THE CHEMICAL PHYSICS OF SOLID SURFACES
THE CHEMICAL PHYSICS OF SOLID SURFACES
Volume 1 CLEAN SOLID SURFACES Volume 2 ADSORPTION AT SOLID SURFACES Volume 3 CHEMISORPTION SYSTEMS Volume 4 FUNDAMENTAL STUDIES OF HETEROGENEOUS CATALYSIS Volume 5 SURFACE PROPERTIES OF ELECTRONIC MATERIALS Volume 6 COADSORPTION, PROMOTERS AND POISONS Volume 7 PHASE TRANSITIONS AND ADSORBATE RESTRUCTURING AT METAL SURFACES Volume 8 GROWTH AND PROPERTIES OF ULTRATHIN EPITAXIAL LAYERS Volume 9 OXIDE SURFACES
THE CHEMICAL PHYSICS OF SOLID SURFACES
EDITED BY
D.P.WOODRUFF B.Sc. (Bristol), Ph.D., D.Sc. (Warwick) Professor of Physics, University of Warwick
VOLUME 9
OXIDE SURFACES
2001 ELSEVIER AMSTERDAM - LONDON - NEW YORK - OXFORD - PARIS - SHANNON - TOKYO
ELSEVIER SCIENCE B.V. Sara Burgerhartstraat 25 RO. Box 211, 1000 AE Amsterdam, The Netherlands © 2001 Elsevier Science B.V. All rights reserved This work is protected under copyright by Elsevier Science, and the following terms and conditions apply to its use: Photocopying Single photocopies of single chapters may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic copying, copying for advertising or promotional purposes, resale, and all forms of document delivery. Special rates are available for educational institutions that wish to make photocopies for non-profit educational classroom use. Permissions may be sought directly from Elsevier Science Global Rights Department, PO Box 800, Oxford 0X5 1DX, UK; phone: (+44) 1865 843830, fax: (+44) 1865 853333, e-mail:
[email protected]. You may also contact Global Rights directly through Elsevier's home page (http://www.elsevier.nl), by selecting 'Obtaining Permissions. In the USA, users may clear permissions and make payments through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; phone: (+1) (978) 7508400, fax: (+1) (978) 7504744, and in the UK through the Copyright Licensing Agency Rapid Clearance Service (CLARCS), 90 Tottenham Court Road, London W1P OLR UK; phone: (+44) 207 631 5555; fax: (+44) 207 631 5500. Other countries may have a local reprographic rights agency for payments. Derivative Works Tables of contents may be reproduced for internal circulation, but permission of Elsevier Science is required for external resale or distribution of such material. Permission of the Publisher is required for all other derivative works, including compilations and translations. Electronic Storage or Usage Permission of the Publisher is required to store or use electronically any material contained in this work, including any chapter or part of a chapter. Except as outlined above, no part of this work may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the Publisher. Address permissions requests to: Elsevier Science Global Rights Department, at the mail, fax and e-mail addresses noted above. Notice No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made.
ISBN 0-444-50745-0 (Vol. 9) ISBN 0-444-41971-3 (Series) First edition 2001 Library of Congress Cataloging in Publication Data A catalog record from the Library of Congress has been applied for. © T h e paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of Paper). Printed in The Netherlands.
Contributors to Volume 9 A. BARBIER
CEA-Grenoble, Departement de Recherche Fondamentale sur la Matiere Condensee, SP2M, 17, rue des Martyrs, 38054 Grenoble Cedex, France
M.A. BARTEAU
Center for Catal3l;ic Science and Technology, Department of Chemical Engineering, University of Delaware, Newark, DE 19716, USA
S.A. CHAMBERS
Environmental Molecular Sciences Laboratory, Pacific Northwest National Laboratory, PO Box 999, MS K8-93 Richland, WA 99352, USA
C.C. CHUSUEI
Department of Chemistry, P.O. Box 30012, Texas A&M University, College Station, TX 77842-3012, USA
B.G. DANIELS
Surface Science Research Centre and Chemistry Department, Manchester University, Manchester M l 3 9PL, United Kingdom
U. DIEBOLD
Department of Physics, Tulane University, New Orleans, LA 70118, USA
R.G. EGDELL
Inorganic Chemistry Laboratory, University of Oxford South Parks Road, Oxford 0X1 3QR, United Kingdom
H.-J. FREUND
Fritz-Haber-Institut der Max-Planck-Gesellschaft, Department of Chemical Physics, Faradayweg 4-6, 14195 Beriin, Germany
D.W. GOODMAN
Department of Chemistry, P.O. Box 30012, Texas A&M University, College Station, TX 77842-3012, USA
B.E. HAYDEN
Department of Chemistry, University of Southampton, Highfield, Southampton SO 17 IB J, United Kingdom
V.E. HENRICH
Department of Applied Physics, Yale University, P.O.Box 208284, New Haven, CT 06520, USA
K. HERMANN
Fritz-Haber-Institut der Max-Planck-Gesellschaft, Faradayweg 4-6, 14195 Berlin, Germany
H. KUHLENBECK
Fritz-Haber-Institut der Max-Planck-Gesellschaft, Department of Chemical Physics, Faradayweg 4-6, 14195 Berlin, Germany
R. LINDSAY
Surface Science Research Centre and Chemistry Department, Manchester University, Manchester Ml3 9PL, United Kingdom
D.C. MEIER
Department of Chemistry, P.O. Box 30012, Texas A&M University, College Station, TX 77842-3012, USA
C. NOGUERA
Laboratoire de Physique des Solides, UMR CNRS 8502, Universite Paris-Sud, 91405 Orsay, France
G. PACCHIONI
Dipartimento di Scienza dei Materiali, Universita di Milano-Bicocca, Istituto Nazionale per la Fisica della Materia, via Cozzi, 53, 20125 Milano, Italy
G. RENAUD
CEA-Grenoble, Departement de Recherche Fondamentale sur la Matiere Condensee, SP2M, 17, rue des Martyrs, 38054 Grenoble Cedex, France
T. RISSE
Fritz-Haber-Institut der Max-Planck-Gesellschaft, Department of Chemical Physics, Faradayweg 4-6, 14195 Berlin, Germany
G.S. ROHRER
Department of Materials Science and Engineering, Carnegie Mellon University, Pittsburgh, PA 15213-3890, USA
J.L. DE SEGOVIA
Institute de Ciencia de Materiales de Madrid, CSIC E-28049 Cantoblanco Madrid, Spain
A.B. SHERRILL
Center for Catalytic Science and Technology, Department of Chemical Engineering, University of Delaware, Newark, DE 19716, USA
G. THORNTON
Surface Science Research Centre and Chemistry Department, Manchester University, Manchester Ml3 9PL, United Kingdom
E.M. WILLIAMS
Surface Science Research Centre, The University of Liverpool, Liverpool L69 3BX, United Kingdom
M.WITKO
Institute of Catalysis and Surface Chemistry, Polish Academy of Sciences, ul. Niezapominajek, 30239 Cracaw, Poland
Preface During the late 1960s and 1970s the commercial availability of ultra-high vacuum (UHV) systems allowed the development of a plethora of new techniques which were devised to probe materials in a surface-specific fashion, and this in turn led to the creation of modem surface science; the study of the structural, electronic and chemical properties of extremely well-characterised surfaces on an atomic scale. When David King and I first conceived this series of volumes in the later 1970s our objective was to recognise the growing maturity of this new scientific discipline which was already starting to apply these techniques in a combined fashion to understand surface processes. In the 20 years since the first volume was published, this perception has certainly proved to be well-founded, and while new techniques have continued to appear, they have rapidly been assimilated into the general armoury of methods (the increasing pervasiveness of scanning probe microscopies is very evident in the current volume), and it is the combination of methods which has proved most effective. The topic of the present volume, Oxide Surfaces, has been in many ways a 'Cinderella' of surface science; while the importance of oxide surfaces has been long recognised, the perceived difficulty of preparing and probing oxide surfaces (especially insulating oxide surfaces) meant that there was far more attention devoted in the mainstream of the subject to metal and semiconductor surfaces. There were notable exceptions, and some of the authors contributing to this volume have made major contributions to the topic throughout the whole period of development of modem surface science. The study of oxide surfaces has been a major growth area in the last few years, partly encouraged by new methods including the use of epitaxial oxide films grown on metallic substrates. These films are more amenable to study by electron spectroscopies, but raise new issues of surface characterisation and associated properties. This volume provides a view of some of the main areas of development and of recent progress in the study of well-characterised oxide surfaces. The first chapter by Henrich, one of the pioneers of modem surface studies of oxides, and co-author of the first text on the subject, provides an overview of the subject and relates the remaining chapters to this overview. Chapters 2 to 4, by Noguera, by Pacchioni and by Hermann and Witko, are concemed with the theory of oxides surfaces; they cover a range of materialsfi*omsimple rocksalt stmctures such as MgO through to the complexity of transition metal oxides, and also present some complementary methods of modelling and calculation. These theoretical studies also address the key issue of surface defects, and cover some aspects of adsorption at oxide surfaces. In some ways oxide surfaces is a topic in which theory was, for some years, ahead of experiment, and hence unchallenged. This was especially tme in the predictions and
measurements of surface structure, but substantial experimental advances have been made in the last few years in quantitative structure determination of oxide surfaces and this progress is surveyed in Chapters 5 and 6; Renaud and Barbier focus on the achievements of surface X-ray scattering methods (an approach which circumvents one of the main problems of surface charging when using electron probes), while Lindsay, Daniels and Thornton concentrate on other methods, and especially on adsorbate structure determination. A closely related issue is the structural characterisation of thin epitaxial oxide films, and this is described in Chapter 7 by Chambers. Crucial complementary structural information on oxide surfaces structures, especially of Ti02, has been provided in the last few years by scanning probe microscopies, and this is the focus of Diebold's Chapter 11. Because of the important role played by surface stoichiometry of oxides in both structure and reactivity, this work also relates quite closely to the content of Chapters 8 (Freund, Kuhlenbeck and Risse), 9 (Chusuei, Meier and Goodman), 10 (Sherrill and Barteau) and 12 (Rohrer) which all address different aspects of the chemistry and reactivity (including photochemistry) of oxide surfaces (including those of thin films), an area which constitutes one of the primary motivations for the study of oxide surfaces. Chapters 13 (Hayden) and 14 (Egdell) discuss two general aspects of oxide surfaces, their vibrational and electronic properties; while the titles suggest an emphasis on physical properties, vibrational spectroscopies, in particular, are also an important probe of adsorbate interactions and are especially relevant to chemical properties. The final chapter by de Segovia and Williams describes another topic of especial relevance to oxide surfaces and closely related to their photochemistry, that of DIET (desorption induced by electronic transitions).
January 2001
D.P.Woodruff
Contents Preface
ix
Chapter 1 (V.E. Henrich) Metal oxide surfaces and interfaces: concepts and issues 1. Introduction 2. Interesting bulk properties 2.1 Electronic properties 2.2 Crystal structure 3. From bulk to surface 4. Electrons on oxide surfaces 5. Oxide-gas, -liquid and -solid interfaces 6. Adsorption on metal oxides 7. Endnote Acknowledgements References
1 1 2 2 8 9 16 21 25 31 33 33
Chapter 2 (C. Noguera) Clean oxide surfaces: a theoretical review 1. Introduction 2. Numerical methods 2.1 Classical approaches 2.2 Semi-empirical quantum methods 2.3 The ab initio HF method and methods beyond 2.4 The ab initio Density Functional Theory and methods beyond 2.4.1 The Self-Interaction Correction (SIC) method 2.4.2 The GW method 2.4.3 The LDA+U method 2.5 General features for simulating surfaces 3. Review of literature 3.1 Fluorite MO2 and anti-fluorite M2O structures 3.2 Rocksalt structure 3.3 Corundum structure 3.4 Rutile structure 3.5 Wurtzite structure 3.6 Perovskite structure 3.7 Other oxides 4. Discussion of non-polar stoichiometric surfaces 4.1 Atomic structure 4.2 Energetics 4.3 Electron distribution 4.4 One particle excitations 4.5 Two particle excitations 5. Non-stoichiometric surfaces 5.1 Atomic relaxation 5.2 Charge and spin distribution
35 35 36 36 37 38 39 40 41 42 43 44 45 45 49 51 53 54 57 59 59 61 62 65 68 69 70 70
5.3 Spectroscopic signature 5.4 Energetics 6. Polar surfaces 6.1 Criterion for surface polarity 6.2 Models of electronic structure in iono-covalent materials: application to polar surfaces 6.3 Surface processes relevant for polarity healing 6.4 Summary 7. Conclusion References
72 73 75 76 78 80 83 85 86
Chapter 3 (G. Pacchioni) Theory of point defects at the MgO surface 1. Introduction 2. Quantum chemical description of ionic crystals 2.1 Cluster models 2.2 Embedding schemes 2.3 Solutions of the Schrodinger equation 3. Role and nature of defects at the MgO surface 3.1 Low-coordinated cations 3.2 Low-coordinated anions 3.3 Hydroxyl groups 3.4 Anion vacancies 3.4.1 Charge distribution 3.4.2 Hyperfme interactions 3.4.3 Stability and formation energies 3.4.4 Energy levels and ionisation potentials 3.4.5 Barriers to vacancies diffusion 3.4.6 Optical spectra 3.4.7 Chemical reactivity 3.5 Cation vacancies 3.6 Divacancies 3.7 Impurity atoms 3.80" radical anions 3.9(lll)microfacets 4. Conclusions Acknowledgements References
94 95 96 98 99 101 103 107 109 110 111 113 115 116 117 118 120 121 122 124 126 128 129 129 130
Chapter 4 (K. Hermann and M. Witko) Theory of physical and chemical behaviour of transition metal oxides: vanadium and molybdenum oxides 1. Introduction 2. Vanadium oxides 2.1 Vanadium oxide bulk systems 2.1.1 Geometric structure 2.1.2 Electronic properties
136 139 140 140 142
2.2 Vanadium oxide surfaces 2.2.1 Geometric structure 2.2.2 Physical and chemical properties (a) Properties of clean surfaces (b) Surface oxygen vacancies (c) Adsorption 3. Molybdenum oxides 3.1 Molybdenum oxide bulk systems 3.1.1 Geometric structure 3.1.2 Electronic structure 3.2 Molybdenum oxide surfaces 3.2.1 Geometric structure 3.2.2 Physical and chemical properties (a) Properties of clean surfaces (b) Surface oxygen vacancies (c) Adsorption Synopsis Acknowledgement References
148 148 152 152 157 162 168 170 170 172 172 172 175 175 182 186 190 191 191
Chapter 5 (R. Lindsay, B.G. Daniels and G. Thornton) Geometry of adsorbates on metal oxide surfaces 1. Introduction 2. Magnesium oxide - MgO 2.1MgO(100)-H2O 2.2MgO(100)-C2H2 2.3MgO(100)-CO 2.4MgO(100)-CO2 2.5MgO(100)-Ca 2.6MgO(100)-Ag 2.7MgO(100)-Pd 2.8MgO(100)-Ni 2.9MgO(100)-Fe 3. Nickel oxide-NiO 3.1NiO(100)-NO 3.2NiO(100)-CO 3.3NiO(100)-H2S 3.4NiO(lll)-NO 3.5NiO(lll)-CO 4. Titanium dioxide - Ti02 4.1TiO2(110)-HCOOH 4.2 Ti02(l 10) - H3CCOOH 4.3 Ti02(l 10) - H5C2COOH 4.4 Ti02(l 10) - H5C6COOH 4.5 Ti02(l 10) - bi-isonicotinic acid 4.6TiO2(110)-Na 4.7 Ti02(l 10) - Na/C02 4.8 Ti02(l 10) - (NH4)6Mo7024.4H20
199 202 202 206 207 208 209 210 212 212 213 213 214 216 217 218 219 219 221 223 224 224 225 226 227 228
4.9 Ti02(l 10) - {Rh(CO)2Cl}2 4.10TiO2(110)-SO2 4.11TiO2(110)-K 5. Aluminium oxide - AI2O3 5.1a-Al2O3(0001)-Cu 5.2 y-like AbOsCOOO 1) - di-tert-butyl nitroxide 6. Chromium oxide - Cr203 6.1Cr2O3(0001)-CO2 6.2Cr2O3(0001)-NO 7. Iron oxide - FexOy 7.1 FeO(lll)-H5C6CHCH2 7.2 Fe304(l 11) - H5C6CHCH2
230 231 232 233 233 234 235 235 236 236 237 237
8. Zinc oxide-ZnO
238
8.1 ZnO(loiO) - HCOOH
238
8.2ZnO(10i0)-CO2
^^^
8.3ZnO(10i0)-C6H6
^^^
8.4ZnO(10i0)-C5H5N
240
8.5 ZnO (lOiO) - C6N3H5,C7N2H6, C7N2H6, C7N3H7
241
8.6ZnO(000i)-CO
242
8.7ZnO(000i)-CO2
243
8.8 ZnO (OOOi) - HCOOH 8.9ZnO(000i)-C5H5N 8.10 ZnO (OOOi)-K 8.11ZnO(0001)-CO 8.12ZnO(0001)-CO2 8.13 ZnO(OOOl) - HCOOH 8.14 ZnO(OOOl) - CH3OH 8.15ZnO(0001)-C6H6 8.16 ZnO(OOOl) - CeHsOH 8.17 ZnO(OOOl) - C5H5N 9. Copper (I) oxide - CU2O 9.1 C u 2 0 ( l l l ) - C H 3 0 H Acknowledgements References
243 244 244 246 246 247 247 247 248 248 249 249 249 250
Chapter 6 (G. Renaud and A. Barbier) Atomic structure of oxide surfaces by surface X-ray scattering 1. Introduction 2. X-ray scattering by surfaces 2.1 Grazing incidence X-rays 2.2 Basic X-ray scattering 2.3 Diffraction by a surface 2.4 Instrumental considerations 3. Surfaces of oxide single crystals
256 258 258 259 260 262 262
3.1 Specific sample preparation and requirements 3.2 MgO(OOl) surface 3.3 a-Al2O3(0001)(lxl) and reconstructed surfaces 3.4 TiO2(110Hlxl) and (001Hlx3) surfaces 3.5 NiO(l 1 l)-p(2x2) reconstructions 3.5.1 NiO( 111 )-p(2x2) structure after air annealing 3.5.2 NiO(l 1 l)-p(2x2) structure after controlled reduction 3.5.3 NiO(l 1 l)-p(2x2) comparison and conclusion 3.6CoO(lll)andMnO(lll) 3.7 ZnO(OOOl) and ZnO (1010) 3.8 SrTiOsCOOl) 3.9Cr2O3(0001Hlxl) 4. Structure of thin oxides films 4.1 NiO(l 1 l)-p(2x2) thin film structure 4.2 AI2O3 layer on NiAl 4.3 Fe203 and Fe304 layers 4.4 Cr203 layers 5. Conclusions and future outlook Acknowledgements References
262 263 266 273 276 277 279 281 281 284 286 288 289 289 292 293 293 294 295 296
Chapter 7 (S.A. Chambers) Structure of thin epitaxial oxide films and their surfaces 1. Introduction 2. Oxidation chemistry and its effect on surface and film structure 2.1 Fe304(l 11) and a-Fe3O4(0001) on Pt(l 11) and a-A^OsCOOOl) 2.2 Cr02 on Ti02(l 10) & (100), and a-Cr203 on a-Al203 3. Lattice mismatch and residual strain fields 3.1 (a-Fe203/a-Cr203)n on a-Al203 4. Crystal symmetry mismatch 4.1 Fe3O4onMgO(001) 5. Conclusions Acknowledgements References
301 303 303 309 313 314 316 317 320 323 323
Chapter 8 (H.-J. Freund, H. Kuhlenbeck and T. Risse) Molecules on well-structured oxide surfaces 1. Introduction 2. Surfaces of oxides with rocksalt structure 3. Surfaces of oxides with corundum structure 4. Surfaces of oxides with rutile structure 5. Surfaces of oxides with layered structures 6. Synopsis References
326 327 341 359 363 366 367
Chapter 9 (C.C. Chusuei, D.C. Meier and D.W. Goodman) Atomic-scale chemical and electronic structure studies of well-defined metal oxide surfaces 1. Introduction 2. Magnesium oxide 2.1 Adsorption of benzene on MgO/Mo(l 00) studied by MIES/UPS 2.2 MIES and TPD analysis of CH3OH and D2O adsorption on MgO/Mo(l00) 2.3 Surface defect characterisation of MgO by MIES 3.Titania 3.1 Structural and electronic properties of Au on Ti02( 110) 3.2 Adsorption of CO on Ai/Ti02 4. Thin-film alumina and mixed-metal oxides 4.1 STM imaging of AI2O3 thin films 4.2 NO adsorption on MgO-NiO and CaO-NiO 5. High pressure STM studies of supported gold clusters 5.1 Imaging Au on Ti02(l 10) at elevated pressures with variable temperature and pressure STM 5.2 Imaging solution-deposited Au6(PPh3)6[BF4]2 on Ti02(l 10) 6. Summary Acknowledgements Reference
373 374 374 378 381 385 385 388 392 393 395 399 400 402 404 405 405
Chapter 10 (A.B. Sherrill and M.A. Barteau) Principles of reactivity from studies of organic reactions on model oxide surfaces 1. Introduction and scope 2. Titanium dioxide as a model for transition metal oxides 3. Case study I: formic acid as a probe of surface properties 3.1 Formic acid decomposition on metal oxides 3.2 The participation of oxygen vacancies in formic acid decomposition 3.3 Tracking formic acid decomposition with scanning probe microscopy 3.4 Catalytic reactions of formic acid on titanium dioxide (110) 3.5 Extension to higher carboxylic acids 4. Case study II: thermal and photoreactions of methanol on Ti02 4.1 Thermal reactions of methanol on Ti02 single crystals 4.2 Thermal reactions of methanol on Ti02 powders 4.3 Thermal reactions of methanol on Sn02 - another rutile oxide 4.4 CH3OH decomposition on fluorite metal oxides 4.5 Decomposition of higher alcohols on titanium dioxide single crystal surfaces 4.6 Higher alcohols: extension to poly crystalline Ti02 and other materials 4.7 Photoreactions of alcohols on Ti02 5. Summary References
409 409 412 412 415 418 420 422 424 425 428 429 430 432 435 437 439 440
Chapter 11 (U. Diebold) The structure of Ti02 surfaces 1. Introduction 2. Bulk structures 2.1 Bulk defects 3. The structure of the rutile Ti02(l 10) surface 3.1 The rutile TiOiCl 10)(lxl) surface 3.1.1 Bulk truncation 3.1.2 Relaxations 3.1.3 Appearance in STM and AFM 3.1.4 Surface defects Step edges Oxygen vacancies created by annealing Oxygen vacancies created by other means Line defects Impurities Crystallographic shear planes 3.2 Reconstructions 3.2.1 Reconstruction under reducing conditions: the structure(s) of the (1x2) phase 3.2.2 Restructuring under oxidising conditions 3.3 Recommendations for surface preparation 4. The structure of the rutile (100) surface 4.1 The TiO2(100)(lxl) surface 4.2 Reconstructions 4.2.1 The microfacet model of the rutile TiO2(100)(lx3) surface 4.2.2 Is the simple microfacet model valid? 5. Rutile (001) 6. Vicinal surfaces 7. Anatase surfaces 7.1 Anatase(lOl) 7.2 Anatase (001) 8. Conclusion References
443 444 445 446 447 447 449 451 455 455 457 459 459 460 461 463 463 465 469 470 470 472 472 473 474 475 476 476 478 479 480
Chapter 12 (G.S. Rohrer) The anisotropy of metal oxide surface properties 1. Introduction 2. Surface character-property relationships 2.1 Particulate oxidation catalysts 2.2 Single crystal surfaces 2.3 Thin films 3. Factors determining surface morphology 4. Factors influencing surface stoichiometry 4.1 The solid-vapour equilibrium 5. Reaction anisotropics on M0O3 surfaces 6. The orientation dependence of the photochemical reactivity of Ti02
485 486 486 488 489 490 496 497 501 506
7. Conclusion Acknowledgement References
510 510 511
Chapter 13 (B.E. Hayden) Vibrational spectroscopy of oxide surfaces 1. Introduction 2. Theoretical and experimental considerations 2.1 High resolution electron energy loss spectroscopy (HREELS) 2.2 Reflection absorption infrared spectroscopy (RAIRS) 3. Surface vibrational modes of the substrate 4. Vibrational spectroscopy of adsorbates 5. Vibrational spectroscopy of adsorbates on oxide supported metals References
514 515 516 521 530 532 539 546
Chapter 14 (R.G. Egdell) Electronic structure of oxide surfaces 1. Introduction 2. Electronic structure of oxides: general considerations 2.1 Models of oxide electronic structure 2.2 Covalency and band dispersion in oxides 2.3 Surface states in oxides 3. Experimental techniques for probing electronic structure of oxides 3.1 Photoemission and inverse photoemission spectroscopies 3.2 Electron energy loss spectroscopy 3.3 Scanning tunnelling spectroscopy 4. Selected case histories 4.1 The rocksah MgO(lOO) surface 4.2 Transition metal rocksalt monoxide (100) surfaces 4.3 Sn02 4.4 Ti02 4.4.1 Comparison of Ti02 and Sn02 4.4.2 Ti02(l 10) 4.4.3 Ti02(l00) 4.5 WOsandNaxWOs 4.5.1 WO3(001) 4.5.2 NaxWOsCOOl) 5. Concluding remarks References
550 551 551 552 553 554 554 556 557 558 558 565 572 577 577 578 585 588 588 593 598 598
Chapter 15 (J.L. de Segovia and E.M. Williams) Desorption induced by electronic transitions, DIET, at oxide surfaces 1. Historical introduction to ESD and PSD 2. Ejection mechanisms: the Feibelman-Knotek model 3. DIET at oxide surfaces 3.1 Oxidised surfaces
608 614 617 618
3.2 Stoichiometric oxides 3.2.1 The Ti02 model surface 3.2.2 The TiO2(100)(lx3) and (110)(lx2) reconstructed surfaces 3.2.3 Some further oxide surfaces studied by DIET 4. Adsorption and reactivity at oxides by DIET 4.1 Small molecules 4.2 Large molecules 5. Summary of relevant kinetics parameters 6. Conclusions Acknowledgements
618 618 622 623 625 625 631 63 8 638 638
Index
645
This Page Intentionally Left Blank
Oxide Surfaces D.P. Woodruff, editor © 2001 Elsevier Science B. V. All rights reserved.
Chapter 1
Metal oxide surfaces and interfaces: concepts and issues Victor E. Henrich Department of Applied Physics, Yale University, P.O. Box 208284, New Haven, CT 06520, USA 1. INTRODUCTION The physics and chemistry of metal oxide surfaces, from the perspective of surface-science experiments and theoretical calculations on well characterized single-crystal samples, was reviewed up to 1992 in Ref 1. Since that time, however, several new areas have emerged in the study of oxide surfaces, and powerful new techniques are being applied to oxides. In 1992, only a few groups had used scanning probe microscopies (SPM) to study metal oxides. A tremendous amount has now been learned from such studies. One of the most important discoveries is that metal oxide surfaces can be far more complex than had been anticipated [2]. Another area that had barely begun in 1992 was the growth of epitaxial oxide films on either single-crystal metal or other oxide substrates. That field has now yielded detailed studies of the surface and chemisorption properties of insulating oxides that cannot be studied by many techniques on bulk crystals because of surface charging. Interesting surface structures have also been produced that are not stable on bulk crystals. Still another new area of oxide surface science is in situ studies of oxide surfaces and chemical reactions in aqueous solution; such oxide-aqueous solution interfacial properties are extremely important in geochemistry [3]. The present volume summarizes the current state of the art in many aspects of the chemistry and physics of metal oxide surfaces. While it cannot be exhaustive, it should prove to be a valuable resource for those interested in the present — and future — of metal^ oxide surface research. The technological and commercial interest in metal oxide surfaces and interfaces has also continued to grow. For decades, oxide surfaces have played a key role in corrosion protection, catalysis, sensors, fuel cells, ceramics, etc. Over the last few years, totally new devices and technologies that rely on the properties of oxide surfaces and interfaces have emerged. Non-volatile ferroelectric field-
effect transistors have been fabricated that will be extremely useful in memory appHcations [4]. Novel metal-oxide-metal spin valve structures are being explored for high-density magnetic storage readout [5]. The properties of oxideoxide interfaces have been invoked in using the ferroelectric field effect to drive a quantum phase transition between normal and superconducting modes in thin oxide superconductor films in order to produce a new type of solid-state switch (an idea that was first proposed in 1956, but has only recently been experimentally realized) [6,7]. The list goes on and on, but the point is that metal oxide surfaces and interfaces will play an ever more important role in a wide raage of future technologies. Before delving into the detailed material contained in the other chapters of this volume, there are some important concepts and issues — some of them unique to oxides — that should be discussed. This chapter highlights a few of those and considers ways in which one should think about the properties of metal oxides. For more information, the reader should consult Refs. 1 and 8 - 1 1 .
2. INTERESTING BULK PROPERTIES Before discussing the surface properties of metal oxides, it is instructive to consider some of the important bulk electronic and geometric properties that characterize them and determine their behavior. 2.1. Electronic properties Perhaps the most important property of metal oxides that determines their electronic structure is their ionicitv. The driving force for this is the desire of oxygen ions in crystal structures to have a closed shell 2s^p^ electronic configuration, thus making the oxygen ions formally O^" (i.e., the "oxide" ion). (Formal ionic charge is an extremely useful chemical concept; it is commonly used in describing ionic materials, and we shall use it here. However, the actual distribution of electron density in a crystal structure is too complicated to fit neatly into such a simple concept [1].) The most highly ionic metal oxides are nontransition-metal oxides such as MgO, whose electron charge density in the crystal closely resembles that of an assembly of Mg^^ and O^" ions. It should be noted that the "free" O^" ion does not exist, however; it is unstable by 7.5 eV compared to the free O atom unless it is stabilized by surrounding positively charged ligands [12]. Covalent bonding is much more important in the later transition-metal oxides such as NiO and La2Cu04 [9]. However, even these relatively covalent metal oxides have a significant ionic component to their bonding and thus also posses the properties that derive from ionicity. tonicity in metal oxides is also discussed in Chaps. 2, 4 and 14. Since they are composed of charged ions, an extremely important property of metal oxides, as for all ionic compounds, is their stoichiometry. Because
Coulomb forces are both large and long range, there is a strong driving force for any ionic material to maintain local charge neutrality. This neutrality must be maintained, both in the bulk and at surfaces and interfaces, on a spatial scale of only a few atoms. As will be discussed throughout much of this book, charge neutrality plays a dominant role in both surface geometric and electronic properties of metal oxides. However, it is equally important in the bulk. For example, a single missing O ion (an "F center") in MgO can exist as a neutral center (i.e., two electrons reside at the defect to compensate for the charge on the missing O^" ion), or it can have only one electron, or no electrons, associated with it. In the last two cases, the defect has a net charge of 1+ or 2+, respectively. Such charged F centers exist, but they must be accompanied by defects of opposite sign nearby in order to maintain local charge neutrality. Charge neutrality is also responsible for the increased catalytic activity of some metal oxides. This is the case when MgO is substitutionally doped with lithium. Since the only possible ionic state for lithium is Li^, an adjacent oxygen ion must be O" (formally). Since it does not have a closed shell electronic configuration, O" is much more chemically reactive than O^", and it is believed that the unique catalytic properties of Li-doped MgO result from the presence of a Li^O" site [10]. This and other defects in MgO are discussed further in Chap. 3. The ionic nature of metal oxides manifests itself in strong spatial localization of electronic charge in the crystal structure. Since both the cations and anions prefer to exist in a particular valence state in a given crystal structure, it costs energy to add electrons to or remove them from an ion. Thus the energy state of a particular electron is strongly dependent on the number and state of other electrons around it; this is referred to as electron correlation [1,9]. Correlation is the antithesis of simple free-electron band theory, in which electronic states are calculated for a particular crystal structure, regardless of electron occupation, and electrons are then placed in those states until the proper total electronic charge is obtained. As an example, consider electrical conduction in a rocksalt transition-metal oxide such as NiO. This is shown schematically in Fig. 1. The ground state electron configuration of the nickel ions is Ni^^ [the oxygen ions are (formally) taken to be O^"], or 3d^ (d" in Fig. 1). Since only the cations have incomplete shells, electrical conduction corresponds to removing a d-electron from one cation and placing in on a nearby cation: d" d"" -^ d^'^ d"^\ As shown in the lower half of Fig. 1, there is an energy cost for this process of U, which is referred to as the Coulomb repulsion energy, or "Hubbard U". For NiO, U is about 5 - 7 eV. Since this is an enormous energy compared to kT at room temperature (0.025 eV), NiO an insulator even though its 3d shell is only partially occupied. The Hubbard U has no analogy in free-electron band theory.
Charge transfer energies U and A d" d" - d"-' cf ^'
d"-d"^^L
•J
L = hole on O " ligand
n-1
2w
ja+1
d"-' d^ W + W
A
d'd^^^L
u n n
d d
Fig. 1. Schematic representation of the Coulomb repulsion energy, U, and the charge transfer energy, A, in an ionic compound. The lower figure shows the energy levels of the system for those excitations (see text).
Also shown in Fig. 1 is the excitation of an electron from an O^" ion to an empty cation orbital; this is referred to as an interatomic charge transfer excitation (usually called simply a "charge transfer excitation"). In this process, an electron is removed from an O^" ion (L denotes a hole on an oxide ion) and is placed on a cation, making its configuration d^^^. The energy required for this process is the charge transfer energy. A; for NiO, A - 4 eV. This excitation does have an analog in band theory; A is the bandgap energy. However, it cannot be calculated from free-electron band theory since correlation is still present in the change of valence state of both the anion and the cation. Free-electron band theory is able to adequately describe the ground states of some metal oxides (e.g., non-transition-metal oxides and d^ transition-metal oxides), but, because of correlation, it cannot be used for excited states [1,9]; see Chap. 2. However, the terminology and general concepts of band theory are still useful, and it is common to use them in discussing oxides; we will do so in this chapter. For simple d^ metal oxides such as MgO and Ti02, the (predominately) O 2p orbitals correspond to the valence band, and the lowest energy empty cation orbitals correspond to the conduction band [9]. The correspondence becomes less clear in d"" transition-metal oxides, for example, since the highest energy occupied orbitals may be on the cations, and the oxide may be an insulator although the d-orbitals are only partially occupied (e.g., NiO). However, in cases where the use of band terminology would not cause confusion, it is often used. A useful distinction exists between oxides of the transition metals and those of the pre- or post-transition metals (i.e., non-transition-metal oxides) [1]. Nontransition-metal oxides are ones in which the cation bonding orbitals are of s or p character. In the alkali metal oxides, the cations can only exist in a 1+ valence state; for alkaline earth oxides, the cations are always 2+. The important oxides of Group IIIB metals, AI2O3 and In203, can only have 3+ cations. One important exception to this is tin, which can have either a 2+ or 4+ valence. (This may not be surprising, however, since Sn is in Group IVB along with diamond and the highly covalent semiconductors silicon and germanium.) Thus both SnO and Sn02 are stable bulk compounds, and some of the defect properties of Sn02 more nearly resemble those of transition-metal oxides than of other non-transition-metal oxides. Transition-metal cations, on the other hand, in which bonding occurs via delectron orbitals, can generally exist in more than one stable oxidation state; the partially filled d-orbitals also give the oxide a component of directional covalent bonding. For example, molybdenum can exist as Mo^^, Mo^^, Mo"^^, Mo^^ and Mo^^; manganese has stable valences states of Mn^^, Mn^^, Mn"^^, Mn^^ and Mn^^. This opens up a whole new dimension of complexity in both bulk and surface electronic structure. Bulk phases of most of the oxidation states are sta-
ble, each one having a different geometric and electronic structure than the others. New types of bulk defects also become allowed. For example, if an O atom is removed from M0O3, the two electrons that must be left behind in the vicinity of the O-vacancy in order to maintain local charge neutrality can occupy dorbitals on two neighboring Mo ions, changing them from Mo^^ to Mo^^; or they could both reside on the same Mo ion, making it Mo"^^. The polyvalent oxides of molybdenum and vanadium are discussed in Chaps. 2, 4 and 12. Oxides of the lanthanide rare earth elements share some of the properties of transition-metal oxides, at least for cations that can have two stable valence states. (None of the lanthanide rare earth cations have more than two ionic valence states.) Oxides of those elements that can only have a single ionic valence are subject to the limitations imposed on similar non-transition-metal oxides. One actinide rare-earth oxide, UO2, has understandably received quite a bit of attention from surface scientists [1]. Since U can exist in four non-zero valence states, UO2 behaves more like the transition-metal oxides. The electronic properties of rare-earth oxides differ from those of transition-metal oxides, however, because of the presence of partially filled f-electron shells, where the f-electrons are spatially more highly localized than are d-electrons. In some transition-metal oxide systems, including those of Ti, V and Mo, a far larger number of stable bulk phases can exist than there are cation valence states (see Chaps. 4, 11 and 12). This occurs through the formation of structures, called Magneli phases, in which single-crystal regions of one cation valence structure are separated by regularly spaced shear planes on which lower valence cations exist [9]. An example is the class of compounds Tin02n-i, in which regions of the rutile Ti02 structure are separated by shear planes occupied by Ti^^ cations, having a local structure similar to that of corundum Ti203. These planes have been observed intersecting the surface of heavily reduced Ti02 crystals by STM; see the discussion in Chaps. 11 and 14. The rutile-based Magneh phases Tin02-n that have been prepared include: TigOis, TiyOo, Ti60ii, Ti509,Ti407andTi305[13].
Another result of the existence of multiple cation valence states in transition-metal oxides is that the population of those states can easily be changed by optical excitation; that is shown as the charge transfer excitation in Fig. 1. The change in population of cation valence states changes the local electronic structure, which can change the chemical activity of surfaces. For example, the corundum oxide hematite, a-Fe203, contains all Fe^^ cations. When light of energy greater than A is incident on the material, electrons can be excited from O^' anions to Fe cations, changing their valence state to Fe^' [14]. This is the other stable valence state for iron, and the excited state may exist long enough to produce measurable changes in chemical activity. This particular system is of in-
terest in solar photochemistry, since A lies in the visible region of the solar spectrum; this will be discussed in more detail in § 6 below. Some transition-metal oxides exhibit metal-insulator transitions as a function of temperature, pressure, composition, etc. that may produce large changes in bulk electronic structure (see Chap. 4). In corundum V2O3, for example, there is a metal-insulator transition at about 150 K in which the sample's conductivity changes by seven orders-of-magnitude within a few degrees K [9]. The electronic behavior becomes more complex if V2O3 is doped with up to 2 at. % Cr (which goes in substitutionally). Cr-doped V2O3 exhibits the same large metalinsulator transition near 150 K, where it goes from a low-temperature antiferromagnetic insulating phase to a high-temperature paramagnetic metallic phase [15]. However, it displays an additional transition from the paramagnetic metallic phase to a higher-temperature paramagnetic insulating phase at a temperature (around room temperature) that depends on the Cr content. The latter transition produces about a two order-of-magnitude change in conductivity. In this particular system, there is a structural phase change (from monoclinic to corundum) at the low-temperature transition, but no symmetry change across the high-temperature one. This high-temperature transition has recently been investigated for use in positive-thermal-coefficient thermistors [16]. The strong electron correlation effects in metal oxides complicate the interpretation of excited-state spectra in a wide range of experimental techniques [1]. In materials that are well described by itinerant electron band models, such as many metals and elemental semiconductors, electrons can be excited, or added to or removed from the material, without significantly altering the energy level structure of the remaining electrons. In correlated materials, however, changing the state of an electron changes the energies of other electrons in the system, and the concept of a single-electron density-of-states becomes meaningless [9]. Consider two common, powerful experimental methods that are used to determine the spectrum of filled and empty electronic levels in solids: photoemission, and inverse photoemission (or Brehmstrahlung isochromat spectroscopy) [17]. In photoemission, an incident photon excites an electron from its ground state to an allowed, normally empty state above the vacuum level, from which the photoexcited electron is emitted into vacuum, where its kinetic energy is measured with an electron spectrometer. Without correlation, the initial state of the electron can be determined by subtracting its kinetic energy in vacuum from the incident photon energy, properly taking the work function of the surface into account. However, if the initial state of the system (i.e., the atom or ion that will be excited) contained n electrons, then the final state contains only {n-l) electrons. When the photoexcited electron is removed, the other electrons in the vicinity (on both the excited atom and its neighbors) change their energies. Thus the state that the photoemitted electron senses will not be the initial ground state
of the system. In late transition-metal oxides such as NiO, for example, photoemission spectra of both valence and core-level electrons exhibit multiple features (sometimes called the "main line" and "satellites", or "well screened" and "poorly screened" final states) that arise from different electron configurations seen by different exiting electrons [1,18]. Thus the interpretation of experimental photoemission spectra is not straightforward; it is usually necessary to compare the experimental spectra with the results of a theoretical model that includes both the n- and (/7-l)-electron states [18]. Additional aspects of photoemission from oxides are considered in Chap. 14. In inverse photoemission, an electron is added to the system in some allowed state above the vacuum level, and the energy of the photon emitted as it falls into some lower energy allowed (normally empty) state is measured. Without correlation, it is straightforward to determine the spectrum of empty electronic states in such an experiment in a manner analogous to that for photoemission. However, the ground, or initial, state of the system has n electrons, while the final state has (n+\). In correlated materials, the addition of the extra electron perturbs the energies of other electrons, so that the empty-state levels involved in the process are not those of the ^-electron ground state. Again comparison with calculations for both the n- and (/i+l)-electron states is required. 2.2. Crystal structure Metal oxide crystal structures consist of metal cations coordinated in various ways by oxygen ligands [1,9,19]. Most common oxides have their cations octahedrally coordinated with six O ions, although in different crystal structures the octahedron will be distorted in different ways. Such structures include rocksalt, rutile, corundum, anatase and molybdenum trioxide. A few metal oxides have tetrahedrally coordinated cations, such as the wurtzite structure of ZnO. The spinel and inverse-spinel structures have a mixture of octahedrally and tetrahedrally coordinated cations. In the perovskite structure, ABO3, the B cation is octahedrally coordinated, while the A ion has twelve nearest-neighbor O ligands. Other ligand coordinations also occur, such as the planar configuration in some high-Tc oxide superconductors. However, the predominance of octahedral and tetrahedral ligand coordination leads to one of the common ways of graphically representing metal oxide crystal structures. Instead of showing the individual ions, the structure is drawn as a collection of octahedra or tetrahedra, where adjacent polyhedra share common corners, edges or faces [9]; this type of representation is shown in Fig. 2 for three oxide crystal structures whose cations are octahedrally coordinated. In the rhenium trioxide structure, all octahedra share only comers. In rutile, both cornerand edge-sharing are present, and corundum contains a mixture of edge- and face-sharing. This is a widely used representation (see Figs. 2 and 10 in Chap. 12), and it is extremely useful for understanding bulk structures such as shear
IV Vlv V y
(a)
(b)
\ /
M' \
Z-
(c)
Fig. 2. Connected polyhedra representation of the bulk crystal structures of (a), rhenium trioxide; (b) rutile; and (c) corundum. [From Ref 9, by permission of Oxford University Press.]
planes in Magneli phases, etc. However, it is not a particularly good starting point for picturing atomic arrangements on surfaces. A better approach for that purpose is a ball-and-stick model, where all atoms or ions, and their relative positions, are represented [19]. (The one misleading feature of such models for ionic crystals is that the sticks which hold the balls together may be mistaken for directional chemical bonds.) Another good way to schematically show crystal structures is with juxtaposed spheres having the proper relative ionic radii [1,19], such as those in Figs. 3 - 5 below; this method is used extensively in other chapters in this volume. The bulk crystal structures of many oxides are discussed at length in Refs. 9 and 19. Our interest here is in determining the geometric structure of various surfaces that can exist on bulk oxide crystals. 3. FROM BULK TO SURFACE When one considers how a semiconductor crystal will separate in order to create a surface along a particular crystallographic plane, one of the considerations is to break the smallest number of directional covalent bonds. Ionic crystals must be considered somewhat differently [1]. Li a highly ionic oxide such as MgO, the ions are nearly spherical distributions of electronic charge; there are no directional bonds. Since cations in ionic crystals are stabilized by their neighboring anions, and visa versa, one should think in terms of retaining as many ligands around each surface ion as possible (i.e., minimizing coordinative un-
10
saturation). Symmetry dictates that, where possible, the two opposing surfaces formed by cleaving along a plane should be identical. A third, and very important, consideration is not to induce any net electric dipole moment normal to the surface. That will be the case if each plane parallel to the surface has zero net charge; such a plane is said to be "charge neutral". (In the terminology of semiconductor surfaces, that is referred to as "autocompensation" [20]; see also Chaps. 7 and 11 in this volume.) Surfaces that do have a net dipole moment are called "polar" surfaces, and the dipole electric field leads to a divergence in the energy of the crystal [21], making the surfaces unstable; see Chap. 2. (There are indications from work on iron oxides that the situation may be more complex if there is a large amount of interlayer relaxation at the surface [22].) A good example of these principles is the rocksalt structure; we will consider MgO to be specific. In the bulk crystal structure, both cations and anions have six nearest-neighbor ligands in an octahedral configuration; this is the ideal configuration for stabilizing both the O^' and Mg^^ ions. Consider three lowindex surfaces of this crystal: (100), (110) and (111). The only way to form a (100) surface is to cleave the crystal between two bulk (100) planes. The resulting surface consists of an alternating square array of Mg and O ions, with each having five nearest-neighbor ligands (see Fig. 3 below). Since each atomic plane parallel to the (100) surface is atomically flat and contains an equal number of Mg and O ions, it is charge neutral, and there is no net dipole moment in any direction. Both of the surfaces created by the cleave also have exactly the same structure. For the (110) surface, the only possible cleavage plane produces two identical surfaces which have alternating lines of Mg and O ions, with each ion having four nearest-neighbor ligands. Planes parallel to the surface are charge neutral, so there is no net dipole moment. The (111) surface is fundamentally different than either the (100) or (110). There is also only one plane along which one can fracture the bulk crystal structure, but it lies between a plane of Mg^' ions and a plane of O^" ions. The two surfaces produced are fundamentally different in that one consists of an entire plane of Mg ions, while the other contains only O ions. Since planes parallel to the surface have a net charge, and the sign of that charge alternates from plane to plane, there is also a large net dipole moment normal to the surface. To make matters worse, each ion in the surface plane has only three nearest-neighbor ligands, compared to the six that it would prefer to have for stability. Needless to say, the rocksalt (111) surface is not stable; it must either reconstruct or be stabilized by the presence of charged species adsorbed on the surface. If an ionic crystal has two opposite polar faces, then, in order for the total crystal to be charge neutral, the two opposing faces must be of opposite types. This effect can readily be observed macroscopically; a well known case is ZnO. ZnO has a tetrahedrally coordinated wurtzite crystal structure; in the [0001] direction, it consists of alternating planes of Zn^^ and O^' ions [1]. When a sample
11
is prepared with opposite {0001} faces, one of them must consist of exposed O ions [the (OOOT) face], and the other will have exposed Zn ions [the (0001) face]. [Both surfaces will relax so as to try to increase the coordination of the exposed surface ions; this relaxation is greater on the (0001) face since the smaller Zn^^ ions can more easily move down between the larger O^" ions.] The O- and Znfaces etch differently; aa HCl etch produces a macroscopically smooth (0001) surface, but a rough, matte (OOOT) surface [23]. This is, in fact, a simple way to determine the orientation of a sample. The above considerations are borne out experimentally on most rocksalt ionic compounds. For example, when magnesium metal is burned, the tiny MgO smoke particles that are formed are almost perfect cubes (see Fig. 2.4 in Ref. 1). The need to form a non-polar surface and to maximize the ligand coordination of surface ions makes the (100) surface energy much lower than that of other possible surfaces in the rocksalt structure. This is also manifest in the cubic shape of grains of table salt, NaCl. The (110) surface of MgO, whose ions are only four-fold coordinated, is also much less stable than the (100) surface [24]. The various types of atomic reconstructions and rearrangements that exist on oxide surfaces, and their implications for surface electronic structure, are discussed in Chaps. 2, 6 - 8, 11, 12 and 14 in this book and will not be discussed here. However, one interesting point should be mentioned in connection with the MgO surfaces described above. It is possible that the reduced ligand coordination of surface ions could lead to charge transfer between them, with a consequent change in the valence state of surface ions. For MgO, one might expect surface ions to be less ionic than those in the bulk, say Mg^ and O". This was studied for MgO (100) by electron energy loss spectroscopy (EELS) [25]. The experimental results suggested that the valence state of the surface ions was very nearly Mg^^ and O^'; i.e., that little such charge transfer actually occurs. This concept will be considered again below in connection with steps on oxide surfaces. It is not always straightforward to determine what the surface structure of a crystal will be along a particular crystallographic plane. A good example of this is the (0001) surface of the very important corundum structure. The corundum crystal structure is trigonal and can be considered as planes of nearly closepacked O ions perpendicular to the three-fold [0001] axis, separated by planes of metal cations. A first glance at the structure would thus suggest that the crystal would have to be separated between a plane of O anions and one of metal cations, which would result in polar surfaces similar to rocksalt (111). A closer look at the structure, however, reveals that the plane of cations in not atomically flat; one-half of the cations lie closer to one of the 0-ion planes, and the other half lie closer to the other 0-ion plane. If one separates the crystal between the two planes of cations, keeping each metal cation with the O plane to which it is
12
closest (i.e., more tightly bound), the resulting surfaces are charge neutral, have no net dipole moment, and both faces of the cleave have identical structures. Experimental evidence on the corundum metal oxides that have been studied to date suggests that this is indeed the case [1]. In the above discussion, we considered how a crystal should cleave along a particular plane that we chose. However, the crystal may have other ideas; in fact, it is not in general possible to predict from the simple arguments above how easily a crystal will separate along a particular plane. Rocksalt is a simple case, and many ionic rocksalt compounds cleave well — and only — along (100). Some metal oxides having the corundum structure, including Ti203 and Cr203, cleave well along the (10T2) plane (which is the plane that contains the empty octahedral cation sites in this structure) [1], while corundum itself, aAI2O3, does not exhibit any easy cleavage planes. An even more striking case is the rutile structure of Ti02. Its most stable surface plane is (110), which is shown in Fig. 5 below and will be discussed in connection with surface defects. In the bulk rutile structure, the Ti cations are octahedrally coordinated with O ions. On the (110) surface, one-half of the cations still have the bulk six-fold ligand coordination, with the other half having five ligands. This surface has been thoroughly studied by a variety of techniques, as is describe in later chapters, and the predicted structure is indeed observed. However, it is not a cleavage plane; there are no easy cleavage planes in rutile. In spite of this, the considerations of surface stability discussed above are important even for surfaces that must be prepared by methods other than cleaving. Surface morphology is considered more generally in Chap. 12. A few metal oxides have layered bulk structures; two of these — V2O5 and M0O3 — are discussed at length in Chaps. 4, 8 and 12 below. In these compounds, there is strong ionic-covalent bonding within a layer consisting of several planes of atoms, but between these layers there is only weak van der Waals bonding (the layers are charge neutral and do not contain any net dipole moment). The oxides thus peel apart hke mica or graphite. The resulting surfaces expose almost entirely O ions, have a structure almost exactly the same as in the bulk crystal, and are chemically quite inert. The criteria given above for stabiHty of crystal surfaces can also be applied to steps on surfaces. Again consider the rocksalt MgO structure for example. Figure 3 shows an idealized ball model of this (100) surface, including missing Mg and O ions and a monatomic height step. The ions on flat terraces each have five nearest-neighbor ligands. In order to form a step from one terrace to another, the ligand coordination of the atoms along the top edge of the step must be reduced. However, the most stable steps should be those in which the ligand coordination is reduced as little as possible. For rocksalt, that is the step in the [010] direction shown. Each ion along the top edge of the step has only four
13
Rocksalt
[010] [001]
Fig. 3. Model of the rocksalt (100) surface. Large circles are O anions, small circles are metal cations. A [010] step to another (100) terrace is shown, as are both missing anion and missing cation point defects.
ligands; however, the stoichiometry of the crystal is maintained along the step, and the step-edge ions can thus retain the same valence states that they have on the terraces (nearly Mg^^ and O^" in this case). This is indeed the step direction that is observed on MgO (100) surfaces. Another example of steps on oxide surfaces is shown in Fig. 4 for the corundum structure. (One anion point defect is also shown in the figure.) As mentioned above, some corundum metal oxides cleave well along the (10T2) plane, along which the empty octahedral cations sites in the crystal are arranged; cleaving along this plane minimizes coordinative unsaturation of the surface ions. From applying the above principles, it was concluded that the most stable monatomic height step on this surface would be that in the [0221] direction shown [26]. As for the rocksalt case, the step-edge cation coordination is reduced from five on the terraces to four along the top of the step. However, the correct stoichiometry of the crystal is maintained, so that no change in ion va-
14
Corundum
(1012)
[0221] Fig 4.
Model of the comndum (1012) surface, including a [0221] step to another (1012) terrace, and an 0-vacancy point defect.
lence state is required. Subsequent AFM measurements on cleaved Cr203 (lOT 2) showed that the predominant step direction was indeed [0221] [27]. The most common type of nonstoichiometry observed on metal oxide surfaces is the formation of O-vacancy point defects associated with surface reduction [1]. The electronic properties of the metal cations in a metal oxide play an important role in the geometric structure of reduced surfaces. Non-transitionmetal oxides, whose cations have only one stable valence state, have a limited amount of freedom in the defect structures that can be formed. Good examples of that are IMgO and AI2O3. In either oxide, the removal of a single O ion would require that two electrons occupy the defect site, as discussed for bulk defects above (unless oppositely charged defects are present nearby to maintain local charge neutrality). Since the lowest-lying empty state on the cation is several eV above the energy of the O 2p orbitals, they are not accessible. This problem can be circumvented, however, if both anion and cation vacancies are created in
15
Stoichiometric proportion. This appears to occur for MgO; when MgO (100) surfaces are bombarded by inert gas ion ions (which reduces virtually all transition-metal oxide surfaces by preferential removal of O atoms), Auger and XPS spectra indicate very httle deviation of the surface stoichiometry from that of the bulk [1]. Also, no significant changes are observed in the surface electronic states in this process [28]. The complexity of surface defect structures on MgO will be treated thoroughly in Chap. 3. For AI2O3, slightly reduced surfaces can be produced that exhibit complex reconstructions; one of these yields (Vsi x V3I) R ± tan"\V3/ll) LEED patterns, for example (see Chap. 6). Such reconstructions are believed to be ordered structures of point defects [1]; however, the surface stoichiometry is again close to that of the bulk. (The electronic properties of these surfaces have not been studied in detail.) The situation is very different for transition-metal oxides, where it is energetically easy to change the population of surface cation d-orbitals. When the surface of a transition-metal oxide is reduced, the electrons associated with the O vacancies can often populate normally empty d-orbitals on adjacent cations; an example of this is given in the discussion of surface electronic structure in § 4 below. A system whose reduced surface structures have been studied in some detail consists of the oxides of iron. Single crystals of both a-Fe203 (hematite) and Fe304 (magnetite) have been reduced and the resulting surface structural changes monitored by LEED or SPM. The structure of the surface layer, which is generally only 1 - 2 monolayers thick, depends on the preparation conditions (i.e., temperature, oxygen partial pressure, time). On a-Fe203 (0001) surfaces, overlayers of Fe304 (111) and Fei.^^O (111) have been observed [29], On Fe304 (111) surfaces, an Fci.^O (111) overlayer structure has also been identified [30]. Unlike the case of non-transition-metal oxides, where the surface cannot really change composition appreciably due to charge balance constraints, thin layers of entirely different metal oxides can be formed on transition-metal oxides. Thus transition-metal oxides offer a much richer array of surface defect properties than do non-transition-metal oxides. Various types of defects on oxide surfaces will be treated in depth in many chapters in this book. An interesting, but troublesome, surface structural effect occurs on one of the important Cu-oxide high-Tc superconductors, YBa2Cu307.;,. The oxygen atoms are quite mobile in that crystal structure, and bulk samples can be easily oxidized or reduced by annealing in an appropriate ambient. However, when single crystals are cleaved in vacuum, some of the O atoms evaporate from the surface, leaving the surface reduced relative to the bulk. In order to study the properties of stoichiometric YBa2Cu307-:v surfaces in vacuum, it was found necessary to cleave the samples at temperatures near 20 K in order to retain the bulk oxygen stoichiometry at the surface [31].
16
4. ELECTRONS ON OXIDE SURFACES Simply from symmetry considerations, the electronic structure of any surface, where the atoms or ions are necessarily coordinatively unsaturated, should be different from that of the bulk. The magnitude and type of the differences between surface and bulk electronic structure depend on the particular oxide, of course; this is addressed in Chaps. 2, 4 and 14 in this volume. However, a few general observations can be made. Since most "surface-sensitive" techniques sample at least a few atomic planes into the sample, it is difficult to experimentally separate the electronic structure of the outermost plane of atoms from that of the planes below. Theoretical calculations are able to clearly separate surface from bulk electronic structure, of course; it is common to calculate a separate electronic density-ofstates for each plane in the crystal structure ("layer density-of-states"). Significant changes from the bulk electronic structure are sometimes found for the surface planes in calculations. However, it is difficult to confirm those results experimentally [1]. In some oxides, the bandgap at the surface has been observed to narrow compared to that of the bulk. The measured core-level binding energies of partially coordinated surface atoms are often shifted, by as much as an eV, from their bulk values [32]; these are referred to as "surface core-level shifts". However, the experimental separation of surface from bulk electronic structure is at present far from satisfactory. The one case in which surface electronic structure can be clearly observed experimentally is for non-stoichiometric surface defects. As in the case of the bulk structure, the surface stoichiometry of an ionic compound plays a crucial role in its surface electronic properties because of the necessity of maintaining local charge neutrality. As a specific example, consider the (110) surface of rutile Ti02; this surface is shown in Fig. 5, and is also discussed in Chaps. 10, 11 and 14. As discussed above, half of the cations on the stoichiometric surface retain their bulk six-fold ligand coordination, while the other half are five-fold coordinated. Rows of "bridging" O ions lie above the six-fold cations. Two Ovacancy point defects are shown schematically in Fig. 5, one a missing bridging O ion, and the other an in-plane O ion. The formal valence of all of the Ti ions on the stoichiometric surface is 4+, and the O ions are taken to be 2-. The stoichiometric surface is charge neutral and has no net dipole moment (to see this, one must consider the plane of five- and six-fold Ti ions, as well as the O ions in that plane and the bridging O ions on either side of it). When one bridging O ion is removed, creating an 0-vacancy point defect, there must be (formally) two electrons remaining at the defect site in order to maintain charge neutrality. They cannot occupy orbitals on neighboring O^" ions because of their closed-shell configuration; the first available empty O orbital would be 3s, which is several eV higher in energy. However, the 3d orbitals on
17
Rutile
[110]
Fig. 5. Model of the rutile (110) surface. Two types of 0-vacancy point defect are shown.
the two (now five-fold) Ti ions immediately adjacent to the vacancy are empty, and it does not cost much energy to place one (or more) of the electrons in them. Thus the schematic picture of such a point defect consists of two Ti^^ ions, each having a 3d^ electron configuration, adjacent to the O vacancy [1]. Needless to say, this population of some of the Ti 3d orbitals gives a very different electronic structure than that of the stoichiometric surface (or the bulk crystal). Stoichiometric Ti02 is an insulator with a charge-transfer bandgap of about 3.1 eV at room temperature [9]. Figure 6 presents a series of angle-integrated photoemission spectra, taken with the photon energies indicated, for an ntype bulk reduced Ti02 (HO) sample whose surface is very nearly stoichiometric [33]; its Fermi level, Ep, is pinned at the bottom of the Ti 3d band. (The location of EF was determined from spectra taken from a gold foil in electrical contact with the Ti02 sample.) There is no emission from the bulk bandgap region. The emission from about 3 to 9 eV is from the O 2p orbitals (i.e., the valence band); emission below that band is inelastic background. The creation of surface O-vacancy defects drastically changes the electronic structure of the surface. Figure 7 presents similar spectra for the sample after its surface has been
18
N(E)
10 5 lUnding Energy (cV)
0
Fig. 6. Series of angle-integrated UPS spectra form stoichiometric Ti02 (110) taken for photon energies between 30 and 76 eV.
slightly reduced by bombardment with inert Ar* ions. In addition to the O 2p emission, electrons are now emitted from a band of states lying slightly below EF in the bulk bandgap. Those are the electrons that are localized at the Ovacancy defects. By following the intensity of that "Ti 3d" emission as a function of photon energy across the Ti 3p -> 3d optical absorption threshold (i.e.,
19
N(E)
15
10
5
Binding Energy (eV) Fig. 7. Series of angle-integrated UPS spectra form reduced Ti02 (110) taken for photon energies between 30 and 76 eV.
by using "resonant photoemission" [1]), the states can be shown to be predominantly of Ti 3d character, as predicted above. This type of O-vacancy defect, with accompanying changes in the population of empty orbitals on neighboring metal cations, is typical of virtually all transition-metal oxides and will be discussed in depth for specific oxides in the following chapters. It is the most important type of defect on metal oxide surfaces.
20
As the photon energy is varied in both Figs. 6 and 7, the emission from the 0 2p band is seen to change amplitude and shape. Since there are no oxygen intra-atomic optical absorption transitions in the photon energy range used here, that emission should not change (i.e., "resonate") if the states were of purely oxygen character. However, in any solid there must be overlap between the electronic orbitals of neighboring atoms or ions; that is chemical bonding. In Ti02, both the O 2p and the Ti 3d orbitals overlap in the region of space between the ion cores; this is referred to as hybridization, or covalent bonding. Thus the "valence band" contains electrons with primarily O 2p character, but also some Ti 3d (and, it turns out, Ti 4s,p) character [33]; the resonant effects seen in this photon range are due to the presence of the Ti 3d wavefunctions in that band. The amount of Ti 3d hybridization in the O 2p band is believed to be of the order of 10 - 20 %. Similarly there is some admixture of O 2p orbitals in the normally empty Ti 3d band (although that hybridization is not seen in a filledstate experiment such as photoemission). This way of looking at hybridization shows why the idea of a formal valence on cations and anions cannot be an exact one. In discussing the electronic structure of 0-vacancy point defects above, we said that the O 2p orbitals were completely occupied. That is not quite true, because hybridization puts some of the O 2p wavefunction into empty states above Ep. Thus there are not quite eight 2s,p electrons on each O ion, and thus their valence state is slightly less negative than 2-. Similarly there is some population (in the valence band) of Ti 3d orbitals, so the Ti ions actually have a valence slightly less positive than 4+. That is one way of saying that the bonding in Ti02 is not purely ionic, but that it also contains some covalent character. There has been recent interest in a different type of defect on transitionmetal oxide surfaces: photoexcited reduced cations. The interest in this comes from atmospheric chemistry and will be discussed further in § 6 below. The idea is to use photons whose energy is greater than the bandgap of the oxide to create charge-transfer excitations of the type shown in Fig. 1, as mentioned in § 1 above. The systems that that have been investigated to date on single-crystal samples include Ti02 [34] and a-Fe203 [14]. The latter was chosen because it is a significant component of tropospheric aerosol particles and thus has the potential to do solar photochemistry. In a-Fe203, high signal-to-noise UPS spectra are able to resolve the changes in valence-band emission when the sample is illuminated with greater-than-bandgap light, compared to the sample in the dark. In the ground state of a-Fe203, all cations are Fe^^. Increased emission at the upper edge of the valence band is seen upon illumination, at the same energy that Fe^^ states are found on a-Fe203 surfaces that are reduced by removal of surface O atoms. Population of the states is completely reversible. These are transient, short-lived excitations, but it appears that enough of them can be ere-
21
ated to observe photochemical effects in the adsorption of SO2 on the surface; this is described in § 6 below. 5. OXIDE-GAS, -LIQUID AND -SOLID INTERFACES Several of the chapters in this book are concerned with the electronic and geometric properties of clean metal oxide surfaces; i.e., the oxide-vacuum interface. Theoretical and experimental determination of clean surface properties is crucial to understanding the physics of metal oxides. This physics can be rather complicated, due to the ionic nature of the system and the strong effect of electron correlation. Some of this work is at the cutting edge of current solid-state physics. Many other chapters are concerned with how metal oxide surfaces react chemically with atoms and molecules. Chaps. 6 - 9 deal with oxide films grown epitaxially on bulk metals or other oxides. The interfacial properties of oxides are where much of their technological, commercial and environmental utility lies, and these fields will be expanding tremendously over the next few years. It is important to consider the connection between the two types of studies. One often refers to the "pressure gap" that separates vacuum studies of chemisorption and catalysis from commercial catalytic reactions, which generally run above —often well above — atmospheric pressure. There is simply no way to properly simulate high pressure conditions in a surface analysis system. Reactions can be run in an attached reaction chamber, which is then pumped out and the sample transferred, under vacuum, into an analysis system equipped for electron, ion and photon spectroscopies. However, except for some optical and x-ray methods that can be performed in situ, the surface analytical tools are not measuring the system under reaction conditions. This gap is well recognized, and both the low- and high-pressure communities keep it in mind when comparing their results. A similar gap exists between clean surface experiments and any adsorption or interface study. For example, an oxide surface may exhibit relaxation or reconstruction due to the reduction in symmetry and ligand coordination at the surface. In the case of ZnO (0001), for example, the small Zn ions relax into the close-packed plane of O ions in order to be more completely surrounded by oxygen, and to reduce the surface dipole moment. But if an adsorbate — almost any adsorbate — is placed on such a surface, it will become an additional ligand for some of the surface ions. Adsorbates therefore, although they are not exactly the atoms or ions that would be present in the bulk structure, do increase the ligand coordination of surface ions. In general, this tends to reduce the amount of relaxation or reconstruction that was present on the clean surface. Thus, one cannot assume that the surface structure that was determined for the oxide-vacuum interface remains when a gas, liquid or other solid is placed on the surface.
22
In general, the surface will tend more toward an ideal termination of the bulk lattice. Similar effects can also occur in surface electronic structure when a moiety is weakly physisorbed onto the surface. The surface core-level shifts measured at the vacuum interface are reduced when atoms or molecules are physisorbed onto the surface. Changes may also occur in the valence electronic structure upon physisorption, such as the disappearance of intrinsic surface states on metals and semiconductors. In chemisorption, adsorbate-substrate interactions are sufficiently strong that one can no longer assume that the surface structure — electronic or geometric — that existed in vacuum remains unchanged. Relaxation and reconstruction that existed in vacuum will be altered not only because of a change in ligand coordination, but because of changes in electronic structure, whether substrate-adsorbate orbital overlap in an acid/base interaction or actual charge transfer in a redox reaction. Vacuum surface properties certainly serve as a guide for what will happen at an interface. However, one should not consider a chemisorbed adsorbate as a perturbation on the clean surface; the adsorbate-substrate complex should be considered as a new system. Many chapters in this book consider chemisorption and catalysis on metal oxide surfaces. Some of the basic ideas of adsorption on oxides will be discussed in § 6 below. An obvious example of the strong effect that chemisorbed molecules can have on surface properties can be seen in Figs. 6 and 7 above for 0-vacancy point defects on Ti02. In Fig. 7, the band of electron emission just below Ep originates from electrons trapped in d-orbitals on Ti cations adjacent to the defect. When the reduced surface is exposed to O2 at room temperature, the defect bandgap emission band almost completely vanishes; the photoemission spectra look very similar to those from the stoichiometric surface in Fig. 6. In this simple case, it is believed that O2 dissociates at 0-vacancy defect sites, removing electrons from the adjacent cations in order to become O^' ions. While there have been thousands of studies of the interaction of gaseous atoms and molecules with oxide surfaces, much less work has been done on the oxide-liquid interface. As mentioned above, this is largely because of the limited number and type of experimental techniques that can penetrate the liquid and access the interface. The greatest interest in oxide-liquid interfaces is from the geochemistry community, where oxide-water interfaces are crucially important in both terrestrial and atmospheric geochemistry. The current state of metal oxide-aqueous solution interfaces was recently reviewed in the report of the Department of Energy, Chemical Sciences Division, Workshop on Metal Oxide Surfaces and Their Interactions with Aqueous Solutions and Microbial Organisms [3]. That report is much broader than the scope of this book, and it addresses a wide range of problems and possibilities in that field. Suffice it to
23
say that oxide-water interfaces in nature are extremely complex, and it will take many years of research before we fully understand them. One of the most active current research areas in metal oxide surface science is oxide-solid interfaces. Work to date falls into three broad categories. The earliest work was the growth of metal films on single-crystal oxide substrates. One of the motivations there was to produce model systems to study the basic properties of oxide-supported metal catalysts. Quite a few groups have been involved in that work, and several papers have reviewed parts of the field [35-37]. An example of this approach is presented in Chap. 9 here. The second area of interest is growth of epitaxial metal oxide films on single crystals of metals. That work was initiated largely in order to expand the types of surface analytical techniques that could be used to study metal oxide surface properties. In the discussion of photoemission spectra on Ti02 in § 4 above, it was mentioned that the normally insulating Ti02 samples were slightly bulk reduced to make them A?-type; this is so that their surfaces would not charge up electrically during photoemission measurements, or other measurements that involved the use of charged particles. While that technique works well for many transition-metal oxides, it does not work for most non-transition-metal oxides. For example, two extremely important oxides are MgO and AI2O3; the former has a bandgap of 7.8 eV, the latter 9.5 eV. They are thus very good insulators, and they cannot be doped sufficiently by reduction or the addition of impurities to make them reasonably good electrical conductors. An alternative approach is to grow thin single-crystal films of those oxides on conducting (usually metal) substrates. If the oxide film thickness is kept below 2 - 3 nm, quantum mechanical tunneling through the film will remove any excess charge that might tend to build up on the surface. Not all oxides or crystal faces can be grown in this way, but there has been a great deal of success for some metal oxides. The quality of films obtainable is not always as good as that for single crystals, due to lattice mismatch and strain between the film and the substrate, the consequent formation of domains in the oxide film, etc. However, interesting surface structures, including polar surfaces, that would not be stable on the bulk oxide can be produced in thin films. Some publications have reviewed that work as well [38,39]. Chapters 6 - 9 here treat various aspects of the growth of metal oxides on metal substrates. Chapters 8 and 9 consider adsorption and reaction on some of those films. The last area is oxide-oxide interfaces. While metal oxide films have been deposited commercially on oxide substrates for many years, and many such systems are of technological importance, the surface science of oxide-oxide interfaces is in its infancy [39,40]. Some oxide-on-oxide growth is discussed in Chaps. 6, 7 and 9. However, its future importance warrants some additional comments on the field. The constraints on the structure of oxide-oxide interfaces are more stringent than for oxide-metal interfaces since both components
24
of the interface are (at least partially) ionic compounds. Not only must there be geometric alignment between substrate and overlayer atoms at the interface, but ligand charges must be matched appropriately. As an example, consider the simple case of the formation of an edge dislocation in a rocksalt compound such as MgO. Such dislocations have a very high energy of formation and rarely occur. They can be pictured by removing one-half of an atomic plane of ions from the bulk structure and then healing the cut by moving the adjacent planes together to fill the gap [41]. Consider removing part of a (100) plane from MgO. When the planes adjacent to the cut are brought together, the entire half-plane would consist of Mg^^ (or O^') ions on one plane directly opposite Mg^^ (or O^) ions on the other plane. There would thus be a large net Coulomb repulsion between the half-planes, which would tend to blow the crystal apart. While that is an extreme example, similar effects can occur at the interface between two different ionic materials, whether they have different geometric structures or simply different lattice constants. A stable interface can only develop if both geometrical and electronic constraints are met. The electronic properties of oxide-oxide interfaces affect a wide range of processes and physical phenomena; the basic physics and chemistry of oxideoxide interface formation are also extremely interesting, as they push the limits of solid-state theory. Several recent technological developments illustrate the importance of oxide-oxide interfaces. Field-effect transistors (FETs) have been successfully fabricated by using metal oxides for both the active conduction medium and the ferroelectric gate [4,42-45]. Such ferroelectric field-effect transistors are useful in memory applications since they are non-volatile; memory retention of the order of hours has been demonstrated in Lai.;cCa;cMn03 LaAlOs devices, and for more than ten days at room temperature in (Pb,La)(Zr,Ti)03 - Lai.;,Sr^Cu04 devices. In such devices, the electronic properties of the oxide-oxide interface are crucial [46], since "any imperfections at the interface, such as the formation of undesirable phases or electronic trapping states, will seriously degrade the performance of the device" [4]. A different type of interfacial effect occurs in superconductor-normal-superconductor (SNS) junctions. In SNS junctions fabricated with high-Tc cuprate superconductors and normal oxide barriers (e.g., CaRuOs, La;^Sri.j,Co03, etc.), the junction resistance is found to be far higher than expected theoretically, and its dependence on device fabrication parameters indicates that it is primarily of interfacial origin [47]. Oxide-oxide interfaces are also extremely important in optical guided wave device structures [48]. Recent work on the growth of perovskites on MgO showed that the quality of the interface depended strongly on which perovskite plane [BaO or Ti02 for the (100) plane of BaTi03] was grown first on the MgO (100) substrate [49]. The reason for the drastic difference in growth mode arises from interface electrostatics, i.e., ion-ion near-neighbor interactions. Very recently, novel ferromagnetic spin order has been observed in
25
LaFeOs - LaCrOs superlattices; the individual oxide layers were between 2.3 and 16.1 A thick [50]. With layers that thin, the quality and properties of the oxide-oxide interface are crucial. Other applications were mentioned in §1 above. Although there have only been a few surface science studies of the atomicscale properties of oxide-oxide interfaces to date, some of which are discussed in Chap. 7, extremely high quality interfaces have been fabricated for many years. A good example is superlattices of Fe304 and NiO, which have been grown with individual oxide layer thickness a small as one Fe304 unit cell and two NiO unit cells [51]. Figure 7.18 in Ref. 1 shows the high quality of the resulting superlattices from their low-angle x-ray diffraction spectra. 6. ADSORPTION ON METAL OXIDES Since the surfaces of metal oxides contain both negatively charged anions and positively charged cations — often with more than one cation valence state present —many avenues of adsorption are available. The various chemisorption and physisorption possibilities on metal oxides are discussed at length in Ref. 1; they will only be briefly summarized here. Using the scheme of Ref. 1, a major distinction can be made between: Non-dissociative or molecular adsorption and Dissociative adsorption. While this distinction is generally a straightforward one, it says nothing about the chemical nature of the bonding or about charge transfer between the surface and the adsorbate. There are four main types of interaction that can occur: weak electrostatic or dispersion acid/base or donor/acceptor oxidation/reduction with electron transfer and oxidation/reduction with oxygen transfer.
26
The first of these implies physisorption only. It dominates for many atoms and molecules at low temperatures, but it is of little interest catalytically. Therefore it has not been as thoroughly studied as the other types. In acid/base, or donor/acceptor, reactions, bonding results form the overlap of filled orbitals on the "donor" and empty orbitals on the "acceptor". Surface cations are generally Lewis acids and act as electron acceptors, while surface O ions are Lewis bases and can donate electrons to acceptor adsorbates. In lower oxides of the transition metals (i.e., in which the cations are in an oxidation state lower than their maximal valency), cations may also be able to donate electrons in an acid/base reaction. Although one talks of donating and accepting electrons in acid/base reactions, the electrons are in no sense free, and there is no actual electron transfer involved. This type of bonding can be either molecular or dissociative. Actual electron transfer does occur in oxidation/reduction, or "redox", reactions. In this type of reaction, there is a change in the oxidation state of the adsorbate. A simple example is the chemisorption of an alkali atom, in which it becomes a 1+ ion, transferring its outer electron to empty electron orbitals of the substrate. It is the large electric dipole moment created by this charge transfer process that lowers the work function of surfaces on which alkali atoms are adsorbed (e.g., "cesiation") by up to several eV. This type of bonding is generally strong, and it can also be either molecular or dissociative. A unique property of metal oxides is their ability to also exchange O ions with adsorbates; this property makes them very useful as partial oxidation catalysts. Such ion transport is a type of redox reaction in that charge (here in the form of an O^" ion) is really transferred from the substrate to the adsorbate. A simple example of a redox reaction with O ion transfer is the oxidation of CO: CO + O'lattice ^ CO2 + 2 e". The oxide surface becomes reduced in this process; if this were a step in a catalytic reaction, some other O-containing species would have to donate an O atom back to the substrate in order for the reaction to continue. Like the other types of reaction, O-transfer adsorption can be either molecular or dissociative. An interesting example of this type of redox reaction — and the role that surface defects can play — is the interaction of CO with the surface of V2O3. This has been recently studied on both pure and Cr-doped V2O3 (0001) singlecrystal surfaces by ultraviolet photoemission, among other techniques [52]. V2O3 doped with 1.5 at.% Cr is an insulator at room temperature, with the V 3d electrons (each cation has a 3(f configuration) forming a band that lies just below Ep. This is shown as the heavy curve in the UPS spectra in Fig. 8(a). The other curves in Fig. 8(a) show how the spectrum changes when the stoichiomet-
27
Clean surface 100 L CO 103 L CO
N(E)
AN(E)
Binding Energy [eV]
Fig. 8. (a) UPS spectra of stoichiometric, insulating Cr(l .5%): V2O3 (0001) at 273K, both clean and exposed to CO (hv = 40.8 eV). (b) Differences created by subtracting the clean surface spectrum from the CO-exposed spectra.
ric, low-defect-density surface is exposed to the amounts of CO indicated at room temperature. Up to 10^ L, there is virtually no effect on the spectra; this can be seen most clearly in the difference spectra (where the clean-surface spectrum is subtracted from each of the other spectra) in Fig. 8(b). However, above a certain threshold, the UPS spectrum changes dramatically. Emission from the O 2p band decreases, while there is increased emission from the V 3d band and at several eV below the O 2p band; the emission near 11 eV is at the location expected for the CO 4a molecular orbital. XPS spectra of the V 2p core levels
28
show that some of the V^^ cations have been reduced to lower valence states in the process; carbon is also present on the surface. That this process involves irreversible removal of O ions from the V2O3 surface can be seen by heating the surface to remove the adsorbed moieties. This is shown in Fig. 9, where the surface was first exposed to 10"^ L CO at room temperature, and then heated sequentially to the temperatures shown; a clean-surface spectrum is included for reference. After heating to 473 K, no carbon remained on the surface, and heating to higher temperatures produced no additional changes in the UPS spectra. However, instead of simply desorbing the adsorbed moiety and returning to something close to the clean surface, more complex changes occur in the spectra. While emission in the CO 4a and O 2p band regions does move in the direction of the clean surface spectrum, the intensity of emission form the V 3d band increases, indicating further reduction of the surface. Coupled with the absence of carbon above 473 K, this suggests that the CO desorbed as CO2, taking surface lattice O ions with it. The threshold behavior seen in Fig. 8 suggests that CO does not react readily with the nearly defect-free stoichiometric surface, but that some O ions must be removed, forming 0-vacancy point defects, before a significant reaction can
10-* L CO at 273 K 323 K 373 K 423 K 473 K Clean surface
N(E)
14
12
10
8
6
4
2
Binding Energy [eV]
Fig. 9. UPS spectra of CO-exposed Cr(1.5%):V203 (0001) heated to 473 K; the clean surface is included for comparison, (hv = 40.8 eV)
29
take place. This can be confirmed by exposing a CO-reduced and then annealed surface (i.e., the final state in Fig. 9) to additional CO; this is shown in Fig. 10. While the spectral changes are smaller here, the difference spectra in Fig. 10(b) show that CO interacts for all exposures used; there is no reaction threshold as there was for the stoichiometric surface. When the surface in Fig. 10, after exposure to 10^^ L CO, is heated to remove the adsorbed species, the changes are similar to those in Fig. 9, with the intensity of the V 3d band increasing as CO2 is desorbed from the surface, removing still more lattice oxygen. When the above cycle is repeated several times, the reduction of the surface continues to increase, as shown in Fig. 11. The dashed spec-
CO-reduced surface 100 L CO lO^LCO
N(E)
0-
•«••••««•«
^'•""^vj^:.
.,j^ir^
' • ^ i . ^ '
AN(E)
(b) Differences 1—^ 14
12
10
1—'—r 4 2
Binding Energy [eV] Fig. 10. (a) UPS spectra of slightly reduced, insulating Cr(1.5%):V203 (0001) exposed to CO at 273 K (hv = 40.8 eV). Prior to the CO exposures shown in this figure, the (stoichiometric) surface was reduced by one cycle of CO exposure and desorption. (b) Difference created by subtracting the first spectrum from successive CO-exposed spectra.
30
• Clean surface • One CO/desorb cycle "Two cycles -Three cycles - Four cycles / ^^
N(E)
10
8
6
4
2
Binding Energy [eV] Fig. 11. UPS spectra showing the effects of repeated cycles of CO exposure (10"* L at 273 K) followed by desorption at 473 K (hv = 40.8 eV). The first spectrum is of clean, stoichiometric Cr:V203; successive spectra were taken following desorption of the CO dose.
tra in Fig. 11 are taken sequentially after each cycle of exposure to 10"^ L CO followed by desorption to 473 K. The four cycles shown have reduced the sur face sufficiently that it should be thought of as a surface layer of a lower oxide of vanadium rather than just reduction of V2O3. (Similar experiments on pure V2O3 showed that the effect was not due to the presence of substitutional Cr.) There has been recent interest in a somewhat different aspect of adsorption and reaction on metal oxides: photocatalysis. The interest stems partially from that role that some transition-metal oxides can play in photochemical reactions in the atmosphere. Atmospheric aerosol particles can act as substrates to catalyze heterogeneous photochemical reactions in the troposphere. Most tropospheric aerosols are silicates, aluminosilicates and salts whose bandgaps are larger than the cutoff of solar radiation in the troposphere (about 4.3 eV); they are thus unable to participate directly in photoexcited reactions. However, transition-metal oxides that have much smaller bandgaps also occur as aerosols — the most prevalent ones are the oxides of iron and manganese — and these materials may thus undergo charge-transfer excitations (discussed above) in the pres-
31
ence of sunlight through electron-hole pair creation, which enable photochemical reactions for which the material would otherwise be inert. Some of the earliest work on such heterogeneous photocatalysis using transition-metal oxides was performed by Yates' group using Ti02; Ref. 34 is an excellent review of the photochemistry of that material. Chapters 10 and 12 in this volume also discusses photochemistry. Recent experiments have focused on the potential role that hematite, aFe203, might play in the oxidation of SO2 in the formation of acid rain [14,53]. Single-crystal hematite was used as a model for atmospheric aerosol particles, and solar radiation was simulated by use of a focused 200 W Hg(Xe) arc lamp (the integrated intensity reaching the sample for photon energies greater than the charge-transfer bandgap of hematite, 2.2 eV, was roughly 70 times the solar flux in that wavelength range). Figure 12(a) shows photoemission spectra for stoichiometric a-Fe2O3(0001) at 300 K, clean and following exposure to 10^ L SO2 while being irradiated by the Hg(Xe) lamp. The dominant changes produced after SO2 exposure are increases in intensity at 4.5 and 10 eV. These can be seen more clearly by subtracting the clean-surface spectrum from the spectra after SO2 exposure; this is done in Fig. 12(b) both for exposure during UV irradiation [i.e., the data in Fig. 12(a)] and for exposure in the dark. The shape of the two differences is the same, indicating that the same moiety adsorbs both in the dark and under illumination; that moiety is believed to be either SOs^' or S04^". However, the amount of SO2 adsorbed is much greater under illumination. The increased S02-hematite interaction is believed to result from the formation of transient Fe^^ ions on the a-Fe203 surface by the greater-than-bandgap illumination (see § 4 above); these Fe^^ cations are much more reactive chemically than are the maximal valency Fe^^ cations that are present in a-Fe203 in the dark. The adsorption that does occur in the dark is believed to do so at a small density of Fe^^ sites associated with surface O-vacancy sites that are always present on the surface. The range of adsorption processes that can occur on metal oxide surfaces is very broad; these will be discussed in many of the other chapters in this book. Chapter 5 considers in detail the atomic positions of adsorbed moieties on several different oxide surfaces. The use of vibrational spectroscopies as a complement to electronic techniques is discussed in Chap. 13. Chapter 15 considers desorption from oxide surfaces induced by incident electrons or photons. 7. ENDNOTE All of the knowledge that has been gained about the properties of metal oxide surfaces and interfaces, particularly over the past 8 - 1 0 years, would fill many volumes this size. There are also several dozen research groups currently
32
a-Fe2O3(0001) exposed to SO2at300K
(a)
UPS spectra
N(E)
0-
- | — I — I — \ — I — I — I — I — I — I — I — I — I — I — r
16
14
12
10 8 6 4 Binding Energy [eV]
(b)
2
differences (xl2)
UV-irradiated during dose NoUV irradiation AN(E)
\r
1
16
\
14
I
I
12
I
I
I
i~
~i—I—\—I—I—\—I—r
6 4 10 8 Binding Energy [eV]
2
0
Fig. 12. (a) He ii UPS spectra for stoichiometric a-Fe203 (0001) at 300 K, clean and following exposure to 10^ L SO2 [while irradiated by the Hg(Xe) arc lamp]; (b) Differences between the S02-exposed spectrum and the clean-surface spectrum (x 12), displaying features arising from adsorbate molecular orbitals.
33
involved in expanding that knowledge base even further. However, the authors who have contributed to the current volume, and the range of topics covered here, comprise a representative cross-section of the field. This volume should thus serve both as a useful portal for researchers wanting to enter the field, and a resource for those of us already in it. ACKNOWLEDGMENTS Some of the research reported here was partially supported by NSF Grant CTS-9610140, and by The Petroleum Research Fund Grant 28797-ACS. REFERENCES [I] [2] [3]
[4] [5] [6] [7] [8] [9] [10] [II] [12] [13] [14] [15] [16] [17] [18]
V.E. Henrich and P. A. Cox, The Surface Science of Metal Oxides, Cambridge University Press, Cambridge, 1994. D.A. Bonnell, Prog. Surf. Sci., 57 (1998) 187. G.E. Brown, Jr., V.E. Henrich, W.H. Casey, D.L. Clark, C. Egleston, 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, Chem. Rev., 99 (1999) 77. S. Mathews, R. Ramesh, T. Venkatesan and J. Benedetto, Science, 276 (1997) 238. S. Araki, M. Sano, S. Li, Y. Tsuchiya, O. Redon, T. Sasaki, N. Ito, K. Terunuma, H. Morita and M. Matsuzaki, J. Appl. Phys., 89 (2000) 5377. CH. Ahn, J.-M. Triscone, N. Archibald, M. Decroux, R.H. Hammond, T.H. Geballe, 0. Fischer and M.R. Beasley, Science, 269 (1995) 373. C.H. Ahn, S. Gariglio, P. Paruch, T. Tybell, L. Antognazza and J.-M. Triscone, Science, 284(1999)1152. C. Noguera, Physics and Chemistry at Oxide Surfaces, Cambridge University Press, Cambridge, 1996. P.A. Cox, Transition Metal Oxides, Clarendon Press, Oxford, 1992. H.H. Kung, Transition Metal Oxides. Surface Chemistry and Catalysis, Elsevier, Amsterdam, 1989. H.-J. Freund, H. Kuhlenbeck and V. Staemmler, Rep. Prog. Phys., 59 (1996) 283. C. Kittel, Introduction to Solid State Physics, Fifth ed., John Wiley & Sons, New York, 1976. T.B. Reed, Free Energy of Formation of Binary Compounds, MIT Press, Cambridge, 1971. D.S. Toledano, E.R. Dufresne and V.E. Henrich, J. Vac. Sci. Technol., 16 (1998) 1050. H. Kuwamoto, J.M Honig and J. Appel, Phys. Rev. B, 22 (1980) 2626. H.R. Kokabi, M. Rapeaux, J.A. Aymami and G. Desgardin, Mater. Sci. Eng. B, 38 (1996) 80. D.P. Woodruff and T.A. Delchar, Modern Techniques of Surface Science, Cambridge University Press, Cambridge, 1986. GA. Sawatzky, in; J. Kanamori and A. Kotani (Eds.), Core Level Spectroscopy in Condensed Systems, Springer Verlag Series in Solid State Sciences, Vol. 81, SpringerVerlag, Berlin, 1988.
34
[19] R.J. Lad, in: W.N. Unertl (Ed.), Handbook of Surface Science, Vol. 1, Elsevier, Amsterdam, 1996, p. 185. [20] C.B. Duke, Chem. Rev., 96 (1996) 1237. [21] There is, however, some recent theoretical work on ways to deal with polar surfaces. See, for example, J.H. Harding, Surf. Sci., 422 (1999) 87; and, M.H. Finnis, phys. stat. sol. A, 166(1998)398. [22] S.A. Chambers and S.I. Yi, Surf. Sci., 439 (1999) L785. [23] R.R. Gay, M.H. Nodine, V.E. Henrich, H.J. Zeiger and E.I. Solomon, J. Amer. Chem. Soc, 102(1980)6752. [24] V.E. Henrich, Surf Sci., 57 (1976) 385. [25] V.E. Henrich, G. Dresselhaus and H.J. Zeiger, Phys. Rev. B, 22 (1980) 4764. [26] V.E. Henrich, personal communication. [27] R.J. Lad and M.D. Antonik, Ceramic Trans., 24 (1991) 359. [28] L.H. Tjeng, A.R. Vos and G A Sawatzky, Surf. Sci., 235 (1990) 269. [29] R.J. Lad and V.E. Henrich, Surf Sci., 193 (1988) 81. [30] N.G. Condon, P.M. Liebsle, T. Parker, A.R. Lennie, D.J. Vaughan and G. Thornton,, Phys. Rev. B, 55 (1997) 15885. [31] J.G. Tobin, C.G. Olson, C. Gu, J.Z. Liu, F.R. Sohal, M.J. Fluss, R.H. Howell, J.C. O'Brien, H.B. Radousky and P A . Sterne, Phys. Rev. B, 45 (1992) 5563. [32] PH. Citrin and G.K. Wertheim, Phys. Rev. B, 27 (1983) 3176. [33] Z. Zhang, S.-P. Jeng and V.E. Henrich, Phys. Rev. B, 43 (1991) 12004. [34] A.L. Linsebigler, G. Lu and J.T. Yates Jr., Chem. Rev., 95 (1995) 735. [35] M. Baumer and H.-J. Freund, Prog. Surf Sci., 61 (1999) 127. [36] T.E. Madey, U. Diebold and J.M. Pan, in: H.J. Freund and E. Umbach (Eds), Adsorption on Ordered Surfaces of Ionic Solids and Thin Films, Springer, Berlin, 1993, p. 147. [37] C.T. Campbell, Surf. Sci. Reports, 27 (1997) 1. [38] S.C. Street, C. Xu and D.W. Goodman, Annu. Rev. Phys. Chem., 48 (1997) 43. [39] S.A. Chambers, Surf Sci. Reports (in press). [40] R.J. Lad, Surf Rev. Lett., 2 (1995) 109. [41] A.L. Ruoff, Introduction to Materials Science, Prentice-Hall, Englewood Cliffs, NJ, 1972. [42] Y. Watanabe, Phys. Rev. B, 57 (1998) R5563. [43] S.A. Campbell, D.C. Gilmer, X.-C. Wang, M.-T. Hsieh, H.-S. Kim, W.L. Gladfelter and J. Yan, IEEE Trans. Electr. Dev., 44 (1997) 104. [44] Y. Watanabe, M. Tanamura and Y. Matsumoto, Jpn. J. Appl. Phys., 35 (1996) 1564. [45] Y. Watanabe, Appl. Phys. Lett., 66 (1995) 1770. [46] T.S. Kim, C.H. Kim and M.H. Oh, J. Appl. Phys., 76 (1994) 4316. [47] K. Char, L. Antognazza and T.H. Geballe, Appl. Phys. Lett., 63 (1993) 2420. [48] A.M. Glass, Mater. Res. Soc. Bull., 13 (1988) 16. [49] R. A. McKee, F.J. Walker, E D . Sprecht, G.E. Jellison, Jr., LA. Boatner and J.H. Harding, Phys. Rev. Lett., 72 (1994) 2741. [50] K. Ueda, H. Tabata and T. Kawai, Science, 280 (1998) 1064. [51] D M . Lind, S.D. Berry, G Chem, H. Mathias and L.R. Testardi, Phys. Rev. B, 45 (1992) 1838. [52] D.S. Toledano, V.E. Henrich and P. Metcalf, J. Vac. Sci. Technol. A, 18 (2000) 1906. [53] D.S. Toledano and V.E. Henrich, J. Phys. Chem. (in press).
Oxide Surfaces D.P. Woodruff, editor © 2001 Elsevier Science B. V. All rights reserved.
35
Chapter 2
Clean Oxide Surfaces: a theoretical review Claudine Noguera Laboratoire de Physique des Solides, UMR CNRS 8502, Universite Paris-Sud, 91405 Orsay, France 1. I N T R O D U C T I O N The last five years have witnessed a tremendous effort to better produce, characterise and study insulating oxide surfaces. Several reasons stem for the rapid development of this field. They are related to experimental considerations — a better control of the fabrication of surfaces, a more thorough use of advanced spectroscopic or structural tools — but also to a more accurate recognition of the technological importance of high quality oxide surfaces in catalysis, magnetic recording, as sensors or as constituents of artificial nano-materials. A simultaneous effort to simulate these surfaces has been performed. While most theoretical works quoted in the Henrich's and Noguera's books [1,2] five years ago, referred to semi-empirical methods, first principles approaches are now routinely used to calculate the ground state properties of surfaces. The delay, with respect to the simulation of semi-conductor surfaces, may be assigned to the larger computational resources necessary to quantum-mechanically treat oxygen and transition metal atoms. However, ab initio methods find their limitations as soon as defects break the symmetry of the surface or when large cell reconstructions take place. In addition, most of them are based on a variational principle which limits their use to the prediction of ground state properties. Accounting quantitatively for quasi-particle spectra, optical excitations or d -^ d excitations on oxide surfaces remains a challenge. The same is true as regards analytical theories describing the basic microscopic processes at work on these surfaces. These considerations will be examplified here on three aspects of oxide
36
surface physics, namely (i) the non-polar stoichiometric surfaces, (ii) the oxygen deficient non-polar surfaces and finally (iii) the polar surfaces. The specific properties of the first family are induced by the low coordination of the surface atoms and are shared by other low-dimensional oxide systems, such as non-supported clusters or ultra-thin films [3]. In the second family, the presence of oxygen vacancies is responsible for strong electron redistributions in the surface layers which can result in a higher surface reactivity and/or surface reconstructions. A similar situation is met in the third family due to the strong electrostatic polar instability [4]. After a presentation of the available numerical methods in Section 2 and a review of simulation works in Section 3, we will consider successively the three families of surfaces in Sections 4, 5 and 6, from the point of view of their generic properties and the theoretical arguments to understand them. Finally, we will conclude on open questions for future investigations. 2. NUMERICAL METHODS The first historical step in the simulation of insulating oxides relied on the idea that most of these materials are highly ionic. It has led to the development of classical models of cohesion in which the leading forces result from Coulomb interactions between ionic charges. These approaches usually give a quite reasonable description of atomic positions and vibrations, but cannot yield information on electronic degrees of freedom, which require quantum treatments. As a function of time, the latters have evolved from semi-empirical — tight-binding or semi-empirical Hartree-Fock (HF) — to first principles methods, in which there are no adjustable parameters. In parallel, quantum-chemistry all-electron approaches allow a treatment of electron correlations beyond the HF approximation, but only for small finite size systems. We will stress the strengths and limitations of these methods when applied to oxide surfaces, and we will also place special emphasis on some methods which can describe excited state properties, or ground state properties of highly correlated oxides. 2.1. Classical approaches In classical approaches, the lattice energy E, which is the diff'erence between the total energy of the system and that of its component ions at infinity, is expanded into a sum of pair, triplet, etc terms:
E = lEEij + l Z Eijk + ...
(1)
Keeping only the Eij terms yields the so-called pair-potential approximation. The pair interactions include Coulomb (l/r^j dependence), van der
37
Waals {1/rfj dependence) and short range terms. The latters result from PauU's principle, which impedes the overlap of closed electronic shells. It may be taken under a Lennard-Jones ( l/rjj variation), or a under a Born Mayer form {Bij exp{—rij/p)). The parameters entering Eij are fitted, so that the properties of the perfect lattice are reproduced. In the applications to surfaces, they are kept equal to their bulk values, in particular the ionic charges Qi. Many body effects are included in the triplet and higher order terms in Eq. (1). The triplet terms, for example, give energy variations with bond angles. In addition, low-coordinated atoms, which are submitted to non-vanishing electric fields, may be strongly polarized. The shell model [5] was designed to describe these effects, in an approximate way. Rather than being written under empirical forms, potentials can also be derived from the functional expression of the free ion energies as a function of the electron density [6]. Corrections may also be included to account for the instantaneous environment in which an ion finds itself, through modifications of the ionic radii, shapes and charge distributions [7,8]. A good review of the state-of-the-art interatomic potentials in solid state chemistry can be found in Ref. 9. In order to predict the ground state geometry, energy minimization with respect to the ionic positions can be performed by various techniques: static ones — steepest descent, conjugate gradient methods — or dynamical ones — damped dynamics, or molecular dynamics [10]. The latters allow an exploration of the configurational space, through the technique of simulated annealing. Classical simulations have long been performed to predict surface phenomena. Due to the increasing availability of first principles calculations for medium size systems, they are nowadays essentially restricted to large size systems, like zeolites, grain boundaries, complex interfaces, etc. 2.2. Semi-empirical quantumi methods At the lowest level of sophistication of quantum treatments, the tightbinding method and the semi-empirical HF method reduce the complexity of the interacting electron system to the diagonalization of an effective one-electron Hamiltonian matrix, whose elements contain empirical parameters. The electronic wave functions are expanded on a minimal basis set of atomic or Slater orbitals centered on the atoms and usually restricted to valence orbitals. The matrix elements are self-consistently determined or not, depending upon the method. In the tight-binding method, the elements of the Hamiltonian matrix are treated as adjustable parameters to be fitted to experimental or first-
38
principles calculation results. To estimate the total energy, an additional short-range repulsion term is added to the electronic contribution. The dependence of the parameters upon interatomic distances are fitted in order to account for the bulk equilibrium structure and elastic constants. It is assumed that the parameters are transferable from bulk to surfaces [11]. This method has been widely used in the field of semi-conductors, and has yielded a good qualitative description of their surface properties. However, in more ionic systems, it completely disregards the variations of the electrostatic potential with the local environment of the atoms (see below), which is the reason for some wrong predictions made in the past. The semi-empirical HF approaches can be viewed as self-consistent tightbinding methods, since they incorporate the self-consistent relationship between charges and electrostatic potentials in the expression of the Hamiltonian matrix elements. The empirical parameters either fix the spatial dependence of the basis set wave functions, or are incorporated directly into some Hamiltonian matrix elements [12-14]. In the application to oxide surfaces, these methods can account for the shifts of the atomic orbital effective levels, i.e. for the surface modification of electronegativity, for example (Section 4.4). This is a very important point, both for obtaining reliable eigen-energy spectra, and to properly describe the reactivity of certain types of atoms as a function of their environment. It can also account for the surface charge redistribution, and for its influence on the structural degrees of freedom, contrary to classical approaches. These methods can be combined with geometry optimization as well as with molecular dynamics algorithms, with forces obtained from the gradients of the total quantum energy [10]. This equally applies to all quantum methods, quoted in the following. 2.3. T h e ab initio H F method and methods beyond The HF method treats electron-electron interactions at a mean field level, with the Hartree and exchange interactions exactly written. The method can be implemented either in its spin restricted form (RHF), for closed shell systems, or in the unrestricted form (UHF) for open-shell or strongly correlated systems. In the first case, the one-electron orbitals are identical for electrons of both spin directions, while UHF can account for a non-uniform spin density. The one electron orbitals, which are determined in the course of the self-consistent resolution of the HF equations, are expanded on an over-complete basis set of optimized variational functions. The HF method has been implemented on periodic systems [15], including bulk and surface crystalline materials. It has proved very useful in the description of magnetic insulators, but it has also successfully been used for describing surface properties of a large number of simple oxide surfaces.
39
One of the very serious problems with the HF approach is the unscreened nature of the Coulomb and exchange interactions. The bare value of a typical intra-atomic Coulomb integral U is in the range 15-20 eV, while screening in solids weakens it by more than a factor of 3. A large overestimation of the HOMO-LUMO (HOMO= Highest Occupied Molecular Orbital; LUMO=Lowest Unoccupied Molecular Orbital) gap in insulators results. However, there have been quite accurate predictions of gap values or d ^ d excitation energies, based on HF total energy differences (Sections 4.4 and 4.5). Quantum chemical approaches have long tried to solve N-electron Hamiltonians of finite systems, beyond HF, at levels of increasing sophistication. The essential ingredient is the expansion of the ground state and excited state N-electron wave functions on a basis set of Slater determinants, built from one-electron HF eigen-functions. This so-called configuration interaction method (CI) includes explicitely electron correlation effects. Depending upon the basis set size and the number and nature of configurations kept, ground state and some types of excited state properties may be calculated [16]. By essence, these methods can only treat finite size systems containing a limited number of electrons, although some hybrid DFT-CI schemes (DFT== Density Functional Theory, see Section 2.4) try to deal with extended systems [17]. Bulks or surfaces are represented by a "quantum part" containing a small number of atoms, which is embedded in an array of point charges. Varying the cluster size gives hints on the validity of the representation. However, due to the heavy demand in computational resources, these methods, usually, can only treat few atoms. 2.4. The ab initio Density Functional Theory and methods beyond The DFT provides another route to include correlation effects. It gives a prescription for calculating the ground state energy of an assembly of atoms at fixed positions, as a function of the electron density n{r) [18]. Assuming that there exists a system of non-interacting electrons with the same n{r)j the ground state energy of the interacting system may be expressed as a function of the eigen-functions and eigen-values of the non-interacting system. The latters are solutions of one-electron self-consistent equations, named Kohn Sham equations, which include a Hartree potential, an exchange term, due to the Pauli's principle and a correlation term. No exact expression exists for the correlation energy functional. The Local Density Approximation (LDA) replaces the exchange and correlation potential Vxc, which is by essence non-local, by a local potential built from the properties of the homogeneous electron gas (HEG). At each point f of the inhomogeneous system, the exchange correlation energy is
40
equal to that of the HEG: £'^^^(n), with the uniform electron density n functionally replaced by the actual density n{r). In open-shell or magnetic systems, it is possible to build an exchange and correlation functional which depends on both the electron density and the spin density. This is the Local Spin Density Approximation (LSDA). Excellent reviews of DFTLDA or LSDA may be found in Ref. 19-21 One step beyond the LDA, the GGA (generalized gradient approximation) represents a semi-local approximation, in which not only the density but its gradient are locally taken into account [22-24]. The GGA corrects a large part of the systematic overestimation of the LDA cohesive energies. In particular, at surfaces, the GGA yields a much better description of adsorption phenomena than the LDA. In order to solve the Kohn Sham equations, an expansion of the oneelectron wave functions on a basis set is performed. Both localized basis sets and plane wave ones are currently used. Localized basis sets have the advantage of their small size. However they are attached to the atomic positions, which yields non-zero Pulay forces in geometry optimization and molecular dynamics. Plane waves, on the other hand, provide a uniform sampling of space, whatever the specific conformation of the system; they are independent of the atomic positions, but they require the use of pseudopotentials to mimick core electrons and a very large number of vectors is necessary in standard surface calculations. The density functional theory is a theory for the ground state. It cannot predict excitation properties. In particular, there is no theoretical justification to identify the Kohn-Sham eigen-values with quasi-particle energies, although this is currently done. In addition, DFT-LDA fails in localizing electrons in highly correlated systems. Several schemes have been proposed, in order to overcome these drawbacks. However, their computational cost is usually very high and, at present, they cannot be used routinely in surface physics. Despite this fact, we will now present them, because there is no doubt that an improved understanding of oxide surface properties will be gained thanks to them, in the coming years. 2.4-1' The Self-Interaction Correction (SIC) method In the mean-field approximation for electron-electron interactions, the Hartree U[n] and exchange Ex energies read: rrr ^ f i'^ A ,nir)n{r') U\n] = d^rd^r'^-^—^ ^^ J 2\r-r'\ Ex = - Y: /dVdV^-^^^-^^^^'^^^^^^^^^^^"!j^^^^^^^^^
(2)
41
Ex involves a sum over orbital (a, a') and spin (cr) quantum numbers, of a product of one-particle orbitals "ipao- with occupation numbers faa obeying Fermi statistics. U[n] includes a spurious Coulomb self-interaction, as can be realized if Eq. (2) is applied to a single electron system. This Coulomb self-interaction is exactly cancelled out by the a = a^ terms in Ex (selfexchange term). In the DFT-LSDA approach, the local approximation to the exchange functional achieves only partial cancellation of the self-Coulomb term. Only for orbitals which are delocalized over the whole system does this self-interaction vanish. In the general case, for finite systems and localized states in extended systems, it leads to systematic errors, which have been summarized in Ref. 25. This work proposes a method for SIC, in which the LSDA exchange-correlation energy functional £'^^^^[n>|^,nj, which depends upon the electron densities for t and I spins, is replaced by: Ell^ = E ^ f A ^ , „ j _ Y: U[n,,,] + El!^^[n,,,,
0]
(3)
a,cr
Modified Kohn-Sham equations result which contain an orbital-dependent local potential. The correction usually leads to a greater localization of the electronic wave functions. It was shown that SIC heals most of the LSDA drawbacks, including a large part of the gap problem in insulators and it corrects the bad shape of the exchange hole around electrons. In addition, in atoms, the negative of the SIC eigen-energies closely approximates the relaxed excitation energies for electron removal. 24.2. The GW method The calculation of a quasi-particle (QP) spectrum in a crystal requires the resolution of an equation of the type: - ^ + 14xt(r) + VH(r-) (4) in which E^j^ and (t>^%{'f^ are the quasi-particle energy and wave function in a band n at a given k point in the Brillouin zone. Kxt(r) i^ ^^^ ionic potential, V]i{r) is the Hartree potential and E is the self-energy operator, which is in general non local, non-Hermitian and energy-dependent. It contains the exchange and correlation effects and is state [nk) dependent. By comparison, in the Kohn Sham equations, E is replaced by T4c(r)^(^"" f^), and the resulting eigen-values fail in reproducing quasiparticle energies in solids, as has been stressed by many authors [26-28]. The estimation of E(r, r';£^^^) is a very difficult task. A possible approximation is the GW one, [29,30] in which a perturbation expansion for
42
the self-energy is constructed and stopped at the first order: E(f, P; Lj) = i /_"^J ^e+i^^'
G{r, ?• u + J) W{r, P; a;')
(5)
In Eq. (5), G is the one-particle Green's function, W is the screened Coulomb interaction and (5 = 0"^. The real part of the self-energy contains a screened exchange contribution, which requires an explicit calculation of the dielectric matrix of the system, and a Coulomb-hole term which takes into account the actual presence of the quasi-particle (excess electron or hole) in the system and its screening by the surrounding electrons. Although many improvements have been proposed for the calculation of the dielectric response function from first principles [31,32], still this stage is both computer time and memory very demanding when a large basis set is used for the description of the electronic structure. Efficient GW methods have thus been developed, in which a model dielectric function is used to mimick the screening properties of the system under study [28,33,34]. It should be noted that the resolution of Eq. (4) may be performed self-consistently or in a perturbative way with respect to E(r, r';£J^^) — 14c(^)- In the second case, HF as well as DFT-LDA eigen-states and eigenenergies may be used as starting points for the implementation of the GW approximation. 24.3. The LDA-hU method The LDA+U method assumes an approximate non-local and frequency independent form for T,{r^r^; ^nk)^ which makes it a HF-type theory with a screened exchange. The idea consists in describing delocalized s — p electrons by an LDA-type approach, and in adding a Hubbard term ^C/ E [ui the occupancy operator of the di orbital) to the d- or /-electron Hamiltonian [35-37]. This procedure stabilizes the occupied LDA bands, and destabilizes the unoccupied ones. This jump in energy between occupied and unoccupied states, which exists in an exact DFT theory, as well as the correlated jump of potential when the number of electrons goes through an integer value [38], are absent in the LDA, which thus misses an important contribution to the gap value. Taking into account exchange interactions (J) and the non-sphericity of electron-electron interactions (dependence of U and J upon the orbitals i), the orbital-dependent potential reads (C/eff = U — J/2):
43
Fig. 1: Domains of the projected atomic structure of the Q;-Al203-(\/3l x \/3l)R±9° reconstruction, where the unit cells as well as domain walls are drawn. The two constituting Al planes are shown separately, with evidence of one being much better ordered than the other. Numerical relaxation has shown that the ordered layer could be associated to the outermost layer and the more disordered one to the layer adjacent to the substrate (from Ref. 43).
Via{r)
= VLDA{r) +
E{UiJ-Ues)n^J-a 3
+ YliUij
- Jij - Ues)nja
+ C/effCo ~ ^*>) ~~ 7*^
(^)
The LDA+U theory may be regarded as an approximate GW method [37]. The screened Coulomb and exchange parameters U and J are usually estimated in a supercell approximation [39]. However, there is some arbitrariness in the choice of the localized orbitals when performing the partitioning of the Hamiltonian. A further step in the improvement of LDA+U consists in adding dynamical effects — frequency dependence in S(r, r';a;). This may be performed using a DMFT-type approach (DMFT= Dynamical Mean Field Theory) [40] as part of the so-called LDA++ approaches [41]. 2.5. General features for simulating surfaces Whatever the choice of the quantum mechanical method, two representations are currently used to simulate semi-infinite surfaces: slabs or clusters. Only the CI method is restricted to cluster geometries. In slab calculations, a finite number of layers mimicks the semi-infinite system, with a two-dimensional (2D) translational periodicity. A minimal thickness dmin is required, so that the layers in the slab centre display bulk characteristics. Practically speaking, dmin should be at least equal to twice the damping length of surface relaxation effects, which depend upon the surface orientation. In plane wave codes, the slab is periodically repeated
44
in the direction perpendicular to the surface, in order to make finite the required number of plane waves. The procedure generates spurious electric fields and energy contributions: in order to minimize them, one has to introduce thick enough vacuum regions between the repeated slabs and has to construct symmetric slabs with a zero total dipole moment. This question is particularly crucial for polar orientations. Cluster calculations seek to determine surface properties by simulating only the local active sites [42]. They are usually implemented when it is believed that local bonding prevails in the electronic structure, when the property under study is local by essence, or when one wishes to model excited states through CI methods. It has been realized that a correct modelling requires the effect of the electrostatic potential created by the ions outside the cluster: the clusters are thus embedded in an array of point charges and/or in effective core potentials to describe atoms at the fringe of the clusters. A good embedding is more and more necessary as the cluster size decreases. Treating non-stoichiometric surfaces requires special care in several respects. First, the long range elastic deformations (surface stress) induced by the defects are not easy to account for in a realistic way, neither in periodic calculations, nor in cluster models. When isolated vacancies are simulated in slab calculations, the size of the 2D unit cell, which is, most of the time, fixed by computational rather than physical constraints, corresponds to vacancy densities by far larger than realistic concentrations. At present, large size reconstructions, such as the (\/3T x \/31)R9^ on the a-Al2O3(0001) surface (Fig. 1) are still beyond the possibilities of quantum simulations. The presence of oxygen vacancies is usually associated to a strong redistribution of electrons, some of them being trapped at the vacancy site. When using localized basis sets to expand the electronic wave functions, it is mandatory either to place virtual orbitals at the location of the missing oxygen or to use strongly polarized orbitals on the neighbouring cations, in order to span the whole space for the electron density (Section 5.2). Finally, in order to account for strong electron redistributions, selfconsistent approaches which properly solve the Poisson's equation have to be used. The same is true for polar surfaces, whose stability relies on a well-defined electrostatic condition. 3. REVIEW OF LITTERATURE This section summarizes the present stage of our knowledge on clean oxide surfaces, obtained from numerical works. It is presented according to the oxide bulk crystal structures, with special emphasis on those oxides which have been more thoroughly studied.
45
3.1. Fluorite MO2 and anti-fluorite M2O structures The anti-fluorite structure consists in an fee oxygen lattice, in which every 0 - 0 pair is bridged by two metal ions, of charge 4-1. It is found in highly ionic oxides [44], like alkali-oxides Li20, Na20 and CS2O. The fluorite structure is similar, with a mere exchange of oxygens and cations, as in Zr02, Ce02, UO2, etc. The non-polar surfaces of lowest Miller indices are the (111) and (110) surfaces. The former is more dense than the latter, with a single broken bond per formula unit, to be compared to 2 on the (110). The (100) surface is polar. Various works have shown that the (111) surface has always a lower energy than the (110), whether Li20 [45-47], Ce02 [48,49], UO2 [50], or Zr02 [45] are concerned. Relaxation eflFects depend upon the oxide: on 1102(111), inward atomic displacements take place, while on the (110) surface, there are small vertical atomic displacements associated to large inplane displacements [50]; on tetragonal ZrO2(001), oxygens move inwards and cations outwards [51], while the (110) and (101) surfaces display some rumpling, with no preferred outward displacements of the surface oxygens [52]. The relaxation strength is an increasing function of the unrelaxed surface energy. The modifications of electronic structure at the Li20 surfaces include a shift of the oxygen Is core levels, and a narrowing of the oxygen-derived valence band (VB) width, especially on the (110) surface. No surface states are found in the bulk band gap [46,47]. In an effort to interpret STM images (STM= Scanning Tunneling Microscopy) of U 0 2 ( l l l ) , electronic charge density maps were obtained, using a DFT-LDA-i-U approach, and confirm the high degree of ionicity of the oxide [53]. Noticeable charge redistributions take place at the tetragonal (001) Zr02 surface [51]. 3.2. Rocksalt structure The rocksalt structure consists in two interpenetrating fee lattices of anions and cations, in which all atoms are in an octahedral environment. It is met in alkaline-earth oxides (MgO, CaO, SrO, BaO) and in some transition metal oxides like TiO, VO, MnO, FeO, CoO, NiO, etc, with cations in a +2 oxidation state. The non-polar surfaces of lowest Miller indices are the (100) and (110) surfaces: they have neutral layers, with a^ many cations as oxygen ions, and their outermost atoms are 5- and 4-fold coordinated, respectively. Actually, planar surfaces can only be produced along the (100) orientation. The polar direction of lowest indices is (111): it has an hexagonal 2D unit cell, three-fold coordinated surface atoms and equidistant layers of either metal or oxygen composition. As regards the (100) surface, all theoretical works predict very small atomic displacements, on MgO [45,54-69], as well as on CaO, SrO, BaO
46 0.10 0.08 0.06 0.04 0.(»
1«llay«rofMnpty| •bovt 0 sltet ml
^iiiwHwibw [t
4.0
a«^
I' turfac* layer
"^2.0
1 0.0 I 4.0
1 St •ubturtac* lay«r
» 2.0 UJ H 0.0
"I
•"I
»W|W"I
|
»»l»i
a 2.0
CM
O 0.0
N I ¥*<»' 3rd MibturtM* layar
••I
-M
-4.0 ^ . 0
0.0
2.0
ENERGY (eV)
^'•'U.O
-*.0 -AJIt -2.0 0.0
»^m
2.0
ENERGY (eV)
Fig. 2: Projected DOS of Ni 3d states (left pannel) and O 2p states (right pannel) for ions belonging to the first four layers of NiO(lOO). The left and right top pannels show the density of Is states corresponding to empty spheres in the vacuum regions above the Ni and O sites, respectively. The chemical potential is in the middle of the band gap (from Ref. 74)
[45,58-60,70] NiO [71,72] and FeO [73]. A rumpling with the oxygen atoms outwards is found on the alkaline earth oxide surface, in agreement with older and more recent structural determinations. On NiO(100) however, the sign of the rumpling is reversed, the Ni being located more outwards than the oxygens. The (110) surface energy is found to be about twice that of (100); the relaxation is much larger and yields a larger energy stabilization. Authors do not really agree on the charge redistribution which takes place at the MgO(lOO) surface, because the ionicity of the oxide appears to be nearly total in HF approaches [61,63], while in DFT and semi-empirical methods, the Mg-0 bonds present a small but non-negligible covalent character [60,66]. The surface projected Density of States (DOS) displays surface states just at the top of the VB and bottom of the conduction band (CB), originating from a reduction of the Madelung potential on the surface atoms. These states also exist on NiO(100) [74] and CoO(100)[75] and were shown to play a role in the formation of STM images (Fig. 2). A small reduction of the HOMO-LUMO gap results in MgO(lOO) as well as CaO(lOO) [60,64,67,68], and NiO(lOO) [74], which is qualitatively consistent with experimental observation [76-79]. Electronic d ^ d excitations at the NiO (100) surface have been calculated either with the CI method on small clusters [80-82], or with a periodic ah initio HF method by means of total energy differences [83]. Both ap-
47
Fig. 3: Iso-density surfaces (about 15% of maximum density) for the two gap states per (2\/2 X 2\/2) 2D unit cellassociated with a 25% concentration of vacancies in a close packed configuration, (a) on the left: bonding combination of the two localized states trapped at the vacancy sites; (b) on the right, antibonding combination. A top view of the surface is presented with oxygen atoms as black dots and magnesiums as light spheres. The location of the vacancies is indicated by black boxes, (from Ref. 69)
proaches reproduce satisfactorily surface-specific transitions, which are due to the modification of the crystal field splitting of the surface d orbitals. It was further shown that monotonic variations of the d —> d excitation energies take place, whose sign depends upon the nature of the orbitals involved, when the Ni coordination number Z varies in thin film geometries [84,85]. This result agrees with suggestions by Freund et al. [86] that d -^ d excitation spectroscopy could be used as a tool for characterizing under-coordinated atoms on NiO surfaces. Finally, the characteristics of the NiO surface antiferromagnetism have been looked at and in particular the value of the direct Jd and superexchange Jse coupling constants at the surface. A reduction of Jge was found on NiO(lOO) and in thin films, which is a function of the Ni coordination [84,85,87]. Oxygen vacancies on MgO(lOO) have been described by numerous authors [69,88-97]. The formation energy Eyf of an isolated vacancy in bulk MgO is large of the order of 10 eV. It decreases monotonically when the vacancy is located closer and closer to the surface [91], or in surface sites of
48
lower and lower coordination Z (step edges, corners, etc) [91,95]. Outward atomic displacements around the vacancy take place, whose strength also increases as Z decreases. In the case of neutral vacancies, each missing oxygen leaves two electrons behind itself, which are redistributed in the surface layers. They are strongly trapped at the site of the vacancy by the electrostatic potential of the surrounding Mg. However, the degree of localisation found in the calculation is dependent upon the basis set used [92,95,98]. These electrons occupy an energy level in the band gap region and can be excited into empty states. Optical transitions as well as electron addition or substraction energies have been calculated using CI [96] or total energy difference [97] approaches. The EPR spectrum (EPR= Electron Paramagnetic Resonance) of charged vacancies or divacancies has also been simulated [95,99]. In an effort to understand the strength and nature of vacancy-vacancy interactions, finite vacancy concentrations and various vacancy arrangements have been considered [69,92,93]. In the most advanced simulation, effective repulsions between vacancies have been found, presenting a non-monotonic behaviour as a function of vacancy separation. The nature of the gap states changes as a function of vacancy concentration: there is a competition between localized states trapped at the vacancy sites (Fig. 3) and delocalized states on surface magnesiums, the latter starting to be occupied above a critical concentration. In an ionic picture, the two-dimensional unit cell of the polar (111) orientation bears a charge ±2, so that a reduction of charge by a factor of 2 is required in the outer layers to heal polarity (Section 6). When charge compensation is provided by changes in stoichiometry, simple electrostatic arguments suggest two stable surface configurations. One is obtained by removing every other atom in the outermost layer, which yields a missing row surface structure with a (2 x 1) reconstructed unit cell. The second stable surface configuration, named the octo-polar (2 x 2) reconstruction [100], is obtained by removing 75% of the atoms in the outermost layer and 25% in the layer beneath, in a way which produces nano {100} facets. All quantum mechanical studies of the unreconstructed MgO(lll) surface [101-103] find a metallicity of the outer layers, which arises from large shifts and overlap of the surface VB and CB. This produces the reduction of charge in the surface layers which fulfills the electrostatic criterion. At variance with ionic models, which predict an infinite surface energy, the calculated one is large {E^ > 5 J.m"^), but not diverging. This however suggests that the unreconstructed surface is not the lowest energy configuration. Simulations of reconstructed surface [102,104-107] have considered the (2x1), the octopolar (2x2), the micro-facetted (\/3x \/3)R30° and (2\/3x
49
2\/3)R30° surfaces. The (2x2) surface energy is found to be lower than the (2x1) one, which supports the observation of the octo-polar (2x2) configuration on the parent oxide NiO(lll). In FeO(lll) however, experiments show that a two-layer film grown on Pt(lOO) and (111) substrates does not reconstruct. Its outer layer is made of oxygen atoms and the interlayer spacing is strongly reduced with respect to bulk FeO (0.68 A versus 1.25 A). The subsequent reduction of the dipole moment is consistent with a quantitative interpretation of STM images [108]. 3.3. Corundum structure Aluminium oxide as well as transition metal sesquoxides (Cr203, Ti203, V2O3, Fe203) crystallize in the corundum structure. The bulk unit cell contains six formula units with metal atoms in octahedral environment and four-fold coordinated oxygen atoms. The surface orientation which has been most thoroughly studied is the basal (0001) surface. Along (0001), the layer sequence is M/O3/M, with a very short inter-unit distance (e.g. 0.485 A in AI2O3). As a result, three chemically distinct terminations may be produced, which expose a single metal layer (M/O3/M/...), an oxygen layer (O3/M/M/...), or two metal layers (M/M/O3/...). Only the first one is non-polar and stable, on which the surface metal atoms are three-fold coordinated. The (lOTO) (the prismatic face), the (0lT2), the (1120) and the(lOll) orientations have also been considered. The structure of the unreconstructed non-polar (0001) surfaces has been the suject of all types of numerical studies, in AI2O3 [45,109-116], Cr203 [117-119] and Fe203 [120-122]. All approaches predict large surface relaxations. However, the predicted amount of relaxation depends upon the symmetry constraints which are assumed in the modelization. Surprisingly, in Al2O3(0001), even when all atomic positions are relaxed, there remains a sizeable discrepancy between ab initio and exprimental [123-125] results, which is larger than the standard error bars of the theoretical methods. As regards other surface orientations, early classical simulations had pointed out that their relative stability was strongly dependent upon relaxation effects [45]. The prismatic face turns out to have a surface energy very close to that of the basal face. An interesting point raised in Ref. 45 is the different behaviour of AI2O3 and Fe203 in this respect. While in AI2O3, the (0001) surface is found to be the most stable, in Fe203, it is the (0112), with an energy just below the (0001) surface. In addition, the prismatic Fe203 face has a higher surface energy than the (1120) face. Such differences manifest themselves in the crystal morphologies of the two oxides.
50
355
Oxygen-poor
Oxygen-rich
-2.0 -1.0 Ho-Ho(gas) (eV)
Fig. 4: Surface energies of different Fe2O3(0001) surface terminations, /io(gas) is the chemical potential per oxygen atom of molecular O2. The allowed range of /^ - /io(gas) is indicated by the vertical dotted lines. Solid lines show results for relaxed geometries, and dashed lines give, for comparison, results for unrelaxed surfaces (from Ref. 122).
The non-polar AI2O3 (0001) surface is insulating and retains a large part of the ionic character of the bulk [115]. The surface DOS displays empty surface states derived from under-coordinated aluminium atoms. The position of these surface states is very sensitive to the degree of relaxation [110,113,126] and differs according to the calculations. The authors thus do not agree on the existence of a gap narrowing at the surface, which is however found to be of the order of 1 eV by EELS experiments [126] (EELS= Electron Energy loss Spectroscopy). The Gibbs free energy of all other terminations of AI2O3 (0001) compatible with a (1 x 1) unit cell turns out to be higher than that of the non-polar surface, in the whole range of accessible oxygen chemical potential [127,128], and a nearly similar situation is found for Fe203 (0001), except at high oxygen partial pressure where the oxygen termination is stabilized [122] (Fig. 4). Energies of surface d -^ d and charge transfer excitations have been calculated on Cr2O3(0001). A reduction of ionicity at the surface was found [129,130]. The structure of ultra-thin AI2O3 films, grown on metal substrates, was shown experimentally to be more closely related to 7-AI2O3 than to aAI2O3, with part of the aluminium atoms located in tetrahedral sites [131136]. Based on ab initio DFT-LDA calculations, a model for a 5 A-thick layer on-top an Al(lll) substrate has been proposed, in which the first oxygen layer in contact with the substrate is hexagonally packed as in the 0 ( 1 X 1)/A1(111) system. Aluminium atoms bind on-top those interfacial oxygens and to three oxygens in the layer above, thus adopting a tetrahedral environment. The outer oxygen layer is strongly relaxed inwards so as to bring its atoms nearly coplanar to the aluminiums [137].
51
© #^i|
i
Fig. 5: Iso-density map of an excess electron state on TiO2(110). Titanium and oxygen atoms are marked by small and large circles, respectively, and the semi-infinite system is represented by a slab with vacuum above and below and lateral periodic boundary conditions (from Ref. 157).
3.4. Rutile structure Sn02 as well as several transition metal oxides of 1:2 composition, such as Ti02 and VO2, crystallize in the rutile structure. Their cation-oxygen bonding usually presents a large part of covalent character. The cations and oxygens are six- and three-fold coordinated, respectively. The three most stable surfaces in the rutile structure are the (110), (100) and (001) surfaces. They display arrays of two-fold-coordinated oxygen atoms — named bridging oxygens — and under-coordinated titaniums in the outer layers, with either 5-fold coordination — in the case of (110) and (100) surfaces — or 4-fold coordination —for the (001) surface. The TiO2(110) face has emerged as one of the most important oxide surfaces, which makes it a model system to understand the surface physics and chemistry of transition metal oxides.
52
Many numerical studies have been devoted to the TiO2(110) surface structure [58-60,138-147]. All approaches predict large surface relaxations, with inward displacement of the undercoordinated atoms and outward displacement of the fully coordinated atoms in the close vicinity of the surface. However, despite important efforts to increase the numerical accuracy of the calculations, by a careful check of the slab and vacuum thicknesses, etc [147], the calculated displacement of the bridging oxygens remains significantly underestimated with respect to the experimental value derived from GIXD experiments [148] (GIXD= Grazing Incidence X-Ray Diffraction). More recently, the surface structure of TiO2(110) has been re-examined by Li-impact collision ion scattering spectroscopy (ICISS) [149], which finds a relaxation of the bridging oxygens smaller than in Ref. 148, and even smaller than the calculated values. The TiO2(110) surface is found to be insulating, with no deep surface states in the bulk band gap [1]. There are oxygen derived surface states located just above the top of the bulk VB and the bottom of the CB is made of 3d orbitals located on the 5-fold coordinated titaniums [138,150156]. These local DOS features were used to interpret STM image contrast [154,156]. The analysis of an excess electron state on TiO2(110) [157] shows that the 5-fold coordinated titaniums experience the lowest Madelung potential on the unrelaxed surface and are thus the electron-acceptor sites. However, in the presence of surface relaxation, the Madelung potential is the lowest on the titaniums located just below them in the sub-surface layer, and the excess charge is thus shared between these two types of titaniums (Fig. 5). The calculated HOMO-LUMO gap is narrower at the surface than in the bulk [58-60,139,140,152,155,156]. The difference is largely reduced on relaxed surfaces because the contraction of the 0-Ti bondlength partly balances the bond breaking effect in the surface Madelung potential. The two other surfaces of low Miller indices, namely the (100) and (001) surfaces, are less stable than the (110) [58-60,138,140,141,143,152,158-160]. This is especially true for the (001) surface. Their atomic and electronic characteristics are qualitatively similar to those of the (110) surface. Aside from a first attempt to model isolated oxygen vacancies on TiO2(110) by a cluster model [153], more recent self-consistent works have considered large densities of vacancies, associated with (1 x 1) [142,143,146,156,161], (1 X 2) [143,146] or (2 x 1) [143,146,156,162] unit cells. The complete removal of bridging oxygens leads to a (1 x 1) configuration, while half removal leads to (1 x 2) or (2 x 1) reconstructions, depending upon whether one oxygen upon two is missing in each bridging oxygen row, or whether one row upon two is missing. In addition, a (2 x 1) structure may also be produced by adding rows according to a model first proposed by On-
53
ishi et al. [163]. Removing half of the bridging oxygens in each row costs less energy than removing half of the rows [143], and the (2 x 1) added row model is found to be more stable than the missing row [162]. In addition, calculated densities of states of the former [156,162] are in better agreement with STM images. DFT-LDA approaches usually predict that non-stoichiometric surfaces are metallic. However, insulating solutions are found for the ( 1 x 1 ) configuration using UHF approach [142] and for the (2 X 1) using DFT-LSDA [146]. In the latter case, a symmetry breaking takes place between the two titaniums close to a vacancy and the excess electrons localize preferentially on one of them. The presence of vacancies is thus associated with a spin polarisation of the surface layer; the ground state of the (1 x 1) surface is anti-ferromagnetic. Finally, we mention studies of the rutile SnO2(110) surface [145,164-166]. Despite the absence of partially filled d orbitals on Sn cations, the surface characteristics of Sn02 are qualitatively similar to those of Ti02. The same is true for the non-stoichiometric (1x1) and (1x2) surfaces, which present a distribution of defect states in the gap [166]. The authors, however, argue that some quantitative diflFerences with respect to Ti02 take place, which are due to the much larger polarisability of the Tin atoms, compared to the Titaniums. 3.5. Wurtzite structure There exist three polymorphs of ZnO, with zinc-blende, wurtzite and rocksalt structures. Atoms are tetra-coordinated in the two first forms, while in the latter, which is stabilized at higher pressure, the local environment is octahedral. The most thoroughly studied phase is the wurtzite one. Non polar (lOTO) and (1120) as well as polar (0001) or (OOOT) planar surfaces can be prepared. Relaxation effects at the non-polar surfaces are important [56,167-172]. In particular, on the (1010) surface, there is a sizeable contraction of the intra-plane as well as interplane Zn-0 bondlengths in the vicinity of the surface, with a very small rotation {6 ^ 3°) of the surface ZnO dimers which pushes the oxygens outwards. The short Zn-0 bond can be viewed as a double bond. At variance, based on earlier tight binding models [56], a bond-length conserving mechanism of relaxation had been proposed in which a much larger rotation of the ZnO dimers was taking place {9 ^ 18°). More recently [173], it was shown that the dimer bondlength is very sensitive to the choice of the exchange correlation functional, and that the rotation angle 0 had been previously underestimated because of an unsuflScient slab thickness. The most recent calculated value 9^5° falls close to the lower edge of the experimental error bar (11.5° =b 5°) [174]. The theoretical prediction of the surface gap width is especially diflScult
54
in ZnO because it nearly vanishes in DFT-LDA calculations. The work done in Ref. 173, which makes use of another exchange and correlation functional, yields a correct bulk band gap of 3.2 eV and predicts a tiny reduction at surface (3.12 eV). Along the polar (0001) or (OOOT) directions, ZnO exhibits a Zn/0/... stacking of the hexagonal type with interplane distances equal to i?i=0.61 A and i?2 =1.99 A inside and in between the double layers, respectively. The surface is Zn-terminated or 0-terminated, respectively, with three-fold coordinated atoms. The condition for stabilization of the polar orientation requires surface charge densities equal to a' ^ .75a (Section 6). Experimentally, unreconstructed or reconstructed terminations can be obtained, depending upon the surface preparation conditions. In particular, both cleaved and polished surfaces submitted to ion bombardment below 600° exhibit a p(l x 1) LEED diagram (LEED= Low Energy Electron Diffraction). Annealing at higher temperature may lead to ( 2 x 2 ) , {VS X V3), (3 X 3) or (4^3 x 4^3) reconstructions, some of them being possibly due to surface contamination [175-180]. Using electrostatic considerations, Nosker et al. [181] have predicted that the polar surface may be stabilized either by removing 1/4 of the outer atoms on each termination, leading to a reconstructed configuration, or by facetting, the latter process yielding a lower surface energy. Quantum mechanical approaches [182-186], on the other hand, have shown that a spontaneous charge redistribution takes place on the stoichiometric terminations, associated to a partial filling of states at the Fermi level. With this in mind, early observations of unreconstructed stoichiometric ZnO(OOOl) or (0001) (1 X 1) surfaces, with no anomalous electronic structure, seemed puzzling. A recent GIXD study [187] which finds partial occupancy of surface sites, without long range order of the surface vacancies, partly clarifies the problem and should be confronted with numerical simulations in the future. 3.6, Perovskite s t r u c t u r e The perovskite ABO3 structure possesses a cubic unit cell with the A cations at the cube vertices, the B cations at the cube centre and the oxygens at the face centers. The lowest Miller index face is the (100), which can exhibit either an SrO or a Ti02 termination in SrTiOa. Using formal charges yields an estimation of zero for the dipole moment in the repeat unit. However, we discussed elsewhere the "weakly polar" character of this surface [4,188]. The (110) and (111) orientations, on the other hand, are polar with equidistant layers (i?i = R2) in their repeat units, of the B/AO3-, and 02/ABO-type, respectively. The layers formally bear charge densities equal to ±4 per 2D unit cell, and require compensating charges
55
(a)
(b)
IA. .^-A
hK /vH
(c)
^
(d)
ii
llld J
h~^'\ -2.0 ENERGY(eV)
0.0
Fig. 6: Density of states of ideal and non-stoichiometric SrTiO3(100) slabs, (a) 1 x 1 cell; (b) \/2 X y/2 cell; (c) >/5 x v ^ cell; (d) ideal slab. The energy scale of each graph is adjusted to Fermi level (from Ref. 210).
equal to ± 2 (Section 6). The surface relaxation on both SrTiOa (001) terminations has been calculated by various numerical approaches [45,189-193]. The surface rumpling is usually reasonably accounted for, but all calculations predict an inward relaxation for the Ti02 termination, in contradiction to experiments [194-196]. A particular attention has been focused on the energy of surface states, since the first study based on non-self-consistent calculations predicted that they were located deep in the gap [197,198]. All subsequent selfconsistent calculations have contradicted this prediction [191,192,199,200], in agreement with photoemission and EELS results [201,202]. When calculated [191,192], the surface energy is rather low, an indication that no surface instability takes place, and there is no evidence of anomalous filling of electronic states. On defective surfaces, for example after Ar bombardment, surface states appear in the band gap region [203]. They were assigned to Ti^"*'-0-vacancy complexes [202]. Various reconstructions have been observed on vacuum annealed surfaces: (2 x 2) [204], (x75 x v/5)R26.6° [205-208], c(2 x 2)[209]. They were assigned to an ordering of oxygen vacancies, the latter reconstruction being possibly due to Ca segregation on the surface. Simula-
56
tions of oxygen vacancies agree on the existence of defect states in the gap [199,200,210]. They predict that the (\/2 x V2) and (\/5 x \/5)R26.6° reconstructed surfaces with one oxygen vacancy per unit cell have a metallic character [200,210] (Fig. 6). The excess electrons are shared between the vacancy and its two neighbouring titaniums, in agreement with STM images. The calculated interaction energies between vacancies depends not only upon the termination, but also upon the inequivalent positions of the vacancy pair on first neighbour sites, on the Ti02 face. On the SrO termination, the vacancies are predicted to repell each other [94]. Due to the lack of knowledge of the precise surface stoichiometry in the experiments, it is presently diflScult to propose a structure for the reconstructed surfaces. The SrTiOa (111) [211-214] and (110) polar surfaces [215-217] and the BaTi03(lll) surface [218] have been produced and studied. At variance with rocksalt polar surfaces, many of these investigations suggest that one can obtain non-reconstructed quasi-planar polar surfaces. It should be realized that the perovskite structure is such that there exist ordered configurations of vacancies in the surface layers compatible with (1x1) diffraction patterns. In addition, SrTiO3(110) displays a variety of reconstructions, such as c(2 x 6) [215,217], under reducing conditions. No precise determination of the layer stoichiometry has been performed. The atomic and electronic structures of SrTiOa (111) and (110) (1 x 1) surfaces have been investigated using a total energy, semi-empirical HF method [219,220]. Terminations of various stoichiometrics, whether compensated or not, exhibit strong electron redistributions, which suppress the macroscopic component of the dipole moment. On stoichiometric surfaces (e.g. SrOa or T i ( l l l ) layers), this is an expected result, but redistributions are also present on compensated surfaces (e.g. Sr02 or T i O ( l l l ) layers), leading to an anomalous filling of surfaces states and anomalous charge states for some surface atoms. All terminations were found to be insulating, a result assigned to the specific value ±2 of the compensating charges, which allows the complete filling of one surface state per 2D unit cell. The average surface energies were found to be rather low, a necessary condition for the obtention of planar surfaces. The relative stability of terminations of different stoichiometrics was calculated as a function of oxygen chemical potential and it was anticipated that non-stoichiometric reconstructions are not necessarily efficient mechanisms to stabilize the surface. It would be interesting, in the future, to assess whether surface stability is due to the presence of disordered surface vacancies, ordered vacancies, or whether the stoichiometric surface, with a complete filling of cation-derived surface states, has indeed a low surface energy, as predicted theoretically.
57
3.7. Other oxides Other oxide surfaces have been the subject of less numerical investigations. We will quote here some studies performed on Fe304, Si02, V2O5 and M0O3. Magnetite Fe304 crystallizes in the inverse spinel structure. It is based on a slightly distorted face centered cubic lattice of oxygen atoms, with the iron ions located in the tetrahedral A and octahedral B interstitial sites. Formally, A sites are occupied by ferric ions Fe^"^, while B sites contain an equal amount of ferrous Fe^"*" and ferric Fe^"*" ions. Magnetite possesses two polar surfaces of low index, the (100) and (111) faces. Along the (111) direction, the stacking sequence contains two formula units and is described by: Fe^/40/3FeB / 4 0 / F e ^ / F e 5 . Six different terminations may be produced. STM images [221] show surface protrusions which are assigned to the topmost Fe atoms. An ab initio HF study in a slab geometry [222] predicts that the termination by an iron bilayer (Fe^/Fe^/40... sequence) is energetically favored, thanks to an exchange of iron and oxygen layers inside the slab, which reduces the total dipole moment. It is suggested that the stability of an Fe monolayer-terminated slab (Fe^/40... sequence) may be enhanced by adsorption of hydrogen: the calculated relaxations then agree with the LEED structure refinement performed in Ref. 223. Crystalline Si02 surfaces have not been the subject of intensive structural or electronic determinations, because their large gap induces severe charging problems in spectroscopic experiments. There are several allotropes, such as a-quartz, /?-tridymite, and /3-cristobalite. The energetics of the lowest index a-quartz surfaces has been modelled by classical potential approaches [224] and the electronic structure of unrelaxed (0001) and (1010) surfaces by a semi-empirical HF method [58,60]. According to the latters, the (1010) face has a higher surface energy, and displays larger ionic charges and smaller band gaps than the bulk. On the /?-tridymite (1010) and the /3-cristobalite (110) surfaces, a tight-binding total energy calculation predicts that relaxation effects involve bond rotations, shifting the oxygens outwards and the silicon atoms inwards. This bondlength conservation mechanism allows a rehybridization of non-bonding oxygen orbitals and thus a lowering of electronic energy [56]. No deep surface states are found in the bulk band gap. Vanadium oxides of various stoichiometrics exist, related to the different oxidation states of the cation. V2O5 has the vanadium atoms with the highest valency and a dP configuration. It is insulating at low temperature, with an orthorhombic layered structure made of VO5 pyramids sharing corners and edges. A transition towards a metallic phase takes
58
0(1)
0(2)
0(3)
Fig. 7: Crystal structure of orthorhombic V2O5 with netplane stacking along (010). Vanadium (oxygen) centers are shown by large (small) balls. Inequivalent oxygen centers, 0(1), 0(2), 0(3), with 1, 2 and 3 coordination number, respectively, are labelled accordingly. (From Ref. 229).
place at high temperature. DFT calculations appUed to an orthorhombic ¥205(010) surface, either in a cluster [225,226] or in a periodic slab [227] geometry have evidenced a spectroscopic signature of the three types of surface oxygen in the Local DOS (LDOS) [228]: the terminal oxygens belonging to vanadyl groups, the two-fold bridging oxygens and the threefold oxygens (Fig. 7). These oxygens are differently charged and present a distinct reactivity towards electrophilic attack by adsorbates. The energy to remove them is also different, increasing from 6.68 eV to 7.26 eV as the oxygen coordination number grows [229]. Molybdenum oxide crystallizes in an orthorhombic structure built from a piling of bilayers along the (010) direction. The bilayers are made of MoOe octahedra sharing oxygen atoms both in and between sub-layers. They are weakly coupled to each others by van der Waals forces. Ab initio DFT cluster calculations have examined the electronic structure of the ideal (010) and (100) surfaces [230]. The first one is obtained by cleavage of the crystal and involves three types of oxygens: terminal oxygens bound to a single Molybdenum atom, asymmetrically bridging oxygen with one short and one long Mo-0 bond-lengths, and symmetrically bridging oxygens with two short and one long Mo-0 bonds. The (100) surface on the other hand is less dense and only the two first oxygen types are found. Mulliken population analysis shows large charge variations between bulk and surface atoms, and an especially large reduction of charge on the terminal and asymmetrically bridging oxygens. This inhomogeneous charge distribution manifests itself in the spatial variations of the surface electrostatic potential, an important quantity to understand the binding of ad-particles on surfaces.
59
4. DISCUSSION OF NON-POLAR STOICHIOMETRIC SURFACES This section summarises some properties common to non-polar stoichiometric oxide surfaces and presents theoretical arguments to explain them. Their specificities come from the local environment of the surface sites, which have a lower coordination number than in the bulk. From this point of view, a close parallel with unsupported clusters or ultra-thin films can be established [3]. We will not explicitely consider here the properties associated to structural defects, such as steps or kinks, for the reason of space limitation. However, most of the time, the same concepts as those akin to terrace sites apply, but with an even larger strength since the local environment is more reduced. We will successively analyse structural characteristics, energetics, electron distribution, one-particle and two-particle excitations. 4.1. Atomic structure There are numerous examples of bond-length modifications when local environments of the atoms involved in the bond change. The usual trend is a contraction when the atom coordination Z decreases. The table of ionic radii given in Ref. 231 illustrates this effect. An increase in Z is always associated with an increase of the ionic radius, and since first neighbour interatomic distances are assumed to be the sum of two ionic radii, both quantities follow the same trends. The same is true at oxide surfaces. Usually, inward relaxation is observed, whose strength is larger on more open surfaces. Systematic calculations on low index faces of rocksalt oxides examplify this tendency: the (001) surface, on which atoms are 5-fold coordinated, is hardly relaxed, while the (110) and (211) surfaces, with respectively 4-fold and 3-fold coordinated atoms, display increasing inward relaxation, associated to increasing bond-length contraction. In-plane bond-lengths are also affected, which yields a surface stress, and the same is true for bond-lengths close to surface defects like steps or kinks. It can thus be concluded that bond-length contraction around under-coordinated atoms is a general phenomenon and simple analytical models of cohesion, whether applied to ionic or covalent bonding, can account for this effect [2]. However, very often, relaxation effects involve more complex atomic displacements, yielding bond-angle distortions and/or surface rumpling. For example, a relaxation mechanism via bond rotation, has been invoked for Si02 and AI2O3 surfaces [56,110]. In the latter case, the huge contraction of the first interplane distance (shortening by «.16 A with respect to its mean value in the bulk) is not reflected strictly into the Al-0 bond-lengths
60
[0001]
® - > [loToi [OiTo]
Fig. 8: Illustration of the sapphire (0001) surface: truncated bulk and best fit model obtained from crystal truncation rods measurements. The two top planes are strongly shifted with respect to their bulk positions in a nearly bond-length conservative displacement (from Ref. 124).
because the oxygens in the second layer move radially away from the surface Al (Fig. 8). The Al-0 bond-lengths are only shortened by 4.5%, according to GIXD studies [124], and the relaxation is said to nearly obey a bond-length conservation principle. Surface layers which contain both oxygens and cations may also rumple, when forces of different strength act on the different atoms. Rumphng and relaxation effects often occur simultaneously. The rumpling amplitude is weak on dense rocksalt oxide surfaces - of the order of 0.05 A on MgO(lOO) - and much stronger in more covalent systems - about 0.4-0.5 A on the non-polar ZnO surfaces. Two mechanisms have been proposed in the literature to account for it. The first one relies on the existence of different electrostatic forces on the anions and cations, originating from polarization effects [232]. The most polarizable atom experiences the largest force [233,234]. On MgO(lOO), due to the small polarizability of magnesiums, Verwey's model predicts the oxygens to be located more outwards than the magnesiums [235], in agreement with exprimental refinements of the surface structure. On NiO(lOO), on the other hand, classical pair potential calculations [71,72] predict a rumpling of opposite sign, due to the larger polarizability of Ni ions. To our knowledge, no systematic extension of Verwey's model to other oxide surfaces has been performed. For example, rumpling has been found on the outer TiO layer of the (110) face of rutile Ti02, as well as on zirconia [52] and on C e 0 2 ( l l l ) [48] surfaces, but the
61
electrostatic arguments should be re-examined for each specific geometry, especially when cations and oxygens do not occupy symmetry-equivalent positions. On compound semi-conductor surfaces, the rumpling is generally much stronger than on rocksalt oxide surfaces, although the anion and cation polarizabilities are comparable [11,236]. In these materials, it was argued that it is easier to bend a bond than to change its length, because the radial elastic constant is higher than the angular ones. The structural distortions thus obey a bond-length conservation principle, whenever possible. In addition, according to an argument first proposed by Chadi [237], a surface rumpling with the anions outwards favors the hybridization of the anion dangling bonds with the atomic orbitals in the underlying layer. This lowers their energy and pushes them into the valence band. Other covalent forces, such as those due to 0 - 0 orbital overlap can also lead to a surface rumpling [60]. 4.2. Energetics In insulating oxides, the surface energies Eg range from 0.5 to 1 J/m^, for the most stable orientations, but quantitatively, very few measurements have been performed [238,239]. Surface energies have several contributions, among which an electrostatic and a covalent one, but it is difficult to quantify them, because no general consensus upon energy partitioning exists among theorists. When several surface orientations of a given compound are simulated with the same numerical method, surface energies display a monotonic increase with the number of broken bonds at the surface. Qualitatively, the models of cohesion developed for relaxation effects account for the gross dependence of the surface energy upon Z [2]. But for the sake of comparing two surfaces with a nearly equivalent number of broken bonds, the value of the Madelung potential acting on surface atoms is a better indicator of stability than the coordination. Smaller Madelung potentials are associated to less stable surfaces. Surfaces are always stabilised by relaxation effects. Since the latters are larger on more open surfaces, the energy lowering AEg is also larger. In their study of zirconia surfaces, Christensen et al. [52] note a linear relation between unrelaxed and relaxed surface energies {AEs/Eg constant) for both tetragonal and monoclinic faces, but this does not seem to be a general feature, as proved by the values of AEg/Eg for MgO found in Ref. 45. When two surface orientations of a given oxide correspond to numbers of broken bonds which are close to each other, the energy lowering depends upon the details of the structure. Inversions of stability between unrelaxed and relaxed surfaces may result. Within a given crystal structure, the na-
62
ture of the oxygen-metal bonding also contributes in a subtle way, yielding different surface orderings and thus different crystal morphologies. This was examphfied in Ref. 45 for a-Al203 and a-Fe203 surfaces. 4.3. Electron distribution The electron distribution in a compound reflects the topology of the system and the characteristics of the bonding. It was proved by Hohenberg and Kohn [18] that the ground state energy is univoquely specified by the electron density n{f)^ which is the quantity self-consistently determined in DFT-type numerical approaches. n{r) sensitively changes when the local environment of the atoms is modified and thus the electron distribution at surfaces may differ noticeably from the bulk distribution. Discussions of surface effects usually rely on integrated quantities of n(r), such as the ionic charges, although the latters are not measurable quantities. There are several ways to define ionic charges. When the eigen-states of the effective one-electron Hamiltonian are expanded on a localised basis set, the probability of presence of the electrons contains a 'site' contribution (square modulus of the projection of the wave function on the sites) and a 'bond' term, related to the overlap of the basis functions. In standard Mulliken population analysis [16], each bond contribution is equally shared between the atom pair. The charges then depend sensitively upon the choice of the basis set and it is meaningless to compare absolute values obtained with different methods. Only charge variations within a given method bear significance. Another way to define ionic charges consists in partitioning space into elementary volumes associated to each atom. One method has been proposed by Bader [240,241]. Bader noted that, although the concept of atoms seems to lose significance when one considers the total electron density in a molecule or in a condensed phase, chemical intuition still relies on the notion that a molecule or a solid is a collection of atoms linked by a network of bonds. Consequently, Bader proposes to define an "atom in molecule" as a closed system, which can be described by a Schrodinger equation, and whose volume is defined in such a way that no electron flux passes through its surface. The mathematical condition which defines the partitioning of space into atomic bassins is thus: Vn(r) • N{r) = 0
(7)
in which N{f) is the unit vector normal to the surface at f. Usually, and more specifically in ionic systems, each atomic bassin includes one nucleus. The advantage of Bader's approach is that the partitioning of space is uniquely defined, and in particular is independent of the basis set used in the electronic structure calculation. A Bader analysis was for example
63
Fig. 9: Electron transfer per bond A as a function of the difference SVMad in the Madelung potentials (in Hartree per electron) acting on the Ti and O in neutral stoichiometric clusters (filled circles), in bulk rutile Ti02 (plus), and on the TiO2(110) surface (diamonds) (from Ref. 243).
performed on bulk MgO and AI2O3 [242], and on free clusters and surfaces ofTi02 [243]. In most of the surfaces which have been considered numerically, it appears that surface charges display only minor variations with respect to their bulk values. This comes from the fact that charges are a complex result of oxygen-cation orbital hybridization and local environment factors which compete with one another. The Bond Transfer Method that we have recently worked out [244] makes this distinction more explicit. It can be shown, starting from the ionic limit in which oxygens and cations bear formal charges (Qo and Q^, respectively), that the sharing of electrons due to orbital hybridization is fully accounted for by positive quantities AoiCj^ defined for each bond. For simple insulating oxides with a completely filled oxygen-derived VB, the charges on an oxygen Oj or a cation Cj are expressed as: (8) Cj
Oi
The summation runs over the first neighbour atoms. Each transfer A, being related to a bond, concerns two atoms and appears in the expression of their charges. This description presents some analogy with the Bond Orbital Model [14]. As expected AoiCj goes to zero in the fully ionic limit (zero resonance integrals /3oiA,Cj//) and increases when the degree of covalency /3o.x,Cjfi/{^Cj —
64
Fig. 10: Electron transfer per bond A as a function of the Ti-0 bond length d (atomic units) in neutral (filled circles), anionic (down triangles) and cationic (up triangles) stoichiometric clusters, in non-stoichiometric clusters (stars), in bulk rutile Ti02 (plus) and at the (110) surface (diamonds) (from Ref. 243).
CQ.) of the bond increases (ecj and CQ. the relevant effective atomic orbital energies for the bond). In the framework of a Hartree theory, the effective atomic orbital energies Ci defined as the diagonal matrix elements of the self consistent Hartree operator H on the atomic orbital basis set, read: ei = {i\H\i)
e f - UiQi - Vi
(9)
In this expression, e\ is the atomic orbital energy in the isolated neutral atoms i] Ui is the intra-atomic electron-electron repulsion integral, Qi is the ionic charge, and K the Madelung potential exerted by the surrounding charges. Since in most compounds, oxygens are surrounded by cations and viceversa, the Madelung potential is positive on O and negative on C, which shifts their effective levels towards lower and higher energies, respectively [2]. The Madelung potential acting on surface atoms is usually smaller than its bulk counterpart. A decrease of the eQ. — CQ. energy difference results. Surface relaxation also modifies the values of the resonance integrals /?OiA,Cj/i with respect to the bulk. The contraction of the anion-cation bonds, usually observed at surfaces, corresponds to a strengthening of the |3oiX,Cj^J,• The AoiCj at the surface — and the same is true around any kind of under-coordinated atom — are thus enhanced with respect to the bulk, which means that the anion-cation bonding at surfaces is more covalent
65
than in the bulk [243] (Fig. 9). On the other hand, the number of atoms on which the summation in Eq. (8) runs is smaller at the surface than in the bulk. The topological factors thus compete with the electronic ones, leading usually to a rough cancellation of the effects. It is likely that having surface charges close to bulk charges also avoids to create huge electric fields and is favorable from an energetic point of view. It was shown in Ref. 243, devoted to a study oiTinOm clusters, TiO2(110) surface and bulk rutile Ti02, that, although the AoiCj are not measurable quantities, their variations are strongly and systematically correlated to variations of bond-lengths (Fig. 10). They are thus pertinent quantities to discuss modifications of the iono-covalent character at surfaces. 4.4. One particle excitations We considere in this section one particle excitations, i.e. electron removal (hole creation) and electron addition processes. We will restrict our considerations to charge transfer insulators [245]. We first recall the relationship between the first ionization potential /, the electron affinity A and the MuUiken electronegativity x ^^ ^^ atom. According to MuUiken, x = —dE/dN is equal to the derivative of the total energy E of an atom with respect to its electron number N. It is the opposite of the chemical potential, which, itself, is equal to either minus / or minus A^ depending upon whether substraction {dN < 0) or addition (dN > 0) of electrons is considered. According to Koopman's and Janak's theorems [246] — the latter may be considered as a generalization of the former in the framework of DFT — / and A are equal to minus the HOMO or LUMO levels enoMO ^^^ ^LUMOI respectively [19], provided that no electronic relaxation takes place. In stoichiometric charge transfer insulating oxides, and more specifically when under-coordinated sites are involved, the HOMO level, at the top of the VB, is usually of non-bonding character and localized on the oxygens: CHOMO ^ ^o- The LUMO level, at the bottom of the CB, has a prevalent cation character CLUMO ~ ^c- Using the expressions Eq. (9) for the effective atomic energies, one can write the MuUiken electronegativity under the following form: Xi = x'^ + UiQi + Vi
(10)
which ascertains that the electronegativity Xi oi the atom in the solid [2] is different from that of the isolated neutral atom x?- Since reactivity, acidbase strengths, etc, are often discussed in terms of the electronegativity of the active sites, Eq. (9) suggests how they vary as a function of the local environment of the surface atoms involved in the reaction: the effective levels of the surface cations (oxygen) being lower (higher) than in the bulk, electron removal is easier to achieve on the oxygens experiencing
66
the smallest Madelung potential. Similarly, electron addition preferentially takes place on the non-bonding cationic orbital of lowest energy, i.e. subject to the lowest Madelung potential (in absolute value). However, electron affinities and ionization potentials are not accurately given by the opposite of the HOMO or LUMO energies, because of the existence of electronic relaxation effects. To obtain quantitative estimates, a simulation of both the neutral and the ionized systems is required. Noting E{N) the ground state energy of the neutral system with N electrons, and E{N — 1) and E{N + 1) the ground state energies of the ionized systems with an excess hole and electron, respectively, / and A are defined by the following total energy differences: / = E{N - 1) - E{N)
A = E{N) - E{N + 1)
(11)
Vertical and adiabatic quantities are obtained if the atomic structure of the neutral system is frozen upon ionization or not. Total energy differences yield much more reliable results for / and A than HOMO and LUMO energies derived from Fock or Kohn-Sham eigenvalues. One way of measuring the "gap" G, in insulators, consists in superposing direct and inverse photoemission spectra (quasi-particle spectra of the compound), and recording the smallest energy difference between them. G is thus related to the ionization potential and electron affinity by: G = I-A
= E{N + 1) + E{N - 1) - 2E{N)
(12)
The excess electron, relevant for the estimation of the electron affinity, and the hole — relevant for the ionization potential — are never simultaneously present in the system and do not interact. If one neglects electron relaxation effects, G ^ ^LUMO — ^HOMO^ but this energy difference does not give an accurate estimate of G: there is a systematic under-estimation in the DFT-LSDA, and a systematic over-estimation in UHF. However, part of the variation of G with the conformation, size and dimensionality of the system results from one-electron processes, and is reasonably described by the variations of eiuMO — ^HOMO- AS discussed above, the ionic contribution €c — eo is smaller at the surface than in the bulk. This acts towards a narrowing of the gap, and the effect is stronger on more open surfaces. It should be noted that gap reduction at surfaces is always overestimated if relaxation effects are neglected, because the latters, by inducing bond-length contractions, strengthen the surface Madelung potential. The prediction of metallicity on NiO(llO) [247], based on unrelaxed geometry, should thus be reconsidered at the light of the present remark. The qualitative behaviour of G as a function of the local surface environment has been confirmed by systematic studies of planar [60] and
67
A / ionised state
conduction bai
-10.0
-5.0
0.0
5.0 10.0 Energy (eV)
15.0
20.0
Fig. 11: Oxygen-projected unoccupied DOS of an NiO(lOO) monolayer. Plain and dashed lines refer to the positively ionized and neutral layers, respectively (from Ref. 85).
stepped surfaces [102,248]. Even in the most advanced calculations of surface gap width, such as the LDA+U approach of Ref. 74, the small gap reduction on NiO(lOO) was assigned to a Madelung potential effect. Incidentally, this explains that calculations of the surface electronic structure, using non-self-consistent methods — in which the effective atomic orbital energies are assumed to be the same throughout the material —, generally conclude that the surface gap width is of the same order as the bulk gap, while it is found to be reduced when self-consistent methods are used. Actually, for the two oxide surfaces on which precise gap measurements have been performed, namely MgO(lOO) [76,77] and Cr2O3(0001) [133], G is smaller than in the bulk in agreement with Madelung potential arguments. It has been argued that, while the HOMO-LUMO energy difference suffers from systematic errors, a more reliable estimate of G can be obtained from the quasi-particle spectrum of the ionized system. For example, in NiO, the presence of an excess hole leads to the formation of a narrow band of unoccupied states in the VB region [249] (Fig. 11). It has been stressed [250] that the gap between this band and the CB edge approximates the optical/conductivity band gap. In LiiNiO [251] and bulk NiO [249], this method yields a value of 4 eV, in agreement with optical absorption measurements [252]. It was also used to estimate G in an NiO(lOO) monolayer [85]. A first approximation to the total quasi-particle spectrum, is given by the DOS. As was the case for / , A and G, a large part of the LDOS modifications on under-coordinated atoms is well accounted for by an effective one electron approach, because the shape of the LDOS reveals the pecu-
68
liarities of the hybridization between oxygen and cation orbitals. The DOS shapes at surfaces usually present two distinctive features: first, there is a systematic shift of the surface VB and CB, leading to a narrowing of the gap, as discussed above. In addition a systematic narrowing of the bands takes place, which is mainly due to the reduction of the surface coordination numbers [2]. To our knowledge, there has been no GW calculation of quasi particle spectra on oxide surfaces. 4.5. Two-particle excitations Electron-hole excitations can be produced in photon absorption or electron energy loss experiments: one distinguishes d -^ d excitations, and charge transfer excitations. d -^ d excitations show up in EELS spectroscopy as peaks in the region of the optical gap, for oxides having a sufficiently large band-gap and involving cations with a partially filled d shell. They correspond to optically forbidden transitions within the d manifold. In CI approaches, in an embedded cluster geometry, the ground state as well as some types of excited state are determined, which allows a calculation of the d -> d transition energies. Surface and bulk d -> d excitations in NiO(lOO) [80,81] and Cr2O3(0001) [129,130] have been obtained in that way. Another method consists in applying a variational procedure with some symmetry breaking to obtain specific types of excited states. The self-consistent loop then converges towards the state of lowest energy within the assumed symmetry class. This was done, in an ab initio periodic UHF approach in NiO [83,85], and the d -^ d transition energies were obtained by total energy differences. To our knowledge, there has been no similar attempt based on the DFT method, because the very localised nature of the d electrons in NiO is badly accounted for by DFT. The nature and number of d —> d transitions in an oxide are determined by the place of the cation in the transition metal series (filling of the d levels) and the crystal field splitting of the d levels. At a surface, an additional lifting of degeneracy usually takes place, induced by the lower symmetry of the local environment of the surface atoms, d -^ d transitions may thus take place with an energy lower than in the bulk, which was indeed observed experimentally. Moreover, the dependence of given di —)dj transition energies in NiO as a function of environment was studied [84]. It is nearly linear with respect to the coordination Z of the Ni^"^ ion, with a positive of negative slope according to whether the final or initial state of the transition is d^^^ the orbital pointing perpendicular to the surface (Fig. 12). Charge transfer excitations simulated by quasi-particle techniques lack the electron-hole interaction (excitonic effect) because the two excess charges
69
fe 5
4i 3 2 1
(z2—x2-y2)
(x2-y2 —z2) (xy—z2) (xz/yz-^z2/x2-y2) - (xy^x2-y2)
0 coordination
Fig. 12: Evolution of calculated d -> d excitation energies in NiO (001) monolayer {Z = 4), NiO (001)/MgO (001) bilayers {Z = 5) and MgO (001)/NiO (001)/MgO (001) trilayers (Z = 6), as a function of the Ni coordination. Lines are drawn to guide the eye (From Ref. 84).
are never simultaneously present in the system. In order to interpret optical absorption experiments or EELS measurements, excitonic effects, have to be explicitely taken into account. The gap value obtained in this way will be denoted G*. Embedded-cluster CI techniques may be used to simulate charge transfer excitations. This was done in a study of planar [253] or defective [254,255] MgO surfaces. The results support the argument of decreasing G* as Z decreases. They are consistent with the arguments of Madelung potentials developed above. Total energy difference techniques also yields values of G*. This procedure was followed in order to determine charge transfer gaps in various geometries of the NiO(100) monolayer, thanks to symmetry breaking tricks [84]. An increase of G* with the coordination number of the Ni ions was found in a series including the unsupported NiO monolayer, a NiO/MgO bilayer and a MgO/NiO/MgO trilayer, which again may be rationalized with Madelung potential arguments. On the other hand, a variation of G* with an opposite slope is found between the monolayer and bulk NiO {G* =5.1 eV and 4 eV, respectively), whose origin remains at present unclear. 5. NON-STOICHIOMETRIC SURFACES Non-stoichiometry in oxides is not a rare phenomenon, and especially at surfaces, where the process of annealing in ultra high vacuum after surface cleaning, induces a desorption of oxygen. Depending upon the annealing temperature and time, surfaces with various oxygen contents may be produced, whose exact stoichiometry is not easily quantified. However, some bench marks can be found when the vacancies order and give rise
70
to reconstructed structures, which allows to prepare surfaces in a roughly reproducible manner. There is a peculiarity of oxide surfaces, which is not yet understood, that no intrinsic reconstruction has ever been observed, at variance with metals or semi-conductors. In this section, we will discuss energetic, structural and electronic signatures of isolated or ordered oxygen vacancies at surfaces. We will only consider the case of neutral oxygen vacancies, in which neutral atoms are removed from the surface and leave two electrons behind them. 5.1. Atomic relaxation Structural relaxations are induced around vacancies by the breaking of oxygen-cation bonds. The importance and sign of the atomic displacements depends upon the crystal structure, the surface orientation, and more generally upon the coordination number of the missing oxygen. The general trend is a displacement of the neighbouring cations away from the vacancy, leading to a contraction of the remaining cation-oxygen bonds. This effect is qualitatively similar to relaxations on stoichiometric surfaces, as discussed in Section 4.1. It relies on the property of interatomic distances to decrease when the coordination number Z of the partner atoms is reduced. When a vacancy is created on an MgO(lOO) surface, for example, the neighbouring cations, which are five-fold coordinated on the perfect surface, become four-fold coordinated, and their displacement away from the vacancy shortens some surface Mg-0 bonds [69,91,95]. This effect is also found on SrTiO3(100) [210] and on TiO2(110) [143,249]. Its strength increases close to surface defect sites — step edges, kinks —, whose Z is smaller. However, there are exceptions to this rule, when a cation-cation bond across the vacancy site may be formed. This is the case in bulk Si02, where the creation of an oxygen vacancy breaks two 0-Si bonds. The atomic displacements of the two neighbouring silicon atoms are directed inwards in such a way that the Si-Si distance across the vacancy site is reduced by about 0.5 A with respect to its value in the perfect oxide [256]. 5.2. Charge and spin distribution The spatial distribution of the two electrons associated to a neutral oxygen vacancy depends upon the degree of covalency of the oxygen-cation bond, and on topological factors: the oxide crystal structure and the coordination of the vacancy site. In some cases, as in MgO, the electrons are trapped in the vacancy site by the strong electrostatic potential exerted by the neighbouring cations, thus forming a surface F centre (denoted Fg centre). In more covalent oxides, Ti02 and SrTiOs for example [89,210], the electron redistribution is more diffuse and reaches the neighbouring
71 d2
43
Fig. 13: Characteristics of the saddle point configuration in the diflFusion path of an oxygen vacancy on MgO(lOO) (top view), (a) atomic configuration; (b) iso-density surface for the vacancy gap state, plotted at about 15% of the maximum state density. The magnesiums and oxygens are shown as large and small circles, respectively (from Ref. 69).
cations and sometimes the second neighbour oxygens. In all cases, the electron density is asymmetric with a larger contribution above the surface plane than below. Similar distortions are also found when the vacancies are located at surface defect sites. However, the precise localization of the excess electrons is not straightforwardly found in numerical simulations, and contradictory predictions have been made as a result of different choices in the internal parameters of the computation. One of them concerns the basis set on which the wave functions are expanded. According to whether it consists of plane waves or localized functions, and, in this last case, according to whether or not virtual orbitals are introduced at the site of the vacancy, space is thoroughly spanned or not and electrons can or cannot redistribute freely in order to minimize energy. The absence of orbitals at the vacancy site, for example, prevents a correct description of an Fg centre et constrains the excess electrons to redistribute on the neighbouring cations [92]. To examplify the difficulty, we show on Fig. 13 the excess electron distribution during the migration of an oxygen vacancy in the MgO(lOO) surface plane: when the oxygen atom reaches the saddle point between two lattice positions, the excess electrons are equally shared between these two sites [69]. It is quite clear that describing this distribution with a localized basis set requires an extreme care. In addition, when localized basis sets are used, it is important to introduce polarization orbitals on the neighbouring cations, to account for their response to the strong electrostatic fields created by their highly asymmetric environment. Part of the differences found between non-stoichiometric Sn02 and Ti02 surfaces was for example
72
assigned to the larger polarisability of the former [166]. The MuUiken charges are also extremely sensitive to the choice of the basis set. This is due to the fact that bond charges, which are equally shared between the two partner atoms, result from the overlap between the neighbouring orbitals [98]. To my knowledge, no Bader analysis — which is basis set independent — ha.s been performed on non-stoichiometric surfaces. In any Ccuse, a consensus now exists that the direct view of charge density maps give more reliable — although not quantitative — information than a MuUiken charge analysis on these surfaces. The electron distribution also depends sensitively upon the treatment of exchange and correlations. For example, a comparison of DFT-LSDA and UHF simulations of non-stoichiometric bulk Ti02 shows that the electron density is larger in the vacancy site in UHF than in LSDA [161]. Similarly, on the reconstructed (2 x 1) TiO2(110) surface, a symmetry breaking induced by spin degrees of freedom — and consequently absent in DFTLDA calculations — takes place in which the excess charge is gathered on a single titanium, rather than equally shared between two titaniums. It was argued that the complete filling of spin-unpaired orbitals is more effective in reducing on-site repulsion — which is a dominant factor in narrow d band compounds — than the partial occupancy of spin-paired orbitals [161,249]. 5.3. Spectroscopic signature Spectroscopic experiments have revealed several specific signatures of the presence of oxygen vacancies, in the electronic spectra. The two electrons left by the missing oxygen fill one additional electronic level, usually located in the the gap of the oxide and which pins the Fermi level. The density of states is distorted, with a transfer of weight from non-bonding states at the top of the VB to bonding states at the bottom of the VB. Finally, core level shifts on the neighbouring cation(s) take place. The numerical studies, quoted in Section 3, support these observations. Due to the difficulties in reproducing correctly gap widths by standard ab initio methods, the position of the vacancy level e^ in the band gap is only given reliably through its distance in energy to the top of the VB. One should note that several factors affect the value of Cy. First, a good prediction requires a correct account of self-consistency effects, because the electron redistribution induces non-negligible electrostatic potentials. All the effects mentioned above — basis set limitations, choice of the exchangecorrelation functional, etc —, which may change the electron localization, can also shift Cy, since energies and wave functions are simultaneously determined in the self-consistent procedure. In addition, by modifying the Madelung potential at the vacancy site, relaxation effects around a vacancy
73
5r
0.0
Fig. 14: Band structure of a fully oxygen defective ( 1 x 1 ) MgO(lOO) surface along the three symmetry lines J-F-M of the 2D Brillouin Zone, as obtained through the FP-LMTO calculation (Full Potential- Linear Muffin-Tin Orbital method). The dashed horizontal line represents the Fermi level, black dots (star) indicate the energy positions of the filled (empty) Bloch states at F calculated in a (2v^ x 2\/2) supercell. The dashed line in the gap of the projected bulk bandstructure gives the dispersion of the F^ centre band. The dashed-dotted line is used for the surface conduction band of lowest energy (from Ref. 69).
can move Cy, This is especially important when the vacancy is located at step edges or kinks, where large atomic displacements take place. It was shown that optical transitions associated to F centres on sites of low coordination of MgO are red shifted with respect to their bulk value, due to a reduced Pauh repulsion in the excited state [96]. The nature of the gap states may change as a function of the vacancy concentration c. On MgO (100), at low values of c, the Fg levels broaden into a band, which is completely filled. The band width was interpreted within a simple tight-binding model, involving effective hopping processes between F^ orbitals on neighbouring vacancies [69]. Above a critical concentration, this band overlaps the surface conduction band, and the surface becomes metallic (Fig. 14). Two types of quantum states, available for the excess electrons, thus coexist on the surface: the Fg orbitals close to the vacancy sites, and surface CB states localized on the surface magnesiums. On rutile Ti02 (HO) surfaces with large vacancy concentrations, it was shown that the metaUic state is less stable than an insulating spin polarized state [249]. 5.4. Energetics The energetics of non-stoichiometric surfaces can be characterized in two different ways. One consists in defining the vacancy formation energy Eyf. It is the energy required to extract one neutral oxygen atom from the
74
surface and to send it at infinite distance into vacuum, either as an isolated atom or as part of an oxygen molecule. In the first case, for example, Eyf is equal to: ^vf =
Est - Enan-st "
-j^
NE{0)
(13)
with Est^ Enon-st ctud £"(0) the energies of the stoichiometric surface, the non-stoichiometric surface with N vacancies, and the neutral oxygen atom, respectively. Most of the simulations quoted in Section 3 give values of Eyf^ which allows to discuss its variations as a function of the oxide, the surface orientation and the coordination Z of the vacancy site at the surface. For a given oxide, E^f systematically decreases as Z gets smaller, i.e. as the number of broken oxygen-cation bonds gets smaller. This is checked, for example, for various vacancy sites on the MgO(lOO) surface [69,91,95] , and on inequivalent oxygen sites on the ¥205(010) surface [229]. Another approach consists in comparing the energetics of surfaces with different oxygen contents, to understand which configuration is the most stable under given experimental conditions. The surface is then assumed to be in contact with an external reservoir. The relevant thermodynamical potential is no longer the internal energy E of the surface, but rather its grand potential Vt\ n = E + PV-TS-Y:f^iNi
(14)
i
For typical pressures P and temperatures T, the PV and —TS terms can usually be neglected, f] is a function of the chemical potentials //j of the different species i, which are not independent variables. For a binary system XnOm, for example, the surface is in equilibrium both with the bulk oxide and with the outer oxygen atmosphere. The first condition yields the relationship niix + mjio = I^XnOm between the cation and oxygen chemical potentials and the bulk energy per formula unit /J^XnOm • Knowing the internal energy for a given surface termination, Q can be calculated, according to Eq. (14), for the whole range of accessible values of /JLOStability of non-stoichiometric Al2O3(0001) [127,128], Fe2O3(0001) [122] and SrTi03(lll) and (110) [220] surfaces has been discussed in this context, and a typical Q — f{fio) graph is shown in Fig. 4. The nature and strength of vacancy-vacancy interactions have scarcely been investigated, although they drive the thermodynamics of vacancy ordering on defective surfaces, and are thus key quantities to understand the numerous reconstructions observed on surfaces annealed in vacuum [1,2]. However, the ordering process may be inhibited by kinetic effects, relying on parameters, such as the activation energy for vacancy diffusion and the temperature. On MgO(lOO), vacancies have been found to weakly repell
75
each other, in a non-monotonic way as a function of their distance [69]. The overall repulsion is typical of an hybridization process — between two Fg orbitals — in which both bonding and anti-bonding states are filled. The non-monotonic behaviour is a consequence of the surface topology, the second and third neighbour interactions being differently mediated, by magnesiums and oxygens, respectively. On this surface, no strong tendency towards order could be found, in agreement with experimental observation. On SrTiO3(100), the same problem was addressed [94], but the interactions are of a different nature due to the more diffuse character of the electron redistribution. An interesting point concerns the comparison between the Ti02 and SrO terminations. On SrO, the excess electrons are shared between the vacancy site in the outermost layer and the titanium located in the sub-surface layer, thus producing a vertical electric dipole. On the Ti02 termination, the excess electrons are spread over the vacancy site and its two titanium neighbours in the surface layer [199], resulting in an horizontal quadrupolar distribution. This difference manifests itself in the sign of vacancy-vacancy interactions on the two terminations, which are found to be repulsive on SrO, and either repulsive or attractive on Ti02, depending upon the relative orientation of the two quadrupoles under consideration [94]. Finally, activation barriers have been determined for the diffusion of a vacancy in the MgO(lOO) surface, between sub-surface and surface layers, and close to surface defects [69,95]. As expected, the barrier is nearly twice lower in the first case than in the second one. Moreover, it is of the order of .20 to .25 times the formation energy, a quite reasonable ratio, in view of similar results on metal surfaces [257]. 6. POLAR SURFACES Polarity represents a peculiarity of some surface orientations in compounds involving simultaneously atoms of different electronegativity. It has been thoroughly studied in semi-conductor compounds [236,258]. A polar orientation is such that each repeat unit in the direction perpendicular to the surface bears a non-zero dipole moment. An electrostatic instability results from the presence of this macroscopic dipole, which can only be cancelled by the presence of compensating charges in the outer planes. This can be achieved either by a deep modification of the surface electronic structure — total or partial filling of surface states, sometimes leading to surface metallization— or by strong changes in the surface stoichiometry — spontaneous desorption of atoms, faceting, large cell reconstructions due to the ordering of surface vacancies, etc. Polar oxide surfaces present a large diversity, compared to sp^ semi-conductors. They exhibit a vast
76
number of crystallographic structures — rocksalt, corundum, spinel, inverse spinel, wurtzite, perovskite, for the simplest ones — which reflect the subtle mixing of ionicity and covalency in the metal-oxygen bonding and the specificities of the d electrons in transition metal oxides. In addition, mixed valence compounds, such as magnetite Fe304, can form when metal atoms with several oxidation states are involved, and, playing with experimental parameters, such as temperature, partial oxygen pressure, etc., allows to stabilize oxides of different stoichiometrics. It is clear that these peculiarities demand a generalization, if not a total reconsideration, of some theoretical concepts, a work which is still in its first stages [4]. In this section, we will discuss successively classical electrostatic arguments, some models of electronic structure which are useful in this context and finally the various processes which can stabilise a polar surface, including non-stoichiometric reconstructions. 6.1. Criterion for surface polarity According to classical electrostatic criteria, the stability of a compound surface depends on the characteristics of the charge distribution in the structural unit which repeats itself in the direction perpendicular to the surface [259]. Type 1 or 2 surfaces — with neutral or charged layers, respectively — have a zero dipole moment fl in their repeat unit and are thus potentially stable. At variance, polar type 3 surfaces have a diverging electrostatic surface energy [181] due to the presence of a non-zero dipole moment not only on the outer layers (which would not distinguish them from non-polar rumpled or reconstructed surfaces), but on all the repeat units throughout the material. The total dipole moment is thus proportional to the number of repeat units and so is the electrostatic contribution to the surface energy per unit area (Fig. 15). This is the origin of the surface instability. The estimation of the dipole moment in the repeat unit is not always an easy matter. In binary compounds, the difference in electronegativity of the constituents readily points out the existence and sign of the charge transfer between the ions. In simple crystal structures and for orientations such that layers containing cations only and anions only alternate, the presence of a dipole moment in the repeat unit is thus unquestionable, whatever the charge values — provided they are non-zero. These surfaces are undoubtly polar surfaces. The same is also true for some surfaces of ternary compounds, such as the (110) and (111) faces of ABO3 perovskites. For example, the (110) and (111) repeat units contain alternating ABO and O2 layers in the first case, and alternating AO3 and B layers in the second one, which are undoubtly charged, because, in each case, one of the layers contains a single constituent.
77 -a 0 -a a -CG
4iio
(a)
l.4n(5R.
-GO-GO
-GG 4710 ^7 4jto_^2?j
(b) -47ia ^
J_4-
/?j+/?2
Fig. 15: Spatial variations of the electrostatic field E and potential F in a slab cut along a polar direction.
There are less obvious cases, among which is the (100) perovskite sur~ face. SrTiO3(100) , for example, presents alternating layers of SrO and Ti02 composition. If formal charges are assigned to the ions (Sr^"^, Ti'*"^ and 0^~), each layer is charge neutral, the repeat unit bears no dipole moment, and the orientation is considered as non-polar. This is the statement most often encountered in the literature. However, SrTiOs is not fully ionic. Its gap width, equal to 3 eV, places it on the border line between semiconductors and insulators and the Ti-0 bond presents a nonnegligible part of covalent character. The actual charges are thus likely not equal to the formal ones and Qsr + Qo and Qxi + 2Qo are not likely to vanish. SrTiO3(100) should thus be considered as a polar surface and this examplifies how careful one should be in the classification of surfaces. In addition, the surface orientation h is not always sufficient to fully characterize a semi-infinite system, especially when various terminations may be produced. In the rutile structure, for example, the bulk repeat unit in the (110) direction is made of three layers of O and (M0)2 composition, and there exist three chemically inequivalent terminations, which expose a single oxygen layer ( 0 / ( M 0 ) 2 / 0 sequence), two oxygen layers ( 0 / 0 / ( M 0 ) 2 sequence) or one mixed cation-oxygen layer ( ( M 0 ) 2 / 0 / 0 sequence). Only in the first case, the repeat unit bears no dipole mo-
78
merit. Similarly, on the basal (0001) face of the corundum structure, three chemically distinct terminations may be produced (Section 3.3). Only the M/O3/M termination is non-polar. According to classical electrostatics, polar surfaces are thus unstable. However, it can be shown [2,94] that the macroscopic dipole moment can be cancelled out by modifications of the charge densities on the outer layers. Assuming that m outer layers have a charge density CTJ different from the bulk {\aj\ 7«^ cr for 1 < j < m) and that the layer spacing is alternatively equal to Ri and i?27 the electrostatic condition for surface stability reads:
E ^j =
^m+l
— 5 -
(_!)"» _ -R2 - -Ri i?2 + -Rl
(15)
Several scenarios, in which either the charge or the stoichiometry of the layers is changed can thus lead to polarity healing: • one or several surface layers have their composition which differs from the bulk stoichiometry. Reconstruction or terracing will result, depending upon how the vacancies or adatoms order. If no order takes place, no information on the surface stoichiometry can be extracted from surface diffraction patterns, unless a quantitative analysis is performed. • foreign atoms or ions, coming from the residual atmosphere in the experimental set-up, provide the charge compensation. • on stoichiometric surfaces, charge compensation may be provided by the electron redistribution which takes place in response to the polar electrostatic field. This is well examplified in self-consistent electronic structure calculations. Which process actually takes place depends upon energetic as well as kinetic considerations. As will become clear in the following, if stoichiometric ideal polar surfaces are not observed, this is never because their surface energy diverges. There always exist enough electronic degrees of freedom in a material to reach charge compensation through the third mechanism. However, in most cases, the resulting surface energy is high and other processes may be more efficient. 6.2. Models of electronic structure in iono-covalent application to polar surfaces The characteristics of the charge distribution in the bulk and layers are thus key factors to understand the physics of polar zeroth-order description of the electronic structure of oxides
materials: in the outer surfaces. A is the fully
79
ionic model, in which formal charges are assigned to bulk as well as to surface atoms. This model successfully assesses whether, in a binary oxide, a bulk repeat unit bears a non-zero dipole moment, and can predict the stable surface terminations in some complex structures. It applies well to highly ionic materials, but cannot be justified in more covalent compounds or when surface charge redistributions take place. For example, according to it, the surface energy of stoichiometric SrTiOs and BaTi03(lll) and (110) surfaces should be infinite, as well as those of the oxygen or iron bilayer terminations of a-Fe203 (0001), in contradiction with both experiment or ab initio calculations. In addition, it may even give wrong predictions on the nature (polar or not) of a surface, in ternary compounds, as in the case ofSrTiO3(100) In tetrahedral semi-conductor compounds, on the other hand, the most widely used criterion for the stability of polar surfaces relies on specific fillings of dangling bonds — sps atomic orbitals which remain unpaired due to surface bond breaking. The criterion is referred to as the autocompensating model [260,261]. It states that a surface is stable if anionderived and cation-derived dangling bonds are occupied and empty, respectively, because such a filling induces a net lowering of the surface energy. According to electron counting rules, it has been realized that, at polar surfaces, the condition of auto-compensation implies that Eq. (15) is fulfilled, and the surfaces are then said to be "charge neutral" [262]. Later, the auto-compensation principle was generalized to describe mineral surfaces [263]. assuming that the electron number per bond is the same at the surface and in the bulk (but not necessarily equal to 2 as in tetraedral semiconductors). This principle has been applied for example to a-Al2O3(0001) [110], Fe3O4(100) [264-267] and a-Fe2O3(0001) [268,269]. This generalization may be useful in a number of cases. However, the vocabulary taken from the physics of semiconductors is highly unsuited to oxides, because, in most cases, dangling bonds have no physical significance. The conservation of the number of electrons per bond between bulk and surface is also not a well founded assumption. The description of the charge distribution in terms of electron transfer per bond [244] given in Section 4.3, allows to make a bridge between the two limits and can account for the metal-oxygen bonding in the whole range of ionicities. We will use it in the following to understand the mechanisms at work on polar surfaces. It has to be noted that, for this purpose, it will never be necessary to know quantitatively the values of the parameters AoiCj, neither in the bulk nor at surfaces. It is the principle of electron sharing per bond which is the key ingredient.
80
6.3. Surface processes relevant for polarity healing We will successively discuss surface relaxation effects, changes of covalency in the outer layers, partial fillings of surface states and stoichiometry changes. It is often thought that surface relaxation is one of the most efficient processes to stabilize polar surfaces, especially open surfaces, which experience large atomic displacements. Actually, two aspects of the problem should be distinguished. First, as far as the electrostatic criterion Eq. (15) is concerned, the macroscopic component of the dipole moment is entirely determined by the charges and layer spacings in the bulk repeat unit. Consequently, in the vicinity of the surface, a mere contraction or expansion of the interlayer spacings, not accompanied by a change in the charge densities, can never heal polarity However, once charge compensation is achieved — by whatever mechanism —, surface relaxation induces a lowering of surface energy, as it does on non-polar surfaces, and the strength of the effect increases as the coordination of the surface atoms gets lower [60] (c.f Section 4.1), as exemplified on the a-Al203 (0001) surface (Section 3.3), and on the reconstructed (2 x 1) NiO(lll) surface [270]. Since charge compensation requires a modification of the charge density, changes of covalency at the surface are often assumed to heal the polarity. With the help of the bond transfer model, one can show that this statement is incorrect, as far as semi-infinite polar surfaces are concerned. It is useful to make a distinction between weakly polar surfaces, in which the dipole moment in the repeat unit is entirely due to covalent effects, and truely polar surfaces whose dipole moment contains an integer contribution. As already said, in the fully ionic limit, the first ones are considered as nonpolar, while the second ones are recognized as polar. A prototype of the first family is SrTiO3(100). Applying the Bond Transfer Model [244], it is found that the charge densities per two-dimensional unit cell, on the Ti02 and SrO bulk layers, read (JB = IQxi + 2(5o| = QST + Qo — 2ATi-o — 8Asr-o, in terms of the electron transfers Axi-o and Asr-o per Ti-0 and SrO bonds. The dipole moment in the bulk repeat unit is thus non-zero. However, it has no integer contribution and depends only on covalent effects. On the (100) surface, due to the change in local environment, due to possible structural distortions and due to shifts of atomic levels, redistributions of charge take place (Fig. 16). On the Ti02 termination, for instance, assuming modified Afpi_o and Agj._o values yields atomic charges which differ from the bulk in two layers. However, it is found that a\ + a2 = Axi-o — 4Asr-o, which fulfills the condition ^1 + cr2 — —crB/2, Eq. (15), whatever the specific values of the A and A' parameters. On SrTiO3(100), the bond breaking mechanism, by itself.
81 Ti02
-f^o
«-^-»
MgO(lll)
SrTiO3(001)
0 =oxygen
000000 0
• =titanium
# =strontium or magnesium
Fig. 16: Electron transfers per bonds introduced in the modelization of an SrTiO3(001) surface (left pannel) and of an MgO(lll) surface (right pannel). On SrTiO3(001), specific transfers A^i-o ^ind Asi._o (represented by thick arrows) are introduced inside the surface layer and in between the surface and sub-surface layers for T i - 0 bonds and Sr-0 bonds involving surface atoms. On M g O ( l l l ) , only the first inter-plane transfer is assumed to be modified, (from Ref. 4).
thus yields the charge redistribution which suppresses the divergent part of the electrostatic potential. The surface charges are different from the bulk, both because the bond covalency in the surface layers is different from the bulk (A' ^ A) and because the coordination of surface atoms is reduced with respect to the bulk. However, the model shows that only the second factor is effective for polarity healing, while the change of covalency plays no role in the charge compensation process. On the other hand, on truely polar surfaces, in which the dipole moment borne by the bulk repeat unit contains an integer contribution a change in covalency in the surface layers cannot either heal polarity, MgO(lll) is representative of this family. The charge density per ( 1 x 1 ) unit cell on bulk (111) layers is equal to GB — ±(2 — 6A). On a stoichiometric (111) magnesium termination, surface atoms are three-fold coordinated to atoms located in the underlying layers. Their charges QMgi are different from those in the bulk. But the same is also true for oxygen atoms in the sub-surface layer (n = 2) because A' 7^^ A (Fig. 16). (^Mgi and QQ^ are such that (Ji -h (J2 = 3A. This does not fulfill the electrostatic criterion, which, instead, requires that ai + a2 = 3A — 1. The surface layers have a deficit of one electron, which is independent of the values of A' and A. It should be noted that the same result is obtained in a fully ionic picture (A' = A = 0) or when neglecting the change of covalency (A' = A). Neither in this family, thus, can charge compensation be achieved by change of covalency at the surface. What we have demonstrated here, is very general: a mere change of
82
(b)
I
E p
vty^ o.o
Energy ( e V )
-4.0
-
O.O Energy ( e V )
Fig. 17: Calculated DOS for five-layer symmetric MgO(lll) slabs, with oxygen (a) and magnesium (b) terminations. Plain and dashed lines represent the DOS in the slab and in the bulk, respectively. The top and lower panels refer to projections on majority and minority spin states, respectively. All DOS have been convoluted by a 0.1 eV wide Gaussian function (from Ref. 103).
covalency in the outer layer can never provide the compensating charges, because it concerns several layers whose contributions cancel out in the expression of the electrostatic criterion. On these stoichiometric polar surfaces, charge compensation can only be achieved by a partial/total filling of CB surface states or depletion of VB surface state. The Bond Transfer Model helps understanding that this filling / can be determined by ionic arguments, even in very covalent materials, and that it is independent of the specificities of hybridization in the outer layers [244]. In the case of MgO(lll), for example, follovi^ing the charge analysis just given above, charge compensation is only achieved if / = 1/2. The analysis of the oxygen termination leads to the symmetric conclusion that a surface valence band has to be half-filled. It should be noted that the condition / = 1/2 is independent of the precise values of the electron transfers per bond. It would have been similar using a fully ionic picture. The result that we have obtianed is not restricted to ionic oxides such as MgO. On ZnO(OOOl) or SrTi03(lll) or (110), the occupancy of surface states is also dictated by the values of the formal charges and of the interlayer spacings Ri and i?2, despite the fact that these oxides are rather covalent. The ionic limit can thus be currently used to estimate the value of / required for charge compensation. Two cases occur. In the first case, / is non-integer. The surface states are only partly filled or empty and the surface layers have a "metallic" character ^ ^We use here the term "metallic" for the sake of simplicity, meaning that electronic excitations of zero energy can be
83
(Fig. 17). This occurs on the rocksalt oxide (111) surfaces (/ = 1/2), the ZnO(OOOl) surface (/ = 1/4), the oxygen termination of corundum oxide (0001) surface (/ = 1.5), etc. These partial fillings were indeed found in the quantum calculations. For some other polar surfaces, / is integer, and the surface can remain insulating. This takes place for example, on the (111) or (110) polar surfaces of SrTiOa (/ = 1). Charge compensation can also be achieved by changing the atomic density in the outer layers. Again, according to the bond transfer model, the required modifications of stoichiometry are found to be entirely determined by ionic arguments, and not by the specificities of hybridization in the outer layers. For example, in the case of MgO(lll), in a two-dimensional supercell containing M surface Mg, from which one atom is removed, the electrostatic criterion requires that / = (M — 2)/(2M — 2). This relationship is independent of the degree of covalency of the Mg-0 bond Its simplest achievement, with no filled conduction band ( / = 0), is obtained for M = 2. This amounts to removing half of the surface atoms from the outermost layer. A similar reasoning, assuming modifications of stoichiometry in two surface layers instead of one, would lead to the octopolar surface configuration. It is thus possible to predict which stoichiometry, in the surface layers, compatible with an insulating band structure, yields charge compensation, only on the basis of ionic considerations, even in very covalent compounds. This does not mean that covalency effects are absent, or that they are the same as in the bulk, but rather that they cancel out in the condition for charge compensation. Charge compensated nonstoichiometric surfaces, such as the (2 x 2) octopolar reconstructed rocksalt (111) surfaces, have electronic properties which present no anomaly. However, due to the presence of surface atoms with low coordination numbers, interesting phenomena for applications, such as a reduction of the gap, an increase in basicity of surface oxygens, or an increase in acidity of surface metal atoms can be expected, as on open non-polar surfaces [58,59,271,272]. 6.4. Summary Polar oxide surfaces present a wide variety of electronic and atomic characteristics, which are dependent upon the crystal structure, the ionicity of the metal-oxygen bonding, the surface orientation and its stoichiometry. The nature of the microscopic processes responsible for the cancellation of polarity provides a means to introduce a classification among these surfaces. Weakly polar surfaces are met each time that the dipole moment in produced. However, when the Fermi level crosses a narrow band, as met on the oxygen termination of MgO(lll) for example, and when the surface metallization would survive in the limit of zero bond width (no electron delocalization, zero value of the resonance integrals), the system should rather be named an "open-shell" system
84
the repeat unit is a function of the electron transfers per bond only and does not involve integer contributions, as in the case of SrTiO3(100). On stoichiometric surfaces, the charge compensation may be assigned to the effect of bond breaking. The surface LDOS differs from the bulk one because of the reduced local environment of the surface atoms (different effective atomic orbital energies, bond-length modifications, change in the coordination numbers). However, the terminations are insulating, with filled oxygen-derived surface states and empty metal-derived surface states. These surfaces have low surface energies and can be produced stoichiometric and planar. We have called them previously weakly polar surfaces^ because their polar instability is weak. For fully polar surfaces, the surface dipole in the repeat unit contains an integer contribution, independently of the values of the electron transfer per bond. For semi-infinite surfaces, it was shown that a mere change of covalency in the outer layers can never provide the compensating charges, because it modifies the charges of several types of atoms whose contributions cancel out in the expression of the electrostatic criterion. Similarly, and for the same reason, surface relaxation cannot provide the compensating charges. Only a modification in the filling of surface states or a change in the stoichiometry of the surface layers may yield charge compensation. It turns out that if stoichiometric polar surfaces are unstable, this is never because they present a diverging electrostatic surface energy. The compounds have enough degrees of freedom, and in particular enough flexibility of their electronic structure in response to the surface potential, to heal the polarity while remaining stoichiometric. If a partial filling of surface states is required — and, as shown above, this can be estimated assuming a fully ionic limit —, as in the rocksalt (111), wurtzite (0001) or (0001), etc surfaces, the terminations present a "metallic" character. This is usually not favourable, from an energetic point of view, as recognized in the expression of the auto-compensation principle. On other stoichiometric polar surfaces, an integer filling of surface states is required. This is the case on SrTiOa or BaTiOs (110) and (111) surfaces. The surfaces thus may remain insulating but a conduction state is located below the Fermi Energy Ep- Self-consistent calculations give a hint that this electronic structure is not associated with a high surface energy. They support preliminary experimental observations of non-reconstructed perovskite (111) and (110) surfaces. This suggests that the autocompensation principle (filled anion-derived surface states and empty cation-derived surface states) should be extended to include all insulating surface configurations, whatever the nature of the filled and empty states. However, in perovskite compounds, the absence of reconstruction is not necessarily
85
synonymous with stoichiometry, and further work should assess this point in the future, by determining quantitatively surface compositions. Polarity may also be healed by the removal of a certain percentage of atoms in the outer layers. When the vacancies order, most of the time, this leads to surface reconstructions. The surface concentrations compatible with a vanishing macroscopic dipole moment can be correctly estimated within a fully ionic picture. Low energy configurations are expected to be insulating, and indeed, on non-polar oxide surfaces, a correlation has been found between the stability of a surface orientation and the surface gap value [58,59]. It is clear that much work remains to be done to extend our understanding to polar surfaces of transition metal oxides in which the cations have partially filled d orbitals. An especially challenging issue is related to mixed valence metal oxides, such as Fe304, in which the cations exist under two oxidation states. In addition, considering the rapid development of ultra-thin film synthesis and characterization, a simultaneous effort should be performed on the theoretical side to settle the conditions of stability of polar films. More generally, on the experimental side, it seems that one of the present bottlenecks is in a quantitative determination of the surface stoichiometry, an information of prominent interest to interpret the presence or absence of reconstruction. 7. CONCLUSION We end this review on clean oxide surfaces by stressing some directions in which theoretical improvements can be expected in a near future One deals with the ah initio description of electronic excited states. These include the attachment or removal of electrons, the account of direct or inverse photo-emission spectra, and the electron-hole excitations of the d -> d or charge transfer type. Advanced methods are presently under development to account for them: the GW method, the SIC method, the LDA-hU method, etc. However, they imply an increased computation cost, which is not routinely accessible for complex systems, such as most oxide surfaces. These methods are also expected to open the field of strongly correlated materials, among which transition metal oxides, which have important technological applications: high-Tc superconductivity, giant magneto-resistance, etc. A second challenge is in the understanding of surface magnetism: elementary spin-spin interactions, atomic magnetic moments on surface sites, collective properties such as surface magnetic ordering, spin polarised transport across oxide ultra-thin barriers, etc. We have not developed this problematics here, because, until now, there have been very few related the-
86
oretical studies. However, new beautiful experimental results have been recently published which should encourage theorists to join this field. Finally, understanding the reconstructions of oxide surfaces from the point of view of non-stoichiometry and/or polarity represents a necessary step for the production and use of high quality nano-structured surfaces. On the experimental side, it seems that one of the present bottlenecks is in a quantitative determination and control of the surface stoichiometry. On the theoretical side, there are strong limitations to apply ab initio methods to reconstructed surfaces because of their computational cost. In addition, it is not yet clear why intrinsic reconstructions do not exist and what are the driving forces for vacancy ordering. No doubt that these points will be seriously considered in a near future if one is to use reconstructed oxide surfaces as substrates to grow artificial structures. REFERENCES I] V.E. Henrich and P.A. Cox, The Surface Science of Metal Oxides, Cambridge University Press, Cambridge, 1994. 2] C. Noguera, Physics and Chemistry at Oxide Surfaces, Cambridge University Press, Cambridge, 1996. 3] C. Noguera, Surf. Rev. Letters , (2001) under press. 4] C. Noguera, J. Phys. Condensed Matter, 12 (2000) R367. 5] B. G. Dick and A. W. Overhauser, Phys. Rev., 112 (1958) 90. 6] W. C. Mackrodt and R. F. Stewart, J. Phys. C, 12 (1979) 431. 7] P. A. Madden and M. Wilson, Chem. Soc. Rev., 25 (1996) 339. 8] M. Wilson and P. A. Madden, Faraday Discussion, 106 (1997) 339. 9] A. M. Stoneham and J. H. Harding, Ann. Rev. Phys. Chem., 37 (1986) 53. 10] M. C. Payne, M. P. Teter, D. C. Allan, T. A. Arias and J. D. Joannopoulos, Rev. Mod. Phys., 64 (1992) 1045. II] J. P. LaFemina, Surf. Sci. Rep., 16 (1992) 133. 12] J. A. Pople, D. P. Sentry and G. A. Segal, J. Chem. Phys., 43 (1965) S129. 13] J. A. Pople, D. P. Sentry and G. A. Segal, J. Chem. Phys., 44 (1965) 3289. 14] W. A. Harrison, Electronic Structure and the Properties of Solids, Freeman and co, San Francisco, 1980. 15] C. Pisani, R. Dovesi and C. Roetti, Hartree-Fock ab initio treatment of crystalline systems. Lecture Notes in Chemistry, Vol. 48, Springer Verlag, Berlin, 1992. 16] A. Szabo et N.S. Ostlund, Modern Quantum Chemistry: introduction to advanced electronic structure theory, MacGraw Hill, 1989. 17] N. Govind, Y. A. Wang and E. A. Carter, J. Chem. Phys., 110 (1999) 7677. 18] P. Hohenberg and W. Kohn, Phys. Rev., 136 (1964) B864. 19] R. G. Parr and W. Yang, Density functional theory of atoms and molecules, Oxford University Press, Oxford, 1989. 20] R. Jones and O. Gunnarsson, Rev. Mod. Phys., 61 (1989) 689. 21] A. Nagy, Phys. Rep., 298 (1998) 1. 22] J. P. Perdew, J. A. Chevary, S. H. Vosko, K. A. Jackson, M. R. Pederson, D. J. Singh and C. Fiolhais, Phys. Rev. B, 46 (1992) 6671. [23] J. P. Perdew, K. Burke and Y. Wang, Phys. Rev. B, 54 (1996) 16533.
87
J. P. Perdew, K. Burke and M. Ernzerhof, Phys. Rev. Letters, 77 (1996) 3865. J. P. Perdew and A. Zunger, Phys. Rev. B, 23 (1981) 5048. M.S. Hybertsen and S.G. Louie, Phys. Rev. B, 34 (1986) 5390. R. Godby, M. Schliiter and L. J. Sham, Phys. Rev. B, 37 (1988) 10159. F. Bechstedt, Adv. Solid State Phys., 32 (1992) 161. L. Hedin, Phys. Rev., 139 (1965) A796. L. Hedin and S. Lundqvist, Solid State Phys., 23 (1969) 1. M.S. Hybertsen and S.G. Louie, Phys. Rev. B, 37 (1988) 2733. L. Reining, G. Onida and R. W. Godby, Phys. Rev. B, 56 (1997) 4301. F.Bechstedt and R. Del Sole, Phys. Rev. B, 38 (1988) 7710. F. Gygi and A. Baldereschi, Phys. Rev. Letters, 62 (1989) 2160. V. I. Anisimov, J. Zaanen and O. K. Andersen, Phys. Rev. B, 44 (1991) 943. V. I. Anisimov, I. V. Solovyev, M. A. Korotin, M. T. Czyzyk and G. A. Sawatzky, Phys. Rev. B, 48 (1993) 16929. [37" V. I. Anisimov, F. Aryasetiawan and A. I. Lichtenstein J. Phys. Condens. Matter, 9 (1997) 767. [38 J. P. Perdew, R. G. Parr, M. Levy and J. L. Balduz, Jr., Phys. Rev. Letters, 49 (1982) 1691. [39 V. I. Anisimov, and O. Gunnarsson Phys. Rev. Letters, 55 (1985) 418. [40^ A. Georges, G. Kotliar and M. rozenberg. Rev. Mod. Phys., 68 (1996) 13. [41' A. I. Lichtenstein and M. I. Katsnelson, Phys. Rev. B, 57 (1998) 6884. [42 G. Pacchioni, P. S. Bagus and F. Parmigiani (Eds) Cluster models for surface and bulk phenomena, NATO ASI Series B, Vol. 283, Plenum, New York, 1992. [43 G. Renaud, B. Villette, I. Vilfan, A. Bourret, Phys. Rev. Letters, 73 (1994) 1825. [44' J. C. Phillips, Rev. Mod. Phys., 42 (1970) 317. [45^ W. C. Mackrodt, Phys. Chem. Miner., 15 (1988) 228. [46^ A. Lichanot, M. Gelize, C. Larrieu and C. Pisani, J. Phys. Chem. Sol., 52 (1991) 1155 [47 T. Ouazzani, A. Lichanot and C. Pisani, J. Phys. Chem. Sol., 56 (1995) 915. [48 T. X. T. Sayle, S. C. Parker and C. R. A. Catlow, Surf. Sci., 316 (1994) 329. [49^ J. C. Conesa, Surf. Sci., 339 (1995) 337. [50 P. W. Tasker, Surf. Sci., 37 (1979) 315. [51 R. Orlando, C. Pisani, E. Ruiz and P. Sautet, Surf. Sci., 275 (1992) 482. [52 A. Christensen and E. A. Carter, Phys. Rev. B, 58 (1998) 8050. [53 M. R. Castell, A. S. L. Dudarev, C. Muggelberg, A. P. Sutton, G. A. D. Briggs and D. T. Goddard, J. Vac. Sci. Technol. A, 16 (1998) 1055. [54^ A. J. Martin and H. Bilz, Phys. Rev. B, 19 (1979) 6593. [55^ F. W. de Wette, W. Kress and U. Schroder, Phys. Rev. B, 32 (1985) 4143. [56^ J. P. LaFemina and C. B. Duke, J. Vac. Sci. Technol. A, 9 (1991) 1847. [57- S. Russo and C. Noguera, Surf. Sci., 251-252 (1991) 1081. [58^ J. Goniakowski and C. Noguera, Surf. Sci., 319 (1994) 68. [59 J. Goniakowski and C. Noguera, Surf. Sci., 319 (1994) 81. [60 J. Goniakowski and C. Noguera, Surf. Sci., 323 (1995) 129. [61 M. Causa, R. Dovesi, C. Pisani and C. Roetti, Surf. Sci., 175 (1986) 551. [62 M. Causa, R. Dovesi, E. Kotomin and C. Pisani, J. Phys. C: Solid State Phys., 20 (1987) 4983. [63 C. Sousa, J. A. Mejias, G. Pacchioni and F. Illas, Chem. Phys. Letters, 249 (1996) 123. [64 C. Satoko, M. Tsukada and H. Adachi, J. Phys. Soc. Jpn., 45 (1978) 1333. [24 [25 [26 [27; [28' [29 [30 [31 [32 [33^ [34 [35' [36
[65] A. Gibson, R. Haydock and J. P. LaFemina, J. Vac. Sci. Technol. A, 10 (1992) 2361. 66] U. Birkenheuer, J. C. Boettger and N. Rosch, J. Chem. Phys., 100 (1994) 6826. 67] S. Pugh and M. J. Gillan, Surf. Sci., 320 (1994) 331. 68] U.Schonberger and F. Aryasetiawan, Phys. Rev. B, 52 (1995) 8788. 69] F. Finocchi, J. Goniakowski and C. Noguera, Phys. Rev. B, 59 (1999) 5178. 70] E. A. Colbourn and W. C. Mackrodt, Sol. St. Ionics, 8 (1983) 221. 71] P. W. Tasker and D. M. Duffy, Surf. Sci., 137 (1984) 91. 72] P. M. Oliver, G. W. Watson and S. C. Parker, Phys. Rev. B, 52 (1995) 5323. 73] J. Kendrick, E. A. Colbourn and W. C. Mackrodt, Radiation Effects, 73 (1983) 259. 74] S. L. Dudarev, A. I. Liechtenstein, M. R. Castell, G. A. D. Briggs and A. P. Sutton, Phys. Rev. B, 56 (1997) 4900. 75] M. R. Castell, S. L. Dudarev, G. A. D. Briggs and A. P. Sutton, Phys. Rev. B, 59 (1999) 7342. 76] J. W. He and P. J. M0ller, Phys. Stat. Sol. B, 133 (1986) 687. 77] F. Didier and J. Jupille, Surf. Sci., 307-309 (1994) 587. 78] A. R. Protheroe, A. Steinbrunn and T. E. Gallon, J. Phys. C: Solid State Phys., 15 (1982) 4951. 79] A. R. Protheroe, A. Steinbrunn and T. E. Gallon, Surf. Sci., 126 (1983) 534. 80] A. Freitag, V. Staemmler, D. Cappus, C. A. Ventrice, Jr, K. Al-Shamery, H. Kuhlenbeck and H. J. Freund, Chem. Phys. Letters, 210 (1993) 10. 81] M. Geleijns, C. de-Graaf, R. Broer and W. C. Nieuwpoort, Surf. Sci., 421 (1999) 106. 82] S. Shouheng and V. Staemmler, Phys. Rev. B, 52 (1995) 12345. 83] W. C. Mackrodt and C. Noguera, Surf. Sci. Letters, 457 (1999) 386. 84] W. C. Mackrodt, C. Noguera and N. L. Allan, Faraday Discussion, 114 (1999) 105. 85] C. Noguera and W. C. Mackrodt, J. Phys. Condensed Matter, 12 (2000) 2163. 86] H. J. Freund, H. Kuhlenbeck and V. Staemmler, Rep. Prog. Phys., 59 (1996) 283. 87] C. de-Graaf, F. lUas, R. Broer and W. C. Nieuwpoort, J. Chem. Phys., 106 (1997) 3287. 88] R. R. Sharma and A. M. Stoneham, J. Chem. Soc. Faraday Trans. II, 72 (1976) 913. 89] M. Tsukada, H. Adachi and C. Satoko, Prog. Surf. Sci., 14 (1983) 113. 90] A. Gibson, R. Haydock and J. P. LaFemina, Appl. Surf. Sci., 72 (1993) 285. 91] L. N. Kantorovich, J. M. Holender and M. J. Gillan, Surf. Sci., 343 (1995) 221. 92] E. Castanier and C. Noguera, Surf. Sci., 364 (1996) 1. 93] E. Castanier and C. Noguera, Surf. Sci., 364 (1996) 17. 94] C. Noguera, A. Pojani, F. Finocchi and J. Goniakowski, NATO ASI Series E, 331 (1997) 455. 95] G. Pacchioni and P. Pescarmona, Surf. Sci., 412-413 (1998) 657. 96] C. Sousa, G. Pacchioni and F. Illas, Surf. Sci., 429 (1999) 217. 97] P. V. Sushko, A. L. Shluger and C. R. A. Catlow, Surf. Sci., 450 (2000) 153. 98] G. Pacchioni, A. M. Ferrari and G. lerano, Faraday Discussion, 106 (1997) 155. 99] L. Ojamae and C. Pisani, J. Chem. Phys., 109 (1998) 10984. 100] D. Wolf, Phys. Rev. Letters, 68 (1992) 3315. 101] M. Tsukada and T. Hoshino, J. Phys. Soc. Jpn., 51 (1982) 2562. 102] A. Pojani, F. Finocchi, J. Goniakowski and C. Noguera, Surf. Sci., 387 (1997) 354. 103] J. Goniakowski and C. Noguera, Phys. Rev. B, 60 (1999) 16120. 104] G. W. Watson J. Chem. Soc. Faraday Trans. II, 92 (1996) 433. 105] M. Baudin, M. Wojcik and K. Hermansson, Surf. Sci., 375 (1997) 374.
89 106] K. Hermansson, M. Baudin, B. Ensing, M. Alfredsson and M. Wojcik, J. Chem. Phys., 109 (1998) 7515. 107] S. C. Parker, N. H. de Leeuw and S. E. Redfern, Faraday Discussion, 114 (1999) 381. 108] H. C. Galloway, P. Sautet and M. Salmeron, Phys. Rev. B, 54 (1996) R11145. 109] W. C. Mackrodt, Phil. Trans. Roy. Soc. London A, 341 (1992) 301. 110] T. J. Godin and J. P. LaFemina, Phys. Rev. B, 49 (1994) 7691. I l l ] C. Pisani, M. Causa, R. Dovesi and C. Roetti, Prog. Surf. Sci., 25 (1987) 119. 112] M. Causa, C. Pisani and C. Roetti, Surf. Sci., 215 (1989) 259. 113] V. E. Puchin, J. D. Gale, A. L. Schluger, E. A. Kotomin, J. Giinster, M. Brause and V. Kempter, Surf. Sci., 370 (1997) 190. 114] I. Manassidis and M. J. Gillan, Surf. Sci., 285 (1993) L517. 115] I. Manassidis and M. J. Gillan, J. Am. Ceram. Soc, 77 (1994) 335. 116] C. Verdozzi, D. R. Jennison, P. A. Schultz and M. P. Sears, Phys. Rev. Letters, 82 (1999) 799. 117] C. Rehbein, N. M. Harrison and A. Wander, Phys. Rev. B, 54 (1996) 14066. 118] C. Rehbein, F. Michel, N. M. Harrison and A. Wander, Surf. Rev. Letters, 5 (1998) 337. 119] P. J. Lawrence, S. C. Parker and P. W. Tasker, J. Am. Ceram. Soc, 71 (1988) C389. 120] W. C. Mackrodt, R. J. Davey, S. N. Black and R. Docherty, J. Cryst. Growth, 88 (1987) 441. 121] E. Wasserman, J. R. Rustad, A. R. Felmy, B. P. Hay and J. W. Halley, Surf. Sci., 385 (1997) 217. 122] X. G. Wang, W. Weiss, S. K. Shaikhutdinov, M. Ritter, M. Petersen, F. Wagner, R. Schlogl and M. Scheffler, Phys. Rev. Letters, 81 (1998) 1038. 123] P. Guenard, G. Renaud, A. Barbier and M. Gautier-Soyer, Mater. Res. Soc Symp. P r o c , 437 (1996) 15. 124] P. Guenard, G. Renaud, A. Barbier and M. Gautier-Soyer, Surf. Rev. Letters, 5 (1998) 321. 125] J. Ahn and J. W. Rabelais, Surf. Sci., 388 (1997) 121. 126] M. Gautier-Soyer, F. Jollet and C. Noguera, Surf. Sci., 352-354 (1996) 755. 127] I. Batyrev, A. Alavi and M. W. Finnis, Faraday Discussion, 114 (2000) 33. 128] X. G. Wang, A. Chaka and M. Scheffler, Phys. Rev. Letters, 84 (2000) 3650. 129] M. Bender, D. Ehrlich, I. N. Yakovkin, F. Rohr, M. Baumer, H. Kuhlenbeck, H. J. Freund and V. Staemmler, J. Phys. Condensed Matter, 7 (1995) 5289. 130] J. A. Mejias, V. Staemmler and H. J. Freund, J. Phys. Condensed Matter, 11 (1999) 7881. 131] R. M. Jaeger, H. Kuhlenbeck, H. J Freund, M. Wuttig, W. Hoffmann, R. Franchy and H. Ibach, Surf. Sci., 259 (1991) 235. 132] J. Libuda, F. Winkelmann, M. Baumer,H. J Freund, T. Bertrams, H. Neddermeyer and K. Muller, Surf. Sci., 318 (1994) 61. 133] H. Kuhlenbeck and H. J. Freund in: D. A. King, D. P. Woodruff (Eds.) Growth and properties of ultrathin epitaxial layers. The Chemical Physics of Solid Surfaces, Vol. 8, Elsevier Amsterdam 1997. 134] C. Becker, J. Kandler, H. Raaf, R. Linke, T. Pelster, M. Drager, M. Tanemura and K. Wandelt, J. Vac. Sci. Technol. A, 16 (1998) 1000. 135] P. J. Chen and D. W. Goodman, Surf. Sci., 312 (1994) L767. 136] Y. T. Wu, E. Garfunkel and T. E. Madey, Surf. Sci., 365 (1996) 337.
90 [137] D. R. Jennison, C. Verdozzi, P. A. Schultz and M. P. Sears, Phys. Rev. B, 59 (1999) R15605. [138] J. Purton, D. W. Bullett, P. M. Oliver and S. C. Parker, Surf. Sci., 336 (1995) 166. [139] A. Stashans, S. Lunell, R. Bergstrom, A. Hagfeldt, S. E. Lindquist, Phys. Rev. B, 53 (1996) 159. [140] P. K. Shelling, N. Yu and J. W. Halley, Phys. Rev. B, 58 (1998) 1279. [141] P. Reinhardt and B. A. Hess, Phys. Rev. B, 50 (1994) 12015. [142] W. C. Mackrodt, E. A. Simson and N. M. Harrison, Surf. Sci., 384 (1997) 192. [143] M. Ramamoorthy, D. Vanderbilt and R. D. King-Smith, Phys. Rev. B, 49 (1994) 16721. [144] D. Vogtenhuber, R. Podloucky, A. Neckel, S. G. Steinemann and A. J. Freeman, Phys. Rev. B, 49 (1994) 2099. [145] J. Goniakowski, J. M. Holender, L. N. Kantorovich, M. J. Gillan and J. A. White, Phys. Rev. B, 53 (1996) 957. [146] P. J. D. Lindan, N. M. Harrison, M. J. Gillan and J. A. White, Phys. Rev. B, 55 (1997) 15919. [147] S. P. Bates, G. Kresse and M. J. Gillan, Surf. Sci., 385 (1997) 386. [148] G. Charlton, P. B. Howes, C. L. NickUn, P. Steadman, J. S. G. Taylor, C. A. Muryn, S. P. Harte, J. Mercer, R. McGrath, D. Norman, T. S. Turner and G. Thornton, Phys. Rev. Letters, 78 (1997) 495. [149] E. Asari, T. Suzuki, H. Kawanowa, J. Ahn, W. Hayami, T. Aizawa and R. Souda, Phys. Rev. B, 61 (2000) 5679. [150] S. Munnix and M. Schmeits, Phys. Rev. B, 30 (1984) 2202. [151] A. E. Taverner, P. C. Hollamby, P. S. Aldridge, R. G. Egdell and W. C. Mackrodt, Surf. Sci., 287-288 (1993) 653. [152] R. V. Kasowski and R. H. Tait, Phys. Rev. B, 20 (1979) 5168. [153] M. Tsukada, C. Satoko and H. Adachi, J. Phys. Soc. Jpn., 47 (1979) 1610. [154] K. O. Ng and D. Vanderbilt, Phys. Rev. B, 56 (1997) 10544. [155] A. J. Paxton and L. Thien-Nga, Phys. Rev. B, 57 (1998) 1579. [156] S. Kimura and M. Tsukada, Appl. Surf. Sci., 130-132 (1998) 587. [157] T. Albaret, F. Finocchi, A. De Vita and C. Noguera, to be published [158] P. M. Oliver, S. C. Parker, J. Purton and D. W. Bullett, Surf. Sci., 307-309 (1994) 1200. [159] P. J. D. Lindan, N. M. Harrison, J. M. Holender, M. J. Gillan and M. C. Payne, Surf. Sci., 364 (1996) 431. [160] J. Muscat, N. M. Harrison and G. Thornton, Phys. Rev. B, 59 (1999) 2320. [161] P. J. D. Lindan, J. Muscat, S. Bates, N. M. Harrison and M.J. Gillan, Faraday Discussion, 106 (1997) 135. [162] M. Ramamoorthy, R. D. King-Smith and D. Vanderbilt, Phys. Rev. B, 49 (1994) 7709. [163] H. Onishi and Y. Iwasawa, Phys. Rev. Letters, 76 (1996) 791. [164] P. A. Mulheran and J. H. Harding, Model. Simul. Mater. Sci. Engin., 1 (1992) 39. [165] T. J. Godin and J. P. LaFemina, Phys. Rev. B, 47 (1993) 6518. [166] L Manassidis, J. Goniakowski, L. N. Kantorovich and M. J. Gillan, Surf. Sci., 339 (1995) 258. [167] D. H. Lee and J. D. Joannopoulos, J. Vac. Sci. Technol., 17 (1980) 309. [168] I. Ivanov and J. Pollmann, Phys. Rev. B, 24 (1984) 7275. [169] Y. R. Wang and C. B. Duke, Surf. Sci., 192 (1987) 309. [170] A. J. Skinner and J. P. LaFemina, Phys. Rev. B, 45 (1992) 3557.
91 [171] [172] [173] [174] [175] [176] [177]
J. E. Jaffe, N. M. Harrison and A. C. Hess, Phys. Rev. B, 49 (1994) 11153. P. Schroer, P. Kriiger and J. PoUmann, Phys. Rev. B, 49 (1994) 17092. A. Wander and N. M. Harrison, Surf. Sci. 457 (2000) L342. C. B. Duke, R. J. Meyer, A. Paton and P. Mark, Phys. Rev. B, 18 (1978) 4225. M. F. Chung and H. E. Farnsworth, Surf. Sci., 22 (1970) 93. J. D. Levine, A. Willis, W. R. Bottoms and P. Mark, Surf. Sci., 29 (1972) 144. L. Fiermans, E. Arijs, J. Vennik and W. Maenhout-van der Vorst, Surf. Sci., 39 (1973) 357. [178] H. van Hove and R. Leysen, Phys. Stat. Sol. A, 9 (1972) 361. [179] R. Leysen, G. van Orshaegen, H. van Hove and A. Neyens, Phys. Stat. Sol. A, 18 (1973) 613. [180] S. C. Chang and P. Mark, Surf. Sci., 46 (1974) 293. [181] R. W. Nosker, P. Mark and J.D. Levine, Surf. Sci., 19 (1970) 291. [182] M. Tsukada, E. Miyazaki and H. Adachi, J. Phys. Soc. Jpn., 50 (1981) 3032. [183] R. Kuwabara, H. Adachi and T. Morimoto, Surf. Sci., 193 (1988) 271. [184] J. B. L. Martins, E. Longo, J. Andres, Int. J, Quantum. Chem.: Quantum Chem. Symposium, 27 (1993) 643. [185] A. Gibson, unpublished [186] C. Noguera, to be published [187] N. Jedrecy, M. Sauvage-Simkin and R. Pinchaux, Appl. Surf. Sci. 162 (2000) 69. [188] J. Goniakowski and C. Noguera, Surf. Sci., 365 (1996) L657. [189] J. Prade, U. Schroder, W. Kress, F. W. de-Wette and A. D. Kulkarni J. Phys. Condensed Matter, 5 (1993) 1. [190] E. Heifets and E. A. Kotomin, Thin Solid Films, 358 (2000) 1. [191] J. Padilla and D. Vanderbilt, Surf. Sci., 418 (1998) 64. [192] Z. Q. Li, J. L. Zhu, C. Q. Wu, Z. Tang and Y. Kawazoe, Phys. Rev. B, 58 (1998) 8075. [193] C. Cheng, K. Kunc and M. H. Lee, Phys. Rev. B, 62 (2000) 10409. [194] N. Bickel, G. Schmidt, K. Heinz and K. Muller, Phys. Rev. Letters, 62 (1989) 2009. [195] T. Hikita, T. Hanada, M. Kudo and M. Kawai, J. Vac. Sci. Technol. A, 11 (1993) 2649. [196] T. Hikita, T. Hanada, M. Kudo and M. Kawai, Surf. Sci., 287-288 (1993) 377. [197] T. Wolfram, R. Hurst and F. J. Morin, Phys. Rev. B, 7 (1973) 1677. [198] T. Wolfram, R. Hurst and F. J. Morin, Phys. Rev. B, 15 (1977) 1151. [199] M. Tsukada, C. Satoko and H. Adachi, J. Phys. Soc. Jpn., 48 (1980) 200. [200] S. Kimura, J. Yamauchi, M. Tsukada and S. Watanabe, Phys. Rev. B, 51 (1995) 11049. [201] R. A. Powell and W. E. Spicer, Phys. Rev. B, 13 (1976) 2601. [202] V. E. Henrich, G. Dresselhaus and H. J. Zeiger, Phys. Rev. B, 17 (1978) 4908. [203] R. Courths, B. Cord and H. Saalfeld, Sol. State Com., 70 (1989) 1047. [204] T. Matsumoto, H. Tanaka, T. Kawai and S. Kawai, Surf. Sci., 278 (1992) L153. [205] H. Tanaka, T. Matsumoto, T. Kawai and S. Kawai, Jpn. J. Appl. Phys. Parti, 32 (1993) 1405. [206] H. Tanaka, T. Matsumoto, T. Kawai and S. Kawai, Surf. Sci., 318 (1994) 29. [207] T. Matsumoto, H. Tanaka, K. Kouguchi, T. Kawai and S. Kawai, Surf. Sci., 312 (1994) 21. [208] A. Kubo, H. Tanaka, H. Tabata,T. Matsumoto and T. Kawai, Jpn. J. Appl. Phys. Part2, 35 (1996) L1692.
92 [209] [210] [211] [212] [213]
J. E. T. Andersen and P. J. M0ller, Appl. Phys. Lett., 56 (1990) 1847. S. Kimura and M. Tsukada, Appl. Surf. Sci., 121-122 (1997) 195. S. Ferrer and G. A. Somorjai, Surf. Sci., 94 (1980) 41. H. Tanaka and T. Kawai, Surf. Sci., 365 (1996) 437. W. M. Sigmund, M. Rotov, Q. D. Jiang, J. Brunen, J. Zegenhagen and F. Aldinger, Appl. Phys. A, 64 (1997) 219. [214] S. Sekiguchi, M. Fujimoto, M. G. Kang, S. Koizumi, S. B. Cho and J. Tanaka, Jpn. J. Appl. Phys., 37 (1998) 4140. [215] Y. Aiura, H. Bando, Y. Nishihara, Y. Haruyama, S. Kodaira, T. Komeda, Y. Sakisaka, T. Maryuama and H. Kato, Advances in Superconductivity, VI, Springer, Tokyo, 1994. [216] H. Bando, Y. Aiura, Y. Haruyama, T. Shimizu and Y. Nishihara, J. Vac. Sci. Technol. B, 13 (1995) 1150. [217] J. Brunen and J. Zegenhagen, Surf. Sci., 389 (1997) 349. [218] C. Hagendorf, K. M. Schindler, T. Doege and H. Neddermeyer, Surf. Sci., 402 (1998) 581. [219] A. Pojani, F. Finocchi and C. Noguera, App. Surf. Sci., 142 (1999) 177. [220] A. Pojani, F. Finocchi and C. Noguera, Surf. Sci., 442 (1999) 179. [221] S. K. Shaikhutdinov, M. Ritter, X. G. Wang, H. Over and W. Weiss, Phys. Rev. B, 60 (1999) 11062. [222] J. Ahdjoudj, C. Martinsky, C. Minot, M. A. Van Hove and G. A. Somorjai, Surf. Sci., 443 (1999) 133. [223] A. Barbieri, W. Weiss, M. A. Van Hove and G. A. Somorjai, Surf. Sci., 302 (1994) 259. [224] N. H. de-Leeuw, F. Manon-Higgins and S. C. Parker, J. Phys. Chem. B, 103 (1999) 1270. [225] M. Witko, K. Hermann and R. Tokarz, J. El. Spectr. Rel. Phenom., 69 (1994) 89. [226] A. Michalak, M. Witko and K. Hermann, Surf. Sci., 375 (1997) 385. [227] A. Chakrabarti, K. Hermann, R. Druzinic, M. Witko, F. Wagner and M. Petersen, Phys. Rev. B, 59 (1999) 10583. [228] K. Hermann, M. Witko, R. Druzinic, A. Chakrabarti, B. Tepper, M. Eisner, A. Gorschluter, H. Kuhlenbeck and H. J. Freund, J. El. Spectr. Rel. Phenom., 98-99 (1999) 245. [229] K. Hermann, M. Witko and R. Druzinic, Faraday Discussion, 114 (2000) 53. [230] A. Michalak, K. Hermann and M. Witko, Surf. Sci., 366 (1996) 323. [231] R. D. Shannon, Acta Cryst. A, 32 (1976) 751. [232] E. J. W. Verwey, Reel. Trav. Chim., 65 (1946) 521. [233] F. W. de-Wette, W. Kress and U. Schroder, Phys. Rev. B, 32 (1985) 4143. [234] M. D. Towler, N. M. Harrison and M. I. McCarthy, Phys. Rev. B, 52 (1995) 5375. [235] G. V. Lewis and C. R. A. Catlow, J. Phys. C, Solid State Phys., 18 (1985) 1149. [236] M. Lannoo and P. Friedel, Atomic and electronic structure of surfaces. Springer Series in Surface Science, Vol. 16, Springer Verlag, Berlin, Heidelberg, New York, 1991. [237] D. J. Chadi, Phys. Rev. Letters, 43 (1979) 43. [238] G. Jura and C. W. Garland, J. Am. Chem. Soc, 74 (1952) 6033. [239] D. T. Livey and P. Murray, J. Am. Ceram. Soc, 39 (1956) 363. [240] R. F. W. Bader, Atoms in molecules: A Quantum Theory, Oxford University Press, Oxford, 1990. [241] R. F. W. Bader, Chem. Rev., 91 (1991) 983.
93 [242' [243^ [244^ [245^ [246' [247'
C. Sousa, F. Illas, C. Bo and J. M. Poblet, Chem. Phys. Letters, 215 (1993) 97. T. Albaret, F. Finocchi, C. Noguera, Faraday Discussion, 114 (1999) 285. C. Noguera, A. Pojani and F. Finocchi, to be published. J. Zaanen, G. A. Sawatzky and J. W. Allen, Phys. Rev. Letters, 55 (1985) 418. ] J. F. Janak, Phys. Rev. B,18 (1978) 7165. J. J. M. Pothuizen, O. Cohen and G. A. Sawatzky, in: J. S. Speck, D. K. Fork, R. M. Wolf and T. Shiosaki (Eds.), Epitaxial Oxide Thin Films II, MRS Proceedings 401 (1996) 501 [248' J. Goniakowski and C. Noguera, Surf. Sci., 340 (1995) 191. [249^ W. C. Mackrodt, Her. Bunsenges. Phys. Chem., 101 (1997) 169. [250 R. Hesper, L. H. Tjeng and G. A. Sawatzky, Europhys. Letters, 40 (1997)177. [251 W. C. Mackrodt, N. M. Harrison, V. R. Saunders, N. L. Allan and M. D. Towler, Chem. Phys. Letters, 250 (1996) 66. [252 V. Propach, D. Reinen, H. Drenkhaln and H. Miiller Buschbaum, Z. Naturforsch. B, 33 (1978) 619. [253 A. L. Shluger, R. W. Grimes, C. R. A. Catlow and N. Itoh, J. Phys. Condensed Matter, 3 (1991) 8027. [254 P. V. Sushko and A. L. Shluger, Surf. Sci., 421 (1999) L157. [255 A. L. Shluger, P. V. Sushko and L. N. Kantorovich, Phys. Rev. B, 59 (1999) 2417. [256 G. Pacchioni, N. Lopez and F. Illas , Faraday Discussion, 114 (2000) 209. [257 M. C. Desjonqueres and D. Spanjaard, Concepts in Surface Physics, Springer Series in Surface Science, Vol. 30, Springer Verlag, Berlin, Heidelberg, New York, 1993. [258" A. Kahn, Surf. Rev. Letters, 3 (1996) 1579. [259^ P.W. Tasker, J. Phys. C: Solid State Physics, 12 (1979) 4977. [260^ H. H. Farrell, J. P. Harbison and L. D. Peterson, J. Vac. Sci Technol. B, 5 (1987) 1482. [261 M. D. Pashley, Phys. Rev. B, 40 (1989) 10481. [262^ W. A. Harrison, J. Vac. Sci Technol., 16 (1979) 1492. [263^ J. P. LaFemina, in: W. N. Unertl (Ed.), Handbook of Surface Science I, Elsevier, 1996 , Ch 4, 137. [264 Y. J. Kim, Y. Gao and S. A. Chambers, Surf. Sci., 371 (1997) 358. [265 S. A. Chambers and S. A. Joyce, Surf. Sci., 420 (1999) 111. [266 T. Fujii, F. M. F. de Groot, G. A. Sawatzky, F. C. Voogt, T. Hibma and K. Okada, Phys. Rev. B, 59 (1999) 3195. [267 F. C. Voogt, T. Fujii, P. J. M. Smulders, L. Niesen, M. A. James and T. Hibma, Phys. Rev. B, 60 (1999) 11193. [268 W. F. J. Fontijn, R. M. Wolf, R. Metselaar and P. J. van der Zaag, Thin Solid Films, 292 (1997) 270. [269 Y. Gao and S. A. Chambers, J. Cryst. Growth, 174 (1997) 446. [270| P. M. Oliver, S. C. Parker, W. C. Mackrodt, Modelling Simul. Mater. Sci. Eng., 1 (1993) 755. [271 C. Noguera, J. Adh. 57 (1996) 91. [272 F. Finocchi, C. Noguera, J. Adh. under press
94
Oxide Surfaces D.P. Woodruff, editor © 2001 Elsevier Science B. V. All rights reserved.
Chapter 3
Theory of point defects at the MgO surface Gianfranco Pacchioni Dipartimento di Scienza dei Material!, Universita di Milano-Bicocca, Istituto Nazionale per la Fisica della Materia, via Cozzi, 53 - 20125 Milano, Italy
1. INTRODUCTION Oxides surfaces are finding continuous new applications in advanced technologies like in corrosion protection, coating for thermal applications, in catalysis as inert supports or directly as catalysts, in microelectronics for their dielectric properties; films of magnetic oxides are integral components in magnetic recording devices and many microporous materials are based on oxides. For all these reasons there is a considerable effort to better characterize the surface and the interface of oxide materials [1,2]. New experimental methods have been developed to growth well defined films of oxides on metal supports in UHV conditions. This approach allows in principle to overcome some difficulties connected to the use of electron spectroscopies for the study of insulating materials, as most of the oxides are. Recently, some very good review articles have been published on this subject [3-10]. It is no surprise that the increasing experimental activity has stimulated a parallel computational activity based on high-quality first principle calculations. One important information coming from calculations is the structure of the oxide surface. Oxide surfaces are often heavily reconstructed or simply relaxed compared to the truncated bulk, and the experimental determination of the surface structure is often not easy. To this end, reliable classical potentials have been developed, in particular for the study of ionic crystals and of covalent solids. Nowadays, first principle band structure calculations making use of large supercells can also be used. These methods, although quite expensive from the point of view of the size of the calculations, provide results which are in excellent agreement with the experimental determinations. Band structure calculations, usually based on plane waves basis sets and on the density functional (DFT) approach represent the most appropriate computational
95
method to study solids. However, an alternative approach to band structure has been developed in the past two decades with the aim of describing in chemical terms interactions occurring at surfaces. This is known as the cluster approach [11] and it has been applied to a variety of problems in surface science. As it will be explicitly discussed below, it has advantages and disadvantages with respect to band structure supercell methods. One of the problems where the use of the cluster approach is more appealing is in the study of surface defects. Only recently it has been recognized by the surface science community that these centers are often the most interesting ones from the point of view of the physical and chemical properties of a material. Several chemical reactions taking place at an oxide surface are directly or indirectly connected to the presence of point and extended defects. Unfortunately, defect centers are elusive species because of their low concentration even in the bulk material, and their identification by spectroscopic methods can be rather difficult. This is even more dramatic when one is interested in surface defects because the distinction from bulk defects may be extremely subtle. For all these reasons the theoretical modeling of defect centers at the surface of oxides is attracting an increasing interest. In this review we will give a general overview of the most common defects occurring at the surface of ionic oxides and in particular of MgO, an ionic material which has been at the center of considerable attention in the last years as a model system, but also because of its potential and technological applications in catalytic processes. 2. QUANTUM CHEMICAL DESCRIPTION OF IONIC CRYSTALS The electronic structure of solids is usually described in terms of band structure. To this end, a unit cell containing a given number of atoms is periodically repeated in three-dimensions to account for the "infinite" nature of the crystalline solid and the Schrodinger equation is solved for the atoms in the unit cell subject to periodic boundary conditions [12]. This approach can also be extended to the study of point defects in the bulk or at the surface, although the ab initio simulation of defects in crystals is not simple. One should in fact include a "reasonable" description of the "infinite" host system while accounting for the rupture of the translational symmetry induced by the defect. One way to circumvent this problem is the supercell approach in which an artificial periodic structure is created where the defect is translationally reproduced in correspondence to a given super-lattice of the host. This procedure allows the use of efficient computer programs designed for the treatment of periodic systems and has indeed been followed by several authors to study defects using either DFT and plane waves approaches [13-15] or Hartree-Fock-based methods with localized atomic orbitals [16,17].
96
The presence of the defect in the unit cell, however, results in a periodic repetition of the defect in the three directions of space, hence in an extremely high concentration. The only way to reduce the defect concentration is to increase the size of the unit cell, a solution which implies a very large computational cost. Nowadays, periodic calculations for supercells containing several tens of atoms are routinely done. Even for large supercells containing -100 atoms, however, the defect concentration is orders of magnitude larger than in real systems where 10^^-10^^ defects/cm^ are usually present [18,19]. At these concentrations one can reasonably assume that the defects are isolated and do not interact with each other. The supercell approach is therefore based on the assumption, largely verified, that neutral defects do not interact appreciably except when they are very close to each other [20], so that rapid convergence is achieved with increasing size of the supercell. With charged defects the supercell approach is feasible but less reliable because of the long range Coulomb interaction between the defects [21]. Methods to include correction terms to account for these spurious interactions have been proposed [22]. 2.1 Cluster models An alternative approach to the periodic band structure methods to study solids is the cluster approach [11]. Here one considers explicitly only a finite number of atoms to describe a section of the solid while the rest of the crystal (or of the amorphous material) is treated in a more or less simplified way (embedding). The main conceptual difference is that in the cluster approach one uses molecular orbitals, MO, instead of delocalized bands. The description of the electronic properties is thus done in terms of local orbitals, allowing one to treat problems in solids with the typical language of chemistry, the language of orbitals. This is particularly useful when dealing with surface problems and with the reactivity of a solid surface. In fact, the interaction of gas-phase molecules with a solid surface can be described in exactly the same way as the interaction of two molecules. Of course, also the cluster model is not free from limitations. The most serious one is that the effect of the surrounding is often taken into account in a more or less approximate way, thus leading to some uncertainties in the absolute values of the computed quantities. It is also possible that some properties are described differently depending on the size of the cluster used. It is therefore necessary to check the results versus cluster size and shape. Examples of this aspect will be discussed in more detail below. The advantages, beside a smaller computational cost, are that in describing point defects an infinite dilution is considered so that no mutual interaction of the defects is present in the model and that theoretical methods derived from quantum chemistry can be applied. This latter is an important advantage which should not be underestimated. In fact, in this way it is possible (a) to explicitly include correlation effects in the calculations through, for instance, a configuration
97
interaction (CI) procedure (see below) and (b) to treat exactly the non-local exchange as in the Hartree-Fock formalism; in DFT in fact the exchange is taken into account in an approximate way through the exchange-correlation functional. We briefly mention here two examples where the use of quantumchemical techniques is extremely important for the correct description of the electronic structure of point defects. The first one is the calculation of excited state properties like for instance optical transitions and intensities, and therefore optical spectra. While some examples of excited-state calculations based on time-dependent DFT are now available [23], the application to solid state problems is still in its infancy. The use of cluster models and CI methods represents a valid approach to the study of excited states of point defects in sohd matrices, and excellent results have been obtained in recent years for the study of optical spectra in Si02 [24-27] and MgO [28,29]. The second problem is that of spin localization in paramagnetic defects. Recent DFT calculations based on supercells and plane waves have described the [A104]° center in silica, an Al impurity replacing a four-coordinated Si atom in the Si02 lattice, as a tetrahedrally coordinated Al atom with a hole delocalized over the four O neighbors [30-32]. This description contradicts the experimental evidence from ESR studies of a hole completely localized on a single O atom [33], a physical picture supported by HF cluster calculations [34,35]. The reason for the incorrect description of the center in DFT is the lack of self-interaction correction and the tendency to delocalize the hole. Recent results obtained at various levels of theory have clearly shown the important role of exact exchange for the proper description of this center [36]. Therefore, cluster calculations represent an alternative way of describing localized defects in ionic crystals. The problem is to introduce in a reasonable way the effect of the rest of the crystal. Completely different strategies can be adopted to "embed" clusters of largely covalent oxides, like Si02, or of very ionic oxides, like MgO. In Si02 and related materials the cluster dangling bonds are usually saturated by H atoms [37]. The positions of the cluster atoms can be initially those of one crystalline phase; then the H atoms are kept fixed while the position of all the other atoms of the cluster are reoptimized. The fixed H atoms provide a simple representation of the mechanical embedding of the solid matrix [24-26]. As an alternative, fully relaxed clusters have also been used [27]. It is believed that these latter systems are better models of the amorphous phase of the material. The saturation of the dangling bonds with H atoms is an important aspect of the embedding, but not the only one. In fact, in this way one neglects the crystalline Madelung field. While this term is of moderate importance in silica and other dielectrics, it may play a role in the description of charged defects and becomes essential in solids with more pronounced ionic character.
98
2.2 Embedding schemes The very ionic nature of MgO implies that the Madelung potential is explicitly included. Indeed, several properties of MgO are incorrectly described if the long range Coulomb interactions are not taken into account [38]. A simple approach is to surround the cluster of Mg and O ions by a large array of point charges (PC) ±2 to reproduce the Madelung field of the host at the central region of the cluster [39]. However, the PC's polarize the oxide anions at the cluster border and cause an incorrect behavior of the electrostatic potential [40]. The problem can be substantially reduced by placing at the position of the +2 PC's around the cluster an effective core potential, ECP, representing the finite size of the Mg^"^ core [41]. No basis functions are associated to the ECP [42] which accounts for the Pauli or exchange repulsion of the O^" valence electrons with the surrounding. This is a simplified approach to the more rigorous ab initio model potential (AIMP) method [43,44] but is computationally simple and reliable. In the AIMP approach the grid of bare charges is replaced by a grid of AIMP's which account not only for the long-range Coulomb interaction but also for the quantum mechanical short-range requirements of exchange and orthogonality without introducing explicitly extra electrons in the model. The addition of the ECP's to the cluster results in a better representation of the electrostatic potential and in substantially different stabilities of the charge states of the defect. For instance, the ionization potentials of oxygen vacancies (the F centers described in the next section) are lowered by about 3 eV when the cluster is embedded in ECP+PC's. What is still missing from this simplified approach is the polarization of the host crystal by the defect. This effect can be particularly important for charged defects. The polarization, Epoi, induced by a charge on the surrounding lattice can be estimated by means of the classical Bom formula [45]: Epoi= - ( l - l / 8 ) q V 2 R
(1)
where 8 is the dielectric constant of MgO, q is the absolute value of the charge and R is the radius of the spherical cavity where the charge is distributed. Since a certain degree of ambiguity remains in the definition of R, this correction is only qualitative. A more refined approach which has been used for the study of the ground state of F centers in MgO [46] makes use of the ICECAP program [47]. In this approach instead of PC's the cluster is surrounded by polarizable ions described according to the shell model [48,49]; in this way the polarization response of the host is taken into account self-consistently up to infinite distance. In the shell model an ion is represented by a point core and a shell connected by a spring to simulate its dipole polarizability. A similar method has been applied recently to the study of energy states of defect sites at the MgO
99
surface [50]. A quantum cluster has been embedded in a finite array of PC's placed at the lattice sites, Fig. 1. The part of the ions closest to the quantum cluster has been treated by the shell model in such a way that they interact among themselves and the quantum cluster via specific interatomic potentials. The novelty of the approach lies in the fact that the positions of the cores and shells of the ions are optimized in response to the changes in charge density distribution within the quantum cluster to minimize the total energy of the system [50]. Quantum Cluster
a),
Fixed ions (Region II)
Shell model (Region I)
Fig. 1. Clusters of surface and comer sites of the MgO surface with embedding in shell models and point charges. Reproduced from ref. [50]. Copyright 2000 Elsevier.
An alternative, more rigorous approach has been developed in recent years by Pisani and coworkers [51-53]. It is named perturbed cluster method and is based on the EMBED computer program [54]. With this approach the properties of neutral and charged defects at the surface of MgO have been studied at the HE and MP2 levels [55]. The method relies on the knowledge of the one-electron Green function G^ for the unperturbed host crystal which is obtained by means of the periodic program CRYSTAL [56,57]. A cluster (C) containing the defect is defined with respect to the rest of the host (H). The molecular solution for the cluster C in the field of H is corrected selfconsistently by exploiting the information contained in G^ in order to allow a proper coupling of the local wave function to that of the outer region. 2.3 Solutions of the Schrodinger equation Two general groups of methodologies are used to solve the Schrodinger equation in combination with cluster models, the Hartree-Fock (HE) approach and related methods to include correlation effects like M0ller-Plesset perturbation theory (MP2) or configuration interaction (CI) [58,59] and the Density Eunctional Theory (DET) approach [59,60].
100
In HF theory the energy has the form EHF = ENUCL +
+ 1/2 - 1/2
(2)
where ENUCL is the nuclear repulsion energy, P is the density matrix, is the one electron energy including kinetic and potential energy terms, 1/2 is the classical coulomb repulsion term and -1/2 is the exchange energy resulting from the quantum nature of the electrons. Thus, the HF method does not include electron correlation which must be included explicitly through perturbation theory at the second (MP2) or higher order (MPS and MP4) or through configuration interaction (CI) techniques. In the Kohn-Sham formulation of DFT the exact HF exchange for a single determinant is replaced by a general expression, the exchange correlation functional, which can include both exchange and electron correlation energy terms: EKS
= ENUCL + + 1/2 + Ex[P] + Ec[P]
(3)
where Ex[P] is the exchange functional and Ec[P] is the correlation functional. HF can be seen as a special case of DFT where Ex[P] is given by the exchange integral -1/2 and Ec = 0. In this respect cluster models offer the opportunity to use the so-called hybrid approaches where the exchange interaction is described partially by the HF method. For example, in the popular B3LYP hybrid method the HF exchange is mixed in with the DF exchange while the correlation, Ec[P], is treated by the Lee, Yang and Parr, LYP, correlation functional [61]. In particular, the exchange treatment is based on the Becke 3-parameters approach [62] where a set of parameters is derived in order to accurately describe the thermochemistry of a large number of molecules. While this introduces some empiricism in a first principle theory, it has been shown that the B3LYP exchange-correlation functional describes in a much more accurate way the energetics of reactions at surfaces than most of the commonly used exchange-correlation functional. This has been proved in particular for the adsorption of metal atoms on MgO [63,64]. However, exceptions are also known and in general the choice of the exchange-correlation functional remains an open problem. The main difference between HF and DFT is that in the HF approach the objective is the determination of the wave function of the system while in DFT the quantity of interest is the charge density p; in fact, the Hohenberg-Kohn theorem [65] ensures that the energy E of a system is a functional of p which takes its minimum value EQ for the ground state density po. Despite this profound conceptual difference, the two methods, HF+correlation and DFT, provide in ultimate analysis very similar results and the choice of one approach or the other is largely matter of convenience.
101
3. ROLE AND NATURE OF DEFECTS AT THE MgO SURFACE Often the most important properties of materials are directly or indirectly connected to the presence of defects and in particular of point defects [18]. These centers determine the optical, electronic and transport properties of the material and usually dominate the chemistry of its surface. A detailed understanding and a control at atomistic level of the nature (and concentration) of point defects in oxides is therefore of fundamental importance to synthesize new materials with well defined properties. This has lead in recent years to the birth of the new field of defect engineering. Of course, before to be created in controlled conditions point defects have to be known in all aspects of their physico-chemical properties. The accurate theoretical description of the electronic structure of point defects in oxides is essential for the understanding of their structure-properties relationship. MgO is a particularly well studied oxide; the structure of the (100) single crystal surface is extremely flat, clean, and stoichiometric. Recent grazing incident X-ray scattering experiments have shown that both relaxation, -0.56±0.4%, and rumpling, 1.07±0.5%, are extremely small [66]. However, no real crystal surface consists of only idealized terraces. A great effort has been undertaken in recent years to better characterize the MgO surface, in particular for polycrystalline or thin-film forms which in some cases exhibit an heterogeneous surface, due to the presence of various sites. All these sites can be considered as defects. The identification and classification of the defects is of fundamental importance. In fact, the presence of appreciable concentrations of defects can change completely the chemical behavior of the surface. A typical example is that of the reaction of CO on MgO (see § 3.1). Table 1 Summary of most important surface defects in MgO Defect Symbol Schematic description low-coordinated cation Mg'^nc (n=3,4) coordinatively unsaturated cation low-coordinated anion 0'-nc(n=3,4) coordinatively unsaturated anion hydroxyl group (OH) proton attached to O^' anion vacancy F"^-'nc(m=0,l,2;n=3,4,5) missing oxygen with trapped electrons cation vacancy V""-„c(m=0,l,2;n=3,4,5) missing cation with holes at O neighbors divacancy cation and anion vacancy VMgVo substitutional cation (M) or anion (X) impurity atoms M"VO"-; Mg^'-ZX^oxygen radical hole trapped at O anion 0-nc (n=3,4,5) small ensemble of Mg3c or Osc ions (111) microfacets none
102
While the single crystal (100) surface is completely unreactive, a defect-rich polycrystalline MgO surface exhibits a rich and complex chemistry when interacting with CO. In the following we provide a summary of the most important kinds of defects at the MgO surface, Table 1, and we discuss their electronic structure and chemical behavior, in view of the e