DFT. The choice of the exchange–correlation potential representing the electron–electron interactions is much like the basis set, but worse, because, while we know that a larger basis set will always give a truer picture of theoretical predictions, the list of DFT exchange–correlation functionals is always growing. Many of these are highly specialized, designed to offer improvements only in small areas (i.e., the calculation of reaction barriers) without any systematic improvement over the more standard functionals in other respects. Further, the world of geochemistry is filled with water, a solvent influenced by weak interactions, such as dispersion and hydrogen bonding, which are difficult to describe using DFT. While, in geochemical problems, we mostly work with systems sufficiently large that we have to use DFT, it’s a good idea to check the DFT results with MO calculations on small systems. For example, one finds that one cannot afford to run accurate MO methods (large basis MP2 or CCSD(T)) for the Mg H2 O 6 12H2 O2 + model that was put together to represent the aquo ion nor the Mg(CO3)6.18Ca embedded cluster that was put together to represent the Mg2 + -bearing calcite. On the other hand, these methods might be feasible for both Mg H2 O 6 2 + and the Mg −CO3 dimer. The B3LYP/6-311++G(2d,2p) calculation that was being contemplated for the large system can now be checked against the MP2/aug-cc-pVTZ calculation on the small system. Say, for example, that check shows that, for the small system, the fractionation factor calculated using DFT is 3.2‰ higher than the fractionation factor calculated with the MO method. First, this indicates immediately that the DFT calculation for the large system is not going to give accuracy better than 3.2‰. However, given a known offset between the DFT result and the MO result on a very similar but much smaller system, a reasonable person might accept a correction of 3.2‰ to the fractionation factor calculated with DFT on the large system as an estimate of a value that might be obtained if accurate MO calculations were possible on the large system. Putting together an effective ladder of successively cross-checked molecular models often gives good insights into the likely accuracy of the calculations and allows informed estimates to be made about the likely magnitude and direction of corrections that would be expected at higher levels of theory. Putting dispersion/van der Waals interactions into DFT in the most consistent and general way is an active area of research, likely to have an important impact on geochemical applications (see, e.g., Fornaro et al., 2014). Finally, choosing where to put all the atoms to best represent the system of interest is a crucial task. This is in some ways the heart of what is called “molecular modeling.” In gas-phase molecules and bulk solids (with sufficiently small unit cells), the representations are straightforward. Aqueous

Computational Isotope Geochemistry

163

environments, however, are dynamic and may need to be represented by more than one configuration. Surface environments such as surface metal centers or sorbed surface complexes are rarely known with any certainty. Often molecular simulation methods such as molecular dynamics or Monte Carlo methods can be used to generate such configurations. Careful, imaginative construction of representative environments can be one of the most difficult (and most fun!) aspects of doing electronic structure calculations for complex problems involving interfaces or solvated species, where precise configurations are not known. In my own work, I have tended to prefer to represent exchange sites in minerals through embedded clusters rather than as periodic solids. In part, this is because much of that work was focused on isotope fractionation between aquo ions and minerals, and I wanted to use the same methods on both the aquo ion and the mineral system to achieve the best possible level of error cancellation. The molecules-as-minerals idea goes back to Gibbs’ 1982 paper (Gibbs, 1982). Then, as now, the motivation for this is that much more flexible computational methods are available for clusters than for periodic solids. If one insists on representing sites in minerals as periodic solids, using anything other than pure DFT functionals is very expensive. MO methods beyond Hartree–Fock, such as MP2 or CCSD(T), have not been implemented in accessible solidstate codes. True, there are the DFT+U (the so-called Hubbard model where an extra repulsive term is added to, e.g., d electrons in a transition metal atom which “see” each other much more explicitly than could be captured in an average, mean field theory; Anisimov et al., 1997), but such methods are mainly used to improve band gaps rather than achieve better general model chemistries. Setting up clusters to represent crystalline and surface environments is straightforward with the right molecular modeling tools (a useful code for this is CrystalMaker™). As an example, let’s set up a calculation for modeling a CO3 group in calcite. Since there is only a single type of CO3 group, we can imagine taking a central carbonate molecule, all eight cations attached to this central group, and then the CO3 ions attached to those eight (other than the central carbonate). The bonds from the outer shell of CO3 ions to the outer Ca2 + ions are replaced by terminating nuclear centers (sometimes called “link atoms”) with a +2/6 charge to match the Pauling bond strength (charge/coordination number) coming into the oxygen atoms on the carbonate from these outer Ca2 + ions. The central carbonate group, the Ca2 + ions attached to it, and all oxygens attached to the Ca2 + ions are chosen as vibrationally active, and the rest of the atoms are fixed in their measured positions. The central CO3Ca8O33 molecule in Figure 5.1 then vibrates within the fixed outer rind. In this way, a “molecule” has been created which ought to have fractionation characteristics very similar to the periodic solid from which the molecule was created. Within this molecule it is straightforward to use hybrid functionals or to put an aug-ccpVTZ basis on the central carbonate. In principle, one can even treat the central carbonate unit at the MP2 level embedded within a DFT treatment of the rest of the cluster (Tuma and Sauer, 2006). For systems investigated so far (Fe2O3 (Rustad and Dixon, 2009), CaCO3 (Rustad et al., 2008), Mg1− x Fex O and Mg1− x Fex SiO3 (Rustad and Yin, 2009)), the embedded cluster type of approach gives excellent agreement with full lattice dynamics treatments. One issue with the use of clusters is that one has to know the mineral structure. This might be a problem if calculations are to be done at high pressure and the lattice parameters are not known. Another problem is the inconsistency in plugging the experimentally determined structure into, for example, a DFT method that wouldn’t recover exactly the measured structure. DFT in the GGA tends to overestimate bond lengths and lattice parameters can be overestimated by 1–2%. A possible work-around would be to optimize the crystal structure with DFT and then construct the cluster from the DFT-optimized structure. On the

164

Molecular Modeling of Geochemical Reactions

other hand, it’s usually a good idea to use/impose the experimental structural information if one has access to it. This is a good strategy even for highly accurate calculations on small molecules (Helgaker et al., 2008). The requirements for clusters representing aquo ions depend strongly on whether the aquo ion is a cation or an anion. Cations usually can be represented by symmetric clusters with an explicit first and (partial) second solvation shell that are then embedded in a continuum representation of the rest of the solvent (see Figure 5.2). First-and-partial-second-shell solvent representations for cations are available in the literature for cations in tetrahedral and octahedral coordination. Anions require many explicit water molecules, at least thirty, and, in my experience, do not respond predictably to embedding in a continuum solvent. (Note that the cluster representing HCO3 − aq in Figure 5.1 is only the core of a much larger cluster and would not, by itself, be sufficient for accurate computation of the fractionation factor for 12 C − 13 C exchange between the bicarbonate aquo ion and calcite.) Because there are no “standard” structures available for clusters of this size, the only way to generate them is through molecular dynamics or Monte Carlo methods. If there are no interaction potentials available, then one has to perform ab initio molecular dynamics studies to generate a sufficient number of representative conformers (10–20). Surface environments may also be of interest. Although this area has not been extensively investigated, experience so far suggests that surface effects are small. For example, Rustad and Dixon (2009) looked at the RPFR for iron isotope exchange in hematite in the bulk as well as at the hematite (012) surface with both molecular and dissociative water adsorption and saw, surprisingly, almost no difference at all between the bulk RPFR and the RPFR values of both types of surface environments. More studies will be required before any general conclusions about surface isotope effects can be drawn. Of course, surface environments can have higher concentrations of elements in different oxidation states, for example, iron atoms at iron oxide surfaces may be partially reduced (Henderson et al., 1998; Williams and Scherer, 2004), and this will certainly have consequences for the isotopic composition of the surface iron atoms (Frierdich et al., 2014; Handler et al., 2009). A major problem in application of electronic structure methods to isotope fractionation problems today is that variations in basis sets, electron–electron interaction models, and environmental representations can easily result in a random walk in combinations of these factors. In particular, reasonable choices for each of these factors can cause 1–2‰ variations in positive and negative directions,

S6

Figure 5.2 Two different cluster representations M H2 O with explicit first and second shell water molecules.

Th 6

12H2 O n+ for octahedrally coordinated aquo ions

Computational Isotope Geochemistry

165

and it is often possible to find some combination that gives an “expected” answer. Individual investigators no doubt can recognize this sort of pathology—no conscientious investigator would try nine combinations of three basis sets and three exchange–correlation functionals, find the one that works, and publish the “successful” calculation without mentioning the failure of the others. If the random walk is carried out over multiple research groups, however, the combination that agrees with “experiment” is the one that tends to get published and the end result is the same as if a single investigator “cherry-picked” a certain combination without telling the scientific community about the other failures. Nothing is really learned this way, unless one can step back and find some combination that tends to work well in a variety of situations. At this point in time, even in relatively well-defined chemical environments, such as ions in solution, at aqueous–mineral interfaces, in oxide, silicate, and carbonate minerals, there seems to be no prescription for success in terms of exchange– correlation functional choices. Thus for any given problem, it is best to try a range of basis sets and exchange–correlation functionals and structural environments to generate some reasonable range of predictions before coming to any strong conclusions. To help illustrate these issues, consider Figure 5.3 showing the relative error in the computed equilibrium constants for isotope exchange reactions involving small molecules, originally studied by Richet et al. (1977) over a wide range of DFT exchange–correlation functionals and basis sets. Clearly there are certain basis sets (cc-pVDZ) and exchange–correlation functionals (m06-hf ) that tend to perform worse than others overall. However, the figure also shows that the performance is highly element specific, with the (aug)cc-pVDZ basis set being notably worse than other basis sets for oxygen exchange reactions. The DZVP(2) family of basis sets (Godbout et al., 1992), which were constructed specifically to be used in DFT calculations in the early days of application of DFT to molecules, do, in fact, appear to perform as well as the more expensive triple zeta basis sets, overall, but again, this is element specific, with the DZVP(2) family performing less well for oxygen exchange reactions. For carbon, LDA seems to perform very well with a wide range of basis sets. It should be kept in mind that these reactions never involve hydrogen-bonded systems with weak interactions and are thus not representative of most of the reactions that would really be of interest in low-temperature geochemistry. For example, the HCTH-407 functional (Boese and Handy, 2001) is one of the better-performing DFT functionals in Figure 5.3 but does not do nearly as well when hydrogen-bonded systems are considered. Nevertheless, Figure 5.3 drives home the point that there are, at this point in time, no “magic” DFT functionals that work particularly well for computing isotope exchange equilibria over a broad range of chemical systems. Example: Calculating the 11B–10B isotope fractionation factor for B(OH)3(aq) and B(OH)4−(aq) The fractionation of 11B and 10B between B(OH)3 (aq) and B OH 4 − aq has been used in paleoclimate studies to estimate the pH of the oceans on time scales of 20 million year (Hemming and Hanson, 1992). The main assumption is that marine carbonates only incorporate B OH 4 − and not B(OH)3 (although it seems less and less likely the more is known about carbonate crystal growth mechanisms; see Demichelis et al., 2011). To make the points made in the previous discussion more concrete and to show the utility of making a ladder of multiple techniques and system size, consider the calculation of the 11 B − 10 B fractionation between B(OH)3 (aq) and B OH 4 − aq : 10

B OH

3

aq + 11 B OH

− 4

aq = 11 B OH

3

aq + 10 B OH

4

57

as studied by Rustad et al. (2010a). Figure 5.4 shows the equilibrium constant for the reaction as a function of the number of solvating water molecules. For the 32-water case both B(OH)3 and B OH 4 − have 32 waters, taken from 10 independent configurations from an ab initio molecular dynamics simulation (AIMD) of B(OH)3 (aq) and B OH 4 − aq . For the intermediate points

166

Molecular Modeling of Geochemical Reactions DFT b88+perd86 xper91 m06–L m06–hf m06–2x b31yp becke97 becke97–1 becke97–2 becke97–3 mpw1k pbe0 perdew LDA optx heth heth–407 m06 cc–pVDZ

cc–pVTZ

aug–cc–pVDZ

aug–cc–pVTZ

DZVP (DFT)

DZVP2 (DFT)

cc–pVDZ

cc–pVTZ

aug–cc–pVDZ

aug–cc–pVTZ

DZVP (DFT)

DZVP2 (DFT)

6–311++G (3df,3pd)

6–311++G (2d,2p)

6–31+G*

6–311G**

6–31G**

Total

6–31G*

PBE

Basis set DFT b88+perd86 xper91 m06–L m06–hf m06–2x b31yp becke97 becke97–1 becke97–2 becke97–3 mpw1k pbe0 perdew LDA optx heth heth–407 m06 6–311++G (3df,3pd)

6–311++G (2d,2p)

6–31+G*

6–311G**

6–31G**

Carbon

6–31G*

PBE

Basis set

Figure 5.3 Error over small molecules in Richet et al. (5.3). Taken from P. Zarzycki and J. R. Rustad (A review of hydrogen, carbon, nitrogen, oxygen, sulphur, and chlorine stable isotope fractionation among gaseous molecules: a quantum chemical study, unpublished).

Computational Isotope Geochemistry DFT b88+perd86 xper91 m06–L m06–hf m06–2x b31yp becke97 becke97–1 becke97–2 becke97–3 mpw1k pbe0 perdew LDA optx heth heth–407 m06 cc–pVDZ

cc–pVTZ

aug–cc–pVDZ

aug–cc–pVTZ

DZVP (DFT)

DZVP2 (DFT)

cc–pVDZ

cc–pVTZ

aug–cc–pVDZ

aug–cc–pVTZ

DZVP (DFT)

DZVP2 (DFT)

6–311++G (3df,3pd)

6–311++G (2d,2p)

6–31+G*

6–311G**

6–31G**

Oxygen

6–31G*

PBE

Basis set DFT b88+perd86 xper91 m06–L m06–hf m06–2x b31yp becke97 becke97–1 becke97–2 becke97–3 mpw1k pbe0 perdew LDA optx heth heth–407 m06 6–311++G (3df,3pd)

6–311++G (2d,2p)

6–31+G*

6–311G**

6–31G**

Nitrogen

6–31G*

PBE

Basis set

Figure 5.3

(Continued)

167

168

Molecular Modeling of Geochemical Reactions 1.04

1.038

1.036

α34 (25°C)

BP86/AUG2

HCTH/ATZP BLYP/ATZP B3LYP/AUG2

1.034

OLYP/ATZP B3LYP/ATZP

MP2/AUG4

X3LYP/ATZP BP86/ATZP PW91/ATZP PBE/ATZP Experiment klochko et al. (2006)

BP86/AUG3

1.032

MP2/AUG2 MP2/AUG3

1.03

1.028 MP2/AUG3 (extrapolated)

1.026 0

5

10

15

20

25

30

35

40

# of solvating waters

Figure 5.4 Equilibrium constant for Reaction 5.3 for various model chemistries as a function of the number of solvating waters. Dashed lines are extrapolations.

B(OH)3 has six solvation waters because this was the hydration number determined in the AIMD calculations. Two systems are used for B OH 4 − , one having 8 waters and another having 11 waters. The B OH 4 11H2 O − cluster was chosen because AIMD simulations of the borate ion in solution give a second solvation shell of approximately 11 waters. B OH 4 8H2 O − was chosen because of its high symmetry (even B OH 4 11H2 O − with its C1 symmetry was too large for MP2/aug-cc-pVTZ calculation in 2008–2009 when this work was done). The conclusions that can be drawn from Figure 5.3 are extremely powerful. Based on the smaller clusters using MP2/aug-cc-pV(D,T,Q)Z, one arrives at a pretty convincing estimate of the MP2/aug-cc-pVTZ calculation for the B OH 4 11H2 O − system, even though the calculation could not actually be carried out. Further, the almost uniform decrease in going from B OH 4 11H2 O − to B OH 4 32H2 O − using DFT gives a fair level of confidence about what would be found had it been possible to do the MP2/aug-ccpVTZ calculations on the B(OH)3.32H2O and B OH 4 32H2 O − clusters. As far as the DFT calculations go, based on the BP86/AUG2 and BP86/AUG3 entries, one can anticipate a drop of 3–4‰ in going from the aug-cc-pVDZ to aug-cc-pVTZ basis. This would happen to bring the B3LYP/aug-cc-pVTZ estimate very close to the experimental value. So while it looks like the hybrid B3LYP functional is not as close to experiment as the pure functionals such as PBE, PW91, and BP86, part of the reason for that is that the 6-311+G∗∗ basis does not give converged results. In a sense, we made up for using a less reliable functional by also using a relatively poor basis set. This no doubt sounds like a lot of speculation, but this kind of activity is essential for getting some perspective on the calculations. Another result that can be anticipated is that after improving the basis sets, the B3LYP/aug-cc-pVTZ calculations are likely to be in better agreement with experiment than the MP2/aug-cc-pVTZ calculations. Because it is generally known that the

Computational Isotope Geochemistry

169

MP2/aug-cc-pVTZ calculations will be much more reliable in general for hydrogen-bonded systems, the apparently better agreement for the B3LYP calculations is an alert that there is likely something else going on that is not accounted for, such as anharmonicity. Thus, it might be found that a hybrid DFT functional with a converged basis set gives essentially the same answer as an MP2 calculation with a converged basis set plus an anharmonic correction. The bottom line is that it’s easy to stumble around with these kinds of small variations, occasionally finding the experimental value at various intermediate points, and then claiming success only to find that the same combination fails on a different system. “Right for the right reason” is a common phrase in computational chemistry, and it’s a good idea to keep this in mind when calculating isotope fractionation factors. The disconcerting conclusion from Figure 5.4 is that MP2/aug-cc-pVTZ, with at least 32 water molecules, is clearly what is needed, in terms of choices of model chemistries, yet these calculations are not going to be available for some time. Again, it is important to remember that all these considerations apply to doing calculations at near chemical accuracy. The calculations of Liu and Tossell (2005) and Zeebe (2005) played a huge role in providing motivation to go back and look at the 1.19 value for the equilibrium constant of Reaction 5.7, as taken from Kotaka and Kakihana (1977) (that turned out to have the mode assignments wrong for B OH 4 − ), and in providing the motivation for the key experiments of Byrne et al. (2006), which, at long last, provided a correct value as a foundation for paleoclimate studies. To my mind this is one of the premier successes of quantum chemistry in geology. Here we are actually approaching what I would call “molecular geology,” as opposed to “molecular geochemistry.” The latter implies the study of molecular-level processes in geochemistry, which may be interesting, but are usually far removed from saying very much about how we interpret the rock record. The term “molecular geology” is much more powerful and implies that these kinds of studies actually made a difference in how we interpret Earth history. The fact that, through the boron proxy, the community misinterpreted 20 million years of Earth history because of a wrongly assigned vibrational mode in B OH 4 − in the original work of Kotaka and Kakihana (1977) is a cogent illustration of the “house of cards” that is built up in these kinds of efforts. These investigators, who were working in the nuclear industry, never thought, of course, that their model would be used in this way.

5.6

Relationship to Empirical Estimates

At this point, the reader may be wondering why one does not use experimentally determined frequencies in systems for which those are available. The problem with this approach is that calculated harmonic frequencies are often compared to measured anharmonic frequencies without a harmonic correction. Unless the necessary measurements to extract actual harmonic frequencies from the experimental measurements have been done (and this is extremely rare for any system of geochemical interest), we cannot learn anything quantitative by making comparisons with anharmonic spectra. A great example is the work of Deines (2004) on the carbonate system. This paper presents a painstaking review of the literature, with the author critically compiling vibrational frequencies on a series of carbonate minerals with both the aragonite and calcite structures from multiple sources, and then using empirical force fields to estimate the isotopic substitution-induced shift for each of the frequencies, also including contributions from acoustic vibrations and external vibrations. This all must have taken years, and since the error in the measured vibrational frequencies is much smaller than the intrinsic error in the calculated frequencies, one might at first think that this is maybe a superior approach to first-principles calculations. The problem with this approach is that it is not a good idea to plug anharmonic frequencies in a harmonic partition function. Comparing

170

Molecular Modeling of Geochemical Reactions 10.0

mag dol rho ara cer mag (deines) dol (deines) ara (deines) rho (deines) cer (deines)

8.0

6.0

1000 In(β/βcal)

4.0

2.0 106/T2

0.0 0

2

4

6

8

10

12

14

–2.0

–4.0

–6.0

–8.0

–10.0

Figure 5.5 13C/12C fractionation relative to calcite for a series of carbonate minerals. Solid lines with markers are calculated from first-principles, and dashed lines are estimated empirically. ara, aragonite; cer, cerussite; dol, dolomite; mag, magnesite; rho, rhodochrosite. Taken from Deines (2004). Reproduced with permission of Elsevier.

Deines’ estimates to estimates from first-principles calculations in Figure 5.5, it is apparent that while there is overall reasonable qualitative agreement, there are important quantitative differences. Where there are data, such as for calcite–CO2 and aragonite–calcite, the first-principles calculations are in better agreement with experiment. In the end, the first-principles estimates for the 12 C − 13 C carbonate–calcite fractionation factors are probably more reliable, even though they took much less time and effort to make and even though there are substantial disagreements between the measured anharmonic vibrational frequencies and the calculated harmonic frequencies. Rustad and Bylaska (2007) initially tried such an approach from a computational point of view in the borate–boric acid system, using classical molecular dynamics to calculate the vibrational spectrum of these species in solution (by Fourier transforming the velocity autocorrelation function), running separate simulations of 10B(OH)3 (aq), 11B(OH)3 (aq), 11 B OH 4 − aq , and 11 B OH 4 − aq . The idea was that the simulations would reveal the vibrational spectrum very clearly, make vibrational mode assignments to determine the vibrational multiplicity, and calculate the harmonic partition function from these assignments. The advantage over other first-principles estimates for the boron isotope fractionation factor (such as carried out Liu and Tossell, 2005) would be that the results would sample many configurations for each aquo ion, not just one, as had been done in previous work. The work was done extremely carefully with sophisticated Monte Carlo uncertainty estimates, and the

Computational Isotope Geochemistry

171

authors thought the results were going to be revolutionary. Instead, they were ridiculous. The fractionation factor went in the opposite direction, with B OH 4 − being enriched in the heavy isotope. When the authors tried extracting a single configuration from the molecular dynamics simulation, optimized it in periodic boundary conditions, and calculated the frequencies as outlined previously in the harmonic approximation, a value of 1.028 was obtained with the PBE functional and a planewave basis with a 90 hartree cutoff, agreeing almost perfectly with the Hartree–Fock 6-31G∗ calculation of Liu and Tossell (2005). Figure 5.4 shows that this level of agreement was to some extent fortuitous; however, it was deflating to go through all that work only to find that all the fancy techniques were superfluous and, in the end, reproduce almost exactly the very calculation on which we were trying to improve. But the study served as an illustrative lesson reinforcing how important it is to use harmonic frequencies in a harmonic partition function. Looking at the real vibrational density of states in aqueous solution was, however, instrumental in discovering the misassigned vibrational modes in the Kotaka and Kakihana (1977) model that helped dislodge the erroneous 1.19 that was latched onto by the boron pH proxy community, so the effort was not totally fruitless. The issue of harmonic frequencies versus measured frequencies brings up an additional point about the practice of scaling calculated frequencies to better match measured frequencies. It has long been realized that there are systematic errors in calculated frequencies and there has been much work done on finding ways to correct these errors through the use of scaling factors (Scott and Radom, 1996). In doing this for isotope exchange equilibria, one has to be sure to use scaling factors that have been designed to recover harmonic frequencies or better zero-point energies. It is probably not a good idea to, for example, find one’s own scaling factors by correcting calculated harmonic frequencies to measured anharmonic frequencies and then using the “corrected” frequencies to calculate the harmonic partition function.

5.7

Beyond the Harmonic Approximation

The next big step in the ab initio computation of isotope fractionation factors is to incorporate anharmonicity. This is especially necessary for hydrogen isotopes where the predictions from harmonic theory are not very useful. One way to go beyond the harmonic approximation is to use path integral molecular dynamics (PIMD) methods to treat the quantum aspects of the dynamics. In AIMD, the forces are calculated from quantum mechanics, but the nuclei move classically in response to these forces. There are no isotope effects in classical AIMD simulations at equilibrium, as the equilibrium fractionation is strictly a quantum mechanical effect. To represent quantum delocalization, PIMD treats each nucleus as many quantum replicas connected by effective temperatureand mass-dependent springs, each of the replicas behaving as classical particles. So instead of just one, for example, H atom, one has, for example, 10 or 20 replicas all slightly displaced from one another. Quantum dynamics is the whole point of PIMD, and in these calculations the mass of the exchanging atom can be gradually perturbed from one isotope into another and free energy of this imposed perturbation can be evaluated from what are now standard molecular dynamics free energy techniques (of course, one could do this also in a classical system, gradually change an H atom into a D atom, but the energy change would be zero). This is a highly elegant and general approach, free from any assumptions about harmonicity and free of the myriad complexities of explicit anharmonic vibrational corrections, but is only now beginning to be applied to the simplest systems (Webb and Miller, 2014). While the initial results are extremely promising, time will tell whether the pitfalls associated with achieving equilibrium and statistical convergence of the free energy evaluation would ever make this a generally feasible approach.

172

Molecular Modeling of Geochemical Reactions

Another way to account for anharmonicity is to make anharmonic corrections to the zero-point energies. These are very complicated but have been coded up as a “black box” add-on to the Gaussian suite of electronic structure programs. A good discussion of these methods from a geochemical perspective can be found in Liu et al. (2010). About the only thing that can be said for certain is that one definitely must include anharmonic effects for hydrogen–deuterium isotope fractionation. For heavier isotopes, it is not possible at this time to make any general statements. It seems likely that the treatment of anharmonicity will end up being another factor in the basis set, electron–electron model, and environmental representation grid that has to be systematically explored, on a variety of system sizes, such as summarized in Figure 5.4, to draw reliable quantitative estimates from first-principles calculations.

5.8 Kinetic Isotope Effects Kinetic isotope effects have also been widely studied with computational methods. In the harmonic approximation, no really new principles are involved. The heavy isotopes bury themselves more deeply in the potential well associated with the reactants and are thus harder to lift out and slower to react. Of course, they also bury themselves more deeply into the transition state; however, in the transition state the bonding is much weaker and the effect is much less, and so the net result is that the heavy isotopes tend to react more slowly. The barrier heights may be more challenging to predict and the transition states more challenging to locate. In addition, there is an effect that comes about because the efficiency of the transmission across the barrier is, in principle, dependent on mass. For example, if we somehow pushed the whole system slowly up to the top of the barrier from the reactant side and then just stopped, let go, and observed which direction it would fall, the probability that it would ultimately fall to the product side of the reaction depends not only on how the system is pushed to the top but also depends, in general, on the masses of all the atoms in the system. Thus, there is an isotope effect associated with the probability of actually crossing the barrier, as well as an isotope effect on the barrier height itself. This concept was recently applied by Hofmann et al. (2012) to look at isotope effects on water exchange kinetics. In a somewhat similar vein, Kavner et al. (2008) have worked out a generalization of the Marcus theory of electron transfer to include isotope fractionation driven by redox reactions. Such process as mass dependence of barrier transmission can in principle be simulated directly using quantum dynamics methods such as PIMD discussed previously. Certainly it will be exciting to see what kinds of information can be obtained from such simulations.

5.9 Summary and Prognosis In a work such as this, one feels compelled to make some sort of summary statement, which usually includes some view of what the future of the subject at hand might look like. One could say overall that first-principles calculations have given the community fast access to reasonably accurate models that, in the past, had to be constructed by hand in a relatively labor-intensive process that required a great deal of ingenuity on the part of the investigator (i.e., Deines, 2004; Richet et al., 1977). More importantly, the calculations allow modeling of arbitrary environments, such as surfaces, for which it would have been almost impossible to build empirical force fields from vibrational spectroscopy. Even aqueous environments are too difficult for the construction of reliable empirical models, as shown in the boron isotope example discussed previously. Moreover, the first-principles calculations often (though not always; see Richet et al., 1977) improve significantly on the empirical approach by

Computational Isotope Geochemistry

173

giving harmonic vibrational frequencies directly, without requiring extraction of harmonic frequencies from anharmonic measurements. It is clear, however, that the field has to evolve before the prediction of equilibrium fractionation factors to 0.1–1‰ accuracy can be achieved for general systems (as shown in Figure 5.4). Given the inherent difficulties in applying quantitative ab initio calculations (which grow as at least the fifth power of system size) to complex geochemical systems, it seems that improvements will come from DFT. Computational geochemists will pay close attention to the research going on in the area of improving DFT calculations on systems with hydrogen bonding and dispersion interactions, as these are likely to improve isotope fractionation calculations for mineral/aqueous systems. In addition, one could imagine a concerted program to invent a new DFT functional that was specifically constructed for the purpose of accurate calculation of isotope fractionation factors in geochemical systems. Such an effort would naturally have to be restricted, for example, to oxide-water systems (oxide here being broadly defined to include oxides, silicates, carbonates, sulfates, phosphates). First-principles calculations have already helped advance the science of isotope geochemistry, having played an important role in helping formulate the field of clumped isotopes (Eiler, 2007; Schauble et al., 2006) as well as elucidating the role of nuclear volume effects in heavy elements (Schauble, 2007). A fascinating step in analytical work is the ability to measure position-specific isotopic compositions in molecules and minerals (Eiler, 2013). Such measurements promise an entirely new window into geochemical processes. Interpretation of position-specific isotope signatures can be greatly aided by calculations (Rustad, 2009; Rustad and Zarzycki, 2008); no doubt we can look to much more work in this area in future years.

Acknowledgments The author would like to thank Brooke H. Dallas, Jason Boettger, and James D. Kubicki for insightful reviews and suggestions and Piotr Zarzycki for providing Figure 5.3.

References Anisimov V. I., Aryasetiawan F., and Lichtenstein A. I. (1997) First-principles calculations of the electronic structure and spectra of strongly correlated systems: the LDA+U method. Journal of Physics: Condensed Matter 9, 767–808. Badger R. M. (1934) A relation between internuclear distances and bond force constants. Journal of Chemical Physics 2, 128–132. Bartlett R. J. and Musiał M. (2007) Coupled-cluster theory in quantum chemistry. Reviews of Modern Physics 79, 291–352. Bigeleisen J. and Mayer M. G. (1947) Calculation of equilibrium constants for isotopic exchange reactions. Journal of Chemical Physics 15, 261–267. Bloechl P. E. (1992) Projector augmented-wave method. Physical Review B 50, 17953–17979. Boese A. D. and Handy N. C. (2001) A new parametrization of exchange-correlation generalized gradient approximation functionals. Journal of Chemical Physics 114, 5497–5503. Byrne R. H., Yao W., Klochko K., Tossell J. A., and Kaufman A. J. (2006) Experimental evaluation of the isotopic exchange equilibrium 10B(OH)3 + 11B(OH)4− = 11B(OH)3 + 10B(OH)4− in aqueous solution. Deep Sea Research Part I: Oceanographic Research Papers 53, 684–688. Cramer C. J. (2004) Essentials of Computational Chemistry: Theories and Models, 2nd Edition, John Wiley & Sons, Ltd: Chichester, UK, 618 pages. Deines P. (2004) Carbon isotope effects in carbonate systems. Geochimica et Cosmochimica Acta 68, 2659–2679. Demichelis R., Raiteri P., Gale J. D., Quigley D., and Gebauer D. (2011) Stable prenucleation mineral clusters are liquid-like ionic polymers. Nature Communications 2, 590. Dunning T. H., Jr. (1989) Gaussian basis sets for use in correlated molecular calculations. I. The atoms boron through neon and hydrogen. Journal of Chemical Physics 90, 1007–1023.

174

Molecular Modeling of Geochemical Reactions

Eiler J. M. (2007) “Clumped-isotope” geochemistry—the study of naturally-occurring, multiply-substituted isotopologues. Earth and Planetary Science Letters 262, 309–327. Eiler J. M. (2013) The isotopic anatomies of molecules and minerals. Annual Review of Earth and Planetary Sciences 41, 411–441. Fornaro T., Biczysko M., Monti S., and Barone V. (2014) Dispersion-corrected DFT approaches for anharmonic vibrational frequency calculations: nucleobases and their dimers. Physical Chemistry Chemical Physics 16, 10112–10128. Frierdich A. J., Beard B. L., Reddy T. R., Scherer M. M., and Johnson C. M. (2014) Iron isotope fractionation between aqueous Fe(II) and goethite revisited: new insights based on a multi-direction approach to equilibrium and isotopic exchange rate modification. Geochimica et Cosmochimica Acta 139, 383–398. Frisch M. J., Trucks G. W., Schlegel H. B., Scuseria G. E., Robb M. A., Cheeseman J. R., Montgomery J. A., Jr., Vreven T., Kudin K. N., Burant J. C., Millam J. M., Iyengar S. S., Tomasi J., Barone V., Mennucci B., Cossi M., Scalmani G., Rega N., Petersson G. A., Nakatsuji H., Hada M., Ehara M., Toyota K., Fukuda R., Hasegawa J., Ishida M., Nakajima T., Honda Y., Kitao O., Nakai H., Klene M., Li X., Knox J. E., Hratchian H. P., Cross J. B., Bakken V., Adamo C., Jaramillo J., Gomperts R., Stratmann R. E., Yazyev O., Austin A. J., Cammi R., Pomelli C., Ochterski J. W., Ayala P. Y., Morokuma K., Voth G. A., Salvador P., Dannenberg J. J., Zakrzewski V. G., Dapprich S., Daniels A. D., Strain M. C., Farkas O., Malick D. K., Rabuck A. D., Raghavachari K., Foresman J. B., Ortiz J. V., Cui Q., Baboul A. G., Clifford S., Cioslowski J., Stefanov B. B., Liu G., Liashenko A., Piskorz P., Komaromi I., Martin R. L., Fox D. J., Keith T., Al-Laham M. A., Peng C. Y., Nanayakkara A., Challacombe M., Gill P. M. W., Johnson B., Chen W., Wong M. W., Gonzalez C., and Pople J. A. (2004) Gaussian 03, Revision C.02, Gaussian, Inc.: Wallingford, CT. Gibbs G. V. (1982) Molecules as models for bonding in silicates. American Mineralogist 67, 421–450. Godbout N., Salahub D. R., Andzelm J., and Wimmer E. (1992) Optimization of Gaussian-type basis sets for local spin-density calculations: 1. Boron through neon, optimization technique and validation. Canadian Journal of Chemistry 70, 560–571. Handler R. M., Beard B. L., Johnson C. M., and Scherer M. M. (2009) Atom exchange between aqueous Fe(II) and goethite: an Fe isotope tracer study. Environmental Science and Technology 43, 1102–1107. Helgaker T., Jorgensen P., and Olsen J. (2000) Molecular Electronic Structure Theory, John Wiley & Sons, Inc.: New York, 944 pages. Helgaker T., Klopper W., and Tew D. P. (2008) Quantitative quantum chemistry. Molecular Physics 106, 2107–2143. Hemming N. G. and Hanson G. N. (1992) Boron isotopic composition and concentration in modern marine carbonates. Geochimica et Cosmochimica Acta 56, 537–543. Henderson M. A., Joyce S. A., and Rustad J. R. (1998) Interaction of water with the (1 × 1) and (2 × 1) surfaces of α-Fe2O3 (012). Surface Science 417, 66–81. Hofmann A. E., Bourg I. C., and DePaolo D. J. (2012) Ion desolvation as a mechanism for kinetic isotope fractionation in aqueous systems. Proceedings of the National Academy of Sciences of the United States of America 109, 18689–18694. Kavner A., John S. G., Sass S., and Boyle E. A. (2008) Redox-driven stable isotope fractionation in transition metals: application to Zn electroplating. Geochimica et Cosmochimica Acta 72, 1731–1741. Kendall R. A., Dunning T. H., Jr., and Harrison R. J. (1992) Electron-affinities of the 1st-row atoms revisited— systematic basis-sets and wave-functions. Journal of Chemical Physics 96, 6796–6809. Kieffer S. W. (1982) Thermodynamics and lattice vibrations of minerals: 5. Applications to phase equilibria, isotopic fractionation, and high-pressure thermodynamic properties. Reviews of Geophysics and Space Physics 20, 827–849. Klochko K., Kaufman A. J., Yao W. S., Byrne R. H., and Tossell J. A. (2006) Experimental measurement of boron isotope fractionation in seawater. Earth and Planetary Science Letters 248 (1–2), 276–285. 10.1016/j. epsl.2006.05.034 Kohn W. and Sham L. J. (1965) Self-consistent equations including exchange and correlation effects. Physical Review 140, A1133–A1138. Kotaka M. and Kakihana H. (1977) Equilibrium constants for boron isotope-exchange reactions. Bulletin of the Research Laboratory for Nuclear Reactors 2, 1–12. Liu Y. and Tossell J. A. (2005) Ab initio molecular orbital calculations for boron isotope fractionations on boric acids and borates. Geochimica et Cosmochimica Acta 69, 3995–4006. Liu Q., Tossell J. A., and Liu Y. (2010) On the proper use of the Bigeleisen–Mayer equation and corrections to it in the calculation of isotopic fractionation equilibrium constants. Geochimica et Cosmochimica Acta 74, 6965–6983. Møller C. and Plesset M. S. (1934) Note on an approximation treatment for many-electron systems. Physical Review 46, 618–622.

Computational Isotope Geochemistry

175

Press W. H., Flannery B. P., and Teukolsky S. A. (1986) Numerical Recipes, Cambridge University Press: Cambridge/New York. Richet P., Bottinga Y., and Javoy M. (1977) A review of hydrogen, carbon, nitrogen, oxygen, sulphur, and chlorine stable isotope fractionation among gaseous molecules. Annual Review of Earth and Planetary Sciences 5, 65–110. Rustad J. R. (2009) Ab initio calculation of the carbon isotopic signatures of amino acids. Organic Geochemistry 40, 720–723. Rustad J. R. and Bylaska E. J. (2007) Ab initio calculation of isotopic fractionation in B(OH)3(aq) and B(OH)4− (aq). Journal of the American Chemical Society 129, 2222–2223. Rustad J. R. and Dixon D. A. (2009) Prediction of iron-isotope fractionation between hematite (α-Fe2O3) and ferric and ferrous iron in aqueous solution from density functional theory. Journal of Physical Chemistry A 113, 12249–12255. Rustad J. R. and Yin Q.-Z. (2009) Iron isotope fractionation in the Earth’s lower mantle. Nature Geoscience 2, 514–518. Rustad J. R. and Zarzycki P. (2008) Calculation of site-specific carbon isotope fractionation in pedogenic oxide minerals. Proceedings of the National Academy of Sciences of the United States of America 105, 10297–10301. Rustad J. R., Bylaska E. J., Jackson V. E., and Dixon D. A. (2010a) Calculation of boron-isotope fractionation between B(OH)3 (aq) and B(OH)4− (aq). Geochimica et Cosmochimica Acta 74, 2843–2850. Rustad J. R., Casey W. H., Yin Q.-Z., Bylaska E. J., Felmy A. R., Bogatko S. A., Jackson V. E., and Dixon D. A. (2010b) Isotopic fractionation of Mg2+, Ca2+, and Fe2+ (aq) with carbonate minerals. Geochimica et Cosmochimica Acta 74, 6301–6323. Rustad J. R., Nelmes S. L., Jackson V. E., and Dixon D. A. (2008) Quantum-chemical calculations of carbonisotope fractionation in CO2 (g), aqueous carbonate species and carbonate minerals. Journal of Physical Chemistry A 112, 542–555. Schauble E. A. (2004) Applying stable isotope fractionation theory to new systems. Reviews in Mineralogy and Geochemistry 55, 65–111. Schauble E. A. (2007) Role of nuclear volume in driving equilibrium stable isotope fractionation of mercury, thallium, and other very heavy elements. Geochimica et Cosmochimica Acta 71, 2170–2189. Schauble E. A., Ghosh P., and Eiler J. M. (2006) Preferential formation of bonds in carbonate minerals, estimated using first-principles lattice dynamics. Geochimica et Cosmochimica Acta 70, 2510–2529. Schwarz K. and Blaha P. (2003) Solid state calculations using WIEN2K. Computational Materials Science 28, 259–273. Scott A. P. and Radom L. (1996) Harmonic vibrational frequencies: an evaluation of Hartree-Fock, MollerPlesset, quadratic configuration interaction, density functional theory and semiempirical scale factors. Journal of Physical Chemistry A 100, 16502–16513. Szabo A. and Ostlund N. S. (1996) Modern Quantum Chemistry: Introduction to Advanced Electronic Structure Theory, Dover Publications: Mineola, NY, 480 pages. Troullier N. and Martins J. L. (1991) Efficient pseudopotentials for plane-wave calculations. Physical Review B 43, 1993–2006. Tuma C. and Sauer J. (2006) Treating dispersion effects in extended systems by hybrid MP2 : DFT calculations—protonation of isobutene in zeolite ferrierite. Physical Chemistry Chemical Physics 8, 3955–3965. Urey H. C. (1947) The thermodynamic properties of isotopic substances. Journal of the Chemical Society, 562–581. Valiev M., Bylaska E. J., Govind N., Kowalski K., Straatsma T. P., van Dam H. J. J., Wang D., Nieplocha J., Apra E., Windus T. L., and de Jong W. A. (2010) NWChem: a comprehensive and scalable open-source solution for large scale molecular simulations. Computer Physics Communications 181, 1477–1489. Webb M. A. and Miller T. F., III (2014) Position-specific and clumped stable isotope studies: comparison of the Urey and path-integral approaches for carbon dioxide, nitrous oxide, methane, and propane. Journal of Physical Chemistry A 118, 467–474. Williams A. G. and Scherer M. M. (2004) Spectroscopic evidence for Fe(II)-Fe(III) electron transfer at the iron oxide-water interface. Environmental Science and Technology 38, 4782–4790. Wolfsberg M., Hook W. A., Paneth P., and Rebelo L. P. N. (2009) Isotope Effects in the Chemical Geological and Bio Sciences, Springer: Dordrecht/New York, 466 pages. Zeebe R. E. (2005) Stable boron isotope fractionation between dissolved B(OH)3 and B(OH)4−. Geochimica et Cosmochimica Acta 69, 2753–2766.

6 Organic and Contaminant Geochemistry Daniel Tunega,1 Martin H. Gerzabek,1 Georg Haberhauer,1 Hans Lischka,2,3 and Adelia J. A. Aquino1,2 1

6.1

Institute for Soil Research, University of Natural Resources and Life Sciences, Vienna, Austria 2 Department of Chemistry and Biochemistry, Texas Tech University, Lubbock, TX, USA 3 Institute for Theoretical Chemistry, University of Vienna, Vienna, Austria

Introduction

Environmental pollution is one of the major problems of our civilization mainly due to urban, industrial and agricultural human activities. Organic and contaminant geochemistry comprises a broad range of studies of physical and chemical processes influencing behavior, distribution, and fate of organic species and contaminants on the Earth and in extraterrestrial materials. Characterization and understanding of these processes is crucial from many aspects, for example, for determination of environmental risk or for control and management of surface and ground water resources, hazardous wastes, contaminated soils and sediments, and geologic repositories (Appelo and Postma, 2005; Berkowitz et al., 2008). Soil represents a main buffer and filter in the environment, through which pollutant substances can migrate and enter into a food chain or ground water reservoirs (Sposito, 1984; Sparks, 1999). Soil is unconsolidated, complex and heterogeneous matter consisting of inorganic and organic materials, water, air, and living organisms and the overall behavior of the soil is determined by the chemical, physical and biological properties of its components (Wolfe and Seiber, 1993; Sparks, 1995; Hornsby et al., 1996; Yaron et al., 1996). Therefore, polluting substances can undergo complex physical, chemical and biological transformation processes in the soil. The behavior in the environmental systems also depends on their chemical nature and properties. For example, nonpolar hydrophobic polycyclic aromatic hydrocarbons (PAHs) can be sequestered in the environment due to strong interactions with carbonaceous materials often appearing in soils and sediments (e.g., black carbon) (Lohmann et al., 2005). There is a special focus on persistent organic pollutants of natural or anthropogenic origin with long half-lives. A typical example is Molecular Modeling of Geochemical Reactions: An Introduction, First Edition. Edited by James D. Kubicki. © 2016 John Wiley & Sons, Ltd. Published 2016 by John Wiley & Sons, Ltd.

178

Molecular Modeling of Geochemical Reactions

dichlorodiphenyltrichloroethane (DDT) having a soil half-life in a range of 2–15 years (US Environmental Protection Agency, 1989; Augustijn-Beckers et al., 1994). These substances are usually resistant to chemical, photolytic, or biodegradation processes and exhibit typically high lipid solubility (low water solubility) leading to their bioaccumulation in fatty tissues of living organisms. As soils are open systems, external factors such as temperature, radiation, precipitation, wetting/ drying cycles, or soil management affect the behavior and transportation of pollutants in the environment. The activity of microorganisms in soils significantly contributes to the chemical transformations of polluting substances. All those physical, chemical, and biological processes may be complicated and obscured appearing at very different time and spatial scales. For example, a photochemical degradation of some pollutant can be usually a fast process but produced metabolites (eventually with a higher toxicity than the parent material) can have much higher resistance to further (bio)mineralization processes. Moreover, the simultaneous action of various processes can have a synergetic effect (positive or negative) on persistence, sequestration, bioavailability, and degradation of pollutants (Konda et al., 2002). Although a variety of physicochemical processes affect the fate of contaminants in natural environments, the sorption to the solid soil matrix is one of the most important phenomena determining other processes such as transport, diffusion, runoff, leaching, or volatilization. Sorption/desorption processes comprise a number of subprocesses with different kinetic rates and operating over broad time scales ranging from hours through years. Owing to the complicated structure and architecture of soil matrices (e.g., existence of pores from nano- to mesopore sizes), standard adsorption processes are often accompanied by other processes such as pore condensation. Sorption/desorption characteristics of soils and sediments are of particular concern because of further contamination, for example, of groundwater systems or for a design of effective waste management and remediation techniques (Rombke et al., 1996; Weber et al., 2001). Although soils are structurally and chemically very complex heterogeneous natural geosorbents, classical adsorption models such as Langmuir or Freundlich are often used to characterize adsorption behavior of the contaminants in soils. There are more complex theories and models for simulation of sorption processes in soils, and details can be found, for example, in Sparks (1999), Sparks et al. (1999), and Weber et al. (1991). Coexistence of inorganic mineral particles, organic soft matter (natural/soil organic matter, (NOM/SOM), and their mutual associates and/or aggregates represents a base for highly heterogeneous surfaces with numerous chemically reactive functional groups, structural defects, hydrophilic and hydrophobic sites that are reflecting different binding mechanisms of adsorbed pollutant species, and, consequently, also great differences in adsorption energies. The origin of the binding forces in soil adsorption processes is governed by the different types of interaction mechanisms such as ionic, hydrogen, and covalent bonding, charge-transfer and electron donor-acceptor mechanisms, van der Waals forces, ligand exchange, complexation, hydrophobic bonding, partitioning, or cooperative interactions among adsorbed molecules (Weber et al., 1991; Senesi, 1992; Delle Site, 2001; Konda et al., 2002). The formation, structure, and stability of adsorption complexes are strongly affected by the presence of water in soil systems. Usually, water is present in soils in unsaturated conditions and its content is strongly variable depending upon external conditions (humidity, precipitation). Water can be concentrated in hydrophilic domains of soil matrices (e.g., polar groups of NOM or mineral surfaces), and its strong solvation potential can destabilize, for example, hydrogenbonded complexes. On the other hand, nonpolar groups and functionalities (e.g., aliphatic chains, aromatic systems) can form water repellent hydrophobic domains able to accumulate nonpolar species (Schaumann, 2005; Schaumann and LeBoeuf, 2005; Schaumann and Bertmer, 2008; Schneckenburger et al., 2012). It is difficult to distinguish and quantify a contribution of individual (pure) components to overall sorption processes from the standard adsorption studies on natural, complex, and heterogeneous geosorbents such as soils and sediments. Therefore, there is a necessity of case studies with better

Organic and Contaminant Geochemistry

179

defined probe materials such as pure minerals, more homogeneous organic matrices (e.g., cellulose), or organic phases extractable from soils (e.g., humic acids (HAs)). However, the traditional batch sorption studies usually provide adsorption isotherms and corresponding adsorption constants (e.g., KL, KF) or partitioning coefficients (e.g., Kd, Kow, KOC) that are macroscopic characteristics telling little about sorption mechanisms at a molecular scale. Traditional large-scale experimental studies are more and more extended and combined with specific studies using advanced experimental techniques able to descend to the molecular scale with an effort to explain fundamental processes and to understand molecular mechanisms that control disposition and interactions of chemical contaminants in the natural geosorbents. These molecular-scale investigations comprise a variety of spectroscopic, microscopic, and imagining techniques such as AFM, STM, XPS, ESEM, SIMS, or EXAFS that contribute significantly to revealing the molecular origin of interactions of contaminants that can be a solid basis for an accurate prediction of the fate of these chemicals in geosorbent systems. Several comprehensive reviews on application of spectroscopic and microscopic techniques in molecular environmental geochemistry are available in the literature (O’Day, 1999; Fenter et al., 2002; Al-Abadleh and Grassian, 2003; Henderson et al., 2014). The complex character of interactions of contaminants with natural geosorbents and many factors that can affect them represent a great challenge for experimental investigations. In the effort to understand elementary mechanisms of these interactions at molecular scale, computational chemistry offers powerful and effective tools. They can contribute significantly to the exploration and identification of basic mechanisms controlling driving forces of kinetics and thermodynamics of contaminants in geochemical processes. Moreover, the achievements in computational geochemistry can be of more general importance for various branches of science and vice versa. For example, the statistical QSAR method, which has been used for years in the prediction of physicochemical properties or theoretical molecular descriptors (e.g., toxicity, carcinogenicity) of chemicals in organic and pharmacologic chemistry, was applied successfully in a prediction of the free energy of hydration and aqueous solubility for a set of pesticide compounds (Klamt et al., 2002, 2009). The knowledge and experiences gathered during the last decades with computer simulations in many fields of chemistry and physics (e.g., material design, biochemistry, pharmacology, mineralogy) can be successfully applied in organic and contaminant geochemistry. In addition to exploring basic mechanisms, molecular simulations can contribute to the explanation and interpretation of the experimental observations that are frequently complicated and ambiguous. Therefore, theoretical methods represent an increasingly powerful complement for experimentalists. For example, Bucheli and Gustaffson (2000, 2003) presented Kd values for the adsorption of PAHs and PBCs by soot, an important natural geosorbent in soils and sediments. These studies were complemented with quantum chemical (QC) calculations performed by Kubicki (2005, 2006) finding a good correlation between calculated adsorption energies (ΔEads) and Kd values. Another advantage of the molecular modeling methods can be found in their capability to predict physical and chemical properties of organic contaminants in cases when experiments would be impractical or too time-consuming. Computer simulations can be used in building a variety of physical models and hypotheses and their testing and validation. One of the biggest challenges in molecular modeling is bridging different temporal and spatial scales (i.e., the size of the systems and the length of processes) in the experiment and in the molecular simulations. A typical example is the prediction of thermodynamic quantities that are classical macroscopic observables. 6.1.1

Review Examples of Molecular Modeling Applications in Organic and Contaminant Geochemistry

The previous section indicates that there is a broad field open for the application of the molecular simulation methods in the research of geochemical surfaces and interfaces. In fact, there are

180

Molecular Modeling of Geochemical Reactions

numerous papers on molecular simulations regarding the interactions of a variety of chemical substances (e.g., heavy metals, radionuclides, small organic molecules, amino acids, peptides, DNA) with various geosorbents available, and it is practically impossible to report all of them. This section focuses on representative examples where molecular simulation methods have been used in the investigation of mechanism of interactions between organic contaminants and geochemically relevant geosorbents such as typical minerals, NOM, and black carbon. The organic contaminants studied in the presented examples comprise pesticides, fragments and components of crude oils and oil sands, PAHs, antibiotics, highly energetic explosives, and organic toxicants. From this review our own papers are excluded as they are discussed in detail in Section 6.3. 6.1.1.1

Pesticides

The adsorption of two pesticides (atrazine and dinitro-ortho-cresol (DNOC)) on montmorillonite surfaces was modeled by means of a classical force field molecular dynamics (FF-MD) approach (see Chapter 2) (Yu et al., 2003). It was shown that in the absence of water, organic compounds form complexes of a maximal contact with the surface. In the presence of a sufficient amount of water, the adsorbed molecules are strongly destabilized and immersed in the aqueous phase. An experimental adsorption study of dioxin by smectite clays saturated with different exchangeable cations was complemented by the FF-MD simulations (Liu et al., 2012a). The sorption of dioxin depends strongly on the hydration of the exchangeable cations, the surface charge density of the smectite clay, and the location (tetrahedral vs. octahedral) of isomorphous substitution in the clay. MD simulations confirmed the observed trends and also showed the important role of dioxin oxygen in the surface complexation. The FF-MD simulations were used to estimate relative clay–organic interaction enthalpies for a series of nitroaromatic solutes and hydrated, K-saturated montmorillonite, for comparison with experimental adsorption isotherm data for the same clay–nitroaromatic systems (Aggarwal et al., 2007). The trend of the computed interaction enthalpies (e.g., −234 ± 17 kJ/mol for trinitrobenzene and −154 ± 16 kJ/mol for p-nitrobenzene) agreed modestly well with the trend of adsorption maxima from the experiments. The mechanism and relatively high sorption of dibenzo-p-dioxin (DD) by Cs-saponite were studied experimentally and theoretically (Liu et al., 2009). Ab initio calculations explained geometrical structures of the DD molecules in the interlayer space and showed that Cs+ interacts with dioxin ring oxygen atoms and benzene ring π-electrons. Combined Monte Carlo (MC) and MD techniques have been used in the study of atrazine behavior in saturated sands (Cosoli et al., 2010). Diffusion coefficients, binding energies, and concentration profiles have been determined for interactions of atrazine with silica. The results confirm a moderate atrazine adhesion onto silica framework and a more favorable tendency to bind with water. Density functional theory (DFT) (see Chapter 1) was used in simulations of adsorption of 2-methyl-4-chlorophenoxyacetic acid (MCPA) and 2-(4-chlorophenoxy)-2-methylpropanoic acid both in neutral and ionized forms on a model surface of muscovite (Ramalho et al., 2013). The ionized adsorbates interact more strongly with the surface than do their neutral forms. Site-specific bonding of herbicides in SOM complexes was performed by Négre et al. (2001). In the work the interactions of the imidazole herbicides imazapyr, imazethapyr, and imazaquin with HAs and SOM models were investigated both experimentally and by molecular mechanics (MM) calculations. The modeling calculations showed that there are stable complex structures involving mostly van der Waals and electrostatic interactions in agreement with hydrophobic bonding and charge transfer interface as specified by the experimental findings in the work.

Organic and Contaminant Geochemistry

181

The formation processes of complexes between a three dimensional (3D) model of HA and a successively increasing number (1–30) of diethyl phthalate (DEP) molecules were studied by means of classical MM (Schulten et al., 2001). The intermolecular forces that stabilize the complexes were found to be mainly driven by hydrogen bonds and van der Waals interactions. It was observed that absorption of DEP molecules occurs in internal voids of the HA structure for low concentrations whereas for high concentrations, adsorption prevails on the outer surface of the HA structure. Blotevogel et al. (2010) investigated redox-promoted degradation processes to predict potential degradation pathways by using DFT calculations. The authors identified the thermodynamic conditions required for the redox reactions taking hexamethylphosphoramide (HMPA) as a test case. HMPA is a broadly used solvent with the possibility to act as a groundwater pollutant. Later a more detailed DFT study was performed with the aim to predict the HMPA persistence in aqueous phase (Blotevogel et al., 2011). The authors concluded that hydrolysis of the P─N bond is the only thermodynamically stable way that could drive to the HMPA degradation under reducing conditions. 6.1.1.2

Antibiotics

Frequently used in the farming praxis, antibiotics represent a serious potential risk for a contamination of soils. In several papers molecular simulations were used in the study of interactions of antibiotics with soil components contributing to understanding their biological availability and retention in natural and engineered soil environments. The binding mechanism of one of tetracycline antibiotics (oxytetracycline) on smectite clays was studied experimentally (X-ray, IR, and NMR) accompanied by MC molecular simulations (Aristilde et al., 2010). An impact of pH was observed on the binding mechanism that involves the adsorbed moiety via the protonated dimethylamino group. MC simulations indicated that interlayer spacing and charge localization of clay layers dictate favorable binding conformations of the intercalated tetracycline molecules facilitating multiple interactions, in agreement with the spectroscopic data. Aristilde and Sposito (2010) studied interactions between a widely used fluoroquinolone antibiotic ciprofloxacin (Cipro) and a well-tested molecular model of humic substance (HS) (Schulten and Schnitzer, 1993; Sutton et al., 2005) in a polar environment through classical MD simulations. The authors presented a detailed analysis characterizing adsorption of Cipro by both protonated and metal cation-bearing HS. The results stressed the capacity of Cipro to participate in multiple H-bonding through their polar groups demonstrating its high affinity for the HS especially if its acidic groups are fully protonated. At circumneutral pH, HS-complexed divalent metal cations could increase binding through carboxylate groups of the antimicrobial structure forming a ternary HS–metal complex. 6.1.1.3

Explosives and Organic Toxicants

Sarin and soman, organophosphorus compounds, are nerve agents used in chemical weapons. Their interactions with magnesium oxide and clay minerals were studied using cluster models for minerals and Møller–Plesset perturbation theory to the second order (MP2) and B3LYP methods (both combined with the 6–31G(d) basis set) (Michalková et al., 2004a, b, 2006). The cluster models of clay minerals represented tetrahedral edges. The charge of the systems and a termination of the mineral fragment determine the strength of the intermolecular interactions. In the neutral complexes, sarin and soman are physisorbed on the mineral fragments through hydrogen bonds. The covalent chemical bonding is formed between a phosphorus atom of sarin/soman and oxygen atom from the negatively charged fragments. Based on the calculated Gibbs free energies, the authors predicted that only chemically bound complexes are thermodynamically stable at room temperature.

182

Molecular Modeling of Geochemical Reactions

In the work by Tsendra et al. (2014), a cluster approach extended to the quantum mechanical/ molecular mechanical (QM/MM) methodology was applied using several density functionals and the MP2 method to simulate the adsorption of selected nitrogen-containing compounds, 2,4,6-trinitrotoluene (TNT), 2,4-dinitrotoluene (DNT), 2,4-dinitroanisole (DNAN), and 3-nitro1,2,4-triazole-5-one (NTO) on the hydroxylated (100) surface of α-quartz. The structural properties were calculated using the M06-2X and PBE functionals both including the dispersion correction (D3) in the calculations. The MP2 method was used to calculate the adsorption energies. Although all molecules are physisorbed, the silica surface showed a different sorption affinity toward the chemicals as a consequence of their electronic structure. Adsorption occurs through multiple hydrogen bonds between the polar functional groups of studied chemicals and surface silanol groups. NTO was found to be the most strongly adsorbed. 6.1.1.4

Oil Components

Exploitation of crude oil and oil sand resources represents another high potential risk for the environmental pollution by various oil components. Structurally complex substances such as resins, asphaltenes, and bitumens or their simpler fragment models have already been investigated in several works. The review by Greenfield (2011) discusses the structure and properties of resins, asphaltenes, and bitumens achieved from molecular dynamics (MD), quantum mechanics (QM), coarse graining, and thermodynamic model approaches. In the paper by Murgich et al. (1998), FF-MD calculations were performed on the model asphaltene and resins adsorbed on neutral surfaces of kaolinite in vacuum. Van der Waals interaction provides the largest contribution (60–70%) to the binding mechanism. Smaller contributions come from Coulombic interactions (20–30%) and hydrogen bonds (10% or less). The variation in the polarity of the resin introduced only minor changes in the energy of interaction with the surfaces. It was also determined that the resin has stronger interaction with the surfaces of kaolinite than asphaltene, indicating that these surfaces may perturb the aggregates formed by the organic molecules that are found in crude oil. Molecular-scale sorption, diffusion, and distribution of asphaltene, resin, aromatic, and saturate fractions of heavy crude oil on quartz surface were studied using FF-MD simulation by Wu et al. (2013). Similarly to the work by Murgich et al. (1998), the authors concluded that despite of the variety of the adsorbed substances, the main contribution to the interactions is represented by the van der Waals energy. Another conclusion was that the most likely oil distribution on quartz surface was those aromatics and saturates transported randomly into and out of the complex consisting of asphaltenes surrounded by resins, which was influenced by temperature. The binding energies of benzene, n-hexane, pyridine, 2-propanol, and water interacting with the kaolinite surfaces were calculated at the DFT level using the exchange-hole dipole moment dispersion model (Johnson and Otero-de-la-Roza, 2012). It was documented that the hydrophilic alumina surface has a stronger affinity to all investigated molecules than the hydrophobic siloxane surface. Hydrogen bonding for pyridine, 2-propanol, and water, OH π interactions for benzene, and CH O interactions for n-hexane were found as the dominant noncovalent interactions. A combination of MM and DFT methods was used in the calculation of the charge distribution and the adsorption energy on the (001) surface of hematite (α-Fe2O3) for a set of 12 molecules representing fragments of resins and asphaltenes (Murgich et al., 2001). The results showed that noncovalent bonding is preferential in the surface complexes. Molecules with high aromaticity and low H/C ratio show high adsorption energy that corresponds with the experimental finding for asphaltenic deposits extracted from wells (Murgich et al., 1999).

Organic and Contaminant Geochemistry

183

Decane, methylbenzene, pyridine, and acetic acid were selected as components in crude oils with a different polarity, and their adsorption behavior on water-wet silica surface was investigated by classical MD simulation (Zhong et al., 2013). The simulation results indicated that polar components could penetrate through water film and adsorb on silica surface, while it was difficult for nonpolar components. It was analyzed that the adsorption capability of oil components is related to three factors: interaction between oil components and silica surface, penetration in water film, and competitive adsorption with water molecules. The authors proposed a two-step adsorption process on the base of the molecular simulations. The polar oil components preferentially absorb on mineral surface promoting further adsorption of nonpolar components. 6.1.1.5

Aromatic Contaminants

The physicochemical properties and the binding characteristics of 18 PAH molecules docking to molecular structural models of fulvic acids (FA), HA, and SOM were studied by Saparpakorn et al. (2007) performing QC semiempirical calculations. The FA, HA, and SOM models were taken from Buffle et al. (1977), Stevenson (1982), and Schulten and Schnitzer (1993), respectively. From the docked conformations of PAHs onto FA, HA, and SOM structures, it was found that π–π interactions and H-bonding were significant for the binding of the PAHs. The final docked energies demonstrated that the PAHs bind to SOM more strongly than to both forms of HA and FA individually. MM and QM studies were carried out by Kubicki and Apitz (1999) to investigate interactions of several NOM models with different organic compounds. The goal was to test different computational methodology for estimating the stability of the isolated systems and the sorption mechanism of organic compounds in NOM. Followed work by Kubicki (2000) comprised energy minimization and dynamics simulations on n-hexane soot model interacting with pyrene including validation of different FF approaches in providing representative soot nanoparticle structures and their properties. Ab initio calculations were used to investigate the electronic and energetic behavior accompanying adsorption of aromatic molecules such as four polychlorinated dibenzo-p-dioxin molecules of different polarities onto an insulating hydrophobic (001) surface of pyrophyllite (Austen et al., 2008). The fairly weak interactions were observed to be dominated by local electrostatics rather than global multipoles or hybridization. A small transfer of electron density was observed from the molecule to the surface. Adsorption of benzene and benzene-1,4-diol on two silica surface models with either hydrophobic or hydrophilic properties was studied by means of calculations based on local Gaussian basis functions and the B3LYP-D functional (dispersion corrected) on periodic structural models (Rimola et al., 2010). It was found that the inclusion of dispersion corrections to the functional dramatically affects both the intermolecular geometries and the adsorption energies. The adsorption of the aromatic molecules on the hydrophobic silica surface is dictated by dispersion and weak CH O(Si)O interactions. For hydrophilic surfaces dispersion is still large despite the fact that adsorption energies are almost doubled with respect to the hydrophobic surface due to weak hydrogen bonding through OH π interactions. The adsorption of benzene and several PAHs on the carbonaceous surfaces from the gas phase and water solution was investigated using several different levels of theory including DFT, MP2, and CCSD(T) (Scott et al., 2012). Both periodic and cluster approaches were used in the study. Good agreement was revealed for the theoretical and experimental adsorption energies of benzene and PAHs adsorbed on the modeled carbon surfaces. The work was later extended on nitrogencontaining aromatic compounds (e.g., pyridine) applying M06-2X and BLYP-D2 functionals

184

Molecular Modeling of Geochemical Reactions

(Scott et al., 2014). The most stable systems were in a parallel orientation toward the modeled carbon surface. The calculated adsorption enthalpies, Gibbs free energies at 298.15 K, and partition coefficients were found to be in a good agreement with available experimental data (Table 1 in Scott et al. (2014)).

6.2 Molecular Modeling Methods This section briefly summarizes molecular modeling methods, tools, and techniques relevant for organic and contaminant geochemistry. Details can be found in many excellent textbooks and reviews available in the literature. Moreover, some other chapters of this volume (Chapters 1–3) also describe particular theoretical methods and modeling techniques and their importance for geochemistry in more detailed aspects. Generally, computational chemistry offers a broad scale of modeling methods dealing with models at a molecular scale. The molecular modeling comprises two basic groups of methods outgoing from (i) MM and (ii) QM, respectively. In the MM methods, classical particles represent a model system consisting of interacting atoms and the main task is to find proper, relatively simple analytical expressions able to describe a potential energy hypersurface of the system under study. On the other hand, in QM, systems are represented as a set of mutually interacting nuclei and electrons and the whole information about the interacting system is keyed in its wave function Ψ. Important physical properties of the system can be achieved by a solution of the Schrödinger equation. In several textbooks the reader can find a comprehensive overview of fundamentals of both QM and MM methods (Leach, 2001; Young, 2001; Frenkel and Smit, 2002; Cramer, 2004). 6.2.1

Molecular Mechanics: Brief Summary

Only basic principles of MM and some practical comments are addressed here. For more details we recommend, for example, Burkert and Allinger (1982) and Rappé and Casewit (1997). Chapter 2 also provides information on MM methods and their applications in geochemistry. MM or force field (FF) methods use analytical expressions for calculation of the potential energy, U(R), for the given configuration of atoms, R. The potential energy of the system is expressed usually in a simple analytical form containing empirical parameters (FF parameters) expressed in terms of interatomic potential functions of interacting atoms in the system. In the classical MM the atoms are represented as the smallest dimensionless entities (particles) having no internal structure. Thus, in this type of simulation, electrons are not explicitly present in the system, but point charges can be assigned to these particles. Complete analytical expression of the potential energy function of the many-body system is a difficult task. Such function can be expressed as a series over sum of two-, three-, and high-order terms, which are functions of geometrical variables, for example, distances and angles. In practice, such series are truncated and high-order terms are neglected and only two- and eventually also three- or four-body interactions are expressed in an analytical form. According to the nature of interactions, the total potential energy of a system is frequently partitioned to the following energy components: UTot = UBond + UAngle + UTors + UCoul + UVDW

61

The first three terms in this equation represent bond-stretching, bond angle, and torsional terms, respectively, defining bonding energy of the system. The last two terms express nonbonding energy terms: the Coulombic, UCoul, and the van der Waals energy, UVDW.

Organic and Contaminant Geochemistry

185

There are many ways how to express individual components, and here only a frequently used approach is briefly described. For describing bond-stretching and bond angle terms, a simple approximation in a form of harmonic potential is used in a following way: UBond = Kb r− r0

2

62

UAngle = Ka θ −θ0

2

63

where Kb and Ka are empirical force constants, r is the separation distance between two atoms (bond), θ is a bond angle for three sequentially bonded atoms, and r0 and θ0 are distance and bond angle for the equilibrium state (corresponding to the lowest energy), respectively. The harmonic approximation is suitable for states, in which atoms are near the equilibrium configuration and atoms oscillate around the equilibrium positions with small deviations. In fact, the harmonic potential can describe most of the vibrations well but for larger deviations it fails. In such a case a more complex function is required to be able to describe an anharmonic nature of vibrations or even more also bond dissociation limit. Such function is, for example, Morse potential with three parameters. In practice, force constants in Equations 6.2 and 6.3 can be obtained from the analysis of the vibrational spectra. The torsion potential, UTors, expresses rotational energy of four atoms connected in sequence (dihedral angle). This potential is periodic and requires an expansion of periodic functions such as a Fourier series. Typical expression for UTors used in many FF programs is UTors = V1 2 1 + cos ω + V2 2 1 + cos 2ω + V3 2 1 + cos 3ω

64

were V1–3 are empirical coefficients and ω is a torsional angle. Proper torsional functions are necessary for a correct description of conformations of large flexible molecules (e.g., proteins), and particular terms are related to a chemical nature and hybridization of particular atoms involved in the torsional bonding. From two nonbonding terms in Equation 6.1, Coulombic potential has a long-range character, whereas UVDW term is of a short-range nature. From classical electrostatics, Coulombic interaction energy can be expressed in a multipole expansion. In the most common practice, only the zerothorder expansion is used to express electrostatic interactions by using point charges assigned to the particular atoms in the model. Then, the electrostatic term is expressed as the sum of pairwise interactions of point charges where pair interactions are a function of the 1/rij, with rij being the distance between charges qi and qj assigned to atoms i and j1: UCoul =

qi qj rij

65

The summation runs over all atomic pairs while avoiding duplications. If the atomic charges are of opposite sign, Equation 6.5 gives a negative (attractive) energy, and if the charges are of the same sign, the result of the summation is a positive (repulsive) energy. The summation in Equation 6.5 is critical for periodic systems due to long-range nature of the electrostatic interactions and requires special mathematical methods to obtain its proper convergence (e.g., Ewald summation (Darden et al., 1993)). The second nonbonding term in Equation 6.1 has a physical meaning in interactions of fluctuating electron densities of atoms or molecules, and it is usually decomposed onto two parts. The first part 1

Atomic units are used for physical constants in equations.

186

Molecular Modeling of Geochemical Reactions

is attractive and is referred as the London dispersion term that is proportional 1/r6. The second component describes Pauli repulsion, which rapidly increases with decreasing distance between atoms and is frequently expressed as a 1/r12 function. The combination of the attractive dispersion and repulsion terms is often expressed in the form of the Lennard–Jones function (also known as LJ or 12–6 potential): UVDW =

4εij

σ ij rij

12

–

σ ij rij

6

66

where εij is a depth of the LJ potential between two atoms (i and j) and σ ij constant is related to a distance at the minimum of LJ potential as rij0 = 2 1 6 σ ij . The LJ potential is used in many FF because of its simple expression (there are also alternative expressions for LJ potential). The nonbonding interactions can be expressed also by different types of potentials such as Buckingham potential (exponential function describes repulsion part) or Morse potential. In specific cases, also 9–6 potential can be used, for example, for simulations of interactions at solid surfaces (Heinz et al., 2008). Apart from the electrostatic potential (ESP), the LJ potential rapidly diminishes with the distance and in practical calculations a cutoff distance is applied (typically 5–10 Å) to reduce the summation in Equation 6.6. The most crucial task in the FF methods is to find proper constants. There are, in principle, two basic ways how to obtain FF parameters. One way is by fitting parameters to experimental data (e.g., structural, thermodynamic, or spectroscopic) for a set of molecules. The second way comprises fitting to QM calculations (e.g., potential energy surface (PES)) on molecular models (or molecular clusters, periodic slab or bulk models if parameters for solids and surfaces are requested). In both ways bonding and also non-bonding terms can be determined. A sensitive part of the FF potentials is the Coulombic contribution due to its long-range nature and the fact that atomic charges cannot be assigned arbitrarily. There are several ways for the assignment of atomic charges. However, the most convenient approach is based on accurate QM calculations. Calculations are usually performed on molecules, molecular clusters, or simple periodic solid systems. The most typical method for obtaining charges is their fitting to the quantum chemically calculated ESP, which is derived from the electron densities. There are various approaches such as CHELP (Chirlian and Francl, 1987), CHELPG (Breneman and Wiberg, 1990), RESP (Bayly et al., 1993), or MK (Singh and Kollman, 1984). A required condition for the determination of the FF parameters and atomic charges is a mutual balance between the Coulombic and nonbonding energy terms. Thus, the combination of the atomic charges from one type of FFs with another one has to be carefully considered. The development of the FF parameters requires their backward validation to ensure that molecular simulations will reproduce energies and properties with a required accuracy. This validation can be performed on some set of molecules, which are similar to those used in the parameter fitting and calculated molecular properties can be compared either with experimental data and/or accurate QC calculations. Praxis showed that it is practically impossible to develop “universal” FF parameters for each element in the periodic table that would be able to reproduce satisfactorily properties of that element in different chemical situations. Instead of that FF parameters are developed for several atomic types of the same element. Carbon atom with its different hybridization states is the classical example. Nowadays the molecular simulation methods offer a broad range of various FF programs usually developed for particular sets of molecular systems. Many of the FF parameters are developed for organic and bioorganic molecules and macromolecules (carbohydrates, peptides, proteins, nucleic

Organic and Contaminant Geochemistry

187

acids, polymers) such as AMBER (Pearlman et al., 1995), CHARMM (MacKerell et al., 1998), GROMACS (Hess et al., 2008), MMFF (Halgren, 1996), CVFF (Sun et al., 1994), UFF (Rappé et al., 1992), and OPLS-AA (Jorgensen et al., 1996). A comprehensive overview can be found in table 2.1 in the textbook by Cramer (2004). Special attention was devoted to developing FF models for water because it is an important polar solvent and also has unusual properties. From many existing FF descriptions of water, SPC (Berendsen et al., 1981), SPC/E (Berendsen et al., 1987), and TIP3P/TIP4P (Jorgensen et al., 1983) models can be assigned to the most frequently used in the simulations. Apart from many available FF parameters for organic molecules, a development of FF parameters for inorganic materials was not such intensive and more specialized FFs have been developed for specific types of inorganic materials (e.g., EAM-FF for metals (Daw et al., 1993) or Buckingham-type potentials for spinels (Fang et al., 2000)). Such FFs are usually less transferable and could contain additional specific energy terms. One of the main reasons for such situation is in more complex chemistry of inorganic substances (composition, bonding) comparing to organic species. Moreover, the accuracy of the FF parameters developed for bulk solids can be problematic for modeling of the surfaces and interfacial properties such as hydration energies, surface tensions, and interaction energies. In addition, nonbonding terms in the energy expressions for inorganic and organic compounds, which are necessary for the simulation of organic–inorganic interfaces, are often not compatible. In spite of that, there is an effort to develop FFs able to overcome depicted problems and using a similar platform as in the standard FF for organics to be able to simulate inorganic compounds and their interfaces with a good accuracy. For example, CLAYFF is a FF developed for simulation of hydrated crystalline compounds (e.g., clay minerals) and their interfaces with fluid phases, which comprises harmonic bond-stretching and angle-bending, Coulombic, and LJ terms (Cygan et al., 2004). Recently, a new effort in modeling of interfaces has been presented in a form of INTERFACE-FF (Heinz et al., 2013). This FF operates as an extension of common harmonic FFs using the same functional form and combination rules. The development is focused on a production of thermodynamically consistent FF parameters for simulations of inorganic–organic and inorganic–biomolecular interfaces. Currently, the INTERFACE-FF includes parameters for aluminosilicates, metals, oxides, sulfates, and apatites. Another long-standing limitation of classical FFs is inability to simulate chemical reactions (i.e., bond-breaking and bond-forming processes). This limitation can be overcome by developing new types of FF, generally called reactive FFs (Liang et al., 2013). ReaxFF is a typical representative of this class of FFs originally developed for reactions of hydrocarbons (van Duin et al., 2001).

6.2.2

Quantum Mechanics: Overview

This section only briefly summarizes QM methods. Principles can be found in many excellent books (e.g., Hehre, 1998; Lowe and Peterson, 2005; Shankar, 1994; Szabo and Ostlund, 1989), and also Chapter 1 in this book provides an overview. The heart of the QM is the Schrödinger equation, which describes the space and time dependence of a system consisting of interacting electrons and nuclei. Mathematically, the Schrödinger equation is the second-order partial differential equation, and its nonrelativistic time-independent form is HΨ = EΨ

67

where Ĥ is the differential Hamiltonian operator, E is the total energy, and Ψ is the wave function of the system. The Hamiltonian can be expressed as the sum of the kinetic and potential energy operators:

188

Molecular Modeling of Geochemical Reactions

1 2 ∇ + V R, r 68 2m In Equation 6.8, m is the mass of the ith particle, ∇2 is the Laplacian operator, and V(R, r) is the operator describing the potential energy associated with the Coulombic interactions of all electrons and nuclei in the system. The existence of a wave function is one of the postulates in QM and has to satisfy certain constraints. It is interpreted in terms of a probability to find electrons in a configurational space and is used for obtaining the energy of the system. Exact solution of the Schrödinger equation for manyparticle system is not possible, and for practical applications its solution requires some approximations. For example, the Born–Oppenheimer approximation separates effectively nuclear and electronic motion; thus, electronic wave function only parametrically depends on the positions of nuclei. QM-based methods can be classified to four basic groups: Hartree–Fock (HF), ab initio correlation (post-HF methods), DFT, and semiempirical methods. HF and post-HF methods involve a 3N-dimensional antisymmetric wave function for a system of N electrons. The HF approach is the simplest one as it considers exchange for a single determinant wave function. The HF method is almost universally used as the reference independent particle model. Therefore, quantum chemistry approaches are based on the HF method as a first approximation (Charlotte, 1987). The electronic correlation energy neglected in the HF method is included in the configuration interaction (CI) method (Szalay et al., 2012). This method possesses a large flexibility in computing various types of electronic structures and excitations but is computationally very expensive. Møller–Plesset (MP) perturbation theory (1934) is another post-HF ab initio method. It corrects the HF method by including electron correlation effects using the Rayleigh–Schrödinger perturbation theory (RS-PT), most commonly to second order (MP2). MP2 is computationally efficient and can be used for many practical applications in geochemistry. Coupled cluster (CC) theory (Čížek, 1991; Stanton and Bartlett, 1993; Paldus, 2005) formulates the electronic Schrodinger equation as a nonlinear equation, allowing the calculation of size-consistent high-precision approximations of the ground state solution. The form of CC, which includes singles and doubles and noniterative triples (CCSD(T)), is considered to be the most accurate method routinely available. Applications are available (as for MP2) for cases where the HF wave function is a good starting point. A fundamental issue to be considered in the application of ab initio-based methods is the size of the system, the type of property desired to be analyzed, and the computational time required. The QC methods are computationally intensive scaling with O(N4) and more especially the correlated post-HF methods. Therefore, in this section we will emphasize the DFT-based methods as they are by far the most frequently used methods to describe large molecular systems of the size of several hundred atoms. The popularity of DFT has roots in its economy and efficiency comparing to the standard HF and post-HF methods as scaling is O N 3 . H=−

6.2.2.1

Density Functional Theory: Basic Principles

DFT defines the electronic states of atoms, molecules, and solids, in terms of the 3D electronic density of the system. It means a great simplification over the wave function-based methods (HF, CI, MP, and CC). The method is based on Hohenberg–Kohn theorems (1964), and subsequently Kohn and Sham have shown a practical way to apply it (1965). The Hohenberg–Kohn Theorems. The first Hohenberg–Kohn theorem establishes that the electron density exclusively determines the Hamiltonian operator and therefore all the properties of the system. This first theorem states that the external potential Vext r is (to within a constant) a unique functional of ρ r . Since in turn Vext r fixes Ĥ it is clear that the full many-particle ground state is a

Organic and Contaminant Geochemistry

189

unique functional of ρ r . Thus, ρ r determines the number of electrons N and Vext r and consequentially all the properties of the system such as the kinetic energy (T[ρ]), the potential energy V(T[ρ]), and the total energy (E[ρ]). The total energy can be written as E ρ = T ρ + ENe ρ + Eee ρ = ρ r VNe r d r + FHK ρ

69

FHK ρ = T ρ + Eee

6 10

In Equation 6.10, FHK[ρ] denotes a universal functional that contains the kinetic energy T[ρ] and the electron–electron interaction, Eee[ρ]. This functional is the holy grail of DFT. The subscript Ne in Equation 6.9 is regarding to the electron-nuclear interaction. It is important to emphasize that this functional is completely independent of the system; it applies well to an atom as well to large molecules. If the functional FHK[ρ] was known, it would be possible to solve the Schrodinger equation exactly. The electronic interaction functional can be expressed as Eee ρ =

1 2

ρ r1 ρ r2 dr1 dr2 + Encl ρ = J ρ + Encl ρ r12

6 11

In Equation 6.11, Encl[ρ] is the nonclassical contribution to the Eee[ρ] term encompassing all the effects of self-interaction correction, exchange, and Coulomb correlation. J[ρ] is the classical Coulomb part. To find expressions describing the T [ρ] and Encl[ρ] functionals is the key challenge in DFT. The second Hohenberg–Kohn theorem establishes a variational principle, which can be expressed as E0 ≤ E ρ = T ρ + ENe ρ + Eee ρ

6 12

These two theorems lead to the major DFT statement: δ E ρ −μ

ρ r dr −N

=0

6 13

Equation 6.13 means that the ground state energy and density correspond to the minimum of a functional E[ρ] with the constraint that the density refers to the correct number of N electrons. The Lagrange multiplier related to this constraint is the electronic chemical potential μ. Thus, we arrive at the interesting conclusion that there is a universal functional E[ρ] which, if it is known, could be used in Equation 6.13 to compute the exact ground state density and energy. The Kohn–Sham Approach. The equations of Kohn and Sham, published in 1965, turn DFT into a practical tool (Kohn and Sham, 1965). The Kohn–Sham approach to DFT provides an exact description of the interacting many-particle systems in terms of an effective noninteracting particle system. The effective potential of the Kohn–Sham system (noninteracting particles) can be completely described by the electron density of the interacting system. The total energy functional in the Kohn–Sham approach contains the kinetic energy, electron–electron interaction, and interaction with the external potential terms: E ρ = T ρ + Vext ρ + Vee ρ

6 14

190

Molecular Modeling of Geochemical Reactions

The external potential is defined as Vext ρ = V ext ρ r dr

6 15

T[ρ] and Vee[ρ] are unknown functionals. In the fictitious system of noninteracting particles idealized by Kohn and Sham, the kinetic energy functional is described by means of a single determinant wave functions in N orbitals, ϕi: TS ρ = −

1 2

N

ϕi ∇2 ϕi

6 16

i

Obviously TS, the kinetic energy of the noninteracting particles, is not equal to the true kinetic energy of the real system. Kohn and Sham accounted for it by writing F[ρ] as F ρ = TS ρ + J ρ + EXC ρ

6 17

In Equation 6.17, EXC is denominated as exchange–correlation energy and contains everything that is unknown. It can be defined based on Equation 6.17 as shown below. EXC ρ = T ρ −TS ρ + Eee ρ −J ρ

6 18

Using the variational principle in order to minimize the energy expression under the constraint ϕi ϕj = δij , one arrives at the final Kohn–Sham equations: 1 − ∇2 2

ρ r2 + VXC r1 − r12 VS r1 =

M A

ZA r1A

ϕi =

1 − ∇2 + VS r1 2

ρ r2 d r2 + VXC r1 − r12

M A

ZA r1A

ϕi =

i ϕi

6 19

6 20

This set of nonlinear Kohn–Sham equations describes the performance of noninteracting particles in an effective local potential. For the exact local potential, the “orbitals” yield the exact ground state density through Equation 6.19 and the exact ground state energy through Equation 6.20. 6.2.2.2

DFT Functionals: General Overview

The major problem in DFT theory is the fact that the exact functionals for exchange and correlation are not known except for the homogeneous electron gas. However, the success of DFT relies on the existence of approximations permitting to calculate electronic structure and physical quantities with a good accuracy. In the starting era of DFT, the local density approximation (LDA) became a frequently used approximation, especially in solid-state physics (Parr and Yang, 1989). In LDA, the exchange–correlation energy functional terms depend simply on the electron density at the coordinate, where the functional is calculated. The LDA was later extended to the local spin-density approximation (LSDA) by the inclusion of the electron-spin term (Vosko et al., 1980). The improvement of the LDA functionals led to the generalized gradient approximation (GGA) method (Langreth and Mehl, 1983; Becke, 1988; Perdew et al., 1992) by including gradients of the electron density to the functional forms. Within the frame of the GGA theory, numerous exchange– correlation functionals have been developed for chemical applications (Zhao and Truhlar, 2008a). Among them PBE (Perdew et al., 1996), PW91 (Perdew and Wang, 1992), and BLYP

Organic and Contaminant Geochemistry

191

(Becke, 1988; Lee et al., 1988; Miehlich et al., 1989) can be considered as frequently used functionals in many types of DFT calculations on molecules or solid-state materials. The B3LYP functional has achieved a high popularity, especially in the chemical community (Becke, 1993). This functional belongs to a hybrid type, in which HF exact exchange is mixed with DFT exchange–correlation using adjustable parameters. These parameters are fitted on a training set of molecules. The popularity of B3LYP lies in its relatively good accuracy in predicting of many properties including the thermochemical parameters for a wide range of chemical species. From the semiempirical methods it is worthy to briefly introduce density functional tight binding (DFTB) approach as it is based on the DFT framework with the inclusion of empirical parameters (Seifert et al., 1996; Elstner et al., 1998). It introduces several approximations that reduce the calculations of complicated electronic integrals. The Hamiltonian and overlap matrices contain oneand two-center contributions only. They are calculated and tabulated in advance as functions of the distance between atomic pairs. Therefore, this method (and its self-consistent charge extension, SCC-DFTB) is orders of magnitudes faster than the standard DFT methods. The improvements known from the standard DFT methods such as dispersion corrections are also adapted in the DFTB method. Thus, this approach offers an acceptable quality in comparison to standard DFT methods and can be used for systems of thousands of atoms. The foundations of the DFTB method are reviewed by Seifert and Joswig (2012). However, the DFTB, as each parametric method, has certain limitations, for example, transferability of parameters for the same element in different chemical environments. 6.2.2.3

Dispersion Corrections to DFT

DFT represents one of the most successful theoretical methods for investigation of structure and properties of molecules and solids in various fields of material research because it holds an excellent balance of computational costs and reached accuracy. DFT methods are even faster and efficient for periodic solid systems if they combine atomic basis sets expressed in plane waves with atomic pseudopotentials. However, it has also been realized early that DFT is not accurate enough in a prediction of properties such as charge transfer, band gap, and excited states and in some specific physical situations. One big problem of the standard DFT is that it is incorrect in the description of weak nonbonding interactions such as dispersion, π–π stacking, or hydrogen bonding. The fundamental reason for this failure is the inability of the method to account for a nonlocal electron correlation effect. Therefore, the successful application of the standard DFT functionals on systems with dominating weak interactions (e.g., molecular crystals) is rather limited. A lot of effort has been devoted in a systematic and active development of DFT functionals to overcome this problem. The development of new methods is conducted at several levels of theory—from purely empirical corrections to the standard DFT functionals to rigorous dispersion functionals derived from first principles. The van der Waals density functional (vdW-DF) method is going beyond the local and semilocal DFT approximation by employing a nonlocal functional to approximate the correlation. This first-principles approach deals with the van der Waals forces by including nonlocal correlations from the electron response to the electrodynamic field (Dion et al., 2004; Thonhauser et al., 2007; Lee et al., 2010). Another group of the improved DFT methods includes flexible hybrid meta-GGA functionals generally called Minnesota functionals. They are, for example, able to cover dispersion interactions in various nonbonding complexes for shorter distances, and the functionals such as M05-2X and M06-2X (Zhao and Truhlar, 2008b; Zhao et al., 2005) have increasing popularity in using by the QC community. The class of methods based on a combination of conventional functionals with empirical dispersion corrections became quickly popular because of its effectivity and simple implementation.

192

Molecular Modeling of Geochemical Reactions

The dispersion corrections include r6 term (eventually also higher terms) and respective methods are generally known as DFT-D methods (Grimme, 2004, 2006, 2011; Grimme et al., 2010). Another method, based on empirical corrections to the standard DFT functionals, is known as atom-centered dispersion-correcting pseudopotential (DCP) method (von Lilienfeld et al., 2004; DiLabio, 2008). An alternative method to the DFT-D approach grounded on system-dependent dispersion coefficients calculated from the first principles was originally suggested by Tkatchenko and Scheffler (vdW-TS method) (2009). The vdW-TS method was later improved by the addition of the selfconsistently screened approach and Hirshfeld partitioning for getting the atomic polarizabilities (Ruiz et al., 2012) and atomic charges (Bučko et al., 2014), respectively. The problem of dispersion corrections in DFT is extensively studied, and there are numerous benchmark calculations and reviews available in the literature (e.g., Johnson et al., 2009; Bučko et al., 2010; Marom et al., 2011; Goerigk, 2014). In 2014, the DFT theory celebrates 50 years and its history, development, and perspectives can be found in the paper by Becke (2014). 6.2.3

Molecular Modeling Techniques: Summary

This section briefly summarizes three basic molecular modeling techniques: energy minimization, MD, and MC. There are many textbooks providing details on the molecular modeling techniques (some of them are cited in this chapter, e.g., Leach, 2001; Young, 2001; Frenkel and Smit, 2002; Cramer, 2004). Chapters 1 and 2 of this book also give much more details on the available sophisticated algorithms. 6.2.3.1

Energy Minimization

PES is a complex function of atomic position; for a system with N atoms, it is a function of 3N–6 internal or 3N Cartesian coordinates. The aim of the energy minimization is to obtain the most stable configuration (the state with the lowest energy) for the system under study. In other words, the minimum of the PES is found by varying geometrical parameters. This method is also referred as a geometry optimization or a structural relaxation. It is expected that in the energy minimum the system has the lowest potential energy and the forces on all atoms are zero and the system is in equilibrium state. Note that “thermodynamic” condition for the optimization is T = 0 K. The global minimum is required for the calculation of molecular properties, for example, vibrational states. There are a variety of optimization procedures. Algorithms employing energy derivatives belong to the most frequently used methods (e.g., steepest descent, complex conjugate gradient, or Newton–Raphson). Such minimization procedures involve the calculation of the potential energy of the initial configuration and the potential energy derivatives with respect to atomic coordinates. New coordinates of the system are adjusted according to computed energy derivatives and are expected that the new configuration will have lower energy than the previous one. This procedure is repeated until defined tolerances for energies, gradients, forces, and deviations of atomic coordinates between successive steps are achieved. For complex energy hypersurfaces it is possible to use methods employing genetic algorithm or (MD) simulated annealing. During the optimization procedures some restrictions or constrains can be imposed. In this way, for example, symmetry can be preserved or positions of selected atoms can be fixed. For systems with translational periodicity, lattice parameters of the computational cell can be optimized as well. 6.2.3.2

Molecular Dynamics

MD is a deterministic method, which describes the movement of the system in the phase space defined by the coordinates and momenta of the particles in the system. The purpose of the MD

Organic and Contaminant Geochemistry

193

is finding a trajectory of the system in the phase space by solving Newton’s equations of motion. For many-body system (ensemble of particles), analytical solution is not possible and a numerical solution has to be used indeed. The core of MD is in the calculation of the forces acting on the particles in the system. The force for each particle can be determined as a derivative of the potential energy U with respect to the change in the particle’s position. In the classical MD, the potential energy describes interactions among the particles in the system and is expressed in a particular type of the FF. Knowing positions, forces, and masses of the particles, it is possible to find new positions by performing a numerical integration of the equations of motion over the selected time interval. Usually, the MD has the following steps. At the beginning, for a given initial configuration of the system, initial velocities are assigned to the particles (e.g., using Boltzmann distribution of velocities). Then, forces are derived from the potential energy followed by the numerical integration of the equations of motion over a very small time step, Δt. From this integration new positions of the particles are obtained and new potential energy is calculated. The procedure is repeated over a desired time length. The time increment in the numerical solution has to be satisfactorily small. It is recommended to use a time step that is satisfactorily smaller than the period of the highest vibrational mode in the system. Thus, typical time steps in the MD simulations are few femtoseconds. The repeated step-by-step procedure results in a series of structural changes (configurations) over time. Several integration algorithms are used in MD, for example, Verlet, velocity Verlet, or leapfrog method (Verlet, 1967; Van Gunsteren and Berendsen, 1977; Ferrario and Ryckaert, 1985). By calculating a partition function of the simulated system, it is possible to characterize its macroscopic behavior from the thermodynamic point of view. Various characteristics of the system can be calculated applying statistical ensemble averages of the various instantaneous values obtained during the MD run, for example, structural or energetic parameters. Moreover, time correlation functions (e.g., of velocities) can be used for the calculation of transport coefficients or power spectra (corresponding to vibrational spectra). MD simulations are also suitable for the prediction of transport properties such as diffusion and viscosity coefficients, but this type of the calculations usually requires a very long time. Typical ensembles used in MD calculations are microcanonical (NVE—constant number of particles, volume, and energy), canonical (NVT—constant number of particles, volume, and temperature), or isothermal–isobaric (NpT—constant number of particles, pressure, and temperature). In NVT simulations, the system is embedded into a thermostat to keep a desired temperature using, for example, scaling of velocities or an external thermostat (e.g., Berendsen et al. (1984)). The prediction of the thermodynamic quantities such as free energies and their changes is one of the most wished outputs from MD. Combination of the statistical averages with thermodynamic integration approach allows calculation of thermodynamic quantities such as free energies and entropy values. According to the ergodic principle, ensemble and time averages should be equal for the infinite time. However, in practice, MD evolves a finite-sized molecular configuration in limited time and possible errors have to be estimated carefully. Even though realistic classical MD simulations can approach milliseconds (Lindorff-Larsen et al., 2011), there are events that would not be sampled satisfactorily at such time scale. Such events are addressed as rare and are linked with barriers crossing on the energy hypersurface (e.g., phase transitions, chemical reactions, or protein folding). The simulations of these processes are challenging, and the corresponding theories and methods are under an intense development. For example, Hartmann et al. (2014) and Dellago and Hummer (2014) summarize MD methods and techniques for the calculations of the free energies and characterization of rare events. Owing to a rapid progress in developing the DFT theory and increasing power of the computer sources, in 1985 appeared the pioneering work linking MD method with DFT theory (Car and Parrinello, 1985). This method, often referred as first-principles MD (FPMD) or ab initio MD (AIMD),

194

Molecular Modeling of Geochemical Reactions

is an accurate atomistic simulation and combines a QM description of electrons with a classical description of atomic nuclei. The Newton equations of motion are solved for all nuclei as in the classical MD. However, the forces acting on the nuclei are derived from a calculation of the electronic energy of the system at each discrete time step of the trajectory. Time and size scales in AIMD are smaller than in the classical MD because the electronic structure calculation is the most timeconsuming part. Classical MD scales with O(NlnN) or O(N2) (depending on a type of FF) and the large-scale FF-MD simulations can be performed on models of approximately 1000 nm in size and milliseconds, while the AIMD scales with O(N3) having the time length approaching to microseconds and the systems consisting of thousands atoms. Indeed, the AIMD has at least two main advantages. Firstly, this method is not parameter dependent and can be used for any system, and the accuracy is determined by the accuracy of the DFT approach used. Secondly, the AIMD can be taken for the systems, where electronic properties and atomic dynamics are significantly dependent. It means that the AIMD can be used also for the simulation of real chemical reactions where chemical bonds are changed. Such simulations cannot be performed in the framework of the classical MD except when using a specific reactive FF (see Chapter 2).

6.2.3.3

Monte Carlo Method

MC is a simulation method widely used in molecular modeling of chemical systems. Unlike MD methods, MC methods are stochastic techniques based on a statistical probability. They are frequently used not only in chemistry but, for example, also in economics. The use of MC methods to model physical problems allows us to examine complex systems that cannot easily be solved by integral calculus or other numerical methods. With MC methods, a large system can be sampled in a number of random configurations, and those data can be used to describe the system as a whole. The most popular MC scheme is the Metropolis MC method (1953). In this method, the PES of the system, which is defined by a FF approach, is scanned over randomly generated configurations. The initial potential energy is calculated for the first configuration, and then, particles in the system (e.g., atoms) are randomly displaced to a new configuration for which the potential energy is again calculated. If this energy is smaller than the previous one, the system is in the more stable configuration and this state is accepted to the statistical ensemble. However, if the new energy is higher than the previous one, the difference between both energies is compared to a random number via a Boltzmann distribution. If the value is smaller than the random number, the new configuration is accepted and used for the generation of other new one. In the opposite, if the value is bigger than the random number, the configuration is rejected and the original configuration is used for the generation of the next configuration. In this way, the PES can be mapped for millions of configurations. Applying statistical ensemble averages various properties such as thermodynamic quantities can be calculated. MC simulations are preferable if there are large intramolecular energy barriers, which can result into molecules being trapped in a few low-energy conformations in MD simulation. Randomly generated moves in the MC simulation can more readily lead to a barrier crossing. In the opposite case, the simulation of liquids by MD can be more effective than by MC. There is a large probability of selecting random moves in the MC simulation of liquids, in which two or more molecules can overlap leading to a large number of rejected configurations. This results in dropping of the efficiency of the MC sampling. However, there are improvements of the MC methods leading to a better performance, for example, configurational bias MC (Siepmann and Frenkel, 1992) or hybrid MC (Duane et al., 1987). Grand canonical MC version (constant chemical potential (μ), V, and T) is very suitable for the study of the interfacial phenomena, for example, for modeling of adsorption isotherm (Snurr et al., 1993; Liu and Monson, 2005).

Organic and Contaminant Geochemistry

6.2.4

195

Models: Clusters, Periodic Systems, and Environmental Effects

Molecular modeling of geochemical processes at surfaces and interfacial phenomena requires a special attention in building proper structural model covering both solid surface and interacting phase (molecule, liquid phase). Clusters are molecular models prepared by cutting of a fragment, which should reflect typical features of the interacting site plus representative large part of the solid phase and surrounding solvent molecules. If there are dangling bonds, they should be saturated (e.g., by addition of hydrogen atoms or pseudo atoms). Cluster method has several drawbacks such as missing long-range interactions, potentially imbalanced stoichiometry, or unwanted impact of the terminal atoms. Additionally, computed physical properties usually depend on the cluster size (Sauer, 1989; Pacchioni et al., 1992). Moreover, the computed properties obviously do not converge continuously with the increasing size of the clusters but often oscillate, which brings difficulties in their extrapolation. For example, Sukrat et al. (2012) showed that the cluster size and cavity structure are very important for predicting energetic barriers for the proton exchange reactions of C2–C4 alkanes in ZSM-5 zeolite. A decrement up to 20 kcal/mol was observed when employing the periodic model instead of using the small cluster model. The problem of the missing environment in the cluster model calculations is overcome in embedding methods. Continuum solvation models can also be assigned to this group. They are developed for the calculation of solvent effects and are based on the self-consistent reaction field (SCRF) approach (Rinaldi and Rivail, 1973). The polarizable continuum model (PCM) and its variants (Miertuš et al., 1981; Cances et al., 1997; Tomasi et al., 2005) and/or the conductor-like screening model (COSMO) (Klamt and Schürmann, 1993; Klamt et al., 1998) belong to the most widely used SCRF methods in quantum chemistry. Generally, these methods use a polarizable continuum representing solvent, in which a cavity for a solute molecule is created. The molecular free energy of solvation is computed as the sum of electrostatic, dispersion–repulsion, charge transfer, and cavitation energy terms. For more details, see the review on the QM continuum solvation models (Tomasi et al., 2005). The simplest embedding scheme is a model cluster inserted in a background of classical point charges, which found numerous applications, for example, in the description of surfaces of ionic solids. More sophisticated schemes include embedding of the cluster treated at the QM level in the backbone of atoms described by MM potential. These hybrid schemes are denoted as QM/ MM methods. They link the main advantages of QM (accuracy) and MM (speed) methods. Moreover, QM/MM schemes can be used in the study of chemical reactions in “realistic” environments that is not possible through the classical MM methods. Further, the QM/MM method is highly appropriate for large macromolecules or associates/aggregates of various smaller organic species (e.g., NOM) including their environments. There are numerous variants of the QM/MM schemes that can be used in different applications in various fields. Details on the methodology, development, advantages, problems, and applications can be found in several reviews (e.g., Gao and Thompson, 1998; Friesner and Guallar, 2005; Vreven and Morokuma, 2006; Senn and Thiel, 2009; Canuto and Sabin, 2010). Bulk properties of the solid-state structures with the translational periodicity are simulated by using a computational cell corresponding to a crystallographic unit cell or its multiplications (supercell approach). The later approach is also carried out in the simulations of liquids and aperiodic solids. A different level of complexity for calculations with the periodic boundary conditions is the modeling of the solid-state surfaces. A periodic supercell slab model is another alternative to the cluster and embedding schemes. The slab is cut from the bulk structure along specific directions usually dictated by the desired surface termination that is determined by the crystallographic planes

196

Molecular Modeling of Geochemical Reactions

(in the case of periodic solids). The slab is periodic in two dimensions, whereas a vacuum is imposed in the third dimension. This vacuum should be adequately thick to avoid artificial interactions between slabs in the neighboring cells if the codes with only possibility of 3D computational cells are used. There are many additional factors that must be taken into account in the slab construction. For example, the slab has two surfaces and both should be the same if possible. This should minimize an artificial polarization effect in the perpendicular direction with respect to the surface. The thickness of the slab is another factor. The electronic structure in the bulk and at the surface is significantly different due to the broken bonds, not fully coordinated atoms at the surface, and the relaxed uppermost atomic planes. Thus, the surface properties planned to be studied have to be checked on a convergence with respect to the slab thickness. This convergence is usually different for systems with localized and delocalized bonding states. In the case of solids composed from two or more types of atoms arises a question about what atomic plane is the topmost. As the surfaces are formed from the unsaturated atoms with respect to the bulk having dangling bonds, these atoms are chemically reactive. Unlike vacuum termination in “realistic” environment (with air gases or liquid phase) they react and a final chemically stable termination can be formed (e.g., from hydroxyl groups). These forms of the surfaces are usually a subject of investigations of the interactions of the molecular species and/or solid–liquid interfaces, respectively. In such investigations it is important to use a slab also having two periodic dimensions large enough to minimize lateral interactions between the interacting molecule and its replicas in the neighboring cells and also to avoid an artificial periodicity for amorphous liquid phase imposed on the surface. Technically, solid–liquid interfaces are modeled in two ways: a water slab of defined thickness is deposited on the surface and then vacuum is inserted above to separate the slab in the neighboring cells or the water slab is confined between two surfaces of the solid slab. In the latter case the dimension of the computational cell in the perpendicular direction with respect to the surface has to be tuned with respect to some physical parameter, for example, pressure. There are many works reporting on molecular simulations of properties and interactions of the solid-state surfaces and interfaces and also Chapter 8 in this book offers examples.

6.3 Applications This section summarizes primarily the results of our research group achieved approximately in the last decade. The examples represent mainly molecular modeling studies of interactions and mechanisms of binding several polar and nonpolar organic contaminants with typical minerals and natural organic matter (NOM). It has to be emphasized again that a characterization of structure, composition, and properties of solid phases is very complex subject for both experiment and molecular modeling. Mineral surface termination, surface morphology, hydration, hydroxylation, redox processes, charging, exchange processes, distribution, and characterization of surface sites and defects represent a complex set of problems that are intensively studied as it is evidenced by numerous papers in this field (see, e.g., other chapters in this book especially Chapter 8). An even more complex situation we face is in the case of NOM. NOM, as the largest pool of the organic carbon on the Earth, plays an important role in many biogeochemical processes including binding, stabilization, and biodegradation of the organic contaminants (Senesi et al., 2009). NOM has a heterogeneous composition consisting of various groups of organic molecules possessing an extremely complex 3D macromolecular structure that is not completely known. Therefore, only approximate models are suggested that are built on knowledge from an elemental analysis and types and distribution of various functional groups achieved from experiments. Several models have been discussed how the NOM components are held together. The most frequently used models are the

Organic and Contaminant Geochemistry

197

polymer and supramolecular models. In the polymer model, NOM is considered as a macromolecule containing amorphous and crystalline domains (Ghosh and Schnitzer, 1980; Leboeuf and Weber, 1997; Xing and Pignatello, 1997). The supramolecular model (Cozzolino et al., 2001; Piccolo, 2002) defines NOM as a physicochemical body dominated by nonbonding interactions such as hydrogen bonds, van der Waals, and weak hydrophobic interactions, which keep individual molecules of primary structure together. It is important to note that a micellar or membrane-like model taking into account the amphiphilic character of HSs was also suggested (Wershaw, 1993). Even though there is no definitive consensus on the structure of NOM, the functional groups of the organic molecules present in their composition have been well described. These include carboxyl, hydroxyl, phenolic, alcohol, carbonyl, and methoxy groups. Among these, carboxyl and phenolic groups are regarded the key functional groups responsible for the adsorption of NOM by soil minerals in forming stable organomineral aggregates (Sutton and Sposito, 2006; von Lützow et al., 2006). 6.3.1

Modeling of Surface Complexes of Polar Phenoxyacetic Acid-Based Herbicides with Iron Oxyhydroxides and Clay Minerals

The contamination of soils and water resources increases progressively with growing production and application of chemicals for agricultural activities that represents a serious problem throughout the world (Kurtz, 1990; Mathys, 1994; Larson et al., 1997). Nearly two million tons of pesticides are applied to agricultural land worldwide each year (Fenner et al., 2013). From a rich group of pesticides, derivatives of phenoxyacetic acid represent a broad spectrum of herbicides extensively used in agriculture. Their behavior in soil environments (solubility, transport, adsorption–desorption, chemical resistance, and biodegradation) is governed by their chemical structure (Scheme 6.1). Molecular structures of these modern herbicides contain polar carboxylic group. This group is responsible for the relatively high chemical activity of these herbicides in interactions with soil components and also contributes to a reduction of their persistence in the environment allowing a better control (Kästner et al., 2014). For example, MCPA is currently used to reduce the spread of annual and perennial broad-leaved weeds on land under grasses and in vineyards. The agronomic MCPA dosage is typically as high as 1–2.5 kg/ha, and it has been classified by the US Environmental Protection Agency as a potential groundwater contaminant (Walker and Lawrence, 1992). In MCPA, one methyl group of the phenoxy moiety is replaced by a chlorine atom, whereas in 2,4-dichlorophenoxyacetic acid (2,4-D) both methyl groups are replaced by Cl atoms (Scheme 6.1). This substitution increases their polarity and, consequently, acidity and solubility (Haberhauer et al., 2000, 2001). Typical half-life of MCPA in soils is about 24 days (Thorstensen and Lode, 2001), and also a combination of processes such as dilution, sorption, and degradation contributes to disappearance of herbicides from the environment (Hiller et al., 2012). Many experiments studied sorption/desorption processes of phenoxyacetic acid herbicides in soils and/or soil components, respectively (Bolan and Baskaran, 1996; Pignatello and Xing, 1996; DePaolis and Kukkonen, 1997; Sannino et al., 1997; Susarla et al., 1997; Benoit et al., 1998; Celis and Koskinen, 1999; Celis et al., 1999; Cox et al., 2000; Haberhauer et al., 2000, 2001; Clausen et al., 2001; O

O

O

O OH

MCPA Cl

Scheme 6.1

OH 2,4-D Cl

Cl

Skeletal formula of MCPA and 2,4-D molecules.

198

Molecular Modeling of Geochemical Reactions

Vasudevan et al., 2002; Spadotto and Hornsby, 2003). These experiments describe sorption from aqueous solutions to various solid matrices and provide mainly the soil–water distribution coefficient Kd and very rarely the adsorption energy or enthalpy calculated from fits to experimental data. It was observed that the adsorbed amount depends strongly on pH and a presence of electrolyte (e.g., CaCl2) in solution (Cox et al., 2000; Vasudevan et al., 2002; Spadotto and Hornsby, 2003). The experiments demonstrated lower sorption capacity of the clay minerals at pH > 4 (what is above a pKa value of 2,4D (2.8) and MCPA (3.1)). A review on the distribution coefficients for sorbed phenoxyacetic acidbased herbicides in soils and soil components showed a mean logKd of 0.16 ± 0.55 l/kg (2,4-D) and −0.10 ± 0.56 l/kg (MCPA), respectively (Werner et al., 2013). 6.3.1.1

Structure and Properties of Kaolinite, Montmorillonite, and Goethite

Clay minerals such as kaolinite and smectites and natural iron oxyhydroxides are common soil minerals. They are potential geosorbents for polar agrochemicals in soil having significant impact on the soil sorption capacity. Phenoxyacetic acids exist in the anionic form at pH >3–4 and may be physisorbed through deprotonated anionic carboxyl groups to positively charged (i.e., protonated) surface sites on minerals below their point of zero charge (PZC) (pHPZC). Therefore, the low pHPZC values of soil clays such as kaolinite or montmorillonite (Table 6.1) limit their sorption capacity with respect to anionic herbicides at ambient soil pH values (6–8), and direct sorption from the solution to neutral or negatively charged surfaces is not strongly preferred. Sorption of 2,4-D on clays was enhanced in the presence of ionic solution and was supposed a formation of a cation-bridged complex between herbicide anion and a negatively charged surface site of clay (Clausen et al., 2001). A ligand exchange mechanism was assumed in the sorption of 2,4-D to the synthetic chlorite-like complexes (Al(OH)x coated montmorillonite surfaces) in the presence of acetate or phosphate buffers at pH 5–6 (Sannino et al., 1997). The neutral form of phenoxyacetic acids dominates at low pH

Table 6.1

Typical physical and chemical characteristics of kaolinite, montmorillonite, and goethite.a

Geosorbent

Kaolinite

Montmorillonite

Goethite

Chemical composition

Al2Si2O5(OH)4

Nax + y Si8−x Alx Al4−y Mgy

α-FeOOH

pH PZC SSAb (m2/g) CECc (cmolc/kg) Dominant/ typical surface Permanent surface charge (|e|) Surface termination

~4 10–30 3–15 001

O20 OH 4 nH2 O 4–6 300–600 80–150 001

7.5–9.5 50–200 Up to 100 110

~0

0.1–0.5

~0

Surface OH (octahedral) basal surface oxygen atoms (tetrahedral)

Basal surface oxygen atoms + compensating cations (e.g., Na + )

Plates, discs

Irregular flakes

–OH μ-OH μ3-OH Idles, rods

1.25