Solid State Materials Chemistry

This book explores the fascinating world of functional materials from the perspective of
those who are tasked with inventing them, solid state chemists. Written in a clear and
accessible style, this book provides a modern-day treatment of solid state materials chemistry
for graduate and advanced undergraduate level courses. With over 330 problems and 400
original figures, this essential reference covers a wide range of materials in a holistic manner,
including inorganic and organic, crystalline and amorphous, bulk and nanocrystals.
The introductory chapters cover topics such as crystal structures, defects, diffusion in
solids, chemical bonding, and electronic band structure. Later chapters focus on important
classes of functional materials including pigments, phosphors, dielectric materials, magnets,
metals, semiconductors, superconductors, nonlinear optical materials, battery materials,
zeolites, metal–organic framework materials, and glasses. The technological applications
and synthesis methods used to prepare the materials that drive modern society are high-
lighted throughout.

Patrick M. Woodward is a Professor in the Department of Chemistry and Biochemistry and

holds a courtesy appointment in the Department of Physics at Ohio State University. He is
best known for his studies of the structures and properties of perovskite-related materials.
He has served as chair of the Solid State Chemistry Gordon Conference (2018), Associate
Editor of the Journal of Solid State Chemistry (2006–2011), and Vice President of the
Neutron Scattering Society of America (2014–2018). He is co-author of the widely used
general chemistry textbook, Chemistry: The Central Science (Pearson Education Limited,
2018). Patrick is a recipient of an NSF Career Award (2001), a Sloan Research Fellowship
(2004), a Leverhulme Visiting Professorship (2017), and is a Fellow of the American
Chemical Society (2020).

Pavel Karen is a Professor in the Department of Chemistry at the University of Oslo. His
interests include inorganic reaction chemistry, solid state synthesis methods, crystallog-
raphy, phase relations and thermodynamics, and point-defect chemistry; all components
of his teaching portfolio. He is interested in the relationship between structure and properties
of less common inorganic solids, such as mixed-valence oxides. Crystal structures are studied
by X-ray and neutron diffraction, local structures by Mössbauer spectroscopy, and valence-
mixing by calorimetry. He is co-author of the chapter Phase Diagrams and Thermodynamic
Properties in the Handbook on the Physics and Chemistry of the Rare Earths, Volume 30,
High-Temperature Superconductors (Elsevier, 2000). Pavel is a member of the American
Chemical Society, the American Crystallographic Association, and of the International
Union of Pure and Applied Chemistry’s Division II and the Interdivisional Committee on
Terminology, Nomenclature and Symbols.

John S. O. Evans is a Chemistry Professor at Durham University where he served as Head of

Chemistry from 2009 to 2014. His research interests are in the synthesis and properties of
(mainly) inorganic materials, their structural chemistry, and their real-world applications. In
recent years he has worked, inter alia, on negative thermal expansion, symmetry properties
of phase transitions, new oxide-chalcogenides and energy-related materials. He has a long-
standing interest in developing powder diffraction methods and is co-author of Rietveld
Refinement: Practical Powder Diffraction Pattern Analysis using TOPAS (De Gruyter, 2019).
John was awarded the 1997 Meldola prize of the Royal Society of Chemistry, and was co-
awarded the 2015 Royal Society of Chemistry Teamwork in Innovation Award for work
with industry.

Thomas Vogt is the Educational Foundation Endowed Chair in the Department of

Chemistry and Biochemistry, Director of the NanoCenter and adjunct Professor in the
Department of Philosophy at the University of South Carolina. His work focuses on
establishing structure–property relationships of solid state materials using X-ray and neu-
tron scattering and electron microscopy. He is recognized as Fellow of the American
Physical Society, the American Association for the Advancement of Science, the Institute
of Advanced Study at Durham University, and the Neutron Scattering Society of America.
Thomas received the Carolina Trustee Professorship of the Board of Trustees in 2018 as well
as the University of South Carolina’s Educational Foundation Award for Research in
Science, Mathematics, and Engineering in 2019.
Solid State Materials Chemistry

Patrick M. Woodward
Ohio State University

Pavel Karen
Universitetet i Oslo

John S. O. Evans
Durham University

Thomas Vogt
University of South Carolina
University Printing House, Cambridge CB2 8BS, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre,
New Delhi – 110025, India
79 Anson Road, #06–04/06, Singapore 079906

Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9780521873253
DOI: 10.1017/9781139025348
© Patrick M. Woodward, Pavel Karen, John S. O. Evans, and Thomas Vogt 2021
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2021
Printed in the United Kingdom by TJ International Ltd, Padstow Cornwall
A catalogue record for this publication is available from the British Library.
ISBN 978-0-521-87325-3 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Contents

Preface page xvii

Acknowledgments xix

1 Structures of Crystalline Materials 1

1.1 Symmetry 1
1.1.1 Translational Symmetry 2
1.1.2 Rotational Symmetry 3
1.1.3 Crystallographic Point Groups and Crystal Systems 5
1.1.4 Bravais Lattices 5
1.1.5 Introduction to Space Groups 8
1.1.6 Symmetry Elements That Combine Rotation and Translation 9
1.1.7 Space-Group Symbols 11
1.1.8 Description of a Crystal Structure 12
1.2 Databases 13
1.3 Composition 14
1.3.1 Coordination, Stoichiometry, and Connectivity 15
1.3.2 The Generalized 8−N Rule 17
1.4 Structural Principles 18
1.4.1 Packing of Spheres 19
1.4.2 Filling Holes 22
1.4.3 Network Structures 28
1.4.4 Polyhedral Structures 32
1.5 Structures of Selected Materials 38
1.5.1 The Spinel Structure 38
1.5.2 The Garnet Structure 39
1.5.3 Perovskite Structures 40
1.5.4 Silicates 44
1.5.5 Zeolites 46
1.5.6 Zintl Phases 47
1.6 Problems 48
1.7 Further Reading 51
1.8 References 51

v
vi Contents

2 Defects and More Complex Structures 54

2.1 Point Defects in Crystalline Elemental Solids 54
2.2 Intrinsic Point Defects in Compounds 55
2.3 Thermodynamics of Vacancy Formation 58
2.4 Extrinsic Defects 61
2.5 Solid Solutions and Vegard’s Law 63
2.6 Kröger–Vink Notation 65
2.7 Line Defects in Metals 66
2.7.1 Edge Dislocations 66
2.7.2 Screw Dislocations 66
2.8 Planar Defects in Materials 67
2.8.1 Stacking Faults 67
2.8.2 Twinning 68
2.8.3 Antiphase Boundaries 72
2.8.4 Crystallographic Shear Structures 74
2.9 Gross Nonstoichiometry and Defect Ordering 75
2.10 Incommensurate Structures 78
2.11 Infinitely Adaptive Structures 80
2.12 Problems 81
2.13 Further Reading 85
2.14 References 85

3 Defect Chemistry and Nonstoichiometry 87

3.1 Narrow Nonstoichiometry in Oxides 87
3.1.1 Point Defects in a Pure Stoichiometric Oxide 87
3.1.2 Point Defects upon Oxidation/Reduction of the Stoichiometric
Oxide 88
3.1.3 Equilibrium Equations for Oxidative and Reductive
Nonstoichiometry 89
3.1.4 Defect Equilibria for Schottky-Type Redox Compensation 90
3.1.5 Acceptor-Doped Oxides 93
3.1.6 Donor-Doped Oxides 94
3.1.7 Solid Solubility of Dopants 94
3.1.8 Cautionary Note on Defect Models in Pure Oxides 96
3.2 Wide Nonstoichiometry in Oxides 98
3.3 Point Defects and Diffusion 99
3.3.1 Point-Defect Movements 101
3.3.2 Random Hopping 103
3.3.3 Hopping Under a Driving Force 104
 Contents vii

3.3.4 Hopping Under a Concentration Gradient 105

3.3.5 Hopping Under an Electric Field 106
3.3.6 Relationship between Conductivity and Diffusivity 107
3.3.7 Ambipolar Diffusion 108
3.3.8 Temperature Dependence of Diffusivity 111
3.3.9 Diffusivity and Redox Defect Equilibria 111
3.3.10 Outline of Non-Steady-State Diffusion 112
3.3.11 Cautionary Note on Diffusion in Real Materials 114
3.4 Problems 115
3.5 Further Reading 118
3.6 References 118

4 Phase Diagrams and Phase Transitions 120

4.1 Phase Diagrams 120
4.2 Two-Component Phase Diagrams 123
4.2.1 Without Compound Formation 123
4.2.2 With Compound Formation 125
4.2.3 Solid-Solution Formation 128
4.3 Three-Component Phase Diagrams 131
4.4 Structural Phase Transitions 135
4.4.1 Classification of Phase Transitions 136
4.4.2 Symmetry and Order Parameters 137
4.4.3 Introduction to Landau Theory 140
4.4.4 Second-Order Transitions 141
4.4.5 First-Order and Tricritical Transitions 144
4.4.6 Phonons, Soft Modes, and Displacive Transitions 147
4.5 Problems 150
4.6 Further Reading 152
4.7 References 153

5 Chemical Bonding 154

5.1 Ionic Bonding 154
5.1.1 Coulombic Potential Energy 154
5.1.2 Lattice Energy and the Born–Mayer Equation 156
5.1.3 Experimental versus Calculated Lattice-Formation Energies 158
5.2 Atomic Orbitals 161
5.2.1 Energies of Atomic Orbitals 166
5.2.2 Sizes of Atomic Orbitals 168
5.3 Molecular-Orbital Theory 169
5.3.1 Homonuclear Diatomics: H2+ and H2 169
5.3.2 The Heteronuclear Diatomic Case: HHe 173
viii Contents

5.3.3 Orbital Overlap and Symmetry 174

5.3.4 Combination of σ and π Bonding: O2 175
5.3.5 Symmetry-Adapted Linear Combinations (SALCs) 177
5.3.6 Simple Polyatomic Molecules: BeH2 and CH4 179
5.3.7 Conjugated π Bonding: C6H6 181
5.3.8 Transition-Metal Complexes: [CrCl6]3− and [CoCl4]2− 183
5.3.9 High- and Low-Spin Configurations 186
5.3.10 Jahn–Teller Distortions 188
5.4 Bond Valences 190
5.5 Problems 195
5.6 Further Reading 198
5.7 References 199

6 Electronic Band Structure 200

6.1 The Band Structure of a Hydrogen-Atom Chain 200
6.1.1 The Electronic Structures of Cyclic HN Molecules 201
6.1.2 Translational Symmetry and the Bloch Function 202
6.1.3 The Quantum Number k 203
6.1.4 Visualizing Crystal Orbitals 204
6.1.5 Band-Structure Diagrams 207
6.1.6 Density-of-States (DOS) Plots 209
6.2 The Band Structure of a Chain of H2 Molecules 210
6.3 Electrical and Optical Properties 213
6.3.1 Metals, Semiconductors, and Insulators 213
6.3.2 Direct- versus Indirect-Gap Semiconductors 214
6.4 Representing Band Structures in Higher Dimensions 215
6.4.1 Crystal Orbitals in Two Dimensions 215
6.4.2 Crystal Orbitals in Three Dimensions 219
6.5 Band Structures of Two-Dimensional Materials 220
6.5.1 Graphene 221
6.5.2 CuO22− Square Lattice 223
6.6 Band Structures of Three-Dimensional Materials 227
6.6.1 α-Polonium 227
6.6.2 Diamond 228
6.6.3 Elemental Semiconductors 230
6.6.4 Rhenium Trioxide 231
6.6.5 Perovskites 233
6.7 Problems 237
6.8 Further Reading 241
6.9 References 242
 Contents ix

7 Optical Materials 243

7.1 Light, Color, and Electronic Excitations 243
7.2 Pigments, Dyes, and Gemstones 245
7.3 Transitions between d Orbitals (d-to-d Excitations) 246
7.3.1 Ligand- and Crystal-Field Theory 246
7.3.2 Absorption Spectra and Spectroscopic Terms 248
7.3.3 Correlation Diagrams 252
7.3.4 Selection Rules and Absorption Intensity 255
7.4 Charge-Transfer Excitations 258
7.4.1 Ligand-to-Metal Charge Transfer 259
7.4.2 Metal-to-Metal Charge Transfer 260
7.5 Compound Semiconductors 261
7.5.1 Optical Absorbance, Band Gap, and Color 262
7.5.2 Electronegativity, Orbital Overlap, and Band Gap 263
7.6 Conjugated Organic Molecules 265
7.7 Luminescence 267
7.8 Photoluminescence 268
7.8.1 Components of a Phosphor 268
7.8.2 Radiative Return to the Ground State 270
7.8.3 Thermal Quenching 272
7.8.4 Lanthanoid Activators 274
7.8.5 Non-Lanthanoid Activators 279
7.8.6 Energy Transfer 281
7.8.7 Sensitizers 283
7.8.8 Concentration Quenching and Cross Relaxation 284
7.8.9 Up-Conversion Photoluminescence 285
7.9 Electroluminescence 287
7.9.1 Inorganic Light-Emitting Diodes (LEDs) 287
7.9.2 Organic Light-Emitting Diodes (OLEDs) 289
7.10 Materials for Lighting 291
7.10.1 Fluorescent Lamp Phosphors 292
7.10.2 Phosphor-Converted LEDs for White Light 293
7.11 Problems 294
7.12 Further Reading 298
7.13 References 299

8 Dielectrics and Nonlinear Optical Materials 301

8.1 Dielectric Properties 301
8.1.1 Dielectric Permittivity and Susceptibility 302
x Contents

8.1.2 Polarization and the Clausius–Mossotti Equation 303

8.1.3 Microscopic Mechanisms of Polarizability 305
8.1.4 Frequency Dependence of the Dielectric Response 306
8.1.5 Dielectric Loss 308
8.2 Dielectric Polarizabilities and the Additivity Rule 309
8.3 Crystallographic Symmetry and Dielectric Properties 313
8.4 Pyroelectricity and Ferroelectricity 314
8.4.1 Ferroelectricity in BaTiO3 314
8.4.2 Antiferroelectricity 319
8.5 Piezoelectricity 321
8.6 Local Bonding Considerations in Non-Centrosymmetric Materials 324
8.6.1 Second-Order Jahn–Teller Distortions with d 0 Cations 325
8.6.2 Second-Order Jahn–Teller Distortions with s2p0 Cations 327
8.7 Nonlinear Optical Materials 330
8.8 Nonlinear Susceptibility and Phase Matching 331
8.9 Important SHG Materials 334
8.9.1 KH2PO4 336
8.9.2 KTiOPO4 336
8.9.3 Niobates and Tantalates 338
8.9.4 Organic and Polymer NLO Materials 339
8.9.5 Borates 340
8.10 Problems 343
8.11 Further Reading 346
8.12 References 346

9 Magnetic Materials 349

9.1 Magnetic Materials and Their Applications 349
9.2 Physics of Magnetism 349
9.2.1 Bar Magnets and Atomic Magnets 349
9.2.2 Magnetic Intensity, Induction, Energy, Susceptibility, and
Permeability 352
9.2.3 Unit Systems in Magnetism 355
9.3 Types of Magnetic Materials 356
9.4 Atomic Origins of Magnetism 357
9.4.1 Electron Movements Contributing to Magnetism and Their
Quantization 357
9.4.2 Atomic Magnetic Moments 359
9.4.3 Magnetic Moments for 3d Ions in Compounds 363
9.4.4 Magnetic Moments for 4f Ions in Compounds 366
9.4.5 Note on Magnetic Moments of 4d and 5d Metals in Compounds 366
 Contents xi

9.5 Diamagnetism 367

9.6 Paramagnetism 367
9.6.1 Curie and Curie–Weiss Paramagnetism 368
9.6.2 Pauli Paramagnetism 371
9.7 Antiferromagnetism 372
9.8 Superexchange Interactions 374
9.9 Ferromagnetism 377
9.9.1 Ferromagnetic Insulators and Half-Metals 381
9.9.2 Ferromagnetic Metals 382
9.9.3 Superferromagnets 384
9.10 Ferrimagnetism 385
9.11 Frustrated Systems and Spin Glasses 387
9.12 Magnetoelectric Multiferroics 388
9.13 Molecular and Organic Magnets 389
9.14 Problems 391
9.15 Further Reading 394
9.16 References 394

10 Conducting Materials 396

10.1 Conducting Materials 396
10.2 Metals 398
10.2.1 Drude Model 398
10.2.2 Free-Electron Model 402
10.2.3 Fermi–Dirac Distribution 403
10.2.4 Carrier Concentration 405
10.2.5 Carrier Mobility and Effective Mass 406
10.2.6 Fermi Velocity 407
10.2.7 Scattering Mechanisms 409
10.2.8 Band Structure and Conductivity of Aluminum 411
10.2.9 Band Structures and Conductivity of Transition Metals 412
10.3 Semiconductors 414
10.3.1 Carrier Concentrations in Intrinsic Semiconductors 414
10.3.2 Doping 416
10.3.3 Carrier Concentrations and Fermi Energies in Doped
Semiconductors 419
10.3.4 Conductivity 421
10.3.5 p–n Junctions 422
10.3.6 Light-Emitting Diodes and Photovoltaic Cells 425
10.3.7 Transistors 426
xii Contents

10.4 Transition-Metal Compounds 428

10.4.1 Electron Repulsion: The Hubbard Model 428
10.4.2 Transition-Metal Compounds with the NaCl-Type Structure 431
10.4.3 Transition-Metal Compounds with the Perovskite Structure 434
10.5 Organic Conductors 437
10.5.1 Conducting Polymers 438
10.5.2 Polycyclic Aromatic Hydrocarbons 441
10.5.3 Charge-Transfer Salts 443
10.6 Carbon 445
10.6.1 Graphene 445
10.6.2 Carbon Nanotubes 447
10.7 Problems 451
10.8 Further Reading 454
10.9 References 455

11 Magnetotransport Materials 457

11.1 Magnetotransport and Its Applications 457
11.2 Charge, Orbital, and Spin Ordering in Iron Oxides 458
11.2.1 The Verwey Transition in Magnetite, Fe3O4 458
11.2.2 Double-Cell Perovskite, YBaFe2O5 460
11.2.3 CaFeO3 and SrFeO3 462
11.3 Charge and Orbital Ordering in Perovskite-Type Manganites 465
11.3.1 Spin and Orbital Ordering in CaMnO3 and LaMnO3 465
11.3.2 The La1−xCaxMnO3 Phase Diagram 468
11.3.3 Tuning the Colossal Magnetoresistance 470
11.4 Half-Metals and Spin-Polarized Transport 472
11.4.1 Magnetoresistant Properties of Half-Metals 472
11.4.2 CrO2 476
11.4.3 Heusler Alloys 477
11.4.4 Half-Metals with Valence-Mixing Itinerant Electrons 480
11.5 Problems 481
11.6 Further Reading 483
11.7 References 483

12 Superconductivity 486
12.1 Overview of Superconductivity 486
12.2 Properties of Superconductors 488
12.3 Origins of Superconductivity and BCS Theory 492
12.4 C60-Derived Superconductors 500
12.5 Molecular Superconductors 505
 Contents xiii

12.6 BaBiO3 Perovskite Superconductors 509

12.7 Cuprate Superconductors 511
12.7.1 La2CuO4 “214” Materials 512
12.7.2 YBa2Cu3O7−δ “YBCO” or “123” Materials 513
12.7.3 Other Cuprates 516
12.7.4 Electronic Properties of Cuprates 517
12.8 Iron Pnictides and Related Superconductors 521
12.9 Problems 523
12.10 Further Reading 526
12.11 References 526

13 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation

Chemistry 529
13.1 Electrochemical Cells and Batteries 529
13.2 Fuel Cells 532
13.3 Conductivity in Ionic Compounds 533
13.4 Superionic Conductors 536
13.4.1 AgI: A Cation Superionic Conductor 536
13.4.2 PbF2: An Anionic Superionic Conductor 539
13.5 Cation Conductors 540
13.5.1 Sodium β-alumina 540
13.5.2 Other Ceramic Cation Conductors 542
13.5.3 Polymeric Cation Conductors 543
13.6 Proton Conductors 545
13.6.1 Water-Containing Proton Conductors 546
13.6.2 Acid Salts 547
13.6.3 Perovskite Proton Conductors 548
13.7 Oxide-Ion Conductors 549
13.7.1 Fluorite-Type Oxide-Ion Conductors 552
13.7.2 Perovskite, Aurivillius, Brownmillerite, and Other Oxide
Conductors 553
13.7.3 SOFC Electrode Materials and Mixed Conductors 555
13.8 Intercalation Chemistry and Its Applications 555
13.8.1 Graphite Intercalation Chemistry 556
13.8.2 Lithium Intercalation Chemistry and Battery Electrodes 559
13.8.3 Lithium-Ion Batteries with Oxide Cathodes 561
13.8.4 Electrochemical Characteristics of Lithium Batteries 568
13.8.5 Other Lithium Battery Electrode Materials 569
13.9 Problems 573
13.10 Further Reading 576
13.11 References 576
xiv Contents

14 Zeolites and Other Porous Materials 579

14.1 Zeolites 579
14.1.1 Representative Structures of Zeolites 581
14.1.2 Roles of Template Molecules in Zeolite Synthesis 586
14.1.3 Zeolites in Catalysis 588
14.1.4 Ion-Exchange Properties 593
14.1.5 Drying Agents, Molecular Sieving, and Sorption 595
14.1.6 AlPOs and Related Materials 596
14.2 Mesoporous Aluminosilicates 597
14.3 Other Porous Oxide Materials 600
14.4 Metal–Organic Frameworks (MOFs) 605
14.4.1 MOF Structures 605
14.4.2 Some Applications of MOFs 608
14.5 Problems 612
14.6 Further Reading 615
14.7 References 616

15 Amorphous and Disordered Materials 619

15.1 The Atomic Structure of Glasses 620
15.2 Topology and the Structure of Glasses 622
15.3 Oxide Glasses 625
15.4 Optical Properties and Refractive Index 625
15.5 Optical Fibers 631
15.6 Nucleation and Growth 633
15.7 The Glass Transition 634
15.8 Strong and Fragile Behavior of Liquids and Melts 639
15.9 Low-Temperature Dynamics of Amorphous Materials 642
15.10 Electronic Properties: Anderson Localization 644
15.11 Metallic Glasses 647
15.12 Problems 651
15.13 Further Reading 652
15.14 References 652

Appendix A: Crystallographic Point Groups in Schönflies Symbolism 655

Appendix B: International Tables for Crystallography 656
Appendix C: Nomenclature of Silicates 661
Appendix D: Bond-Valence Parameters in Solids 662
Appendix E: The Effect of a Magnetic Field on a Moving Charge 663
Appendix F: Coupling j–j 664
 Contents xv

Appendix G: The Langevin Function 665

Appendix H: The Brillouin Function 666
Appendix I: Measuring and Analyzing Magnetic Properties 670
Appendix J: Fundamental Constants of Exact Value 672
References for Appendices 673
Index 674
Preface

Functional materials are an integral part of daily life. As an example, consider the materials
that underpin smartphone technology. The integrated circuitry is made from complex
patterns of semiconductors, metallic conductors, and insulators. Organic light-emitting
diodes convert electrical signals from the processor into a vibrant high-resolution color
display. The display is protected by a screen made from tough but lightweight Gorilla®
glass, which is coated with a transparent conducting oxide to make the screen responsive to
the touch of a finger. Magnetic materials are used in the speakers, a lithium-ion battery
powers the device, specific dielectric materials are used to receive and isolate a call once the
signal reaches a base station, and the list goes on.
This book explores the fascinating world of functional materials from the perspective of
those who are tasked with inventing them, solid state chemists. We therefore adopt the
chemist’s definition of a material as a substance whose structure and properties are con-
trolled at the atomic level to produce a specific function. Returning to our example, a modern
smartphone contains over half of the non-radioactive elements on the periodic table. A few
are used in their elemental form, but in most cases the desired function can only be achieved
by combining elements to form compounds. With the periodic table as a palette, how does
the chemist design and synthesize the mind-boggling variety of functional materials that
future technologies depend upon? That question is the topic this book explores.
The book is written specifically with teaching in mind and is intended primarily for use in
upper-level undergraduate or graduate level courses. While our perspective is that of
a chemist, the book is accessible to physicists and engineers as well. Mathematical details
are given where they add deeper understanding, but the focus is always on relating the
properties of a material to the characteristics of the atoms and molecules from which it is
built.
The first six chapters cover the fundamentals of extended solids: crystal structures, defects,
reactivity, phase diagrams, phase transitions, chemical bonding, and band structure. The
remaining chapters, each of which is organized around a specific property or class of
materials, show how the properties of modern functional materials can be understood
from these fundamental concepts. Recognizing that the field of solid state chemistry is
much more expansive than can be covered in a single course, the later chapters are designed
to be largely independent of each other. This organization provides the instructor freedom to
tailor a course to cover those materials that are most relevant for their students.
Coverage of inorganic and organic materials is interwoven throughout the book to place
the emphasis on properties. To keep the scope at a manageable level, neither synthesis nor

xvii
xviii Preface

characterization are covered in detail. Instead, boxes on synthetic methods and character-
ization methods are placed throughout the book to highlight specific examples. In a similar
vein, boxes are used to describe how the properties of nanoscale solids differ from bulk
materials (Nanoscale Concepts), and to highlight important technological applications of
materials (Materials Spotlight). Students learn by practice, and, in this spirit, we have
included dozens of problems at the end of each chapter to allow students to test their
understanding of the concepts covered in the chapter. Instructors can obtain a full set of
worked solutions on request.
We hope that this book will be a valuable source of learning for the next generations of
solid state scientists and engineers and a resource for those who already work in this
fascinating field.

Patrick Woodward
Pavel Karen
John Evans
Thomas Vogt
Acknowledgments

We are indebted to several organizations and countless people for their support and encour-
agement. PMW would like to acknowledge the Leverhulme Foundation for supporting his
stay at Durham University as a Visiting Professor during the 2017–2018 academic year. TV
spent the beginning of 2018 as a Fellow at Durham University’s Institute of Advanced Study.
He is grateful to Linda Crowe and the rest of the team at the institute, as well as David
Wilkinson, Principal at St. John’s College, for their warm hospitality and for providing an
environment so conducive to scholarly work. These overlapping stays in Durham were
instrumental in making the final push to finish this book. PMW is grateful for many years
of support from the Solid State Materials Chemistry program of the National Science
Foundation. PMW and JSOE thank Arthur Sleight for inspiration and early career mentor-
ing. In these data-dominated times, PK is grateful to the now Professor Jiří Hanika for
teaching Fortran programming in the 1970/1 course Computational Technology at the
VŠCHT in Prague, and to the now Ing. František Hovorka, CSc, for the idea of taking an
external typing course during the sophomore year 1967/8 at the SPŠCH Praha. We all would
like to thank both colleagues and students who provided key feedback on early versions of
the chapters. Finally, we thank our families for their patience and support over this long
journey.

xix
1 Structures of Crystalline Materials

This book is about functional materials—those that perform a task or a technological

operation. By the end of the book, we’ll see that all useful properties can ultimately be traced
back to structure and dynamics at the atomic level of materials. Understanding structure is
therefore of crucial importance.
In this chapter we’ll investigate the structures of crystalline materials—those in which
atomic arrangements are repeated periodically in three-dimensional (3D) space. Non-
crystalline materials are covered in Chapter 15. In the first section, we will discuss the
symmetry and crystallography concepts that are important for the description of crystalline
substances. A brief introduction to structure databases will follow. In the third section, we’ll
cover the nomenclature and electron counting rules needed to understand the composition of
solids; before learning in the fourth section how structures are built up by packing spheres,
connecting coordination polyhedra, or via networks. In the fifth and last section, we’ll
discuss some structure types encountered later in the book. In addition to the figures and
descriptions given in the chapter, readers might find it useful to draw models of important
structures with the included structural coordinates.

1.1 Symmetry
In the first section of this chapter, we’ll develop the language required to describe the
structures of crystalline compounds. We know from everyday life that such materials
frequently display an amazing regularity and symmetry on the macroscopic scale—salt
crystals can grow as “perfect” cubes and many minerals and gemstones display wonderfully
symmetric facets. The origin of this macroscopic symmetry can ultimately be traced back to
the symmetry that’s present at the atomic scale (Å or 10−10 m). This local symmetry is
replicated millions of times by translational symmetry to produce symmetric macroscopic
objects. It’s perhaps worth stating at the outset that there is nothing “magical” about the
high-symmetry structures materials adopt. As we will see in Chapter 5, local bonding

1
2 Structures of Crystalline Materials

interactions have inherent symmetry, and dense packing of such units is favorable
energetically.

1.1.1 Translational Symmetry

To describe a crystal structure, it is useful to introduce the concept of lattice; a spatial pattern
of points of equal and equally oriented surroundings. We then define a motif, which might be
a small group of atoms, a molecule, or a collection of several molecules. If we associate this
motif with each of the lattice points, a crystal structure is built as shown in Figure 1.1. We can
think of this “association” as re-drawing the motif at a constant displacement from each
lattice point. The lattice is an operator of translational symmetry of crystal structures, of their
periodicity. We can see in Figure 1.1 that the translational symmetry defined by the lattice
produces a structure in which the individual atoms in the motif achieve a sensible bonding
environment.
The small spatial segment that fully represents the entire structure upon periodic repetition
is called the unit cell.1 We can use the analogy of tiles (2D) or bricks (3D) being stacked side
by side. In 3D, the unit cell is a parallelepiped, a box whose sides are parallelograms. The size
and shape of the unit cell is described with three lattice vectors a, b, c of lengths a, b, c, and
angles α, β, γ. The angle α is between b and c, β between a and c, and γ between a and b.
Together, a, b, c, α, β, γ are called lattice parameters or unit-cell parameters.
Positions of atoms inside the unit cell are expressed with relative or fractional coordin-
ates x,y,z in terms of fractions of the lattice vectors that define the unit cell. Fractional
coordinates define the position (or radius) vector r from the unit-cell origin to the atomic
position as r = xa + yb + zc. They can always be expressed on a 0 to 1 scale. Because of
translational symmetry, a coordinate of 1.2 is equivalent to 0.2, or a coordinate −0.2 is
equivalent to −0.2 + 1 = 0.8.

Figure 1.1 Association of an atomic motif with a lattice produces a crystal structure.

1
The following rules apply for choosing the unit cell: (a) its rotational symmetry is the same as that of the lattice, (b)
the edges and angles are made as similar to each other as possible, (c) the number of right angles is maximized, and (d)
the volume is minimized. Where applicable, the origin coincides with the inversion center of symmetry (Section
1.1.2). In some cases (Section 1.1.4), the cell contains more than one lattice point.
 1.1 Symmetry 3

Figure 1.2 Description by indices of lattice planes and lattice directions with respect to the unit cell. An extra
cell is drawn in order to show cases where planes facing the origin intercept the lattice vector at negative values.

Directions in the lattice are given with a [uvw] notation. When a line is drawn from the
origin, parallel to the desired direction, then u,v,w are simply the relative/fractional coordin-
ates of any point that line goes through, multiplied to give integer values. If it goes through
½,1,½, the direction is [121]. Symbols [242] and [484] would represent the same direction.
A set of symmetrically equivalent lattice directions, such as [100], [010], [001], [−100], [0−10],
[00−1] in a cubic lattice, is collectively referred to using angle brackets, ⟨100⟩.
It’s often useful to define a set of parallel equidistant planes in a lattice. These are
represented by an hkl notation. Starting with the plane that contains the origin, an hkl set
of equidistant planes divides the unit-cell vector a into h sections, b into k sections and c into
l sections (Figure 1.2). Dividing into 0 sections is possible and means that the set of planes is
parallel with that axis. If the plane that faces the origin crosses a, b, or c at negative values,
the appropriate h, k, or l of that set has negative sign (often put above the index, like h). An
(hkl) symbol refers to a plane or to a crystal face (Miller indices). A set of their symmetry-
equivalent orientations is denoted in curly brackets, {hkl}.

1.1.2 Rotational Symmetry

A point group2 is a set of symmetry operations that fulfill the mathematical require-
ments of being a group3 and act on an isolated geometrical object. The number of

2
At least one point of the object remains unshifted under point-group symmetry operations. Elementary knowledge of
point-group symmetry may be an advantage for the reader; see Further Reading.
3
A group must have a closure (combination of two elements yields an element of the group), fulfill the mathematic
associative law (the result of combination of the elements is independent of the order they are applied), have an
identity (an element that converts other elements into themselves), and have an inversion (every element has an
inverse element; when combined together, they yield the identity element).
4 Structures of Crystalline Materials

Figure 1.3 Inversion center (symbol −1 or 1) at the origin operates on an object at the endpoint of
the object’s radius vector r. The comma symbolizes a change in handedness (left- to right-hand
relationship).

Figure 1.4 Operation of the twofold rotation axis (of symbols 2 in text and in graphics) on an object at
the point r.

symmetry operations defines the order of the group; the higher the order, the higher
the symmetry. The symmetry operations are performed by the elements of point
symmetry; the identity, inversion center, mirror plane, rotation axis, and rotoinver-
sion axis. All objects possess the identity; other symmetry elements may or may not
be present.
Let’s start with the inversion center. Figure 1.3 shows a point x,y,z represented by its radius
vector r and the effect on that point of an inversion center at the origin of the coordinate
system (shown by a small circle). Inversion moves x,y,z to −x,−y,−z. Mathematically, we can
describe this transformation of r to r 0 with the equation r 0 = R ·r, where R is the matrix of the
point-symmetry element, describing its operation.4 Inversion is given the symbol −1, often
typeset as 1.
Symmetry elements that unify a point in lattice space with another one by rotating it in
steps of 1/n (n = 1, 2, 3, 4, 6) of the full circle are called n-fold rotation axes. The full-circle
rotation (n = 1) is the identity, a twofold axis (n = 2) rotates by ½ of the full circle and has
symbol 2, etc. As can be seen in Figure 1.4, twofold rotation around the z axis moves a point
of fractional coordinates x,y,z to −x,−y,z. A rotoinversion axis is a single element, the
operation of which combines rotation and inversion. However, the twofold rotoinversion
axis −2, shown in Figure 1.5, operates like a mirror plane, m, as can be also demonstrated by
4
The three columns in this matrix are the products of the particular symmetry operation on the end-points of the
respective unit-cell vectors (1,0,0), (0,1,0), and (0,0,1). This is conveniently used to set up the matrix.
 1.1 Symmetry 5

Figure 1.5 Rotoinversion axis −2 is identical with mirror m.

multiplying together the matrices representing 2 and −1. Since m ≡ −2, all point-symmetry
elements are in fact elements of rotational symmetry; the rotational axes 1, 2, 3, 4, 6 and the
rotoinversion axes −1, −2, −3, −4, −6. Accordingly, the operation matrix of each of them has
the symbol R.

1.1.3 Crystallographic Point Groups and Crystal Systems

Due to the infinite number of rotation axes, infinitely many point groups are possible for
isolated objects. However, in crystal structures, the translational symmetry of space filling is
only compatible with a small number of rotation axes. Consider that we can tile a plane
perfectly with identical rectangular tiles (twofold axis present), isosceles triangles (threefold),
squares (fourfold), or hexagons (sixfold), but we can’t with pentagons (fivefold) or heptagons
(sevenfold), etc. This argument extends to the 3D space filled by the “bricks” of unit cells
(Figure 1.1). The point groups with symmetry elements 1, 2, 3, 4, 6, 1, m (≡ 2), 3, 4, 6, which
are compatible with translational symmetry, are called crystallographic point groups, also
known as crystal classes.
There are 32 crystallographic point groups and they are classified into seven crystal
systems; cubic, tetragonal, hexagonal, trigonal, orthorhombic, monoclinic, and triclinic.
Each crystal system is defined by its minimum point-group symmetry (Table 1.1, see also
Appendix A). If we take the cubic system as an example, the minimum-symmetry point
group has symbol 23. It means that at each lattice point, the twofold axes along the x-, y-, and
z-coordinate-system axes repeat the threefold axis of the symbol along all body diagonals of
the adjacent cells. These threefold axes are easier to visualize and remember as the symmetry
condition for the cubic crystal system. So if the actual crystal structure has four intersecting
threefold axes, it is cubic. If you do not see intersecting threefold axes, the structure cannot be
cubic even if the unit cell has right angles and equal edges.

1.1.4 Bravais Lattices

As noted earlier, a lattice is a collection of points with identical surroundings. Having the
highest rotational symmetry of each crystal system, 14 types of Bravais lattices are possible
6 Structures of Crystalline Materials

Table 1.1 Sorting 32 crystallographic point groups into seven crystal systems.

Crystal system Minimum symmetry Higher-symmetry point groups

Triclinic 1 1
Monoclinic 2, m 2/m
Orthorhombic 222 mm2, mmm
Tetragonal 4, 4 4/m, 422, 4mm, 4 2m, 4/mmm
Hexagonal 6, 6 6/m, 622, 6mm, 6m2, 6/mmm
Trigonal 3, 3 32, 3m, 3m
Cubic 23 m3, 4 3m, 432, m3m

(Figure 1.6), which represent 14 types of translational symmetry in 3D lattice space5 of relative
coordinates (Section 1.1.1). Some unit cells have lattice points only at the corners, and their
lattices are called primitive Bravais lattices, labeled with symbol P. Others have lattice points
located also at the centers of some or all unit-cell faces or at the unit-cell center, and these are
called centered Bravais lattices. The body-centered lattice has symbol I (the cell has an
additional lattice point at ½a + ½b + ½c relative to a P lattice). The face-centered lattices
have symbol F when all unit-cell faces are centered. Symbol A, B, or C is used when just two
opposite unit-cell sides are centered, along one direction, a, b, or c. Thus, C-centering6 adds an
additional lattice point at ½a + ½b relative to a P lattice. A special type of centering, R, occurs
in the hexagonal lattice, Figure 1.6. This R-centered lattice is equivalent to a P lattice with
a rhombohedral7 unit cell, see Table 1.2. The rhombohedral lattice occurs only in those
structures of the trigonal crystal system that carry the symbol R in their symmetry description.
The remaining trigonal structures are described with a primitive hexagonal cell P. It is often
convenient to express also the R structures on a hexagonal cell (not just the P). Having triple
the volume of the rhombohedral cell, the hexagonal cell contains three rhombohedral lattice
points: 0,0,0 and ⅔,⅓,⅓ with ⅓,⅔,⅔, shown in Figure 1.6 labelled as hR.
Figure 1.7 gives an idea why only certain types of centering are possible for certain crystal
systems. For example, a C-centered tetragonal cell could always be described with a smaller
primitive cell, an F-centered tetragonal with a smaller I-centered cell. Similarly, a monoclinic
B cell becomes a smaller P cell, monoclinic F cell becomes a smaller C cell, and a monoclinic
I cell is equivalent to a C cell via an A cell rotated around b. However, there are cases where it
is useful to choose a non-standard Bravais cell; for example in order to illustrate similarity
between two structures.

5
The lattice space defines the orientation and angles of the coordinate-system axes applied to each unit cell of this
space.
6
While International Tables for Crystallography use the British-English forms “centre”, “centring”, and “centred”,
the alternative spellings “center”, “centering”, and “centered” prevail in the USA.
7
Terms like trigonal, tetragonal, and hexagonal originate in the rotational symmetry; the term rhombohedral implies
that the Bravais cell is a rhombohedron, hence it refers to the lattice.
 1.1 Symmetry 7

Figure 1.6 Standard settings of the 14 Bravais lattices. Lower-case letters: a = anorthic/triclinic, m =
monoclinic, o = orthorhombic, t = tetragonal, h = hexagonal and c = cubic. Upper-case letters refer
to centering. The primitive rhombohedral lattice is often described on its equivalent hR cell.
8 Structures of Crystalline Materials

Table 1.2 Metrics of lattices. Parameters not listed can take

any values.

Lattice Conditions for lattice parameters

Triclinic (anorthic) None*

Monoclinic α = γ = 90°
Orthorhombic α = β = γ = 90°
Tetragonal a = b, α = β = γ = 90°
Rhombohedral a = b = c, α = β = γ
Hexagonal a = b, α = β = 90°, γ = 120°
Cubic a = b = c, α = β = γ = 90°
* For limits on triclinic angles, see ref. [1].

Figure 1.7 Non-standard centering of monoclinic (top) and tetragonal Bravais cells (bottom). Drawing
two cells reveals the true cell of the same crystal system.

1.1.5 Introduction to Space Groups

We’ve seen that there are 32 crystallographic point groups and 14 Bravais lattices. Their
combination gives a total of 230 space groups (we’ll see where this number comes from in
Section 1.1.7). A space group defines both the translational symmetry (type of the Bravais
lattice) and the rotational symmetry (point group) of the structure. Of the 230 space groups, 73
do not involve any symmetry other than that already present in the Bravais lattice and point
group alone; these are called symmorphic space groups. The remaining 157 space groups are non-
symmorphic and possess translations by suitable fractions of lattice vectors, brought about by
screw axes or glide planes. Before we explain these two terms, a note on symmetry operators.
When discussing point groups, it was convenient to introduce the matrix operator R that
acts on a point r of the relative-coordinate vector r such that r 0 = R · r. Because space groups
 1.1 Symmetry 9

may include the above-mentioned additional fractional translations, this symbolic language
is extended into a Seitz operator, (R | t) = R · r + t, which combines rotations (matrix R) and
the possible translations (vector t). If we consider a symmetry element that involves no
translations, such as the inversion 1, the Seitz symbol is (R | 0) hence (1 | 0). A rotational axis
has a direction [uvw] that must be included. The Seitz symbol (2[001] | 0) then refers to
a twofold rotation around the z axis. A plane has a direction as well—the direction of its
normal vector that is oriented perpendicular to the plane. As an example, the mirror in the xy
plane of Figure 1.5 has the Seitz symbol (m[001] | 0). The Seitz operators that include the
fractional translations are explained in the subsection below.

1.1.6 Symmetry Elements That Combine Rotation and Translation

The periodicity of crystal structures (their translational symmetry) means that the set of
rotational symmetry elements repeats at each lattice point. This creates additional symmetry
elements in between lattice points and may give rise to two types of symmetry elements that
aren’t present in isolated molecules,8 screw axes and glide planes. A screw axis combines
rotation with translation along the axis of rotation. An NM screw axis (M < N) rotates
anticlockwise by 360/N degrees while shifting the image by a distance equal to an M/N
fraction of the lattice periodicity along that axis. As an example, the twofold screw axis 21 in
Figure 1.8 rotates by 180° and shifts by ½ of the vector c (the axis is along z). A 63 axis rotates
by increments of 60° and each time shifts by 3/6 = ½ of c.
Figure 1.9 shows the symmetry operations of axes 62 and 64, illustrating that screw axes
NM and N(N−M) produce mirror images of each other. The “tailed hexagon” is the graphic
symbol of the sixfold screw axes.
A glide plane operates as a mirror plus a plane-parallel shift by half a vector length between
two lattice points. When the shift is half of one lattice vector (either a, b, or c), the glide is
called an axial glide plane a, or b, or c. The operation of the, say, c glide plane is a reflection

Figure 1.8 The screw axis 21[001] rotates 180° and shifts by ½ of the vector c. The graphical symbol of 21
parallel to the drawing plane is a half-arrow.

8
In the point groups used for isolated molecules, all symmetry elements intersect at a point; in space groups this is no
longer true.
10 Structures of Crystalline Materials

Figure 1.9 Screw axes rotate anticlockwise. Axes 62 and 64 produce mirror images of each other. The 62
axis rotates by 60° and translates by 2/6 c, the 64 axis rotates by 60° and translates by 4/6 of the unit-cell
length c. Due to translational symmetry, integers are subtracted from fractional coordinates ≥1, such as
3 /3 ≡ 0 or 10 /3 ≡ ⅓.

Figure 1.10 The glide plane c[100] creates a mirror image shifted by c/2.

followed by shift along c by half the c-length. This is shown in Figure 1.10 with the c-glide
plane oriented in the direction of x.9 Note that an equally oriented plane with a translation
along y would be called a b glide. In rare cases (five of the 230 space groups), centering creates
a situation when the two alternative shift directions along a glide plane result in the same
point. The symbol a, b, or c of such a glide plane is then changed to e (for equivalent or
either), a double glide plane.
When the shift length is half the sum of two lattice vectors (half the vector to the diagonal
lattice point), we have a diagonal glide plane, denoted by the symbol n. An example for a/2 +
b/2 is in Figure 1.11.
When the shift is half of the F- or I-centering vector, we have a diamond glide plane,
symbol d. This shift can be decomposed into components along lattice vectors (a /4) + (b /4)
or (a /4) + (c /4) or (b /4) + (c /4) for the F-centered orthorhombic and cubic lattices, and into
(a/4) + (b/4) +(c/4) for the I-centered tetragonal and cubic lattices.

9
As noted above (Section 1.1.5), the direction of a plane is the direction of the plane’s normal.
 1.1 Symmetry 11

Figure 1.11 An orthorhombic diagonal glide n[001] in the xy plane, with its graphical symbol on top left
(see also Appendix B). It creates a mirror image shifted by a/2 and b/2. The subsequent operation of the
same diagonal glide then recreates the original point in the next unit cell.

1.1.7 Space-Group Symbols

Shorthand symbols are used to name space groups. These consist of one of the Bravais-lattice
symbols, P, F, I, R, A, B, C, followed by the symbol of the crystallographic point group, in
which screw axes and glide planes can replace ordinary rotation axes and mirror planes when
appropriate, for example Pnma. Non-symmorphic space groups are thus recognizable by the
presence of symbols such as a, b, c, e, n, d, 21, 32, 42, etc., which indicate the fractional lattice
translations. For quick understanding, it is helpful to know the orientation of the rotational-
symmetry elements listed in the standard10 space-group symbols (Table 1.3). Consider
I 4/m c m (with individual posts separated for clarity) as an example. The presence of
a fourfold axis, combined with the absence of threefold axes, tells us that this is
a tetragonal space group (the I specifies that the Bravais lattice is body-centered). The first
post 4/m means that both the fourfold axis and the mirror m are in the direction of c, hence 4
is perpendicular to m. The second post tells us that there are glide planes c in the directions of
both edges of the square face (edges a and b) of the unit cell. The third post tells us that there
are mirror planes in the direction of the diagonal of this square face.
The triclinic crystal system has a P lattice and is compatible with only two symmetry
elements; identity and inversion (Table 1.1). There are therefore only two triclinic space
groups, those with symbols P1 and P1. As listed in Table 1.1, the monoclinic crystal system
allows three types of rotational symmetry; the 2, m, and 2/m crystallographic point groups.
There are two Bravais lattices (Figure 1.7) available; primitive, P, and base-centered, C. This
results in six symmorphic space groups; P2, Pm, P2/m, C2, Cm, C2/m. Ten non-symmorphic
monoclinic groups would be obtained by replacing 2 with screw axes 21 and m with glide
planes c; P21, Pc, P21/m, P2/c, P21/c, C21, Cc, C21/m, C2/c, C21/c. However, not all of these
10 space groups are unique. This is because the combination of C-centering and the rotation
axis 2 generates its “own” 21 (Figure 1.12). In the standard setting, they appear parallel to
b at x = ¼ and ¾. For this reason, C21 ≡ C2, C21/m ≡ C2/m, C21/c ≡ C2/c, and there are not 16

10
Standard orientations are those with unit-cell axes a, b, c chosen by an agreed set of rules, as listed in the
International Tables for Crystallography, Volume A. Permutations are possible but non-standard.
12 Structures of Crystalline Materials

Table 1.3 Directions of the three posts in standard space-group symbols. Trigonal directions
refer to hexagonal unit cell.

Crystal system Post 1 Post 2 Post 3

Monoclinic Perpendicular to the plane of the monoclinic angle (standard, || b)

Orthorhombic Edge a Edge b Edge c
Tetragonal 4(4) || c Square edges Square diagonals
Trigonal 3(3) || c
Rhombus edges Rhombus diagonal, longer
Hexagonal 6(6) || c
Cubic Edges 3(3) in body diagonals Face diagonals

Figure 1.12 Operation of a twofold axis 2[010] combined with monoclinic centering C (lattice points are
marked with ×) makes a pattern of points related by screw axes 21[010] at x = ¼ and ¾ (marked with half-
arrow heads).

but only 13 unique monoclinic space groups possible. Similar arguments can be developed
for all crystal systems leading to a total of 230 unique 3D space groups.
Symmetry information for all space groups is given in the International Tables for
Crystallography, Volume A (for more details, see Appendix B of this book). The tables
list, inter alia, the positions of all symmetry elements and how an atom at an initial point x,y,
z is reproduced in the unit cell by the symmetry elements present.

1.1.8 Description of a Crystal Structure

A crystal structure is defined by its space-group symmetry, unit-cell parameters, and by the
coordinates of the atoms in its asymmetric unit. The asymmetric unit contains only the
atomic coordinates in the unit cell that are crystallographically unique, all other atomic
positions are created by the symmetry-element operations. Atoms whose coordinates lie on
symmetry elements are said to be on special positions, those which don’t, are on a general
position (x,y,z, Appendix B). Different positions are often referred to using Wyckoff site
labels. The Wyckoff site is labeled with a number and a letter (e.g. 8a). The number is the
multiplicity of a site; it gives the number of equivalent points (atoms) generated by
the available symmetry operations of that site from any one of them (the “original point”).
The letter labels the sites for a given space group. The highest-symmetry site is listed as a and
 1.2 Databases 13

Figure 1.13 Data needed to describe the crystal structure of nickel arsenide.

the lower-symmetry sites as b, c, d. . . (each with its own set of coordinates for symmetry
equivalent points) in the International Tables for Crystallography, Volume A.
For easy drawing, some figures will show auxiliary information that defines the crystal
structure; its space group, unit-cell parameters, and fractional coordinates of atoms in the
asymmetric unit. For example, in Figure 1.13, a nickel atom is listed in the site 2a at 0 0 0 and
an arsenic atom in 2c at ⅓ ⅔ ¼. Application of symmetry operations in a structure-drawing
software (or a look at pages of the space group number 194 in the International Tables,
Volume A) yields an additional Ni at 0 0 ½ and an additional As at ⅔ ⅓ ¾, making a total of
two formula units of NiAs per unit cell (Z = 2).

1.2 Databases
In this chapter, we’ll introduce just a few of many structure types behind millions of
individual crystal structures. Crystallography information comes in databases. The two
with the longest tradition are the Inorganic Crystal Structure Database (ICSD) at the
Fachinformationszentrum Karlsruhe in Germany and the Cambridge Structural Database
(CSD), curated by the Cambridge Crystallographic Data Center. As of October 2020, the
ICSD contained 232012 entries, the CSD 1094733. In addition, National Institute of
Standards and Technology (NIST) has developed a wider Inorganic Structure database
(NIST ICSD) that combines full structure data (mostly from ICSD) with the unit-cell
identification data set curated there as NIST Crystal Data since 1963.
Databases provide powerful software tools for searching known structures, exploring
similarities between materials, and for “data mining” to identify important chemical
and structural trends. The structural information is exported in a standardized
form of the crystallographic information file (cif) that is used as input for various
crystallographic software packages. Some databases provide calculation of diffraction
patterns, useful to identify materials. While the full databases require a license, there
are several open-access resources online. Try a search on “icsd demo” or “database of
zeolite structures” or “mineralogy database” or the “Crystallography Open Database”.
Useful demo-version utilities for plotting crystal structures from cif files are searchable
as well.
14 Structures of Crystalline Materials

1.3 Composition
With the essential ideas of symmetry and crystallographic concepts in place, we’ll spend the rest of
this chapter describing principles that rationalize structures as composites of different elements
and entities. Materials adopt what at first appears to be a bewildering variety of different atomic
arrangements in 3D space. However, many of these arrangements can be readily understood by
considering relatively simple structural rules that concern stoichiometry and connectivity patterns
in compounds. For those purposes, it will be convenient to introduce the crystal-chemical formula
[2] that summarizes structural information. These formulas list each crystallographically non-
equivalent atom11 in the structure separately and contain information about its coordination to
other atoms in superscripted square brackets. The coordination numbers in the brackets follow
the same array as the atoms in the formula, and are separated by a comma, while any bonding to
atoms of the same element comes after a semicolon. In addition to the coordination number, the
bond geometry may be indicated by a lower-case letter (l = line, n = not in a line or plane, t =
tetrahedron, o = octahedron, y = pyramid, p = prism, c = cube, co = cuboctahedron, etc.). To
keep the formula as simple as possible, it’s common to include only connectivities of direct
chemical bonds. These simpler formulas will generally be used in this chapter. The nomenclature
is best illustrated with examples, such as those given in Table 1.4.

Table 1.4 Examples of crystal-chemical formulas for compounds; full = full* neighborhood, simpler =
direct neighborhood (bonding).

Full crystal-chemical formula

Formula Simpler crystal-chemical formula Brief description of bonding

SiO2 Si[4t;]O2[2;] Si tetrahedrally coordinated by 4O while

Si[4t]O2[2] O is coordinated by 2Si
SrTiO3 Sr[8,12co;]Ti[8,6o;]O3[4,2;] Sr in cuboctahedron of 12 O and Ti in
Sr[12co]Ti[6o]O3[4,2] octahedron of 6O
FeS2 Fe[6o;]S2[3;1] Fe in octahedron of 6S; (S–S)2− units
Fe[6o]S2[3;1] present
MgAl2O4 Mg[12,4t;]Al2[6,6o;]O4[(1,3)t;] Mg in tetrahedral and Al in octahedral
Mg[4t]Al2[6o]O4[1,3] coordination
Y3Fe5O12 Y3[4,6,8c;]Fe2[6,6,6o;]Fe3[6,4,4t;]O12[(2,1,1)t;] Y in a cube of 8O; two Fe sites, one
Y3[8c]Fe2[6o]Fe3[4t]O12[2,1,1] octahedral, one tetrahedral

* Coordination numbers refer to the array of the other atoms in the formula, such as Sr 8-coordinated with Ti and
12-coordinated with a cuboctahedron of O in SrTiO3. Possible bonds to atoms of the same kind follow after the
semicolon, such as 1 in FeS2.

11
Apart from chemical identity, crystallographically identical atoms have identical environments.
 1.3 Composition 15

1.3.1 Coordination, Stoichiometry, and Connectivity

If ionic charges are included, the crystal-chemical formula of a binary compound will
incorporate three important balances: The electroneutrality balance, the connectivity bal-
ance, and the bond-valence balance (Figure 1.14). Let’s assume a binary compound CmAn of
two non-identical elements, each at one unique site; one assigned as a “cation” C of charge
number c, the other as an “anion” A of charge number a. The crystal-chemical formula is
cþ ½N; a ½M;
Cm An , where N and M are the coordination numbers of C and A, respectively. The
electroneutrality balance requires that m × c = n × a. The connectivity balance requires
that m × N = n × M because there must be an equal number of CA and AC connections.
Lastly, c/N = a/M is the bond-valence balance [3].12
Understanding these balances allows us to make simple structural predictions from the
chemical formula.13 Consider SiO2 as an example. We know from inorganic chemistry that
silicon favors tetrahedral coordination, so the connectivity balance 1 × 4 = 2 × M gives M = 2
for the coordination number of oxygen (Table 1.4).
For compounds having more than two sites, balances analogous to those for the two-site
formula in Figure 1.14 can be set up. As an example, the mineral hausmannite, Mn3O4 (a
spinel, Section 1.5.1), has Mn2+ in tetrahedral and Mn3+ in octahedral coordination. There is
one oxygen site. Using this information, we write the formula Mn2+[4t]Mn3+2[6o]O4[x,y] with
the coordination numbers around oxygen as unknowns. Two connectivity balances can be
written, 1 × 4 = 4 × x for Mn2+ and 2 × 6 = 4 × y for Mn3+, and we see that each oxygen is
coordinated by one Mn2+ and three Mn3+.
For extended structures14 of ternary and higher-component phases, a bond graph is
informative as it shows the connectivity visually, see Figure 1.15. The symbolism of
a bond graph differs from symbolism of formulas used by molecular chemists, so let’s
again illustrate it with the simple case of SiO2. One starts with the crystal-chemical formula
(Table 1.4) and writes down one Si symbol surrounded by two O symbols. Then a line is
drawn for each bonding connection between two atoms. The lines tell us that each oxygen is
coordinated by two silicon atoms and each silicon is coordinated by four oxygens (the two
lines between Si and O thus do not mean a double bond). A quick examination of the bond

Figure 1.14 Three balances in the crystal-chemical formula of a two-site compound.

12
Termed “the electrostatic valence principle” in ref. [3]. More on bond valences is in Section 5.4.
13
Such estimates can be useful, see Box 1.1.
14
Crystal structures with bond networks that extend over the entire volume.
16 Structures of Crystalline Materials

graphs of TiO2 and CaF2 shows that as the cation coordination number increases from 4 to 6
to 8, the anion coordination number increases from 2 to 3 to 4, consistent with the connect-
ivity balance.

Figure 1.15 Examples of bond graphs. In V2O5, five oxygens occupy three crystallographically different
sites of differing coordinations to V.

Box 1.1 Synthetic Methods: Preparation of Na3N

Most binary ionic compounds were discovered long ago. A number of these compounds, such
as NaCl and KCl, have played pivotal roles in the development and advancement of civiliza-
tions. In principle, such compounds form easily; one brings the two elements together and
initiates the reaction. It is surprising to find that, despite repeated attempts, at the end of the
twentieth century no one had been able to prepare sodium nitride, Na3N. This changed in 2002
when Fischer and Jansen reported the first successful synthesis [4]. Like many binary com-
pounds, it was prepared from the elements. However, the preparation conditions were hardly
typical. Atomic beams of the two components were generated separately in a microwave
plasma and passed into a vacuum chamber where they condensed onto a sapphire substrate
cooled to 77 K. At such a low temperature, the elements don’t react when deposited onto the
substrate, but upon heating to 200 K they begin to form crystalline Na3N.
This compound adopts the cubic ReO3 structure (see Table 1.9), with the cation and anion
positions reversed. As might be expected for such an elusive compound, Na3N is not very
stable, it decomposes above 360 K.
Why is sodium nitride so difficult to prepare and why is it unstable even after it forms? An
important clue comes from the crystal-chemical formula, Na3[2l]N[6o]. Despite its large size, sodium
is coordinated by only two nitride ions in a linear geometry. Compare this with stable transition
metal nitrides such as ScN, ZrN, and CrN, where the NaCl-type structure gives a cation coordin-
ation number of six. The reason for the abnormally low cation coordination number in Na3N is
a simple but unavoidable topological consequence of the stoichiometry. The C3A stoichiometry
means that if the nitride ion is 6-coordinated (larger coordination numbers are rare for N3−) the
sodium ion can only be 2-coordinated. This simple link between stoichiometry and coordination
number [5] is the fundamental reason why the synthesis of Na3N has proved so difficult. Typical
coordination numbers for Na+ range from 6 to 12; the coordination number of two is an extreme
outlier. Let’s also note that the densest-packing principle (Section 1.4.2) supports C3A of only the
smallest cations C occupying voids among the largest anions A, such as Na3As or Li3N.
 1.3 Composition 17

1.3.2 The Generalized 8−N Rule

Atoms bond in patterns that yield stable electron configurations. For an electronegative sp atom
of N valence electrons, the 8−N rule is valid: Valence electrons short of 8 are obtained in bonds.
Carbon of 4 valence electrons obtains the missing 4 electrons by forming 4 bonds; phosphorus
obtains the missing 3 electrons by forming 3 bonds, etc., as illustrated in Figure 1.16.
In a binary compound, the more electronegative sp element, the one that attracts and holds
electrons more, will maintain the 8−N rule in order to achieve a stable configuration of the
noble gas. Most electropositive sp elements will tend to lose their valence electrons to also
achieve a noble-gas configuration. The ionic approximation is therefore a convenient way of
recognizing the stable configurations in a compound.
Consider our binary solid CmAn (Section 1.3.1) of sp atoms. The valence-electron count per
anion A, VECA, is calculated from the stoichiometry and from valence-electron numbers eC
and eA:

VECA ¼ ðm  eC þ n  eA Þ=n: (1.1)

When VECA = 8, both atoms achieve the noble-gas configuration as noted above. When VECA >
8, the excess electrons remain with the cation forming cation–cation bonds or cation-localized
electron pairs. When VECA < 8, the atom A obtains the missing electrons by forming A–A bonds
so that each A has an octet. The formal expression of this generalized 8−N rule [6, 7] for CmAn is:
m
VECA ¼ 8 þ CC   AA; (1.2)
n

where the variable CC is the number of electrons per “cation” C that form C–C bonds or are
localized at the cation as lone pairs, and AA is the number of electrons per “anion” A that
form A–A bonds. The generalized 8−N rule is useful for analysis of structures of nonmetallic
sp phases. Let’s illustrate this using GaSe and SnCl2 with VECA > 8 and CdSb and CaC2 with
VECA < 8.
In GaSe, gallium has three valence electrons, selenium six, VECA = 9. The excess electron will
remain with the Ga cation (CC = 1, Equation (1.2) and form single-bonded (Ga–Ga)4+ pairs. In

Figure 1.16 Structures of selected p elements drawn to equal scale.

18 Structures of Crystalline Materials

GaSe SnCl2

Figure 1.17 Examples of structures with VECA > 8, in which the excess electrons per anion A are localized
at the cation (smaller spheres) as bonds or as lone pairs.

CdSb CaC2

Figure 1.18 Examples of structures with VECA < 8, when “electrons missing to 8” at the anion A are
obtained by sharing in bonds between A atoms.

SnCl2, tin has four valence electrons, chlorine seven, VECA = 9; one electron in excess. Given
two Cl per Sn, two electrons remain at Sn (CC = 2) forming the lone electron pair of Sn2+. The
structures of GaSe and SnCl2 are shown in Figure 1.17.
In CdSb, the cadmium atom has two valence electrons,15 antimony five, and VECA = 7.
The missing electron is obtained by sharing between two Sb atoms that form Sb24− single-
bonded dumbbells (AA = 1) isoelectronic with I2 (Figure 1.18, left). In CaC2, calcium has two
valence electrons, carbon four, and VECA = 5. The three missing electrons are obtained by
sharing in three two-electron bonds (AA = 3), and triple-bonded C22− pairs isoelectronic with
N2 are present in the crystal structure of CaC2 (Figure 1.18, right).

1.4 Structural Principles

In this section, we’ll see that the principles of arranging atoms, ions, or molecules vary
according to the type of forces that hold them together. In some structures, the building units
are packed as efficiently as possible, in others directional covalent bonding gives rise to
networks. In yet other structures it is the local chemical bonding environment of a central
atom that dictates everything else about the structure. Appropriate visualization and under-
standing of such structures, as well as rationalizing their physical and chemical properties,
depends on identifying the underlying principles upon which they were built.
15
4d orbitals in Cd are sufficiently low in energy that they can be considered part of the core.
 1.4 Structural Principles 19

1.4.1 Packing of Spheres

The structures of many materials can be understood by exploring how hard spheres pack. If we
start in 2D, Figure 1.19 shows two symmetric ways of packing circles and their corresponding unit
cells; it’s clear that the packing of Figure 1.19b is more space-efficient. The packing efficiency,
calculated as the area fraction filled by circles, is 0.785 and 0.907 for the two arrangements.
The densest16 3D packing of spheres is based on stacking 2D layers of the type in Figure
1.19b. A second layer of spheres nestles in the dimples of the first layer and is therefore
laterally shifted relative to it (Figure 1.20). For the third layer, there are two choices for
positioning the spheres. One choice reverses the shift direction so that the spheres in the third
layer lie directly above those in the first; the other choice continues shifting in the same
direction, in which case the third layer does not lie directly above the first layer. When we
view the layers side on (Figure 1.21) the repeat unit is either AB or ABC. The former
arrangement is called hexagonal closest packing (hcp), and the latter cubic closest packing
(ccp). Both sequences fill space equally efficiently (74.0% of space is occupied; 26.0% is voids
between spheres) and would be energetically equivalent for hard spheres. For elements that
adopt these arrangements, such as the solid noble gases and many metals, orbital symmetries
make one or the other arrangement slightly more stable.
Specific combinations of cubic and hexagonal stacking sequences can be stabilized in
some structures; rare-earth metals are typical examples. Figure 1.22 shows how to
analyze the sphere packing occurring in α-La. The first step is to orient the unit cell so
that we have an on-top view of the densely packed layers (against c in hexagonal
structures) and the direction of subsequent shifts is horizontal; as shown in Figure
1.22 bottom. The structure is then rotated 90° so that the normals of the layers point
up. We can then identify the closest-packed sphere layers as being A, B, C, mark the
subsequent shifts as arrows, and assign letter c to a layer that has local cubic environ-
ment and h that has local hexagonal environment according to those shifts. The repeti-
tive sequence of these letters is the basis of the Jagodzinski–Wyckoff notation of stacking
sequences. The Ramsdell symbol gives the number of closest-packed sphere layers per
unit cell of a Bravais lattice and a letter to signify the Bravais lattice (H for hexagonal,
R for rhombohedral, C for cubic). As an example, the Ramsdell symbol for α-La is 4H.

Figure 1.19 Two

high-symmetry pack-
ings of circles. After
Kepler [8].

16
Although these arrangements have long been suggested to be the densest form of packing of equal spheres (J.
Kepler’s conjecture in ref. [8], apparently responding to the English Admiralty task to determine the most efficient
packing of cannonballs on ships), a widely accepted proof was only published in 1998 by Thomas Hales. A panel of
referees was 99% certain the proof was correct.
20 Structures of Crystalline Materials

Figure 1.20 Two choices for placement of the third layer of densely packed spheres. One line of the
subsequent layer shifts is used to analyze a densest packing.

Figure 1.21 Cubic (ccp) and hexagonal (hcp) closest packing of equal spheres. Edges of 8-unit cells are
drawn for ccp, with body diagonals perpendicular to the layers. The two unit cells are below; face-
centered cubic and primitive hexagonal.
 1.4 Structural Principles 21

Figure 1.22 How to analyze sphere packing in hexagonal structures (α-La shown).

Figure 1.23 Spheres in contact in unit cells of primitive and body-centered cubic structures of metals.

In addition to the densest-packed arrangements, two other simple high-symmetry pack-

ings are found in metals and are shown in Figure 1.23; the primitive cubic arrangement on
the left (adopted by α-Po) and the body-centered cubic (bcc) arrangement on the right. Note
that in ccp (Figure 1.21), the spheres are in contact along the face diagonal of the unit cell, in
bcc along the body diagonal, and in the primitive cubic packing along the cell edge. The
preferred structures for metallic elements are given in Figure 1.24.
Although the densest packing adopted by most elemental metals suggests non-directional
bonds, this is true only as a first approximation. It has been shown [9] that bonding electrons
may concentrate in certain directions; in aluminum, they prevail in voids formed by every
four mutually touching atoms of the cubic closest packing.
22 Structures of Crystalline Materials

Figure 1.24 Structure types commonly adopted by metallic elements.

Box 1.2 Nanoscale Concepts: Structure changes!

At constant temperature and pressure, lattice parameters are constant. Except when the crystal
enters the nano-size range. Then the lattice parameter contracts, depending of course on the
crystal’s shape. The culprit is the surface of the nanocrystal. At a surface, only a portion of the
atoms’ bonding ability is utilized, the remaining bonds are open-ended, so-called “dangling
bonds”. The surface layer is therefore at a substantially higher energy level than the bulk.
A system tends to minimize its energy either by increasing the entropy of itself, or of its
surroundings by the bond-formation heat. The surface area is therefore minimized to keep
the amount of dangling bonds to a minimum. The force to minimize the area is proportional to
the excess energy of the surface over the bulk. If high enough, it will compress the entire crystal.
The interplay of the surface and bulk has consequences even for the crystal structure adopted.
While tungsten metal is bcc (like Fe in Figure 1.23), its nanoclusters are ccp, with a face-centered
cubic (fcc) unit cell, when they have fewer than 7000 atoms [10]. Why? Whereas the bcc packing
of spheres fills ~68% of space, fcc fills ~74%. The space volume per atom is then smaller for fcc by
a factor of 68/74. The same atoms would pack a smaller
pffiffiffi fcc sphere, of surface smaller than bcc by
a factor of about (68/74)2/3 = 0.945; exactly ð3=4Þ 3 2. As the proportion of the surface and of its
higher energy content increases with decreasing size of our originally bcc nanoparticle, the
energy that can be released by minimizing the surface eventually exceeds the energy needed to
form the normally less favored but denser fcc tungsten packing in the bulk, hence the transition.

1.4.2 Filling Holes

As suggested in the previous section, voids are left between spheres of the atomic densest
packing. There are two types of voids in hcp and ccp arrangements; tetrahedral holes and
octahedral holes (Figure 1.25). Four spheres surround a tetrahedral hole; three from one layer
touching, one from the next. Six spheres surround an octahedral hole; three from one layer
and three from the next. One octahedral and two tetrahedral holes are present per each sphere.
 1.4 Structural Principles 23

Table 1.5 Binary-compound structure types derived by full or fractional filling of octahedral and
tetrahedral holes in planes between closest-packed layers of A.

Filling in total General Structure type Structure type The plane-filling

of holes formula hcp of anions ccp of anions sequence

All octahedral CA NiAs NaCl All full

½ octahedral CA2 CdI2 CdCl2 Empty and full
½ octahedral CA2 CaCl2 All ½ full
⅓ octahedral CA3 BiI3 YCl3 Empty and ⅔ full
⅓ octahedral CA3 RuBr3 All ⅓ full
⅔ octahedral C2A3 La2O3 Empty, full, full *
⅔ octahedral C2A3 Al2O3 All ⅔ full
All tetrahedral C2A ** Li2O All full
½ tetrahedral CA ZnS wurtzite ZnS sphalerite All ½ full
All C3A Na3As All full
*Lanthanum fills the octahedral holes in a strongly off-center manner that provides bonding across the empty hole
plane with one of the oxygen atoms, by which La achieves coordination number 7. **Repulsion of cations in the
two face-sharing tetrahedra prevents this type of hole filling.

Figure 1.25 Tetrahedral and octahedral holes between layers of ccp and hcp spheres. Note that the label
C of the subsequent ccp shifts is partly obscured.

Structures of many simple compounds can be described in terms of smaller atoms (usually
cations) occupying the octahedral and/or tetrahedral holes in closest-packed arrangements
of larger atoms (usually anions).17 Table 1.5 summarizes several structure types that can be
described in this way. They are discussed in the following paragraphs.
Let’s start a more detailed account by analyzing filling of octahedral holes in densest
packed arrays. The two-step structure-orientation process is shown in Figure 1.26 for the
respective prototypes, NiAs and NaCl. In both, the cation-occupied octahedral holes are

17
The closest-packed spheres (atoms) begin to touch once the radius of the atom in the octahedral holes becomes less
than √2 − 1 ≈ 0.414 of the sphere radius. For tetrahedral holes this occurs when the radius of the smaller atom
becomes less than (√3/√2) − 1 ≈ 0.225 of the closest-packed sphere radius. Hence these radius ratios are the minima
for the given coordination to occur, at least in the hard-sphere approximation.
24 Structures of Crystalline Materials

Figure 1.26 The hcp and ccp prototypes NiAs (left) and NaCl (right). Identification of octahedral holes
(black) by reorienting the structure from a general view to the top-on view (below) and finally to the side-
on view (above) of the closest-packed layers.

Figure 1.27 Coordination of octahedral holes in NiAs and NaCl structures.

located in the intermediate planes between the closest-packed layers of anion spheres, and
the sphere stacking sequence is immediately recognizable.
In NiAs, the closest-packed As atoms have six nearest Ni neighbors forming a trigonal
prism (Ni[6o]As[6p]). In NaCl, both cation and anion arrays have the same arrangement in
space and both are octahedrally coordinated; Na[6o]Cl[6o]. Either of them can therefore be
considered closest packed, though this description better suits the larger Cl− anion. Figure
1.27 illustrates how the Ni coordination octahedra in NiAs share opposite faces (in addition
to sharing all edges), whereas the octahedra in NaCl do not (only edges are shared). In
strongly ionic compounds, the NaCl-type structure is therefore preferred over NiAs in order
to avoid the added electrostatic repulsion across the shared octahedral face.
 1.4 Structural Principles 25

Figure 1.28 Composition CA2 via filling octahedral holes in hcp: CdI2 (fully filling every second plane of
holes) and CaCl2 (half filling every plane).

Composition CA2 with “cations” C in half the octahedral holes is achieved in two ways;
either by alternating one empty and one filled plane of octahedral holes, or by half filling each
plane (Table 1.5). Figure 1.28 compares these two arrangements in the hcp array; the
alternately filled arrangement is called the CdI2 structure type18 and the homogeneously half-
filled one is the CaCl2 structure type.19 Note that the symmetry of the CaCl2 structure is
reduced to orthorhombic by the rectangular pattern of holes.
The alternating empty and filled octahedral planes create a layered structure because the
anions face each other across each empty plane of octahedral holes.20 The polyhedral
representation in Figure 1.29 top shows two types of such isolated layers; a CA2 layer that
has all octahedral holes filled and a CA3 layer with two-thirds of the holes filled. Surprisingly,
these two octahedral layers give rise to four structural arrangements. This is because each can
stack in either an hcp or ccp array.
The two-thirds filling of every plane of octahedral holes in hcp also occurs in the structure
of corundum, Al2[6o]O3[4t], an important refractory material. Figure 1.30 shows this two-
thirds filling and how it subdivides the NiAs-type infinite columns of octahedra into pairs.
The two octahedra share one face and their central Al3+ cations repel each other due to the
proximity of their ionic charges.
Let’s now focus on filling tetrahedral holes. A complete filling of all the tetrahedral holes in
a ccp array leads to the composition C2A of materials such as Li2O. However, the traditional

18
Perversely, CdI2 tends to adopt numerous stacking sequences rather than this simple hcp structure.
19
The structure of rutile, TiO2, is related, but with anions distorted such that they no longer lie in perfect layers. TiO2 is
then better understood in terms of corner- and edge-sharing octahedra (see Figure 1.45).
20
The layers are held together by weak electrostatic forces of electric dipoles generated by random fluctuation of the
electronic charge in one object, such as the anionic layer, generating a corresponding inverse-charge fluctuation in
the other such object (van der Waals forces, see Section 5.1.3).
26 Structures of Crystalline Materials

Figure 1.29 Top: Polyhedral representation of a single layer of octahedral holes, fully occupied (left) and
two-thirds occupied (right). Bottom: Structure types formed when these layers are packed such that their
anions form a ccp or hcp array.

Figure 1.30 Sequence of octahedral-hole filling in hcp oxygens in corundum. The hole plane is two-thirds
filled (left; every third hole is empty). All hole sequences in a straight line along c are full, full, empty (side-
on view on right). The formed Al2[6]O3[4] has pairs of octahedra sharing faces, with repulsion of their
central Al3+ ions.
 1.4 Structural Principles 27

Figure 1.31 CaF2: Identification of tetrahedral holes (black) in the side-on view of the ccp layer stacking
of Ca atoms (gray large spheres). Notice a tetrahedral hole both above and below each densest-packed Ca
atom, hence CaF2.

Figure 1.32 Identification of tetrahedral-hole planes (black dots) in the side-on view of the hcp (left) and
ccp (right) stacking of two ZnS modifications.

prototype structure used is CaF2,21 and has a cation ccp array with the fluoride anions in all
of the tetrahedral holes. Figure 1.31 illustrates its pattern of densest packing and hole filling.
There are no known structures based on an hcp array with all the tetrahedral holes filled.
This can be understood with reference to Figure 1.25 where the face sharing of tetrahedra
would have the cations unreasonably close (closer than the cation–anion distance). The
prototype structures based on filling half of the tetrahedral holes in hcp and ccp are the
wurtzite and sphalerite (zinc blende) polymorphs of ZnS. Identification of their sphere
packing and hole filling is illustrated in Figure 1.32.
Polyhedral coordination of the filled tetrahedral holes in unit cells of the prototype ccp and
hcp structures is compared in Figure 1.33. More about building up structures from polyhe-
dra is given in Section 1.4.4.
Structures of some ordered metal alloys can be described as densest-packed arrays with
filled holes. An example is the family of Heusler alloys that has a variety of important

21
Consequently, Li2O is considered an anti-fluorite type.
28 Structures of Crystalline Materials

CaF2 ZnS sphalerite ZnS wurtzite

S
Ca S

Figure 1.33 Coordination of tetrahedral holes in CaF2, sphalerite, and wurtzite.

Figure 1.34 Unit cells of the XYZ and X2YZ Heusler alloys.

magnetic and electronic properties. Two closely related types form, XYZ and X2YZ, the
former called half-Heusler alloys. Figure 1.34 shows that both are derived from a ccp of
X atoms in which Y and Z occupy tetrahedral holes in an ordered manner, giving the XYZ
structure. In the structure of X2YZ, the added X fills all octahedral holes.

1.4.3 Network Structures

For many materials, the directional character of the chemical bonding around each atom in
the structure is of prime importance. The most natural way to view such structures is to
consider them as networks. This often helps to reveal the relationships and similarities
among many simple structures. In more complex materials, the field of coordination poly-
mers (Chapter 14) has driven developments in the nomenclature [11, 12], taxonomy [13], and
classification [14] of networks. Here, we’ll consider only the most elementary terms.
Networks with a single type of vertex are uninodal networks; all vertices are N-connected.
An example of a 3-connected uninodal network is graphene (the single sheet of graphite,
Figure 1.35), in which carbon forms three sigma bonds and the fourth valence electron is
delocalized.
Networks with low coordination numbers may be characterized by a vertex symbol [11].
The vertex symbol contains one post for each angle that occurs at the vertex. The post is
 1.4 Structural Principles 29

Vertex symbol 616161

Three angles at the vertex, each angle is part of one 6-membered ring

Figure 1.35 A 3-connected uninodal network of graphene.

Figure 1.36 Two 3-connected networks of polysilicide anions in CaSi2 (puckered planar) and SrSi2 (3D).
One 10-membered ring is highlighted with a dashed line.

a number with a subscript. The number gives the size of the smallest ring that a particular
angle at the vertex is part of. The subscript identifies how many rings are joined at this
particular angle. As an example, the vertex symbol for graphene has three posts because
there are three angles at the vertex. Each angle forms part of only one smallest, 6-membered,
ring so the vertex symbol is 616161.22
Let’s consider the silicide anions in CaSi2 and SrSi2 as other examples of 3-connected
networks. Using the generalized 8−N rule (Section 1.3.2), both compounds have VECA = 5,
and the three missing electrons are obtained by forming three Si–Si single bonds at each Si. The
net representing the silicide-anion network on the left of Figure 1.36 has the same vertex
symbol 616161 as graphene, but differs in that it is puckered due to accommodation of one
nonbonding electron pair at each Si. The standalone crystal-chemical formula for this anion

22
Following Wells, a symbol (6,3) is sometimes used for this honeycomb of 6-membered rings and 3-connected
vertices. Do not confuse these Wells symbols with the N,M-connected binodal nets!
30 Structures of Crystalline Materials

Figure 1.37 Two 4-connected uninodal nets—cubic and hexagonal diamond.

would be 2∞ Si−, where the subscript and superscript to the left denote its infinite 2D nature. The
network in SrSi2 on the right of Figure 1.36 has vertex symbol 105105105. There are three angles
at the vertex, and each angle is a part of five 10-membered rings in this 3∞ Si− 3D network.23
Examples of 4-connected nets are shown in Figure 1.37. Both concern elemental carbon
forming uninodal nets of the vertex symbol 626262626262. The symbol has six posts because
there are six angles at each tetrahedral vertex, where each angle is part of two 6-membered
rings. For uninodal networks with higher connectivity, we have already encountered in
Figure 1.23 the primitive cubic structure of polonium, which is a 6-connected network,
and body-centered cubic α-Fe that can be considered as an 8-connected network.
A binodal network has two different types of vertices. When characterized using the
½N; a ½M;
crystal-chemical formula for a binary phase, Ccþ m An , two different vertices are seen,
an N-connected vertex C and an M-connected vertex A. We say that a binodal network has
an N,M-connected net. The relationship between uninodal and binodal nets is instructive for
understanding similarities among structures.
One type of network-based similarity is site ordering.24 In its simplest case, identical
vertices of a uninodal network become occupied by two different atoms in an ordered
manner, forming a binodal network. Three examples of this relationship are shown in
Figure 1.38. Site ordering removes the equivalence of sites that were symmetry related,

23
The only 3-connected 3D network that has all three bonds and angles around the vertex equal.
24
One of the homeotypical relationships between two structures defined in ref. [2]. Another such homeotypism is the
formation of distortion variants upon decrease in the space-group symmetry.
 1.4 Structural Principles 31

Figure 1.38 Examples of site ordering of uninodal networks (top) to create binodal networks (bottom).

which means that the ordered structure, a superstructure, has a lower point symmetry (e.g.
d3m to 43m; as in diamond to sphalerite in Figure 1.38), or translational symmetry (the unit
cell is multiplied; as in Po to NaCl in Figure 1.38), or both. If the site ordering multiplies the
original cell, the new cell is called a supercell.
A second important relationship is network expansion where a linker is placed between
a pair of vertices [11]. For example, we can take the diamond-type net of elemental silicon
and put an oxygen atom mid-way between each node, obtaining the high-temperature
structure of cristobalite, SiO2. This and other examples are shown in Figure 1.39 on the
same set of uninodal nets as in Figure 1.38. The linker does not have to be a single atom; the
primitive cubic 6-connected net expanded with a −C≡N− linker is the generic ingredient of
the Prussian-blue-type structures.
A final relationship worth mentioning is vertex decoration—replacing a vertex with
a group of vertices. When this group is a cluster, it is common to describe this as network
augmenting [11]. An example is the relationship between CaTe and CaB6 in Figure 1.40.
CaTe itself is an example of a material with the CsCl-type structure, which in turn can be
considered as a site ordering of the α-Fe bcc array of Figure 1.23.
Vertex decoration and network expansion may occur simultaneously. This is a good
approach for visualizing open metal–organic frameworks (MOFs). The example in Figure
1.41 starts with oxygen atoms in a cubic Po-type network, which are “decorated” by zinc to
[Zn4O]6+ and then linked with the terephthalate anion [O2CC6H4CO2]2−. This forms a ReO3-
type network of these cations and anions.
32 Structures of Crystalline Materials

Figure 1.39 Examples of network expansion from uninodal to binodal networks.

Figure 1.40 Decoration (augmenting) of anion sites in the CsCl-type structure.

In some cases, two interpenetrating nets run throughout a structure without ever crossing
each other. Cu2O is one such example. In the right-hand side of Figure 1.42, we see two
2,4-connected nets of cristobalite type which never intersect. In each net, CuI adopts its
preferred coordination number, 2, and oxygen is tetrahedrally coordinated, Cu2[2l]O[4t].

1.4.4 Polyhedral Structures

In previous sections, we have seen various approaches that help visualize extended struc-
tures. In functional materials, however, the local chemistry is often the key to desired
properties. Coordination polyhedra are an efficient way to illustrate these local bonding
environments, and, often, to visualize the entire structure. We’ll find in Chapter 6 that
focusing on coordination polyhedra is also a convenient start for developing electronic band-
 1.4 Structural Principles 33

Figure 1.41 Combined site decoration and network expansion in a Zn4O(O2CC6H4CO2)3 MOF.

Cu
O

Cu2O

Å
Wyck. x y z
2a 0 0 0
Cu 4b

Figure 1.42 Two interpenetrated nets (one solid, one dashed) in cuprite, Cu2O.

structure ideas and understanding certain properties. In Section 1.5.4 and in Chapter 14,
we’ll discuss how SiO4 tetrahedra are the essential building block of many silicate structures.
Building from the ideas in Section 1.4.3, Table 1.6 contains examples of common coord-
cþ ½N; a ½M;
ination polyhedra for the “cation” and “anion” in the Cm An compounds of binodal
N,M-connected nets. Many of the structures that were earlier described as based on filling
octahedral or tetrahedral holes are also found in this table.
Given the chemical preference to form a certain coordination polyhedron, can we make
any generalizations how such equal polyhedra connect into networks? Let’s take SiO2 as an
34 Structures of Crystalline Materials

Table 1.6 Common coordinations in some binodal N,M-connected nets.25

N,M Cation coordination Anion coordination Example

4,2 Tetrahedron Linear (bent) SiO2 (in quartz, cristobalite, tridymite)

4,3 Tetrahedron Triangle Si3N4
4,3 Square Triangle Pt3O4
4,4 Tetrahedron Tetrahedron ZnS (in zinc blende, wurtzite)
4,4 Square Tetrahedron PtS
4,4 Square Square NbO
6,2 Octahedron Linear ReO3
6,3 Octahedron Triangle TiO2 (in rutile, anatase, brookite)
6,4 Octahedron Tetrahedron Al2O3 (in corundum)
6,6 Octahedron Octahedron NaCl
6,6 Octahedron Trigonal prism NiAs
8,4 Cube Tetrahedron CaF2
8,8 Cube Cube CsCl

example and assume tetrahedral coordination of Si. If both oxygens are equivalent, the
coordination number of O will be 2 (see the connectivity-balance argument in Section 1.3.1),
and the composition of this polyhedral network can be conveniently described as SiO4/2. This
is an example of a Niggli formula, in which every polyhedral vertex is listed by its connectiv-
ity. Accordingly, SiO4/2 is a silicon dioxide with four 2-connected O vertices of the polyhe-
dron around Si, meaning that each of the four oxygen vertices is shared with another
tetrahedron to form a network of corner-sharing tetrahedra.
When the vertices do not have the same connectivity, the subscript in the Niggli formula is
written as a sum of fractions. Various connectivities can therefore be combined to give the same
composition. SiO4/2, SiO3/3+1/1, and SiO2/4+1/2+1/1 all refer to SiO2 built of tetrahedra, but of
different connectivities at the vertices.26 Yet, looking at known CA2 phases, we do not find the
latter two structures. Why? Sharing three or four tetrahedra at a common vertex is not prohibi-
tive; we do find 3-connected tetrahedra in Si3N4 (Niggli formula SiN4/3) or 4-connected tetrahe-
dra in SiC (Niggli formula SiC4/4). Empirically, we can conclude that it is favorable for the oxygen
vertices in SiO2 to have the same bonding environments. When an identical environment is not
possible, atoms of the same element usually prefer their environments to be as similar as possible.
This preference was postulated by Linus Pauling [3] as the rule of parsimony: “The number of
essentially different kinds of constituents in a crystal tends to be small.”
While this rule aids estimates of polyhedral connectivity in many compounds, does it
predict the structure type? In general it does not; several possible spatial arrangements
25
Note that geometries may depart slightly from ideal depending on local site symmetry. Examples do not necessarily
represent the only networks that can occur for a given N,M-connected net.
26
SiO4/2 means four 2-connected vertices, SiO3/3+1/1 three 3-connected and one 1-connected vertex, and SiO2/4+1/2+1/1
two 4-connected, one 2-connected, and one 1-connected vertex in the tetrahedron.
 1.4 Structural Principles 35

Figure 1.43 Structures made of 2-connected tetrahedral vertices.

usually exist. We can see this by considering just a few of the structural variants for the
4,2-connected binodal network of CA2 materials. Both the quartz and cristobalite structures
contain corner-sharing tetrahedra but have different 3D arrangements. The OCu2 structure
(Figure 1.43) is related to cristobalite, but more complex due to the interpenetration of two
networks shown in Figure 1.42. The HgI2 structure also has vertices shared by two tetrahedra
but is layered (Figure 1.43).
Let’s consider next the structural possibilities for CA6 octahedra sharing vertices, common
in functional materials. Starting with the highest n/m composition ratio for CmAn, CA6 in
octahedral coordination represents an isolated octahedron. CA5 can be written as CA2/2+4/1
where there are two 2-connected and four 1-connected vertices.27 Such structures contain
variously oriented infinite corner-connected octahedral chains. CA4 requires four octahedral
vertices shared (CA4/2+2/1), and various clustered chains, layers, or networks are possible, as
illustrated in Figure 1.44. With all six vertices shared, we have formula CA3 (CA6/2) and the
octahedral network found in ReO3 and ccp perovskites (Section 1.5.3). Other connectivity
patterns are possible if we allow octahedra to share edges instead of just vertices. We’ve already
encountered this in the YCl3/BiI3 structures of Figure 1.29, which contain octahedra that share
three edges of six vertices shared by two octahedra each, giving again the Niggli formula CA6/2.
As the n/m ratio decreases below three, the average anion coordination number must
continue to increase, and shared edges become a necessity. There are several paths to
composition CA2. The octahedra must share at least two edges, and the simplest way to do
this yields the rutile structure of TiO2 that contains infinite chains of octahedra sharing two
opposite edges with the two remaining corners linking the chains together (Figure 1.45). The
brookite and anatase polymorphs of TiO2 are other possibilities.
In general, edge and face sharing is more frequent for octahedra than for tetrahedra. This
is because shared edges between CA4 tetrahedra lead to very short distances between cations.
Face-sharing tetrahedra are not known (we have already encountered the rule that only half
27
Two of six octahedral vertices are shared between two octahedra. Or, stated in the language of coordination
chemistry, there are two bridging and four terminal ligand atoms.
36 Structures of Crystalline Materials

Figure 1.44 Examples of structures with identical octahedra sharing four vertices.

Figure 1.45 Edge- and corner-sharing octahedra in TiO2 modifications. Two unit cells are drawn for
rutile and anatase.
 1.4 Structural Principles 37

Box 1.3 Nanoscale Concepts: Polymorphism of TiO2 nanocrystals

As the proportion of the crystal’s surface atoms versus bulk atoms increases towards the
ultimate unity for the absolutely smallest cluster, the proportion of the surface’s higher energy
content also increases. Consider TiO2 (Figure 1.45). The surface and bulk energies of its three
modifications are as follows [15]:

TiO2 Surface energy (J/m2) Relative bulk energy (kJ/mol)

Anatase 0.4 2.6

Brookite 1.0 0.7
Rutile 2.2 0

Rutile has the most stable bulk, yet its surface has the highest energy (sharing fewest
octahedral edges in its bulk yields the most dangling bonds at the surface). Below particle
sizes of about 35 nm [16], the increasing surface-energy proportion destabilizes the normally
stable rutile in favor of brookite. Below 11 nm [16], TiO2 adopts the anatase structure that
shares all octahedral edges and has the fewest dangling bonds at the surface. In the opposite
direction, this affects the crystallization of TiO2, as detailed below.
Since crystals grow from their smallest seed (called the nucleus), TiO2 obtained by precipita-
tion from an acidic TiCl4 solution is anatase (instantaneous formation, smallest crystallites).
This complies with Ostwald’s step rule—the least stable polymorph often crystallizes first [17].
The metastable anatase will turn into the stable rutile only after prolonged aging that allows the
crystallites to join and recrystallize. The size and structure of nanoparticles synthesized from
solutions depends on the extent of this aggregative growth, called Ostwald ripening, in which
the tiniest crystals dissolve and regrow on the surface of larger crystals, ultimately minimizing
the surface-energy proportion of the system.

of tetrahedral holes in an hcp can be filled). In contrast, face-sharing octahedra are found in
NiAs (Figure 1.26), corundum (Figure 1.30), the hexagonal perovskites (Figure 1.54), and
other structures.
At least one coordination polyhedron exists for every coordination number. As the
coordination number increases, several polyhedra become possible; such as the cube, square
antiprism, and dodecadeltahedron for coordination number 8, or cuboctahedron, anti-
cuboctahedron, and icosahedron for 12. However, upon further increase, the number of
known structural examples quickly decreases. One of the highest coordination numbers
occurs in the SmCo5-type structure of superferromagnets. Each Sm in this alloy is sur-
rounded by 18 Co atoms that form a 6-capped hexagonal prism. The six capping Co
atoms are 3-coordinated, and the twelve prismatic Co atoms are 4-coordinated, giving
a Niggli formula SmCo6/3+12/4 (Figure 1.46).
38 Structures of Crystalline Materials

Figure 1.46 Polyhedral representation of the crystal structure of SmCo5.

1.5 Structures of Selected Materials

1.5.1 The Spinel Structure
The mineral spinel, MgAl2O4, gives its name to a structure type in which two metals of
similar atomic sizes are accommodated in tetrahedral and octahedral sites in a ccp array of
anions. A simple formula to remember is [tetrahedron]1[octahedron]2O4. A detailed ana-
lysis reveals that one-eighth of the tetrahedral and a half of the octahedral ccp holes are
occupied in a non-trivial stacking sequence. Table 1.7 illustrates the ubiquity of spinels.
Two limiting types of site occupation occur. The first is MgAl2O4 with each ion at its
respective site, [Mg2+][4t][Al3+Al3+][6o]O4, termed a normal spinel. In the inverse spinel, half the
octahedral ions are exchanged with the tetrahedron, as in magnetite [Fe3+][4t][Fe2+Fe3+][6o]O4.
Magnetite forms an inverse spinel due to the ligand-field stabilization energy (LFSE) of the
high-spin d 6 Fe2+ being larger at octahedral than at tetrahedral sites, while high-spin d 5 Fe3+
has no LFSE. LFSE of d 8 Ni2+ makes [Ga3+][4t][Ni2+Ga3+][6o]O4 an inverse spinel, and LFSE
of d 4 Mn3+ stabilizes normal spinel [Mn2+][4t][Mn3+Mn3+][6o]O4 with no LFSE from
d 5 Mn2+. A spinel of similar ions of no LFSE may show an intersite disorder, such as
[Mn2+1−xFe3+x][4t][Fe3+2−xMn2+x][6o]O4, depending in general on several size- and bond-
ing-related factors.
The connectivity of the cation coordination polyhedra in the spinel structure is hard to
visualize, but one key feature is the network of edge-sharing octahedra (Figure 1.47, left).
The visualization in Figure 1.47 right is useful for rationalizing the magnetic and electronic
properties of spinels that are discussed in more detail in Chapters 9 and 11.
Table 1.7 Oxidation-state combinations for spinels.

Tetrahedron Octahedron Chemical formula of an example

1 3 4 LiMn2O4 (a = 8.245 Å)
2 3 3 ZnFe2O4 (a = 8.442 Å)
2 4 2 Fe2TiO4 (a = 8.521 Å)
3 2 3 Fe3O4 (a = 8.394 Å)
4 2 2 Ni2SiO4 (a = 8.045 Å)
5 1 2 LiZnNbO4 (a = 6.082, c = 8.403 Å)
6 1 1 Na2WO4 (a = 9.108 Å)
 1.5 Structures of Selected Materials 39

Figure 1.47 Left: Coordination polyhedra of cations within the unit cell of spinel. Right: The unit-cell
content in its simplest representation (the cubes are empty).

1.5.2 The Garnet Structure

The garnet structure type is found in many important magnetic, optical, and magneto-
optical materials. Many solid state lasers use yttrium aluminum garnet (YAG) doped with
~1% Nd as the active laser medium. Garnet is a complex, high-symmetry structure, and the
simplest formula to remember is [cube]3[octahedron]2[tetrahedron]3O12. Table 1.8 shows
that a variety of metal atoms can be accommodated at these sites. Intersite disorder is again
common, particularly for garnet minerals like pyrope, Mg3Al2Si3O12, almandine,
Fe3Al2Si3O12, and others.
The crystal structure of garnet is not easy to visualize. The crystal-chemical formula
together with a polyhedral illustration (Figure 1.48) provides some idea about coordinations
and connectivities. The tetrahedra connect to octahedra by corner sharing, whereas three
tetrahedral and all six octahedral edges are shared with the rather deformed cubes. As in
spinel, all oxygens have a distorted tetrahedral coordination.

Table 1.8 Oxidation states in garnets [cube]3[octahedron]2[tetrahedron]3O12.

Cube Octahedron Tetrahedron Examples

2 6 2 Ca3Te2Zn3O12 a = 10.930 Å
3 3 3 Y3Fe2Fe3O12 a = 12.376 Å
2 3 4 Mg3Al2Si3O12 a = 11.459 Å
1 3 5 Na3Sc2V3O12 a = 10.913 Å
40 Structures of Crystalline Materials

Figure 1.48 Coordination polyhedra in one-eighth of the garnet unit cell, which corresponds to a single
formula unit.

1.5.3 Perovskite Structures

The mineral perovskite, CaTiO3, lends its name to a vast family of structures where adjoined
MX6/2 octahedra form 12-coordinated voids that can be filled by larger atoms A so that the
composition becomes AMX3. Perovskites can also be viewed in terms of closest-packed AX3
layers in which M occupy one-quarter of the octahedral holes—those that are formed solely by
X. Although ccp is most common, perovskites with hcp of AX3 layers are also known. Figure
1.49 illustrates the ccp case on an ideal cubic perovskite where M fits the octahedral holes exactly.

Figure 1.49 Octahedral network in an ideal cubic perovskite (origin at A is chosen).

The ccp array will be more stable than hcp until the increasing size of A prevents the MX6/2
octahedra from linking at corners. On the other hand, much smaller A atoms than this
maximum can be accommodated while keeping the corners linked. This can be evaluated
with the Goldschmidt tolerance factor [18] (Figure 1.50) in terms of the ionic radii of A, M,
and X. In particular, the Shannon radii [19] for 6-coordinated M, 12-coordinated A, and
2-coordinated O are consistent with Goldschmidt’s size considerations: Cubic perovskites, of
ideal ratio of the AX and MX bond lengths AX/(MX√2) = 1 as in Figure 1.50, top left, are
 1.5 Structures of Selected Materials 41

Figure 1.50 Tolerance factor t and two ways it can become smaller than unity.

stabilized over the t range ~1.04 to ~0.98 [20] by minor bonding compromises. When t ≳ 1.04,
the hcp stacking is often observed. In the t range between ~0.99 and ~0.83, perovskites are
stabilized by structural distortions as discussed below.
Figure 1.50 illustrates two scenarios for t < 1: either it is due to M being too large and
expanding the MX6 octahedron, or due to A being too small so that A and X are no longer in
contact. The drawing shows that both cases effectively mean that the A atom is smaller than
one that would fit exactly. The structure responds to this size mismatch by a coupled rotation
of the corner-linked MX6/2 octahedra, which expands them in a fixed frame of constant unit
cell, accommodating the large M and bringing X closer to A. This octahedral tilting can
profoundly influence the magnetic and electronic properties of perovskites.
While symmetry analysis [21, 22] is the most rigorous approach to describe the various
types of tilting, a simple combinatorics of these coupled rotations is the basis for the Glazer
tilt classification [23]. It considers a model of linked octahedra and evaluates all combin-
ations of rotations along the three 4-fold axes of the octahedron in Figure 1.51. In the
network, twisting the central octahedron anticlockwise mechanically requires its in-plane
neighbors to twist clockwise in a cooperative manner. The plane of octahedra below the one
under consideration, however, is free to rotate in either the same or the opposite sense. Let’s
start with the simplest case of no rotation at all. It is given a three-letter symbol a0a0a0
because equal rotations about different axes of the octahedron are symbolized by use of the
same letters. A uniaxial rotation is then denoted with “c” as the last letter (a0a0c+ and a0a0c−
in Figure 1.51) where superscripts + or − specify whether the octahedra in adjacent layers
rotate in the same direction (+, in phase) or in opposite direction (−, out of phase). For
biaxial rotation of the octahedron, the letter “b” is added with a superscript sign, such as in
42 Structures of Crystalline Materials

a⁰ a⁰ c+ a⁰ a⁰ c–

Figure 1.51 Octahedron can rotate about each of its fourfold axes (center). Uniaxial rotations: In-phase
between successive planes (left), out-of-phase (right).

a0b+c+ (two unequal rotations in phase), for triaxial rotation all three letters have superscript
signs, such as in a+b+c+ (three unequal rotations in phase). Since for equal rotations about
different axes the same letter is used, we may have a0b+b+ (two equal rotations in phase) or
a−a−b+ (two equal rotations out of phase) or a+a+a+ (three equal rotations in phase). In total,
this combinatorics yields 23 Glazer tilt systems in ref. [23].
Symmetry analysis [21] shows that there are 15 unique Glazer tilts28 (some tilts also do not
form), but three are particularly frequent [24]. One of them is a0a0a0 for the cubic perovskite
of Pm3m symmetry. SrTiO3 is one of many examples. Another frequent tilt is a−a−a−, which
lowers the symmetry to rhombohedral, R3c. This tilt is equivalent to a single rotation of an
octahedron around its threefold axis. Figure 1.52 shows this rotation for LaNiO3 and views
of the structure along all three 4-fold axes of the octahedron. These three projections
illustrate that all three tilts are equal and out of phase. The high symmetry extends to the
coordination polyhedra, fulfilling Pauling’s parsimony rule and providing the most favor-
able ionic bonding of all tilted perovskites. This becomes important as the charge number of
the A cation increases. The effect this tilt has on the A-atom coordination is to bring three of
the twelve X atoms closer to A while moving three others away, hence approaching the
coordination number 9. As the progressing rotation contracts only three of these nine A–X
bonds, the range is limited, and the a−a−a− tilting is rare when t < 0.97.
The most common tilt is a+b−b− that lowers the symmetry to orthorhombic Pnma. It again
accommodates the smaller-than-optimal atom A by bringing some of the twelve X neighbors
closer to increase their bond strength more than the rest of them lose it (Section 5.4). An
example in Figure 1.53 is LaFeO3, where La has eight close O neighbors and the remaining
four at a longer distance. Unlike a−a−a−, this tilt allows A to deviate from the center of its
coordination polyhedron and is therefore prevalent among perovskites with t < 0.97.

28
The reduction from 23 to 15 tilts can be understood by careful consideration of the relationship between symmetry
elements and the magnitude of the tilts imposed by Glazer’s notation. The number of unique tilts drops from 23 to
17 upon realizing that symmetry cannot restrain tilts of a different sense (in-phase versus out-of-phase) to be of the
same magnitude. As an example, a+a−a− has the same symmetry as a+b−b−, and the latter thus covers both cases.
Even if you forced the in-phase tilt to be of equal magnitude to the out-of-phase tilts, there would be no symmetry
element that constrained it to be equal, hence, in a real crystal, it would never be more than approximately equal.
For a full explanation of how symmetry reduces the number of allowed tilt systems, see ref. [21].
 1.5 Structures of Selected Materials 43

Figure 1.52 Rotation of Glazer’s tilt a−a−a−. The actual physical rotation of eight corner-linked
octahedra as viewed looking down the threefold axis (top right) and as separated into rotations around
the three tilt axes of the octahedron (bottom).

Figure 1.53 Glazer’s tilt a+b−b−. The octahedron rotates around all three of its axes, in phase around one
axis and out of phase around the remaining two.
44 Structures of Crystalline Materials

Figure 1.54 Left: Comparison of ccp perovskite with a polytype of (hc)2 stacking sequence. Right: The 4H
BaMnO3 phase as an example of the latter.

Perovskites with hcp packing of AX3 layers are less common, because the ABAB layer
repetition means that the MX6/2 octahedra share faces, forming columns. In between these
columns there is a lot of space to accommodate large soft A atoms such as barium. These so-
called hexagonal perovskites are therefore formed when the Goldschmidt tolerance factor
t increases significantly over 1, that is, when the A cation is too large for its site. Initially, ordered
intergrowths of ccp with hcp occur. Figure 1.54 gives an example of 4H BaMnO3 (t = 1.11) that
alternates ccp and hcp sequences. A completely hexagonal stacking is found for BaNiO3, with
t of 1.13, the structure of which contains infinite columns of face-sharing octahedra.
The compositional variety of perovskites is extremely rich. Table 1.9 shows examples of
cubic AMX3 perovskite-type structures for diverse oxidation-state combinations. At the
bottom of the table, the anti-perovskites even contain a large cation at the X site. Several
perovskites have the A site vacant; ReO3 (our cover star) is cubic while the related WO3 has
its WO6/2 octahedra tilted and distorted.

1.5.4 Silicates
Silicates are multicomponent oxides based on silicon–oxygen frameworks made up of
predominantly corner-sharing tetrahedra. Silicon in these tetrahedra may in part be replaced
 1.5 Structures of Selected Materials 45

Table 1.9 Oxidation-state combinations for AMX3 perovskites.

A M X Examples (cubic or nearly cubic chosen)

+4 +5 −3 ThTaN3 a = 4.02 Å
+3 +3 −2 LaAlO3 a = 3.82 Å
+2 +4 −2 SrTiO3 a = 3.90 Å
+1 +5 −2 CsIO3 a = 4.67 Å
+6 −2 □ReO3 a = 3.73 Å
+2 +1 −1 BaLiF3 a = 4.00 Å, BaLiH3 a = 4.02 Å
+1 +2 −1 KMgF3 a = 3.97 Å, (CH3NH3)PbBr3 a = 5.93 Å
+3 −1 □ScF3 a = 4.00 Å
−1 −2 +1 AuORb3 a = 5.50 Å, ISAg3 a = 4.90 Å
−3 −3 +2 SbNCa3 a = 4.85 Å, AuNCa3 a = 4.82 Å
−4 −2 +2 GeOCa3 a = 4.73 Å

by a neighboring element, typically aluminum. Very few silicates contain Si in octahedral

coordination,29 very few feature edge-sharing tetrahedra, and the oxygen vertices are gener-
ally 2-coordinated. Silicates are the most common minerals in the Earth’s crust and represent
about one-third of known inorganic crystal structure types. Although some crystalline
silicates, in particular zeolites, have use as functional materials, silicate materials find
applications predominantly due to their mechanical properties. The oldest examples of
artificial composite materials are silicates such as porcelain, the exquisite properties of
which are caused by a texture of fibrous crystals embedded in a glassy matrix.
It’s impossible to cover the vast array of silicate structures here; instead we’ll focus on simple
ideas and nomenclature in common use. We’ve already seen one way of understanding silicate
structures in Section 1.4.3, where we described how the cristobalite modification of SiO2 is
derived via expanding the cubic-diamond network of Si and the how the tridymite modifica-
tion of SiO2 is derived from lonsdaleite. Along this line of thinking, replacement of
every second Si in cristobalite with Al and compensation of the thus-formed negative charge
by placing Na+ into the cavities in the tetrahedral network leads to carnegieite NaAlSiO4. In
a similar way, tridymite can be expanded to the structure of nepheline (Figure 1.55). Such
homeotypism occurs also between the quartz modification of SiO2 and β-eucryptite, LiAlSiO4,
which is a low-expansion material commonly used as a catalyst support in car exhausts.
The coarsest chemical categorization of silicate anions is based on the network dimensional-
ity. Silicates are classified as oligo- and cyclo- for finite polyanions, catena- for chains, phyllo- for
layered anions, and the general name of tectosilicates is used for 3D networks.30 Examples of
1D and 2D networks are in Figure 1.56. Their chemical compositions can be determined from

29
Common only at high pressure in phases such as MgSiO3.
30
From Greek, “olígos” few, “phýllo” leaf, “tekton” builder; and Latin, “catena” chain.
46 Structures of Crystalline Materials

Figure 1.55 Homeotypism between tridymite and nepheline.

Figure 1.56 Examples of cyclo silicate, catena silicate, and phyllosilicate anions.

the vertex connectivities with Niggli formulas. As an example, all the anions in the top row of
Figure 1.56 have the same elemental composition. All have silicon coordination tetrahedra with
two terminal and two shared oxygens, (SiO2+2/2)2–. It is clear that, for infinite polyanions,
neither the chemical formula nor the chemical name is very informative of the structural
arrangement. A more detailed and informative nomenclature is given in Appendix C.

1.5.5 Zeolites
From the discussion above, it should be clear that the connected tetrahedra in silicates may
form an almost infinite number of shapes. One technologically important subgroup of
tectosilicates is the zeolites. Zeolites contain interconnected cages made of SiO4/2 and
AlO4/2 tetrahedra, with pores in between them where guest ions, atoms, or molecules can
be accommodated. Materials with pores of 2.5–20 Å in diameter are referred to as micropor-
ous, those with sizes 20–500 Å, mesoporous. Owing to the ease with which non-framework
chemical species can be exchanged on the large inner surface area, zeolite properties can be
 1.5 Structures of Selected Materials 47

Figure 1.57 Graphical representations of the alumosilicate cages in sodalite.

tuned to fit many technological applications (discussed in Chapter 14). One common zeolite
cage is the sodalite cage, built from 24 tetrahedra as shown in Figure 1.57. The visualization
of the interconnected zeolite cages is greatly simplified by omitting the tetrahedra and
connecting only their central atoms with a straight line (removing O atoms, the inverse of
network expansion).
The connectivity of the cages is accomplished by sharing (Figure 1.57) or by connecting
their equivalent faces. When all square faces of the sodalite cage are shared at the Si atom
with other sodalite cages, the sodalite structure is formed and contains small pores that do
not qualify as micropores. However, when all square faces of sodalite cages are connected via
Si–O–Si bonds, forming a neck between the cages, the so-called zeolite A is obtained. When
four of the eight hexagonal faces of the sodalite cage are connected to other sodalite cages,
zeolite X/Y (faujasite) is formed (Figure 14.2). Owing to important industrial applications,
a zeolite nomenclature [25] has been adopted by the International Union of Pure and
Applied Chemistry (IUPAC).

1.5.6 Zintl Phases

Zintl phases are compounds of two elements one would consider as metals or semimetals
(typically an alkali or alkaline-earth metal and a post-transition metal). Unlike conventional
alloys and intermetallics, Zintl phases have fixed stoichiometry and physical properties that
are atypical for a metal. They are diamagnetic, brittle, and exhibit low electrical conductivity.
Their composition follows the Zintl–Klemm concept of the more electropositive metal
behaving as a cation that provides electrons for the more electronegative metal to form
polyanions in which electrons are shared to complete the octet [26]. The composition can
thus be treated by the generalized 8−N rule of Section 1.3.2. As an example, although we
normally think of thallium as an electropositive metal, it forms a Zintl phase NaTl with the
even more electropositive Na. Because VECA = 4, four electrons per Tl must be shared,
48 Structures of Crystalline Materials

Figure 1.58 The Zintl phases NaTl and SrGa2 with VECA = 4 and AA = 4 form anion networks that are
electronically and structurally related to diamond and graphite.

Figure 1.59 Zintl phases KGe and CaSi with anions isoelectronic to elemental nonmetals.

forming four two-electron Tl–Tl bonds at each Tl. Indeed, the crystal structure in Figure 1.58
has a polyanion network 3∞ Tl− analogous to diamond. Similarly, for SrGa2 (Figure 1.58),
VECA = 4 and four electrons per Ga are shared—three in single bonds, one via conjugation
as in graphite. The crystal-chemical formula of the anion is 2∞ Ga−.
In KGe, VECA = 5, and three two-electron bonds per Ge are formed (Figure 1.59). In the
crystal structure of KGe, germanium occurs in Ge44− cages isoelectronic with molecules of
white phosphorus. Another example is CaSi, where VECA = 6, suggesting two two-electron
bonds per Si (Figure 1.59). In the crystal structure of CaSi, silicon forms polyanion chains 1∞ Si2−,
isoelectronic with the fibrous modification of elemental sulfur, and similar to the structure of
tellurium shown in Figure 1.16.

1.6 Problems
(Note that some require a structure-drawing program.)

1.1 Write down the ⟨111⟩ set of symmetry-equivalent directions in a cubic lattice.
 1.6 Problems 49

1.2 By analogy with Table 1.1, determine the number and type of 2D crystal systems via
considering their possible minimum symmetry elements and sketching their Bravais
lattices.
1.3 Write down indices for the following sets of equidistant planes:

c c c c

b b b b
a a a a

1.4 Sketch a set of equidistant 113 planes in a cubic unit cell.

1.5 Sketch a set of equidistant 113 planes in a cubic unit cell.
1.6 State the Bravais lattice and write down the crystallographic point-group symbol for
a structure of space-group symbol: (a) C2/m, (b) Fmm2, (c) I4/mmm, (d) P312, (e) R3m,
(f) P6m2, (g) F23, (h) P213, (i) Ia3d.
1.7 Is it possible for a c glide plane to have the direction of: (a) a axis, (b) b axis, (c) c axis? If
not, why is this not allowed?
1.8 Write down crystal-chemical formulas for two-site binary compounds CrN, Cr2O3, and
CrO2 with octahedrally coordinated chromium.
1.9 Write down the three balances expressed in the crystal-chemical formulas of the phases
in the previous problem.
1.10 Draw the bond graph for the mineral spinel MgAl2O4 (Figure 1.47).
1.11 Convert the bond graphs in Figure 1.15 into crystal-chemical formulas.
1.12 Describe or sketch a structure for layered V2O5 that is consistent with the bond-graph
representation in Figure 1.15.
1.13 Use the generalized 8−N rule to identify whether anion–anion or cation–cation bonds
are present for the following compounds (assume no cation-localized nonbonding
electron pairs): (a) Na2Tl, (b) SrSb2, (c) BaTe2, (d) InSe.
1.14 Suggest the anion bonding that might occur in MgB2 and MgC2.
1.15 Calculate the percentage of available space that’s taken up by touching spheres in
primitive and body-centered cubic arrangements.
1.16 Imagine a coordination polyhedron with a cation at the center. Now treat the ions as
hard spheres and reduce the size of the cation until the anions just touch. What is the
radius r of the cation for anions of unit radius in the following coordinations: (a) cube,
(b) octahedron, and (c) tetrahedron. Hint: Body diagonal is √3 and face diagonal √2
times the cube edge.
1.17 From unit-cell parameters in figures in this chapter, calculate the following shortest
distances: (a) Na–Cl and Na–Na in NaCl, (b) Ni–Ni in NiAs, (c) Ca–F in CaF2,
50 Structures of Crystalline Materials

(d) C–C in diamond, (e) Ti–Ti in TiO2 (rutile), and (f) Ti–O, Sr–O, and O–O in the cubic
perovskite SrTiO3 (a = 3.90 Å).
1.18 With a structure-drawing program, determine the stacking sequence in Tb (data for
plotting, see Section 1.1.8: P63/mmc, a = 3.068, c = 14.87 Å, Tb in 2b at 0 0 ¼ and 4f at
⅓ ⅔ 0.083). Suggest the Ramsdell symbol and the Jagodzinski–Wyckoff notation.
1.19 Determine the stacking sequence of Br in trigonal CdBr2 (R3m in hexagonal setting; a =
3.965, c = 18.70 Å, Cd in 3a at 0 0 0, Br in 6c at 0 0 0.25).
1.20 In the low-temperature form of CrCl3, every second plane of octahedral holes is
occupied. What is the filling fraction of the occupied planes? Check the result by a side-
on view of the closest-packed layers perpendicular to their subsequent shifts (R3 in
hexagonal setting; a = 5.94, c = 17.3 Å, Cr in 6c at 0 0 ⅓, Cl in 18f at ⅔ 0 0.0757). What is
the type of densest packing?
1.21 La in solid LaBr3 has a tricapped trigonal-prismatic coordination. What is the coord-
ination number of Br? Is it possible that the tricapped trigonal prisms share only
corners, or would you expect sharing of edges and/or faces? Check the result by viewing
the structure and constructing polyhedra around La (P63/m, a = 7.971, c = 4.522 Å. La
in 2c at ⅓ ⅔ ¼, Br in 6h at 0.3849 0.2988 ¼).
1.22 Write the Niggli formula and the simple crystal-chemical formula for the CrO3 struc-
ture that contains chains of corner-sharing chromium-centered tetrahedra.
1.23 Construct a bond graph or Niggli formula to determine if it is possible for all anions to
be equivalent in a structure of tetrahedrally coordinated cations and stoichiometry of
C2A3? Which alternative Niggli formula complies best with the rule of parsimony?
1.24 Using the Niggli formula and the rule of parsimony, determine the stoichiometry that
results from sharing (a) all corners, (b) all edges, and (c) all faces of a cation-centered
cube of anions. Note the structure prototype where you recognize it.
1.25 Write the Niggli formula for C3N4 made of identical CN4 tetrahedra. How many
different types of nitrogen vertices are there? What is the coordination number of each?
1.26 In β-Li3N, nitrogen is 11-coordinated. Write down the Niggli formula of the NLi3
polyhedron.
1.27 ReO3 has a 3D network of octahedrally coordinated rhenium. Determine the
N, M-connectivity for this binodal network
1.28 Identify the type of derivative network relationship between: (a) CaO (Fm3m, a =
4.778 Å, O in 4a at 0 0 0, Ca in 4b at ½ ½ ½) and CaO2 (I4/mmm, a = 3.56 Å, c = 5.95 Å,
Ca in 2a at 0 0 0, O in 4e at 0 0 0.394), (b) SrMoO3 (Pm3m, a = 3.965 Å, Sr in 1b at ½ ½
½, Mo in 1a at 0 0 0, O in 3d at ½ 0 0) and the high-temperature Sr2FeMoO6 phase
(Fm3m, a = 7.93 Å, Sr in 8c at ¼ ¼ ¼, Mo in 4b at ½ ½ ½, Fe in 4a at 0 0 0, O in 24e at
0.253 0 0).
1.29 Which type of similarity do you see between the Laves phase MgCu2 (Fd3m, a =
7.034 Å, Mg in 8a at 0 0 0, Cu in 16d at ⅝ ⅝ ⅝) and the spinel in Figure 1.47?
1.30 Rewrite the chemical formulas in Table 1.7 into simplified crystal-chemical formulas of
spinel.
 1.8 References 51

1.31 Analyze hole filling in the spinel structure of ccp oxygens. (a) Given the general formula
[tetrahedron]1[octahedron]2O4, calculate the total fractions of tetrahedral and octahe-
dral holes filled. (b) Use a structure-drawing program to orient the structure and
identify the type of filled holes between layers of densely packed oxygen atoms. (c)
Determine the sequence and the fraction of each hole filling.
1.32 With a structure-drawing program, determine which of the three most common tilts the
perovskite prototype CaTiO3 (ICSD 62149) adopts.
1.33 Give the Niggli formula for a cyclosilicate anion containing 3 Si.
1.34 What is the formula of the infinite alumosilicate anion in sodalite?
1.35 Use VECA modified for the stable 18-electron configuration of Kr to justify the
network of Cu tetrahedra in MgCu2.
1.36 Suggest the bonding present in the Zintl phase LiAs.
1.37 Suggest the bonding present in the Zintl phase KIn.
1.38 Suggest the bonding present in the Zintl phase CaIn2. Verify your suggestions with the
ICSD and a structure-drawing program.

1.7 Further Reading

A.F. Wells, “Structural Inorganic Chemistry” (1984) Oxford University Press.
B.G. Hyde, S. Andersson, “Inorganic Crystal Structures” (1989) Wiley.
F.A. Cotton, “Chemical Applications of Group Theory” (1990) Wiley.
G. Burns, A.M. Glazer, “Space Groups for Solid State Scientists” 2nd edition (1990) Academic Press.
M. O’Keeffe, B.G. Hyde, “Crystal Structures I: Patterns and Symmetry” (1996) Mineralogical Society
of America.
A. Vincent, “Molecular Symmetry and Group Theory” (2001) Wiley.
R. Tilley, “Crystals and Crystal Structures” (2006) Wiley.
U. Müller, “Inorganic Structural Chemistry” 2nd edition (2007) Wiley.
M. de Graef, M.E. McHenry, “Structure of Materials: An Introduction to Crystallography, Diffraction
and Symmetry” (2007) Cambridge University Press.
C. Giacovazzo, H.L. Monaco, G. Artioli, D. Viterbo, M. Milaneso, G. Ferraris, G. Gilli, P. Gilli,
G. Zanotti, M. Catti, “Fundamentals of Crystallography” 3rd edition (2011) IUCr/Oxford
University Press.
U. Müller, “Symmetry Relationships between Crystal Structures” (2013) IUCr/Oxford University
Press.

1.8 References
[1] J. Foadi, G. Evans, “On the allowed values for the triclinic unit-cell angles” Acta Crystallogr.
Sect. A 67 (2011), 93–95.
52 Structures of Crystalline Materials

[2] J. Lima de Faria, E. Hellner, F. Liebau, E. Makovický, E. Parthé, “Nomenclature of inorganic

structure types” Acta Crystallogr. Sect. A 46 (1990), 1–11.
[3] L. Pauling, “The principles determining the structure of complex ionic crystals” J. Am. Chem.
Soc. 51 (1929), 1010–1026.
[4] D. Fisher, M. Jansen, “Synthesis and structure of Na3N” Angew. Chem. Int. Ed. Engl. 41 (2002),
1755–1756.
[5] M. O’Keeffe, B.G. Hyde, “Stoichiometry and the structure and stability of inorganic solids”
Nature 309 (1984), 411–414.
[6] A. Kjekshus, “The general (8-N) rule and its relationship to the octet rule” Acta Chem. Scand. 18
(1964), 2379–2384, and references therein.
[7] E. Parthé, “Valence-electron concentration rules and diagrams for diamagnetic, nonmetallic ion
covalent compounds with tetrahedrally coordinated anions” Acta Crystallogr. 29 (1973),
2808–2815.
[8] J. Kepler, “Strena seu de nive sexangula” Godfried Tambach, Frankfurt am Main (1611).
[9] P.N.H. Nakashima, A.E. Smith, J. Etheridge, B.C. Muddle, “The bonding electron density in
aluminum” Science 331 (2011), 1583–1586.
[10] H.K. Kim, S.H. Huh, J.W. Park, J.W. Jeong, G.H. Lee, “The cluster size dependence of thermal
stabilities of both molybdenum and tungsten nanoclusters” Chem. Phys. Lett. 354 (2002),
165–172.
[11] M. O’Keeffe, M. Eddaoudi, H. Li, T. Reineke, O.M. Yaghi, “Frameworks for extended solids:
geometrical design principles” J. Solid State Chem. 152 (2000), 3–20.
[12] V.A. Blatov, M. O’Keeffe, D.M. Proserpio, “Vertex-, face-, Schläfli-, and Delaney symbols in
nets, polyhedra and tilings: recommended terminology” CrystEngComm 12 (2010), 44–48.
[13] O. Delgado-Friedrichs, M. O’Keeffe, O.M. Yaghi, “Taxonomy of periodic nets and the design of
materials” Phys. Chem. Chem. Phys. 9 (2007), 1035–1043.
[14] O. Delgado-Friedrichs, M.D. Foster, M. O’Keeffe, D.M. Proserpio, M.M.J. Treacy,
O.M. Yaghi, “What do we know about three-periodic nets?” J. Solid State Chem. 178 (2005),
2533–2554.
[15] M.R. Ranade, A. Nawrotsky, H.Z. Zhang, J.F. Banfield, S.H. Elder, A. Zaban, P.H. Borse,
S.K. Kulkarni, G.S. Doran, H.J. Whitfield, “Energetics of nanocrystalline TiO2” Proc. Natl.
Acad. Sci. USA 99 (2002), 6476–6481.
[16] H. Zhang, J.F. Banfield, “Understanding polymorphic phase transformation behavior during
growth of nanocrystalline aggregates: insights from TiO2” J. Phys. Chem. B 104 (2000),
3481–3487.
[17] W. Ostwald, “Studien über die Bildung und Umwandlung fester Körper. 1. Abhandlung:
Übersättigung und Überkaltung” Z. Physik. Chem. 22 (1897), 289–330.
[18] V.M. Goldschmidt, “Laws of crystal chemistry” Naturvissenschaften 14 (1926), 477–485.
[19] R.D. Shannon, “Revised effective ionic radii and systematic studies of interatomic distances in
halides and chalcogenides” Acta Crystallogr. Sect. A 32 (1976), 751–767.
[20] M.W. Lufaso, P.M. Woodward, “Predictions of the crystal structures of perovskites using the
software program SPuDS” Acta Crystallogr. Sect. B 57 (2001), 725–738.
[21] C.J. Howard, H.T. Stokes, “Group theoretical analysis of octahedral tilting in perovskites” Acta
Crystallogr. Sect. B 54 (1998), 782–789.
[22] D. Wang, R.J. Angel, “Octahedral tilts, symmetry-adapted displacive modes and polyhedral
volume ratios in perovskite structures” Acta Crystallogr. Sect. B 67 (2011), 302–314.
 1.8 References 53

[23] A.M. Glazer, “Classification of tilted octahedra in perovskites” Acta Crystallogr. Sect. B 28
(1972), 3384–3392.
[24] P.M. Woodward, “Octahedral tilting in perovskites. II. Structure stabilizing forces” Acta
Crystallogr. Sect. B 53 (1997), 44–66.
[25] L.B. McCusker, F. Liebau, G. Engelhardt, “Nomenclature of structural and compositional
characteristics of ordered microporous and mesoporous materials with inorganic hosts” Pure
Appl. Chem. 73 (2001), 381–394.
[26] E. Zintl, “Intermetallische Verbindungen” Angew. Chem. 52 (1939), 1–6.
2 Defects and More Complex Structures

We have seen in Chapter 1 that the solid state world is dominated by long-range order and
beauty. Crystalline materials contain highly symmetric arrangements of atoms that are
regularly repeated over millions of unit cells. In this chapter, we will question how realistic
this picture is.
In reality, there are a number of ways in which crystalline materials deviate from perfect
long-range order and contain imperfections or disorder. This can occur via “mistakes” in the
atomic arrangement of a pure material or via the introduction of impurity atoms giving rise to
chemical disorder. These defects can occur locally or extend over lines, planes, or 3D volumes
of materials. Such effects, even when they occur at very low levels, are vitally important to the
chemical and physical properties of materials. They turn low-value minerals into precious
gemstones; soft iron into strong and corrosion-resistant stainless steel; and they control the
semiconducting properties of silicon in the transistors powering modern electronics.
This chapter also introduces a variety of ways in which materials can deviate from having
simple stoichiometric formulae. This can occur either via the presence of defects or chemical
substitutions in a material or can have a variety of more complex structural origins. In later
chapters, we will see how these various effects influence many of the important properties of
functional materials.

2.1 Point Defects in Crystalline Elemental Solids

We have seen that the structures of many elements can be described in terms of regular arrays
of spherical atoms. At the local level, this order can be perturbed by three different types of
point defects; vacancies, interstitials, and substitutional disorder. These are shown schemat-
ically in Figure 2.1.
A vacancy occurs when an atom is missing from a site in the structure as shown in Figure
2.1, left. An interstitial defect occurs when an extra atom sits in a site that would not normally
be occupied. An interstitial site can be occupied either by an atom of the same type that

54
 2.2 Intrinsic Point Defects in Compounds 55

vacancy interstitial substitution

Figure 2.1 Different types of point defects.

makes up the structure, or by an impurity atom. It is common, for example, for small non-
metallic elements such as C, N, O, and H to occupy a fraction of interstitial positions in the
structures of transition metals. Such materials can be of great technological importance:
C interstitials in iron greatly increase its mechanical strength; and Pd can store around 0.6%
by mass of H interstitials, which is of interest for hydrogen storage (see Box 2.1).
The third common type of defect is substitutional disorder where a foreign atom adopts
a site in the structure of a pure element. At low levels, foreign aliovalent1 atoms are
frequently referred to as dopants and can significantly alter chemical (Chapter 3), electronic
(Chapters 6 and 10), and optical (Chapter 7) properties of a material. Doping silicon with
low levels of Al or P, for example, leads to the formation of p- or n-type semiconductors,
respectively. At higher levels of doping, one typically refers to solid-solution or alloy
formation. This process can again be used to tailor a material’s properties. Real materials
exist with any one of these basic types of defects or with various combinations of them.

2.2 Intrinsic Point Defects in Compounds

Similar defects to those depicted in Figure 2.1 for elements can occur in ionic compounds,2
though with the additional constraint that one must maintain overall electrical neutrality in
the crystal. If, for example, one simply removed an Na+ ion from a compound such as NaCl
to create a defect, the crystal would end up with a negative charge. Defects in ionic
compounds can only occur in ways that avoid such charge build-up. Defects that can
occur in pure compounds are called intrinsic defects.
A Schottky defect is a cation vacancy compensated by an appropriate number of anion
vacancies (Figure 2.2), i.e. it is a missing formula unit. For a binary phase MX, the
number of vacancies on cation and anion sites will be equal; for MX2, two X− anion
vacancies would be needed for every M2+ vacancy. In terms of ionic charges, removing M+

1
Of different valence or oxidation state to the atom it replaces; isovalent means of the same valence.
2
Our discussion in this chapter will largely be in terms of formal ionic charges in materials. This need not, of course,
imply that ionic bonding is actually dominant.
56 Defects and More Complex Structures

Box 2.1 Materials Spotlight: Hydrogen-storage materials

Efficient energy production, storage, and use is one of the major challenges facing the modern
world. Both developed and developing economies are heavily reliant on fossil fuels in all areas
of energy use. Fossil fuels are limited in supply and polluting when burnt. One alternative fuel
under consideration for automotive applications is hydrogen, which produces benign H2O.
Safe and efficient storage of H2 is a major issue [1, 2]. H2 is gaseous at temperatures above
20.3 K and while its heat of combustion per unit mass is high compared to other fuels (H2
120 MJ/kg, gasoline 44.5 MJ/kg) the value per unit volume is low: at standard conditions,
hydrogen gas has ΔHcomb = 9.6 kJ/L and even as a liquid ΔHcomb = 8.4 MJ/L compared to 31.2
MJ/L for gasoline. For automotive use, the US Department of Energy state specific targets of
a gravimetric energy density of 7.9 MJ/kg (2.2 kW h), corresponding to 6.6 wt% H2, and
a volumetric energy density of 6.1 MJ/L at the time of writing. In addition, the fuel storage in
a car needs to be reversible (~1000 times) and allow refueling in 3 minutes—significant
challenges!
One possibility is to store H2 as a high-pressure gas or as a liquid at 20 K. It costs about 15%
of the energy content to pressurize H2 to 700 bar or about 20% to liquefy it (plus around 2%
per day to keep it cold). For these and other technological reasons, high-pressure seems the
more viable storage solution.
Chemical solutions to the H2 storage problem can be divided into two main categories—
those that rely on chemisorption and formation of chemical bonds, and those that rely on
physisorption of H2. In the chemisorption case, the stability of the compounds formed means
that a key challenge is to provide materials that decompose to release H2 at a low enough
temperature (below approximately 90 °C) to be compatible with, for example, polymer elec-
trolyte membranes in fuel cells. For physisorption, the challenge is the opposite: to provide
systems that still hold H2 at room temperature and ambient pressure. The adsorption of H2 by
carbon nanotubes, graphene, polymeric materials, and the metal–organic frameworks of
Chapter 14 has received significant attention [3]. Currently, no materials come close to targets
under ambient conditions and any adsorption-based storage systems will rely on liquid-N2
temperatures and the use of several tens of bar pressures.
The alloy LaNi5 is one potential storage material as the number of H atoms that can be stored
in interstitial sites (forming LaNi5H6) per unit volume is around double that of H2 liquid at its
boiling point. The reversibility of H2 uptake is good, and the H2 pressure at room temperature is
around 2 bar. Unfortunately, the mass content of hydrogen is too low at ~1.4 wt%, and La is
expensive. Similar problems face CoNi5H4 (1.1 wt%) and PdH0.6 (0.6 wt%). The latter decom-
poses close to room temperature and shows excellent reversibility, but suffers from the high cost
of Pd metal.
The only way to achieve a higher mass content of hydrogen in a storage compound is to focus on
hydrides from the top rows of the periodic table. MgH2 is one material of interest and can practically
store around 7.6 wt% H2, though its decomposition temperature is rather high (~330 °C) and
reversibility is poor. Other materials under investigation include complex hydrides such as NaAlH4.
When small amounts of Ti or other metals are included as a catalyst, this material can reversibly
produce H2 by two processes:
 2.2 Intrinsic Point Defects in Compounds 57

Box 2.1 (cont.)

NaAlH4 → 1/3Na3AlH6 + 2/3Al + H2 (3.7 wt% H2, > 33 °C)

Na3AlH6 → 3NaH + Al + 3/2H2 (1.8 wt% H2, > 110 °C)

Although NaAlH4 could theoretically release 5.5 wt% H2, in practice only around 4% can be
stored reversibly. In addition, the kinetics are slow.
An interesting strategy for reducing the decomposition temperature of a hydrogen-storage
material is to either destabilize the hydrogenated form or stabilize the dehydrogenated form. For
example, LiBH4 releases ~13.6 wt% H2 in an endothermic reaction, LiBH4 = LiH + B + 3/2H2 (ΔH =
67 kJ/mol H2). If one mixes it with MgH2, the mass hydrogen content is reduced to 11.4 % but the
exothermic formation of MgB2 in the reaction, LiBH4 + ½MgH2 → LiH + ½MgB2 + 2H2, makes
the total reaction less endothermic (ΔH = 25 kJ mol/H2), and the release temperature lower.
Other materials under investigation include the lithium amide–imide–nitride system (LiNH2–Li2
NH–Li3N, which stores 10.4 wt% H2 and in which nonstoichiometric materials are believed to play
a key role [4]), ammonia borane (BH3NH3, ~14 wt% release from 100–180 ºC) and related lithium/
sodium amidoboranes (LiNH2BH3/NaNH2BH3, ~11/~7.5 wt% at ~90 °C), though reversibility of
all these systems remains an issue.
Even if the chemical challenges can be met, there are still significant engineering challenges
for practical H2-fueled cars. For example, there is significant heat produced when either the
endothermic chemical processes described above are reversed on refueling, or the heat of
adsorption for cryo-adsorption systems is released. Heavy heat exchangers capable of handling
hundreds of kilowatts would be required for chemical systems capable of acceptable refueling
rates, reducing the gravimetric storage capacity of the system. More information on hydrogen-
storage materials can be found via www.doe.gov.

Schottky Frenkel

Figure 2.2 Schottky and Frenkel defects. Black and white circles represent cations and anions.

from its structural site will perturb the local electroneutrality—the negative charge on
surrounding X− ions is no longer cancelled by the charge of the cation. The uncompen-
sated 1− charge is therefore formally assigned to the cation vacancy. The anion vacancy
created elsewhere in the crystal at the same time will carry an analogous 1+ charge. These
58 Defects and More Complex Structures

ideas are formalized in the Kröger–Vink notation outlined in Section 2.6. In a Schottky
defect, the negative and positive charges will attract each other, which can lead to vacancy
clustering in the structure (see Section 2.9). One might also expect that on creation of
a defect, the surrounding structure may relax or distort. This is indeed the case but may be
difficult to detect using conventional structural characterization techniques (which tend to
probe average structures rather than local structures), particularly at low defect concen-
trations. In Section 2.9, we will see how substantial long-range structural changes can
occur at larger defect concentrations.
A Frenkel defect consists of a vacancy and a corresponding interstitial ion (of the same
charge) to maintain charge balance. Frenkel defects can occur for cations or for anions. AgI
commonly shows cation-Frenkel defects, CaF2 anion-Frenkel defects. The number of
Frenkel defects increases with the temperature, and materials with a high concentration of
Frenkel defects can have significant ionic mobility at high temperatures. We’ll see in Chapter
13 that this defect type gives rise to cationic conductivity in AgI and anionic conductivity in
CaF2-type materials.
We’ll mention other types of disorder/defects possible in compounds only briefly. For many
materials, intersite disorder is possible, for example the (Mn2+1−xFe3+x)(Fe3+1−x/2Mn2+x/2)2O4
spinels we discussed in Section 1.5.1, or disordered alloys such as FePt (see Section 4.4.2). In
Chapter 13, we’ll also encounter ions that are rotationally disordered in the solid state.
The final intrinsic point defect we’ll discuss is a color center. It can be described as electrons
trapped on vacant anion sites by the positive charge of the vacancy. Isolated trapped
electrons are known as F centers. More complex color centers consisting of pairs (M centers)
or triplets (R centers) of electrons or trapped holes are also known. The shape of the
potential holding the electron in an F center is complex but it can give rise to a localized
s-like state as well as a more extended p-like shape. Excitation from one state to the other
frequently leads to light absorption in the visible region giving rise to characteristic colors.
F centers can be generated in a variety of ways including by exposing alkali halides to excess
alkali-metal vapor or by X-ray irradiation. In nature, they give rise to blue forms of calcite
and feldspar as well as green diamonds.

2.3 Thermodynamics of Vacancy Formation

What factors determine how many defects will be present in a material? Let’s consider an
ideal crystal in two dimensions, formed by a single type of atom, such as the one drawn in
Figure 2.3. There are N0 atomic sites in the entire crystal. We can consider the formation of
a vacancy as the process of moving one atom from its regular site to a new site on the surface
of the crystal. The number of sites becomes N0 + 1. What is the Gibbs energy of creating
n such isolated, non-interacting, vacancies?
The main driving force behind vacancy creation is the increase in configurational entropy:
there are many possible ways to position the vacancy. The configurational entropy
 2.3 Thermodynamics of Vacancy Formation 59

Figure 2.3 Formation of a vacancy in an ideal crystal.

Figure 2.4 Gibbs energy of formation of n vacancies in an ideal crystal with N0 sites.

contribution to ΔG is −kT∙lnΩ per n vacancies formed. In this expression, Ω is the number of

ways n vacancies can be arranged over the N0 + n sites, which is given by Ω = (N0 + n)!/(N0!n!)
and k is the Boltzmann constant.3 Note that −kT∙lnΩ (Figure 2.4) is not a linear function of
the number of vacancies. There is also a smaller vibrational contribution to the entropy,
ΔSvib, which arises on vacancy formation. The overall favorable increase in entropy is
counteracted by the positive enthalpy of vacancy formation. The enthalpy term is positive
as the chemical bonds broken on vacancy formation are only partially compensated by the
formation of new bonds at the surface site.
At low vacancy concentrations, the ΔH and ΔSvib terms depend simply on the number of
vacancies n. Since ΔG = ΔH − TΔS, we can express the total change in Gibbs energy for
formation of n vacancies as ΔG = n(ΔH − TΔSvib) − kT∙lnΩ. This expression is plotted as
a function of n in Figure 2.4. The nonlinear nature of the configurational-entropy term
means that the minimum in ΔG always occurs at a non-zero value of n; that is, a small
equilibrium number of vacancies, neq, will always be formed. The value of neq can be
calculated by finding the minimum in ΔG by differentiation:
 
d ðN0 þ nÞ!
nDH  nTDSvib  kT ln ¼0 (2.1)
dn N0 !n!

3
Boltzmann constant k = 1.380649×10−23 J/K.
60 Defects and More Complex Structures

The factorial term can be treated by applying Stirling’s formula, ln(x!) ≈ xlnx − x for large x,
which on differentiation leads to:
 
N0 þ neq
DH  TDSvib  kT ln ¼0 (2.2)
neq

Note that the term in brackets involving neq is simply the inverse of the fractional concentra-
tion of vacancies formed, xv. Rearrangement shows that the fractional concentration of
vacancies depends solely on temperature and the two parameters, ΔH and ΔSvib:
   
neq DSvib DH
xv ¼ ¼ exp exp (2.3)
N0 þ neq k kT

The form of this equation suggests that the concentration of non-interacting point defects at
equilibrium4 follows the mass-action law. We’ll investigate the use of such equilibria in Chapter 3.
We can approximate the absolute number of defects by assuming N0 >> neq, which gives:
   
DSvib DH
neq ≈ N0 exp exp (2.4)
k kT

A similar expression would hold for the formation of isolated interstitials. We can see from
this expression that either a low ΔH or high T will favor the formation of defects.
For Schottky defects in an ionic material, one must consider the number of ways of
arranging both cation and anion vacancies. The configurational entropy term will be
−2kT∙ln Ω. A similar derivation gives the equilibrium number of Schottky defect pairs as:
   
DSvib DHS
neq;Schottky ≈ N0 exp exp (2.5)
2k 2kT

where ΔHS is the formation enthalpy of a Schottky defect pair.

For Frenkel defects, one generates n vacancies on the N0 sites and places them on
n interstitial sites out of a possible number of NI. There will be a configurational entropy
term of the form N0/I!/(N0/I –n)!n! for both vacancies and interstitials. One can then show that:
   
½ DSvib DHF
neq;Frenkel ≈ ðN0 N1 Þ exp exp (2.6)
2k 2kT

Table 2.1 gives molar enthalpies and entropies of defect formation for various compounds
from which the number of defects can be readily estimated.5 In Cu metal at 1000 °C, for example,

4
Reaching such equilibrium in solids usually requires high temperatures, it’s easy to “trap” non-equilibrium numbers
of defects on cooling.
5
Note the need to convert between the Boltzmann constant k, with units J/K, and the molar gas constant R, with units
J/(K mol), when using molar quantities; NAk = R.
 2.4 Extrinsic Defects 61

Table 2.1 ΔHf and ΔSf for defect formation from ref. [5]. Schottky values concern
two vacancies such that the energy per vacancy is similar for all three categories.

Defect type Material ΔHf, kJ/mol ΔSf, J/(K mol)

Vacancies Cu 123 21
Ag 105 12
Au 91 8.3
Al 72 18
Schottky NaCl 235 81
NaI 193 63
KCl 245 75
KI 244 86
Frenkel AgCl 140–150 45–101
AgBr 109–124 55–101

one would estimate xv as around 10−4. If one considers that a 1 cm3 block of Cu contains around
8.5×1022 atoms, the total number of defects (~1019) is considerable. It’s worth emphasizing that
thermodynamics gives the equilibrium number of defects ignoring kinetics—if a real sample has
been rapidly cooled from high temperatures, the actual defect concentration might be much
higher than the equilibrium value calculated at the temperature of the cooled sample. Defect
concentration calculations are illustrated and expanded in the end-of-chapter problems.

2.4 Extrinsic Defects

Defects involving new chemical species are called extrinsic defects. For elemental solids,
we’ve seen that foreign atoms can be substituted6 for atoms of the host material to produce
what can be described as doped systems at low levels of incorporation or alloys at higher
levels. This is called substitutional disorder. Substitutional disorder is also common, and
often deliberately targeted, in ionic compounds. As with intrinsic defects, it is important to
consider charge balance. In an ionic oxide MO, one can envisage replacing M2+ with another
2+ metal to give a solid solution M1−xM 0 xO (so-called isovalent substitution); a typical
example might be Co1−xMnxO that is known to form a solid solution for all values of x.
A common and technologically important group of materials based on isovalent cation
substitution are luminescent phosphors that rely on doping optically active rare-earth atoms
such as Nd and Eu into optically inactive hosts such as La2O2S. More examples of this
phenomenon are given in Chapter 7.

6
Loose language by soccer commentators means that there’s often confusion about usage of the verb to substitute;
indeed Fowler [Modern English Usage (1964) Oxford] amusingly describes the verb as a “treacherously double edged
sword”! The correct usage is that if we “substitute A for B”, B is removed and A is put in its place. “Beckham is
substituted for Rooney”, means Rooney leaves the pitch. If in doubt use the verb “replace”.
62 Defects and More Complex Structures

In an aliovalent substitution, an atom is replaced by one with a different oxidation state. An

example might be the substitution of Ca2+ for Zr4+ in ZrO2. The introduction of a lower-valent
ion means that the substituted site has a formal negative charge that must be compensated (Ca2+
has insufficient charge to compensate the negative charge of O2− ions surrounding the former Zr4+
site). In this case, charge balance is predominantly achieved through the introduction of vacancies
on the anion lattice. For every Zr4+ that is replaced by Ca2+ (charge difference of 2 units),
a corresponding O2− vacancy is required giving rise to a material with formula Zr1−xCaxO2−x.
If one substituted Y3+ for Zr4+ (charge difference of 1), one oxide-ion vacancy for every two Y3+
ions would be required; Zr1−xYxO2−x/2. There is a variety of possible charge compensations
involving substitution, vacancies, or interstitials; these are summarized in Table 2.2.
In semiconducting materials such as Si, dopants with either more or fewer valence
electrons can be introduced: a phosphorus atom is able to donate an electron to the
conduction band and is said to be a donor; an aluminum atom accepts electrons from the
valence band (or equivalently adds holes to the valence band) and is said to be an acceptor.
We’ll explore the case of Li+ acceptor and Al3+ donor doping in NiO in detail in Chapter 3.
In cases where the elements present have chemically accessible lower or higher oxidation states,
it is possible to achieve charge balance via redox compensation without the introduction of
vacancies. If we take the example of La2CuO4, which contains Cu in the +2 oxidation state, it
is possible to replace 7.5% of the available La3+ sites with Sr2+ to produce La1.85Sr0.15CuO4. We
can understand the charge balance in this material in terms of copper changing its oxidation state
from +2 to +2.15 on average. A similar Cu oxidation can also be achieved by incorporating extra
oxygen in interstitial sites of La2CuO4, which produces La2CuO4.075. Again, we can understand
the charge balance by considering oxidation of Cu from +2 to +2.15 manifested as a mixture of
0.85Cu2+ and 0.15Cu3+ on the Cu site. Both processes convert insulating antiferromagnetic La2
CuO4 into a superconductor at low temperatures (see Chapter 12).
The range of different types of cation substitutions, combined with the range of different
ways in which charge can be compensated, leads to the development of remarkably complex
materials from chemically simple starting points. Even for a binary oxide such as NiO, where
oxidation/reduction can lead to a nonstoichiometric formula, there are many possibilities.
On oxidation, one can incorporate excess oxygen or create Ni vacancies; on reduction, create

Table 2.2 Charge compensation mechanisms in cation-doped materials with fixed integer
oxidation states. Symbol ⃞ represents a vacant site.

Dopant charge Compensated by Host Dopant Substituted material

Higher positive Cation vacancy NaCl Ca 2+

Na1−2xCax⃞ xCl
Higher positive Anion addition CaF2 Y3+ Ca1−xYxF2+x
Lower positive Cation addition SiO2 Al3+ LixSi1−xAlxO2
Lower positive Anion vacancy ZrO2 Ca2+ Zr1−xCaxO2−x⃞ x
Any Double substitution CaAl2Si2O8 Na+ (Ca1−xNax)(Al2−xSi2+x)O8
 2.5 Solid Solutions and Vegard’s Law 63

Table 2.3 Redox compensations in materials (oxidation states can change).

Redox-active
metal Defect or dopant Examples Importance

Oxidized Cation vacancy Fe1−xO Vacancy clustering

LiCoO2→Li1−xCoO2 Battery material
Reduced Cation interstitial TiS2→LixTiS2 Battery material
WO3→NaxWO3 Electrochromic

Oxidized Anion interstitial La2CuO4→La2CuO4+x Superconductor

UO2→UO2+x Structural evolution
Reduced Anion vacancy YBa2Cu3O7→YBa2Cu3O7−x Superconductor
WO3→WO3−x Crystallographic shear
Oxidized Lower-valent cation La2CuO4→La2−xSrxCuO4 Superconductor
LaMnO3→La1−xCaxMnO3 Magnetoresistance

Reduced Higher-valent cation Nd2CuO4→Nd2−xCexCuO4 Superconductor

CaMnO3→Ca1−xLaxMnO3 Magnetoresistance

oxygen vacancies or incorporate excess Ni. The ways how to treat these possibilities in
combination are explored quantitatively in Chapter 3.1. Table 2.3 summarizes some of the
possible substitutional mechanisms and gives examples of materials in each category that will
be discussed in later chapters.
Just as for cations, substitutions for anions are possible, and there is a similar range of
charge-compensation mechanisms. Solid solutions based on anion substitution are generally
less common than those based on cation substitution. This is principally because the range of
anions of similar size/charge is smaller than the range of metals. However, in later chapters
we will meet important materials that can be understood in terms of substitution of O2− for
F−, N3−, or S2− ions.

2.5 Solid Solutions and Vegard’s Law

At low levels of substitution and for certain element combinations, substitutions will occur
randomly throughout a crystal structure leading to a solid solution. In many cases, the
properties of solid solutions evolve smoothly as one changes the degree of substitution. This
enables fine tuning of important parameters. As an example, the unit-cell parameters and
volume of a material often vary smoothly from that of the host (e.g. AY) towards those of the
substitutent (BY), as the degree of substitution x in A1−xBxY is increased. Such a material is
said to follow Vegard’s law (Figure 2.5); for the unit-cell parameter a as an example:

aðxÞ ¼ xaBY þ ð1  xÞaAY (2.7)

64 Defects and More Complex Structures

5.664 3.86
(a) (b)

pseudo-cubic cell parameter (Å)

5.662 3.84

3.82
cell parameter (Å)

5.660
3.80
5.658
3.78
5.656
3.76

5.654 3.74

5.652 3.72
0 0.5 1 0 0.5 1
x in Alx Ga1–x As x in CaRu1–x Mnx O3

Figure 2.5 (a) Unit-cell parameter of the AlxGa1−xAs thin films [6] and (b) pseudo-cubic cell parameter of
CaMnxRu1−xO3 [7]. Dotted lines show the Vegard’s-law prediction.

One can then estimate the degree of substitution directly by determining the cell parameters
and rearranging this expression to give x ¼ ½aðxÞ  aAY =ðaBY  aAY Þ.
Control of cell parameters via Vegard’s law can be particularly important when trying to grow
an epitaxial (lattice matching) layer of a material on top of a substrate. For example,
a semiconductor laser emitting between 1.2 and 1.65 μm (i.e. within the transparency window
of optic fibers) can be prepared by sandwiching In1−xGaxP1−yAsy between n- and p-type InP. By
adjusting x and y, it’s possible to exactly match the cell parameter of the sandwich layer to that of
InP. In addition, the band gaps of many semiconductors can also show a Vegard’s-law depend-
ence. While maintaining the same cell parameter, different combinations of x and y may have
different band gaps allowing control over the device’s properties. For a system such as this, our
simple form of Vegard’s law must take account of each component present and becomes:

aðx; yÞ ¼ xy aGaAs þ xð1  yÞ aGaP þ ð1  xÞy aInAs þ ð1  xÞð1  yÞ aInP (2.8)

If one introduces appropriate cell parameters (aGaAs = 5.65 Å, aGaP = 5.45 Å, aInAs = 6.06 Å,
aInP = 5.87 Å), the relationship between x and y to achieve lattice matching simplifies to:
y
x¼ (2.9)
2:21  0:053y

In practice, Vegard’s law is not always followed precisely. For some systems, intermediate
members of a solid solution have smaller unit cells than predicted—a negative deviation from
Vegard’s law. Others have larger unit cells than predicted and show a positive deviation. This
departure is due to the fact that atoms/ions can only be approximately treated as hard
 2.5 Solid Solutions and Vegard’s Law 65

Table 2.4 Examples of Kröger–Vink notation of point defects.

Symbol Example Charge number (effective charge)

vNa 0 +
Na vacancy in Na1−2xCaxCl −1
CaNa• Ca2+ on a Na+ site in Na1−2xCaxCl +1
Fi 0 F− interstitial in Ca1−xYxF2+x −1
vO•• O2− vacancy in Zr1−xCaxO2−x +2

spheres and specific bonding interactions cause small deviations in volume. Even for
AlxGa1−xAs, the high-precision data of Figure 2.5a show a small but significant
positive deviation from Vegard’s law; the departure is such that a Vegard’s-law derived
Al content would be ~3% in error at x = 0.5. The data for CaMnxRu1−xO3 in Figure
2.5b show a much more marked deviation. The authors have interpreted this as being
due to the presence of Mn3+/Ru5+ (as opposed to the expected isovalent Mn4+/Ru4+
substitution) at intermediate compositions.7 Departure from Vegard’s law can therefore
give a useful indication of a change in relevant properties of a material under investigation (see
also the unit-cell parameter discontinuity in YBa2Cu3O7−x upon entering the superconducting
regime in Chapter 12). Many materials also show a maximum range of solid solution, beyond
which it is impossible to perform further substitution, a solid-solubility limit. Departure from
Vegard’s law can reveal when this limit is reached.
Finally, as we will again see in Chapters 3 and 4, the ease of forming solid solutions is often
temperature-dependent, and, as a material is cooled, a solid solution may separate (phase
segregate) into its component phases. This can lead to islands of one phase surrounded by
a matrix of the second.

2.6 Kröger–Vink Notation

When discussing defects in crystals it is often useful to adopt a shorthand notation for the various
types of defects encountered. The most common is the Kröger–Vink notation in which a symbol
of the form ACB is used. Position A is the chemical symbol of the atom concerned or v for vacancy.
Superscript B indicates the effective charge of the defect (remember that a cation vacancy leaves
a local excess of negative charge at the defect site; an anion vacancy an excess of positive charge).
A dot is used for a positively charged defect (i.e. AC•) and a prime for a negatively charged defect
(i.e. AC0 ). Finally, the subscript C identifies the site in the crystal on which the defect occurs via the
symbol of the host atom or a symbol i if the site is interstitial.8 Typical examples are given in Table
2.4 using materials introduced in Table 2.2.

7
Note that the y-axis scales on Figures 2.5a/b are very different: the overall percentage change in cell parameter is ~ 0.1% in
(a) and ~3% in (b). In (b) we plot (volume/4)1/3 to give a pseudo-cubic-cell parameter related to the simple perovskite cell.
8
One may also use an s for a surface site (it is often convenient to think of the atoms that are missing in, e.g. a Schottky
defect as having migrated to the surface of the crystal).
66 Defects and More Complex Structures

2.7 Line Defects in Metals

The defects considered up to now have all been randomly distributed, isolated point defects
where the probability of finding a defect at a given point in a structure is largely independent
of whether nearby sites contain a defect. Energetically favorable defect clustering means,
however, that it is common to find aggregated and extended defects in materials. It is this
type of defect, which, for example, gives metals their characteristic properties of malleability
and ductility and allows them to be worked into useful forms. We’ll consider two basic types
of line defects in metals; edge dislocations and screw dislocations.

2.7.1 Edge Dislocations

An edge dislocation can be envisaged as shown in Figure 2.6a illustrating a 2D slice through
a crystal structure. An extra plane of atoms (running perpendicular to the plane of the paper)
has been inserted into the top half of the crystal. This gives rise to a dislocation line (again
perpendicular to the plane of the paper) shown by a star. This defect is similar to the
interstitial defect of Figure 2.1b but extends over a 2D plane of the structure.
The existence of edge dislocations helps explain why metals can be relatively easily
deformed without cracking or failure. If one applies a force perpendicular to the top half
of the crystal of Figure 2.6, then the “extra” plane of atoms can move to alleviate the imposed
stress. This can occur in a series of steps in which a single line of bonds breaks and then
reforms in the crystal such that the dislocation line moves across the crystal by one lattice
spacing at a time. The plane along which the dislocation line moves is called the slip plane
(shown as a dashed line). In each step only a small number of chemical bonds need to be
broken, which helps rationalize the ease with which metals can be deformed. Alternative
models of deformation would require the simultaneous breakage of entire planes of metal–
metal bonds and would be prohibitively costly in terms of energy. The importance of
dislocation motion also helps explain why minor impurities can have a large influence on
the mechanical properties of metals. Impurities can trap dislocations at specific locations in
a sample and prevent their motion. This process is known as pinning.
The description of deformations of materials via the slippage of atomic planes also rationalizes
why ceramic materials are usually harder, more brittle, and more prone to cleavage than metals.
The delocalized, isotropic nature of metallic bonding allows easy slip-plane formation and
slippage as opposed to covalent solids; in an ionic solid slippage of planes by one atomic unit
is unfavorable due to the repulsive interactions between ions of like charge.

2.7.2 Screw Dislocations

The second common type of line defect in metals is a so-called screw dislocation. One can
imagine this as a “spiral staircase” or “corkscrew” arrangement formed by slicing a crystal to
 2.8 Planar Defects in Materials 67

slip
plane

(a) (b)

(c) (d)

Figure 2.6 An edge dislocation due to an extra plane of atoms (a). Application of a shear stress moves the
edge dislocation through the crystal (b, c) eventually forming a step on the crystal surface (d). Such
processes occur as metals are worked.

its center and sliding adjacent layers of atoms by one atomic plane as shown schematically in
Figure 2.7. As subsequent atoms attach to the growing crystal surface, they will form a spiral
arrangement as shown in the experimental image of SiC on the right of the figure. This is
a line defect, and the line of the dislocation runs up the center of the spiral. Screw dislocations
can move through a crystal under applied stress in a similar manner to edge dislocations.
In practice, the dislocations in metals may appear more complex than simple edge or screw
dislocations. However, most common dislocations can be described as a combination of
these two basic types.

2.8 Planar Defects in Materials

Many defects in materials involve planes of atoms as opposed to the point and line defects
described above. Three categories of planar defects found in technologically important
materials are stacking faults, twins, and antiphase boundaries.

2.8.1 Stacking Faults

Many materials can be described in terms of layers of atoms or groups of atoms that stack
along a certain direction in the structure. These range from metals, which are essentially 3D
in their properties, to materials such as graphite, layered transition-metal sulfides, and
68 Defects and More Complex Structures

Figure 2.7 Left: A schematic picture showing the start of a screw dislocation in a crystal. Right: An image
of a screw dislocation in a crystal of SiC; the width of the inner terraces is ~10 μm. Image courtesy of
Dietmar Siche.

silicate clays which are far more 2D. Section 1.4.1 describes, for example, how the structure
of many metals is based on either hexagonal or cubic closest packing of spheres that repeat in
the layer-stacking sequence ABABAB and ABCABCABC, respectively. Both fill the same
percentage (74.05%) of available space and are close in energy. In fact, some materials switch
from one sequence to another under different conditions (for example Co from AB to ABC
upon heating above 690 K, or Al from ABC to AB under very high pressures). Such materials
may also be prone to faults in the stacking sequence. Stacking faults also occur easily in
layered materials such as those related to the CdCl2 or CdI2 structure types (Figure 1.28). In
some cases, the forces holding together adjacent layers can be so weak or non-specific that
although materials are highly ordered in two dimensions, adjacent layers along the stacking
direction rapidly lose registry. This type of disorder, turbostratic disorder, is commonly
found in graphite, layered clays, and molecular intercalates (Chapter 13).
While stacking faults occur at random in some materials, in others they give rise to ordered
structures. Ordered variants differing by their stacking sequences are referred to as polytypes,
and a schematic example is shown in Figure 2.8. Polytypism (formation of several polytypes)
is commonly observed in the layered halides, structurally related transition-metal dichalco-
genides, SiC, and many other systems. It is a subset of the more general phenomenon of
polymorphism: the existence of more than one structural form of a given material.

2.8.2 Twinning
Crystallographic twinning is an extended defect that is important in areas as diverse as the
mechanical properties of metals, the optical properties of crystals, and the performance of
piezoelectrics and ferroelectrics (Chapter 8). A twinned crystal is defined as an intergrowth of
two or more individual crystals of the same species, in which the different portions are related
by a symmetry operation that doesn’t belong to the point group of the crystal. Each individual
is called a twin component or a twin domain and the symmetry operation relating them a twin
 2.8 Planar Defects in Materials 69

a b
A
A B C A
A A A A

A A A C

C A A A

B B
c
A A A A A A

a b

Figure 2.8 Polytypism of an element. Left: ccp, layer sequence ABC. Right: ABAC.

law. The interface between domains is called a twin boundary or domain boundary, and these
can be several atomic layers thick if there is significant structural rearrangement required
across the boundary. Figure 2.9 contains some examples. Twinning can be detected in
a number of ways. Sometimes it leads to crystals with re-entrant9 angles between faces. In
translucent crystals, it can be seen using cross polarizers in a polarization microscope as
different domains show light extinction at different angles. It is also apparent in single-
crystal diffraction experiments, since each domain will give rise to its own diffraction pattern;
hkl reflections from different domains may or may not overlap in reciprocal space depending
on the twin law. Twinning can be an unwanted complication in these experiments and is one of
the reasons why powder-diffraction methods are often used for structural work.
There are three twin categories; growth twins, deformation (glide) twins, and transformation
twins. Growth twins can arise during the formation of a crystal if it grows from multiple
crystallization nuclei. In Figure 2.9a, we can imagine that two domains have grown from
different nucleation sites at the twin boundary. Deformation twins occur when a shear stress is
applied to a crystal and causes each plane of atoms to move by a fraction of a unit cell relative
to the layer below it. Major changes in the crystal’s macroscopic shape arise from small
individual atomic displacements, and the process can occur very rapidly (Figure 2.9b). This
type of twinning is important in steel-production, mineralogy, and in the superelastic and
shape-memory alloys discussed in Box 2.2.
In functional materials, twinning is most commonly encountered when the shape of the unit cell
has a higher point symmetry than the atomic contents.10 If one imagines producing a crystal by
stacking together such unit cells (like building a wall out of parallelepiped bricks), it is easy to
make a mistake and start placing bricks upside down without disrupting the overall stacking.

9
A re-entrant angle in an irregular polyhedron is an angle inside that polyhedron, which is greater than 180°, such as
on the right-hand side of the crystallite shown in Figure 2.9b.
10
Or close to higher symmetry. For example, a monoclinic cell with a β = 90.05° approximates an orthorhombic cell.
70 Defects and More Complex Structures

twin boundary

twin
twin
Brick Brick boundary
boundary
Brick Brick
Brick Brick

(a) (b) (c)

Figure 2.9 (a) A twinned crystal where a growth fault leads to domains related by a mirror plane.
(b) A deformation twin where atoms in the top domain of the crystal have been displaced by a shear
stress. (c) A high-temperature square lattice (a = b) that distorts into rectangles (a ≈ b) at low temperature will
form a twin-domain structure in which the longer edge (emphasized by an arrow) can be aligned in different
directions, two of which are shown; domains are related by a fourfold rotation of the square lattice.

Figure 2.9a shows a 2D example where the unit cell is a rectangle and thus has two mirror planes
(mm symmetry) while the cell contents (here the word “Brick”) have lower symmetry (here no
symmetry). When a “mistake” is made, a new twin component is formed, related to the first by
one of these mirror planes. This process is only likely to occur if the energy penalty for placing the
unit cell in an alternative orientation is low. One situation where this happens is when a structure
undergoes a phase transition from a high-symmetry to a related low-symmetry form.11 If only
minor structural changes occur, the “lost” symmetry element is likely to act as a twin law giving
rise to a transformation twin.12 A common example is when a material is cubic (a = b = c) at high
temperature but distorts to a lower symmetry tetragonal (a = b ≲ c) structure on cooling. In this
process, it is likely that different regions of the crystal will have their longer c unit-cell axis pointed
in different directions in 3D space, giving rise to different domains (a 2D simplification is shown in
Figure 2.9c where the long cell direction is indicated by an arrow). In fact, doing this will lower the
strain across the entire crystal, such that multidomain crystals usually form on cooling.
In a transformation twin, the domains are formally related by a rotational symmetry
element that is present in the high-symmetry phase but absent in the low-symmetry phase.
They are therefore associated with translationengleiche transitions (see Appendix B) in
which the translational symmetry is retained (all lattice points are kept) but a portion of
the rotational symmetry is lost. The number of twin variants (different twin-domain orienta-
tions that will form) can be predicted from the ratio of the number of rotational symmetry
operations13 per lattice point in the two space groups in question. The ratio is called the index
11
There are numerous examples discussed in Chapter 4 and throughout this book.
12
The spatial arrangement of atoms originally related by symmetry is regenerated across the twin boundary when the
symmetry operation becomes a twin law.
13
These symmetry operations are in a numbered list for each space-group entry under the heading “Symmetry
operations” in the International Tables for Crystallography, Volume A (see Appendix B), one for each point of
the general position. In centered space groups, there is a set associated with each of the centering translations (i.e.
with each lattice point per cell). The number of individual domains can be much higher than the number of variants
obtained from the symmetry-operations ratio per lattice point.
 2.8 Planar Defects in Materials 71

Box 2.2 Materials Spotlight: Shape-memory alloys

Shape-memory alloys are a remarkable family of structural materials that can be mechanically
deformed at low temperature but will regain their original shape on gentle heating. They thus
retain a “memory” of their initial shape. What’s the origin of this memory effect? On cooling,
some materials undergo diffusionless transitions (each atom only moves a small distance
relative to its neighbors) which lead to twinning of the type shown in Figure 2.9b. These are
commonly called martensitic transformations after the well-known example that occurs when
austenite (fcc) transforms irreversibly to lower-symmetry martensite (body-centered tetrag-
onal) during the quenching of steel. Shape-memory alloys undergo reversible martensitic
transitions in which the details of the low-temperature domain structure control the shape of
the crystal. This is shown in the figure below. If a shape-memory alloy is cooled through the
phase transition while restricted to a certain physical shape (step 1 in the figure), a twin
structure with domains of optimal size and arrangement to fit this particular shape will develop.
When this cooled form is mechanically deformed (step 2), the twin-domain boundaries will
move into a new interlocked metastable position. In it, the twin domain that best compensates
the applied stress has grown at the expense of the others, facilitating the deformation. When
that form is heated (step 3) above the temperature of the low- to high-symmetry phase
transition, the domain structure, and hence the mechanically deformed shape, will be lost.
On subsequent cooling (step 4) the original twin structure “stored” in the sample during its
initial manufacture, and hence its shape, is reformed.

1 2 3 4
cool deform heat cool

The best-known shape-memory alloy is a 1:1 alloy of Ni and Ti, commonly known as nitinol
(nickel titanium Naval Ordnance Laboratory); other commercial materials are based on Cu-
rich CuZnAl and CuNiAl alloys. NiTi is cubic at high temperature but on cooling undergoes
a phase transition to a structure with a small monoclinic distortion. This martensitic trans-
formation starts at around 60 °C on cooling and 71 °C on warming, but slight changes in
composition allow the transition temperature to be controlled between −50 °C and 100 °C. For
Cu-based systems, transition temperatures from −180 °C to 200 °C can be achieved. Annealing
temperatures of 500–800 °C are typically used to “store” a specific shape in a sample.
Shape-memory alloys have a range of potential applications. Biomedical implants can be
prepared, which are inserted into the body in a compressed, deformed state and unfold as they
warm to body temperature. Shape-memory fittings to join piping together are available
commercially; these are made as tubes, slightly smaller than the pipes, and then deformed at
72 Defects and More Complex Structures

Box 2.2 (cont.)

low temperature to fit over the two pipes to be joined. On gentle heating, they regain their
original dimensions thus forming a tight seal. So-called two-way shape-memory alloys that
“remember” the shape of both the low- and high-temperature form can also be prepared. This
is achieved by using defects, which create local stresses in a material, to control the places where
individual twin domains form. Typing “shape memory movie” into a search engine will provide
links to a number of on-line movies that give dramatic illustrations of this effect.
A second property of these alloys—their high pseudoelasticity or superelasticity—is
exploited in eyeglass frames that can withstand distortions without damage. Here, one
operates just above the transformation temperature of the material so that the martensitic
transition occurs on application of stress, allowing significant mechanical distortion.
When the stress is removed, the material reverts to the austenite form and the original
shape is regained.

of the translationengleiche subgroup (Appendix B) of the two groups. For example, if a cubic
material with space group Pm3m (#221) undergoes a phase transition to a tetragonal
structure with space group P4/mmm (#123), the former has 48 symmetry operations (or
general points), the latter 16. We therefore expect 48/16 = 3 variants. This result can also be
obtained straightforwardly by considering the point-group order alone, which again changes
from 48 for m3m (Oh in Schönfliess notation) to 16 for 4/mmm (D4h).14

2.8.3 Antiphase Boundaries

A third type of planar fault that can occur in crystals is an antiphase boundary. These are
frequently associated with site ordering in materials. Consider the situation shown in Figure
2.10, which could represent a square array of oxide ions with metal ions of two different
charges (e.g. 1+ and 3+) sitting in the four-coordinate sites. At high temperature, the two
types of metal ions are disordered over their common crystallographic site in the structure.
On cooling, it may become thermodynamically favorable for them to order in, say,
a checkerboard pattern as in the bottom left of Figure 2.10, where cations of low charge
are surrounded by high charge and vice versa. Such an arrangement minimizes electrostatic
repulsions as it maximizes the separation between the 3+ ions.
The second column of Figure 2.10 represents a situation where an antiphase boundary was
formed by two ordered regions meeting out of phase as the crystal cools from its surface. The

14
The point group contains only rotational symmetry operations that leave at least one point unchanged. The order is
the number of operations it contains. We include the Schönflies notation of the point groups (Appendix A) as it’s
familiar to most chemists and most inorganic texts will contain point-group character tables that state the group
order.
 2.8 Planar Defects in Materials 73

cooling

cooling

cooling

no antiphase boundary antiphase boundary

Figure 2.10 Formation of antiphase boundaries. A hypothetical array of oxide anions (small circles)
contains an equimolar mixture of 1+ and 3+ metal cations that are disordered at high temperature (gray
circles) and order (white/black circles) on cooling from two hypothetical seeds at the opposite crystal
surfaces; 50% meet in phase (on the left), 50% out of phase (on the right). The ordered structure here has
a quadrupled unit cell reflecting a loss of translational symmetry.

two components are related by a symmetry element present in the high-symmetry structure
but missing in the low-symmetry structure. In contrast to twinning, antiphase boundaries
form when a subset of translational symmetry elements is lost rather than rotational
elements. Antiphase boundaries are therefore associated with klassengleiche subgroups.15
Clearly, the domain-boundary structure on the bottom right is energetically unfavorable
with respect to that in the bottom left, but considerable atomic rearrangement would be
required to “heal” the fault. The presence and number of antiphase boundaries will depend
strongly on the thermal history of a sample. Real examples where antiphase boundaries
occur include M-site ordering in double perovskites such as A2MM 0 O6 (AM0.5M 0 0.5O3) that
contain a mixture of differently charged M and M 0 ions (e.g. 2+/4+ or 1+/5+) on the
octahedral site. Similar effects are found in metal alloys such as FePt that has a cubic
structure at high temperature, with Fe and Pt disordered over all sites of an fcc lattice, but
orders on cooling to a tetragonal structure with alternate layers of Fe and Pt. Faults occur
when different ordered domains meet such that two layers of Fe or two layers of Pt are
adjacent or the layers grow in different orientations. The ordered material is of interest in
that it has high magnetocrystalline anisotropy making it potentially useful for magnetic data
storage.
15
See Appendix B. Klassengleiche subgroups involve a partial loss of translational symmetry (for example, loss of
some lattice points due to formation of a supercell or a loss of centering.). The point-group symmetry or crystal class
is retained.
74 Defects and More Complex Structures

2.8.4 Crystallographic Shear Structures

A final family of planar defects that can give rise to nonstoichiometric formulae and
structures that initially appear remarkably complex are found in the crystallographic shear
structures. They are also frequently referred to as Magnéli phases after A. Magnéli who first
described Mo9O26 and W20O58. The key to understanding them in terms of extended planar
defects was provided by Wadsley [8]. A good example is provided by the tungsten oxides. We
have seen in Section 1.5.3 that the structure of WO3, a yellow semiconducting material, can
be described in terms of distorted and tilted corner-sharing WO6 octahedra (a distorted
variant of the ReO3 structure). When WO3 is heated under reducing atmospheres, or when it
is reacted with additional W in the absence of oxygen, a series of compounds of general
formula WO3−x (e.g. WO2.9) can be made. These are typically bright blue in color due to
partial reduction of W(VI) to W(V). Although their compositions can be approximated as
WO3−x, x is not a continuous variable but takes certain discrete values. Careful structural
studies have shown that these materials are members of homologous series with general
formulae such as WnO3n−1 and WnO3n−2; a material such as WO2.9 being better formulated as
W20O58, the n = 20 member of the second series.
The structural origin of these general formulae can be understood with respect to
Figure 2.11. If one imagines that partial reduction of WO3 initially removes an entire
plane of oxygen atoms, one would generate the structure in the middle of Figure 2.11,
in which W atoms on either side of the missing plane are only five-coordinate. If each
W in the right-hand five-coordinate plane is shifted by half a unit cell parallel to the
two in-plane axes of the original unit-cell edge of pseudo-cubic WO3, the sixfold
coordination of each W is regained. We see a double chain in Figure 2.11 of octahedra
that share two edges in addition to sharing their remaining three corners. In the
example of Figure 2.11, the octahedra translate solely in the plane of the figure, but
we could also imagine octahedra moving out of the plane to share apical edges instead
of the equatorial ones.
In reality, many different crystallographic shear planes are possible for WO3. Figure 2.12
shows some of them. Figure 2.12b shows the (101) fault plane. This corresponds to a twin
where all oxygens remain coordinated to two octahedral W atoms, and the formula remains
WO3. As the plane of defects rotates clockwise in Figure 2.12 from (102) to (103) to (001) ≡
(10∞), the 2D projection shows a line of blocks of four edge-sharing octahedra in part (c),
a line of blocks of six octahedra in part (d) and a continuous zig-zag belt of octahedra sharing
two edges in (e), with a corresponding decrease in the oxygen content as the oxygen connect-
ivity increases. Irregular (102) Wadsley defects have been observed for WO3−x with x = 0.002.
By x = 0.05, these defect planes become ordered, giving a WnO3n−1 family of phases (n = 20 for
x = 0.05) with the integer n decreasing as the defect planes get closer on lowering oxygen
content (increasing x). Ordered structures of this type are known down to n ≈ 12 of W12O35
(WO2.92 or x = 0.08). Beyond x ≈ 0.08, the spacing of (102) defects becomes too close, and (103)
defects are observed instead, leading to WnO3n−2 materials that are well characterized for at
 2.9 Gross Nonstoichiometry and Defect Ordering 75

Figure 2.11 A section of the structure of WO3 showing a projection of corner-sharing octahedra.
Removal of a plane of oxygen atoms followed by a displacement along the shear plane recreates
octahedral coordination around each W. The pseudo-cubic unit cell of WO3 is shown as
a dark box.

least n = 26–18 (WO2.92 to WO2.89). As might be expected, similar structures are also observed
when a small amount of Nb5+ is substituted for W6+.
Figure 2.12f shows an extreme case with regularly spaced defects corresponding to
a formula of M2O5; the ideal16 structure of V2O5. The structure of the mixed-anion com-
pound Nb3O7F can similarly be described in terms of crystallographic shear planes of the
type shown in Figure 2.12e separated by a single plane of fully corner-shared octahedra. If
one allows shear planes in two perpendicular directions, a large family of closely related
structures results. An excellent description of the various possibilities is given in the text by
Hyde and Andersson (see Further Reading).
A similar phenomenon occurs in the rutile TiO2 structure leading to a variety of compos-
itions between Ti3O5 (TiO1.67) and TiO2. The shear planes are harder to depict as there’s no
convenient projection to draw, but the resulting structures contain infinite 2D slabs of rutile
separated by regions with a portion of face-sharing octahedra. Depending on the shear plane
involved, materials with formula TinO2n−p result: TiO1.75 to TiO1.89 have 4 < n < 9 and p = 1;
TiO1.89 to TiO1.93 have 9 < n/p < 16 and p > 1; TiO1.93 to TiO1.98 have 16 < n < 40 and p = 1.
It’s also common (and unsurprising) to find intergrowths of different slabs in electron
microscopy images of samples rapidly cooled from high temperature.

2.9 Gross Nonstoichiometry and Defect Ordering

In Section 2.3, we investigated the number of vacancies one might expect to find in a simple ionic
material with fixed oxidation states. For compounds with variable metal oxidation states, defect
16
V2O5 is normally described as layers of edge- and corner-sharing VO5 square pyramids. However, an oxygen from
an adjacent layer makes up the sixth coordination site of the V such that it can be described as a highly distorted
octahedron.
76 Defects and More Complex Structures

a c

(101)

(a) (b)

(102) (103)

(c) (d)

(001)

(e) (f)

Figure 2.12 Possible planar defects for an ideal WO3-type structure. (a) The WO3 structure viewed down
the b axis; the unit cell is bold. (b) A twin boundary. (c–e) Crystallographic shear planes with loss of
oxygen from the structure. (f) The idealized structure of V2O5.

levels can be sufficiently high that defect interactions and ordering become important—this can
be remarkably complex. The different TiOx (0 ≤ x ≤ 2) phases that form when Ti is progressively
oxidized give several examples [9].
Ti itself is an hcp metal. On oxidation, O2− ions initially adopt interstitial sites in the
hcp structure at random before ordering at higher concentrations. At composition Ti6O,
oxygen fills one-third of the octahedral holes in half of the hole planes, at Ti3O, two-thirds
of them, and at Ti2O all of them, with anti-CdI2-type ordering adopted. The structure and
composition probably evolve continuously over much of this range, and Vegard’s law is
 2.9 Gross Nonstoichiometry and Defect Ordering 77

followed up to Ti3O. Early literature [10] suggested that, around TiO0.67–0.75, a structure
related to ε-TaN forms but with high O deficiency; this can equally be described as
O interstitials in the ω-Ti structure. This same structure type has more recently been
reported [11] for a TiO composition synthesized in a Bi flux so there is some doubt about
the earlier work. Around x = 1, TiO more commonly forms with an NaCl-type structure
and a significant number of defects. On the O-deficient side (TiO0.64), up to ~36% of
oxygen sites are vacant; on the O-rich side (TiO1.26), ~23 % of Ti sites and some O sites are
vacant. At the 1:1 composition, ~15 % of both sites are vacant such that we might write
the formula as Ti0.85O0.85. Note that in a case like this, neither elemental analysis nor site
occupancies refined from diffraction data (which merely measure relative scattering from
cation and anion sites) would reveal the presence of defects; density measurements of the
type explored in the end-of-chapter problems are one way of detecting them. At low
temperatures, vacancies in TiO order to give a monoclinic structure, in which a regular
arrangement of one in six cation and anion sites of the NaCl-type structure is empty.
Similarly, Ti4O5 (TiO1.2) orders at low temperature to a structure in which four out of five
cation sites are occupied, and O sites are essentially fully occupied. At higher oxygen
contents, one reaches Ti2O3 (with the corundum structure), Ti3O5 (several polymorphs)17,
Ti4O7, and then the crystallographic shear-plane series TinO2n−1 (with n = 5, 6, . . .)
discussed in Section 2.8.4.
Iron monoxide (wüstite) provides an example where nonstoichiometry and clustering of
defects add significant complexity to its ideally NaCl-type structure. In fact, the homogeneity
range of FeO does not include the stoichiometric (or integer-valence; see Chapter 3) phase.
At 1350 °C, Fe1−xO with 0.06 ≤ x ≤ 0.16 can be prepared; at lower temperatures, the range of
x is smaller. These materials are thermodynamically stable above ~570 °C (below this Fe and
Fe3O4 are stable) and there has been considerable investigation into the defect structures of
quenched samples. The principal defect present is a vacancy on the regular octahedral Fe2+
site (vFe 0 0 ; V for brevity in this FeO case), which is charge-compensated by a small number of
interstitial Fe3+ at nearby tetrahedral holes (Fei••• or T) like those occupied by Zn in ZnS,
sphalerite, Figure 1.32.18 Due to their effective charge, the vFe 0 0 and Fei••• defects attract each
other such that locally the Fe3+ interstitial is surrounded by four Fe2+ vacancies—a
V4T cluster (Figure 2.13). What’s more controversial is how these small vacancy clusters
are arranged on a longer length scale to form larger clusters.19

17
The structural chemistry of Ti3O5 is complex, and several polymorphs (α–δ, λ) exist with different arrangements of
corner-, edge-, or face-sharing TiO6 octahedra. The γ polymorph [S.-H. Hong, S. Asbrink, Acta Crystallogr. Sect.
B 38 (1982), 2570; ICSD 35148] can be viewed as an n = 3 rutile shear structure.
18
We shouldn’t be surprised by this as a similar site is occupied by Fe3+ in Fe3O4 (an inverse spinel structure, i.e. a ccp
of O2− with Fe3+ in one-eighth of the tetrahedral holes and Fe2+/Fe3+ in half the octahedral holes).
19
If the clusters order in three dimensions, one would see extra superstructure peaks in diffraction patterns due to the
larger unit cell involved. Such peaks are indeed observed experimentally, though for some compositions
a commensurate superstructure with asup = 2.5n·aFeO (n is an integer) is formed while at others the superstructure
00 0
is incommensurate (Section 2.11) with asup = (2.51−2.73)n·aFeO—so-called P and P ordering respectively.
78 Defects and More Complex Structures

(a) (b) (c) (d)

Figure 2.13 Iron-vacancy clusters proposed for Fe1−xO: (a) V4T cluster, (b) V10T4, (c) V12T4, and (d) V13
T4. White squares represent Fe2+ vacancies and gray circles interstitial Fe3+ sites. The ccp O2− that lie on
the unmarked vertices of each cube are omitted for clarity. Note that each cube here has the size of one
octant of the NaCl unit cell shown in Figure 1.26.

In one of the early studies on the defect structure of FeO [12], Koch and Cohen proposed
the V13T4 cluster shown in Figure 2.13. Other clusters have been suggested, including the V12
T4 cluster, which, when regularly spaced at a distance of 2.5aFeO (i.e. close to experimentally
observed superstructures), would lead to a composition of Fe0.872O. Yet other workers have
suggested the importance of V10T4 clusters. A single-crystal study [13] investigating both
Bragg diffraction and diffuse scattering on a specific Fe0.943O sample suggests that V13T4 and
V16T5 clusters are important. The experimental data were interpreted in terms of defect
clusters lying on the vertices of a highly distorted cubic lattice with spacing ~2.7aFeO, with
around 50% of these cells containing a defect and 50% being defect-free. The defect and
defect-free regions are not homogeneously distributed through the structure. Interestingly, if
one considers the size of a V13T4 cluster and the fact that for charge balance (its charge
number is 14−), neighboring octahedral sites must contain a portion of Fe3+ ions, close
packing of the overall units would require a cell of ~2.5aFeO. A mixture of larger V16T5 and
V13T4 would require a cell of around 2.7aFeO. These values are consistent with experimental
observations and explored in the end-of-chapter problems. The precise structural picture of
samples quenched to low temperatures is clearly complex, though it’s clear that, at high
temperature, isolated vacancies and smaller defect clusters become more important.

2.10 Incommensurate Structures

A number of materials exist that appear to have full occupancy of atomic sites in the
structure, yet still possess a nonstoichiometric formula. One case occurs in the so-called
incommensurate structures, which can’t easily be described using the simple 3D concept of
one unit cell and space group introduced in Chapter 1. An example is Sn1.17NbS3.17; one of
a number of A1+xBX3+x compounds that were long thought to have a simple ABX3
composition. The origin of the structural complexity in this family is at its heart simple.
The structure of Sn1.17NbS3.17 can be described as alternating layers of NaCl-like SnS and
 2.10 Incommensurate Structures 79

S
NbS2 Nb
S
Sn,S
c SnS c Sn,S
S
a b Nb
NbS2 S
a(NbS2)
12 a(NbS2)
~≈
7 a(SnS)

a(SnS)

Figure 2.14 The intergrowth structure of (SnS)1.17NbS2. The right-hand view shows that b cell parameters
of SnS and NbS2 segments are identical such that the layers fit together in a simple fashion; the left-hand
view shows that the a cell parameters have no simple relationship.

NbS2 (Figure 2.14).20 The relative sizes of the ideal unit cells of the SnS and NbS2 portions of
the structure are such that along b the cells fit together but along a they don’t. In addition,
there is no simple integer ratio of the a cell parameters such that, for example, three units of
SnS would match up with two units of NbS2; the two periodicities do not match.21 It’s
therefore not possible to use a simple multiplied unit cell. The situation is rather like tiling
a bathroom with rows of tiles of two different lengths—if one starts in one corner of the room
and works along the wall, the gaps between the two rows of tiles may never perfectly align.
In fact, for Sn1.17NbS3.17, the cell parameters of the two portions (aSnS = 5.673 Å; aNbS2 =
3.321 Å) are such that the structure approximately matches up after seven SnS and twelve
NbS2 unit cells. The formula of the material can therefore be approximated as (SnS)2×7
(NbS2)12 (2 × 7 as each SnS cell contains two formula units) or (SnS)1.17NbS2 (equivalent to
Sn1.17NbS3.17) and could be approximately described on a supercell with a = 39.8 Å. This
description is, however, rather inelegant, and incommensurately modulated structures such as
this are better described using the language of superspace groups and modulation functions
(structures that are incommensurate in 3D can be conveniently described in a superspace of
(3 + n)D with n added modulation periodicities) [14], but this description is beyond the scope
of this text. This approach may also be applied to commensurately modulated structures,
those with periodicities that match a small multiple. The advantage is a low number of
structural variables, as opposed to working with a large supercell of many atoms.

20
The NbS2 layers are similar to those found in CdI2, but with trigonal prismatic rather than octahedral coordination of Nb.
21
In the original publication [Meetsma et al., Acta Crystallogr. Sect. A 45 (1989), 285–291] the SnS portion of the structure
was described with a = 5.673 Å, b = 5.750 Å, c = 11.760 Å in space group C2mb with Sn at 0 0.25 0.1335 and S at 0.476 0.25
0.0954 and the NbS2 portion with a = 3.321 Å, b = 5.752 Å, c = 11.763 Å in space group Cm2m with Nb at 0 0 0 and S at 0
0.3335 0.1328. The sets of b and c cell parameters measured differ by less than the experimental uncertainty.
80 Defects and More Complex Structures

Sr
Cu

Figure 2.15 Structure of Sr0.73CuO2. The a axis runs horizontally in the plane of the paper.

Figure 2.15 shows a related phenomenon in the structure of Sr0.73CuO2 that has a Sr
periodicity of 3.72 Å and Cu periodicity of around 2.73 Å along the a direction. The Sr:Cu
ratio is thus 2.73/3.72 = 0.73. The Cu atom is always in a square-planar coordination
environment, whereas the Sr coordination varies as you move along a. Many other materials
display structural modulations. For example, the hexagonal perovskite Sr14/11CoO3 [15] has
a misfit modulation where voids between face-sharing octahedral columns accommodate an
excess of the relatively small Sr ions compensated by a partial reduction of Co4+ to Co3+; or
metal alloys such as Zn22Li6 [16] show occupational modulation. Incommensurate structures
have even been found for simple metals such as Ba and Bi under high pressure.

2.11 Infinitely Adaptive Structures

We’ll finish this chapter with a short discussion of other structures that use simple building
principles to accommodate nonstoichiometric and continually variable formulae. In fact, we’ve
seen examples of this behavior already in the shear structures of Section 2.8.4, where, as the shear
plane or plane spacing changes, a variety of closely related structures with variable composition
evolve. Systems where any small compositional change leads to a structure that is unique, even if
closely related to those of neighboring compositions, are called infinitely adaptive.22 Another
example is the family of Y3+ materials of composition Y(O, F)2.13 to Y(O, F)2.20 that can be
formed by the reaction of appropriate quantities of YOF and YF3. Within this composition
range, the excess anions are accommodated in a practically infinite series of very closely related
structures, each fluorite-related (as is YOF itself).
We can understand the structures by considering the fluorite MX2 structure (Figure
1.31) in terms of square grids of X, with a checkerboard arrangement of M above and
below the square grid, forming a slab of edge-sharing XM 4/4 tetrahedra (Figure 2.16, left).
Fluorite itself can be built up by alternating this MX slab with square grids of X along
a stacking axis, giving MX + X = MX2 overall. In the Y3+O1−mF1+2m23 structures, alternate
square grids of X are replaced by a layer of X anions with a triangular grid, which, for an
ideal situation, is denser by a factor of 2/√3 = 1.155. The composition of this ideal situation

22
Using the language of Chapter 4, each composition is a single phase. The compositions are so dense that there are
effectively no two-phase regions, just one solid-solution range.
23
The MX2+δ composition can be expressed in various ways. The general formula can be expressed as YxOx−yFx+2y ≡
YO1−mF1+2m ≡ Y(O1−mFm)F1+m; the final representation emphasizes that anion sites in the square grid are fully occupied.
 2.12 Problems 81

anions

cation

CaF2-structured hexagonal F layer Y5O4F7 Y5(O4F)F6

Y(O1–mFm) slab (YO0.8F1.4 MX 2.2)

Figure 2.16 The structure of Y5O4F7 (right) can be built up from Y(O, F) layers of the fluorite type
alternating with hexagonal F anion layers with a denser packing than in the square grid. In the Y(O, F)
slab, cation sites below the anion layer are shown in paler gray.

becomes MX + 1.155X = MX2.155. Through small adjustments in the size of the triangular
anion nets, m in YO1−mF1+2m can be varied from 0.13 to 0.22 (YO0.87F1.26 to YO0.78F1.44).
The way in which the square and triangular grids match up in the plane perpendicular to
our imagined stacking direction can lead to large unit cells (analogous to the situation
drawn in the lower part of Figure 2.14). In the m = 0.2 case of Y5O4F7 shown on the right of
Figure 2.16, six triangles line up with five squares; in m = 0.167 Y6O5F8, seven triangles line
up with six squares. These ideas are explored in Problem 2.27. In these phases, the Y sites
have local coordination numbers varying between eight (like fluorite) and six, and this type
of behavior is therefore most likely in materials where cations have flexible coordination
environments.
Similar features occur in a range of structures built from different slabs or columns.
Traditionally, the structures have been described using large unit cells, but the continuous
range of structures within a given system means that they are again often better described as
incommensurately modulated structures [17]. While we’ve only touched on a couple of
examples, the complexity and range of structures possible in real materials should be
apparent.

2.12 Problems
2.1 Given that Cu adopts a ccp structure with a cubic cell parameter of 3.615 Å, confirm that the
equilibrium number of vacancies in a 1 cm3 sample at 1000 °C is around 1019. Note the need to
convert between k, in J/K, and R, in J/(K mol), when using molar quantities: NAk = R.
82 Defects and More Complex Structures

2.2 Calculate the fractional number of vacancy sites in Cu at (a) 300 K, (b) 800 K, and (c) its
melting point (1357 K).
2.3 Assuming a unit-cell parameter of 5.62 Å, estimate the equilibrium number of Schottky
defects in a 1 mm3 grain of NaCl at 300 K and at 700 K.
2.4 Derive the equations given in the text for the number of defects at equilibrium, neq, for
Schottky and Frenkel defects. You will find the following expressions useful: for large x,
d  d 
lnðx!Þ≈ xlnðxÞ  x, ðc  xÞlnðc  xÞ ¼ lnðc  xÞ  1, and xlnðxÞ ¼ lnðxÞ þ 1.
dx dx
2.5 A sample of nonstoichiometric nickel oxide (A) was found to contain 77.70% Ni by mass. (a)
Calculate the empirical formula of A and state the two alternatives for the intrinsic defect that
would on its own give rise to this formula. (b) A has the NaCl-type structure and an
experimental density of 6526 kg/m3. Assuming a cell parameter of 4.180 Å, state which of
the two defects is present. (c) State how the cell parameter of A could be determined
experimentally and suggest how it would compare to that of stoichiometric NiO.
2.6 A brown sample of zinc oxide was found to have the hexagonal wurtzite structure with
a = b = 3.2495 Å, c = 5.2069 Å (α = β = 90°; γ = 120°). Chemical analysis gave 80.765% Zn
by mass. Density measurements gave 5810 kg/m3. Determine the formula of the material
and state whether it contains oxygen vacancies or interstitial metal atoms.
2.7 Suggest oxidation states for the metal ions in each of the following materials: (a) FeO,
Fe0.872O, Fe3O4, FeS2; (b) FeTe, Fe1.1Te; (c) LaOFeAs, LaO0.9F0.1FeAs; and (d)
YBaFe2O5, NdBaFe2O5.5, and NdBaCo2O6.
2.8 Suggest oxidation states for the metal ions in each of the following materials: (a) TiS2,
Li0.7TiS2; (b) LaMnO3, La0.8Sr0.2MnO3, La0.5Ca0.5MnO3; (c) La2CuO4, La1.85Ba0.15
CuO4, La2CuO4.075; and (d) BaPbO3, BaBiO3, Ba0.6K0.4BiO3.
2.9 TiO with a 1:1 ratio of Ti:O was synthesized and found to have a NaCl-related structure
with a = 4.1831 Å and an experimental density of 4927 kg/m3. Comment on these values.
2.10 The table below gives cell parameters and densities for a range of TixOy materials.
Determine the defects present in each. From a graph of your results, estimate x for a 1:1
stoichiometric sample TixOx.

z in TiOz Cell (Å) Density (g/cm3)

1.32 4.1608 4.713

1.12 4.1755 4.867
0.69 4.2212 4.992

3
2.11 NbO has an NaCl-related structure, a cell parameter of 4.21 Å and a density of 7.27 g/cm .
Calculate the percentage of vacant sites in the material. Draw a sketch of how the vacancies
can be arranged in an ordered way so as to give square-planar coordination of Nb. What is
the O coordination?
 2.12 Problems 83

2.12 GaAs1−xPx has a unit-cell parameter of 5.59 Å. Calculate x and estimate the band gap
(Eg) of the material, given aGaAs = 5.65 Å, Eg = 1.42 eV; aGaP = 5.45 Å, Eg = 2.24 eV.
2.13 Using the cell parameters quoted in Section 2.5, confirm the form of Equation (2.9).
2.14 Using the cell-parameter information stated in Section 2.5, calculate the cell parameter
expected for In0.76Ga0.24P0.47As0.53. Would this composition be lattice matched to any of
the four possible end members: InP, GaP, InAs, or GaAs?
2.15 State the Kröger–Vink notation for the predominant defects in each of the materials in
Table 2.2.
2.16 Suggest what type of twinning might occur in: (a) an orthorhombic structure with two
cell edges approximately equal; (b) a structure with a monoclinc cell with β = 90.1°; (c) an
orthorhombic structure with cell parameters a = 3.92 Å, b = 11.21 Å, c = 7.88 Å; and (d)
a structure with a conventional primitive monoclinic unit cell with a ≈ c.
2.17 At high temperatures, BaTiO3 has the cubic perovskite structure. On cooling, it under-
goes a series of phase transitions in which the Ti atom moves away from the center of the
TiO6 octahedron, and there are changes in the space group and unit-cell parameters
causing the ferroelectric behavior discussed in Chapter 8. Determine the number of twin-
domain variants (orientations) that could form in the first step when the cubic (a ≈ 4 Å,
space group Pm3m) structure undergoes a phase transition to a tetragonal structure with
space group P4mm and cell parameters a = b ≈ c ≈ 4 Å, α = β = γ = 90°.
2.18 At high temperature, Cu3Au has a disordered fcc structure with space group Fm3m. On
cooling, an ordering transition occurs (similar to that in Figure 4.14) and the low-temperature
structure has space group Pm3m with Au at 0 0 0 (Wyckoff site 1a) and Cu at ½ ½ 0 (3c). (a)
Sketch the low- and high-temperature structures. (b) Is this a translationengleiche or klassen-
gleiche transition? (b) State the number of domain variants in the low-temperature structure
and whether they will be related by twin or antiphase boundaries.
2.19 Like tungsten oxides, molybdenum oxides show a variety of crystallographic shear
structures. The unit cell for one of them is shown here:

The octahedra also share corners with octahedra in identical layers above and below the
84 Defects and More Complex Structures

plane of the paper. What is the composition of this compound? What is the oxidation state
of Mo?
2.20 A mixture of WO3 and MoO3 was heated with a small excess of W and Mo metals
to give a blue single-phase product containing 25.09% O by mass and a 1:1 W:Mo
ratio. Determine the empirical formula of the product and suggest its structure
type.
2.21 Assuming the presence of a V12T4 defect cluster of the type shown in Figure 2.13,
it is possible to build an ordered superstructure for Fe1−xO using a 5a × 5a ×
10a unit cell. This supercell would contain 16 vacancy clusters. Calculate the
composition.
2.22 Assuming that the unit-cell parameter of Fe1−xO is given by Vegard’s law as a = 4.3325 −
0.4103x Å, calculate the experimental density for (a) the hypothetical stoichiometric FeO
and (b) an Fe-deficient material with an Fe:O ratio of 1.075.
2.23 A sample of Fe1−xO has a measured density of 5491 kg/m3 and cell parameter of 4.281 Å.
Estimate the composition based on these two observations.
2.24 Sketch a V4T cluster for the Fe1−xO structure showing both O and vacant Fe sites.
Indicate the nearest shell of occupied octahedral Fe sites relative to the vacant site. State
how many of these sites must on average be occupied by Fe3+ to achieve electroneutrality.
Write down the Kröger–Vink notation of each defect present.
2.25 Assume that the defect structure of a crystal of Fe1−xO can be described in terms
of V13T4 clusters arranged such that they have a cubic cell with a = 2.7 × aFeO,
and that on average 50% of these cells contain a defect cluster. Calculate the value
of x.
2.26 A layered material with approximate composition PbNbS3 was found to have a very
similar X-ray-diffraction pattern to that of Sn1.17NbS3.17. Single-crystal studies revealed
that the PbS portion could be described with cell parameters of a = 5.834 Å, b = 5.801 Å,
c = 11.90 Å and the NbS2 part with cell parameters of a = 3.313 Å, b = 5.801 Å, c =
2 × 11.90 Å (ref. [18]). (a) Give a brief description of the likely structure of “PbNbS3”. (b)
State how large the a axis would need to be to describe this structure using a conventional
crystallographic unit cell. (c) From the cell parameters given, calculate the true compos-
ition of “PbNbS3”. (d) Suggest why the c axis of the NbS2 part of the structure is double
that of the PbS part. (e) Suggest a likely structure for a material of composition
Pb1.14Nb2S5.14.
2.27 The structure of Y(O, F)2+δ materials can be described in terms of a YX slab based on
a square grid of anions alternating with a hexagonal anion grid as shown in Figure 2.16.
(a) Confirm that the composition of a material with ideal anion arrays (all nearest anion–
anion distances equal) would be MX2.155. (b) State whether you’d expect this MX2.155
material to be commensurate or incommensurate. (c) Assume that the triangles are
compressed slightly in the b direction such that horizontal rows of anions align after
five squares and six triangles as in Figure 2.16. Calculate the composition of this material.
Repeat your calculation for a system in which anions align after six squares and after
 2.14 References 85

seven triangles. (d) State the relationship you’d expect between the square and triangular
anion grid in Y7O6F9.

2.13 Further Reading

Defects: F.A. Kröger, “The Chemistry of Imperfect Crystals” (1964) North-Holland; K. Kosuge,
“Chemistry of Nonstoichiometric Compounds” (1994) Oxford Science Publications; O. Toft
Sorensen, “Nonstoichiometric Oxides” (1982) Materials Science Series, Academic Press; R.J.
D. Tilley, “Principles and Applications of Chemical Defects” (1998) CRC Press.
Twinning: A. Putnis, “Introduction to Mineral Sciences” (1992) Cambridge University Press; U. Müller
“Symmetry Relationships between Crystal Structures” (2013) Oxford University Press.
Crystallographic shear: B.G. Hyde, S. Andersson, “Inorganic Crystal Structures” (1989) Wiley.
Modulated structures: S. van Smaalen, “Incommensurate Crystallography” (2007) Oxford University
Press.

2.14 References
[1] A.C. van den Berg, C.O. Aréan, “Materials for hydrogen storage: Current research trends and
perspectives” Chem. Commun. (2008), 668–681.
[2] U. Eberle, M. Felderhoff, F. Schüth, “Chemical and physical solutions for hydrogen storage”
Angew. Chem. Int. Ed. Engl. 48 (2009), 6608–6630.
[3] L.J. Murray, M. Dinca, J.R. Long, “Hydrogen storage in metal–organic frameworks” Chem.
Soc. Rev. 38 (2009), 1294–1314.
[4] W.I.F. David, M.O. Jones, D.H. Gregory, C.M. Jewell, S.R. Johnson, A. Walton, P.P. Edwards,
“A mechanism for non-stoichiometry in the lithium amide/lithium imide hydrogen storage
reaction” J. Am. Chem. Soc. 129 (2007), 1594–1601.
[5] A.R. Allnatt, A.B. Lidiard, “Atomic Transport in Solids” (1993) Cambridge University Press.
[6] S. Gehrsitz, H. Sigg, N. Herres, K. Bachem, K. Kohler, F.K. Reinhart, “Compositional depend-
ence of the elastic constants and the lattice parameter of AlxGa1−xAs” Phys. Rev. B 60 (1999),
11601–11609.
[7] T. Taniguchi, S. Mizusaki, N. Okada, Y. Nagata, S.H. Lai, M.D. Lan, N. Hiraoka, M. Itou,
Y. Sakurai, T.C. Ozawa, Y. Noro, H. Samata, “Crystallographic and magnetic properties of the
mixed valence oxides CaRu1−xMnxO3” Phys. Rev. B 77 (2008), 014406/1–7.
[8] A.D. Wadsley, “Nonstoichiometric metal oxides” Advances in Chemistry Series 39 (1963), 23–36.
[9] J.L. Murray, H.A. Wriedt, “The O–Ti (Oxygen–Titanium) system” Bull. Alloy Phase Diagrams 8
(1987), 148–165.
[10] B.G. Hyde, S. Andersson, “Inorganic Crystal Structures” (1987) J. Wiley and Sons.
[11] A. Shinsaku, D. Bogdanovski, H. Yamane, M. Terauchi, R. Dronskowski, “ε-TiO, a novel stable
polymorph of titanium monoxide” Angew. Chem. Int. Ed. Engl. 55 (2016), 1652–1657.
86 Defects and More Complex Structures

[12] F. Koch, J.B. Cohen, “The defect structure of Fe1−xO” Acta Crystallogr. Sect. B 25 (1969),
275–287.
[13] T.R. Welberry, A.G. Christy, “Defect distribution and the diffuse X-ray diffraction pattern of wustite
Fe1−xO” Phys. Chem. Miner. 24 (1997), 24–28.
[14] T. Wagner, A. Schönleber, “A non-mathematical introduction to the superspace description of
modulated structures” Acta Crystallogr. Sect. B 65 (2009), 249–268.
[15] O. Gourdon, V. Petricek, M. Dusek, P. Bezdicka, S. Durovic, D. Gyepesova, M. Evaina,
“Determination of the modulated structure of Sr14/11CoO3 through a (3 + 1)-dimensional space
description and using non-harmonic ADPs” Acta Crystallogr. Sect. B 55 (1999), 841–848.
[16] V. Pavlyuk, I. Chumak, L. Akselrud, S.Lidin, H. Ehrenberg, “LiZn4−x (x = 0.825) as a (3 + 1)-
dimensional modulated derivative of hexagonal close packing” Acta Crystallogr. Sect. B 70
(2014), 212–217.
[17] S. Schmid, “The yttrium oxide fluoride solid solution described as a composite modulated
structure” Acta Crystallogr. Sect. B 54 (1998), 391–398.
[18] G.A. Wiegers, A. Meetsma, R.J. Haange, S. Van Smaalen, J.L. De Boer, A. Meerschaut,
P. Rabu, J. Rouxel, “The incommensurate misfit layer structure of (PbS)1.14NbS2, ‘PbNbS3’, and
(LaS)1.14NbS2, ‘LaNbS3’: An X-ray diffraction study” Acta Cryst. Sect. B 46 (1990), 324–332.
3 Defect Chemistry and Nonstoichiometry

One key aspect of materials chemistry is the ability to prepare materials with precisely
controlled composition. We’ll see countless times in later chapters that ever minor changes
in chemical composition can hugely influence a material’s properties. Defects of the type
we’ve met in Chapter 2 play a key role in both synthesis and composition control. In the first
sections of this chapter, we’ll investigate how simple ideas of chemical equilibria can give us
qualitative and quantitative insights into defect formation. In the second half of the chapter,
we’ll look at the diffusion of different types of defects, which controls the reactivity of solids
and the properties of some functional materials.

3.1 Narrow Nonstoichiometry in Oxides

We learned in Chapter 2 that entropy favors a certain small number of defects in all extended
solids. In a metal oxide, defects such as metal or oxygen vacancies and interstitials give rise to
a narrow range of oxygen nonstoichiometry around the integer oxidation state of the metal.

3.1.1 Point Defects in a Pure Stoichiometric Oxide

Let’s use NiO as an example. Formation of vacancies (Figure 2.4) is one of several possible
ways for intrinsic ionic defects to occur in the NaCl-type structure of NiO. In Equations (2.3)
to (2.6) we saw that the mass-action law applies to defect formation as if it were a chemical
reaction. Using the Kröger–Vink notation (Section 2.6), we can write “chemical” equations
for the formation of all possible intrinsic-defect pairs in stoichiometric (1:1) NiO. Table 3.1
shows there are four such pairs:
The term “nil” denotes the value of 0 obtained after crossing out the regular structure sites on
both sides of the equation. The first four equations describe structural defects already introduced
in Chapter 2. The last equation describes electronic defects. We symbolize them as electrons and
holes, but in a redox-prone oxide such as NiO, the electron e0 would behave as Ni+ (an aliovalent

87
88 Defect Chemistry and Nonstoichiometry

Table 3.1 Formation equations for the four alternative intrinsic

ionic-defect pairs in NiO and for intrinsic ionization.

Process Reaction

Schottky: nil = vNi 0 0 + vO••

Anti-Schottky: NiNi˟ + OO˟ = Nii•• + Oi 0 0
Cation-Frenkel: NiNi˟ = vNi 0 0 + Nii••
Anion-Frenkel: OO˟ = vO•• + Oi 0 0
Intrinsic ionization: nil = e 0 + h• (2NiNi˟ = NiNi 0 + NiNi•)

defect NiNi 0 in the Kröger–Vink notation), whereas the hole h• would represent the oxidized state
Ni3+ (NiNi•). The last reaction in Table 3.1 is in principle a disproportionation of divalent nickel
into mono- and trivalent defects.1

3.1.2 Point Defects upon Oxidation/Reduction of the Stoichiometric Oxide

The possible changes in the numbers of intrinsic defects on oxidation and reduction of the
stoichiometric NiO are in Figure 3.1. When NiO is reduced within its homogeneity range,2
either an O deficit appears at its regular structural sites or an excess of Ni at interstitial
sites, yielding NiO1−δ or Ni1+δO, respectively. Upon oxidation, either an O excess or
a Ni deficit may occur as NiO1+δ or Ni1−δO. These alternative responses are based on
the four possible intrinsic point-defect reactions in Table 3.1. Which of them actually
dominates in a particular material can only be answered by a rather involved experi-
mental study.
One useful rule is that large closest-packed atoms form defects less readily than
smaller atoms located in the holes. In nickel oxide and other 3d monoxides, oxidation
(adding oxygen) creates metal vacancies vNi 0 0 rather than inserting bulky oxygen
interstitials. Likewise, vCr 0 0 0 forms in Cr2O3 and other corundum-type 3d oxides.
Reduction (oxygen removal) may form interstitials such as Zni•• in ZnO. However, if
the size difference is less pronounced, this approximation fails: CdO with the NaCl-
type structure, for example, forms oxygen vacancies upon reduction. While the defect
type follows from the structure, the propensity for a dominant oxidative or reductive
nonstoichiometry depends on the chemistry. A significant nonstoichiometry often
develops towards another stable oxidation state of the metal. For example, FeO
tends to be oxidized towards FeIII.

1
Chemically, 2Ni2+ = Ni+ + Ni3+. We will see in Chapter 10 that nil = e 0 + h• represents a thermal excitation of an
electron from a valence band to a conduction band, which leaves a hole behind in the valence band. In a redox oxide
such as NiO, the electron becomes trapped as Ni+ close to the conduction band, whereas the hole is trapped as Ni3+
close to the valence band. In Chapter 7, we’ll see that optical excitation also creates an electron–hole pair.
2
By “reduction” or “oxidation” in the context of defects, we’ll mean a trace reduction or trace oxidation that creates
nonstoichiometry but does not decompose the phase.
 3.1 Narrow Nonstoichiometry in Oxides 89

Figure 3.1 The

four alternative point-
defect compensations
for reduction and oxi-
dation of stoichiomet-
ric (1:1) NiO are
anion-Frenkel the four alternative
intrinsic-defect pairs
named on the right.
cation-Frenkel

3.1.3 Equilibrium Equations for Oxidative and Reductive Nonstoichiometry

Just as we did for intrinsic defects in Section 3.1.1, we can write equilibrium equations
for the point-defect changes that occur on oxidation or reduction of an oxide. Let’s
assume O2 to be the only gaseous reactant3 in the intrinsic redox reactions of NiO at
high temperature.4 For clarity, we’ll proceed via an auxiliary equation of a small amount
of oxygen, δO, reacting with NiO. As shown in the middle column of Table 3.2,
oxidation forms holes compensated by the intrinsic negative defect vNi ″ or Oi ″
(Figure 3.1) in an equation balanced in terms of charges, sites (denoted in subscripts),
and chemical elements. Analogously so for reduction and electrons in Table 3.3. Next,
the auxiliary reactions are simplified into the defect-formation equation on the right of
Table 3.2 and Table 3.3 by crossing out species occurring on both sides, changing to O2,
and removing fractions.

Table 3.2 Two alternative compensations of oxidative nonstoichiometry in NiO.

Auxiliary scheme Equation for oxidation

•
vNi ″ NiO + δO → δvNi 0 0 + 2δh + δOO˟ + NiNi˟ + OO˟ O2(g) ⇄ 2vNi 0 0 + 4h• + 2OO˟
Oi ″ NiO + δO → δOi 0 0 + 2δh• + NiNi˟ + OO˟ O2(g) ⇄ 2Oi 0 0 + 4h•

3
An analogous set of equations would have to be set up for every other reacting gas species such as atomic oxygen,
ozone, or nickel vapor.
4
Temperature T provides the needed activation energy (for dissociation of O2 and diffusion of defects) in the form of
thermal energy kT (k is the Boltzmann constant, 1.380649×10−23 J/K, see Appendix J).
90 Defect Chemistry and Nonstoichiometry

Table 3.3 Two alternative compensations of reductive nonstoichiometry in NiO.

Auxiliary scheme Equation for reduction

Nii•• NiO − δO → δNii•• − δNiNi˟ + 2δe 0 − δOO˟ + NiNi˟ + OO ˟ 2OO˟ + 2NiNi˟ ⇄ 2Nii•• + 4e 0 + O2(g)
vO•• NiO − δO → δvO•• + 2δe 0 − δOO˟ + NiNi˟ + OO˟ 2OO˟ ⇄ 2vO•• + 4e 0 + O2(g)

3.1.4 Defect Equilibria for Schottky-Type Redox Compensation

In this section, we’ll explore the redox defect chemistry and nonstoichiometry of NiO if
Schottky intrinsic-defect pairs (cation and anion vacancies) dominate. We will therefore
need to consider two intrinsic-defect pairs—the Schottky pair vO•• and vNi ″ is one, the
other is e 0 and h•. There will be two intrinsic-pair formation reactions and two5 redox-
defect reactions, one for oxidation one for reduction; the equilibria 6 are summarized in
Table 3.4:

Table 3.4 Schottky-type* redox compensation in NiO.

Process Reaction equation Reaction quotient

Schottky nil ⇄ vNi 0 0 + vO•• KS = [vNi 0 0 ][vO••]

Ionization nil ⇄ e 0 + h• Ki = [e 0 ][h•]
Oxidation O2(g) ⇄ 2vNi 0 0 + 4h• + 2OO˟ Kox = [vNi 0 0 ]2[h•]4·pO2−1
Reduction 2OO˟ ⇄ 2vO•• + 4e 0 + O2(g) Kred = [vO••]2[e 0 ]4·pO2
* Reaction quotients (but not the full reaction equations) for anti-Schottky, Frenkel, and anti-Frenkel
intrinsic defect pairs are obtained upon appropriate substitution of defect symbols: Nii•• ≡ vO•• and Oi 0 0 ≡
vNi 0 0 .

Only three of the four reaction quotients (mass-action terms) are independent, as the
equilibrium constants combine to KS 2  Ki 4 ¼ Kox  Kred . We have four unknowns (the frac-
tion of holes, electrons, nickel vacancies, and oxygen vacancies per NiO formula of regular
sites), and we need one more equation. This comes from the electroneutrality condition
requiring equal amounts of positive and negative charges:

2½vNi ″ þ ½e0  ¼ ½h•  þ 2½vO ••  (3.1)

5
Both involve O2 gas. If nickel vapor were present as well, two additional redox equations would follow—one for
oxidation of the vapor by holes into regular nickel sites, Ni(g) + vNi 0 0 + 2h• = NiNi˟, and one for reduction of NiNi˟ to
nickel vapor, NiNi˟ + vO•• + 2e 0 = 2Ni(g). One of their two reaction quotients would be independent of all the others,
which allows us to solve one more variable, the Ni vapor pressure.
6
The unexpressed fractional concentrations of atoms at their regular sites are set to unity due to the low concentration of
defects.
 3.1 Narrow Nonstoichiometry in Oxides 91

The four independent equations can then be chosen and rearranged such that each of them
expresses the fraction of only one type of defect as a function of the partial pressure of
oxygen, pO2, having three equilibrium constants as parameters. The equations7 can be solved
either analytically or numerically to give the fractions of defects. Once this is done, the
oxygen nonstoichiometry [vNi 0 0 ] − [vO•• ] can be evaluated8 as a function of pO2.
Figure 3.2 shows logarithmic plots of defect fractions versus pO2 for two limiting cases of
dominating intrinsic-defect pair; ionic (left), electronic (right).9 The approximate oxygen
nonstoichiometry is given in the upper plots. High oxygen pressures lead to excess oxygen in
the oxide; low pressures to oxygen deficiency. The defect-concentrations in the lower plots
illustrate what causes this simple composition change—changing fractions of electronic
defects [e 0 ] and [h•], and of ionic defects [vNi 0 0 ] and [vO•• ]. High [vNi 0 0 ] and [h• ] emerge upon

Figure 3.2 Oxygen nonstoichiometry (top) and defect fractions per formula (bottom) in NiO with domin-
ant ionic defects (left; KS = 10−6, Ki = 10−9, Kox = 10−9) and with dominant electronic defects (right; KS =
10−12, Ki = 10−6, Kox = 10−9).

7
2KS3/2Kox1/4 + KiKSpO2−1/4[vO•• ]1/2 − Kox1/2pO21/4[vO•• ]3/2 − 2Kox1/4KS1/2[vO••]2 = 0
−2KSKox1/4 − Kox1/2pO21/4[vNi 0 0 ]1/2 + KipO2−1/4[vNi 0 0 ]3/2 + 2Kox1/4[vNi 0 0 ]2 = 0
2KoxpO2 + Ki(Kox pO2)1/2[h•] − (Kox pO2)1/2[h• ]3 − 2KS[h• ]4 = 0
−2KSKi2 − Ki(Kox pO2)1/2[e 0 ] + (Kox pO2)1/2[e 0 ]3 + 2(Kox pO2/Ki2)[e 0 ]4 = 0.
The unit of concentration is the same as in the equilibrium constants.
8
The maximum error due to this simplification of the precise δ = ([vNi 0 0 ] − [vO••])/(1 − [vNi 0 0 ]) in NiO1+δ is 1% of the
nonstoichiometry value at the left edge of the graph in Figure 3.2.
9
PbO is an example of predominant ionic and CuO of predominant electronic defects.
92 Defect Chemistry and Nonstoichiometry

oxidation in oxygen-rich atmospheres, high [vO••] and [e 0 ] upon reduction in oxygen-poor

atmospheres.10 In between, there are two important points: Zero nonstoichiometry is
associated with the point of integer structure where occupied metal and oxygen sites have
the stoichiometric ratio 1:1 and the intrinsic structural defects compensate each other, [vNi 0 0 ]
= [vO••]. The point where the electronic defects compensate each other, [e 0 ] = [h•], is the point
of integer valence.11 For a pure binary oxide, these two points coincide on the pO2 scale.
In several regions of Figure 3.2, the defect fractions have essentially linear variation on the
log–log scale. This occurs in ranges where a pair of mutually compensating defects dominates.
When Schottky vacancies dominate, the electroneutrality condition in Equation (3.1) simpli-
fies to [vNi 0 0 ] = [vO••] = constant. The mass-action equations for oxidation and reduction in the
last two rows in Table 3.4 then show that [h•] is proportional to pO1/4 2
and [e 0 ] is proportional to
−1/4
pO2 . This holds in the central part of the plot in Figure 3.2, left, where the slopes on the
log–log plot are +¼ and −¼, respectively. In Figure 3.2, right, the central range is dominated
by electronic defects, [e 0 ] = [h•] = constant, and analogous substitution gives [vNi 0 0 ] proportional
to pO1/2
2
and [vO••] proportional to pO−1/22
. In the most oxidized region, holes are compensated
by metal vacancies. The electroneutrality condition is 2[vNi 0 0 ] = [h•], and substitution for [h•]
in the mass-action term in the “Oxidation” row of Table 3.4 shows that [vNi0 0 ] is proportional
to pO1/6
2
. In the most reduced region, electrons are compensated by oxygen vacancies. The
electroneutrality condition is [e 0 ] = 2[vO••]. Substitution for [e 0 ] in the mass-action term in the
“Reduction” row of Table 3.4 shows that [vO••] is proportional to pO−1/6 2
. The existence of linear
regions allows construction of approximate diagrams of defect fractions versus pO2 using
straight lines with these slopes—Brouwer diagrams [1], Figure 3.3.

Figure 3.3 Brouwer-diagram sketches corresponding to the previous figure.

10
Recall that h• is equivalent to Ni3+, e 0 to Ni+. 11
Or of integer oxidation state.
 3.1 Narrow Nonstoichiometry in Oxides 93

3.1.5 Acceptor-Doped Oxides

How do things change when we replace a small portion of Ni2+ with a cation of a lower and
fixed oxidation state (such as Li+), which is called acceptor12 doping? All the mass-action
equations from Table 3.4 remain the same. The only difference appears in the electroneu-
trality condition that now includes the additional defect LiNi 0 :

2½vNi 00  þ ½e0  þ ½LiNi 0  ¼ ½h•  þ 2½vO ••  (3.2)

Expressing defect fractions as functions of pO2 can be done in the same way as the pure-oxide
case, except that we now have four parameters that control the defect equilibria; the
three equilibrium constants and the fixed fraction of the acceptor defect, [LiNi 0 ].
Figure 3.4 shows how the defect concentrations of pure NiO in Figure 3.2 change when
[LiNi 0 ] = 0.02. The Li+ acceptor moves the point of integer valence, [e 0 ] = [h•], to lower pO2
because Ni2+ is now more easily oxidized in order to keep the charges balanced. The point of
integer structure, [vNi 0 0 ] = [vO••], moves towards higher pO2 because oxygen vacancies are

Figure 3.4 Defect fractions per formula in Li-doped NiO, [LiNi 0 ] = 0.02, for dominant ionic (left) and
electronic (right) defects with equilibrium constants as in Figure 3.2. The square in the chemical formula
in the top left stands for vacancies.

12
In electronics, an acceptor is a neutral atom with fewer valence electrons than the regular atom. In our context, it is
an atom of a fixed oxidation state lower than the matrix atom it replaces.
94 Defect Chemistry and Nonstoichiometry

more easily formed after LiO1/2 has dissolved in NiO and a higher O2 pressure is then needed
to fill them. Of the two readily formed defects h• and vO•• that compensate the LiNi 0 acceptor,
holes h• dominate around the point of integer structure and vacancies vO•• around the point
of integer valence.
As before, some regions of Figure 3.4 have a practically linear dependence on log pO2. The
reasons for linearity in the extreme oxidized and reduced regions are the same as for
the undoped oxide. The integer-valence point and the integer-structure point split due to
the doping, and we get approximately linear ranges around both. Because vacancies vO••
compensate the acceptor around the point of integer valence, the electroneutrality condition
in Equation (3.2) simplifies there to [LiNi 0 ] = 2[vO••] = constant. As this also keeps [vNi 0 0 ]
constant, mass-action equations for oxidation and reduction in the last two rows in Table 3.4
give [h•] proportional to pO1/42
and [e 0 ] proportional to pO−1/4
2
. Holes h• compensate the
acceptor around the point of integer structure, hence the electroneutrality condition in
Equation (3.2) simplifies to [LiNi 0 ] = [h•] = constant. Because this also keeps [e 0 ] constant,
mass-action equations for oxidation and reduction in the last two rows in Table 3.4 give [vO••]
proportional to pO−1/2
2
and [vNi 0 0 ] proportional to pO1/2
2
. Brouwer diagrams are explored in the
end-of-chapter problems.

3.1.6 Donor-Doped Oxides

Finally, let’s replace a small portion of Ni2+ in NiO with a higher-valent ion of fixed
oxidation state (for example Al3+)—donor13 doping. The mass-action equations from Table
3.4 remain the same, but the electroneutrality condition changes to:

2½vNi 00  þ ½e0  ¼ ½AlNi •  þ ½h•  þ 2½vO ••  (3.3)

Figure 3.5 shows that the defect fractions and oxygen nonstoichiometry as a function of the
partial pressure of oxygen are the inverse of the acceptor case for the same level of doping
[AlNi•]. Relative to the pure oxide, it is now easier to reduce Ni2+ and to create nickel
vacancies. These [e 0 ] and [vNi ″ ] compensate [AlNi•].

3.1.7 Solid Solubility of Dopants

In the previous two subsections on doping, we implicitly assumed that the doping level was
always lower than the actual solid solubility of the dopant. In fact, the donor or acceptor
solubility in oxides depends on the temperature and partial pressure of oxygen. Before we
make some qualitative considerations on the effects of these two variables, let’s consider
a simpler case of isovalent substitution.
For isovalent substitutions, the effect of temperature on solubility is straightforward.
A solution has higher entropy than the sum of its pure components, and increasing
13
In our context, a donor has a fixed oxidation state higher than the matrix atom it replaces.
 3.1 Narrow Nonstoichiometry in Oxides 95

Figure 3.5 Defect fractions per formula in Al-doped NiO, [AlNi•] = 0.02, for dominant ionic (left) and
electronic (right) defects with equilibrium constants as in Figure 3.2.

temperature will increase the entropy term TΔS that drives the dissolution, thus increasing
the solubility. We know empirically that chemically similar isostructural crystals of close
enough14 atomic radii mix completely at high temperatures, but calculations [2] do show that
they would eventually demix at low temperatures.
For aliovalent defects, oxidation and reduction make the picture more complex. Let’s
consider first the effect of temperature. In an oxide15, increased temperature will favor loss of
oxygen (chemical reduction) because the released gas has high entropy. Donor doping makes
reduction easier because we can think of the donor oxide (like AlO3/2 dissolved in NiO) as
bringing excess oxygen into the structure. Increased temperature will therefore favor dissol-
ution of the donor also by promoting reduction (not only by increased TΔS of mixing). The
situation is different for an acceptor. Since an acceptor introduces an oxygen deficit into
the structure, the further oxygen loss at high temperature will disfavor this dissolution, but
the general entropic driving force of dissolution will still be present. The effect of increasing
temperature on the solid solubility of an acceptor therefore can’t be predicted.
The effect of the partial pressure of oxygen can be estimated from the Le Chatelier
principle16. Dissolution of a donor oxide brings excess oxygen, thus a tendency for easy

14
Size differences less than 15% typically allow complete solubility at high temperatures.
15
In particular of a redox-active element such as Ni.
16
Equilibrium will adjust to minimize the effects of external changes.
96 Defect Chemistry and Nonstoichiometry

Table 3.5 Dissolution of Al donor in NiO around the point of integer structure.

Dissolution reaction Mass-action term

Al2O3(s) ⇄ 2e 0 + 2AlNi• ˟
+ 2OO + ½O2(g) Ksd = [AlNi•]2[e 0 ]2pO21/2

Table 3.6 Dissolution of Al donor in NiO around the point of integer valence.

Dissolution reaction Mass-action term

Al2O3(s) ⇄ vNi 0 0 + 2AlNi• + 3OO˟ Ksd = [AlNi•]2[vNi 0 0 ]

loss of O2, and this will be favored at low partial pressures of oxygen. Therefore, low pO2 will
increase donor solubility, whereas high pO2 will suppress it. For an acceptor, the opposite
holds. The dissolved acceptor oxide introduces an oxygen deficit, implying a tendency for
easy inclusion of the missing oxygen into the solid; high pO2 will promote this process and
increase the solid solubility of the acceptor.
The combination of these temperature- and pO2 effects is unambiguous only for dissol-
ution of the donor oxide—high temperature and low pO2 will increase the donor’s solid
solubility. To evaluate this quantitatively, we must include the dissolution reaction of the
donor oxide in the set of equilibrium defect-reaction equations. Let’s consider the dissolution
of the trivalent donor Al in the integer structure of NiO. The defect predominantly compen-
sating the donor will be e 0 (see Figure 3.5 right), and the dissolution reaction will maintain
the 1:1 ratio of the anion and cation sites as shown in Table 3.5.
At equilibrium with excess Al2O3, [AlNi•] is the solid-solubility fraction x of the donor, and
we see from the equilibrium equation that [AlNi•] = [e 0 ]. Substitution into the mass-action
term in Table 3.5 gives x = Ksd¼pO2−⅛, and we observe that the solid solubility of the donor
increases when pO2 decreases.
Similarly, we can evaluate the donor solubility about the point of integer valence. The
predominant defects compensating the donor will be vNi 0 0 (see Figure 3.5 left). Table 3.6
above shows that the donor solubility is independent of the partial pressure of oxygen when
the defect that predominantly compensates the donor is a structural point defect.

3.1.8 Cautionary Note on Defect Models in Pure Oxides

A pitfall of defect modeling in “pure” oxides is that they are never truly pure. In any oxide,
there will be impurities influencing the defect equilibria at some level. Furthermore, some
deceptively simple oxides, such as the NaCl-type wüstite of ideal composition FeO, are
grossly nonstoichiometric and exhibit clustering of defects (Section 2.9). Real materials
 3.1 Narrow Nonstoichiometry in Oxides 97

Box 3.1 Synthetic Methods: Oxygen-nonstoichiometry control in oxides

A precise and homogeneous (non)stoichiometry is important for many functional oxide
materials. All methods to achieve this start with equilibrium of the oxygen exchange between
the point defects in the solid and the surrounding gas atmosphere. Two processes occur. The
first is the surface reaction where O2 splits to, or forms from, two oxide anions and four holes.
This is a redox reaction since the holes represent the oxidized state of the metal. The subsequent
process is a diffusion-driven homogenization, during which the surface-oxygen excess (or
deficit) homogenizes throughout the bulk. The oxygen nonstoichiometry can be controlled
by the following parameters, depending on whether the system is closed or open:

Control parameters Redox reagent System

pO2, T Flowing gas of given pO2 Open

T pO2 buffer Closed
Mass Oxygen getter or source Closed

The open systems use hot flowing reaction atmospheres with defined partial pressures of oxygen.
These can be mixtures of Ar and O2 down to pO2 = 10−4 bar; mixtures of Ar, H2, and H2O
(via H2O ⇄ H2 + ½O2) below pO2 = 10−10 bar; and CO/CO2 (via CO2 ⇄ CO + ½O2) in the
intermediate range. After isothermal equilibration, samples are quenched to low temperature
because otherwise they would oxidize during the cool-down.
The closed systems are usually set up in sealed ampoules. The pO2 buffer is a solid redox-couple
mixture that maintains constant pO2 when it is in excess of the nonstoichiometric sample and not in
contact with it. Like in the previous technique, the constant pO2 value is fixed by the temperature.
Gibbs phase rule (Chapter 4) states that F = C + 2 − P, where F is the degree of freedom, C is the
number of components, and P is the number of phases in equilibrium. As long as the buffer, say,
a homogeneous mixture of Ni and NiO, contains Ni and NiO1−δ at equilibrium, the pO2 above them
remains constant at a given temperature: 2 components + 2 intensive variables (temperature +
pressure) − 3 phases (Ni, NiO, O2) = 1 variable that can be varied independently in the current
phase system without a phase disappearing. The temperature will therefore fix the pO2. Almost any
pO2 can be achieved by a good choice of the redox couple in equilibrium with O2, from low (Ni–
NiO) to very high pO2 (Ag2O2–Ag2O in gold wraps in anvil cells).
The oxygen-getter/oxygen-source technique is similar, but here it is the amount of the added
substance that controls the oxygen taken up or released by the sample. Zirconium metal is
a good getter that completely and rapidly oxidizes into ZrO2 but only if relatively high pO2 levels
are generated by the sample. When very low pO2 is needed to reduce the oxide, Mg, Zn, or Fe
can be used. A common oxygen source is Ag2O, which releases all O2 above 450 °C.
A versatile modification of the getter/source technique is solid state coulometry, where
specific amounts of oxygen are dosed to or from an enclosed sample electrochemically, via
a window made of cubic stabilized zirconia; an excellent conductor of oxide ions (Chapter 13)
yet a poor electronic conductor. When appropriate electric charge is supplied from a Pt
electrode, the corresponding amount of oxide anions moves through the zirconia window.
98 Defect Chemistry and Nonstoichiometry

range from nearly ideal oxides with randomly distributed dilute point defects, such as Cu2O,
NiO, ZnO, and Cr2O3, to quasi-random distributions (CoO), to defect clustering (FeO), to
formation of defect-ordered superstructures (CeO2, PrO2), or infinitely adaptive defect-
ordered structures (e.g. Magnéli phases). Simple clustering, such as dimerization or
donor–vacancy association, is common for impurities (dopants) even in oxides that exhibit
a random distribution of point defects. In general, we must take extreme caution. On the
other hand, the next section shows that some complex oxides can have surprisingly simple
defect equilibria.

3.2 Wide Nonstoichiometry in Oxides

Wide nonstoichiometry ranges can occur between two oxidation states of an atom in
two closely related integer structures. An example is the YBa2Cu3O7−δ high-Tc super-
conductor (0 < δ < 1; Figure 3.6), in which one entire oxygen atom per formula can be
removed. In YBa2Cu3O6, one copper atom has a linear coordination typical of Cu+ and
the other two are Cu2+. In YBa2Cu3O7, the linear coordination becomes square planar,
consistent with Cu3+.
This type of nonstoichiometry can also be treated with defect equilibria. We only have to
decide which of the limiting structures (Figure 3.6) contains the integer valence that
corresponds to the intrinsic situation [e 0 ] = [h• ], where, as Figure 3.2 suggests, properties
related to the concentration of these charge carriers should achieve minimum values.
Electrical conductivity and oxygen diffusivity decrease towards YBa2Cu3O6 (as discussed
in ref. [3]), suggesting it is the integer-valence point. The other limit, YBa2Cu3O7, is then
defined as the integer-structure point by formally considering YBa2Cu3O7−δ as an
acceptor-doped Y3Cu3O7 with 2BaY 0 per formula.

YBa2Cu3O6 YBa2Cu3O7

Cu2+ Cu2+

Ba2+

Cu+ Cu3+

Ba2+

Cu2+ Cu2+

Y3+

Figure 3.6 Two limiting structures of YBa2Cu3O7−δ.

 3.3 Point Defects and Diffusion 99

The intrinsic ionic defects in the YBa2Cu3O7 integer structure are of the anion-Frenkel
type; oxygen vacancies and interstitials. We can then set up the redox compensations in
analogy with Table 3.4 and solve the mass-action equations together with the electroneu-
trality condition. The Frenkel-process reaction is OO x ¼ vO •• þ Oi 00 , the oxidation reaction
is ½O2ðgÞ ¼ 2h• þ Oi 00 , the reduction reaction is OO x ¼ 2e0 þ vO •• þ ½O2ðgÞ . A profound sim-
plification can be achieved by subtracting the Frenkel reaction from the oxidation, which
yields the intuitive reaction of vacancy filling and oxidation:

½O2ðgÞ þ vO •• ¼ OO x þ 2h• (3.4)

Equation (3.4) does not include the e 0 and Oi 0 0 defects important for reduction and oxidation
beyond the YBa2Cu3O6 and YBa2Cu3O7 limits, respectively, but it is reasonably valid for
YBa2Cu3O7−δ within this range. With vacancies counted from the point of integer structure
of the acceptor-doped model, ½vO ••  ≈ δ. With holes counted from the point of integer valence,
½h•  ≈ 2ð1  δÞ. The mass-action term for Equation (3.4) is then:

Kvox ¼ pO2½ 4ð1  δÞ2 =δ (3.5)

Equation (3.5) has a solution for δ as a function of pO2:

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 
Kvox 16 pffiffiffiffiffiffiffi pffiffiffiffiffiffiffi
δ¼1 pO2 þ pO2  pO2 (3.6)
8 Kvox

which describes the entire nonstoichiometry range well for temperatures below 600 °C. As an
example, Kvox = 55 bar−1/2 at 500 °C, and δ = 0.12 (YBa2Cu3O6.88) is calculated. That
matches both experiment [4] and the full defect-model approach within their uncertainties.
Above 600 °C, a progressive deviation from the experimental reality appears for the highest
and lowest values of δ calculated by Equation (3.6).
Further implications of the YBa2Cu3O7−δ nonstoichiometry are in the end-of-chapter
problems and in Chapter 12. This example illustrates that in some cases a simple approxi-
mation works for wide nonstoichiometry.

3.3 Point Defects and Diffusion

In Section 3.2, we discussed point defects taking part in the reaction of O2 with a crystalline
solid. How can oxygen move inside a compact crystal? How long does it take? In fact, the
oxygen speed in YBa2Cu3O7−δ is amazing; one day at 400 °C in O2 is enough to fully oxidize
a pellet of 1 cm in diameter. We can understand this by realizing that point defects are mobile
and disperse homogeneously in a material in order to maximize entropy. This is one of many
examples of a process called diffusion. Solid state diffusion is behind major industrial
processes, such as hardening of steel, fabrication of doped semiconductors, or “filtering”
100 Defect Chemistry and Nonstoichiometry

oxygen or hydrogen via permeable membranes. It also underpins some important technol-
ogy, such as the transport of Li+ in lithium batteries or of O2−/H+ in fuel cells.
Equations describing diffusion are deceptively simple. Fick’s first law states that
 
dc
J¼D  (3.7)
dy

when limited to one17 direction, y. It tells us that the flux J (the flow density of the diffusing
particle in the matrix or substrate18) is directly proportional to the negative gradient of the
concentration c of the particle along the flux direction.19 The minus sign is there because the
flow in positive direction occurs along a gradient of higher to lower concentrations, down
a negative slope, so we have to turn this negative gradient into a positive number. The
proportionality constant D is the diffusivity or diffusion coefficient.20 The diffusivity D is
characteristic for a particle–matrix pair at a given temperature and pressure; it is not
a constant. When the diffusion concerns changes in chemical composition, D is called the
chemical diffusion coefficient and is denoted as DA formally implying diffusion of a neutral
atom A.
Since time does not appear in Equation (3.7), Fick’s first law is valid for steady-state
diffusion, in which the particle flow into a system equals the flow out. An example is
permeation of CO2 (or heat) through a house wall into an open space. The CO2 (heat) flux
is constant in the steady state, and its concentration profile across the wall is a straight line of
negative slope.
Non-steady-state diffusion concerns the spread of diffusing particles as a function of time.
The time, τ, is introduced via a continuity consideration—the concentration increase per
time equals the net incoming flux per unit length (the negatively taken negative gradient of
the particle flux). At every instant, ∂c/∂τ in m−3/s equals −∂J/∂y in (s−1 m−2)/m. Plugging in
J from Equation (3.7) leads to Fick’s second law:
 
∂c ∂ ∂c
¼ D (3.8)
∂τ ∂y ∂y

If D is constant and independent of c, it comes before the differential, ∂c/∂τ = D(∂2c/∂y2). This
equation states that the change of concentration with time is proportional to the curvature of

17
It can be changed to three dimensions by adding terms for x and z, and treating D as a tensor if the material is
anisotropic. In our treatment, we will consider diffusion along one direction in an isotropic matrix of a 3D structure,
for which D behaves as a scalar.
18
In a broad sense, the substrate or matrix is everything except the particle whose diffusion we follow.
19
The flux is the number of atoms [J] = s−1 m−2, or moles, [J] = mol s−1 m−2, or kilograms, [J] = kg s−1 m−2 passing per
unit time through a unit cross-sectional area perpendicular to the flux. The respective concentrations are in m−3, or
mol m−3, or kg m−3.
20
A simple consideration for the units of D and, say, the mole flux: Particles that flow at a rate of 1 mol/s
perpendicularly through a 1 m2 window under a concentration gradient of 1 (mol/m3)/m have diffusivity D =
1 m2/s. Substitute into Equation (3.7).
 3.3 Point Defects and Diffusion 101

the concentration gradient. In a curved gradient, fluxes at any two points are different,
Equation (3.7), and will tend to equalize the concentration. When the diffusion proceeds into
an enclosed space, the curvature of the gradient disappears once the concentration c becomes
equal throughout the bulk. However, if the particle can leave the bulk, a steady state is
eventually obtained.
Let’s conclude by noting that both of Fick’s laws belong to a family of equivalent laws that
include thermal and electrical conduction.21 In the following sections, we shall consider
transport along a field gradient in a crystal, termed lattice (bulk) diffusion.22 Uncharged
species will move down the chemical-potential slope, whereas charged defects will also move
down an electric-potential slope.

3.3.1 Point-Defect Movements

Vacancies and interstitials are the transport vehicles in crystal structures. Atoms at regular
sites jump into neighboring vacancies, whereas interstitial atoms jump between interstitial
sites (Figure 3.7).23 Why do atoms jump?
The free energy of an atom is at a minimum at its regular site. Jumps occur because the
atom’s thermal vibration24 of frequency ν (in s−1) provides a chance of passing the Gibbs-
energy barrier Δ‡Gm for moving to the neighboring vacant site (see Figure 3.9 later on). The
likelihood that a vibration becomes a successful jump follows the Boltzmann probability
distribution,

pB ¼ expðD‡ Gm =kTÞ (3.9)

where kT is the thermal energy. If pB is, say, 0.0002, on average there will be a jump into
a new position every 5000 vibrations. Consider the 2D interstitial in Figure 3.7. It will jump
with frequency pBν, but only one-quarter of these jumps will move it forward by the length

21
The thermal-conduction flux is in J s−1 m−2, the electrical-conduction flux in C s−1 m−2 (where J s−1 = W and C s−1 =
A). The thermal diffusivity in units of J s−1 m−1 K−1 (under gradient of thermal potential in K per length of 1 m) is
normally called thermal conductivity. The electrical diffusivity (under gradient of electric potential in V per length
of 1 m, called electric field) in units C s−1 m−1 V−1 (= S m−1, siemens per meter) is normally called electrical
conductivity. The mass-, thermal- or electrical-conduction flux is generally a vector (in contrast to electric- or
magnetic-induction fluxes treated later in the book, which are scalars, being an integral amount of a vector quantity
over a finite area; a dot product of these two vectors, a scalar).
22
This is in order to distinguish it from diffusion via extended defects and grain boundaries, which is usually faster
than bulk diffusion. However, for high concentrations of point defects, as in many functional materials, bulk
diffusion may dominate.
23
There are also other, more complicated, mechanisms of atom movements.
24
The vibration frequency ν is about 1013 s−1. In the Debye model of atomic oscillators connected by elastic springs of
chemical bonds in a periodic solid, ħν = kΘD, where ħ = h/2π and h is the Planck constant (6.62607015×10−34 J s), k is
the Boltzmann constant (1.380649×10−23 J/K) and ΘD is the Debye temperature of the solid, typically a few
hundreds of kelvins. The kΘD term is the maximum energy (highest frequency) of a sound wave propagating in
the solid. ΘD is proportional to the square root of the “spring constant” (the chemical bond strength).
102 Defect Chemistry and Nonstoichiometry

Figure 3.7 Self-diffusion of an atom via vacancy (left) or interstitial (gray, right).

a along the chosen direction of the diffusion marked with the arrow. The progression rate (in
s−1) of that jump in 2D is rprogression = ¼ pBν.
In general, the progression rate of a jump in the direction of the flux is obtained by
multiplying the jump frequency pBν with the probability pavail that there is a site to jump
to, with the probability pdir that such a site moves the atom forwards by a certain
distance along the direction of the flux, and with a correlation factor fc (described
below) between the atom that has jumped and the space it left behind:

rprogression ¼ fc pdir p avail pB ν (3.10)

All three factors, pavail, pdir, and fc, depend on the crystal structure and the diffusion
mechanism. Let’s consider a body-centered cubic crystal (Nn = 8 nearest neighbors, Figure
1.23) of a metal M with very dilute25 vacancies as the sole defects. Self-diffusion of the atoms
via vacancies will have pavail = 8[vM], which is a probability that a vacancy occurs around
a selected metal atom when the fraction of vacant sites is [vM].26 The atom will then jump into
this vacancy. The probability pdir that this particular vacancy will move the jumping atom
along the positive direction of the unit-cell edge is ½. This is because 4 of the 8 nearest-
neighbor sites advance the atom along this direction (all by a/2). Thus, pavail pdir = 4[vM] in
this example. Now, it is relatively unlikely that there will be another vacancy waiting for our
atom to jump further. Why doesn’t it jump straight back into the vacancy it left behind? The
reason is that it needs to overcome the same energy barrier to jump back. While it waits for
this (for a time dictated by the probability), other atoms may fill that vacancy. The probabil-
ity of the Nn − 1 = 7 neighbors not jumping into the vacancy before our atom is just ½. If they
don’t, our atom jumps back, eliminating its 2 jump chances, and the progression rate of the

25
Vacancies may tend to cluster at temperatures close to the melting point of the solid and speed up the diffusion. In
real materials, line and plane defects, as well as grain boundaries, also contribute to diffusion.
26
Do not confuse the Greek lowercase ν symbol for the frequency with the typographically similar italicized Latin
lowercase v in the symbol vM for the vacancy.
 3.3 Point Defects and Diffusion 103

original jump is affected by a factor of 1−2/7 = 0.714. This rough estimate, fc = 1−2/(Nn−1), of
the correlation factor in Equation (3.10) agrees reasonably with the actual fc = 0.727
obtained by exact and involved derivation for Nn = 8 (see Mehrer in Further Reading).
This approximation works well for Nn > 4.
Instead of the metal atom M, let’s consider the vacancy jump. While the atom M jumping
via the vacancy mechanism has pavail = Nn [vM], the vacancy is jumped into by surrounding
M atoms, the availability of which at a selected neighbor site is pavail = 1−[vM] while pdir and fc
remain the same. Now, what is the jump frequency of this vacancy? If it has Nn neighbors, it
will get jumped into Nn times more often than by just one atom; hence the jump frequency of
the vacancy is NnpBν instead of pBν for the atom M in Equation (3.10). While rprogression(M) =
fc pdirNn[vM]pBν, we find that rprogression(vM) = fc pdirNn(1−[vM])pBν. The ratio of these two
jump progression rates is [vM]/(1−[vM]).
In more general terms, the ratio of the atom- and vacancy-progression rates is a consequence
of the jump balance: as 1 vacancy jumps 1000 times, 1000 atoms jump once. We’ll see soon that
rprogression is directly proportional to the diffusion coefficient D, and we anticipate:

Datom ½vatom 
¼ (3.11)
Dvacancy 1  ½vatom 

For low defect fractions, this simplifies to Datom = Dvacancy [vatom].

3.3.2 Random Hopping

In the previous section, we realized that atoms in solids jump and move even without an
external driving force. In isotropic solids, the movement has equal probability in all direc-
tions. The eventual probability distribution follows from a 2D thought experiment called
a random walk: If you tag an atom on a plane by placing the coordinate cross at it, and let it
“walk” n steps, it will end up some distance from the origin. Do the same with another atom,
and it will end up somewhere else. If you mark the ends of very many such random walks,
there will be no directional preference, only a radial distribution, in 1D.

number of atoms

σ σ

radial displacement 0 radial displacement

Figure 3.8 Radial distribution of atoms after many long random walks from zero.
104 Defect Chemistry and Nonstoichiometry

The radial distribution is binomial, and becomes normally distributed for an infinite
number of steps. It can be shown that the mean square displacement σn2 after n jumps
of equal length λ is equal to the sum of the squares of the individual jump vectors,
σn2 = nλ2. Hence after n steps of random walk, the root mean square displacement
pffiffiffi
from the origin is σ = λ n. The distribution in Figure 3.8 shows that σ is a convenient
measure of the progression of the atoms, only a tiny fraction of which will reach the
maximum distance λn in the given time τ = n/rjump. Problem 3.20 explores these ideas
further.

3.3.3 Hopping Under a Driving Force

At thermal equilibrium, atoms and defects adopt local minima of the Gibbs energy.
A typical energy profile between two interstitial sites is shown in Figure 3.9. In the
absence of any external force, jumps between the sites will occur at random. If,
however, an external field exerting a force F is applied, the jumps will develop
a preferred direction. Such a field may be due to a chemical-potential or an electric-
potential gradient. In the direction of the force, the Gibbs-energy barrier for migration,
Δ‡Gm, will be lowered (Figure 3.9, right) by an energy Em, while it will be increased in
the opposite direction by the same amount. Em is equivalent to the work done by the
force F along the corresponding path λ/2:

Em ≡ ½λF (3.12)

For a jump from site 1 to site 2, the barrier height becomes (Δ‡Gm−Em), and the progression
rate r12 of this jump is r12 = pdir pavail ν exp{(−Δ‡Gm+Em)/kT} in analogy with Equation
(3.10), or r12 = rprogression exp(Em/kT), where rprogression refers to the progression rate under

Em
Gibbs energy

∆‡Gm

λ λ/2

1 2 1 2

Figure 3.9 Gibbs-energy profile between two interstitial sites. Left: No external force. Right: Under an
external force, the forward barrier is lower by Em.
 3.3 Point Defects and Diffusion 105

random movement. When the force F is small, the exponential function can be approximated
by a Taylor series with only the linear term27 retained, exp(Em/kT) ≈ 1+(Em/kT). The frequency
of this jump then becomes r12 ≈ rprogression{1+(Em/kT)}. For jumps in the opposite direction,
the Gibbs-energy barrier is higher by Em and accordingly r21 ≈ rprogression{1−(Em/kT)}.
Combining these terms gives the net rate rnet = r12 − r21 = rprogression(2Em/kT). Substitution for
Em from Equation (3.12) yields rnet = rprogression(λ/kT)F. The net flux J (in s−1 m−2 ) in a given
direction is obtained by multiplying rnet (in s−1) by the distance λ (in m) between the two sites
and by the local volume concentration c of the atoms (in m−3):

λ2 c
J ¼ rprogression F (3.13)
kT

This result provides two important conclusions: (1) The flux depends on the probability of
the random progression in the given direction modified by the effect of the external force. (2)
The flux is linearly proportional to the external force.28 In the following sections, we’ll
consider two such driving forces—the entropy increase upon homogeneous distribution of
defects and the drift of charged defects in an electric field.

3.3.4 Hopping Under a Concentration Gradient

When defects move in a concentration gradient to homogenize their distribution in a sample,
the associated entropy increase decreases the Gibbs energy of the system. The driving force Fi
per atom i migrating in direction y is given by the gradient of the partial molar Gibbs energy
of that atom (gradient of that atom’s chemical potential µi):

dμi
Fi ¼  (3.14)
dy

The negative sign in Equation (3.14) means that the transport proceeds from higher values
of the potential to lower values, i.e. down the (negative) slope of the potential. The chemical
potential per atom i is defined29 as:

μi ¼ μ0i þ kTlnai (3.15)

For a dilute ideal solution of the defect i, the activity ai can be replaced with the volume
concentration ci of the defects (in m−3).30 Considering that dµi/dy = (dµi/dci)(dci/dy), the
force Fi per atom is:

27
f(x) = f(0) + f 0 (0)x. For x = 0.1 in ex, the error is 0.47%, for x = 0.5 the error is 9%.
28
As long as F is sufficiently small so that the linear term of the Taylor-series approximation is sufficient.
29
See any textbook on physical chemistry.
30
If the concentration is expressed in mol/m3, the Boltzmann constant k is replaced with the molar gas constant R =
NAk, where NA is Avogadro’s number (6.02214076×1023 mol−1 exactly).
106 Defect Chemistry and Nonstoichiometry

 
1 dci
Fi ¼  kT  (3.16)
ci dy

Substituting for F in Equation (3.13) gives:

   
λ2 c i 1 dci dci
J ¼ rprogression  kT  ¼ rprogression λ 
2
(3.17)
kT ci dy dy

This reproduces the macroscopic Fick’s first law, Equation (3.7), via microscopic atomic
jumps. We see that the diffusivity Di (in m2 s−1) of the atom (defect) i in an ideal and dilute
solution31 in the solid is

Di ¼ rprogression λ2 (3.18)

where rprogression (in s−1) is the progression rate at a given site, Equation (3.10), under random
movement, and λ is the jump length in the direction of the progression (in m).32 This is an
important result and shows that diffusion under an applied force is controlled by the atom’s
self-diffusion coefficient.

3.3.5 Hopping Under an Electric Field

From electrostatics we know that the force Fi acting on charge qi in an electric field of
intensity E is Fi = qiE.33 When the charged defect i moves down the gradient of the electric
potential V (in volts),34 it experiences the electric-field intensity E = −dV/dy and:
 
dV
Fi ¼ qi  (3.19)
dy

Substituting for F in Equation (3.13) and converting the particle flux Ji into the charge flux ji
by multiplying it with the charge (in C) per particle qi (hence ji = qiJi), we obtain

31
If the solution of the defect in the solid were ideal but not dilute, the backward flux of the solvent (i.e. the solid host
matrix) would have to be included with its own (intrinsic) diffusion coefficient of the participating fluxes. The
combined diffusivity is often called interdiffusivity, denoted as D. e
32
For isotropic cubic structures, the probability pdir that a vacancy jump moves a selected atom forward along the
direction of the flux, when multiplied by the square of the length λ of the projection of the jump onto that direction,
equals ℓ2/6, where ℓ is the actual length of the jump. Furthermore, ℓ2/6 equals a2/Nn, where a is the unit-cell
parameter and Nn is the number of nearest neighbors of the vacancy. These formulae can be conveniently used
instead of pdir λ2 when D is calculated from the combined Equations (3.10) and (3.18).
33
The unit of the electric-field intensity is V/m. One V/m is the intensity of an electric field where a point charge of
1 C (C ≡ A s) experiences a force of 1N (N ≡ J/m ≡ Ws/m ≡ VA s/m). 1 N of force per 1 C of charge then equals 1 V/m
electric field.
34
Note that in this context we use the symbol V for the electric potential. Do not confuse with volume.
 3.3 Point Defects and Diffusion 107

 
dV
ji ¼ σ i  (3.20)
dy

in which σi is the electrical conductivity due to defect i,

λ2 ci q2i
σi ¼ rprogression (3.21)
kT

where rprogression is again the progression rate of Equation (3.10) under random
movement.
Electrical conductivity is an intrinsic material property and will be treated in more detail in
Chapter 10. The conductivity σi represents a fraction ti of the total electrical conductivity σ of
the solid due to defects, and ti is termed the transference number of the point defect i. For flux
of elementary charges under constant electric potential (a steady state), Equation (3.20)
adopts the form j = σE that we’ll recognize in Chapter 10, Equation (10.2), as one possible
expression of Ohm’s law.

3.3.6 Relationship between Conductivity and Diffusivity

Whether the driving force is electrical or chemical, it produces a flux. The flux Ji of the defect
i at a concentration ci under the general driving force Fi, Equation (3.13), can be expressed
using the experimentally available35 random diffusivity Di from Equation (3.18):

Di ci
Ji ¼  Fi (3.22)
kT

It can also be expressed using the experimentally available36 electrical conductivity σi. This
makes use of the Nernst–Einstein relation between σi (in S/m, 1 S ≡ 1 A/V) and Di (in m2/s),37
obtained when Equation (3.21) is expressed with Di of Equation (3.18) for a defect i of charge
qi (in C ≡ A s) and carrier density ci (in38 1/m3):

35
Di is obtained by tracer-diffusion techniques with isotopes. An isotope is traceable either by mass spectrometry after
sectioning the sample, or by measuring radioactivity along the diffusion path if a radioactive isotope is used. As
noted earlier, for vacancy mechanisms the tracer-diffusion coefficient is smaller than a purely random diffusion
model by the correlation factor fc. For interstitial diffusion of dilute defects, fc = 1.
36
σi is obtained by measuring electrical conductivity with electrodes that conduct specifically via the defect i. As an
example, ZrO2 (Chapter 13) blocks electronic carriers while allowing oxide anions through.
37
The Di thus obtained from conductivity (Dconductivity) will not necessarily be the same as the tracer-diffusion
coefficient for the same atom/defect. The ratio Dtracer/Dconductivity is then evaluated as the Haven ratio and reflects
various movement correlations for the charged defects.
38
When the amount of charge-carrying defects is expressed in moles, hence the carrier density is in mol/m3, we replace kT
with RT, and the defect’s electric charge qi = zie with ziF. Here F is the Faraday constant (1 mole of elementary charges:
NAe = 6.02214076×1023 × 1.602176634×10−19 ≈ 96485.3 C) and zi is the defect’s charge number (the integer charge in units
of elementary charge e).
108 Defect Chemistry and Nonstoichiometry

Di σi
¼ 2 (3.23)
kT qi ci

Substituting from here for Di in Equation (3.22) yields the flux in terms of σi:
σi
Ji ¼  Fi (3.24)
q2i

Returning to the Nernst–Einstein relation, we can perform one last step that will lead to the
useful concept of mobility that we’ll encounter in Chapters 10 and 13. Mobility, μi, is the
speed of a species i per unit intensity of the field that moves it. It is obtained by multiplying
Equation (3.23) with the charge qi in coulombs. Both sides of the equation then have units
of m/s per V/m; the left-hand side μi = Diqi /kT expresses the mobility of the charged defect
via its random diffusivity and the right-hand side μi = σi/qici via its electrical conductivity.
Finally, we should note that transport of a single type of charged particle is rare. It is
usually accompanied by transport of a compensating counter charge, forming two fluxes
that together appear as a total flux of neutral atoms. This is explored in the following
section.

3.3.7 Ambipolar Diffusion

Transport of charged defects is not limited to situations where an electric field is applied. In
many cases, transport under a concentration gradient proceeds via charged defects. Clearly,
the transport of a charged defect in the absence of an applied electric field must maintain bulk
neutrality. Consider the oxidation of YBa2Cu3O6 by O2 as discussed in Section 3.2.
Chemically, a copper cation is oxidized while ½O2 is reduced to O2−. As O2− migrates into
the center of the sample, the positive cation charges (holes) follow its motion; we thus have an
influx of both negative (O2−) and positive species (2h•), which together represent the flow of
oxygen as a neutral atom (O). This type of polar transport of nonpolar species is called
ambipolar diffusion.
Equation (3.4) describes oxygen transport in YBa2Cu3O7−δ in terms of defects. At the
surface, O2 oxidizes copper (creates h•) and is reduced to O2− ions that fill vO•• coming from
the interior. The holes migrate inwards. Because these defects are charged, any non-uniform
distribution not only creates a concentration gradient but also a local electric-potential
gradient. A flux under the combined chemical and electric potentials therefore requires
a formal summation of the two forces Fi used separately in Equation (3.14) and Equation
(3.19). The flux of Equation (3.24) then becomes:
 
σi ∂μi ∂V
Ji ¼ 2    qi (3.25)
qi ∂y ∂y
 3.3 Point Defects and Diffusion 109

As stated above, there are two such fluxes, an outward flux of vO•• and a charge-compensat-
ing inward flux of h•. The charge of one mole of oxygen vacancies is +2F, the charge of one
mole of holes is +1F. The participating mole fluxes are then:
   
σv••O ∂μv•• ∂V σh• ∂μh• ∂V
Jv••O ¼ 2  O
 2F and Jh• ¼ 2   F (3.26)
4F ∂y ∂y F ∂y ∂y

Electroneutrality requires the sum of charged fluxes to be zero. The flux of holes must
therefore be twice the flux of oxygen vacancies and run in the opposite direction:

Jh• ¼ 2Jv••O (3.27)

We cannot establish the actual electric potential created inside the solid, so a good way to
proceed is to eliminate the ∂V/∂y terms from Equations (3.26).39 By taking the actual inward
O flux (see above) as the negative of the oxygen vacancy flux, we obtain:
 
1 1 ∂μv•• ∂μh•
JO ¼ Jv••O ¼  O
 2 (3.28)
4F 2 1 1 ∂y ∂y
þ
σv••O σh•

The form of this equation suggests that the flux of O behaves as if driven by its own chemical-
potential gradient—a sum of the gradients of the ionic- and electronic-defect fluxes (of vO••
and h•, respectively):

∂μO ∂μv••O ∂μ •
 ¼ 2 h (3.29)
∂y ∂y ∂y

Note that the combination of the ionic- and electronic-defect conductivities in Equation
(3.28) appears in a mathematical form as if ionic and electronic resistivities40 were summed
and then inverted into the total ambipolar conductivity σO. This can be interpreted as though
the ionic and electronic resistors are connected in series in a circuit.
When the O-flux Equation (3.28) is formally rewritten in terms of ∂μO =∂y and σO,
 
σO ∂μO
JO ¼ 2   (3.30)
4F ∂y

it becomes obvious that a diffusion coefficient for the neutral species O (the chemical
diffusion coefficient DO) can also be defined. At this point, we need to remember that
diffusivities are normally expressed via gradients of concentrations c and not of chemical
potentials (Section 3.3.4). Therefore a conversion ∂μO =∂y ¼ ð∂μO =∂cO Þð∂cO =∂yÞ is performed

39
For example by expressing the coulombic term ∂V/∂y from the equation for holes, substituting it into the equation
for vacancies, substituting for the hole flux from Equation (3.27), and rearranging.
40
Resistivity is the inverse of conductivity.
110 Defect Chemistry and Nonstoichiometry

on Equation (3.30). In the result, we identify DO as the proportionality parameter between

the flux JO and the concentration gradient ∂cO =∂y:
 
σO ∂μO
DO ¼ 2  (3.31)
4F ∂cO

Having defined the chemical diffusion coefficient DO, we can return to the individual point defects
and their concentrations. From Equation (3.29), we see that ∂μO ¼ ∂μv•• þ 2∂μh• , and electro-
O
neutrality dictates that ∂cO ¼ ∂cv••O ¼ ∂ch• =2 (there are two holes per vacancy). This gives:
!
σO ∂μv•• ∂μ •
DO ¼ 2  O
þ4 h (3.32)
4F ∂cv••O ∂ch•

For low concentrations of defects, activity depends on concentration via µ = µ0 + RT lnc;

hence ∂μ=∂c ≈ RT=c, where c is the equilibrium steady-state concentration in mol m−3.41
Converting σO back to the defect conductivities, we obtain:
!
RT 1 1 1
DO ¼ 2   þ (3.33)
F 1 1 4cv••O ch•
þ
σv••O σh•

What remains is to express the separate diffusion coefficients for the ionic- and electronic-
defect components. This is done by substituting for each of the two defect conductivities
from the Nernst–Einstein relation of Equation (3.23), which after rearrangement gives:
!
1 1 1
DO ¼  þ (3.34)
1 1 4cv••O ch•
þ
Dv••O  4cv••O Dh•  ch•

We see that the diffusion coefficient of the chemical element transported depends on the
participating electronic and ionic defects; their individual diffusivities/conductivities, their
charge numbers, and their equilibrium concentrations. The diffusion of the neutral chemical
element is largely controlled by the slower species,42 but the slower flux is somewhat
enhanced by the faster flux. As an example, the enhancement factor for Equation (3.34)
approaches a maximum of DO =Dv••O ¼ 1 þ ð4cv••O =ch• Þ ¼ 3 if Dh• ch• >> Dv••O 4cv••O .

41
For concentrations given in number of particles per m3, R is replaced by k, and F by e.
42
Diffusion of neutral O through a purely oxide-ion conductor, such as Ca- or Y-doped zirconia, is limited by the
material’s low electronic conductivity. It speeds up enormously when a metallic conductor is added to transport the
electrons. That can be a Pt powder mixed with the doped-zirconia powder prior to sintering into the device form, or
an external wire connecting porous Pt electrodes applied onto the two opposite surfaces along the diffusion path. In
contrast, diffusion of O through a good electronic conductor, such as YBa2Cu3O7−δ, is controlled by the oxygen-ion
conductivity of the material.
 3.3 Point Defects and Diffusion 111

3.3.8 Temperature Dependence of Diffusivity

As we have just seen, point defects are the vehicles of mass transport in crystalline solids.
Diffusivity depends on temperature via the energetics of their movement. The energetics may
have two contributions—the cost of making the vehicle and the cost of running it. Both costs
can be expressed in terms of the free-energy change ΔG (the ability of a system to do work),
and the common approximation is to consider the enthalpy and entropy terms in ΔG = ΔH −
TΔS to be independent of temperature.
The making costs are the total Gibbs free energy of formation for the defects relevant in
the particular transport. This is derived in Section 2.3, where Equation (2.3) gives the
equilibrium fractional concentration of the defect as a function of the changes in atomic
vibration entropy (ΔSvib) and enthalpy (ΔHf) upon defect formation. The running costs stem
from Equation (3.9), which represents the increasing Boltzmann probability that at higher
temperatures the atom will be more likely to overcome the Gibbs-energy barrier for migra-
tion: Δ‡Gm = Δ‡Hm − TΔ‡Sm.
Let’s consider a metal atom M diffusing by a vacancy mechanism in its own cubic structure
where it has Nn nearest neighbors. Substituting Δ‡Gm in Equation (3.9) with Δ‡Hm − TΔ‡Sm,
considering that pavail = Nn[vM] in Equation (3.10), expressing [vM] with Equation (2.3),
replacing k with R so that the enthalpy and entropy are per mole, and plugging the result into
Equation (3.18), yields:

DM ¼ λ2 fc pdir Nn ν expðD‡ Sm =R þ DSvib =RÞ  expðD‡ Hm =RT  DHf =RTÞ (3.35)

Here DM has the temperature dependence of the Arrhenius equation, D = D0exp(−EA/RT),

in which the activation energy EA combines the enthalpy contributions, and the pre-
exponential term D0 includes everything else. EA can therefore be obtained from
experimental data by least-squares fitting of the linear slope of lnD versus 1/T.
In some cases, the making costs do not apply and only running costs need be considered.
This is the case for the diffusion of foreign or extrinsic interstitial atoms in low concentra-
tions, for example of solid-solution carbon atoms in iron. Here the vehicle is simply the
extrinsic defect itself, pavail ≈ 1, and the equivalent expression becomes:

Dinterstitial;extrinsic ¼ λ2 fc pdir ν expðD‡ Sm =RÞ  expðD‡ Hm =RTÞ (3.36)

3.3.9 Diffusivity and Redox Defect Equilibria

Let’s now consider diffusion in a metal oxide via a metal- or oxygen-vacancy mechanism.
Here, we’ll have to take into account how the vacancy fraction depends on the equilibria we
explored in Section 3.1. The constituent atom i (metal or oxygen) at its regular lattice site will
not migrate unless it has an adjacent vacancy. The probability of this is pavail = Nn [vi], and we
modify the previous result (take fc = 1 for simplicity) to
112 Defect Chemistry and Nonstoichiometry

Di ¼ Nn ½vi λ2 pdir ν expðD‡ Sm =RÞ  expðD‡ Hm =RTÞ (3.37)

The vehicle production costs not-yet-included concern the vacancy fraction [vi] and follow
from energetics of the point-defect equilibria the vacancy is involved in.
Let’s consider three cases where this [vi] = f(T) function is straightforward. The first is
when the high-temperature vacancy fraction [vi] in the material becomes “frozen” at lower
temperatures when kinetic factors prevent further redox exchange with the reaction atmos-
phere to establish a new equilibrium. The second is when [vi] is independent of temperature
because it is fixed by aliovalent dopants around the point of integer valence (oxygen
vacancies in Figure 3.4 and nickel vacancies in Figure 3.5). The third case is when the
vacancy is the dominant defect and [vi] can be expressed via the temperature dependence
of the equilibrium constant K for the defect-formation reaction in terms of the reaction-
enthalpy and -entropy change as K = exp[(ΔS/R)−ΔH/RT].43 As an example, when NiO is
oxidized, holes are compensated by nickel vacancies (Figure 3.2). The electroneutrality
condition simplifies to 2[vNi 0 0 ] = [h•], and from Table 3.4 we see that [vNi 0 0 ] = (pO2Kox/16)1/6.
Substituting into Equation (3.37) and rearrangement gives the random diffusion coefficient
of Ni in oxidized nonstoichiometric NiO via a vacancy mechanism as
! !
p 16 DSox þ 6D‡ Sm DHox þ 6D‡ Hm
O2
DNi ¼ Nn λ pdir ν exp
2
 exp  (3.38)
16 6R 6RT

where both the Arrhenius activation enthalpy and the pre-exponential factor have two
components—one from the defect formation and one from the defect mobility. As stated
earlier, ν is the (essentially constant) vibration frequency of the atom and pdir is the probabil-
ity that the jump occurs in the flux direction, advancing the atom by the distance λ. See
Footnote 32 in this chapter for alternative expressions of λ2 pdir.

3.3.10 Outline of Non-Steady-State Diffusion

Fick’s second law, Equation (3.8), expanded into three dimensions, has analytical solutions
for time-dependent concentrations of diffusing species across simple shapes such as a plate of
infinite thickness (termed a half-space; diffusion enters from one side only), a plate of finite
thickness (diffusion enters from both sides), a parallelepiped, a cylinder, or a sphere, all
provided D is constant. The principal solutions are in the literature on either particle
diffusion [like Crank or Mehrer in Further Reading] or heat conduction. When D depends
on the concentration c, this must be properly included in the treatment.
As an example, we’ll calculate oxygen-content profiles in a half-space of YBa2Cu3O6
exposed to O2 at 350 °C (Figure 3.10). In this arrangement, the oxygen flux has one direction.
Assuming that the surface reduction of O2 is instantaneous and D is independent of

43
Since ΔG = −RT lnK.
 3.3 Point Defects and Diffusion 113

y
Figure 3.10 Set-up
sketch of oxygen dif-
fusion into half-space
O2(g) of YBa2Cu3O6.

concentration, Equation (3.8) is simplified, and the solution starts by introducing the
dimensionless (half) length,

y=2
u ¼ pffiffiffiffiffiffi ; (3.39)
Dτ

as a new variable.44 This replaces the two variables (τ and y) in Fick’s second law
 2   2   
∂c ∂c ∂c ∂c ∂u ∂c ∂2 c ∂u 2
¼D with one. Given the new variable u, ¼  and ¼  .
∂τ ∂y2 ∂τ ∂u ∂τ ∂y2 ∂u2 ∂y
∂u u ∂u 1
In the latter two expressions, ¼  and ¼ pffiffiffiffiffiffiffiffi are obtained by differentiating u.
∂τ 2τ ∂y 4Dτ
2
dc dc
This yields the differential equation 2 þ 2u ¼ 0. Substituting z ¼ dc=du simplifies it to
du du
dz 1
þ 2uz ¼ 0. Separation of variables gives dz ¼ 2udu that is integrated to
du z
lnz ¼ u þ const, from which z ¼ C  expðu Þ. Putting back z ¼ dc=du yields
2 2
pffiffiffi
π
dc ¼ C  expðu2 Þdu. The integral of expðu2 Þdu equals  erfðuÞ, expressed with the
2
Gaussian error function (erf).45 Given the boundary limits of erf(0) at the concentration
pffiffiffi
c = 1 at u = 0 at any time and erf(∞) = 1 when c = 0 at u = ∞ at any time, C ¼ 2= π.
Integration of the left-hand side of the differential equation gives c = −erf(u) + K. The initial
condition of c = 1 at u = 0 yields K = 1 and the relative concentration change cr:

cr ¼ 1  erfðuÞ (3.40)

In Figure 3.11 (left), our solution of Equation (3.40) is illustrated for DO = 7.144×10−13 m2/s
pffiffiffiffiffiffi as erf(u) ≈ u for low u. Even
at 350 °C. The concentration profile is linear close to the interface
at u = 0.5, erf(u) is 0.5205, which happens at the distance Dτ; see Equation (3.39). This
distance is called the penetration depth (Figure 3.11). It is a characteristic value indicating the
extent of penetration of the diffusant. For diffusion solutions based on the erf function, it is
close to the diffusion half length; a length where a 50% concentration change occurs in

44
This is possible only if both initial and boundary conditions are functions of u only.
45
The erf(u) is the integral of the normal probability distribution from its center at 0 to u. That distribution is the
limiting large-numbers envelope for the discrete binomial distribution, Figure 3.8.
114 Defect Chemistry and Nonstoichiometry

Relative concentration profile 600 500 400 300 t (°C)

cr or x in YBa2Cu3O6+x
1.0
-8

log D (m2/s)
0.8
0.6 -10

0.4 -12
10 40 90 160 250 360
0.2 -14
0.0
0.0 0.5 1.0 1.5 1.0 1.2 1.4 1.6 1.8 2.0
distance (mm) 1000/T (K−1)

Figure 3.11 Left: Profiles of oxygen-content increase inpffiffiffiffiffi

a thick
ffi layer of YBa2Cu3O6 exposed to O2 at
350 °C, after 10, 40, 90, etc. hours. Penetration depths Dτ are marked by dots. Right: Arrhenius-type
temperature dependence of the diffusion coefficient over temperature range 350 to 600 °C.

a given time. Note also that the penetration depth doubles in 22 times that time, triples in 32
times that time, etc.
Equation (3.40) is derived for a plate of infinite thickness. It can also be used for the initial
stages of diffusion into a finite or thin plate, but only until the concentration starts to change
significantly at half the plate thickness. The D values for temperatures of interest are easily
evaluated from published Arrhenius-type temperature dependencies. The D dependence for
our example is plotted in Figure 3.11 (right) and has an activation energy of 129.2 kJ/mol and
D0 of 4.808×10−2 m2/s.

3.3.11 Cautionary Note on Diffusion in Real Materials

Caution needs to be applied when investigating diffusion in real oxides and other
binary compounds with high ionicity. In addition to possible computational difficulties,
problems may be encountered either due to the diffusion model or due to the sample.
The model-related problems concern the assumption of single point defects being the
transport vehicles. In many oxides, these are in reality aggregated defect clusters, the
simplest of which are dimers of interstitials or vacancies. Other model-related problems
include disregarding ambipolar diffusion, and, in variable-temperature studies, the fact
that D may depend on some defect concentration that is chemically variable due to
reaction with the measurement atmosphere. There are also the practical difficulties of
arranging the experiment in a manner that correctly approximates the initial and
boundary conditions of the model. Sample perfection and morphology is also of
concern—in single-crystal measurements, line defects can lead to short-circuit diffusion;
in polycrystalline materials, a similar rapid transport can occur at grain boundaries.
Last, but not least, omnipresent aliovalent impurities will increase diffusivity by creat-
ing point defects as vehicles for the mass transport.
 3.4 Problems 115

3.4 Problems
3.1 Write Kröger–Vink symbols for the following fully charged point defects in NiO: metal
vacancy, oxygen vacancy, lithium acceptor, aluminum donor.
3.2 Name the following defects in TiO2 and write their Kröger–Vink symbols: Al and Nb
dopants at the Ti site; F and N dopants at the O site; oxygen and titanium atoms out of
their regular sites.
3.3 State the type of intrinsic defects would you predict in fluorite-type CeO2 and UO2.
3.4 Recast Table 3.4 into one valid for a hypothetical MO oxide with cation-Frenkel
intrinsic defects.
3.5 Use chemistry or physics to suggest whether ionic or electronic defects will dominate in
pure ZrO2 at high temperatures.
3.6 Consider pure PbO with anion-Frenkel compensation and dominant ionic defects.
What are the slopes of the essentially linear dependences in the three limiting regions of
the plot of defect fractions versus pO2? Sketch the Brouwer diagram.
3.7 In fact, pure PbO may show an anti-Schottky disorder. What differences does this
bring to the results of Problem 3.6?
3.8 Consider an idealized pure CuO with anion-Frenkel compensation and dominant
electronic defects. What are the slopes of the essentially linear functions in the
three limiting regions of the plot of defect fractions versus pO2? Sketch the
Brouwer diagram.
3.9 State the limiting slope of the metal-vacancy fraction versus pO2 in the oxidative-
nonstoichiometry range for (a) Cu2O and (b) Cr2O3, both of Schottky intrinsic
defects.
3.10 Consider a stoichiometric metal oxide of dominant Schottky defects in equilibrium.
Which of these two defects will increase its fraction upon either an acceptor or donor
doping under the same conditions?
3.11 Consider a stoichiometric metal oxide in equilibrium with its dominant electronic
defects. Which of these two defects will increase its fraction upon an acceptor or
donor doping under the same conditions?
3.12 Sketch the Brouwer diagram for Li-doped NiO with dominant electronic defects.
3.13 Sketch the Brouwer diagram for Al-doped NiO with dominant ionic defects.
3.14 Write down the dissolution reaction of ZrO2 in Cr2O3 and state how the solid solubility
will depend on pO2: (a) about the point of integer structure, (b) about the point of
integer valence.
3.15 The equilibrium constant in Equation (3.5) for oxidation of the oxygen vacancy in
YBa2Cu3O7−δ at 500 °C is Kvox = 55 bar−½. (a) State the equilibrium composition of
YBa2Cu3O7−δ at 500 °C in a flow of Ar gas containing 100 ppm O2 and in air, both at
1 bar. (b) State the pO2 at which you would anneal YBa2Cu3O7−δ in order to obtain an
average oxidation state of CuII.
116 Defect Chemistry and Nonstoichiometry

3.16 Derive the simplified expression for gross oxygen nonstoichiometry in NdBaFe2O5+δ
as a function of pO2 assuming that Fe3+ defines the point of integer valence.
3.17 Water flows from a 0.2 cm2 hose at 6 kg per minute. State the mass, volume, mole, and
molecule flux of H2O in SI units.
3.18 Calculate the mass flux of carbon under steady-state diffusion through a steel plate
separating carbon-rich and carbon-poor gases at 700 °C if subsequent analysis gives
carbon concentrations of 1 kg/m3 and 0.5 kg/m3 at the respective depths of 0.5 mm and
1 mm below the surface. Assume 10−11 m2/s as the diffusion coefficient of C in the iron
matrix at this temperature.
3.19 An oxygen-permeable YBa2Cu3O7−δ membrane 1 cm thick operates at 700 °C between
pressurized air at 11 bar and pure O2 at 1 bar. The steady-state O2 gas production is 0.84 mL
at atmospheric pressure and 20 °C (molar volume 24 L/mol) per square meter every second.
Calculate the O2 flux, O atom flux, and, finally, DO with Kvox = 4.3 bar−½ (to obtain the
O gradient across the membrane) and with molar YBa2Cu3O7−δ volume of 10−4 m3/mol.
3.20 Consider a random 1D walk of n equal steps of length λ originating at zero. (a) What is
the probability of deviating by 4λ from the origin in 4 steps? (b) Construct a table of
probabilities of reaching points −nλ . . . + nλ for up to n = 4. (c) Calculate the variance
(i.e. the mean squared deviation from the mean) and its root after 4 steps of unit length.
(d) Derive a formula for the variance σn2 after n steps of length λ and for its root σn.
3.21 Calculate the jump frequency of interstitial carbon in bcc iron at 800 °C, assuming
a vibration frequency ν = 1013 s−1 and an activation energy for hopping Δ‡Gm =
62 kJ/mol. On average, every x-th carbon vibration overcomes the jump barrier;
determine x.
3.22 Assume that bcc iron at 1800 K has a fraction of vacant Fe sites of 0.0001, an Fe atom
vibration frequency ν = 1013 s−1 and an activation energy for hopping Δ‡Gm = 29 kJ/
mol. What is the jump frequency of Fe atoms? On average, how many vibrations are
there between jumps? What is the Fe progression rate at each site?
3.23 Calculate the self-diffusion coefficient of Fe in Problem 3.22, given the unit-cell
parameter a = 2.87 Å.
3.24 Given the diffusivity DC = 10−10 m2/s for interstitial carbon in bcc iron at 800 °C of a =
2.87 Å, estimate the activation energy Δ‡Gm for hopping of the C atoms (ν = 1013 s−1)
via interstitial sites.
3.25 The self-diffusion coefficient in Al (fcc; a = 4.05 Å) at 600 K is DAl = 2×10−16 m2/s.
Assuming an atomic vibration frequency ν = 4×1013 s−1 and an activation energy for
hopping Δ‡Gm = 58 kJ/mol, calculate the site fraction of Al vacancies.
3.26 Verify the statement in this chapter’s Footnote 32 that pdirλ2 = ℓ2/6 = a2/Nn, in which the
distance a is the unit-cell edge, ℓ the actual jump length, λ its projection onto the unit-
cell edge direction, and Nn is the number of nearest neighbors of the vacancy, for (a)
primitive cubic, (b) bcc, and (c) fcc packing of spheres.
 3.4 Problems 117

3.27 Calculate the self-diffusion coefficient in the primitive cubic α-Po (Figure 1.23; a =
3.36 Å) at 500 K via vacancy mechanism, assuming 0.001 vacant sites, ν = 1013 s−1,
Δ‡Gm = 36 kJ/mol. Estimate the diffusivity of the vacancy.
3.28 Calculate the self-diffusion coefficient of α-Po along the direction of the cell edge
(Figure 1.23; a = 3.36 Å) at 500 K via an interstitial mechanism with full availability
of interstitial sites, fc = 1, ν = 1013 s−1, and Δ‡Gm = 40 kJ/mol. Verify the result by
orienting the cube to yield three equivalent jumps for positive progression direction and
three for negative progression direction.
3.29 A single crystal of fcc iron (a = 3.6468 Å) having a few ppm of interstitial carbon
is kept at 1000 °C and isolated from its surroundings. Assume DC =
2.217×10−11 m2/s for the self-diffusion coefficient of interstitial carbon along
a via octahedral holes, site availability pavail = 1, correlation factor fc = 1 for
the interstitial, and atomic vibration frequency ν = 1013 s−1. (a) State the prob-
ability of a vibration of C becoming a jump. (b) State the total length of the path
a carbon atom travels in one minute. (c) State that atom’s root mean square
displacement from its position a minute earlier. (d) Calculate the hopping activa-
tion Gibbs energy per mole of interstitial carbon defects.
3.30 Confirm by dimensional analysis in SI that multiplying Equation (3.23) with the charge
qi yields mobility.
3.31 A rectangle of calcium-stabilized zirconia Zr0.85Ca0.15O1.85 (a = 5.13 Å) is covered with
porous platinum electrodes on its opposite faces, heated to 1100 °C in air, and an
electrical conductivity of 6 S/m is measured. Assuming t(vO••) = 1, calculate the
diffusivity D(vO••) and mobility μ(vO••) of the oxygen vacancies in the bulk of the
sample, neglecting kinetics of surface recombination.
3.32 Assuming that oxide-ion conductivity of YBa2Cu3O7−δ at 500 °C in air is σ(vO••) = 10−2 S/m
and the ionic transference number t(vO••) = 10−6, what is the total electrical conductivity?
3.33 Calculate the enhancement factor DO/Dh• for Equation (3.34).
3.34 Use Equation (3.34) to generalize the value of the limiting ambipolar enhancement for
DA of a neutral atom A, the flux of which consists of a flux of a z-charged ionic defect
and a much faster flux of singly charged electronic defects.
3.35 Figure 3.2 suggests that oxidative nonstoichiometry of NiO at high temperature is
achieved via formation of nickel vacancies and holes. The following data have been
measured [5, 6] at 1100 °C in O2: coefficient of (tracer) random diffusion for nickel
atoms DNi = 10−15 m2/s, total electrical conductivity σ = 65.5 S/m, site fraction 0.0001
of nickel vacancies. Assuming that Ni diffuses with a vacancy mechanism via nickel
sites (never at O site), and given a = 4.20 Å for the NaCl-type cell, calculate for each of
the two majority defects: (a) Concentration per m3, (b) diffusivity, (c) ionic electrical
conductivity and ionic transference coefficient in order to determine whether NiO is an
ionic or electronic conductor, (d) mobility.
3.36 Set up the equation for the temperature dependence of DCr in oxidized Cr2O3 where
chromium vacancies dominate.
118 Defect Chemistry and Nonstoichiometry

pffiffiffiffiffiffi
3.37 Check that u = (y/2)/ Dτ of Equation (3.39) is dimensionless.
3.38 A steel blade has been nitridized by exposure to flowing NH3 at 700 pffiffiffiffiffiffi°C for
1 hour. The hardened layer, estimated from the penetration depth Dτ of the
nitrogen atoms, is 20 μm. In what time would the penetration depth reach 0.1 mm?
3.39 Nitridized steel is hard yet brittle, and this means that the bulk of a steel object must
remain free of nitrogen. Assuming D = 10−9 cm2/s for interstitial diffusion of
nitrogen in steel at 700 °C, estimate the minimum thickness of the above steel
blade such that the nitrogen concentration at its center would increase by no more
than 0.005 of the surface change under 1 (alternatively 25) hour(s) of nitridization.
Assume for simplicity that the concentration at the center is twice the concentration
that would be caused by nitridization from one side only (in reality it will be less than
twice due to decreasing gradient after the two fluxes penetrate each other).
3.40 At 700 °C, an YBa2Cu3O7−δ sphere of 2 cm radius is abruptly exposed to 1 bar O2 gas.
Assuming DO = 10−5 cm2/s and diffusion as the rate-controlling process, calculate how
long it will take before 90% of the total oxidation change occurs 1 mm below the
surface. Under these conditions, the sphere center will oxidize by less than 1%, so that
the sphere can be approximated as an infinite half-space.

3.5 Further Reading

P. Kofstad, “Non-Stoichiometry, Diffusion and Electrical Conductivity in Binary Metal Oxides” (1972)
Wiley.
J. Crank, “The Mathematics of Diffusion” 2nd edition (1975) Clarendon Press.
S. Mrowec (translated to English by S. Marcinkiewicz), “Defects and Diffusion in Solids” (1980)
Elsevier.
D.M. Smyth, “The Defect Chemistry of Metal Oxides” (2000) Oxford University Press.
D.S. Wilkinson, “Mass Transport in Solids and Fluids” (2000) Cambridge University Press.
J. Maier, “Physical Chemistry of Ionic Materials” (2004) Wiley.
H. Mehrer, “Diffusion in Solids” (2007) Springer.
J.-M. Missiaen, “Solid-state spreading and sintering of multiphase materials” Mater. Sci. Eng. A 475
(2008), 2–11.
P. Heitjans, J. Kärger (editors), “Diffusion in Condensed Matter, Methods, Materials, Models” 3rd
edition (2011) Springer.

3.6 References
[1] G. Brouwer, “A general asymptotic solution of reaction equations common in solid-state
chemistry” Philips Res. Rep. 9 (1954), 366–376.
 3.6 References 119

[2] J.C. Schön, I.V. Pentin, M. Jansen, “Ab initio computation of low-temperature phase diagrams
exhibiting miscibility gaps” Phys. Chem. Chem. Phys. 8 (2006), 1778–1784.
[3] P. Karen, “Nonstoichiometry in oxides and its control” J. Solid State Chem. 179 (2006), 3167–3183
(and references therein).
[4] P. Schleger, W.N. Hardy, B.X. Yang, “Thermodynamics of oxygen in YBa2Cu3Ox between 450 °C
and 650 °C” Physica C 176 (1991), 261–273.
[5] W.C. Tripp, N.M. Tallan, “Gravimetric determination of defect concentrations in NiO” J. Am.
Ceram. Soc. 53 (1970), 531–533.
[6] M.L. Volpe, J. Reddy, “Cation self-diffusion and semiconductivity in NiO” J. Chem. Phys. 53
(1970), 1117–1125.
4 Phase Diagrams and Phase Transitions

In this chapter we will consider two separate but related topics; phase diagrams and phase
transitions. Phase diagrams are used throughout materials chemistry to guide the synthesis
of known and new materials. Electronic and magnetic phase diagrams are used to help
summarize the properties of functional materials under different conditions. We’ll see
throughout the later chapters of this book that many functional materials undergo structural
phase transitions that are intimately linked to their properties. The second half of this
chapter will introduce the fundamental concepts needed to understand these.

4.1 Phase Diagrams

Phase diagrams provide a graphical summary of the behavior of chemical systems at
equilibrium as a function of external variables. In materials chemistry, these variables are
usually composition, temperature, and pressure, but could also include effects such as
electric or magnetic field. These diagrams are based on the phase rule first proposed by
Gibbs, which states that for a system in equilibrium:

PþF ¼Cþ2 (4.1)

where P is the number of phases, F the degrees of freedom or variance, and C the number of
components.1 We can define the system as being the part of the universe that we’re interested

1
The derivation of this rule is relatively straightforward and is covered in many physical chemistry texts. Briefly, to
describe the state of a system consisting of P phases, you need temperature, pressure (two variables), and
(P·C) − P mole fractions for P phases of C components each (−P appears because the sum of mole fractions of
each phase is one, hence one fraction per phase is redundant). There are then a total of 2 + (P·C) − P variables. For
phases at equilibrium, chemical potentials for each component must be equal. This gives C(P − 1) independent
equations (conditions). The number of variables that still remain free to vary is F = 2 + (P·C) − P − C(P − 1), which
can be rearranged to the equation given.

120
 4.1 Phase Diagrams 121

critical
point
220.5
d
Solid Liquid

pressu re (ba r)
e b a normal
c boiling
point
1.01
normal
melting
point
-3 Vapor
6x10 triple point

0 0.0098 100 374.4

temperature (ºC)

Figure 4.1 Schematic phase diagram of water.

in, anything else is the surroundings. It is easiest to understand the various terms in the phase
rule through examples.
Gibbs defined a phase as a state of matter that is uniform throughout in terms of both its
chemical composition and its physical state. It thus represents a homogeneous part of the
chemical system that is bounded by a surface such that it can, at least in principle, be
separated from other parts of the system. To take the simple example of H2O (Figure 4.1)
ice, water, and steam are individual phases—each is homogeneous and each could poten-
tially be separated from the other; crystals of ice dispersed in liquid water is therefore a two-
phase system.2 If we consider a mixture of two metals A and B where A forms a dispersion of
droplets within B, it is a two-phase system. If A and B form a solid solution in which atoms are
homogeneously mixed on an atomic length scale, it is a one-phase system. For a material
such as SiO2, which displays polymorphism and has more than one structural form in the
solid state (e.g. the quartz, tridymite and cristobalite modifications), each form is a different
phase.
We can define the components of a system as the minimum number of independently
variable chemical constituents that we need to define the overall composition. The
H2O phase system, for example, contains a single component because H2O has a fixed
chemical formula. Our two metals A and B would be a two-component system. Al2O3 and
SiO2 can also be considered a two-component system if we take Al2O3 and SiO2 as having
fixed composition (no nonstoichiometry).3 This particular two-component system contains
three phases: Al2O3, SiO2, and the compound Al6Si2O13 (often expressed as 3Al2O3·2SiO2 on
phase diagrams).

2
Throughout this chapter, for reasons of pedagogical clarity, single-phase areas in phase diagrams are shaded in gray.
3
With nonstoichiometry considered, Al2O3 and SiO2 would become part of a three-component system; Si, Al, O.
122 Phase Diagrams and Phase Transitions

The number of degrees of freedom can be formally defined as the number of variables
(pressure, temperature, concentration of components) that need to be specified to fully define
the condition of the system. This can be understood with reference to the phase diagram of
water (a one-component system) in Figure 4.1. If we consider a general point on the phase
diagram, we have a single phase (P = 1) present—liquid water for the point labeled a, ice at
point e. For a one-component system, C = 1. The phase rule tells us that the number of
degrees of freedom, F, is C − P + 2 = 2. At a point like a, we can thus vary both the
temperature and the pressure of the system independently of each other, meaning both have
to be specified to define the system. We therefore see that on a single-component phase
diagram like this, a single phase exists over an area.4 If we cool the system until we reach
point b, which is the freezing point of water at the pressure under consideration, we reach the
phase boundary between water and ice represented by the almost vertical solid line. At this
point, two phases (liquid water and ice) are in equilibrium and will remain so indefinitely.
The phase rule says that F = 1 + 2 − 2 = 1, and we therefore need to specify only one variable
to define the system. We can change pressure and temperature and retain two phases but we
can’t change them independently without destroying one phase or the other. If we increase T,
we must decrease p to compensate (i.e. move to point c); if we decrease T, we must increase
p (i.e. move to point d). We can see from the phase rule why, when water freezes to ice at
constant pressure, the temperature of the system remains constant as long as both ice and
water are present and in equilibrium.5
At the so-called triple point, one has three phases (ice, liquid water, and water vapor) in
equilibrium, and F = 1−3 + 2 = 0. If one changes either p or T, one phase will disappear, so no
variables have to be specified to define the system. This is an example of an invariant point;
one with no degrees of freedom. A line on the H2O phase diagram thus represents the
presence of two phases, and a point where three lines intersect represents three phases.
The final point to mention on the water phase diagram is the critical point, which occurs at
the critical temperature and critical pressure. Beyond this point, liquid and vapor have the
same density and can’t be distinguished—a supercritical fluid is the only phase present.
A practical consequence is that above the critical temperature, gas can’t be liquefied by
application of pressure alone.
As the focus of this text is on solid state materials, the majority of the chemical systems
that we’ll consider will be condensed systems, i.e. ones in which the vapor pressures of the
substances involved are negligible compared to atmospheric pressure. The pressure can
therefore be considered as constant, and one degree of freedom is removed from the system.
The phase rule for a condensed system becomes:

4
Note that this is not the case in a phase diagram with two or more components. In a two-component diagram, a single
solid phase is represented by a line and in a three-component diagram by a point.
5
In practice, remaining at equilibrium means using a slow cooling rate. With more rapid cooling, one may not allow
time for crystallites of ice to nucleate and thus obtains a supercooled state. This, however, is not an equilibrium state
of the system.
 4.2 Two-Component Phase Diagrams 123

PþF ¼Cþ1 (4.2)

This is the form of the phase rule that will apply for the rest of this chapter.

4.2 Two-Component Phase Diagrams

4.2.1 Without Compound Formation
Figure 4.2 shows one of the simplest phase diagrams for a two-component (A and B) or
binary condensed system, a system in which A and B form no compounds AmBn.
Components A and B form no solid solutions (x is either zero or one in A1−xBx) but are
completely miscible when molten. The y axis represents the temperature of the system and
the x axis the composition. Composition throughout this chapter is expressed in mole
fraction (0 to 1), though mole percents are also in common use.6 The left-hand y axis of
Figure 4.2 then corresponds to pure A, and we can read from this axis that A melts at T4; the
right-hand axis represents pure B, which melts at T5.
There are four main regions on this phase diagram. At low temperature, as A and B form
no compounds and no solid solution, one has a two-phase region consisting of a mixture of
pure A and pure B solids. When the solid two-phase mixture is heated to temperatures just
above T1, one obtains a mixture of either solid A and liquid, or solid B and liquid, depending

g
b T5
T4 a
Liquid q p m T3
temperature

liquidus
k l
j T2
A + Liquid B + Liquid
c d T
1
e solidus
f
A+B

0.0 0.2 0.4 0.6 0.8 1.0 A F E K J B

A mole fraction B, xB B
1.0 0.8 0.6 0.4 0.2 0.0
mole fraction A, xA
(a) (b)

Figure 4.2 (a) The phase diagram for a binary system of A and B that form no compounds and no solid
solution. (b) The same phase diagram but with labels for specific points discussed in the text. Single-phase
areas are shaded, two-phase areas unshaded.

6
Mole percents are just mole fractions ×100. In some fields (particularly phase diagrams used industrially), the
composition axis may be expressed in weight percent.
124 Phase Diagrams and Phase Transitions

on whether the composition lies to the left or right of composition E. At composition E, the
solid mixture melts directly to form a liquid at point e, called the eutectic point, which
represents the lowest melting temperature in the system. Line cd (Figure 4.2b) where the
first liquid appears is called the solidus. At temperatures above those defined by the line aeb
(called the liquidus), one enters a region where the system is fully molten into a single-phase
liquid.
When interpreting such phase diagrams, it is crucial to realize the difference between
a two-phase and a one-phase region. For a one-phase region (shaded), each point on the
diagram represents a certain state of the phase in terms of composition and temperature. For
example, one can form a liquid with any desired composition at any temperature above the
liquidus. For two-phase regions the situation is very different. An arbitrary point in a two-
phase region does not represent an actual state of a phase, but corresponds to the overall
composition of two phases in equilibrium. At point f in Figure 4.2b (whose composition is
marked as F on the composition axis), one has a two-phase mixture of pure A and pure
B. The relative amount of each phase present can be read directly from the composition axis.
Equivalently (and to prepare ourselves for later diagrams), we can see that the mole fraction
of A present, xA, is given by (distance F to B)/(distance A to B) (i.e. FB/AB) and xB as AF/
AB. Similarly, point j on the diagram contains a mixture of pure B and a liquid. The
composition of the liquid present is given by point k on the diagram (i.e. xA ≈ xB ≈ 0.5).
The solid present is pure B and represented by point l. The relative amounts of liquid of
composition K and pure B present can be determined straightforwardly by the lever rule.
Let’s label the fraction of pure liquid present as xL and the fraction of pure B as xB. The lever
rule7 states that:

xL  kj ¼ xB  jl (4.3)

i.e. the relative amount of B and liquid is given by:

xB =xL ¼ kj=jl (4.4)

Since we know xL + xB = 1, we can eliminate xL from Equation (4.3) to give the mole fraction
of B as:

xB ¼ kj=ðjl þ jkÞ ¼ kj=kl (4.5)

Note again that the amount of B present is proportional to the distance (kj) further from pure B.
Let’s consider what happens when you cool a liquid of a given composition. Firstly, let’s
consider a liquid of composition E (xA ≈ 0.6, xB ≈ 0.4) on our diagram. This liquid has the
special property of transforming directly to two solids on cooling at the eutectic point e. No
region of mixed solid and liquid phases exists. Since three phases (A, B, and liquid) are

7
This can be likened to balancing two masses at different distances from a pivot. At balance, m1d1 = m2d2 where m1
and m2 are the two masses and d1 and d2 the distances to the pivot.
 4.2 Two-Component Phase Diagrams 125

present at e, the number of degrees of freedom is zero (F = C + 1 − P = 2 + 1 − 3 = 0), and this

tells us that the eutectic point e is invariant.8
The behavior of other compositions is more complex. Consider the composition J (xA ≈
0.35, xB ≈ 0.65) initially at point g. When the liquid is cooled to T3, one meets the liquidus
curve at point m where an infinitesimally small amount of pure B will form. Solid B will be in
equilibrium with liquid of composition J at this point. On further cooling to T2 we reach
point j. One still has a liquid in equilibrium with pure B, but the composition of the liquid has
changed to K—it has become richer in A. The relative amounts of solid and liquid present are
given by the lever rule as kj/jl as discussed above. Note that although the liquid and solid have
different compositions, the overall composition of the system does not change. As one
continues to cool, the liquid composition follows the line ke until at temperature T1 the
final liquid of composition E crystallizes. Below T1 one forms a solid mixture of A and B of
the original overall composition. The system will follow this pathway if cooled under
equilibrium conditions. If the system is rapidly cooled, it may follow a different pathway,
but the final phases present would be the same.
We can also use Figure 4.2 to understand the effect that solid impurities have on
melting points. A sample of pure B at temperature T3 will be a solid (B doesn’t melt
until the higher temperature T5). However, if a small amount of A is added, the solid will
be in equilibrium with a small amount of liquid at point m (composition J, xA ≈ 0.35, xB ≈
0.65); i.e. a small portion of B will have dissolved in the liquid. If sufficient A is added so
that the overall composition becomes J, then all of B will dissolve. The impurity A has
lowered the melting point from T5 to T3. Since at this point B has just dissolved, we have
a saturated solution of B in A. As we continue to add A at this temperature, we retain
a single-phase liquid until we reach point p on the liquidus. At this point, a small amount
of solid A will form. As more A is added, the amount of solid will increase and liquid
decrease until at point q we have pure solid A present. The point p represents a saturated
solution of A in B. Since ae and eb represent saturated solutions of A and B, respectively,
point e represents a liquid saturated in both solids. The use of one solid to lower the
melting point of another is the reason why salt is spread on roads to prevent ice forma-
tion. It’s also of great practical use in the growth of crystals from a flux at lower
temperatures than would otherwise be required. Fluxes are chosen to be easily separable
from the crystals of interest, for example by dissolution in water in the case of NaCl/KCl
fluxes.

4.2.2 With Compound Formation

Most of the systems we’ll meet in materials chemistry will be more complex than Figure 4.2—
components that undergo reactions in the solid state leading to new materials are far more
interesting and exploitable than those that don’t! Figure 4.3 shows three possible phase

8
Terms univariant and bivariant are used for F = 1 and F = 2.
126 Phase Diagrams and Phase Transitions

diagrams in which A and B can react to form an intermediate phase AB2. This phase occurs
as a vertical solid line on the phase diagram at xA = ⅓, xB = ⅔. Note that for composition
calculations it is often more convenient to express the formula of AB2 as A1/3B2/3 (a
normalized or pseudoatom formula).
Figure 4.3a shows the most straightforward system. Each of the solid phases A, AB2, and
B melts to form a liquid without decomposition (at temperatures T2, T1, and T3, respectively,
on the diagram); they are said to have a congruent melting point. The phase diagram is then
very similar to that of the simple eutectic system in Figure 4.2. In fact, it can be considered as
two eutectic systems placed side by side. Figure 4.3b shows a situation in which the
compound AB2 is only stable up to a temperature T1 above which it decomposes without
melting to solid A and B. Above this temperature, the diagram is again essentially identical to
Figure 4.2. Other systems show the reverse behavior, i.e. compounds are only stable above
a certain temperature. ZrW2O8 of Figure 4.8 is one such example (see later).
Figure 4.3c represents a more complex situation in which AB2 has an incongruent melting
point, that is, it melts to give a solid and a liquid of different chemical compositions. On
heating AB2 to T2, it decomposes to give a liquid of composition corresponding to point
p and pure B; the relative amounts are given by the lever rule. On further heating, B will
gradually dissolve in the melt and the liquid will move in composition from p to r. At point r,
all the solid dissolves/melts and a single liquid phase results. At point p, three phases are in
equilibrium (liquid, AB2, and B), F = C − P + 1 = 0, and p is therefore an invariant point. As
can be seen from Figure 4.3c, the composition of point p lies outside the composition range of
AB2 to B. The composition of the liquid phase therefore can’t be expressed in terms of
positive quantities of the solid phases with which it is in equilibrium. Such a point is called
a peritectic point. At a peritectic point there is no minimum in the melting curve as there is at
the eutectic point, only a kink.
Systems in which a compound melts incongruently follow relatively complex pathways on
cooling. Consider a liquid at point s on Figure 4.3c. On cooling, the liquidus is met at t, and
liquid and solid B will be in equilibrium. As cooling continues towards T2, the liquid will
move in composition from t to p. At temperatures just below the peritectic point p, the stable
solid phase becomes AB2. To form this solid, a peritectic reaction must occur, in which all the
solid B present must react with the liquid to form solid AB2. A drastic change in the
composition of solids present will therefore occur for a very small temperature change.
The amount of liquid present also changes dramatically at T2. Just above T2 the lever rule
shows the system will contain largely liquid (xL ≈ 0.86 given by uB/pB) and just below T2 far
less liquid (~0.36, uAB2/pAB2). On further cooling, the liquid composition follows curve pe,
and, at temperatures below T1, solid A forms along with solid AB2.
From the complexity of this behavior, it should not be surprising that systems containing
phases that melt incongruently frequently show non-equilibrium behavior. If liquid at point
s is cooled rapidly and/or if solid B falls to the bottom of the crucible during cooling such that
it is essentially removed from the system, the peritectic reaction may not have time to occur
completely and one would therefore observe a non-equilibrium mixture of A, AB2, and B at
 4.2 Two-Component Phase Diagrams 127

(a) T3

T2 Liquid
B+ T1

temperature
Liquid
A + Liquid
AB2
+ Liquid
AB2 + B
A + AB2

0.0 0.2 0.4 0.6 0.8 1.0

A xB AB2 B

(b)

Liquid
t empera tur e

A + Liquid B + Liquid

A+B
T1
A + AB2 AB2 + B

0.0 0.2 0.4 0.6 0.8 1.0

A xB AB2 B

(c)
s
Liquid
r T3
t empera tur e

t B + Liquid
A + Liquid p u
T2
AB2
T1 + Liquid
e
AB2 + B
A + AB2

0.0 0.2 0.4 0.6 0.8 1.0

A xB AB2 B

Figure 4.3 Two-component phase diagrams containing an intermediate phase AB2: (a) AB2 stable to its
melting point. (b) AB2 decomposes to solid A and B before melting. (c) AB2 decomposes on melting.
128 Phase Diagrams and Phase Transitions

low temperature. It should also be clear that if one wants to grow crystals of composition
AB2 from the melt under equilibrium conditions, one should start with a liquid composition
between e and p.

4.2.3 Solid-Solution Formation

We learned in Chapters 2 and 3 that solid solutions are ubiquitous. A schematic phase
diagram for a two-component (C = 2) system that shows complete solid solution in both the
liquid and solid states is shown in Figure 4.4. At high temperature (above the liquidus), one
has a single liquid phase and F = C + 1 − P = 2; at low temperature (below the solidus), one
has a single solid phase and F = 2. In both regions, either temperature or composition can be
varied without changing the number of phases present. Between these two extremes, one has
a two-phase region in which (since F = 1) the composition of the liquid and solid phases is
fixed by the temperature of the system. Consider what happens when a liquid with compos-
ition and temperature corresponding to point c is cooled under equilibrium conditions. At
point dℓ, the liquid of composition dℓ will be in equilibrium with a solid of composition ds, one
richer in B than the starting composition. On cooling to T3, the liquid will have composition
eℓ (richer in A) and the solid es (richer in B); the relative amounts of solid and liquid will be
given as eℓe/ese by the lever rule—approximately equal amounts as drawn. As one cools
further, the liquid becomes progressively richer in A (it follows curve dℓ fℓ), and the amount
of liquid relative to solid decreases. Simultaneously, the solid composition changes from ds to
fs. At T1, one has a single solid phase with the same composition as the original liquid.
The description above again holds only under ideal equilibrium conditions, which would
be very hard to achieve in practice. On more rapid cooling, it is very likely that the B-rich

c
Liquid dl
ds T4
el e
es T3
temperature

fl
liquidus fs T2

solidus
Solid

g T1

0.0 0.2 0.4 0.6 0.8 1.0

A mole fraction B, xB B

Figure 4.4 Phase diagram for a two-component system showing complete solubility in both solid and
liquid states.
 4.2 Two-Component Phase Diagrams 129

Liquid

tempera ture
Ass + Bss + Liquid
Ass Liquid
f Bss
e T1
d g h
Ass + Bss

c
0.0 0.2 0.4 0.6 0.8 1.0
A mole fraction B, xB B

Figure 4.5 Phase diagram for a two-component system in which a limited range of solid solutions is
formed.

crystals formed initially wouldn’t have sufficient time to fully equilibrate with surrounding
liquid. One would then expect crystals whose core is rich in B (compositions as high as ds) and
outer shell rich in A (compositions as A-rich as fℓ). Such effects are frequently observed in
mineralogy. For many applications, the range of compositions from the core to the outside
of the crystal may give rise to deleterious properties, and extended annealing periods may be
required to homogenize samples.
A full range of solid solutions as shown in Figure 4.4 would only be expected for
components that are chemically very similar, and few phase couples have full thermo-
dynamic miscibility at very low temperatures.9 In most systems at equilibrium, one observes
partial solid solution; there is a limit of solubility at a given temperature. Figure 4.5 shows
a typical phase diagram for a two-component system in which no compounds form but there
is solid solution at either end of the composition range. The diagram is directly related to
Figure 4.2a. At the left-hand side of the diagram, we have a region in which A will dissolve
B to form a single-phase solid solution A1−xBx (labeled Ass). The range of solubility generally
increases with temperature (curve cdf ) and reaches a maximum at the solidus temperature
T1. A similar region exists at the B-rich side of the diagram (Bss). At temperatures below T1,
these regions are separated (the boundary is called the solvus) by a two-phase region
containing a mixture of Ass and Bss. The composition corresponding to any point on this
diagram can be found using the same rules we’ve used above. Point g, for example, has an
overall composition of xA ≈ 0.7, xB ≈ 0.3; there will be two phases present, Ass and Bss. As
drawn, Ass will have composition d (A0.9B0.1) and Bss composition h (A0.05B0.95). The mole
fractions of each of the two phases as given by the lever rule are gh/dh for Ass and dg/dh for Bss

9
Many solid solutions can be quenched from high temperatures and remain stable at low temperatures because of
slow kinetics of demixing.
130 Phase Diagrams and Phase Transitions

c
Single Liquid T1
f e g T2

temperature
Two Liquids
Ass + d
Ass Liquid a b
B + Liquid

AASS++BB

0.0 0.2 0.4 0.6 0.8 1.0

A mole fraction B, x B B

Figure 4.6 Phase diagram for a two-component system with limited solid-solution range for A and liquids
rich in A and B immiscible under certain conditions.

(~0.765 A0.9B0.1 and ~0.235 A0.05B0.95). Note that (0.765 × 0.9) + (0.235 × 0.05) = 0.7 and
(0.765 × 0.1) + (0.235 × 0.95) = 0.3, as expected for the overall composition.
In the two examples described above, the components have been fully miscible when molten
and therefore form a single liquid phase. This is not, however, always the case, and we’re
familiar in everyday life with oil and water forming two-phase liquid systems. Liquid immisci-
bility is also encountered in ceramic phase diagrams. The example in Figure 4.6 shows an
immiscibility dome inside which two liquids exist as separate phases. Point e, for example, will
correspond to a mixture of two liquids of compositions corresponding to f and g. Temperature
T1 is called the upper consolute temperature. Along the line abd, two liquids (compositions at
a and b) and a solid (pure B) are in equilibrium. Point b is an invariant point (three phases are
present) called a monotectic point. It is similar to the eutectic point, except that one of the
phases that forms is a solid (here pure B) and the other a liquid (composition at a).
When we discussed the formation of compounds in Section 4.2.2, we moved from simple
eutectic phase diagrams such as Figure 4.2 to the more complex diagrams of Figure 4.3. Each
of these diagrams can be readily extended to allow for the formation of solid solutions in the
same manner as described above. For example, the incongruently melting system of
Figure 4.3c becomes Figure 4.7. While they appear complex at first glance, we can use the
same ideas as described above to read such phase diagrams.
A real-world A–B binary system that combines many of the ideas discussed above is shown
in Figure 4.8 for ZrO2–WO3. The system exhibits two structural forms of ZrO2 (monoclinic
and tetragonal structures) over the temperature range depicted, each of which shows a small
range of solid solution. There is a single AB2 compound ZrW2O8 that is only thermodynam-
ically stable above 1105 °C, melts incongruently, and has a peritectic point at 1257 °C. This
system is discussed further in the end-of-chapter problems.
 4.3 Three-Component Phase Diagrams 131

Liquid

tempera ture
Bss + Liquid

Ass + Liquid AB2 ss

+ Liquid Bss
Ass AB2 ss + Bss
AB2 ss
Ass + AB2 ss

0.0 0.2 0.4 0.6 0.8 1.0

A mole fraction B, xB B

Figure 4.7 Phase diagram for a simple binary system with one intermediate phase AB2 and a range of
solid solutions for each solid phase.

ZrO2 ss tet Liquid

1400 ZrW2O8 +
ZrO2 ss tet + Liquid Liquid WO3
tempera ture (ºC)

1300
+ Liquid
1200 ZrO2 ss tet + ZrW2O8
ZrO2 + ZrW2O8 ZrW2O8 + WO3
ss mon
1100

1000
ZrO2 ss mon + WO3
900 ZrO2 ss mon

0.0 0.2 0.4 0.6 0.8 1.0

ZrO2 mole fraction WO3 WO3

Figure 4.8 Partial phase diagram for the ZrO2–WO3 system at ambient pressure, after [1].

4.3 Three-Component Phase Diagrams

For a three-component or ternary system, we would need five axes (one for each
composition and one each for p and T ) to draw a general phase diagram. In practice,
an equilateral triangle is used to represent the composition of the system, and
temperature is represented by an axis normal to this triangle, forming a prismatic
diagram (Figure 4.9). One such prism is required for each pressure of interest. The
side faces of the prism represent the binary phase diagrams we’ve already discussed.
Due to the complexity of drawing and reading such diagrams, it is far more common
132 Phase Diagrams and Phase Transitions

tem per atur e

+L liquidus
surface C
+L
B+
L B
+L
A+B
B+C
A
C

Figure 4.9 A phase diagram for a three-component eutectic system.

to depict an isothermal section of the prism (Figure 4.10), which represents the phases
present at a specified temperature and pressure. In this chapter, we will only discuss
sub-solidus ternary diagrams and not address crystallization pathways for these
systems.
The reason for using an equilateral triangle for such phase diagrams is that its
geometry provides a natural way of expressing mole fractions xA, xB, xC of the three
components, which must, by definition, sum to 1. This arises from the fact that if you
take any point in an equilateral triangle and draw lines from it to intercept each edge at
right angles then the sum of the length of these lines is equal to the height of the
triangle (Viviani’s theorem). If, as in Figure 4.10a, we draw such a triangle and call its
height 1 then the distances marked xA, xB, and xC will always sum to 1. One corner of
the triangle represents pure A (the others are B and C), and lines drawn parallel to the
opposite edge moving towards that corner represent increasing mole fractions of
A. This is shown in Figure 4.10a where each of the dashed lines represents
a constant mole fraction of A, with the amount of A increasing as one moves from
edge CB towards corner A. If we consider point d, it lies on the line representing xA =
0.6. If we were to draw lines parallel to the other edges, d would lie on lines corres-
ponding to xB = 0.3 (lines parallel to the AC edge) and xC = 0.1 (lines parallel to AB).
A second way of determining composition is to use the triangle rule, which is the equivalent
of the lever rule we used for binary systems. In Figure 4.10b we draw lines from each corner
of the triangle through d to the opposite edge. The amount of each component is given by:
xA ¼ q=ð p þ qÞ; xB ¼ s=ðr þ sÞ ¼ 0:3; xC ¼ u=ðt þ uÞ ¼ 0:1:
 4.3 Three-Component Phase Diagrams 133

(a) A (b) A

0.8 0.2 p 0.8

s

cC
xB

mo
xC 0.6 0.4 u 0.6

fra

le
d d

fra
le
mo

cA
0.4 0.6 0.4
xA
t q r
0.2 0.8 0.2

C B C 0.2 0.4 0.6 0.8 B

mole frac B

(c) A (d) A

0.2 0.8 0.2 0.8

e
cC

cC
mo

mo
xB
0.4 0.6
fra

0.4 xC 0.6
fra
le

le
d f d
fra
le

fra
le
mo

mo
cA

cA
0.6 0.4 0.6 0.4
AC2
h xB'
0.8 0.2 0.8 xC' 0.2
e
g

C 0.2 0.4 0.6 0.8 B C 0.2 0.4 0.6 0.8 B

mole frac B mole frac B

Figure 4.10 Isothermal sections for a three-component system; (a) shows how the mole fraction of a phase
increases as one moves towards the corresponding corner, (b) illustrates the triangle rule for determining
composition, (c) illustrates how composition can be read from a single axis, and (d) shows a line
corresponding to a constant B:C ratio.

A third way of reading compositions, which is often the most useful in practice, is to read
the amount of each component directly from one axis. To do this, we take point d in
Figure 4.10c and draw dashed lines through it parallel to each edge of the triangle and
look where these lines intercept any edge. If we consider edge AB, then the mole fraction of
A is given by f B (as in two-component phase diagrams, the length “furthest” from pure
134 Phase Diagrams and Phase Transitions

A gives its mole fraction), xB as Ae, and xC is the remaining portion of the axis, namely ef.
This method is particularly useful as it also applies to scalene10 triangles. If, in the system of
Figure 4.10c, A and C can react to form an intermediate phase AC2 (or A1/3C2/3 when
normalized), then we might want to consider just the subsystem formed by A, B, and AC2.
This is represented by the triangle A–B–AC2 in Figure 4.10c. We could redraw this subsys-
tem as an equilateral triangle to read off the composition of point d in terms of A, B, and
AC2. However, our original diagram lets us read this information directly. To do this, we can
use edge B–AC2 as we already have construction lines (eh and dg) drawn parallel to the other
two edges, and composition can be read from the interception of these lines with B–AC2. The
mole fraction of AC2 is given by Bg/(B–AC2), the mole fraction of B by (h–AC2)/(B–AC2)
and the mole fraction of A by gh/(B–AC2). These can be measured from the diagram as
xA1/3C2/3 = 0.15, xB = 0.3, xA = 0.55 (using the normalized formula for AC2). Note that
multiplying each mole fraction with the phase composition gives an overall A:B:C ratio of
0.6:0.3:0.1, as expected.
One can see intuitively from these diagrams that the closer a point lies to a given corner,
the richer the composition is in that component. In fact, for points d and e of Figure 4.10d,
which lie on a straight line Ade, the ratio xB/xC = xB 0 /xC 0 , and the dashed line Ae therefore
shows the effect on a system of increasing the amount of A at a fixed B:C ratio. This is true for
any set of compositions that lie on a straight line linking them to a corner.
The sub-solidus phase diagrams in Figure 4.11 are for the relatively simple Al2O3–TiO2
–ZrO2 system and the more complex Y(O)–Ba(O)–Cu(O) system,11 in which the famous
YBa2Cu3O7 superconductor is found. In the second system, the (O) brackets imply that the
oxygen content in some phases on the diagram can be variable (redox behavior of Cu). In
Figure 4.11a, only binary and ternary oxides are formed (all compounds are points on the
edges of the main triangle); in Figure 4.11b quaternary oxides also form. The phases present
for any overall composition can be read in a similar way as for binary phase diagrams.
A point within any single triangle represents a mixture of the three solid phases at its corners
(the amounts of which can be found using the rules above). The edge of any triangle
represents a mixture of the two phases at the corners it links, and a vertex represents
a single phase. Thus under the conditions corresponding to Figure 4.11b, a certain amount
of the YBa2Cu3O7 phase12 (labeled 123 for the Y:Ba:Cu ratio) will be formed from starting
compositions anywhere within the triangle enclosed by points labeled CuO, 211, and
BaCuO2. Pure material will only be formed using the correct 1:2:3 Y:Ba:Cu ratio. In practice,
as we discussed for binary phase diagrams, many systems form solid solutions. These would
be represented by an area located around an ideal stoichiometric point.

10
A scalene triangle is a triangle with no equal sides.
11
Note that in Figure 4.11a the compositions are in mole fractions of Al2O3, whereas in Figure 4.11b in mole fractions
of YO1.5. Al2TiO5 (2:1 Al:Ti) therefore lies halfway along the edge joining Al2O3 and TiO2 in Figure 4.11a, whereas
Y2Cu2O5 (1:1 Y:Cu) lies halfway along the edge joining Y(O) and Cu(O) in Figure 4.11b.
12
Though the oxygen content may not be exactly 7; see Chapters 3 and 12.
 4.4 Structural Phase Transitions 135

(a) ZrO2 (b) Cu(O)

CuO

2
TiO

mo
c

le
c
fra

fra
a Y2Cu2O5 123 BaCuO2

c
le

ZrTiO4

ZrO
b
mo

143

2
Ba2CuO4
211
163
e d
f Al2TiO5 Y2O3 BaO
TiO2 Al2O3 Y(O) BaY2O4 Ba3Y4O9 Ba(O)
mole frac Al2O3

Figure 4.11 Isothermal sub-solidus phase diagrams (neglecting minor solid solubilities) for
(a) TiO2–Al2O3–ZrO2 below 1580 °C [2] and (b) Y(O)–Ba(O)–Cu(O) at 900 °C in O2 [3].

4.4 Structural Phase Transitions

In the first half of this chapter, we have explored the phase diagrams of a number of simple
systems. The concept of phase transitions was implicit to this discussion. In Figure 4.1, for
example, as we heat from point e to point a, there is a phase transition from solid ice to liquid
water. Many materials undergo phase transitions in the solid state where their structure
changes. For example, when α-quartz13 is heated, it undergoes a series of phase transitions.
At 573 °C (the transition temperature or critical temperature, Tc), a relatively subtle transition
to β-quartz occurs, in which the connectivity of the SiO4/2 tetrahedra is unchanged (i.e. no
chemical bonds are broken or formed), but the tilting pattern of tetrahedra changes and the
symmetry increases from trigonal to hexagonal. On further heating, more drastic modifica-
tions occur in which the bonding pattern changes; at 870 °C, a transition to β-tridymite and,
at 1470 °C, to β-cristobalite. The α- to β-quartz transition can be classified as a displacive
phase transition, because no bonds are broken or formed. The higher temperature transi-
tions are classified as reconstructive because they entail a major reorganization of the
structure involving bond breaking and formation.14 The final category of phase transition
we’ll encounter is order‒disorder transitions. Order‒disorder transitions can be further

13
Polymorphs are often distinguished by Greek letters α, β, γ, etc.
14
These different classifications aren’t used uniformly in the literature. For example, we describe a transition as
displacive if atoms only move a short distance, no chemical bonds need to be broken to achieve the conversion, and
there is a group–subgroup relationship (Appendix B) between the two space groups, regardless of whether the
process occurs in an abrupt or continuous fashion. Others choose to restrict the term displacive to continuous phase
transitions.
136 Phase Diagrams and Phase Transitions

subdivided into substitutional disorder (for example atoms ordering on specific lattice sites
such as the FePt example in Section 4.4.2) and orientational disorder where atoms or groups
of atoms change their positions without breaking the fundamental bonding pattern of the
material (such as the phase transitions of C60 described in Chapter 13). We’ll see throughout
the later chapters of this text that structural phase transitions are closely associated with
changes in the physical properties of materials.

4.4.1 Classification of Phase Transitions

The traditional way of classifying phase transitions was suggested by Paul Ehrenfest based
on how chemical potential (which is simply the Gibbs free energy G per mole) changes at the
phase transition. From thermodynamics we know that (∂G/∂p)T = V and (∂G/∂T)p = −S.
Since entropy and enthalpy changes at a phase transition are related by ΔS = ΔH/Tc, it
follows that for a phase transition such as melting or boiling, in which there are abrupt
changes in both volume and entropy/enthalpy, there will be a discontinuity in the first
derivative (or slope) of the free energy with respect to temperature (and pressure). Such
a transition is therefore called a first-order transition. Since constant-pressure heat capacity,
Cp, is defined as (∂H/∂T)p, it will be infinite at the phase transition.
The existence of a significant latent heat (ΔH) means that first-order transitions display
a hysteresis, a difference in phase-transition temperature on warming and cooling. We can
understand this through a simple thought experiment: if a sample is warmed to Tc and held at
precisely this temperature, the phase transition will not initially occur as there is no tempera-
ture gradient between the surroundings and sample to allow the flow of latent heat. It is only
when the surrounding temperature is raised above Tc that latent heat flows and nuclei of the
new phase begin to form. A first-order solid state transition will therefore lag behind the
temperature change causing it in any real experiment. The sample remains at temperature Tc
during the transformation while the heat supplied feeds the higher entropy of the product
(drives the transition). This means that the presence of two phases in coexistence is a criterion
for the first-order phase transition. To obtain any hysteresis intrinsically associated with
a first-order transition, this kinetic factor has to be eliminated by extrapolating temperature-
dependent measurements to the zero rate of temperature change.
A second-order phase transition is one in which there is a discontinuity in the second
derivative of the free energy with respect to temperature or pressure. The volume and the
entropy (and therefore the enthalpy) do not change abruptly at the phase transition, but
quantities such as the volumetric coefficient of thermal expansion or heat capacity do.15 The
dependence on temperature of various thermodynamic quantities for classical first-
and second-order transitions is shown in Figure 4.12.

15
The volumetric coefficient of thermal expansion, αV, as αV ¼ ð1=VÞ∂V=∂T ¼ ð1=VÞ∂2 G=∂P∂T and Cp as
Cp =T ¼ ð1=TÞ∂H=∂T ¼ ∂S=∂T ¼ ∂2 G=∂T 2 .
 4.4 Structural Phase Transitions 137

G V S H Cp 
first
order

T T T T T

G V S H Cp
second
order

T T T T T

Figure 4.12 Variation in thermodynamic quantities at first- and second-order Ehrenfest phase transitions.
The vertical dashed line corresponds to the transition temperature.

W1 W2 Figure 4.13 The Euler

W3 W3 strut. When the load on
the strut is increased
(weights W3 > W2 >
W1), the system will
eventually buckle
causing a reduction in
symmetry.

It is now more common to categorize phase transitions as discontinuous or continuous with

reference to the behavior of entropy or another order parameter (see below) at Tc. At
a discontinuous transition, the order parameter changes abruptly. All discontinuous phase
transitions are first-order in the Ehrenfest classification. During a continuous phase transi-
tion, the order parameter changes in infinitesimally small steps, as in a second-order transi-
tion. These concepts are developed further in the following sections.

4.4.2 Symmetry and Order Parameters

We can define many of the phase transitions we’ll encounter in terms of the changes in
symmetry that occur. Figure 4.13 illustrates the relevance of symmetry using a simple
mechanical model, the Euler strut. If we place increasing weights on top of a flexible plastic
rod, the rod will buckle or bend at some point. When this occurs, the symmetry of the system
is lowered or broken. We can see from the figure that the system could choose to buckle to the
right or the left. Even if the mass was loaded perfectly centrally on the rod, the system would
“choose” one direction or the other—ultimately decided in this ideal scenario by random
thermal displacements of the system. An analogy to the Euler strut example in materials
138 Phase Diagrams and Phase Transitions

Box 4.1 Characterizations: Differential scanning calorimetry (DSC)

Everyone enjoying an iced drink on a hot summer day exploits the fact that latent heat is
needed to melt the ice before the drink’s temperature starts rising. If you plotted the tempera-
ture evolution of a freshly iced cup versus an identical standard second cup where the ice has
just melted, the area between the two time-dependence curves would be proportional to the
latent heat of the ice in the first cup. This is the principle of DSC: a sample and a standard are
warmed at a constant rate, and the temperature of both is “scanned” at regular time intervals,
for example every second.
One patented DSC method adopts a smart twist. It measures the latent heat directly in terms
of the electrical power to supply it. The sample in a tiny aluminum pan, and an equivalent
empty pan for comparison, are each placed in their own tiny thermostat, each having
a miniature hot plate and a temperature sensor; the total of about 1 gram weight. The
electronics is wired up to keep their heating rate constant during the “scan”. The extra power
(J/s) needed for the sample is recorded every second upon heating through the phase transition.
When summed per unit mass of the sample, it gives the latent heat. The apparatus is usually
calibrated using a phase change of a standard of known latent heat and weight. This DSC
method is termed a power-compensation DSC.

sample thermostat reference thermostat

Pt resistance thermometer
heater

He He He He
power difference

The advantage of the method is a quick thermal equilibration that gives a fast response
needed for high rates of heating or cooling, as well as sharp peaks that make it possible to
separate closely occurring transitions. In addition to the latent heat of a discontinuous transi-
tion, it is possible to determine the heat capacity of the individual phases upon heating or
cooling beyond the transition peak.

chemistry could be the phase transitions between perovskite tilt systems outlined in
Chapter 1. In the ideal perovskite structure, one has 180° M–O–M bond angles (this ideal
structure would normally be the high-temperature, low-pressure limit), which can bend away
from 180° on cooling, lowering the symmetry of the system.
As a second model of broken symmetry, consider a simple single-headed arrow that can
fluctuate between pointing to the left and pointing to the right. If it switches between these
two orientations very rapidly, it would appear as a double-headed arrow to an observer. If,
however, its rate of switching was gradually reduced, there would come a point when the
 4.4 Structural Phase Transitions 139

Figure 4.14 Order‒

disorder transition in
FePt alloy.
cooling
Pt

Fe

disordered, fcc ordered, tetragonal

observer would see it “frozen” into one of two possible orientations. Once the arrow adopts
one of the two possible orientations, it changes from appearing double-headed to single-
headed and its symmetry is reduced.16 A structure with an atom dynamically disordered over
two closely separated sites at high temperature (for example, an atom in a double-well
potential), which chooses one of the two sites and orders on cooling, is one example of this
type of behavior. Phase transitions in ferroelectric perovskite materials such as BaTiO3 and
PbTiO3, which are discussed in Chapter 8, can also be understood in a similar way. As the
material is cooled from high temperature, Ti moves away from a position on a dynamic
average at the center of the BO6 octahedron, freezes in an off-center position, and the
material develops a spontaneous polarization. Again, this is associated with a lowering of
symmetry at the phase transition on cooling.
Symmetry changes leading to phase transitions can also be brought about by atomic
ordering. One important example is found in FePt 1:1 alloys. At high temperatures (above
~1300 °C), FePt has the face-centered cubic structure shown on the left of Figure 4.14, in
which Fe and Pt atoms are randomly distributed over all sites of the structure. On cooling, it
is thermodynamically favorable to order Fe and Pt atoms in layers perpendicular to the
original c axis, and the stable room-temperature structure is tetragonal (this is the so-called
L10 phase). Once again we see that ordering of atoms on cooling lowers the symmetry. This
particular transition is of technological relevance as the ordered material has a high
magnetocrystalline anisotropy, making it possible to produce magnetically hard nanoparti-
cles with a range of potential applications.
In each of these examples it is useful to introduce an order parameter, η, to describe the
distortion of the structure relative to the high-symmetry case.17 This is particularly true when
the distortion involves movement of groups of atoms or molecules. In the case of octahedral
tilts in perovskites, the tilt angle can be used to define the order parameter. For B cations
moving off-center in a BO6 octahedron, the shift in fractional coordinate of the metal might

16
Note that the point at which the transition is deemed to have occurred might be influenced by the experimental
method used to observe it. For example, if one used a photographic technique, a camera with a high shutter speed
would record the freezing transition before one with a low shutter speed.
17
The symbol Q is also commonly used for order parameter.
140 Phase Diagrams and Phase Transitions

be the order parameter. For the FePt transition, if we define one lattice site on the right of
Figure 4.14 as being the Fe site and the second as being the Pt site, we can define the order
parameter in terms of the fractional occupancy of each site by “right” and “wrong” atoms. In
a perfectly ordered structure, occPt(Fe) (the fractional occupancy of Pt at the Fe site) would be
0, and occFe(Fe) = 1; in the fully disordered occPt(Fe) = occFe(Fe) = 0.5. An order parameter
could therefore be defined as η = 2[occFe(Fe) − 0.5] and would vary from 0 to 1 for
a disordered to fully ordered material on cooling. Finally, entropy change itself is an order
parameter, such as (Sliquid − S)/(Sliquid − Ssolid) for H2O freezing.
Many readily measurable macroscopic quantities also show a simple dependence on η and
can be used to monitor a phase transition. For example the distortion of cell parameters as
a perovskite changes from cubic to lower symmetry can be expressed as a spontaneous
strain,18 ε = (a − acub)/acub, and it typically depends on either η or η2.19
It’s worth noting that although the examples described above have an order parameter
that could potentially vary continuously (though we’ll see that it needn’t in practice) from the
high- to low-symmetry situation, symmetry itself always changes abruptly—a symmetry
element is either present or absent in the structure. A continuous transition concludes when
the final infinitesimal heating step causes the order to finally disappear, at which point the
new symmetry emerges. For the vast majority of transitions, the high-temperature form has
the higher symmetry.

4.4.3 Introduction to Landau Theory

One of the advantages of introducing the order parameter to describe phase transitions in
solids is that it allows one to use Landau theory20 to describe their thermodynamics. The
basic assumption used in this approach is that the Gibbs free energy of the phase can be
approximated as a simple power series in terms of the order parameter η:

18
Strain (ε, the proportional displacement in shape or volume) is the deformation that occurs when a material is
subjected to a mechanical stress (σ, a force per unit area). For small deformations where materials behave in an
elastic manner, the stress tensor of rank 2 (a 3X
× 3 matrix; diagonalized in engineering to have six non-zero terms) is
related to the strain tensor of rank 2 by σi ¼ cij εj where cij are elastic stiffness coefficients (originally a tensor of
j
nine 3 × 3 matrices; in engineering simplified to 36 cij parameters relating the stress and strain matrices) and
subscripts refer to different directional components. For a uniaxial stress and isotropic body, this simplifies to
Hooke’s law σ = Eε where E is Young’s modulus. See, for example, Elliott [Physics and Chemistry of Solids (1998),
J. Wiley and Sons] for more detail.
19
Transitions that keep group–subgroup relationships can be categorized according to how the translational
symmetry of the lattice changes at the transition: if the lattice centering changes or a superlattice is formed, the
transition is called a zone-boundary transition, otherwise it is called a zone-center transition. Here the “zone”
refers to the Brillouin zone introduced in Chapter 6 as the volume in reciprocal space that lies closest to each
reciprocal lattice point; if lattice points are added or diluted, the zone boundaries change. Zone-boundary
transitions (such as the SrTiO3 perovskite tilting discussed in Section 4.4.6) generally have ε proportional to η2
and zone-center transitions to η.
20
We call this a phenomenological theory as it explains experimental results mathematically without using a rigorous
fundamental physical law.
 4.4 Structural Phase Transitions 141

1 1 1
GðηÞ ¼ G0 þ Aη2 þ Bη4 þ Cη6 þ . . . (4.6)
2 4 6

where G0 represents the part of the free energy that does not change at the phase transition.
The term G(η) − G0 thus represents the excess free energy compared to one that the
unchanged form would possess if no phase transition occurred. Additional terms can be
introduced to this equation if, for example, the order parameter is coupled to properties such
as strain. Usually G is independent of the sign of η such that only even powers are needed, as
in Equation (4.6), and it’s common to adopt the smallest number of terms required to
describe a system.

4.4.4 Second-Order Transitions

Let’s consider the phase transition shown schematically in Figure 4.15, which involves an
oxygen atom being continuously displaced from an ideal site midway between two metal
atoms—this could represent a tilting transition of octahedra in a perovskite. We can define
an order parameter η in terms of the oxygen displacement from the ideal site. At high
temperatures, the oxygen vibrates around its M–O–M midpoint, maintaining high symmetry
(η = 0). At low temperature, the oxygen vibrates around a position displaced to the left or to
the right (η ≠ 0). For this situation, oxygen displacements to the left or right (±η) are
equivalent, justifying the use of even powers of η in Equation (4.6).
Since the linear M–O–M arrangement is stable at high temperature (T >Tc), G(η) at η = 0
must be a minimum. The simplest expression that would produce this has to have a term in
Equation (4.6) dependent on η2 with a positive A coefficient. Below the phase-transition
temperature Tc, the linear M–O–M is no longer stable, implying that G(η = 0) becomes
a local maximum (Figure 4.15). This change requires that A changes sign from positive to
negative at the phase transition. The simplest way to express this is to give A a temperature
dependence such as:

AðTÞ ¼ aðT  Tc Þ (4.7)

with a positive. This means that A(T) is positive for T >Tc and negative for T < Tc. In order
to produce minima in our free-energy curve at η ≠ 0 below Tc, it is now necessary to include
a term in the G(η) series with a positive coefficient, such as one depending on η4. The Gibbs
free-energy expression becomes:

1 1
GðηÞ ¼ G0 þ aðT  Tc Þη2 þ Bη4 (4.8)
2 4

We can see from the form of G(η) curves in Figure 4.15 that because A(T) changes smoothly
with temperature, G(η) also changes smoothly, and the phase transition occurs in a smooth or
continuous fashion. At the transition temperature Tc, the state of both phases is identical, and
142 Phase Diagrams and Phase Transitions

M M

G( )
O O

M M

T>Tc
M M

O T=Tc
O
G0 T<Tc
M M
− 0 +
order parameter

Figure 4.15 Free-energy curves for a second-order displacive phase transition.

the symmetry of the body at this point must contain the symmetry of both phases—the two
space groups must be related by a group–subgroup relationship (Appendix B). In general, the
high-temperature phase has the higher symmetry, and the low-temperature phase has only
a subset of these symmetry elements. This turns out to be a powerful tool when investigating
the possible symmetries of the low-temperature structures for a variety of phase transitions.
A more detailed discussion of this topic can be found in the literature [4–8].
Equation (4.8) lets us investigate many of the important thermodynamic quantities
associated with the phase transition. At any temperature, the equilibrium value of the
order parameter is given by the minimum of the G(η) curve, i.e. where (∂G/∂η) = 0 and
(∂2G/∂η2) > 0. Differentiating Equation (4.8) gives:

∂G
¼ aðT  Tc Þη þ Bη3 ¼ 0 (4.9)
∂η

The three solutions to this equation are:

rffiffiffi
a
η ¼ 0 and η ¼  ðTc  TÞ1=2 (4.10)
B

For T >Tc, η = 0 represents the minimum in Gibbs energy, and there are no other real
solutions. For T < Tc, this solution (η = 0) is a local maximum in Gibbs energy, and the other
two solutions represent minima.
The way in which η varies with temperature can be expressed even more succinctly than
pffiffiffiffiffiffiffiffiffiffiffiffiffi
Equation (4.10). If we make const ¼  aTc =B then:
 4.4 Structural Phase Transitions 143

second order first order

order parameter, η

order parameter, η
Tc Tc

temperature temperature

Figure 4.16 Schematic temperature dependence of the order parameter η in the Landau description of
a second- and first-order transition.

η ¼ const½ðTc  TÞ=Tc β with β ¼ 1=2: (4.11)

The parameter β as a general variable is called the critical exponent. Under the Landau
approximation, the full structural order of η = 1 is acquired at T = 0 such that const = 1 and
η = [(Tc − T)/Tc]½. This simple dependence of the order parameter η on temperature
(Figure 4.16, left) is followed by many second-order transitions, at least close to Tc.
The transition thermodynamics can be derived as follows. Substitution of the non-zero η
from Equation (4.10) into Equation (4.8) gives the excess Gibbs energy acquired by the
transition on cooling from Tc to T as:

a2
DG ¼ G  G0 ¼  ðT  Tc Þ2 (4.12)
4B

Since G = H − TS, we can use S = −dG/dT to derive the excess entropy and enthalpy acquired
by the phase transition on cooling from Tc to T. The full-transition values can be determined
by setting T = 0 into the temperature-based terms for ΔS and ΔH below, or by setting η = 1
with Tc = B/a into their η-based terms:21
a2 1
DS ¼ ðT  Tc Þ ¼  aη2 (4.13)
2B 2
a2 a2 T 1 1
DH ¼ DG þ TDS ¼  ðT  Tc Þ2 þ ðT  Tc Þ ¼  aTc η2 þ Bη4 (4.14)
4B 2B 2 4

The derivations are explored in the end-of-chapter problems. Equation (4.14) shows that the
excess enthalpy of the system is a double-well function, and the positive sign of a and B in
Equation (4.13) means that entropy, as a measure of disorder, decreases as the system
distorts (orders) below Tc.

21
Transition entropies and enthalpies are always reported upon heating, hence would have opposite signs to those
suggested by Equations (4.13) and (4.14).
144 Phase Diagrams and Phase Transitions

The effect of the phase transition on the heat capacity can be found using the relationship
Cp = T(∂S/∂T)p. Below Tc, differentiating Equation (4.13) gives:

∂DS 1 ∂η2 a2 T
DCðT < Tc Þ ¼ T ¼  aT ¼ (4.15)
∂T 2 ∂T 2B
DCðT > Tc Þ ¼ 0 (4.16)

We see that the heat capacity changes abruptly by a2Tc/2B at Tc. Since B is positive, and we
are considering the cooling process, heat capacity shows a positive jump at Tc on cooling to
the low-temperature phase (see Figure 4.12), consistent with Ehrenfest’s definition of
a second-order transition.
Note from this analysis that once a and B are known, all the thermodynamic quantities
associated with the phase transition can be calculated. The values of a and B can be obtained
by measuring Tc and one of the excess thermodynamic quantities. The way in which an order-
parameter approach can be used to describe magnetic transitions is explored in the end-of-chapter
problems.

4.4.5 First-Order and Tricritical Transitions

We can apply similar arguments to investigate a discontinuous or first-order transition,
provided we can relate the two structures by an order parameter. Here, the high- and low-
temperature phases coexist in equilibrium at Tc. This coexistence requires that equal minima
in G(η) occur for different absolute values of η; one at the order parameter η = 0 of the
disordered phase and two at non-zero ±η of the ordered phase.22 The simplest Gibbs free-
energy function that will allow this coexistence has a fourth-order term negative and a sixth-
order term positive:

1 1 1
GðηÞ ¼ G0 þ aðT  T0 Þη2  Bη4 þ Cη6 (4.17)
2 4 6

where a, B, and C are all positive quantities as written. T0 is a temperature slightly

lower than Tc, and we will derive their relationship shortly. This function has a single
minimum at high T, three minima at intermediate temperatures above T0, and two
minima at temperatures below T0, as depicted in Figure 4.17. As in the second-order
case, differentiating Equation (4.17) with respect to η will tell us the order parameter
values that give rise to minima in the free energy. Depending on temperature, there are
up to three minima at:

22
Note that in the Landau model, the order parameter η is often used in a somewhat relaxed way as a transition
parameter and allowed to exceed the value of 1.
 4.4 Structural Phase Transitions 145

G( )
T>T1

T1
Tc

T<Tc G0

T0

T<T0

− 0 +
order parameter

Figure 4.17 Free-energy curves in the Landau model of a first-order phase transition.

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!1=2
Bþ B2  4CaðT  T0 Þ
η ¼ 0 and η ¼  (4.18)
2C

At high temperatures, we have a single η ¼ 0 minimum. On cooling, we first reach

a temperature T1 = T0 + B2/4aC (higher than Tc or T0), where two additional local minima
appear at η ≠ 0, though they have a higher G than the G at η ¼ 0.23 The phase-transition
temperature Tc is reached on further cooling, and is defined as the temperature where the
three minima in the free-energy curve are all equal to G0.24 This can be shown25 to occur at:

3 B2
Tc ¼ T0 þ (4.19)
16 aC

and Tc lies between T1 and T0. Finally, upon cooling through T0, the local η = 0 minimum
disappears and only the two η ≠ 0 minima remain.
Figure 4.18 shows the temperature dependence of the free energy of each minimum,
highlighting how the ordered η ≠ 0 phase becomes thermodynamically stable below Tc. The
temperature dependence of the order parameters for first- and second-order transitions are
compared schematically in Figure 4.16. Unlike Equation (4.10), the form of Equation (4.18)
means that η changes abruptly or discontinuously at a first-order phase transition. The order
parameter jumps from 0 to ±(3B/4C)½ on cooling through Tc, and, for those phase transi-
tions that are finished at this point, it can be constrained to η ± 1 by setting C = ¾B. Note that
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
23
T1 is the temperature where the expression within the square root B2  4CaðT  T0 Þ in Equation (4.18) equals 0.
24
This is our equilibrium condition of ΔG = 0 for the phase coexistence at the transition.
25
At Tc, G = G0 and dG/dT = 0 constrain the Landau coefficients. Performing these two operations on Equation (4.17)
yields two equations that subtract to give η2 = 3B/4C that is substituted into the dG/dT = 0 equation and rearranged
to give Equation (4.19).
146 Phase Diagrams and Phase Transitions

order parameter
free energy

Tc
more 0
stable
Tc more T0 T1
stable

98 100 102 104 106 98 100 102 104 106

temperature temperature

Figure 4.18 Temperature dependence of free energy and order parameter for a first-order transition in the
Landau approximation. Dotted regions T0 to Tc and Tc to T1 show the temperature ranges, discussed in
the text, over which local minima are present in G(η).

for the first-order case, the simple jump in η at Tc doesn’t imply anything about the pattern of
atomic displacements that takes place at the transition. The transition must occur by the
nucleation and growth of one phase in the other, as discussed above.
Just as we did for the second-order transition, we can derive other thermodynamic quan-
tities by substituting our expression for η of Equation (4.18) back into Equation (4.17) to
express G as a function of T. By differentiating with respect to T (as S = −dG/dT), we obtain
the entropy, and from Cp = T(∂S/∂T)p the heat capacity.26 Upon cooling through Tc, the
entropy change is −3aB/8C (negative, the system becomes more ordered).27 The phase
coexistence, hence ΔG = 0 at Tc, means that we can use ΔH = TΔS to obtain:

3aBTc
DHðT ¼ Tc Þ ¼  (4.20)
8C

We see that ΔH has a finite negative value for cooling through Tc.28 Upon heating, this ΔH is
positive and is the latent heat supplied to the equilibrium system of two phases at the transition.
We’ve seen that a differentiator between first- and second-order transitions is the sign of
the η4 contribution to G(η). Using the G(η) expression of Equation (4.6),29 a first-order

 32
B2 =4  aCðT  T0 Þ aBðT  T0 Þ B3
26
The is tedious but results in GðTÞ ¼ G0 
algebra þ  ,
0 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1 3C 2 4C 24C2
a @B B2 A a2 T
SðTÞ ¼ S0  þ  aCðT  T0 Þ , and Cp ðTÞ ¼ Cp0 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi.
2C 2 4 4 B2 =4  aCðT  T0 Þ
27
Details explored in Problem 4.16.
28
Contrast this to Equation (4.14) that shows that ΔH = 0 at T = Tc for a second-order phase transition.
29
Note that in Equation (4.17) we reversed the sign of the η4 term to keep all coefficients positive quantities in our main
discussion.
 4.4 Structural Phase Transitions 147

transition has B < 0 and a second-order, B > 0. The case where B = 0 is called a tricritical
transition, and for these η ∝ ðTc  TÞ1=4 with a critical exponent β = ¼ (see Problem 4.13).
What can cause the sign of B to become negative so that a phase transition becomes first-
order rather than second-order? One opportunity arises when the phase transition distorts
the lattice (also termed spontaneous strain). Our free-energy expression should then include
additional terms to account for this.30 As an example, the displacive transition in SrTiO3 can
be described by including a quadratic coupling term between strain and order parameter,
and the free-energy function can be approximated as:

1 1 1 1
GðηÞ ¼ G0 þ aðT  T0 Þη2 þ Bη4 þ λεs η2 þ cel ε2s þ . . . (4.21)
2 4 2 2

where εs is spontaneous strain, cel an elastic constant, and λ a coupling constant; the final
term in this equation is a Hooke’s-law-like term with elastic energy proportional to the strain
squared. Since the free energy depends on strain, at equilibrium (∂G/∂εs) = 0, which implies
that εs = −λη2/2cel. On introducing this to Equation (4.21) we find:
 
1 1 λ2
GðηÞ ¼ G0 þ aðT  T0 Þη þ
2
B η4 þ . . . (4.22)
2 4 2cel

The effect of strain is then to reduce the coefficient of the η4 term. If the coupling is sufficient
to make the overall η4 coefficient negative, the transition will be first-order.

4.4.6 Phonons, Soft Modes, and Displacive Transitions

The final concept we will discuss for understanding structural distortions at phase transitions
is that of the soft mode. In any molecule or material, atoms are never at rest but undergo
thermal vibrations. In an isolated molecule, we usually discuss these in terms of normal
modes. In H2O, for example, we can describe the possible vibrations in terms of three normal
modes that approximate to a symmetric and an antisymmetric stretch of O–H bonds and
a bending of the H–O–H bond angle. If we wanted to describe a hypothetical distorted water
molecule with one short and one long O–H bond, one recipe would be to imagine freezing the
asymmetric bond stretch at some point along the normal-mode coordinate describing the
H displacement.
In an extended solid, the interactions between neighboring atoms mean that they don’t
vibrate independently. The motion of one atom influences those around it, producing
displacement waves that travel through the crystal, termed lattice modes. The top panel of
Figure 4.19 shows this schematically for an isolated chain of atoms that show transverse
30
In general, excess free energy due to strain will contain terms dependent on εsη and εsη2 (though higher-order terms can
also be included). For zone-center transitions (see Footnote 19) only the linear coupling term is usually required; for
zone-boundary transitions only the quadratic term. More detail can be found in the text by Salje [E.K.H. Salje, “Phase
Transitions in Ferroelastic and Co-Elastic Crystals” (1993) Cambridge University Press].
148 Phase Diagrams and Phase Transitions

transverse

a
λ =10a, k=2 /λ = /5a wave moves through crystal

longitudinal

λ =2a, k= /a

a 2a

Figure 4.19 Phonons. Transverse and longitudinal phonons with λ = 10a in a 1D chain. In the longitu-
dinal case, the wave amplitude represents motion along the chain direction indicated by the small gray
arrows. The lower picture shows how a soft mode with k = π/a can lead to transition to a structure with
doubled unit cell.

displacements from the chain axis with a travelling wave of λ = 10a. The peaks and troughs of
this wave move through the crystal at a velocity (the phase velocity) given by c = λν such that
atoms oscillate at frequency ν. There are also longitudinal modes, in which the travelling-
wave amplitude describes atomic motion to the right (+ amplitude) or left (−) of the time-
averaged position along the chain direction. Lattice modes can always be described by waves
between λ = ∞ (all atoms move in the same direction corresponding to a translation of the
entire crystal) and λ = 2a (atoms in adjacent cells vibrate out of phase). In a normal 3D
crystal, we have to consider waves propagating in all directions, and it becomes convenient to
label them in reciprocal space using a wave vector k with magnitude 2π/λ.31 The language is
analogous to that developed in Chapter 6 to describe band theory and excellent descriptions
can be found in texts such as those by Kittel, Ziman, or Dove (see Further Reading).32 Each
lattice mode or wave will affect all the atoms, and the amplitude of the motion depends on its
frequency and the temperature. The overall motion of atoms will be a superposition of the
motion caused by each of the waves.
To this point we’ve taken a classical view, but in reality the motion of atoms in solids is
determined by quantum mechanics, and the energy of the vibrations is quantized. In the
same way that wave-particle duality allows us to describe light waves in terms of particles
31
As discussed in Chapter 6, the possible values of k (the lengths of k) range from 0 to ±π/a with + and −
corresponding to waves travelling in opposite directions.
32
For a crystal with n atoms in the unit cell, there are 3n different combinations of motion, each of which has a specific
frequency (dependent on interatomic forces) at each value of k; these are called branches. The dependence of ν on
k (labels direction and λ) for each branch is called a phonon-dispersion curve and is analogous to a band-structure
diagram. For a 3D monatomic (n = 1) crystal, there will be three branches; one longitudinal and two transverse. At
k = 0 all atoms move in phase and ν = 0 for each branch. As k changes, the frequency of the different branches will
change.
 4.4 Structural Phase Transitions 149

called photons, vibrational waves can be described as particles called phonons. A phonon is
the quantum unit of vibrational energy in a crystal and its energy is given by hν. When
vibrational waves propagate heat or sound energy through a crystal, it is carried by the
motion of phonons. As temperature is increased, the amplitude of atomic vibrations
increases, and this corresponds to an increase in the number of phonons in the crystal.
Slightly confusingly, the common usage is that the lattice modes describing the vibrations of
atoms are also called phonons.
Anharmonic effects33 in crystals lead to phonon frequencies varying with temperature. If
a material has a relatively low-frequency phonon (or set of related phonons) whose fre-
quency decreases on cooling, we can envisage a situation where the frequency could fall to
zero. At this point, the structure becomes unstable with respect to a permanent distortion
corresponding to the atomic motion described by the phonon—it becomes soft with respect
to the distortion. The phonon becomes frozen into the structure and a displacive phase
transition occurs to a lower-symmetry structure. The phonon involved is called a soft mode.
The atoms now vibrate around their new equilibrium positions and the frequency starts to
increase again on further cooling.
If we return to our simple example of a H2O molecule, freezing in different normal modes
(bends, stretches) will lead to a different distortion of the molecule with potentially different
point-group symmetries. The same is true of soft modes in extended structures, and there is
an elegant language that lets you explore the different symmetries of the structures that could
form (colloquially called child structures) from a high-symmetry (parent) structure,34 depend-
ing on the symmetry properties (decribed using an irreducible representation or irrep) of the
phonon involved. These can be explored through web-based tools such as ISODISTORT
(http://stokes.byu.edu/iso/isodistort.php).
As an example, SrTiO3 undergoes a cubic to tetragonal phase transition on cooling through
~110 K due to a soft mode at k = (½, ½, ½). The atomic motions that freeze into the structure
below Tc are shown schematically in Figure 4.19 and consist of coupled rotations of TiO6
octahedra around c, with adjacent octahedra rotated in opposite directions. This is one of the
tilted perovskite structures discussed in Section 1.5.3. Since k is non-zero, the unit cell of the
low-temperature structure is larger. Similarly, the displacive phase transition between α- and
β-quartz can be described by the softening of a ~200 cm−1 phonon of β-quartz upon cooling.
This phonon is at k = 0, so the unit-cell size remains unchanged in α-quartz.35 In some cases,
the wavelength of the soft mode doesn’t correspond to an integer number of unit cells of the
parent structure. This is another recipe for formation of an incommensurate structure (dis-
cussed in Section 2.10), where the structure can’t be conveniently described using

33
In the simplest treatment of phonons (the harmonic approximation), it’s assumed that the energy of the system
depends only on the square of the relative displacements of adjacent atoms. The energy is then the same as that of
a set of harmonic oscillators. Higher-order contributions to the energy are called anharmonic terms and are
normally treated as a perturbation to the harmonic approximation.
34
More formally, the highest-symmetry structure is called the aristotype and lower-symmetry structures hettotypes.
35
k = 0 means λ = ∞ so that all β-quartz cells undergo the same in-phase distortion into α-quartz.
150 Phase Diagrams and Phase Transitions

a conventional 3D crystallographic description. In fact, there is a narrow 1.3 K region just

below Tc in quartz where the structure is incommensurately modulated for this reason. At
lower temperatures, the soft-mode wavelength is said to “lock in” to the lattice and the
incommensurate modulation disappears. There is more detail about the intricacies of soft
modes and how this simple model applies in real materials in Further Reading.

4.5 Problems

4.1 Give a brief definition of the terms phase (P), component (C), and degrees of freedom (F)
in the condensed matter phase rule P + F = C + 1.
4.2 Refer to the phase diagram depicted below. (a) State which four phases are stable at 100 °C.
(b) What is the name given to the horizontal line separating region 2 from 1 and 3? (c) What
are the approximate melting points of A, AB, and B? (d) What happens if you try and melt
solid AB2? (e) State what phases are present in each of areas 1–9. (f) Do any of the phases
depicted form solid solutions? (g) State the number of phases and degrees of freedom at
points a, w, x, y, and z. (h) Describe what happens when compositions at each of points a to
f are cooled from high temperature. (i) Estimate the relative amounts of solid and liquid
when a composition at point a (xB ≈ 0.065) is cooled to 500 °C, 400 °C, and 300 °C. (j) What
might be observed if composition at point e is cooled rapidly? (k) State the differences
between the peritectic reactions that happen on cooling compositions at points e and f.

a b c d e f
600
8
500 9
temperature ( C)
o

400 w 4 6
1 3
300 x
7
z
200 5
y
2
100

0.0 0.2 0.4 0.6 0.8 1.0

A mole fraction B, xB B

4.3 Using the phase diagram of Figure 4.8: (a) State how you would attempt to prepare
a solid polycrystalline sample of ZrW2O8. (b) State how you would attempt to grow
single crystals of ZrW2O8.
4.4 In the system Al2O3–BaO, five phases stable above 1300 °C were identified: Al2O3, Al12
BaO19, Al2BaO4, Al2Ba3O6, and BaO. Each was found to melt congruently at 2072 °C,
1900 °C, 1811 °C, 1616 °C, and 1918 °C, respectively. Eutectics form at xBaO = 0.11, 0.32,
 4.5 Problems 151

0.66, and 0.87 (relative to Al2O3) with melting points of 1875 °C, 1620 °C, 1480 °C, and
1425 °C. Sketch and fully label the phase diagram of this system.
4.5 Perovskite chemists searching in the CaO–TiO2 system initially found four phases stable
above 1300 °C: CaO, Ca3Ti2O7, CaTiO3, and TiO2. CaO, CaTiO3, and TiO2 were
reported to melt congruently at 2600 °C, 1970 °C, and 1830 °C and Ca3Ti2O7 to melt
incongruently at 1750 °C. Eutectics were reported at xTiO2 = 0.29 and 0.76 with melting
points of 1695 °C and 1460 °C. Sketch and fully label a phase diagram for this system.
4.6 Use the phase rule to explain how a mixture of Ni and NiO can be used to provide
a controlled low oxygen partial pressure in a closed system.
4.7 If the height of the triangle in Figure 4.10a is 1, prove that the distances xA, xB, and xC
must sum to 1. Hint: Write an expression for the total area of triangle ABC in terms of
constituent triangles such as ABd.
4.8 Plot the following compositions on a triangular composition diagram. Comment on the
compositions of points a, b, and c and of points d, c, and e.

Point xA xB xC

a 0.8 0.1 0.1

b 0.4 0.3 0.3
c 0.2 0.4 0.4
d 0.1 0.7 0.2
e 0.3 0.1 0.6

4.9 State the compounds you would expect to form and their relative phase fractions
when oxide mixtures corresponding to points a–f in Figure 4.11 are reacted under
equilibrium conditions.
4.10 The following examples of phase transitions are discussed either in this chapter or in
other parts of the book. In each case, would you describe the transition as reconstruct-
ive, displacive, or order–disorder in nature? State whether or not you would expect to be
able to isolate the high-temperature phase at low temperature. (a) The transition from α-
quartz to β-quartz at 573 °C. (b) The transitions from β-quartz to tridymite (870 °C)
then to β-cristobalite at 1470 °C. (c) The transition at 641 °C of the a0b0c+-tilted
NaNbO3 perovskite (P4/mbm) to the cubic (Pm3m) perovskite a0b0c0. (d)
4.11 State whether symmetry decreases or increases at the water-to-ice phase transition.
4.12 For a ferromagnetic second-order phase transition, assume that the relative magnetiza-
tion M can be used as the order parameter in G(M) = G0 + ½AM2 + ¼BM4. Sketch the
temperature dependence of M on Tc.
4.13 Prove that for a tricritical transition η ¼ ½ðTc  TÞ=Tc  1=4 .
4.14 Show that the expressions given in the text for the critical temperature of a first-order
phase transition [Equation (4.19)] and for the abrupt order parameter change of η =
152 Phase Diagrams and Phase Transitions

±(3B/4C)½ at Tc are consistent with the definition that the three minima in the Gibbs
energy curve are equal to G0 at Tc (Figure 4.17).
4.15 Show that the expression relating Tc and T0 of a first-order phase transition [Equation
(4.19)] and the Gibbs energy expression in Footnote 26 are consistent with the definition
that the three minima in the Gibbs energy curve are equal to G0 at Tc (Figure 4.17).
4.16 Show that the entropy and enthalpy changes for a first-order phase transition are given
by ΔS = −3aB/8C and ΔH = −3aBTc/8C.
4.17 On cooling the cubic perovskite SrZrO3 (space group Pm3m) from high temperature, it
undergoes a phase transition to the tetragonal space group I4/mcm. At the phase transition,
the c cell parameter doubles and a increases by √2. The strain et is related to the difference in
pffiffiffi 
a and c parameters scaled 
back to those of the cubic cell. With appropriate normalization,
et ¼ ð2= 3Þ ðc  aÞ=acub where acub is the cell parameter expected for an undistorted
material at that temperature, and is expected to be proportional to η2. The table below
contains unit-cell parameters at various temperatures from ref. [9] that were later
analyzed in ref. [10]. Plot the temperature evolution of the cell parameters. By assuming
a sensible functional form, predict acub for each temperature below 1360 K. Comment
on your graph. (b) Comment on a plot of et0.5 versus T. (c) From a suitable plot, show
that the data are consistent with the transition being close to tricritical in character and
determine the transition temperature Tc.

T (K) a (Å) c (Å) T (K) acub (Å)

1160 4.1394 4.1514 1360 4.1536

1180 4.1405 4.1520 1380 4.1546
1200 4.1416 4.1524 1400 4.1555
1220 4.1428 4.1529 1420 4.1565
1240 4.1441 4.1532 1440 4.1575
1260 4.1455 4.1535 1460 4.1585
1280 4.1469 4.1539 1480 4.1594
1300 4.1485 4.1542 1500 4.1604
1320 4.1500 4.1542
1340 4.1517 4.1537

4.6 Further Reading

E.M. Levin, C.R. Robbins, H.F. McMurdie, “Phase Diagrams for Ceramists” (1964) American
Ceramic Society.
H. Okamoto, “Desk Handbook: Phase Diagrams for Binary Alloys” (2000) ASM International.
M.T. Dove, “Introduction to Lattice Dynamics” (1993) Cambridge University Press.
M.T. Dove, “Structure and Dynamics”, Oxford Master Series in Condensed Matter Physics (2003)
Oxford University Press.
 4.7 References 153

J.M. Ziman “Electrons and Phonons: The Theory of Transport Phenomena in Solids”, Oxford Classic
Texts in the Physical Sciences (2001) Oxford University Press.
L.D. Landau, E.M. Lifshitz, “Statistical Physics, Course of Theoretical Physics, Volume 5” 3rd edition
(2000) Reed Educational and Professional Publishing Ltd.
C. Kittel, “Introduction to Solid State Physics” (1996) John Wiley and Sons.

4.7 References
[1] L.L.Y. Chang, M.G. Scroger, B. Phillips, “Condensed phase relations in the systems ZrO2–WO2
–WO3 and HfO2–WO2–WO3” J. Am. Ceram. Soc. 50 (1967), 211–215.
[2] A.S. Berezhnoi, N.V. Gul’ko, “The system Al2O3–TiO2–ZrO2” Dopov. Akad. Nauk Ukr. RSR 1
(1955), 77–80.
[3] P. Karen, A. Kjekshus, “Phase diagrams and thermodynamic properties” in “Handbook on the
Physics and Chemistry of Rare Earths, Volume 30”, editors K.A. Gschneidner, Jr., L. Eyring, M.
B. Maple (2000) Elsevier, 229–371.
[4] H. Wondratschek, U. Müller (editors), “International Tables for Crystallography, Volume A1:
Symmetry relations between space groups” (2011) http://dx.doi.org/10.1107/97809553602
060000110.
[5] L.D. Landau, E.M. Lifshitz, “Statistical Physics, Course of Theoretical Physics, Volume 5” 3rd
edition (2000) Reed Educational and Professional Publishing Ltd.
[6] B.J. Campbell, H.T. Stokes, D.E. Tanner, D.M. Hatch, “ISODISPLACE. A web-based tool for
exploring structural distortions” J. Appl. Crystallogr. 39 (2006), 607–614.
[7] H.T. Stokes, D.M. Hatch, “Coupled order parameters in the Landau theory of phase transitions
in solids” Phase Transitions 34 (1991), 53–67.
[8] E.F. Bertaut, “Representation analysis of magnetic structures” Acta Crystallogr. Sect. A 24
(1968), 217–231.
[9] C.J. Howard, K.S. Knight, B.J. Kennedy, E.H. Kisi, “The structural phase transitions in
strontium zirconate revisited” J. Phys. Condens. Matter 12 (2000), L677–L683.
[10] R.E.A. McKnight, C.J. Howard, M.A. Carpenter, “Elastic anomalies associated with transform-
ation sequences in perovskites: I. Strontium zirconate, SrZrO3” J. Phys. Condens. Matter 21
(2009), 015901.
5 Chemical Bonding

Changes in crystal structure invariably lead to changes in physical and/or chemical proper-
ties. In some cases, these changes can be dramatic, as illustrated by the contrasting properties
of the allotropes of carbon (diamond, graphite, graphene, C60, etc.); in other cases they are
subtle but nonetheless important. To understand the relationship between structure and
properties, one must first understand chemical bonding.
We begin this chapter with an overview of ionic bonding. From there we move on to the
properties of atomic orbitals (AOs) and their interactions to form covalent bonds through
the framework of molecular orbital theory. In Chapter 6, we then build upon these principles
to describe the formation of bands in extended solids. In this way, covalent and metallic
bonding can be understood through a common approach.

5.1 Ionic Bonding

Although there are no compounds where the bonding can be described as purely ionic, the
ionic model is a useful approximation for many compounds. We begin our treatment of
bonding with a brief overview of the factors that determine the strength of ionic bonding in
crystalline solids.

5.1.1 Coulombic Potential Energy

The coulombic potential energy, UC, between two ions of charge numbers z1 and z2 separated
by a distance d is:

ðz1 eÞ  ðz2 eÞ
UC ¼ (5.1)
4πε0 d

154
 5.1 Ionic Bonding 155

first nearest second nearest third nearest

neighbors neighbors neighbors
(attractive) (repulsive) (attractive)

Figure 5.1 The NaCl structure and the first, second, and third nearest neighbors to a Cl− ion. The Cl− ions
are gray and the Na+ ions black.

where e is the elementary charge and ε0 is the electric constant.1 To estimate the strength of
ionic bonding in a crystal, we treat the ions as point charges and use Equation (5.1) to capture
all electrostatic interactions in the crystal, both attractive and repulsive.
To illustrate, consider the electrostatic interactions in the NaCl structure shown in Figure
5.1. We begin with the Cl− ion in the center of the unit cell and consider the interaction
between this ion and all other ions in the crystal. The nearest neighbors are the six Na+ ions
that lie at the center of the faces of the unit cell, at a distance d from the central Cl− ion. The
potential energy of this interaction is negative (an attraction) and amounts to six times
Equation (5.1) with z1 and z2 equal to −1 and +1. The second nearest neighbors are 12 Cl−
ions that lie at the center of each edge of the cubic unit cell, at a distance of √2d. The potential
energy of this interaction is positive (a repulsion; z1 = z2 = −1) and is 12/√2 times Equation
(5.1). The third nearest neighbors are eight Na+ ions that lie at the corners of the unit cell, at
a distance of √3d. This interaction is attractive (z1 = −1, z2 = +1), and has a magnitude of 8/√3
times Equation (5.1). This process must be repeated for increasingly distant neighbors
leading to an infinite series, of which the first seven terms are as follows:
 
ðz1 eÞ  ðz2 eÞ 12 8 6 24 24 12
UC ¼ 6  pffiffiffi þ pffiffiffi  pffiffiffi þ pffiffiffi  pffiffiffi  pffiffiffi … (5.2)
4πε0 d 2 3 4 5 6 8

The infinite series expressed in Equation (5.2) does not readily converge because successive
shells tend to alternate between those containing anions and those containing cations.
Convergence to a value near 1.7476 [times Equation (5.1) for the sole Na+Cl− pair] is
obtained if the shells are chosen in such a way that the net charge of each shell is nearly
neutral [1]. This value, which is the same for all ionic compounds with the NaCl-type
structure, is called the Madelung constant, A. The net coulombic potential energy per mole
of NaCl (z1, z2 of opposite signs) is then:

1
UC is calculated from Coulomb’s law, which describes the electrostatic force between two charges, as the work
required to separate the positive and negative ions from their initial distance d to infinity. Note also that the electric
constant of ~8.8542×10−12 C/(V m) is often referred to as the permittivity of free space.
156 Chemical Bonding

Table 5.1 Values of the Madelung constant A for selected structure types.

Structure type Coordination A Structure type Coordination A

[8c] [8c]
Cesium chloride Cs Cl 1.763 Fluorite Ca F2[4t]
[8c]
2.519
Sodium chloride Na[6o] Cl[6o] 1.748 Rutile Ti[6o] O2[3] 2.408
Wurtzite Zn[4t] S[4t] 1.641 Cadmium chloride Cd[6o] Cl2[3n] 2.244
Zinc blende Zn[4t] S[4t] 1.638 Cadmium iodide Cd[6o] I2[3n] 2.192

z1 z2 e2 NA
UC ¼ A (5.3)
4πε0 d

Similar derivations can be carried out for other structure types. The Madelung constants for
some structure types are given in Table 5.1, many others can be found in the literature or
calculated using the Ewald method [2]. We see that the Madelung constant increases as the
coordination number increases from four for zinc blende/wurtzite to six for sodium chloride
to eight for cesium chloride. It is smaller for layered structures with direct anion–anion
contacts (CdI2 or CdCl2) than for structures with the same stoichiometry where the cations
are distributed more uniformly (rutile TiO2). The Madelung constant also increases, almost
proportionally, with the number of ions per formula unit.
We can make several generalizations about the factors that optimize electrostatic inter-
actions. The electrostatic attraction holding the ions together increases as the ionic charges
increase (MgO will have a more negative UC than LiF) and as the cation–anion distance
d decreases (LiF will have a more negative UC than RbBr), though the latter effect is less
dramatic. As a rule, electrostatic interactions are highest in symmetric structures that allow
for efficient packing of ions and high coordination numbers.

5.1.2 Lattice Energy and the Born–Mayer Equation

Of the two alternative ways to define lattice energy (UL), we’ll use the energy of formation of
the ionic crystal at T = 0 K:

aMyþ ðgÞ þ bXz ðgÞ → Ma Xb ðsÞ (5.4)

This process releases energy of the system into the surroundings, hence UL is a negative
number just like the coulombic potential energy UC introduced in the preceding section.
The lattice energy is an equilibrium of attractive and repulsive interactions. To model it
appropriately, we must therefore consider not only the coulombic attraction UC, but also the
repulsive potential energy Ur (Figure 5.2) that arises when electron clouds on the otherwise
attracting neighboring ions approach each other too closely. That repulsion can be
 5.1 Ionic Bonding 157

approximated by a variety of mathematical functions that rise rapidly as the distance

d between the two ions decreases. A common choice is the two-parameter exponential
function:2

Ur ¼ Bed=ρ (5.5)

where B is a magnitude constant that includes NA, and ρ is an empirical constant typically
taken to be 3.45×10−11 m (0.345 Å). Combining the coulombic potential energy of the cation
and anion charges, Equation (5.3), with the repulsive potential energy arising from inter-
actions between electron clouds on neighboring ions, Equation (5.5), we obtain an expres-
sion for the lattice-formation energy UL of the ionic compound of charge numbers z1 and z2:
 
z 1 z 2 e 2 NA
UL ¼ ðUC þ Ur Þ ¼ A þ Bed=ρ (5.6)
4πε0 d

As shown in Figure 5.2, Equation (5.6) has a minimum at the equilibrium interatomic
distance3 d0 [3] that is calculated4 as dUL/dd = 0:

dUL z1 z2 e2 NA Bed0=ρ
¼ A ¼0 (5.7)
dd 4πε0 d0 2 ρ

From here, we can solve Equation (5.7) for B and substitute this expression back into
Equation (5.6), giving the Born–Mayer equation:
 
z1 z2 e2 NA ρ
UL ¼ A 1 (5.8)
4πε0 d0 d0

A generic plot of the coulombic potential energy, the repulsive potential energy, and the
lattice-formation energy of an ionic structure as a function of interatomic distance is shown
in Figure 5.2. For typical interatomic distances, Ur is about 10–20% of UC.

2
It is also common to represent repulsive interactions with the expression, Ur = B/d n, where n has a value between 5 and 12
depending upon the electron configuration of the cation and anion. This approach leads to the Born–Landé equation.
3
This d0 can be determined either from crystallographic studies or estimated from ionic radii (d0 = rC + rA). Ionic radii
are derived from observed interatomic distances in a large database of crystal structures. The most extensive and
widely used set of radii is that determined by Shannon. These radii are listed in two parallel sets, one called crystal
radii (CR) the other ionic radii (IR). Cations have CR that are 0.14 Å larger than their IR, while anions have CR that
are smaller than IR by the same amount. The CR set is chosen to approximate ionic sizes determined from minima in
the electron density between a given cation and anion and is thought to more accurately represent the true sizes of
ions, the IR are chosen to have the same radii for the O2− and F− ions as initially chosen by Linus Pauling. Both sets
give identical estimates of cation–anion distances.
4
Remember from calculus that the first derivative of a function is equal to zero at either a minimum or a maximum of
that function. In this case, dUL/dd = 0 identifies the minimum (most favorable) energy. The fact that it is a minimum
rather than a maximum can be shown by taking the second derivative.
158 Chemical Bonding

U C+ U r repulsive potential energy, U r

energy

0
d0

dU L
=0
U0 dd

coulombic potential energy, U C

interatomic distance, d

Figure 5.2 The coulombic (UC) and repulsive (Ur) potential-energy contributions to the lattice-formation
energy.

5.1.3 Experimental versus Calculated Lattice-Formation Energies

How does the lattice-formation energy calculated with the Born–Mayer equation compare
with experiment? The lattice energy as defined in Equation (5.4) cannot be determined in
a single experiment. Instead, the lattice-formation enthalpy, HL, is determined from several
measurements using a thermochemical cycle called a Born–Haber cycle. As the enthalpy
change ΔH in the Born–Haber cycle is associated with the same reaction as Equation (5.4),
HL ≈ UL, and we will treat the two as numerically equivalent.5
As an example, the Born–Haber cycle for KCl is shown in Figure 5.3. In the left branch,
energy is supplied to decompose the salt into its constituent elements, atomize molecular
chlorine, and evaporate and ionize potassium. In the right branch, energy is released by
chlorine atoms acquiring an electron and by the formation of solid KCl from the gaseous
ions. The enthalpy of the latter step is the lattice-formation energy we are seeking. For KCl,
the value of HL is found to be −718 kJ/mol using this approach.
It is interesting to note that while we normally think of the electron transfer from metal to
nonmetal as a spontaneous process, comparison of the two energies on top of Figure 5.3
shows that for potassium and chlorine this process is endothermic. In fact, this energy is

5
For a process at constant pressure, where the gaseous ions behave ideally, the relationship between lattice-formation
enthalpy and energy is given by the equation, ΔHL = ΔUL + ΔnRT, where Δn is the number of moles of gaseous ions
produced (e.g. Δn = 2 for NaCl, Δn = 3 for MgF2). At T = 0, the lattice-formation energy and enthalpy are equal,
while for finite temperatures the two values differ only slightly (RT = 2.5 kJ/mol at T = 300 K). If one considers the
vibrations of the atoms in the solid and in the gas phase more rigorously, the equations become more complicated.
For more details, the interested reader is directed to an article by H.D.B. Jenkins, J. Chem. Ed. 82 (2005), 950–952.
 5.1 Ionic Bonding 159

Figure 5.3 The Born–Haber cycle for KCl constructed from enthalpies of sublimation for K, atomization
for Cl2, the ionization energy (IE) of K, and the electron-gain enthalpy (Heg) of Cl (note Heg has the same
magnitude but opposite sign as electron affinity).

endothermic for any combination of elements. The energy released upon forming an
extended ionic solid is the driving force behind the spontaneous and often highly exothermic
reactions that occur between metals and nonmetals.
A comparison of experimental lattice-formation enthalpies (HL) and calculated lattice-
formation energies (UL) for alkali-metal halides with NaCl-type structure is given in Figure
5.4. Given the simplicity of this model, the agreement is surprisingly good. Although the
lattice-formation energies are underestimated by 3–6%, the functional dependence is well
reproduced in the experimental data.
To obtain more accurate estimates of lattice-formation energy, we must include additional
factors, such as van der Waals forces, polarization of ions in low-symmetry environments,
and zero-point energy. Each of these effects is briefly discussed below.
Van der Waals forces collectively refer to weak attractive forces arising from induced
dipoles in the electron clouds of neighboring atoms and molecules. They are solely respon-
sible for holding atoms together when noble gases solidify, and they are the dominant force
when neutral molecules condense to form molecular solids. They tend to favor close packing,
which explains why the noble gases crystallize with a ccp structure at low temperatures.
London dispersion forces are the most common type of van der Waals forces.6 They arise
from random fluctuations of the electron density around an otherwise spherically symmetric
6
Other types of van der Waals forces include dipole–induced-dipole forces, and interactions between rotating polar
molecules.
160 Chemical Bonding

-550

fluorides
-650 chlorides
bromides
lattice energy (kJ/mol) iodides

-750

-850

-950

-1050
1.9 2.2 2.5 2.8 3.1 3.4 3.7
cation–anion distance (Å)

Figure 5.4 Experimental lattice-formation enthalpies HL for alkali-metal halides with the NaCl-type
structure are indicated with symbols. The smooth curve shows the lattice-formation energy UL as
calculated with the Born–Mayer equation for comparison.

atom. The dipolar electrostatic field that is temporarily created in one atom induces
a transient dipole of opposite polarity in a neighboring atom. The strength of the van der
Waals force increases with increasing atomic number, but in all cases is much weaker than an
ionic bond. It also drops off more rapidly with increasing distance, varying inversely with the
sixth power of the internuclear separation.
Polarization is a distortion of the electronic charge density of an ion in response to the
electric field created by the surrounding ions. The most important type of polarization in
ionic crystals is the polarization of large soft anions by adjacent cations that are smaller and
often more highly charged. Polarization effects, together with van der Waals forces, are often
invoked to help explain the stability of layered structures with direct anion–anion contacts
like CdCl2 and CdI2 [4]. Polarization of the electron distribution of the halide ions reduces
the anion–anion repulsions across those ccp or hcp hole planes that do not contain cations
(Section 1.4.2) to the point where van der Waals forces are sufficient to stabilize the structure.
The other term that makes a small contribution to the lattice-formation energy is the zero-
point energy of the crystal. The zero-point energy, a quantum-mechanical concept that comes
from the Heisenberg uncertainty principle, is the vibrational energy of the lattice at absolute
zero temperature.
How important are these additional contributions to the lattice-formation energy? As
a rough guide, van der Waals forces make the lattice-formation energy more negative, and
 5.2 Atomic Orbitals 161

the magnitude of this contribution ranges from a few tenths of a percent to roughly 5% of the
total. The zero-point energy is generally less than 1.5% of the total and makes the lattice-
formation energy less negative. Because these two corrections are relatively small and
opposite in sign, they tend to cancel each other. In systems where these corrections are
large, the validity of the ionic model comes into question. More sophisticated approaches are
favored when a high degree of accuracy is important.

5.2 Atomic Orbitals

The concept of a covalent bond is ubiquitous in chemistry. A variety of approaches spanning
a range of complexities can be used to model covalent bonding. Simple approximations such
as Lewis structures and valence-shell electron-pair repulsion (VSEPR) provide useful esti-
mates of molecular geometry and bonding in many cases. At the other end of the spectrum,
advanced computational methods like density-functional theory can be used to make
increasingly accurate predictions for crystalline solids. It is, however, not always easy to
extract chemical insight from such approaches. Semi-empirical methods, such as extended
Hückel theory [5], occupy an intermediate ground and afford insight into the links between
local bonding interactions, crystal structures, and physical properties, a useful attribute that
merits their inclusion in this book. Models of covalent bonding are typically based on the
overlap of AOs. Therefore, we begin our treatment with a closer look at electrons and
orbitals.
We start with the simplest possible case, a one-electron atom, and treat its electron as
though it were a standing wave rather than a particle. The quantum-mechanical behavior of
the electron is described by the partial differential equation first proposed by Erwin
Schrödinger7 in 1926:
" #
2
^ ψðrÞ ¼  ℏ
H ∇ þ VðrÞ ψðrÞ ¼ EψðrÞ
2
(5.9)
2m

where the function ψ(r) describes the electron as awave, a wavefunction. The hamiltonian
operator, Ĥ, is a sum of two terms. The first term  ℏ2 =2m ∇2 represents the kinetic energy
of an electron of mass m (the Laplacian ∇2 is the second partial derivative of the wavefunc-
tion), and the second term V(r) represents the electron’s potential energy as a function of its
position vector r from the nucleus. The ψ(r) determines the energy E of the electron in
Equation (5.9). We will see later how also the size and shape of the electron cloud can be
calculated from ψ(r). Because each AO has a unique ψ(r), we will often refer to ψ(r) as the
orbital wavefunction, but, more precisely, it is the mathematical description of the standing
wave of an electron that occupies a given orbital; an electron wavefunction.

7
This is the non-relativistic, time-independent Schrödinger equation for a single particle moving in an electric field.
162 Chemical Bonding

That wavefunction is typically expressed in spherical polar coordinates as the product of

two functions:

ψn;ℓ;mℓ ¼ Rn;ℓ ðrÞYℓ;mℓ ðθ; ϕÞ (5.10)

In these coordinates, the radial part of the wavefunction, R, describes the variation of the
electron wave as a function of the distance r from the nucleus (here r is a scalar rather than
a vector). The angular part of the wavefunction, Y, describes the shape of the wavefunction
in terms of θ and ϕ. Equations for R(r) and Y(θ, ϕ) can be found in many physical chemistry
textbooks. We won’t go into the detailed mathematics here, but we will qualitatively look at
how the size, shape, and energy of the orbital depends upon the quantum numbers n, ℓ,
and mℓ that determine the wavefunction.
The principal quantum number, n, labels the quantized energy and largely determines the
size of the orbital. The allowed values of n, which are independent of the other quantum
numbers, are positive integers: 1, 2, 3, . . . The orbital angular-momentum quantum number
(sometimes called the azimuthal quantum number), ℓ, labels the quantized orbital angular
momentum and dictates the shape of the orbital. The allowed values of ℓ are the integers
ranging from 0 to n − 1. The values of ℓ are given specific letter designations: ℓ = 0
corresponds to an s orbital, ℓ = 1 to a p orbital, ℓ = 2 to a d orbital, and ℓ = 3 to an
f orbital. Orbitals with the same value of n are said to belong to the same shell, and those with
the same values of n and ℓ are said to belong to the same subshell (2s, 3d, 4p, etc.).
The magnetic quantum number, mℓ, labels the quantized orientation of the angular momen-
tum and dictates the orientation of the orbital. The allowed values of mℓ are the integers
between −ℓ and ℓ. This restriction limits the number of orbitals for each subshell to one s orbital
(mℓ = 0), three p orbitals (mℓ = −1, 0, +1), five d orbitals (mℓ = −2, −1, 0, +1, +2), and seven
f orbitals (mℓ = −3, −2, −1, 0, +1, +2, +3). It is customary to label the orbitals with subscripts
that denote the orientation of the orbital with respect to an arbitrary set of Cartesian axes. The
p orbitals are labeled px, py, and pz, where the subscripts identify the axis along which the lobes
of the orbital point, and the d orbitals are labeled dxy, dxz, dyz, dx2−y2, and dz2. A fourth
quantum number, the spin quantum number, ms, labels the quantized values of spin and can
take only two values; +½ and −½. The spin quantum number does not enter the wavefunction
directly, but through the Pauli exclusion principle8 it limits the maximum occupancy of an AO
to two electrons.
The angular part Y of the wavefunction is a constant when ℓ = 0, and thus the s orbitals are
spherically symmetric. This makes them convenient examples to illustrate how the radial
part of the wavefunction changes with increasing n. The wavefunctions for the 1s, 2s, and
3s orbitals are plotted as a function of r on the left-hand side of Figure 5.5. All three
wavefunctions drop off exponentially upon moving away from the nucleus. The
1s wavefunction is positive (ψ > 0) for all values of r, but the sign of the 2s wavefunction
8
The Pauli exclusion principle states that no two electrons in an atom can exist in the same quantum state and
therefore cannot have identical quantum numbers.
 5.2 Atomic Orbitals 163

radial distribution function, 4πr 2ψ 2(r )(Å−1)

1.5 1.2
1s 1s
wavefunction, ψ(r )(Å−3/2)

1.0 0.8

0.5 0.4

0.0 0.0

radial distribution function, 4πr 2ψ 2(r)(Å−1)

0.4
2s 2s
0.5
0.3
wavefunction, ψ(r )(Å−3/2)

0.3
0.2
radial node

0.1
0.1

-0.1 0.0
radial distribution function, 4πr 2ψ 2(r )(Å−1)

0.20 0.10
3s 3s
wavefunction, ψ(r )(Å−3/2)

0.15 radial nodes 0.08

0.10 0.06

0.05 0.04

0.00 0.02

-0.05 0.00
-8 -6 -4 -2 0 2 4 6 8 0 2 4 6 8 10
distance from nucleus, r (Å) distance from nucleus, r (Å)

Figure 5.5 The wavefunctions for the 1s, 2s, and 3s orbitals of a hydrogen atom (left) and the corresponding
radial distribution functions, 4πr2ψ2(r) (right). The 1s orbital is the ground state, while the other
wavefunctions represent excited states. A cutaway view of each orbital is shown. The dark and light shading
indicates a positive (ψ > 0) or negative (ψ < 0) sign of the wavefunction, respectively.9

9
See also Footnote 15 in this chapter for more discussion on wavefunction sign conventions used in this book.
164 Chemical Bonding

crosses over from positive to negative (ψ < 0) upon moving away from the nucleus. The point
(a surface in three dimensions) where ψ = 0 is called a node. When the node occurs because
the radial part of the wavefunction goes to zero, R(r) = 0, it is called a radial node. For a given
AO, the number of radial nodes is equal to n − ℓ − 1. For example, a 2s orbital has 1 radial
node, 5d orbital has 5 − 2 − 1 = 2 radial nodes.
The sign of wavefunction ψ is arbitrary and has no physical significance. In contrast,
the square of the wavefunction ψ2, which is called the probability density and is positive
everywhere, has a physical meaning. If we integrate the probability density ψ2 over the
surface of a sphere of radius r, we get the radial distribution function, 4πr2ψ2(r). This
function gives the probability of finding an electron inside an infinitesimally thin
hollow sphere at distance r from the nucleus. It is plotted for the 1s, 2s, and
3s orbitals on the right-hand side of Figure 5.5. We see that the most probable location
for the 1s electron is not at the nucleus, where ψ(r) reaches a maximum, but on
a sphere 0.5292 Å (the Bohr radius) away from the nucleus. Increasing the principal
quantum number n increases the number of local maxima in the radial distribution
function and the number of the radial nodes between them. For the ns orbitals plotted
in Figure 5.5 the number of radial nodes is equal to n − 1. The orbital also gets larger
as n increases. In multielectron atoms, the number of radial nodes impacts the attrac-
tion of the electron to the nucleus, and hence the energy of an orbital relative to other
orbitals in the same shell. This has an important impact on the periodic properties and
bonding tendencies of atoms. We will return to this point later.
Orbitals are often depicted by contours called boundary surfaces that enclose a specified
percentage of the electron density of that entire orbital (typically 90–95%, the rest is
outside). For s orbitals, the boundary surface is a sphere whose radius expands with
increasing n. The boundary surfaces of various AOs are shown in Figure 5.6. We see that
the p, d, and f orbitals have two, four, and six lobes, respectively. In each case, the
wavefunction changes sign on going from one lobe of the orbital to its neighboring
lobe.10 The lobes are separated by planes where the angular part of the wavefunction
goes to zero, Y(θ, ϕ) = 0. These “angular nodes” are referred to as nodal planes because of
their shape.11 For a given AO, the number of nodal planes is equal to ℓ. For example, the
d orbital shown in Figure 5.6 has two nodal planes (shown with dashed lines).
Nodal planes play an important role in determining the structures of molecules and solids.
Consider the 2px wavefunction in Figure 5.7. It takes a positive sign (ψ > 0) for positive x and
a negative sign (ψ < 0) for negative x, hence the yz plane is a nodal plane. In the boundary-
surface representation, the sign of the wavefunction is represented by shading, with dark
shading used for the lobe where ψ > 0 and light shading where ψ < 0.

10
Note that the boundary surfaces represent properties of ψ2, the sign of which is always positive, whereas the shading
refers to the sign of the underlying wavefunction ψ.
11
For some of the d and f orbitals, the surface where the angular part of the wavefunction goes to zero is a cone rather than
a plane; the dz2 orbital being one such example. We use the term nodal plane generically to apply to both planes and cones.
 5.2 Atomic Orbitals 165

Figure 5.6 Boundary

surfaces of the angu-
lar part of the wave-
function Y(θ, ϕ) for
selected orbitals. The
s orbital p orbital d orbital f orbital light and dark shad-
ℓ=0 ℓ=1 ℓ=2 ℓ=3
ing represent the sign
of the underlying
wavefunction. The
dashed lines represent
the nodal planes.

0.2
y Figure 5.7 The ampli-
2px tude of 2px wavefunc-
x
tion plotted along the
x-axis line. The con-
wavefunction, ψ(r) (Å−3/2)

0.1 x
tour shape shows the
nodal plane at x = 0,
nodal (yz) plane looking down the
z axis.
0.0

-0.1

-0.2
-8 -6 -4 -2 0 2 4 6 8
r (Å)

The size and energy of an orbital depend on the strength of the attraction between the
orbital’s electron(s) and the nucleus. In one-electron atoms, (H, He+, Li2+, and other so-
called hydrogenic atoms), the allowed energy levels are given by:

Z 2 hcRH
En ¼  (5.11)
n2

where Z is the charge of the nucleus, h is Planck’s constant, c is the speed of light, and RH is
the Rydberg constant. Note that the energy of the orbital only depends upon the nuclear
charge Z and the value of n.
166 Chemical Bonding

3
core

radial probability, 4πr2 ψ 2(r) (Å−1)

3p
electrons
3s

0.04

2
3d
3d
0.00
0 0.2 0.4

3p
1

3s

0
0 1 2 3 4 5 6
distance from nucleus, r (Å)

Figure 5.8 An illustration of the penetration the 3s, 3p, and 3d orbitals into the region occupied by the
core electrons (1s, 2s, and 2p orbitals). The inset shows the region very close to the nucleus.

In a multielectron atom, the charge of the nucleus experienced by a given electron is

partially screened by the other electrons. The reduced positive charge felt by the electron is
called the effective nuclear charge, Z*. For electrons in a given orbital, Z* depends in part
upon the penetration of that orbital’s wavefunction into the region close to the nucleus,
a region dominated by the wavefunctions of core orbitals, as shown in Figure 5.8. Notice
how the radial probability function of the 3s orbital has a small peak close to the nucleus,
whereas the first peak for 3p is located further away, and the electron density of
the 3d orbitals is even lower near the nucleus. As a result, the effective nuclear charge
decreases upon moving from 3s to 3p to 3d orbitals, and the orbital energies follow the
sequence 3s < 3p < 3d. This effect is sufficiently large that by the time valence electrons start
populating the 3d orbitals, the 3s and 3p orbitals behave as core orbitals. The same relation-
ship holds for orbitals with n = 4 or 5. The effect is even more dramatic for the f orbitals.

5.2.1 Energies of Atomic Orbitals

To apply bonding principles in a semiquantitative manner, we need to develop a feel for the
properties of AOs. In this subsection and the one that follows, we will examine the periodic
trends in the energies and sizes of AOs.
Table 5.2 shows calculated energies (in electronvolts12, eV) for the valence orbitals across
most of the periodic table. Not surprisingly, the general periodic trends in orbital energies are
similar to those seen for electronegativity, namely that the valence-orbital energies become
increasingly negative upon moving up and to the right. However, there are disruptions in the

12
An electronvolt is the amount of kinetic energy gained by a single electron accelerating from rest through an electric-
potential difference of one volt. Accounting for the elementary charge of an electron, 1 eV = 1.602176634×10−19 J
exactly.
 5.2 Atomic Orbitals 167

Table 5.2 Valence-orbital energies (in eV) of the elements, obtained from relativistic Dirac–
Fock calculations [6].*

H He
1s −13.5 −24.8

Li Be B C N O F Ne
2s −5.3 −8.4 −13.4 −19.2 −26.0 −33.8 −42.6 −52.3
2p −8.4 −11.0 −13.7 −16.6 −19.7 −23.0
Na Mg Al Si P S Cl Ar
3s −4.9 −6.8 −10.7 −14.7 −19.2 −24.0 −29.1 −34.7
3p −5.7 −7.5 −9.5 −11.5 −13.7 −16.0

K Ca Sc Ti Zn Ga Ge As Se Br Kr
4s −4.0 −5.3 −5.7 −6.0 −8.1 −11.7 −15.4 −19.2 −23.3 −27.6 −32.1
4p −5.6 −7.3 −8.9 −10.6 −12.4 −14.3
3d −9.1 −10.7 −20.6 −31.5 −43.4 −56.3 −70.3 −85.3 −101
Rb Sr Y Zr Cd In Sn Sb Te I Xe
5s −4.0 −4.9 −5.4 −5.8 −7.6 −10.7 −13.8 −16.9 −20.2 −23.7 −27.3
5p −5.3 −6.7 −8.1 −9.6 −11.1 −12.6
4d −6.3 −7.9 −19.5 −27.4 −35.5 −44.0 −53.0 −62.3 −72.2
Cs Ba La Hf Hg Tl Pb Bi Po At Rn
6s −3.5 −4.4 −4.9 −6.5 −8.9 −12.1 −15.3 −18.5 −21.9 −25.3 −28.9
6p −5.2 −6.7 −8.1 −9.5 −11.0 −12.5
5d −6.4 −6.5 −16.5 −23.0 −29.4 −35.9 −42.7 −49.6 −56.8

* The effects of spin–orbit splitting on the energies of the p and d orbitals have been averaged.

smooth periodicity that merit further comment. This is particularly true of the vertical trends
in the s-orbital energies of the p-block elements. Firstly, note that upon moving from
the second period (B–Ne) to the third period (Al–Ar), the s-orbital energy becomes less
negative by a significant amount, while upon moving from the third to the fourth period
(Ga–Kr), the s-orbital energy change is smaller, and, in some cases (Ga, Ge), the 4s orbital of
the fourth-period element has a lower energy than the 3s orbital of its third-period neighbor.
This anomaly can be attributed to the incomplete shielding of the 4s orbital when the
d subshell is filled for the first time. A similar discontinuity is seen upon moving from the
fifth period (In–Xe) to the sixth period (Tl–Rn), where the unexpectedly deep (i.e. large and
negative) orbital energies of s orbitals of the sixth-period elements originate primarily from
relativistic effects.13
Another important feature to glean from Table 5.2 is how significantly the angular-
momentum quantum number ℓ influences the orbital energies. The valence s orbitals are

13
When the charge of the nucleus is large enough that the velocity of an electron in its vicinity approaches the speed of
light, relativistic effects increase the electron’s mass. This effect can effectively be neglected for light elements, but it
becomes more significant as the charge of the nucleus increases, and is largest for electrons in s orbitals, because they
have a non-zero probability at the nucleus.
168 Chemical Bonding

always much deeper in energy than the p orbitals. This is particularly true for oxygen and
fluorine. We can also see a significant change in the energy of the (n − 1)d orbitals relative to
the ns orbitals upon moving from left to right across a period. At the beginning of the
transition-metal block, the energies of the (n − 1)d and ns orbitals are similar; by the end of
the series, the (n − 1)d orbital energy is considerably more negative than the ns orbital energy.
As we continue into the p block of the periodic table, the energy of (n − 1)d orbitals becomes
so negative that these orbitals effectively act as core orbitals.

5.2.2 Sizes of Atomic Orbitals

The wave nature of electrons makes it impossible to precisely define the size of an orbital. The
boundary surfaces shown in Figure 5.6 are one way to approximate the size of an orbital.
Another approach is to determine the radius at which the radial distribution function reaches
a maximum, rmax. Values of rmax for valence orbitals across the periodic table are given in
Table 5.3. Many of the periodic trends seen for orbital energies also hold for orbital radii.
Upon moving up or right, the size of the valence orbitals decreases.
The relative sizes of the valence orbitals for a given atom also contain important informa-
tion. Notice in Table 5.3 that for each value of the orbital quantum number ℓ, the orbitals
belonging to the lowest-energy subshell (1s, 2p, 3d, and 4f) are particularly compact. This has
some significant implications for bonding. For the second-period elements of the p block (B–
Ne), the 2p orbitals are comparable in size to the 2s orbitals, which is not the case for heavier
p-block elements where the p orbitals are substantially larger than the s orbital. As a result,
atoms from the second period form short σ bonds with each other, which facilitates forma-
tion of π bonds (Section 5.3.4).
Notice that for a given atom the (n − 1)d orbitals are much smaller than the ns and np
orbitals, and as a result they interact to a lesser extent with the surrounding atoms. Their
small size is one reason why the (n − 1)d orbitals are often found to be partially filled in
transition-metal compounds. This leads to several useful properties explored later in the
book, including color (Chapter 7), cooperative magnetism (Chapter 9), and metal–insulator
transitions driven by changes in external conditions (Chapters 10 and 11). The small size of
the (n − 1)d orbitals, with respect to the ns and np orbitals of the same atom, is most
pronounced for 3d electrons because this is the first d shell to fill and thus it experiences
less shielding from the nucleus than electrons populating the 4d and 5d orbitals.
Although the lanthanoids and actinoids are not shown in Table 5.3, we may conclude that
the contracted nature of the 4f and 5f orbitals is even more dramatic than for the d orbitals.
For example, the valence-orbital values of rmax for cerium are 2.17 Å for 6s, 1.12 Å for 5d,
and 0.37 Å for 4f. Consequently, the 4f orbitals have minimal interactions with the orbitals of
surrounding atoms and behave more like core orbitals than valence orbitals. The lack of
interaction between the 4f orbitals and the neighboring atoms is responsible for the sharp
lines seen in optical absorption and emission spectra, which make the lanthanoid ions useful
as luminescence centers (Chapter 7). This also makes magnetic exchange interactions
 5.3 Molecular-Orbital Theory 169

Table 5.3 Distances (in Å) at which the radial distribution function reaches a maximum value
according to relativistic Dirac–Fock calculations [6].

H He
1s 0.53 0.30

Li Be B C N O F Ne
2s 1.64 1.09 0.81 0.65 0.54 0.46 0.41 0.36
2p 0.84 0.64 0.52 0.44 0.38 0.34
Na Mg Al Si P S Cl Ar
3s 1.79 1.37 1.11 0.95 0.84 0.75 0.68 0.62
3p 1.42 1.15 0.98 0.85 0.76 0.69

K Ca Sc Ti Zn Ga Ge As Se Br Kr
4s 2.29 1.83 1.71 1.61 1.18 1.04 0.95 0.87 0.81 0.76 0.72
4p 1.39 1.19 1.06 0.96 0.89 0.82
3d 0.60 0.53 0.30 0.29 0.27 0.25 0.24 0.23 0.22
Rb Sr Y Zr Cd In Sn Sb Te I Xe
5s 2.45 2.01 1.85 1.74 1.30 1.24 1.09 1.03 0.97 0.92 0.87
5p 1.56 1.37 1.24 1.15 1.07 1.01
4d 0.96 0.85 0.52 0.51 0.47 0.45 0.43 0.41 0.40
Cs Ba La Hf Hg Tl Pb Bi Po At Rn
6s 2.72 2.27 2.11 1.78 1.22 1.13 1.07 1.01 0.97 0.93 0.89
6p 1.59 1.40 1.28 1.20 1.13 1.07
5d 1.19 0.88 0.61 0.59 0.57 0.54 0.53 0.51 0.49

between lanthanoid ions weak, which explains why cooperative magnetic ordering of such
ions typically occurs far below room temperature if at all (Chapter 9).

5.3 Molecular-Orbital Theory

Having reviewed the properties of AOs, we now consider what happens when AOs interact to
form molecular orbitals. The goal of this treatment is to provide the basic qualitative knowledge
needed to interpret and construct simple molecular-orbital (MO) diagrams. The treatment is
largely non-mathematical and does not require prior knowledge of group theory. In Chapter 6,
we will build on this foundation to model the electronic structures of extended solids.

5.3.1 Homonuclear Diatomics: H2+ and H2

To illustrate the general principles of covalent bonding, we begin by considering the simplest
possible molecule, H2+. To do so we need to describe the behavior of an electron shared by two
nuclei. This is done by defining a wavefunction ψMO that represents a molecular orbital (MO).
MOs are similar to the AOs we’ve already encountered, with the important distinction that they
170 Chemical Bonding

can extend over the entire molecule. One of the most common approaches to defining ψMO is to
treat it as a linear combination of atomic orbitals (LCAO):

ψMO ¼ c1 ψAOð1Þ þ c2 ψAOð2Þ (5.12)

where ψAO(1) and ψAO(2) are AO wavefunctions on atoms 1 and 2, respectively, while c1 and c2
are numerical coefficients. In the Hþ 2 case, ψAO(1) is a 1s orbital on the first hydrogen atom
and ψAO(2) is a 1s orbital on the second atom. The MO wavefunction must meet three criteria;
it must be finite everywhere, single valued, and ψ2 must have an integral. Since the probabil-
ity of finding an electron when integrated over all space must be unity, the coefficients c1 and
c2 are chosen so that the MO wavefunction is normalized, which can be expressed mathem-
atically as:

2ðπ ð
πð∞

ψψ drdθdϕ ¼ 1 (5.13)

0 0 0

The MO diagram for Hþ 14

2 is shown in Figure 5.9. Each orbital is represented by a horizontal
line and each electron by a vertical arrow. The AOs are shown on the sides of the diagram, and
the MOs are shown in the center of the diagram. The vertical axis is an energy scale. A basic
principle of MO theory is that the number of MOs in a molecule is equal to the number of AOs
that combine to form them. In Hþ 2 , there are two MOs: ψ+, which results from constructive
interference of the two H 1s orbitals (their wavefunctions sum upon overlap); and ψ−, which
results from destructive interference of these AOs (their wavefunctions subtract upon overlap).
The ψ+ MO is stabilized with respect to the H 1s orbital energy because constructive interfer-
ence of the two AO wavefunctions increases the electron density between the two positively
charged nuclei, leading to an attraction between the two atoms. This orbital is called a bonding
molecular orbital. Populating this MO stabilizes the molecule with respect to the energy of two
isolated atoms. In the antibonding molecular orbital, ψ−, the electron density between the nuclei
is lowered, and populating ψ− destabilizes the molecule.
The AO interactions that make up each MO are represented by sketches where shading
gives the sign of the AO wavefunction after it is multiplied by the coefficient that determines
its contribution to the MO,15 and its relative size represents the magnitude of that coefficient.
When two otherwise identical H 1s AOs have opposite signs (visually represented with
opposite shading) in an MO, we say the two orbitals have the opposite phase. The exact

14
The energies and orbital coefficients in this and following figures were calculated using extended Hückel theory as
implemented in the Caesar 2.0 software suite.
15
In these sketches, the shading refers to the sign of the atomic orbital wavefunction at the periphery of the atom (i.e. beyond
the outermost radial node) where the interaction with orbitals from neighboring atoms is most significant. For an s orbital,
the sign is the same in all directions, while for p, d, and f orbitals the sign alternates from lobe to lobe. The choice of that
sign is arbitrary, what matters is the relative sign as we move from orbital lobe to lobe and from atom to atom. This relative
sign includes the sign of any numerical coefficient that multiplies the orbital wavefunction.
 5.3 Molecular-Orbital Theory 171

Figure 5.9 Calculated MO diagram for the H2+ molecule. The size of the circle used to represent each AO
is proportional to its coefficient. Shading indicates the sign the AO wavefunction acquires from its
coefficient in the MO wavefunction; ψ > 0 is white and ψ < 0 gray. Arrows represent electrons.

values of the orbital energies and coefficients depend upon the level of theory employed, but
the general features of the MO diagram will be the same regardless of the sophistication of
the theoretical treatment.
The rules for filling MOs with electrons are the same as for AOs. The MOs are filled in
order of increasing energy (Aufbau principle) and each MO can hold two electrons of
opposite spin (Pauli exclusion principle). When MOs are degenerate (i.e. they have the
same energy), the most stable configuration is the one that produces the largest number of
parallel electron spins (Hund’s first rule).
Analytical expressions for the energy and wavefunction for each MO of H2+ are as follows:

H11 þ H12 1
Eþ ¼ ψþ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðψ1 þ ψ2 Þ
1 þ S12 2ð1 þ S12 Þ
H11  H12 1
E ¼ ψ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðψ1  ψ2 Þ (5.14)
1  S12 2ð1  S12 Þ

The wavefunctions and their energies are defined by three parameters: the coulomb integral, H11;
the interaction integral, H12; and the overlap integral, S12. The coulomb integral H11 is simply the
energy of the AO in question, in this case the energy of a hydrogen 1s orbital (−13.5 eV from
Table 5.2). The interaction integral H12 is a measure of the interaction energy that results from
172 Chemical Bonding

the overlap of AOs. It largely determines the magnitude of the energy splitting between the
bonding and antibonding orbitals. The amount of spatial overlap between orbitals is quantified
by the overlap integral S12, whose magnitude depends on the interatomic distance as well as the
size and symmetry of the overlapping orbitals. Numerical values for these variables can be
computed. However, for our purposes, the general features of the MO diagram are more
important than the exact energies of the MOs. The bonding (ψ+) and antibonding (ψ−) MO
wavefunctions are plotted along the internuclear axis in Figure 5.10.
If we add another electron to create a neutral H2 molecule, the computational complexity
of the problem increases due to repulsive electron–electron interactions that are difficult to

ψ1
ψ−= 0.94(ψ1−ψ2)

nodal plane
wavefunction ψ

–ψ2

ψ1 ψ2
wavefunction ψ

ψ+= 0.59(ψ1+ψ2)

-4 -3 -2 -1 0 1 2 3 4
z (Å)

Figure 5.10 Bonding-MO ψ+ (bottom) and antibonding-MO ψ− (top) along the internuclear axis (z) for
an H2+ molecule. The AO wavefunctions ψ1 and ψ2 are shown in gray, and the boundary surfaces of the
MOs are shown in the upper left corner.
 5.3 Molecular-Orbital Theory 173

model. We’ll return to this point later, but for now we will ignore electron–electron inter-
actions. We can use our simple MO diagram to make some predictions about bonding as
electrons are added to the molecule. H2 has two electrons, and this leads to a doubly occupied
bonding orbital ψ+ and an empty antibonding orbital ψ− (electron configuration ψ+2ψ−0).
This electron count produces a single two-electron bond; the H–H bond order is hence 1.16
Increasing the number of electrons in the bonding MO favors a high degree of overlap
between the two AOs. For spherically symmetric 1s orbitals, this is achieved by reducing the
H–H distance. The collapse of one atom on top of the other would give maximal orbital
overlap, but the coulombic repulsion between the hydrogen nuclei prevents this from
happening.17 The equilibrium bond distance represents the most favorable balance of
these two competing interactions.
If we move to an H 2 molecule, the electron count becomes three and the electron
configuration is ψ+ ψ− . We have now partially filled the antibonding MO (ψ−). This
2 1

weakens the bonding (the bond order is now ½), leading to a reduction in the bond
dissociation energy and an increase in the equilibrium bond distance. Intuitively, we would
expect Hþ 
2 and H2 to have bonds of equal strength, but this is not quite correct. Because the
overlap integral S12 is always positive, the energy of the antibonding orbital ψ− is destabilized
more than the bonding orbital ψ+ is stabilized; see Equation (5.14). This is a general property
of MOs that can have implications. In this case it means that the bond in Hþ 2 is somewhat

stronger than the bond in H2 .

5.3.2 The Heteronuclear Diatomic Case: HHe

Let’s increase the complexity of the problem by considering the heteronuclear diatomic
molecule, HeH. The MO diagram for HeH is shown in Figure 5.11. We now must take into
account the different energies of the H 1s and He 1s orbitals. This impacts the bonding in
several ways. Firstly, the AO coefficients in each wavefunction, c1 and c2, are no longer
equal. For the bonding MO, the He 1s orbital coefficient (0.89) is larger than the H 1s orbital
coefficient (0.24). The opposite relationship holds for the antibonding MO. In each case, the
AO closer in energy to the MO makes the larger contribution. This is schematically repre-
sented by the relative sizes of the AOs shown in Figure 5.11. The inequality in the coefficients
c1 and c2 reflects the fact that the electrons are not equally shared, the electron density is
shifted toward the more electronegative He.
The energy difference between overlapping orbitals not only affects the wavefunction, it also
impacts the MO energies. For the sake of comparison, the same bond distance (1.06 Å) was
assumed for the hypothetical HHe molecule in Figure 5.11 as for H2+ in Figure 5.9, yet the
stabilization of the bonding MO in HHe with respect to the He 1s orbital (0.9 eV, see Figure

16
Bond order is calculated as ½[(number of e– in bonding MOs) − (number of e– in antibonding MOs)].
17
For multielectron atoms, repulsions between core electrons play the primary role in limiting the interatomic
distance.
174 Chemical Bonding

Figure 5.11 Calculated MO diagram for the HeH molecule, Shading represents the sign of the AO
wavefunctions when they make up the MOs, ψ > 0 (white) and ψ < 0 (gray).

5.11), is less than the corresponding stabilization of the bonding MO in H2+ (3.1 eV, see Figure
5.9). This illustrates a general feature of MO theory; the stabilization of the bonding MO and
destabilization of the antibonding MO depend upon both the spatial and energetic overlap of
parent AOs (on both S12 and H12).
Although the covalent stabilization of the bonding MO is smaller in the case of HHe, this
does not mean the energy gap between bonding and antibonding orbitals, Δ = E− − E+, is
smaller. Extended Hückel calculations give Δ values of 10.9 eV for H2+ and 17.5 eV for HHe.
The energy gap for HHe is larger because the separation between the two MOs contains an
ionic contribution. The ionic contribution, EI, is the difference in orbital energies of the
H and He 1s orbitals, EI = EH − EHe = −13.5 − (−24.8) = 11.3 eV, while the covalent
contribution accounts for the remaining 6.2 eV of the gap between MOs. In Section 7.5, we
will see that both covalent and ionic contributions must be considered to account for changes
in the band gaps of compound semiconductors.

5.3.3 Orbital Overlap and Symmetry

For the spherically symmetric s orbitals, the overlap integral S depends only upon the radial
portion R(r) of the wavefunction and the interatomic distance. With orbitals that are not
spherically symmetric (p, d, and f orbitals), we must also consider the angular geometry. For
example, consider the overlap integral between a p orbital (see Figure 5.7) and an s orbital on
 5.3 Molecular-Orbital Theory 175

Figure 5.12 Overlap integral S as a function of the angle θ between two orbitals.

a neighboring atom shown in Figure 5.12. In this case, S depends not only upon the
interatomic distance, but also upon the angle θ between the two orbitals. When θ = 90, the
overlap integral is zero for any value of the interatomic distance because the s orbital
overlaps equally with both lobes of the p orbital. The overlap with the lobe of the same
sign yields constructive interference, while the overlap with the opposite-sign lobe is equal in
magnitude, but destructive. This gives an overlap integral S = 0, and the two orbitals are said
to be orthogonal. Combinations of other orbitals can give rise to different angular depend-
encies, such as the overlap between a d orbital (see Figure 5.6) and an s orbital, shown in
Figure 5.12. In general, when S = 0 we will say that orbital mixing is symmetry forbidden.
Such geometries can often be identified visually.

5.3.4 Combination of σ and π Bonding: O2

The final diatomic molecule that we will consider is O2. We now have five AOs per atom to
consider: 1s, 2s, 2px, 2py, 2pz. The 1s orbitals need not be treated explicitly because the large
effective nuclear charge they experience confines their wavefunctions to the region close to
the nucleus, and their overlap with orbitals on the neighboring atom is negligible. As a rule,
we will omit core orbitals when constructing MO diagrams.
The approximate MO diagram for O2 is shown in Figure 5.13. The diagram is approximate
because mixing between MOs formed from the 2s and the 2p orbitals is neglected. We will
return to this point shortly. The vertical separation between the 2s and 2p orbitals in the MO
diagram corresponds to the differences in orbital energy previously discussed (see Table 5.2).
The two lowest-energy MOs, σ(2s) and σ*(2s), are the respective bonding and antibonding
176 Chemical Bonding

Figure 5.13 MO diagram for the O2 molecule (along z), with s–p mixing neglected.

orbitals that arise from overlap of the O 2s orbitals. They are analogous to the MOs already
discussed for H2. The asterisk as a superscript indicates an antibonding orbital.
The overlap of O 2p orbitals depends on the orientation of each orbital with respect to the
internuclear axis, which is defined to run parallel to the z direction by convention. The O 2pz
orbitals point directly at each other and overlap in a “head-on” fashion forming bonding and
antibonding orbitals, σ(2p) and σ*(2p). Here, the bonding MO σ(2p) is obtained by subtracting
wavefunctions of the 2pz orbitals on neighboring atoms, σ(2p) = c1ψ(2pz)1 − c2ψ(2pz)2, where the
coefficients c1 = c2 are chosen to normalize the wavefunction. The minus reverses the signs of the
two lobes of the 2pz wavefunction on atom 2 in such a way that the two AOs overlap
constructively between the nuclei. Whenever we encounter orbitals with nodal planes (p, d, and
f orbitals), multiplying by a negative coefficient inverts the sign of the AO wavefunction at all
points in space, which reverses the shading of each lobe. Whether the orbital has a nodal plane(s)
or not, upon multiplying by a negative coefficient we can say that it has the opposite phase of an
otherwise identical orbital whose coefficient is positive. The antibonding MO σ*(2p) is formed by
adding the two wavefunctions, which leads to destructive interference between the nuclei.
The O 2px and O 2py orbitals are aligned perpendicular to the internuclear axis and overlap
in a “side-on” fashion. This overlap produces bonding and antibonding orbitals π(2p) and
π*(2p).18 Because the overlap of 2px orbitals is the same as the overlap of 2py orbitals, these
orbitals are doubly degenerate (i.e. two MOs with the same energy). The splitting between the σ

18
The number of nodal planes associated with the bonding MO increases from zero for a σ MO, to one for a π MO, to
two for a δ MO (an MO formed by side-on overlap of d orbitals).
 5.3 Molecular-Orbital Theory 177

σ* s–p mixing

− more antibonding
=
σ* (2p) destabilized, E ↑

+ less antibonding
=
σ* (2s) stabilized, E ↓

σ s–p mixing

− = less bonding
σ (2p) destabilized, E ↑

+ = more bonding
σ (2s) stabilized, E ↓

Figure 5.14 MOs of the same symmetry mix: σ(2p) with σ(2s) and σ*(2p) with σ*(2s).

(2p) and σ*(2p) orbitals is larger than the splitting between the π(2p) and π*(2p) because the
head-on overlap of 2pz orbitals is larger than the side-on overlap of 2px (2py) orbitals; H12(σ) >
H12(π) and S12(σ) > S12(π).
How does s–p mixing come into the picture? The π(2p) and π*(2p) MOs do not mix with the
MOs formed by the 2s orbitals because for every constructive overlap there is an equivalent
destructive overlap that cancels it out. However, the σ(2s) and σ(2p) MOs have the same
symmetry and can mix, as can the σ*(2s) and σ*(2p) MOs, both interactions are illustrated in
Figure 5.14. The s–p mixing stabilizes the lower-energy MOs with predominant 2s character
and destabilizes the higher-energy MOs with predominant 2p character. Whenever MOs of
the same symmetry mix, the lower-energy MO is stabilized and the higher-energy MO is
destabilized.
A more accurate MO diagram for O2, where s–p mixing has been included, is shown in
Figure 5.15. The effect of mixing is most evident near the middle of the MO diagram where
the energy of the σ*(2s) MO (now labeled 1σ*) is lowered and that of the σ(2p) MO (now
labeled 2σ) is raised. The amount of s–p mixing depends on a number of factors including the
energy separation of the 2s and 2p orbitals and the overlap integral. The s–p mixing is
sufficiently strong that the 2σ orbital rises above the π-bonding MOs of the diatomic
molecules from Li2 to N2, while for O2 and F2 it remains just below the π MOs.

5.3.5 Symmetry-Adapted Linear Combinations (SALCs)

In the MO diagrams of diatomic molecules, the AOs are placed on the left- and right-
hand sides of the diagram while the molecular orbitals are in the middle. This approach
does not directly translate to larger polyatomic molecules. However, we can retain this
178 Chemical Bonding

Figure 5.15 MO diagram for O2 without (left) and with (right) s–p mixing.

approach if we divide the molecule into two fragments and build the MO diagram from
them. The most common way to do this is to consider the outer atoms, or ligands, as
one fragment and the central atom as the other fragment. The task of combining the
AOs of the ligands into a new set of orbitals, the ligand-group orbitals, is based on
symmetry. The resulting ligand-group orbitals are called symmetry-adapted linear com-
binations or SALCs for short.
To rigorously construct SALCs requires knowledge of group theory, which is beyond the
scope of our treatment. However, once derived, the interactions between ligand-group
SALCs and the central atom can be visually estimated from simple symmetry considerations,
as demonstrated in Figure 5.16. The ligand SALCs in Figure 5.16 are shown for two, four,
and six hydrogens around a central atom possessing s and p valence orbitals, in linear,
tetrahedral, and octahedral geometry. In the linear geometry, one SALC can have a net
overlap with the s orbital of the central atom, and the other SALC with the pz orbital. The px
and py orbitals on the central atom are orthogonal to both SALCs and therefore cannot mix
to form bonding/antibonding MOs.
Moving from linear to tetrahedral geometry, the number of SALCs follows the
number of ligands. The number of orbitals on the central atom and the number of
SALCs are now equal, and all orbitals on the central atom have a ligand SALC with
the appropriate symmetry for bonding. For an octahedral molecule, there are six
SALCs, which a symmetry analysis divides into three groups. The uppermost SALC
in Figure 5.16 does not have a nodal plane and has the correct symmetry to interact
 5.3 Molecular-Orbital Theory 179

ligand SALCs

central linear tetrahedral octahedral

atom (BeH2) (CH4) (SH6)

0 nodal
s
planes

pz

px 1 nodal
plane

py

2 nodal
planes

2 nodal
cones

Figure 5.16 Symmetries of s and p orbitals on a central atom (left) and ligand SALCs for linear, tetrahedral,
and octahedral arrangements of hydrogen ligands.

with the s orbital of the central atom. The next three SALCs possess a single nodal
plane, and each of them has the correct symmetry to interact with one of the p orbitals
on the central atom. There are no orbitals on the central atom to interact with the
lowest two SALCs. However, as we will see later, when the central atom is a transition
metal, these SALCs have the appropriate symmetry to interact with two of the five
d orbitals on the central atom.
The SALCs shown in Figure 5.16 are relevant for more than just molecules containing
H as a ligand. The symmetries and mixing shown in this figure are representative of σ-
bonding interactions between the central atom and ligands in any linear, tetrahedral, or
octahedral molecule containing a main-group central atom. For example, the SALCs shown
in Figure 5.16 are sufficient to describe the bonding in SF6. Armed with knowledge of how
the SALCs mix with orbitals on the central atom, we will now look at the MO diagrams of
some simple polyatomic molecules.

5.3.6 Simple Polyatomic Molecules: BeH2 and CH4

The MO diagram for the triatomic molecule BeH2 is shown in Figure 5.17. As discussed in
the preceding section, the Be 2s and 2pz orbitals interact with the two ligand SALCs to form
two bonding MOs and two antibonding MOs. Their subscripts “g” and “u” are shorthand
180 Chemical Bonding

for gerade and ungerade, respectively,19 and refer to symmetry; for molecules with an
inversion center, the subscript “g” (e.g. σg) is given to MOs that are invariant on inversion,
while the subscript “u” is used for MOs whose wavefunction changes sign on inversion. The
Be 2px and 2py orbitals do not have the correct symmetry to overlap with either of the
H 1s SALCs (S = 0). These orbitals remain strictly nonbonding with 100% Be character.
Thus, we see that MOs are not always delocalized, sometimes they are localized on a single
atom.
The MOs closest to the crossover from filled to empty are typically the most important for
reactivity and properties. They are called the highest-energy occupied molecular orbital
(HOMO), and the lowest-energy unoccupied molecular orbital (LUMO), respectively. The
lowest-energy optical excitations correspond to electronic transitions from the HOMO to the
LUMO. Chemical oxidation corresponds to removal of an electron from the HOMO, while
reduction puts an electron into the originally lowest unoccupied molecular orbital, LUMO.
Many chemical reactions involve interactions between the HOMO on one molecule and the
LUMO on another. Collectively, the HOMO and LUMO are referred to as frontier orbitals.
For BeH2, the HOMO is the bonding σu MO formed from the interaction between the Be 2pz
orbital and the appropriate ligand SALC, while the LUMO is the doubly degenerate set of
nonbonding Be 2p orbitals.

Figure 5.17 MO diagram for BeH2.

19
They originate in group theory. Gerade and ungerade are German for even and odd.
 5.3 Molecular-Orbital Theory 181

t2*

energy a 1*

C 2p t2
a1

t2 H 1s
C 2s SALCs

a1

Figure 5.18 MO diagram for CH4.

The MO diagram for CH4 is shown in Figure 5.18. The C 2s orbital can overlap with the
SALC where all four hydrogen AOs have the same phase to form bonding (a1) and
antibonding (a1*) MOs. Each of the C 2p orbitals interacts with one of the SALCs that
possess a single nodal plane. These interactions give rise to a triply degenerate set of bonding
(t2) and antibonding (t2*) MOs.20 As with BeH2, there is no mixing of the 2s and 2p orbitals
on the central carbon atom. The well-known valence-bond concept of sp3 hybridization in
tetrahedral molecules is useful in that it tells us that the s and all three p orbitals on the central
atom are involved in forming σ bonds. However, the picture of quadruply degenerate
bonding orbitals that is sometimes inferred from this description is not consistent with the
orbital energies obtained from calculations or seen in photoelectron spectra.

5.3.7 Conjugated π Bonding: C6H6

A common feature of many organic functional materials is the presence of a conjugated
π-bonding system. The archetypical example is benzene, C6H6, a highly symmetric planar
molecule possessing 30 valence electrons. The in-plane C 2s, 2px, 2py, and H 1s orbitals

20
We use Mulliken symmetry labels in MO diagrams: Triply degenerate orbitals are labelled t, doubly degenerate
orbitals e, and singly degenerate orbitals either a or b, depending upon the symmetry versus the principal rotation
axis. A subscript of 1 means the MO is symmetric with respect to a twofold rotation axis perpendicular to the
principal axis, while a 2 means it is asymmetric with respect to this axis. A prime symbol signifies an MO that is
symmetric with respect to a mirror plane perpendicular to the principal rotation axis, a double prime indicates the
MO is asymmetric with respect to such a mirror plane.
182 Chemical Bonding

b2g

e2u

energy
e1g

a2u

Figure 5.19 The overlap of C 2pz orbitals to form MOs with π character in C6H6 as viewed nearly
perpendicular to the plane of the molecule (the H atoms do not contribute to the π bonding and are
omitted). The nodal planes for each MO are marked with dashed lines.

overlap to form 12 occupied σ-bonding MOs and 12 unoccupied σ-antibonding MOs. The 24
electrons that occupy bonding MOs are responsible for the network of C–C and C–H σ
bonds that hold the molecule together. The remaining C 2pz orbitals are oriented perpen-
dicular to the plane of the molecule. These orbitals do not contribute to the σ bonding but
they do interact with each other in a π fashion to form an additional six MOs. Just as was the
case with the O2 molecule, the energies of the MOs with π character fall between the σ and σ*
MOs. These π orbitals, which play a major role in determining the chemical and physical
properties of benzene, merit a closer look.
The MO diagram for the π interactions in benzene is shown in Figure 5.19. As required, six
MOs result from the interaction between six C 2pz orbitals. Not surprisingly, the lowest-
energy MO is the one where all six 2pz orbitals have the same phase so that each carbon has
bonding π interactions with its neighbors. It should also be intuitive that the highest-energy
MO is the one where every 2pz orbital wavefunction is out of phase with its nearest neighbors,
leading to antibonding interactions between neighboring carbon atoms. At first glance, the
relative energies of the remaining four MOs may be less obvious. However, we can correlate
the energies of each MO with the number of nodal planes it possesses. The lowest-energy MO
does not have a nodal plane oriented perpendicular to the plane of the molecule.21 Next come
two degenerate MOs, each with one nodal plane, as indicated by the dotted lines in Figure
5.19. The next set of MOs possesses two nodal planes, and the highest-energy MO has three

21
In this discussion, we ignore the nodal plane that coincides with the plane of the molecule, as all six MOs have this
nodal plane in common.
 5.3 Molecular-Orbital Theory 183

nodal planes. The correspondence between the number of nodal planes and the energy should
not be surprising. The presence of a nodal plane between two atoms signals destructive
interference between AO wavefunctions and is characteristic of an antibonding interaction.
A key feature of the MO diagram for benzene is the delocalized nature of the π bonding.
Even in large conjugated molecules, the frontier orbitals span the entire molecule (neglecting
the H atoms) as they do in benzene. Consequently, removing (adding) electrons through
oxidation (reduction) introduces charge carriers that can move from one end of the molecule
to the other. As we will see in later chapters, this means that conjugated organic molecules
can exhibit properties where movement (conductivity) of electrons and/or long-range coup-
ling of electron spins (cooperative magnetism) occur.

5.3.8 Transition-Metal Complexes: [CrCl6]3− and [CoCl4]2−

Many of the functional materials that we will discuss later in the book contain transition
metals. Their optical, electrical, and magnetic properties are often dictated by the energy
levels and occupation of the five d orbitals, whose orientations are shown in Figure 5.20.
Here we focus on the two most common coordination geometries for transition-metal
complexes and compounds, the octahedron and the tetrahedron.
Let’s begin by considering the octahedral anion [CrCl6]3−. The orbital energies of the
chromium 3d (−13.5 eV) and 4s (−6.6 eV) orbitals are much better matched to the Cl
3p orbitals (−13.7 eV) than they are to Cl 3s orbitals (−29.1 eV). If we neglect the Cl
3s orbitals, there are 18 ligand-group orbitals to consider (six Cl atoms × three 3p orbitals
per Cl). Using group theory, these ligand-based orbitals can be divided into seven SALCs:
one singly degenerate SALC with a1g symmetry, one doubly degenerate SALC with eg
symmetry, and five triply degenerate SALCs with t1g, t2g, t2u, t1u, and t1u symmetry (Figure
5.21). The a1g, t1u, and eg ligand SALCs are analogous to the SALCs already shown for the
octahedral SH6 molecule in Figure 5.16. They can form σ-bonding/antibonding MOs with Cr
orbitals of appropriate symmetry; the a1g SALC with Cr 4s, the t1u SALC with Cr 4p, and the
eg SALC with Cr 3dx2−y2 and 3dz2 orbitals.
The final three Cr valence orbitals, 3dxy, 3dxz, and 3dyz, have the correct symmetry to form
bonding interactions with the t2g Cl SALC. This interaction produces π-bonding and π-

z z z z z

y y y y y

x x x x x
dxy dyz dxz dx2−y2 dz2

Figure 5.20 The orientations of the d orbitals with respect to a Cartesian coordinate system.
184 Chemical Bonding

a1g eg t1u t2g t2u t1u t1g

Cl SALC
Cr orbital

No metal orbitals with

appropriate symmetry
s dz2 , dx 2−y2 px, py, pz dxy, dxz, dyz

Figure 5.21 Representative Cl 3p SALCs and corresponding orbitals on the central Cr atom in [CrCl6]3−
grouped by symmetry. For clarity only the atoms and orbitals in the xy plane are shown.

antibonding MOs. The remaining Cl SALCs are nonbonding ligand-based MOs.22 Because
the chlorine ligands are more electronegative than the metal, the bonding MOs have more
ligand character and the antibonding MOs have more metal character.
The MO diagram for [CrCl6]3− is shown in Figure 5.22. Its complexity may be a bit
intimidating at first glance. Fortunately, the chemical and physical properties are dictated
largely by the frontier orbitals. The d 3 configuration of chromium(III) means that the half-
filled t2g (π*) orbitals are the HOMO and the empty eg (σ*) orbitals are the LUMO. When the
d orbitals are partially filled, we need therefore only concern ourselves with the t2g (π*) and eg
(σ*) sets of MOs. When the d orbitals are empty, as is the case when the transition metal has
a d 0 configuration, the nonbonding ligand-group t1g set becomes the HOMO.
The splitting of the d orbitals into two groups results from the fact that two of them, the Cr
3dx2−y2 and 3dz2 orbitals, point directly at the ligands, hence their antibonding (σ*) inter-
action with the eg Cl SALC has a greater overlap than the π* interaction between the
remaining three Cr orbitals, 3dxy, 3dxz, 3dyz, and the t2g Cl SALC. Because the σ* overlap
involving the eg orbitals is more destabilizing than the π* overlap involving the t2g orbitals,
there is a splitting of the antibonding MOs with 3d-orbital parentage into two sets. The
magnitude of the energy separation between the two is referred to as ligand-field splitting, Δ.
Its value depends on the identity of both the ligands and the transition metal. In many cases,
the value of Δ is such that electrons can be excited between d orbitals by visible light, and, as
a result, transition-metal compounds are often colored. We will consider colors and bonding
of these materials in more detail in Chapter 7.
The electron counting used to obtain the occupancy of these important frontier orbitals is
generally straightforward for the transition-metal compounds encountered in solid state
chemistry. Most of the ligands we will encounter (chloride, fluoride, oxide, nitride, sulfide,

22
The situation is complicated somewhat by the fact that there are two different sets of triply degenerate SALCs with
t1u symmetry, both of which are allowed by symmetry to interact with the Cr 4p orbitals. However, the t1u SALC
that is shown in the same column as the Cr 4p orbital in Figure 5.21 contributes more prominently because it can
form both σ and π interactions with each Cr 4p orbital. The t2u and t1g SALCs are strictly nonbonding.
 5.3 Molecular-Orbital Theory 185

Figure 5.22 MO diagram for the octahedral anion [CrCl6]3− (Cl 3s orbitals neglected). The MOs with t1g,
t1u, and t2u symmetry in the middle of the MO diagram are nonbonding Cl 3p SALCs.

etc.) are closed-shell ions in the ionic limit. Therefore, the ligand-based SALCs will be
completely filled. Any remaining electrons go into the t2g (π*) and eg (σ*) MOs that have
predominantly metal d character. Their number is simply the number of the valence electrons
of the transition metal minus its oxidation state.
The MO diagram for the tetrahedral transition-metal anion CoCl42− is shown in Figure
5.23. As the number of ligands is reduced from six to four, the number of ligand-group
orbital SALCs is also reduced, from 18 to 12. Using group theory, the ligand-based orbitals
can be grouped into five SALCs; one singly degenerate SALC with a1 symmetry, one doubly
degenerate SALC with e symmetry, and three triply degenerate SALCs (one with t1, and two
186 Chemical Bonding

Figure 5.23 MO diagram for tetrahedral [CoCl4]2− (Cl 3s orbitals neglected).

with t2 symmetry). The Co 3d orbitals split into a triply degenerate t2 set (dxy, dxz, dyz) and
a doubly degenerate e set (dx2−y2, dz2). The doubly degenerate antibonding e MO lies at lower
energy than the triply degenerate antibonding t2 MO, which is opposite to what we found for
an octahedron. Furthermore, unlike a transition metal in octahedral geometry, neither the t2
set (dxy, dxz, dyz) nor the e set (dx2−y2, dz2) of orbitals (Figure 5.20) is pointing directly at the
ligands. Hence, for a given metal–ligand combination, the ligand-field splitting Δ is always
smaller for a tetrahedron than it is for an octahedron.23

5.3.9 High- and Low-Spin Configurations

Up to this point, we have neglected the effects of electron–electron interactions. However, to
properly understand the electronic structures and bonding of transition-metal compounds,
these effects cannot be ignored. Being negatively charged, electrons repel each other. Due to
their close proximity, the repulsive interaction is largest when the two electrons occupy the
same orbital. The energy penalty for placing two electrons with antiparallel spins in the same
orbital is called the spin-pairing energy, P.
In compounds containing first-row transition-metal ions, the ligand-field splitting energy Δ is
comparable to the spin-pairing energy P. For many combinations of electron count and
23
For example, when Co2+ ions are doped into MgO, their environment is octahedral and Δoct = 1.19 eV. When Co2+
ions are doped into ZnO, their environment is tetrahedral and Δtet = 0.47 eV. These values fall surprisingly close to
Δtet = (4/9)Δoct predicted from simple crystal-field theory discussed in Chapter 7.
 5.3 Molecular-Orbital Theory 187

Figure 5.24 Low-spin (left) and high-spin (right) configurations of Co3+ in an octahedral environment. Top:
Conventional representation without electron–electron interactions included. Bottom: Representation with
the pairing energy P shown explicitly.

coordination geometry, the electron configuration depends on which energy is larger, Δ or P.

Consider, for example, two possible electron configurations for a d 6 Co3+ ion in an octahedral
environment (Figure 5.24). These are normally represented with the energy diagrams shown on
the top of Figure 5.24. An alternative representation is shown on the bottom of the figure. There,
we explicitly see that when P > Δ (right-hand side) a high-spin (HS) configuration is the most
stable configuration, whereas when P < Δ (left-hand side) a low-spin (LS) configuration is
obtained.
The magnetic, optical, and electrical properties of transition-metal compounds depend
upon the spin state (HS or LS) of the transition-metal ion. For a d 6 ion in an octahedral
ligand field, the change is particularly dramatic because the HS ion is paramagnetic while the
LS ion is diamagnetic. The strength and length of the metal–ligand bond also depend upon
the spin state, because populating the σ* eg orbitals weakens the bonding more than
populating the π* t2g orbitals. Hence, LS Co3+ is smaller (0.685 Å; six-coordinate CR
value) than a HS Co3+ ion (0.75 Å) [3].
The question of whether the HS or LS state will be more stable depends on the transition-
metal ion, the ligands, and the coordination geometry. Some general guidelines are as follows:

• The 4d and 5d orbitals are larger than the 3d orbitals (Table 5.3). As a result, they
experience greater overlap with the ligand orbitals, leading to a larger Δ. At the same
188 Chemical Bonding

time, the increased size of the 4d and 5d orbitals decreases electron–electron repulsions,
leading to a smaller P. Consequently, second- and third-row transition-metal ions invari-
ably adopt LS configurations.
• Because Δtetr is small (roughly 4/9 of Δoct), the HS configuration is strongly favored for
tetrahedral coordination.
• The value of Δ also depends on the ligand. The ligands that are most common in solid state
chemistry, O2−, S2−, F−, and Cl−, give relatively small Δ, thereby favoring the HS config-
uration (weak-field ligands). Ligands with very short bonds to the central atom and empty
π* MOs, like CN− and CO, tend to produce large Δ and favor the LS configuration (strong-
field ligands).
• Increasing the metal oxidation state or moving from left to right across the transition-
metal series brings the energy of the metal d orbitals energy closer to the energy of the
ligand orbitals. This increases the covalent mixing, increases Δ, and favors the LS
configuration.

5.3.10 Jahn–Teller Distortions

There are certain combinations of cation electron configurations and ligand geometries that
are electronically unstable with respect to a distortion which lowers the symmetry of the
molecule (the site symmetry in an extended solid). Jahn–Teller distortions are the most
familiar class of electronically driven distortions. The Jahn–Teller theorem states that an
incompletely filled set of otherwise degenerate MOs will undergo a structural distortion that
removes the degeneracy and lowers the energy of these orbitals.
While there are many electron configurations that meet the conditions of the Jahn–Teller
theorem, the most important examples occur for octahedral coordination of either a d 9
(e.g. Cu2+) or a HS d 4 (e.g. Mn3+) ion. These two cases undergo Jahn–Teller distortions
because the doubly degenerate eg set is partially filled. The orbital degeneracy can be
removed either by lengthening the M–O bonds in the z direction and compressing the
bonds in the xy plane or vice versa. The z elongation of the octahedron stabilizes the dz2
orbital (it becomes less antibonding) and destabilizes the dx2−y2 orbital (it becomes more
antibonding) as shown on the right-hand side of Figure 5.25. The xy elongation, shown on
the left-hand side of Figure 5.25, does the opposite.
To a first approximation, both types of distortion lower the energy by an equivalent
amount, yet the distortion that elongates the octahedron along the z axis occurs almost
exclusively. What is the reason for this strong preference? Burdett [7] has shown that the
z-elongated octahedron is favored because its partially or fully occupied 3dz2 orbital is
additionally stabilized over 3dx2−y2 by symmetry-allowed mixing with the empty 4s orbital.
We will see in later chapters how Jahn–Teller distortions play a key role in the crystal
chemistry and physical properties of important classes of materials, such as cuprate super-
conductors (Chapter 12) and magnetoresistive oxides (Chapter 11).
 5.3 Molecular-Orbital Theory 189

compressed elongated
octahedron octahedron
octahedron

d z2 d x2− y2

d z2 d x2− y2
dx2− y2 d z2

dxz dyz dxy

dxz dxy dyz
dxz dyz
dxy

Figure 5.25 Two possible distortions of an octahedron that remove the degeneracy of unequally occupied
eg orbitals for a HS d 4 ion: elongation of the bonds along z (right), and elongation of the bonds in the xy
plane (left). In both instances, the remaining bonds contract to maintain the same degree of metal–ligand
bonding.

Second-order Jahn–Teller (SOJT) distortions are a related class of electronically driven

distortions. SOJT distortions alter the symmetry of the molecule (or crystal) so that two or
more MOs that were orthogonal prior to the distortion can interact. The interaction lowers
the energy of one or more filled MOs and raises the energy of one or more empty MOs. The
driving force for a SOJT distortion is inversely proportional to the energy difference between
the two states that are interacting. Consequently, these states are often the HOMO and the
LUMO. Both the conventional Jahn–Teller distortion (sometimes called a first-order Jahn–
Teller distortion) and the second-order Jahn–Teller distortion stabilize the molecule by
lowering its symmetry, but for a SOJT the molecule need not possess degenerate, partially
occupied MOs as the HOMO.
As an example of the second-order Jahn–Teller distortion, consider the NH3 molecule. We
know that the atoms in NH3 form a trigonal pyramid and not a triangle. Within the VSEPR
framework, the stereochemical influence of the electron lone pair is invoked to explain this,
but we can use MO theory to reach the same conclusion. The MOs for NH3 in both
configurations are shown in Figure 5.26. In planar NH3, the HOMO is a strictly nonbonding
N 2pz orbital (labeled 1a2″). The distortion to nonplanar NH3 enables interaction (mixing)
between the N 2pz orbital and the N 2s—H σ* MO (labeled 2a1 0 ) that is symmetry forbidden
in the planar geometry. This mixing converts the HOMO from a nonbonding MO to
190 Chemical Bonding

Figure 5.26 MO diagrams for planar (left) and nonplanar (right) NH3. The SOJT is driven by the interaction
between the N 2pz nonbonding and N 2s—H σ* MOs that is symmetry forbidden in the planar geometry but
allowed in the pyramidal geometry. The notation used to assign MO labels is explained in Footnote 20 in this
chapter.

a weakly bonding MO, thereby stabilizing the molecule.24 We can recognize the HOMO as
the lone pair predicted for NH3 in VSEPR theory. In solid state chemistry, SOJT distortions
are often associated either with formation of a stereochemically active electron lone pair or
with d 0 ions in octahedral coordination. We will see in Chapter 8 that SOJT distortions play an
important role in the crystal chemistry of dielectric and nonlinear optical materials.

5.4 Bond Valences

The bond-valence concept is based on the idea that an atom has a certain bonding power,
a valence, distributed over the bonds it forms. This concept takes a particularly simple form
in organic chemistry, where each two-electron bond carries a valence of one. It is a powerful
predictive concept to know a priori that in stable organic molecules carbon forms four two-
electron bonds with its neighbors, oxygen forms two bonds, hydrogen forms one bond, etc.

24
SOJT distortions lower the energy of some MOs while raising the energies of others, including some filled MOs,
which can make it difficult to decide whether a particular distortion should lower the overall energy of the molecule.
In most cases, the geometry that achieves the lowest energy for the HOMO also has the lowest total energy.
 5.4 Bond Valences 191

We saw in Section 1.3.1 that one can also assign bond valences in inorganic solids, but,
unlike in most organic compounds, they often take non-integer values. For example, the
bond-valence balance (Figure 1.14) gives a valence of ⅙ for an Na–Cl bond in NaCl, and
a valence of ⅔ for a Ti–O bond in TiO2. Recall that these expectation values are calculated by
dividing the absolute value of the oxidation state of each atom by the atom’s coordination
number. In this section, we expand upon this simple concept by relating valences to experi-
mentally observable quantities, such as bond lengths.
We begin with the valence-sum rule—the valence vi of an atom (here meaning the absolute
value of its oxidation state) is equal to the sum of bond valences vij around it,
X
vi ¼ vij (5.15)
j

where j is the number of bonds formed by atom i, that atom’s coordination number. For
example, the octahedrally coordinated Ti in TiO2 has vTi = 6 × vTi–O = 6 × ⅔ = 4, oxygen has
vO = 3 × vTi–O = 3 × ⅔ = 2.
To relate these values to experimental data, we need to find a quantitative relationship
between bond valence and bond length. While there is more than one function that can be
used to approximate this relationship, the most widely used expression is:
!
R0ij  dij
vij ¼ exp (5.16)
B

where vij and dij are the valence and length, respectively, of the bond between atoms i and j, R0ij
is the length of a bond with a valence of one (a single, two-electron bond), and B is
a “universal” constant usually taken to be 0.37 Å. The values of R0ij are determined empiric-
ally from the crystal structures of known compounds. Each cation–anion pair has its own R0ij
value (see Appendix D for bonds to oxygen).
Owing to the exponential relationship between bond valence and bond length, the bond
valence increases faster upon bond contraction than it decreases upon bond expansion. For
example, if we start with a bond whose distance dij = R0ij and valence vij = 1.0 and shorten it by
~0.256 Å, the valence is doubled (see Figure 5.27). If we lengthen the bond by the same
distance, the valence is halved, but this is a smaller change on an absolute scale.
A consequence of this asymmetry is the distortion theorem—for any ion, lengthening some
of its bonds and shortening others, while keeping the bond-valence sum the same, will always
increase the average bond length. Consequently, the coordination polyhedron about a cation
will have the smallest volume when all anions are equidistant.25

25
An important consequence of the distortion theorem is that a too-small cation placed in a symmetric environment of
anions will move off center to increase the lengths of some bonds and decrease the lengths of others, thus creating
a local dipole moment (Chapter 8).
192 Chemical Bonding

2.02 vij = 2

Rij0 – dij = B ln vij

1.5
bond valence vij

1.01 vij = 1

vij = ½
0.5
vij = ¼
vij = ⅛

0
Rij0 – 0.5 Rij0 Rij0 + 0.5 Rij0 + 1
bond length dij (Å)

Figure 5.27 The variation in the valence of a bond as a function of bond length.

The bond-valence method combines the valence-sum rule in Equation (5.15) with the bond-
length bond-valence relation of Equation (5.16). It works forwards or backwards; either to
evaluate the bond-valence sum at an atom from experimental bond lengths or to predict bond
lengths.
An important application of the former approach is to assess the validity of experimentally
determined crystal structures. To check a structure, the valence of each bond is calculated
from its bond distance using Equation (5.16), and the valence sum is then evaluated for each
crystallographically distinct/inequivalent ion with Equation (5.15). If the bond-valence sum
vi is close to the oxidation state expected for each ion, we can say that the bonding is similar
to that seen in the structures that were used to determine the R0ij values, which implies
a chemically reasonable structure.
Evaluation of the bond-valence sum can also be used to assess atom oxidation states in
compounds. As an example, consider the crystal structure of ilmenite (Figure 5.28) which is an
ordered variant of corundum containing alternating layers of iron- and titanium-centered octa-
hedra sharing faces. Using the notation introduced in Section 1.3, we express the coordinations as
Fe[6o]Ti[6o]O3[2,2]. There are electrostatic repulsions between cations in this structure, which cause
the iron and titanium to shift away from each other, resulting in distorted coordination environ-
ments for both, as shown in Figure 5.28. The assignment of oxidation states is potentially
ambiguous, either Fe2+Ti4+O3 or Fe3+Ti3+O3. Let’s see if we can use bond-valence sums to
determine the correct values. We begin by assuming Fe2+Ti4+O3 and convert the bond distances
Fe–O (2.078 Å, 2.201 Å) and Ti–O (1.874 Å, 2.089 Å) in Figure 5.28 into bond valences:
 5.4 Bond Valences 193

d = 1.874 Å (×3)

TiO6
d = 2.089 Å (×3)
FeO6
d = 2.201 Å (×3)
TiO6
d = 2.078 Å (×3)
FeO6

TiO6

Figure 5.28 The crystal structure of ilmenite, FeTiO3.

   
R0TiO  dTiO 1:815  1:874
vTiO ð1Þ ¼ exp ¼ exp ¼ 0:853
B 0:37
 0   
RTiO  dTiO 1:815  2:089
vTiO ð2Þ ¼ exp ¼ exp ¼ 0:477
B 0:37
 0   
RFeO  dFeO 1:734  2:201
vFeO ð1Þ ¼ exp ¼ exp ¼ 0:283
B 0:37
 0   
RFeO  dFeO 1:734  2:078
vFeO ð2Þ ¼ exp ¼ exp ¼ 0:395 (5.17)
B 0:37

Summing up the valences about each ion gives vTi = (3 × 0.853) + (3 × 0.477) = 3.99 and vFe =
(3 × 0.283) + (3 × 0.395) = 2.03, confirming our guess about the Ti and Fe valences. We can
also calculate the bond-valence sum for oxygen; vO = 0.853 + 0.477 + 0.283 + 0.395 = 2.008
that is reassuringly close to the ideal value. Note that if we had initially guessed Fe3+Ti3+O3,
the R0ij values used in our calculations would have been slightly different (1.759 Å for Fe3+–O
and 1.791 Å for Ti3+–O), yielding bond-valence sums vTi = 3.84 and vFe = 2.17, somewhat
farther from the integer values but still leading to the same conclusion.
We can also work in the opposite direction, using the bond-valence method to predict bond
lengths. Knowing the crystal-chemical formula of the structure in question, we use the valence-
sum rule, Equation (5.15), to convert the oxidation state of an atom into the expected, ideal
values of the bond valences. After rearranging Equation (5.16), we use these ideal valences to
predict the lengths of bonds in much the same way that ionic radii are used:

dij ¼ R0ij  B ln vij (5.18)

194 Chemical Bonding

Figure 5.29 The cubic

perovskite structure of
SrTiO3 showing the
octahedral coordination
of Ti (left) and the
cuboctahedral coordination
of Sr (right).

Let’s consider the cubic perovskite SrTiO3 as an illustration of how bond valences can be
used to predict distances. The coordination numbers are Sr[12co]Ti[6o]O3[4,2] as illustrated by
the bond graph shown in Figure 5.29. We expect that each of the Ti–O bonds will have
a valence of 4/6 = 2/3 and each of the Sr–O bonds a valence of 2/12 = 1/6. Given R0ij parameters of
1.815 Å and 2.118 Å for the Ti4+–O and Sr2+–O bonds, the expected bond distances are
calculated to be 1.965 Å and 2.781 Å, respectively. As discussed in Section 1.5.3, the
undistorted cubic perovskite structure will only be stable if the Sr–O and Ti–O lengths are
appropriately matched. Our distances can be used to calculate the tolerance factor (see
Figure 1.50): t = dA–O/(dM–O√2) = 2.781/(1.965×√2) = 1.001. Recalling from Chapter 1 that
for the cubic perovskite structure to be stable the tolerance factor should be close to 1, we see
that the undistorted structure does satisfy the bonding preferences of both Sr2+ and Ti4+.
If we replace Sr2+ with the smaller Ca2+ and repeat the above calculation, we get a smaller
tolerance factor, t = 0.966, suggesting an octahedral-tilting distortion. At room temperature,
SrTiO3 is a cubic perovskite and CaTiO3 a distorted perovskite (exhibiting a−a−b+ tilting,
Section 1.5.3), in agreement with the bond-valence predictions.
Comparisons of distances obtained from bond valences with those obtained by
summing ionic radii show that there is a good agreement between the two approaches.
Table 5.4 illustrates this for selected symmetric coordination polyhedra. The agreement
should not come as a surprise, as both methods depend on parameters derived from
similar sets of structures. Both capture the same trends as the cation coordination
number and oxidation-state change. However, the bond-valence approach has advan-
tages. For a given cation–anion pair, only one bond-valence parameter R0ij is needed,
whereas with ionic radii a different value is associated with each coordination number
(both for the cation and the anion). An even more important advantage of the bond-
valence approach is the ability to handle distorted environments as easily as symmetric
environments. One limitation of bond-valence parameters is that they are not generally
tabulated for specific spin states (i.e. HS or LS ions of 3d metals), which do affect bond
lengths (Section 5.3.9).
 5.5 Problems 195

Table 5.4 Metal–oxygen bond distances dBV in a polyhedron of coordination

number CN, computed from bond-valence parameters R0 and compared with
distances dIR obtained from ionic radii [3]. The four-coordinate radius of oxygen
has been arbitrarily used for calculating dIR for all entries.

R0 (Å) CN vij dBV (Å) dIR (Å)

Ca2+–O 1.967 6 ⅓ 2.37 2.38

8 ¼ 2.48 2.50
10 ⅕ 2.56 2.61
Mg2+–O 1.693 4 ½ 1.95 1.95
6 ⅓ 2.10 2.10
8 ¼ 2.21 2.25
Zn2+–O 1.704 4 ½ 1.96 1.98
6 ⅓ 2.11 2.12
Al3+–O 1.620 4 ¾ 1.73 1.77
6 ½ 1.88 1.915
Fe2+–O 1.734 4 ½ 1.99 2.01
6 ⅓ 2.14 2.16
Fe3+–O 1.759 4 ¾ 1.87 1.87
6 ½ 2.02 2.025

5.5 Problems
5.1 Consider the infinite series for the Madelung constant of the NaCl-type structure. Its
convergence depends on how the successive terms are chosen. As written in Equation
(5.2), each successive shell contains ions of the same type (either cations or anions). (a)
Calculate the sum of this series for two shells, three shells, etc., up to the full seven shells
listed in Equation (5.2). For each successive shell, determine the total number of cations
and anions surrounding the central anion. (b) What can you say about the convergence
of this series after seven terms? (c) Which of these successive sums is closest to the
Madelung constant value of +1.7476? For which sum is the total charge of the cluster
closest to zero? (d) What can be done to achieve a more rapid convergence of this series?
5.2 Taking into account both attractive and repulsive interactions, derive an equation
analogous to Equation (5.2) for the CsCl structure (Figure 1.40). Include the first four
terms in the Madelung series. Hint: You may find this easier to do in terms of the cell
edge a, and then convert to the interatomic distance d that is normally used in Madelung
formulas.
5.3 CaO adopts the NaCl-type structure with a = 4.80 Å. (a) Use the Born–Mayer equation
to calculate the lattice-formation energy for CaO. (b) How well does this estimate agree
with the value of −3414 kJ/mol determined from the Born–Haber cycle? (c) Calculate
196 Chemical Bonding

the size of the repulsive term as a percentage of the attractive term. (d) Given the fact
that SrO has the same structure, would you expect the lattice-formation energy of SrO
to be larger or smaller than CaO?
5.4 MgO adopts the NaCl-type structure with a = 4.22 Å. (a) Use the Born–Mayer equation to
calculate the lattice-formation energy for MgO. (b) Given this estimate of the lattice energy,
construct a Born–Haber cycle and estimate the second electron gain enthalpy of oxygen,
O−(g) + e− → O2−(g). Sublimation enthalpy of Mg = +147 kJ/mol, bond dissociation energy
of dioxygen = 498 kJ/mol, first ionization energy of Mg = 738 kJ/mol, second ionization
energy of Mg = 1451 kJ/mol, first electron gain enthalpy of oxygen = −141 kJ/mol, enthalpy
of formation of MgO = −602 kJ/mol.
5.5 Why are there no examples of fluorides with the CdI2 or CdCl2 structures (Figure 1.28)?
5.6 With the exception of helium, all noble gases solidify at low temperature. The lack of
ionic or covalent bonding means that atoms are held together by dispersion forces alone.
Given the melting points of the noble gases; Ne = 24 K, Ar = 84 K, Kr = 116 K, Xe =
161 K, what can you say about the strength of the London dispersion forces as the
principal quantum number of the outermost shell increases? What is the explanation for
this trend?
5.7 Classify each of the following statements about nodes in orbital wavefunctions as true or
false: (a) s orbitals have no nodes. (b) The orbital wavefunction always changes sign at
a node. (c) The number of nodal planes is determined by the principal quantum number.
5.8 What are the values of the principal and orbital angular-momentum quantum numbers
for each of the following orbitals? How many radial nodes and nodal planes does each
orbital possess? (a) 4s orbital, (b) 5d orbital, (c) 4f orbital, (d) 2p orbital.
5.9 Use the MO diagram of oxygen to determine the oxygen–oxygen bond order in the
peroxide ion, O2 2 . Will the O–O distance in peroxide be longer or shorter than in O2?
5.10 Construct an MO diagram for trigonal-planar BH3 by analogy with the MOs for
trigonal-planar NH3 in Figure 5.26. Use this diagram to determine the degeneracy
and orbital character of the HOMO and the LUMO.
5.11 Consider these six-coordinate ionic radii (IR values from ref. [3]; see also Footnote 3 in
this chapter) for divalent, first-row transition-metal ions: r(Ti2+) = 0.86 Å, r(V2+) = 0.79
Å, r(Cr2+) = 0.80 Å, r(Mn2+) = 0.83 Å, r(Fe2+) = 0.78 Å, r(Co2+) = 0.745 Å, r(Ni2+) = 0.69
Å. For a fixed oxidation state, the ionic radius normally decreases on moving left to right
across the periodic table due to the increasing effective nuclear charge. Why then does the
radius increase on moving from V2+ to Cr2+ to Mn2+?
5.12 For which d-electron counts are there distinct HS and LS states of an octahedrally
coordinated transition-metal ion?
5.13 In each of the following pairs, one species contains a transition metal in the HS state and
the other in the LS state. Indicate the complex that is most likely to contain the LS ion.
(a) Fe[6]Cl3 and Ru[6]Cl3, (b) [Co[6](NH3)6]3+ and [Co[4]Cl4]−.
5.14 NiO adopts the cubic NaCl-type structure while PtO adopts the cooperite structure
shown below. (a) What factor do you think is responsible for the differing crystal-
 5.5 Problems 197

chemistry preferences of these two compounds? Hint: Consider the splitting and occu-
pation of the d orbitals for each compound. (b) Which structure type do you think PdO
will adopt? Would it be possible to tell from a magnetic measurement?

O Pt

Ni

NiO (rock salt) PtO (cooperite)

5.15 Construct an MO diagram for a linear H2O molecule by analogy with BeH2 in Figure
5.17. (a) Determine the degeneracy and orbital character of the HOMO and the LUMO.
(b) Identify the orbitals on oxygen that participate in bonding to hydrogen. (c) Now
distort the molecule by bending the H–O–H bond and consider how this impacts the
MO diagram. How does the orbital character of the HOMO(s) change? (d) Which
oxygen orbitals now participate in bonding? (e) Is this distortion an example of a first-
or second-order Jahn–Teller distortion?
5.16 MgF2 adopts the rutile structure (Figure 1.45) with a = 4.62 Å and c = 3.04 Å. The bond-
valence parameter R0MgF = 1.581 Å. (a) Use the bond-valence method to predict the
length of the Mg–F bonds. (b) Use the Born–Mayer equation to estimate the lattice-
formation energy for MgF2. (c) Comment on the difference between this value and the
value of −2978 kJ/mol obtained from a Born–Haber cycle. Is the agreement between
calculated and experimental values similar to that observed for the alkali-metal halides
discussed in Section 5.1.3? (d) Would you expect the lattice-formation energy of rutile
(TiO2) to be lower (more stable) than MgF2?
5.17 The structure of ZrV2O7 can be derived from the structure of NaCl by replacing Na+
with Zr4+ and Cl− with V2O74− pyrovanadate groups. The coordination environment of
zirconium is octahedral while the local coordination at vanadium is tetrahedral. One of
the seven oxygen atoms, O(1), does not bond to Zr, while the other six equivalent
oxygens, O(2), bond to both Zr and V. (a) Construct a bond graph (Section 1.3.1) for
ZrV2O7. (b) What are the idealized valences for the V–O(1), V–O(2), and Zr–O(2)
bonds? (c) Given bond-valence parameters R0ZrO = 1.928 Å and R0VO = 1.803 Å calculate
the expected V–O(1), V–O(2), and Zr–O(2) bond distances.
5.18 In LaOF, which has a structure closely related to fluorite, each lanthanum is surrounded
by four oxide and four fluoride ions. Although X-ray-diffraction studies cannot easily
distinguish oxygen from fluorine, two bond distances are seen in the crystal structure:
198 Chemical Bonding

2.42 Å to one anion and 2.60 Å to the other anion. (a) Write the crystal-chemical formula
(Section 1.3) for LaOF. (b) Given the eight-coordinate La3+ radius of 1.30 Å and the four-
coordinate radii for O2− and F− of 1.24 Å and 1.17 Å, respectively, assign the two
observed distances to La–O and La–F bonds. (c) Determine the ideal valences of the
La–F bonds and the La–O bonds. (d) Use the bond valences from part (c) and the bond-
valence parameters R0LaF = 2.02 Å and R0LaO = 2.17 Å to estimate the length of La–F and
La–O bonds in this structure. Does your assignment of the two bonds based on bond
valences agree with the assignment based on ionic radii in part (b)?
5.19 The bond-valence parameters for Ca–F and Mg–F bonds are R0CaF = 1.842 Å and
R0MgF = 1.581 Å. Calculate the closest Ca–F, Mg–F, and F–F distances in MgF2 and
CaF2 assuming (a) a fluorite-type structure, (b) a rutile-type structure. (c) The four-
coordinate radius of F− is 1.31 Å and the three-coordinate radius is 1.30 Å (both IR
values from ref. [3]). A simple-minded analysis based on these radii suggests that F–F
contacts shorter than 2.62 Å will be unfavorable in fluorite and shorter than 2.60 Å
unfavorable in rutile. How do the F–F distances calculated from the cation–anion
distances in parts (a) and (b) and simple geometric considerations compare with these
limiting distances? (d) The bond-valence parameter for Be–F is R0BeF = 1.28 Å. Based on
this value, do you think BeF2 would be stable in the rutile structure?
5.20 MgSiO3 is of interest to geologists because it is abundant in the Earth’s mantle.
Silicon is normally tetrahedrally coordinated by oxygen, but, at the high pressures
found in the mantle, silicon becomes octahedrally coordinated and a perovskite
structure is formed. (a) Given bond-valence parameters R0MgO = 1.693 Å and R0SiO =
1.624 Å, calculate the expected Mg–O and Si–O distances for a cubic perovskite. (b) Use
the distances calculated in part (a) to estimate the tolerance factor t (Section 1.5.3), and
state whether you would expect MgSiO3 to form as a cubic or a distorted perovskite. (c)
Experiments identify MgSiO3 as a distorted perovskite. The Si–O distances are 1.78 Å
(×2), 1.79 Å (×2), and 1.80 Å (×2). How do those compare to your estimate of the Si–O
distance in part (a)? (d) The Mg–O distances are 2.00 Å, 2.06 Å (×2), 2.29 Å (×2), 2.41 Å
(×2), 2.85 Å, 2.96 Å, and 3.11 Å (×2). Calculate the bond-valence sum for Mg2+ as
a qualitative estimate of whether this is a reasonable coordination environment for
Mg2+.

5.6 Further Reading

T.A. Albright, J.K. Burdett, M.H. Whangbo, “Orbital Interactions in Chemistry” (1985) John Wiley
and Sons.
J.K. Burdett, “Chemical Bonding in Solids” (1995) Oxford University Press.
R. Dronskowski, “Computational Chemistry of Solid State Materials” (2005) Wiley−VCH.
D.M.P. Mingos, “Essential Trends in Inorganic Chemistry” (1998) Oxford University Press.
 5.7 References 199

Y. Jean, F. Volatron (translated and edited by J.K. Burdett) “An Introduction to Molecular Orbitals”
(1993) Oxford University Press.
I.D. Brown, “The Chemical Bond in Inorganic Chemistry: The Bond Valence Model” (2006) Oxford
University Press.

5.7 References
[1] R.P. Grosso Jr., J.T. Fermann, W.J. Vining, “An in-depth look at the Madelung constant for cubic
crystal systems” J. Chem. Educ. 78 (2001), 1198–1202.
[2] R.A. Jackson, C.R.A. Catlow, “The Madelung constant” Mol. Simul. 1 (1988), 207–224.
[3] R.D. Shannon, “Revised effective ionic-radii and systematic studies of interatomic distances in
halides and chalcogenides” Acta Crystallogr. Sect. A 32 (1976), 751–767.
[4] M. Wilson, P.A. Madden, “Anion polarization and the stability of layered structures in MX2
systems” J. Phys. Condens. Matter 6 (1994), 159–170.
[5] R. Hoffmann, “An extended Hückel theory. I. Hydrocarbons” J. Chem. Phys. 39 (1963),
1397–1412.
[6] J.P. Desclaux, “Relativistic Dirac–Fock expectation values for atoms with Z = 1 to Z = 120”
At. Data Nucl. Data Tables 12 (1968), 311–406.
[7] J.K. Burdett, “Chemical Bonding in Solids” (1995) Oxford University Press, 242–251.
6 Electronic Band Structure

In Chapter 5 we saw how molecular orbital (MO) diagrams can be used to describe the
electronic structures of molecules. In this chapter we turn our attention to extended solids,
whose electronic structures are represented by band-structure diagrams. The optical, elec-
trical, and magnetic properties of a material are directly linked to its band structure. The
chemical reactivity and catalytic properties of a material depend upon the energy levels and
symmetry of electronic states near the Fermi level; even dielectric and mechanical properties
can be traced to chemical bonding interactions that are intimately linked to the electronic
structure. It is therefore essential to develop a working knowledge of the electronic band
structures of solids before we can begin to understand the behavior of many functional
materials.
In Chapter 5, we learned how MOs can be derived from the overlap of atomic orbitals
(AOs) using the linear combination of atomic orbitals (LCAO) approach. In this chapter,
we will see how the electronic structures of extended crystalline solids can be built up in the
same way. This approach follows directly from MO theory, which makes it particularly
intuitive for chemists. It can be applied to solids with electrons that are localized or delocal-
ized. This is an important advantage because many interesting phenomena arise in materials
with intermediate degrees of electron delocalization.
To make the visualization and mathematics easier, we begin by considering the electronic
structures of 1D systems. The concepts developed in 1D are then extended to describe
electronic structures of 2D and 3D crystals.

6.1 The Band Structure of a Hydrogen-Atom Chain

To introduce the concepts associated the electronic structure of an extended solid, we begin
with the simplest structure we can imagine; an infinite 1D chain of hydrogen atoms. The
following sections develop the band structure of this model system.

200
 6.1 The Band Structure of a Hydrogen-Atom Chain 201

highest-
energy MO

energy

lowest-
energy MO
H2 H6 H10 H14 H50 H∞

Figure 6.1 MO diagrams for H2 and several cyclic HN molecules (N = 6, 10, 14, 50, ∞). For economy of
space, only the electrons in the highest-energy occupied MOs are shown for H50 and occupied MOs are
shaded for H∞. Shading of the MO images indicates the sign individual H 1s wavefunctions acquire in the
LCAO formalism; ψ > 0 is white and ψ < 0 gray.

6.1.1 The Electronic Structures of Cyclic HN Molecules

Before considering an infinite chain, let’s look at the MO diagrams of cyclic all-hydrogen
molecules. Because each atom has an identical environment, such molecules are the most
appropriate finite-sized approximants to an infinite chain. The MO diagrams for H2 and cyclic
HN molecules of 6, 10, 14, and 50 atoms are shown in Figure 6.1 together with a molecule
extrapolated to H∞ so that its MOs become continuous. These MO diagrams possess several
common features that can be understood from the concepts covered in Chapter 5:

• The number of MOs in the molecule is equal to the number of constituent AOs. For
hydrogen, we have one AO per atom: the 1s orbital.
• The bonding interactions between nearest-neighbor atoms range from the completely
bonding lowest-energy MO to the completely antibonding highest-energy MO.
• As the energies of the MOs increase, the number of nodes increases (see discussion of 2p π
MOs of C6H6 in Section 5.3.7).
• As the size of the molecule increases, the MOs become more closely spaced in energy, with
the highest density occurring near the most bonding and most antibonding orbitals.
202 Electronic Band Structure

6.1.2 Translational Symmetry and the Bloch Function

How can we go from molecules with relatively small numbers of atoms to crystals where the
number of atoms becomes practically infinite? In a manner similar to the way molecular chemists
use rotational symmetry to derive the symmetry-adapted linear combinations (SALCs) in an MO
diagram (Section 5.3.5), we will use the translational symmetry (Section 1.1.1) of a crystal to
derive the orbitals that make up the band structure of an extended solid.
To illustrate this, consider a ring of hydrogen atoms HN, where N is very large but still finite. If
N is sufficiently large, we can neglect the curvature of the ring to approximate the translational
symmetry of an infinite linear chain, a periodic 1D “crystal” with a unit cell of length a equal to
the interatomic H–H distance. Just as was the case for the atoms and molecules discussed in
Chapter 5, we will treat each electron in the crystal as a wave, modeled by a wavefunction ψ(r)
(Section 5.2). The wavefunctions in crystals are conceptually no different from those in atoms or
molecules, so, likewise, the energy, momentum, and probability density of an electron in
a crystal can be calculated from its wavefunction. Our ψ(r) is a product of two functions:

ψðrÞ ¼ eikr uðrÞ (6.1)

of distance r from an arbitrary origin.1 The u(r) function follows the periodicity of the chain, u(r) =
u(r + na) where n is an integer that labels the hydrogen atoms, and eikr = cos(kr) + i sin(kr). With
the parameter k (Section 6.1.3) having units of inverse length, the product kr is dimensionless (the
arguments of the cos and sin functions are in radians). The wavefunction ψ(r) of Equation (6.1) is
called the Bloch function and describes crystal orbitals, the infinite-crystal analogs to MOs.
The u(r) in Equation (6.1) is referred to as a basis set. Throughout most of this book, we use
tight-binding methods where AO wavefunctions are the basis set.2 This is the solid state analogue
of MO theory since, in both approaches, each electron wavefunction ψ(r) is a linear combination
of atomic orbitals (LCAO). For our linear chain of hydrogen atoms, the basis set is the sum of
N hydrogen 1s wavefunctions (ψ1s), each centered at one of the H nuclei along the chain. For
each value of k, we combine the H 1s orbitals in a different way to get a unique crystal orbital that
can hold two electrons of opposite spin without violating the Pauli exclusion principle (Section
5.2). The manner in which this is done will become clear in Section 6.1.4 where we examine the
crystal-orbital wavefunctions of the infinite H-atom chain more closely. In Section 6.2, we will
see that when the unit cell contains more than one atom, we can use MOs as the basis set.
The tight-binding model provides a great deal of chemical understanding because the basis
set explicitly retains the identities of the atoms that make up the crystal, but the functions
used as the basis set need not be AOs or centered on individual atoms.3 In fact, most modern

1
We can also express the distance from the origin r = xa, where x is a fractional coordinate, and the interatomic
distance a is the length of the 1D unit cell.
2
Computational implementations of tight-binding theory typically use mathematical functions that approximate AO
wavefunctions, such as Gaussian-type and Slater-type orbitals.
3
The free-electron model takes the rather drastic approach of assuming the periodic electric-field potential created by
the atomic nuclei is uniform throughout the crystal, in which case u(r) = 1. This approximation, while
 6.1 The Band Structure of a Hydrogen-Atom Chain 203

approaches to calculating the electronic structures of crystals rely upon basis sets other than
AOs.4 The details of these methods are beyond the scope of this book, but interested readers
are encouraged to consult Dronskowski’s Computational Chemistry of Solid State Materials,
listed in Further Reading.

6.1.3 The Quantum Number k

Reducing the electronic structure of an entire crystal down to the contents of a single unit cell
greatly simplifies analysis of its electronic structure. This entire simplification is achieved by
introducing the parameter k in the Bloch function, Equation (6.1). Although k can take any
value, we need only concern ourselves with k values within a finite range. To show this, we
return to our HN ring. If we move a distance Na along the length of the ring, we are back
to where we started, and the wavefunction must repeat. Mathematically, this equates to
the introduction of a boundary condition that constrains the electronic wavefunction
to have the same value at the end of the chain that it has at the beginning:

ψðrÞ ¼ ψðr þ NaÞ (6.2)

If we apply this boundary condition to the Bloch function, Equation (6.1), recalling that
u(r) = u(r + na), we obtain:

eikr uðrÞ ¼ eikðrþNaÞ uðrÞ ¼ eikNa eikr uðrÞ ¼ ½cosðkNaÞ þ i sinðkNaÞeikr uðrÞ (6.3)

This equality only holds if the term in square brackets is unity, which only happens when kNa
equals an integer number n of the “sinusoid” periods 2π:

kNa ¼ nð2πÞ (6.4)

Thus, we see that for a finite ring HN, the allowed values of k are quantized. In analogy to MO
theory, a finite ring containing N hydrogen atoms will have one crystal orbital for every atom
in the ring (see Figure 6.1), each with a different value of k.5 If we let n adopt positive and
negative integer values ranging from ±1 to ±N/2,6 we obtain the following set of values for k:

nð2πÞ 2π 4π 6π ðN=2Þ2π
k¼ ¼ ;  ;  ; …  (6.5)
Na Na Na Na Na

computationally simple, is of limited utility (and even more limited accuracy) because it effectively removes chemical
bonding from consideration. We will use the free-electron model as an entry point to study the conductivity of simple
metals in Chapter 10.
4
Most AOs possess radial nodes in the core region, and the rapid oscillations of the wavefunction near these nodes
makes calculations with an atomic orbital basis set numerically expensive. Most calculations do not include the core
electrons and use simpler functions to approximate the wavefunctions of valence electrons in the core region.
5
In a finite crystal, the number of k values will be equal to the number of unit cells in the crystal.
6
For convenience, we will assume that N is an even number so that N/2 is an integer.
204 Electronic Band Structure

a a* = 2π/a

real-space lattice reciprocal-space lattice

Figure 6.2 The 1D real- and reciprocal-space lattice. In the latter, the interval that is marked corresponds
to the first Brillouin zone.

As the number of atoms in the chain N goes to infinity, the difference between successive
values of k becomes infinitesimally small, and k becomes continuous. Interestingly, we see
that the last term in Equation (6.5), which contains the smallest (−π/a) and largest (+π/a)
value of k, does not depend upon the number of atoms in the chain. This is important because
it means that even though k becomes continuous as the chain length goes to infinity, we only
need to consider the values of k that fall in the finite range −π/a ≤ k ≤ +π/a (of width 2π/a).
To further demonstrate the link between translational symmetry and k, we borrow
a concept from X-ray crystallography and transform the real-space lattice into the
reciprocal-space lattice shown in Figure 6.2. We relate the reciprocal-space lattice parameter
a* (with units of inverse length, just like k) to the real-space lattice parameter a through the
relationship a* = 2π/a.7 The range of k values needed to generate all possible crystal orbitals
without duplication, −π/a ≤ k ≤ +π/a, can be expressed in reciprocal-space coordinates as
−a*/2 ≤ k ≤ +a*/2. This interval, which is the unit cell in 1D reciprocal space, defines the first
Brillouin zone. Just as knowing the unit cell and its contents in real space defines the crystal
structure of a material, knowing the energies and wavefunctions of the crystal orbitals
throughout the first Brillouin zone defines its electronic structure.

6.1.4 Visualizing Crystal Orbitals

Let’s consider the analogy between MOs and crystal orbitals constructed using the tight-
binding approach (Section 6.1.2) more closely. Recall from Equation (5.12) that the MO
wavefunctions for the H2 molecule are of the form

ψMO ¼ c1 ψAOð1Þ þ c2 ψAOð2Þ (6.6)

where AO wavefunctions ψAO(1) and ψAO(2) on atoms 1 and 2 are multiplied by coefficients
c1 and c2. For a molecule of N constituent AOs, each MO wavefunction can be written:
X
N
ψðrÞ ¼ cj ψAO ðjÞ where ψAO(j) are individual AO wavefunctions, and the cj coefficients
j¼1

quantify the contribution of each AO to the MO wavefunction:

7
The relation used in X-ray crystallography would be a* = 1/a (less frequently a* = λ/a where λ is the wavelength of the
incident X-ray radiation). Given the sinusoidal character of the eikr term of the Bloch function, it is necessary to set
a* = 2π/a so the electron density |ψ(r)|2 retains the translational symmetry of the real-space lattice.
 6.1 The Band Structure of a Hydrogen-Atom Chain 205

Using the tight-binding model for our very large HN ring, we express the crystal-orbital
wavefunctions as:
X
N
ψðrÞ ¼ N 1=2 eikna ψAOðnÞ ðr  naÞ (6.7)
n¼1

where ψAO(n) is a 1s AO wavefunction (ψ1s) on the nth atom, the nucleus of which is located at
r = na, and the N−1/2 term acts to normalize the crystal-orbital wavefunction. We can see that the
eikna terms in the crystal orbital play the same role as the cj coefficients in an MO. The
contribution of each AO to the crystal-orbital wavefunction is weighted by the magnitude and
sign of this term. To visualize these crystal orbitals ψk for a 1D chain of hydrogen atoms,
consider two cases, k = 0:

ψk ¼ N 1=2 ½eikð1aÞ ψ1sð1Þ þ eikð2aÞ ψ1sð2Þ þ eikð3aÞ ψ1sð3Þ þ eikð4aÞ ψ1sð4Þ þ eikð5aÞ ψ1sð5Þ þ …
ψk¼0 ¼ N 1=2 ½e0 ψ1sð1Þ þ e0 ψ1sð2Þ þ e0 ψ1sð3Þ þ e0 ψ1sð4Þ þ e0 ψ1sð5Þ þ … (6.8)
ψk¼0 ¼ N 1=2 ½ψ1sð1Þ þ ψ1sð2Þ þ ψ1sð3Þ þ ψ1sð4Þ þ ψ1sð5Þ þ …

and k = π/a:
ψk¼π=a ¼ N 1=2 ½eiðπÞ ψ1sð1Þ þ eið2πÞ ψ1sð2Þ þ eið3πÞ ψ1sð3Þ þ eið4πÞ ψ1sð4Þ þ eið5πÞ ψ1sð5Þ þ …
(6.9)
ψk¼π=a ¼ N 1=2 ½ψ1sð1Þ þ ψ1sð2Þ  ψ1sð3Þ þ ψ1sð4Þ  ψ1sð5Þ þ …

where ψ1s(n) represents the 1s AO on the nth atom.

These two crystal orbitals, sketched in the manner used in Chapter 5, are shown in
Figure 6.3, where we can see how the phases of the AOs are altered by the value of k. The
two k values chosen, 0 and π/a, have the smallest and largest absolute values of k within the
first Brillouin zone, and correspond to the most bonding and most antibonding crystal
orbitals, respectively. Orbitals with intermediate values of k will have intermediate energies.
The wavefunctions ψ(r) for these two crystal orbitals and for k = π/2a are plotted in
Figure 6.4. Notice how the eikr term imparts a sinusoidal modulation to the wavefunction

k = π/a

k=0

n = 1 2 3 4 5 6

Figure 6.3 The most bonding (k = 0) and most antibonding (k = π/a) crystal orbitals for an infinite chain
of hydrogen atoms. The spheres represent individual H 1s orbitals, while dark and light shading indicates
the sign the AO wavefunction acquires from the eikr term; ψ > 0 is white and ψ < 0 gray.
206 Electronic Band Structure

k=0

λ=∞

× =
a

k = π/2a

λ= 4a

× =

k = π/a

λ= 2a

× =

u(r) eikr ψ(r)

Figure 6.4 Crystal orbitals derived within the tight-binding approximation at k = 0 (top), π/2a (middle),
and π/a (bottom), for an infinite linear chain of hydrogen atoms. The periodic function u(r) (left), is
composed of the H 1s AO wavefunctions (Figure 5.5) centered at H nuclei at the 1D lattice points (unit-
cell edge a) along the chain. The real parts of the eikr term (middle), and the full Bloch function (right)
obtained according to Equation (6.1), vary with k.8 The imaginary part of the wavefunction is not shown.

that yields its maximum (non-imaginary) value of 1 when kr = 0. It reaches this maximum
value again when kr = 2π, and so on for every integer multiple of 2π.9 The distance r = λ from
one maximum to another is the crystal-orbital wavelength:

8
Since the cos(x) function of a real number x completes its full period from x = 0 to x = 2π, it has a wavelength λ = 2π.
Consequently, cos(2x) has λ = π, and cos(x/2) has λ = 4π.
9
Because only the real part of ψ(r) is plotted, the function ψ2(r) that represents the electron density ρ(r) would appear
to change from cell to cell, but when the imaginary component is also considered this is not the case.
 6.1 The Band Structure of a Hydrogen-Atom Chain 207

λ = 2π/k (6.10)
For a free electron, we can use the de Broglie formula to relate its momentum p and
wavelength λ:

p ¼ h=λ (6.11)

If we insert the wavelength of the crystal orbital, λ = 2π/k, into Equation (6.11), we obtain
a relationship that expresses the momentum of the electron in a crystal in terms of k:

p ¼ hk=2π ¼ ℏk (6.12)

Because we are dealing with an electron in a crystal rather than a free electron, ℏ k is termed
the crystal momentum. Crystal momentum is not quite the same thing as momentum in the
classical sense, but, as we will see later, it plays an important role in both the electrical and
optical properties of a material.

6.1.5 Band-Structure Diagrams

Each crystal orbital has a specific energy E and crystal momentum ℏ k. We can represent
the electronic structure of the crystal by plotting the energies of its crystal orbitals as
a function of k to create what is called a band-structure diagram. The band-structure
diagram for our infinite chain of hydrogen atoms is plotted in Figure 6.5. For a chain of
finite length (i.e. a large ring), we can imagine the curve in this figure as a series of closely
spaced points, each representing a different crystal orbital. There will be one crystal orbital
for each hydrogen atom, just as there is one MO per hydrogen in the MO diagrams shown
in Figure 6.1. As the number of atoms in the crystal increases, the MO-energy points
become more closely spaced. For an infinite crystal, the curve in Figure 6.5 is a continuous
function, an infinite set of crystal-orbital points, each with the same u(r) but a different k;
a function that defines a band of allowed energies. When the unit cell contains more than
one atom, the band-structure diagram will have multiple bands, each derived from
a different u(r), as we will see in Section 6.2.
Despite its simplicity, there are several things we can learn from Figure 6.5. Firstly,
the number of bands is equal to the number of AOs in the unit cell. In this case, we
have one atom per unit cell and one AO per atom, which leads to a single band.
Secondly, we see that the crystal orbitals in the band run “uphill” in energy from k = 0
(bonding) to k = ±π/a (antibonding). The energy span between the top and bottom of
a band is the bandwidth, W.10
In Figure 6.5, the band-structure diagram is plotted for two different H–H distances.
Figure 6.5a corresponds to an interatomic distance of 1.0 Å, while Figure 6.5b shows how the

10
As W increases, we also say that the dispersion of the band increases. Wide bands are often called disperse bands.
208 Electronic Band Structure

20

en er g y (eV) 10

0 W

-10
W

-20
−π /a 0 π /a −π /a 0 π /a
k k

(a) H–H distance = 1.0 Å (b) H–H distance = 2.0 Å

significant orbital overlap: minimal orbital overlap:
wide band (W large) narrow band (W small)

Figure 6.5 The band-structure diagram for an infinite chain of H atoms where the interatomic spacing is
(a) 1.0 Å or (b) 2.0 Å. The crystal-orbital fragments shown on the left correspond to k = 0 and k = ± π/a.

band structure changes if the interatomic distance increases to 2.0 Å. In the former case, the
hydrogen atoms are close together, resulting in a high degree of orbital overlap.
Consequently, the bonding crystal orbital at k = 0 is strongly stabilized with respect to an
isolated H atom, while the antibonding crystal orbitals at k = ±π/a are highly destabilized,
leading to a very wide band of W ≈ 40 eV. The bandwidth is decreased by a factor of 10 when
the interatomic distance becomes 2.0 Å. Narrow bands result when there is little change in
the strength of the bonding/antibonding interactions as the value of k changes, either due to
poor spatial overlap of orbitals from one real-space unit cell to the next (as is the case for the
weakly bonded H-atom chain) or for reasons of symmetry. Electrons that occupy narrow
bands are typically highly localized, either on a single atom or a group of atoms within the
unit cell. As we will see in Chapter 10, bandwidth is an important parameter in determining
the conductivity of a material.
Another point to be made about the band structure of the infinite H-atom chain
concerns the shape of the band-structure diagram. The band structure from k = 0 to
π/a is a mirror image of the band structure from k = 0 to −π/a. This follows from the
mathematical properties of the eikr term in the Bloch function. Because of this sym-
metry, band structures are typically plotted only for positive values of k. We should
also note that the band is not symmetrically distributed about the energy of an isolated
H 1s orbital (E = −13.6 eV), and that the band is flatter at the bottom (near k = 0)
than at the top (near k = ±π/a). This asymmetry, which is more evident for the chain
where the atomic spacing is 1 Å (Figure 6.5), is a consequence of the fact that
antibonding states are destabilized more than bonding states are stabilized, as previ-
ously discussed for the MO diagram of H2+ (Section 5.3.1).
 6.1 The Band Structure of a Hydrogen-Atom Chain 209

6.1.6 Density-of-States (DOS) Plots

A band-structure diagram contains a considerable amount of information; it effectively
contains a MO diagram for each point in k space. In a 3D crystal, particularly one with
a complicated crystal structure, band-structure diagrams can become quite complex. The
additional complexity comes in part because there are more bands (one for each AO in the
unit cell), and in part because there are many more values of k to consider in three dimensions
than in one dimension.
A mechanism for depicting the electronic structure in a simpler form is to convert the
band-structure diagram to a density-of-states (DOS) plot. The DOS plot for the hydrogen
chain with an H–H distance of 2.0 Å is shown in Figure 6.6. The vertical axes of the DOS plot
and of the band-structure diagram are identical; they give the energies of the electronic states.
However, the horizontal axis of a DOS plot is different; it represents the density of states,
N(E), which is the number of allowed energy levels per unit volume of the solid in the energy
range E to E + dE, as dE goes to zero.
A peak in the DOS plot indicates a large number of crystal orbitals with similar energies
and is therefore related to the slope of the E versus k curve. Recall that for a large but
finite chain, the line in Figure 6.6 (left) comprises closely spaced but discrete crystal
orbitals. The range where the band flattens out (in this example near k = 0 and π/a) will
have many crystal orbitals with very similar energies, and the DOS will be high. Where the
band is steep (in this example near k = π/2a), there will be fewer crystal orbitals in the same
energy range and the DOS will be lower. We saw a preview of the double-peaked DOS for
the infinite H-atom chain in the MO diagrams for cyclic all-hydrogen molecules (Figure
6.1), where the MOs were more closely spaced near the bottom and top of the energy
range covered by MOs.

-10 -10

-12
en er g y (eV)

-12
empty states

EF EF
-14 -14 filled states

-16 -16
0 π /a
k DOS, N(E)

Figure 6.6 The band-structure (left) and density-of-states plot (right) for an infinite H-atom chain of 2.0 Å
separation at T = 0 K. EF is the Fermi energy.
210 Electronic Band Structure

DOS plots provide information about energy levels, filling, and widths of bands in
a manner that can be assimilated without knowledge of Bloch functions or an under-
standing of k space. In fact, for depicting the electronic structure of an extended solid,
the DOS plot is in many ways the closest equivalent to the MO diagrams discussed in
Chapter 5. We will see that it is possible to sketch approximate DOS plots from
knowledge of the structure, electron count, and orbital energies in the same way that
approximate MO diagrams can be constructed from the same data. Nonetheless, we
should not forget that useful information is lost upon transforming a band-structure
diagram into a DOS plot.
In band-structure diagrams, just as in MO diagrams, it is important to know which crystal
orbitals are occupied by electrons and which ones are empty. The energy level that separates
the filled states from the empty states at T = 0 K is called the Fermi energy, EF, or Fermi level.
In Figure 6.6, we see that the Fermi energy in our infinite hydrogen chain cuts the band in
such a way that half of the crystal orbitals have energies that fall below EF and half above it.
Such a half-filled band is exactly what we would expect because each H atom contributes one
orbital and one electron. Like the orbitals from which they are formed, each band can hold
two electrons (of opposite spin). Metallic conductivity is often observed in materials where
the Fermi level cuts through a band, leaving it partially occupied. We will return later to the
importance of band filling and the Fermi level on the electrical and optical properties of
materials.

6.2 The Band Structure of a Chain of H2 Molecules

You may have been surprised by the result of the preceding section; that an infinite hydrogen
chain would exhibit metallic conductivity. Although hydrogen is thought to become metallic
at very high pressures, under ambient conditions elemental hydrogen is certainly not metal-
lic. Furthermore, under ambient conditions, a linear chain of hydrogen atoms would not be
stable versus dimerization to form H2 molecules. Let’s take a closer look at how this
dimerization impacts the band structure.
For simplicity, we retain a 1D extended structure, but this time the chain will comprise H2
molecules lined up end to end. In the resulting structure, there are now two different H–H
contacts, a short intramolecular distance (d1 in Figure 6.7) and a longer intermolecular
distance (d2 in Figure 6.7). The new unit cell now holds two hydrogen atoms instead of one,
and we anticipate that the band-structure diagram will contain two bands instead of one.
The periodic function u(r) is more complicated in this case, but otherwise the analysis is
identical to the chain of hydrogen atoms. The basis-set functions to use for u(r) are the MOs
of the H2 molecule; a bonding MO ψ+ and an antibonding MO ψ− (Figure 5.9). Each MO will
give rise to a separate band. The orbital interactions at the center (k = 0) and edge (k = π/a) of
the first Brillouin zone are shown in Figure 6.7. We start by considering the band derived
from the bonding MO (ψ+). At k = 0, the intra- as well as intermolecular interactions are
 6.2 The Band Structure of a Chain of H2 Molecules 211

Figure 6.7 Crystal orbitals in an infinite chain of H2 molecules at k = 0 and k = π/a.

bonding, hence this is the lowest-energy crystal orbital (shown on the bottom in Figure 6.7).
Upon moving to k = π/a, intermolecular interactions become antibonding, whereas intra-
molecular interactions remain bonding. Consequently, the energy of the bonding band (ψ+)
rises; it runs “uphill” from k = 0 to k = π/a.
Next, we consider the band derived from the antibonding MO (ψ−). When k = 0, both the
intra- and intermolecular interactions are antibonding, and, as a result, this is the highest-
energy crystal orbital (shown on the top in Figure 6.7). At k = π/a, the intramolecular
interactions are still antibonding, but the intermolecular interactions are now bonding.
Hence, this band runs “downhill” in energy from k = 0 to k = π/a. Because the intramolecular
H–H distance is shorter than the intermolecular H–H distance (d1 < d2), the band derived
from the bonding MO (ψ+) will be lower in energy than the band derived from the antibond-
ing MO (ψ−) throughout the first Brillouin zone (Section 6.1.3).
The calculated band structure of our H2 molecular chain is shown in Figure 6.8. As
predicted, the lower band runs uphill in energy and the upper band runs downhill in energy.
Each hydrogen atom has one valence electron so there are two electrons per unit cell, enough
to fill the lower band while leaving the upper band empty. Therefore, the Fermi energy lies in
the energy gap between bands. Materials where the Fermi level does not cut through a band
are semiconductors or insulators, depending on the size of the gap.
In a basic sense, we see that the band-structure diagram of the H2 chain could have been
approximated by broadening each of the two H2 MOs into a band. This is our first lesson in how
to use the MO diagram of a chemical building unit as the starting point for approximating the
band structure. The width of the bands depends on the interaction between molecules in
212 Electronic Band Structure

10 10 10
conduction
ψ−
en er g y (eV)

energy (eV)
band

energy (eV)
0 0 0

-10 -10 EF -10 EF

ψ+
valence
band
-20 -20 -20
0 π /a DOS, N(E)
k

Figure 6.8 MO diagram of an isolated H2 molecule (left) together with the band-structure diagram for an
infinite chain of H2 molecules (middle) and the corresponding DOS (right). The Fermi energy EF is
denoted by a dashed line. In the DOS plot, shading indicates filled states. The crystal orbitals pictured are
those at k = 0 and k = π/a.

neighboring unit cells.11 If the intermolecular spacing is decreased, the bands will become wider
and the energy gap between bands smaller. As the intermolecular H–H distance becomes almost
the same as the intramolecular H–H distance, the bonding/antibonding interactions at k = π/a in
the upper and lower bands will be nearly equivalent, and the energy gap between the bands will
become infinitesimally small. When the H–H distances become equal, we revert to a single-atom
unit cell and the metallic band structure of the H-atom chain shown in Figure 6.5.12
We now see why a structure possessing alternating long and short H–H distances
would be more stable than one where the H–H distances are equal; the bond length
alternation creates a band gap, thus lowering the energy of the occupied band at the
expense of the empty band (Figure 6.8). The H-atom chain is not the only system to
undergo this type of distortion. Peierls’ theorem states that any 1D structure possessing
a partially filled band will undergo a distortion that leads to bond-length alternations,
splitting the partially filled band to create a gap between filled and empty bands.
Distortions of this type are called Peierls distortions. A classic real-world example is
the Peierls distortion in polyacetylene (Figure 6.9). If all C–C bonds were the same
length, the electrons in the π orbitals would be fully delocalized, as shown on the left-

11
The band formed from the antibonding MO (ψ−) is wider (more disperse) than the band formed from the bonding
MO (ψ+) because antibonding interactions are more destabilizing than bonding interactions are stabilizing.
12
If we were to artificially use the two-atom unit cell also for the chain where all H–H distances are equal, the real-
space unit cell would be twice that of the true one-atom unit cell, and the length of the first Brillouin zone would be
halved. The band-structure diagram corresponding to this artificial two-atom unit cell can be derived by drawing
a vertical line in the band plot of the H-atom chain (Figure 6.6, left) at k = π/2a and folding the plot over that line so
that the two wings overlap. The resulting band-structure plot shows two bands widely separated at k = 0, but
touching at the folding point (the new k = π/a); the limiting case of the band structure in Figure 6.8. This concept is
called band folding.
 6.3 Electrical and Optical Properties 213

Figure 6.9 The Peierls distortion in polyacetylene.

hand side of Figure 6.9. However, when the chain distorts to give alternating long and
short bonds, the π electrons localize as shown on the right-hand side, and the total
electronic energy of the system is lowered. We’ll examine the band structure of
polyacetylene in more detail in Section 10.5.1.

6.3 Electrical and Optical Properties

The electrical and optical properties of a substance follow directly from its band structure. In
this section, we look at some basic connections between band structure and properties. In
later chapters, we will revisit these topics in detail.

6.3.1 Metals, Semiconductors, and Insulators

Because the H-atom chain and the H2 molecular chain have different band structures, the
properties of these hypothetical chains will also be different. The electronic conductivity is
sensitive to the location of the Fermi level. When EF cuts through a band, as it does for the
H-atom chain, there is almost no energy cost for electrons to move from occupied states just
below EF to empty states just above EF. As a result, the electrons can move in response to the
application of an external driving force, such as an electric field or a temperature gradient.
This explains why metals readily conduct electricity and heat. In contrast, when the Fermi
energy does not cut through a band, as in the H2 chain, a non-negligible amount of energy is
needed to excite the electron from the filled band to the empty band. In this case, the electrical
behavior corresponds to that of a semiconductor or insulator. We will examine these classes
of materials more closely in Chapter 10.
Before going further, we need to develop some nomenclature for discussing semiconduct-
ors. For a semiconductor, the minimum energy difference between the filled band(s) and the
empty band(s) is called the band gap, Eg. Although there is not a precise value of Eg that
separates semiconductors from insulators, as a rough guideline, materials with Eg > 3 eV are
typically considered insulators. When discussing semiconductors/insulators, we will refer to
the filled bands as valence bands and the empty bands as conduction bands. Simple DOS
sketches of a metal, semiconductor, and insulator are shown in Figure 6.10. These block
sketches do not attempt to capture the shape of the various bands as they would be depicted
in an actual DOS plot. Instead, each band (or set of bands) is represented by a rectangle and
shading is used to show their filling. These crude sketches can be a useful tool for quick
approximations of the electronic structure.
214 Electronic Band Structure

metal semiconductor insulator

conduction
band

energy Eg
Eg
EF

valence
band

Figure 6.10 Schematic DOS diagrams for a metal (left), a semiconductor (middle), and an insulator
(right). Occupied crystal orbitals are shaded.

The shading used in the DOS plots in Figure 6.10 is, strictly speaking, only valid when the
temperature is equal to absolute zero (T = 0 K). At finite temperatures, the electron
distribution smears out as some of the electrons are thermally excited from states below EF
to states above EF. This effect has important consequences for the conductivity of a material
and will be considered in detail in Chapter 10.

6.3.2 Direct- versus Indirect-Gap Semiconductors

In Figure 6.8, it is easy to see that the energy separation between the upper and lower bands
changes as a function of k. At k = 0 the two bands are separated by ~34 eV, while at k = π/a,
only by ~10 eV. The band gap is defined as the minimum energy separation between the
highest energy state in the valence band and the lowest energy state in the conduction band,
~10 eV in this case. When the valence-band maximum and the conduction-band minimum
occur at the same value of k, as they do in this example, the material is said to be a direct-gap
semiconductor. When these two points fall at different values of k, the material is an indirect-
gap semiconductor. Schematic representations of these two cases are given in Figure 6.11.
To excite an electron from one band to another requires a transfer of energy from the
incoming photon to the electron. If k does not change upon moving from one band to another
(i.e. the transition is a vertical line in Figure 6.11), the transition only involves transfer of
energy. If k changes, however, it is necessary to alter not only the energy but also the
momentum of the electron, Equation (6.12). The mechanism for doing this requires transfer
of momentum between a lattice vibration (a phonon, see Section 4.4.6) and the electron as the
photon is absorbed. Because this is effectively a three-body process, light is absorbed much
 6.4 Representing Band Structures in Higher Dimensions 215

Figure 6.11 The lowest-energy valence to conduction band transition in a direct-gap (left) and indirect-
gap (right) semiconductor.

more efficiently in a direct-gap semiconductor than it is in an indirect-gap semiconductor. This

has important consequences for the design of optical devices, as we will see in Chapter 7.

6.4 Representing Band Structures in Higher Dimensions

Although there are a few materials whose band structure can be approximated by a 1D model,
most materials need to be treated in two or three dimensions. The 2D case is not just
a conceptual intermediate on the way to the 3D model, it applies to a number of interesting
materials. Surfaces are 2D entities, and their electronic structures are useful constructs for
understanding the chemical processes that occur there, such as heterogeneous catalysis. Many
compounds are made up of covalently bonded layers that are held together by ionic or
dispersion forces. The electronic structures of such materials are largely 2D in character.
Other materials are 2D solids in a strict sense, graphene being the best-known example.
The underlying concepts of band theory don’t change upon increasing the dimensionality, but
the mathematics of the analysis and the representation of the results become more complicated.
We will leave the computational details to computers, but it’s not possible to discuss 2D and 3D
systems without first developing a basic understanding of how they are represented.

6.4.1 Crystal Orbitals in Two Dimensions

Let’s start by recalling that the position of any lattice point in a 2D crystal is given by a real-
space translation vector, T:

T ¼ ua þ vb (6.13)
216 Electronic Band Structure

b*
b

a a*

real-space lattice reciprocal-space lattice

Figure 6.12 The 2D real-space rectangular lattice (left) and its reciprocal-space lattice (right). The shaded
rectangle represents the first Brillouin zone. Note that if b > a in real space, b* < a* in reciprocal space.

where a and b are the basis vectors of the real-space lattice (Chapter 1.1) while u and v are
integers. The real-space unit cell is defined by u = v = 1. The corresponding reciprocal-space
lattice vector, G, is given by:

G ¼ ha þ kb (6.14)

where a* and b* are the basis vectors of the reciprocal-space lattice, while h and k are
integers. In two dimensions, the reciprocal-space lattice vector a* must be perpendicular to
the real-space vector b, and b* must be perpendicular to a. The magnitudes of the reciprocal-
space lattice vectors are defined so that the dot products a∙a* = b∙b* = 2π.
For a 2D rectangular lattice, the reciprocal-space lattice vectors a* of magnitude 2π/a and
b* of magnitude 2π/b are shown in Figure 6.12. As in one dimension, the first Brillouin zone
contains the complete range of k values needed to generate all possible crystal orbitals
without duplication. It is the region in reciprocal space that surrounds k = 0 and contains
all k points that are closer to this point than to any other reciprocal-space lattice point. To
determine the boundaries of the first Brillouin zone, we bisect the reciprocal-space lattice
vectors around one of the lattice points by perpendicular lines.
Our next example is the 2D hexagonal lattice shown in Figure 6.13. To construct the first
Brillouin zone, we first draw lines from an arbitrary lattice point chosen as the origin to each
of its six nearest neighbors. The perpendicular bisectors of these lines define a hexagon that
contains all points closer to k = 0 than to any other reciprocal-space lattice point. Unlike
crystallographic unit cells, the first Brillouin zone is not necessarily a parallelogram in two
dimensions (or a parallelepiped in three dimensions), as this example illustrates.13

13
In crystallography, a locus of points in space closer to a given lattice point than to any other lattice point is called
a Wigner–Seitz cell.
 6.4 Representing Band Structures in Higher Dimensions 217

b
a
b*
a*

real-space lattice reciprocal-space lattice

Figure 6.13 The 2D real-space hexagonal lattice (left) and its reciprocal-space lattice (right). On the right-
hand side, thin lines connect the reciprocal-space lattice point defined as k = 0 with the six neighboring
lattice points, and the shaded hexagon represents the first Brillouin zone.

The real-space hexagonal-lattice vectors a and b have same lengths (|a| = |b|) and the angle
between them is 120°. For calculations in Cartesian coordinate space, we can express them in
terms of orthogonal unit-length vectors that are perpendicular (^x ) and parallel (^y ) to b,14 as
a = [(|a|√3/2)]^x − [|a|/2]^y and b = |b| ^y . Using the relationships a* ⊥ b and b* ⊥ a, one can
show that the angle between reciprocal-space lattice vectors is 60° (Figure 6.13). In terms of
unit vectors in Cartesian coordinate space, the hexagonal reciprocal-space vectors are a* =
[4π/(|a|√3)]^x and b* = [2π/(|a|√3)]^x + [2π/|b|]^y .
In a 2D crystal, the electron wavefunction, which was ψ(r) = eikru(r) in one dimension
(Section 6.1.2), becomes:
ψðrÞ ¼ eikr uðrÞ (6.15)

where r is the position vector or radius vector r = xa + yb (Section 1.1.1) of any point in the
crystal in fractional coordinates x and y, and u(r) is the lattice-periodic function describing
the electric-field potential felt by an electron in the real-space crystal. Whereas k was a scalar
in 1D space confined to the interval −a*/2 ≤ k ≤ +a*/2, in 2D space k becomes a vector, but its
allowed values are still confined to the first Brillouin-zone interval −a*/2, − b*/2 ≤ k ≤ a*/2,
b*/2. It can be expressed with fractional coordinates −½ < m, n < ½ as k = ma* + nb* and
is referred to as the wave vector.15 It gives not only the wavelength of the crystal orbital,
λ = 2π/|k|, but also the direction in which the eik·r term modulates the wavefunction.16
To show the entire electronic structure of a 2D material, we need a 3D plot, two dimen-
sions for k and one for energy. Fortunately, the maxima and minima of many bands are
generally found either at the center of the Brillouin zone or at one of its boundaries.

14
Their lengths are j^x j = j^y j = 1 chosen unit-length in real space or 1 chosen unit-length−1 in reciprocal space.
15
The wave vector is perpendicular to the wave fronts. In two dimensions, the wave fronts are parallel lines that run
through points where the crystal-orbital wavefunction takes a constant value.
16
See Figure 6.17 (right) for a visual illustration of how the sign of the AO wavefunction is modulated by different
values of the wave vector k.
218 Electronic Band Structure

* *
* *
* *

Figure 6.14 The first Brillouin zone and special symmetry points for the 2D square lattice. The magnitudes
of the reciprocal lattice vectors are |a*| = |b*| = 2π/|a|.

Γ Γ
ŷ

b*
K′ M K K′ M K

a* x̂

label wave vector label wave vector

Γ 0 a* + 0 b* Γ 0 xˆ + 0 yˆ
M (1/2) a* + 0 b* M (2π 3a ) xˆ + 0 yˆ
K (1/3) a* + (1/3) b* K (2π 3a ) xˆ + (2π 3a ) yˆ
K′ (2/3) a* − (1/3) b* K′ (2π 3a ) xˆ − (2π 3a ) yˆ

Figure 6.15 The first Brillouin zone and special symmetry points for the 2D hexagonal lattice in terms of
the reciprocal-space lattice vectors a* and b* (left), and the orthogonal axes (right).

Therefore, we can get a good sense of the band structure by plotting the energy as a function
of k along lines that run between various high-symmetry points in the Brillouin zone.17 For
a 2D square lattice, these points are given in Figure 6.14.
The high-symmetry points within the first Brillouin zone of a 2D hexagonal lattice are
shown in Figure 6.15. These points are expressed in terms of the reciprocal-space lattice
vectors a* and b* on the left-hand side of the figure and in terms of the orthogonal unit-
length vectors on the right-hand side of the figure.

17
The Γ point has all point-symmetry elements of the lattice, while the other special points retain some but not all
symmetry elements.
 6.4 Representing Band Structures in Higher Dimensions 219

primitive cubic face-centered cubic

primitive orthorhombic primitive hexagonal

Figure 6.16 3D Brillouin zones of lattices reciprocal to four selected 3D real-space lattices, with high-
symmetry points identified and orthogonal axes drawn.

6.4.2 Crystal Orbitals in Three Dimensions

In three dimensions, the crystal-orbital wavefunctions are completely analogous to the 2D case,
Equation (6.15), but the radius vector r = xa + yb + zc and wave vector k = ma* + nb* + oc* are
given in three dimensions. The reciprocal-space basis vectors are defined in terms of the real-
space lattice vectors by the following relationships:

2π 2π 2π
a ¼ ðb  cÞ b ¼ ðc  aÞ c ¼ ða  bÞ (6.16)
V V V

where V is the real-space unit-cell volume, V = a∙(b × c). From the definition of a vector cross
product,18 a* is perpendicular to b and c, b* is perpendicular to a and c, and c* is perpen-
dicular to a and b. For a cubic lattice with a real-space unit-cell edge a, the reciprocal-space
lattice is also cubic with a cell edge of magnitude a* = 2π/a. Notice that as the real-space unit
cell gets larger, the reciprocal-space cell gets smaller.
We construct the first Brillouin zones of 3D reciprocal-space lattices using the same
approach as in two dimensions, but they are now polyhedra instead of polygons. The first
Brillouin zones for reciprocal-space lattices that correspond to some common 3D real-space
lattices are shown in Figure 6.16, with the high-symmetry points labeled. Those Brillouin
zones derived from primitive real-space lattices have simple shapes expected from the 2D

18
The cross product of two vectors (e.g. b × c) is a vector perpendicular to the vectors from which it is formed with
a magnitude equal to the area of the parallelogram formed by the original two vectors.
220 Electronic Band Structure

analogy. For the face-centered cubic (fcc) lattice, one needs to realize, or simply accept,
that its reciprocal-space lattice is body-centered cubic (bcc). The first Brillouin zone is
then constructed just as it is in one or two dimensions, by bisecting lines connecting
neighboring reciprocal lattice points with perpendicular planes that form the faces of
a polyhedron, which in this case is the truncated octahedron shown in the top right of
Figure 6.16.

6.5 Band Structures of Two-Dimensional Materials

Let’s begin with a simple hypothetical example: the band structure of a 2D square lattice of
hydrogen atoms (Figure 6.17). Using a tight-binding approach (Section 6.1.2), we take u(r) =
ψ1s(r), the H 1s AO wavefunction. At the Γ point, defined as k = 0a* + 0b*, the eik·r term is
equal to 1 for any value of r (eik·r = e0 = 1), as is the coefficient associated with each
H 1s orbital in the 2D Bloch function, Equation (6.15). The resulting crystal-orbital wave-
function is the 2D analog of the 1D wavefunction at k = 0, Equation (6.8). Since the AOs on
neighboring H atoms all have the same phase (their coefficients in Equation (6.15) have the
same sign), they interfere constructively. This condition maximizes the bonding overlap
and therefore corresponds to the lowest-energy point in the first Brillouin zone. At
X (k = ½ a* + 0 b*), the H 1s orbital wavefunction contribution alternates between positive
and negative (due to the changing sign of the eik·r term) in the real-space x direction, but remains

y
55 M
x

35
energy (eV)

15
EF

-5 ┏

-25
Γ X M Γ

Figure 6.17 The band structure of a 2D square lattice of hydrogen atoms at a 1 Å separation. The crystal
orbitals at Γ, X, and M (defined in Figure 6.14) are shown on the right in real-space coordinates. The
spheres represent individual H 1s orbitals, while dark and light shading indicates the sign the AO
wavefunction acquires from the eikr term; ψ > 0 is white and ψ < 0 gray.
 6.5 Band Structures of Two-Dimensional Materials 221

constant in the real-space y direction.19 The energy goes up along the Γ to X line because
half of the nearest-neighbor H–H interactions change from bonding to antibonding.
Finally, the energy reaches a maximum at the M point (k = ½ a* + ½ b*) where all nearest-
neighbor interactions are antibonding. Note that the overall bandwidth W is higher in two
dimensions than in one dimension because the number of bonding/antibonding inter-
actions per H atom has increased.

6.5.1 Graphene
Graphite is one of the most familiar layered materials. It contains planar sheets of sp2
hybridized carbon atoms arranged in a 2D honeycomb pattern (or, if you prefer, chicken
wire). In graphite, the layers are stacked in an ordered way to form a 3D structure; however,
when a single layer of carbon atoms from graphite is isolated, the 2D entity is called
graphene. While it is easy to imagine a graphene sheet, it is quite challenging to isolate layers
that are only a single atom thick. It was not until 2004 that researchers found a route to single
graphene layers [1]. The discovery of graphene created tremendous excitement, leading to
a surge in the study of 2D materials.
The structure of graphene is shown in Figure 6.18. There are two atoms in the hexagonal
unit cell. Each carbon has four valence orbitals (2s, 2px, 2py, 2pz), so we expect a total of eight
bands in the band structure. Our earlier look at the bonding in benzene (Section 5.3.7) is
instructive. Just as in benzene, the 2s, 2px, and 2py orbitals in graphene overlap to form σ

b
*
a -5
energy (eV)

EF

-15
3

2

-25 1

-35
K  M

Figure 6.18 Left: A segment of a graphene sheet with the hexagonal unit-cell rhombus. Right: The band
structure of graphene with σ bands in gray and π and π* bands in black. The three empty σ * bands located
at higher energy are not shown.

19
Imagine the 2D wave at X like a corrugated roof with peaks and troughs parallel with y.
222 Electronic Band Structure

bands while the 2pz orbitals overlap to form π bands. By analogy with the MO diagram of
benzene, we would expect the π band to be the highest-energy valence band and the π* band
to be the lowest-energy conduction band—the respective solid state equivalents of the
highest occupied MO (HOMO) and lowest unoccupied MO (LUMO). Hence, we expect
the properties of graphene to be dominated by these bands.
The 2D band structure of graphene is shown on the right-hand side of Figure 6.18. We see
that over much of the first Brillouin zone, the σ bands are more stable than the π and π*
bands, as expected.20 To understand the orbital overlap that gives rise to the π bands, we
need only consider a single 2pz orbital per atom. Because there are two atoms per unit cell,
there are two bands that originate from the 2pz orbitals; one of them from the π-bonding MO
and one from the π-antibonding MO of the two-atom motif (the atom or group of atoms
associated with each lattice point, Figure 1.1).
The interaction (bonding, antibonding, nonbonding) between carbon atoms in neigh-
boring unit cells is dictated by the value of k, as shown in Figure 6.19. At Γ, eik∙r = 1, hence
all nearest-neighbor interactions in the crystal orbital derived from the π MO are bonding,
while in the corresponding π* MO they are antibonding (Figure 6.19, left). This combin-
ation maximizes the energy separation between the π and π* bands. At M, there are two
bonding and one antibonding nearest-neighbor interactions for the π band and vice versa
for the π* band (Figure 6.19, middle), reducing the energy separation between the two
bands (Figure 6.18).

 M K

π* band

π band

Figure 6.19 The overlap of 2pz orbitals in the π (bottom) and π* (top) bands of graphene at the Γ (left),
M (middle), and K (right) points identified in Figure 6.15. The sizes of the orbitals are drawn proportional
to the magnitude of the crystal-orbital coefficients eik·r. For the crystal orbitals at K, only the real part of
the wavefunction is shown.

20
This condition is not met near Γ where the energies of the σ2 and σ3 bands reach a maximum. At Γ, mixing (i.e. any
bonding or antibonding interaction) between 2s orbitals and 2px/2py orbitals is symmetry forbidden (see Section
6.5.2), thus the σ1 band has pure C 2s character while the σ2 and σ3 bands have pure C 2px/2py character.
 6.5 Band Structures of Two-Dimensional Materials 223

Inspection of the band structure (Figure 6.18) shows that the filled π and empty π* bands
become degenerate at K. A more detailed analysis reveals that the crystal orbitals for both
bands are nonbonding at K. The Fermi level is located at the energy where these bands touch.
This leads to an unusual electronic structure where the valence bands and conduction bands
touch but do not overlap. Materials that possess such an electronic structure are said to be
semimetals.21 We will explore the properties of graphene in more detail in Section 10.6.1.

6.5.2 CuO22− Square Lattice

The CuO22− square lattice (Figure 6.20) is another illustrative and interesting 2D example, as
this layer is present in nearly all high-Tc cuprate superconductors. Within each layer, the
copper atoms are surrounded by four oxygen atoms forming squares that share corners, and
each oxygen atom is linearly coordinated by copper; the Niggli formula is CuO4/22−.
We can approximate the DOS plot from the MO diagram of square-planar CuO46− shown
on the left-hand side of Figure 6.21. The five highest-energy MOs originate from antibonding
Cu 3d–O 2p interactions. Highest is the antibonding interaction between Cu 3dx2−y2 orbitals
and O 2p orbitals that point directly at each other. The other four antibonding Cu 3d-based
MOs are ~2 eV lower in energy.22 The remaining MOs, located between −14 eV and −16 eV,
are either Cu 3d–O 2p bonding MOs or O 2p nonbonding MOs. To facilitate the discussion,
we will group the MOs into three separate energy regions, marked I, II, and III in Figure 6.21.
The contribution of each Cu 3d orbital to the total DOS, the so-called partial density of
states (PDOS), is plotted on the right-hand side of Figure 6.21. We can infer from the small

Cu

Figure 6.20 A segment of the 2D CuO4/22− sheet. The unit cell is shown in black.

21
In a strict sense, a semimetal is a material whose electronic density of states goes to zero at EF but is non-zero for any
finite energy above or below EF. This occurs whenever the conduction and valence bands exactly touch, as they do in
graphene. The term semimetal also applies in a more general manner to systems where the valence and conduction
bands overlap by a small amount, generally at different k points. In graphite, interactions between layers cause the π
and π* bands to overlap slightly.
22
The MO diagram of square-planar CuO46− exhibits d-orbital energy levels similar to those seen for a transition-
metal ion in an elongated octahedral environment (Section 5.3.10).
224 Electronic Band Structure

MO diagram t otal DOS Cu 3d partial DOS

III
-11
energy (eV) EF EF

-13 II

-15

-17

Figure 6.21 The MO diagram of a square-planar CuO46− unit (left), the total-DOS plot (middle), and
a partial-DOS plot for the infinite CuO4/22− layer showing the contributions of the individual Cu
3d orbitals (right).

degree of Cu character that region I, between −17.0 eV and −14.3 eV, is dominated by
oxygen 2p orbital contributions. The Cu contributions in the crystal orbitals of the lower and
middle sections of region I have Cu 3d–O 2p bonding character. The electronic states near the
top of this region, where the Cu contribution is minimal, are best described as
O 2p nonbonding states.
The Cu 3d orbitals make the dominant contribution to regions II and III of the DOS plot. The
dz2, dxz, dyz and dxy orbitals contribute heavily to region II (−14.2 eV to −13.0 eV). These bands
originate from antibonding Cu 3d–O 2p MOs. As the dz2 orbital has a rather weak overlap with
oxygen in the xy plane, it produces a very flat band, which in turn gives rise to a sharp peak in the
DOS. The dxz, dyz, and dxy orbitals have a π* interaction with the oxygen 2p orbitals, the strength
of which depends on k. Finally, region III (−12.1 eV to −11.0 eV) originates from the antibond-
ing Cu 3dx2−y2–O 2p σ* MO.
The full band structure for the CuO4/22− layer is shown in Figure 6.22. The number of
bands can be determined by considering the number of AOs in the motif. The O 2p-orbital
contribution is dominant in region I. We see a total of six bands in this region because there
are two oxygen atoms per unit cell, each with three 2p orbitals.23 The Cu 3d-orbital
contribution is dominant in regions II and III. Hence, we see four bands in region II (one
for each of the doubly occupied Cu 3d orbitals dz2, dxz, dyz, and dxy), and a single half-filled
band associated with the Cu 3dx2−y2 orbital in region III.

23
The square-planar CuO46− unit contains four oxygen atoms and hence there are 4 × 3 = 12 MOs in region I, whereas
the band structure is for a CuO22− sheet which contains two oxygen atoms per unit cell. Hence there 2 × 3 = 6 bands
in region I.
 6.5 Band Structures of Two-Dimensional Materials 225

Once again, we can understand the shape of the bands in Figure 6.22 by considering the
orbital overlap at various points in the Brillouin zone. As this is our first example with more
than one element, let’s take a closer look at how the mixing of copper and oxygen orbitals
varies with changes in k. Consider in Figure 6.23 the band that originates from the anti-
bonding interaction between Cu 3dxy and O 2p orbitals (this band spans −14.2 eV to
−13.0 eV in Figure 6.22). Because this is a π* band, we choose an antibonding interaction
between the Cu 3dxy orbital and two O 2p orbitals within the CuO2 motif (the orbitals that
contribute to the motif are marked with dashed ovals).

III
-11
Cu dx2−y2 EF
EF
(σ* band)
energy (eV)

Cu dz2 band
-13 II
Cu dxy,dxz,dyz
(* bands)

-15
O 2p bands
I

-17
 X M 
total DOS

Figure 6.22 The electronic band structure (left) and the DOS plot (right) of a CuO4/22− layer.

y Γ

x
Mixing is symmetry
forbidden
+
Cu 3dxy and O 2p
remain nonbonding

y M

Cu 3dxy O 2p π* crystal orbital

Figure 6.23 The mixing (or lack thereof) of the Cu 3dxy and O 2p orbitals at Γ and M (defined in Figure
6.14). The orbitals associated with a single CuO2 motif are enclosed in dashed ovals.
226 Electronic Band Structure

Once we have chosen the appropriate orbitals for the motif, the u(x,y) basis set is fixed for
all crystal orbitals that belong to this band. The next step is to determine the phases with
which the AOs contribute to the crystal orbital (as dictated by the eik(x,y) term of the Bloch
function) at a few high-symmetry values of k. At Γ, the phases of the AOs of the motif are
invariant on moving from one unit cell to the next (see Figure 6.23), and we see that the Cu
3dxy and O 2p orbitals have an equal amount of bonding and antibonding overlap (along
both directions). When this happens, orbital mixing (any bonding or antibonding inter-
action) is said to be (translational-) symmetry forbidden, and two nonbonding crystal
orbitals result. One is derived from the O 2p AO (in Figure 6.22, region I, amongst the “O
2p bands”), the other from the Cu 3dxy AO (the minimum-energy crystal orbital of the Cu
3dxy–O 2p π* band in Figure 6.22).24 At M, the phases with which the AOs contribute
to the crystal orbital are such that each O 2p orbital has antibonding interactions
with both Cu neighbors, and the crystal orbital has Cu 3dxy–O 2p π* character.
Using the same approach, we can draw crystal orbitals for the Cu 3dx2−y2–O 2p σ* band
(−12.1 eV to −11.0 eV) and compare them with the Cu 3dxy–O 2p π* crystal orbitals at Γ, M,
and X (Figure 6.24). Although mixing between the Cu 3dx2−y2 and O 2p orbitals is
also symmetry forbidden at Γ (not shown in Figure 6.24), this crystal orbital is not strictly

Γ X M
y

x
Cu 3dx2−y2 − O
σ* band

x
Cu 3dxy − O
π* band

Figure 6.24 Orbital interactions for the σ* (top) and π* (bottom) bands at the Γ (left), X (middle), and
M (right) points (defined in Figure 6.14).

24
The nonbonding O 2p crystal orbital is lower in energy because an isolated O 2p orbital has a lower energy than an
isolated Cu 3d orbital, see Section 5.2.1.
 6.6 Band Structures of Three-Dimensional Materials 227

Cu 3dx2−y2 nonbonding because the O 2s orbitals we have not discussed up to this point have
the proper symmetry to interact with Cu 3dx2−y2. This interaction (see upper left panel of
Figure 6.24) is weaker than the Cu 3d–O 2p interaction due to the poor energetic overlap of
the Cu 3d (−14 eV) and O 2s orbitals (−34 eV), yet strong enough to affect the width of this
band. Specifically, the σ* band would become nonbonding at Γ (Figure 6.22), lowering its
energy and increasing bandwidth, were it not for this antibonding 3dx2−y2–O 2s interaction.
At X, the interactions of Cu 3dxy and Cu 3dx2−y2 orbitals with the O 2p orbitals become
antibonding in the x direction (Figure 6.24, middle), while remaining symmetry forbidden
(nonbonding) in the y direction. At M, the Cu 3d–O 2p interactions become antibonding in
both directions (Figure 6.24, right). Because the energy of a band increases as the number of
antibonding interactions increases, these Cu 3dxy π* and Cu 3dx2−y2 σ* bands run uphill from
Γ to X to M as shown in Figure 6.22.25
Since the optical, magnetic and electrical properties are most sensitive to the bands near the
Fermi level, the filling and width of the Cu 3dx2−y2 band is critical. Given the d 9 electron
configuration of Cu2+, it should not come as a surprise that the Fermi level cuts the Cu 3dx2−y2
band in half (Figure 6.22). As discussed in Section 6.3.1, the presence of a partially filled band
should lead to metallic conductivity. However, this picture changes once electron–electron
interactions are considered, as we will learn in Section 10.4. These interactions stabilize an
antiferromagnetic, insulating ground state in compounds containing CuO22− layers, such as
La2CuO4. Interestingly, when electrons are added to or taken away from the layers, the
properties can change dramatically, as we will see when we take up the topic of superconduct-
ivity in Chapter 12.

6.6 Band Structures of Three-Dimensional Materials

Finally, we are ready to tackle the band structures of 3D materials.26 In this section, we will
consider several materials that build on the principles already established for lower-
dimensional systems, most of which represent important classes of functional materials.

6.6.1 α-Polonium
Perhaps the simplest 3D structure we can imagine is a primitive cubic lattice with a single
atom per unit cell. The only element that crystallizes with this structure is α-Po (Figure 1.23).
The band structure for α-Po is shown in Figure 6.25. There are four bands, one for each of the
valence orbitals of Po; 6s, 6px, 6py, and 6pz. The orbital character of the lowest-energy band is
exclusively Po 6s (due in part to relativistic effects mentioned in Section 5.2.1). Its shape,

25
At the M point, the Cu 3dxy π* band has the highest energy of the four bands that make up region II.
26
In the band-structure diagrams for all 3D materials, the energy scale is arbitrarily set so that the highest filled
electronic state falls at E = 0, as is the customary practice for density-functional theory used to generate these
diagrams.
228 Electronic Band Structure

5
Γ X R

0 EF pz
energy (eV)
-5

px
-10

-15 s
z
y
-20
Γ X M R Γ x

Figure 6.25 The band structure of α-Po, showing the orbital overlap at the Γ, X, and R points (defined in
Figure 6.16) for the 6s, 6px, and 6pz bands. The effects of spin–orbit coupling have not been included in
this calculation.

running uphill from Γ to X to M to R, is the 3D analogue of the band structure of a 2D square

lattice of hydrogen atoms (Figure 6.17).
The three higher-energy bands originate from Po 6p orbitals. These bands are two-thirds
filled, and they are therefore cut by the Fermi level. The right-hand side of Figure 6.25 shows
the crystal orbitals associated with the 6s, 6px, and 6pz bands at Γ, X, and R. At Γ, all three
6p bands are degenerate; each exhibiting strong σ-antibonding interactions with two neighbor-
ing Po atoms and weak π-bonding interactions with their remaining four nearest neighbors.
The 6p bands are also degenerate at the R point, where the σ interactions have become strongly
bonding and the π interactions weakly antibonding. At other points in the first Brillouin zone,
the degeneracy of the 6p bands is lifted. For example, the 6px band reaches a minimum at the
X point where both σ and π nearest-neighbor interactions are bonding. In contrast, the 6py and
6pz bands experience strong antibonding nearest-neighbor σ interactions at the X point.
Bismuth lies one element to the left of polonium in the periodic table. The 6p orbitals
are now half filled. The structure of Bi is related to the primitive cubic structure of α-Po
(Figure 6.26), but distorted so that each atom makes three short bonds (d1 = 3.07 Å) and
three long bonds (d2 = 3.53 Å) to its neighbors. This distortion has a similar origin as the 1D
Peierls distortion illustrated in Figure 6.9. An even more pronounced distortion of this type is
seen in lighter group-15 elements, such as gray arsenic (d1 = 2.51 Å, d2 = 3.15 Å).

6.6.2 Diamond
The crystal structure of diamond has eight atoms in the face-centered cubic unit cell
(Figure 1.16 and 1.37), but only two in the equivalent primitive cell (at 0 0 0 and ¼ ¼ ¼ of
the rhombohedron representing the primitive cell) normally used for band-structure
 6.6 Band Structures of Three-Dimensional Materials 229

α -polonium (cubic) bismuth (rhombohedral)

Figure 6.26 The structure of α-polonium (left) showing eight unit cells compared with an analogous
fragment of the bismuth structure (right). The short and long bonds in bismuth are marked with solid and
dashed lines, respectively.

18

conduction bands
Γs*

(Γs*) 2s−2s antibonding

10 Γp*
energy (eV)

Eg = 5.5 eV
2
(Γp*) 2p−2p antibonding
Γp
valence bands

-6

(Γp) 2p−2p bonding

-14

Γs
-22
W L Γ X W (Γs) 2s−2s bonding

Figure 6.27 The band structure of diamond. Right: Crystal orbitals at the Γ point (only one of the three
degenerate crystal orbitals present at Γp and Γp* is shown).

calculations. Using the smaller primitive cell, eight bands result from overlap of the
2s and 2p valence orbitals on carbon (Figure 6.27). The eight valence electrons (four
per atom) completely fill the lower four bands. Because the interatomic distances are
short (1.545 Å), there is a large degree of orbital overlap stabilizing the filled valence
bands, which are bonding, and destabilizing the empty conduction bands, which are
antibonding. The net result is a large band gap, 5.5 eV. Diamond has an indirect band
gap as the valence-band maximum (at Γ) and the conduction-band minimum (near X)
occur at different values of k.
230 Electronic Band Structure

Let’s take a closer look at the orbital character of the bands at the Γ point. Just as the
2s and 2p orbitals contribute to different MOs in the MO diagram of CH4 (Section 5.3.6),
the 2s and 2p orbitals in diamond contribute to different bands at the Γ point. This lack of
2s and 2p mixing yields just four energies at Γ: singly degenerate crystal orbitals corresponding
to bonding and antibonding overlap of the 2s orbitals, labeled Γs and Γs* in Figure 6.27, and
triply degenerate crystal orbitals arising from bonding and antibonding overlap of the
2p orbitals, labeled Γp and Γp*. The large energy separation between the bonding 2s (Γs) and 2p
(Γp) states results in part from the C 2s orbitals being more stable than the C 2p orbitals by
~8 eV (see Table 5.2) and in part from the fact that the 2s orbital overlap is bonding in all
directions at Γ. In fact, the 2s orbital overlap at Γ is so large that the antibonding 2s crystal
orbital (Γs*) is higher in energy than the antibonding 2p crystal orbital (Γp*).
Moving away from the Γ point, 2s–2p mixing occurs because the non-zero k alters the
phases of the AO contributions to the crystal orbital in a manner that allows the 2s orbitals to
have a non-zero overlap with 2p orbitals on neighboring atoms. Unlike graphene, where the
energy of the 2pz-bonding band reaches a minimum at Γ, here the energies of all three filled
2p bands are stabilized on moving away from Γ. The presence of 2s–2pz mixing throughout
most of the first Brillouin zone in diamond (but not at Γ) is responsible for the fact that
diamond is an insulator while graphite is a semimetal. We see that alternative crystal
structures can produce dramatically different band structures, leading to materials with
strikingly different properties.

6.6.3 Elemental Semiconductors

Semiconductors play a central role in modern electronic devices, and no semiconductor is more
widely used than silicon. The band structures of silicon and its group-14 neighbor germanium
are shown in Figure 6.28. They are qualitatively similar to diamond, but the energies and widths
of the various bands have changed. The increase in interatomic distance (2.35 Å for Si and
2.45 Å for Ge) reduces the orbital overlap, which in turn reduces the stabilization of the valence
bands and the destabilization of the conduction bands. Consequently, the band gaps are much
smaller than that of diamond, 1.1 eV for Si and 0.7 eV for Ge. There are also changes in the
relative energies of the antibonding crystal orbitals at Γ. The decrease in orbital overlap reduces
the splitting between bonding and antibonding states, lowering the energy of Γs* with respect to
Γp*.27 As a result, by the time we reach germanium, Γs* lies below Γp*.
The orbital overlap decreases further upon moving to α-Sn (dSn–Sn = 2.81 Å), where Γs* is
only marginally higher in energy than the valence-band maximum Γp. Not only does this
further reduce the band gap (Eg = 0.1 eV), it reduces the overall stability of the structure. In
fact, tin has two polymorphs; gray tin (α-Sn) is a semiconductor with the diamond structure,
whereas white tin (β-Sn) is a metal with a body-centered tetragonal structure. Moving to the
27
Orbital overlap is the most important factor in elemental semiconductors for dictating the relative energy of
the bands at Γ because the difference in the energies of the s and p orbitals is almost constant from C to Sn (see
Table 5.2).
 6.6 Band Structures of Three-Dimensional Materials 231

Γs*
Γp*
4

Γp*
energy (eV)

Γs*
0 Eg = 1.1 eV Eg = 0.7 eV
Γp
Γp
-4

-8

Γs Γs
-12
W L Γ X W W L Γ X W

silicon germanium

Figure 6.28 The band structures of silicon (left) and germanium (right).

sixth period, we come to Pb where this trend is further exacerbated by relativistic effects
(Section 5.2.1) which contract the 6s orbitals and lower their energy with respect to the
6p orbitals. If Pb were to adopt the diamond structure, the antibonding 6s states would be
populated before the bonding 6p states are full, destabilizing the structure. Consequently, Pb
adopts a cubic closest-packed structure in preference to the diamond structure.

6.6.4 Rhenium Trioxide

The 3D analogue of the square CuO22− lattice is the cubic ReO3 structure (Figure 1.39),
a network of corner-sharing octahedra, with a Niggli formula of ReO6/2. ReO3 is not just
a model case, it is one of the most highly conducting oxides known. Its conductivity of
1.1×107 S/m at room temperature is higher than the conductivity of several elemental metals
(Section 10.1).
The electronic band structure of ReO3 is shown in Figure 6.29. Based on the MO diagram of
an octahedrally coordinated transition-metal ion (Section 5.3.8), we expect the Re 5d and
O 2p orbitals to make the dominant contributions to the bands near the Fermi level. The
octahedral coordination geometry means that the Re 5d orbitals are split into a triply degener-
ate t2g set (dxy, dyz, dxz) that interacts with the O 2p orbitals in a π fashion, and a doubly
degenerate eg set (dx2−y2, dz2) that interacts with the O 2p orbitals in a σ fashion. All Re 5d–O
2p bonding and O 2p nonbonding crystal orbitals are fully occupied. One electron occupies the
Re 5d–O 2p π* set of crystal orbitals, as expected given the 5d1 configuration of Re6+.
232 Electronic Band Structure

Re
12 O

8
Re dx2−y2, dz2
(σ* bands)
energy (eV)
4

Re dxy, dxz, dyz

0 EF (π* bands) EF

-4

-8 O 2p bands

-12
X R M Γ R
DOS, N(E)

Figure 6.29 The crystal structure, band structure, and DOS of ReO3.

The primitive unit cell of the ReO3 structure contains one rhenium atom and three oxygen
atoms. Therefore, we expect (3 × 3) nine bands for the O 2p orbitals and (5 × 1) five bands
for the Re 5d orbitals. We see the nine O 2p-based bands between −12 eV and −4 eV in
Figure 6.29. Six of them run downhill from Γ, where they are mostly nonbonding O 2p crystal
orbitals, to R where they have a Re 5d–O 2p and Re 6s–O 2p bonding character. The
remaining three are confined to a narrow energy window near the top of the valence band
and are largely O 2p nonbonding throughout the first Brillouin zone.
The three bands located between −2.3 eV and +2.6 eV arise from antibonding π* inter-
actions between the triply degenerate t2g set of Re 5d orbitals and O 2p orbitals. The Fermi
level cuts through these three bands. The fact that the Fermi level cuts through a reasonably
wide band (in this case three bands) is responsible for the high electrical conductivity of ReO3.
The two highest-energy bands in Figure 6.29 are the Re 5d–O 2p σ* bands, located between
+2.7 eV and +9.9 eV. The Re 5d eg orbitals make the dominant contribution to these bands.
Because the σ interactions lead to a higher degree of spatial overlap than the π interactions,
the σ* bands are wider than the π* bands, 7.2 eV versus 5.0 eV.
The orbital overlap is shown in Figure 6.30 for three representative bands—a nonbonding
O 2p band, the Re 5dxy–O 2p π* band, and the Re 5dx2−y2–O σ* band—at Γ, X, and M. The
picture is very similar to that discussed earlier for the CuO22− square lattice (Section 6.5.2).
At the Γ point, orbital mixing is symmetry forbidden for Re t2g and O 2p orbitals, which
makes this set of bands strictly Re 5d nonbonding. The σ* set of bands is also (largely) Re
5d nonbonding at Γ. Symmetry allows for Re 5d–O 2s σ* interactions, but they are weak due
to the poor energetic overlap of the O 2s and Re 5d orbitals. At the X point, all three π* bands
have antibonding interactions with their oxygen neighbors along x, while the interactions in
the other two directions remain nonbonding. This leads to a splitting of the three π* bands,
 6.6 Band Structures of Three-Dimensional Materials 233

y
Γ X M
x

Re 5dx2−y2 −Ο
σ* band

Re 5dxy −Ο
π* band

O 2p (t1g)
nonbonding band

Figure 6.30 The orbital interactions for a nonbonding O 2p band (bottom), a Re–O π* band (middle), and
a Re–O σ* band (top) at the Γ, X, and M points (defined in Figure 6.16).

because the overlap of the Re 5dyz orbitals with the x-direction neighbors is negligible,
whereas the Re 5dxy and 5dxz orbitals have significant Re–O interactions along x.
Consequently, the latter two bands, being antibonding, have a higher energy than the Re
5dyz band at X. For similar reasons, the Re 5dx2−y2 band has a higher energy than the Re 5dz2
band at X. At the M point, the Re 5dxy and Re 5dx2−y2 bands are completely antibonding, as
illustrated on the right-hand side of Figure 6.30. Finally, at the R point all five Re 5d bands
are completely antibonding (not shown), with each band reaching its highest energy.
The O 2p (t1g) band depicted in Figure 6.30 is one of the O 2p bands described as
nonbonding in the preceding discussion. In a sense, this description is accurate because the
band lacks any Re 5d character. However, it is not strictly nonbonding if we take O 2p–O
2p interactions into account. If we consider only the nearest-neighbor oxygen interactions,
we see a change from nonbonding at Γ to antibonding at M. As a result, the energy of this
band increases by a relatively small amount, 1.2 eV, on moving from Γ to M.

6.6.5 Perovskites
Just as the band structure of Si is representative of many semiconductors, the electronic
structure of ReO3 is representative of an important group of materials, the perovskites
(Figure 1.49). Throughout the book, we will encounter many functional materials with the
perovskite structure. In perovskites such as SrTiO3, the bonding between the larger A-site
234 Electronic Band Structure

cation (Sr2+) and oxygen is normally quite ionic, while the more electronegative M-site cation
(Ti4+) forms much more covalent bonds with the anion (O2−). It is therefore a reasonable
approximation to assume the A-site cation completely donates its valence electrons to the
MO3n− network. To the extent this approximation is valid, the band structure of a cubic
perovskite is identical to that of ReO3. This approximation is generally quite good for bands
near the Fermi level. The filling of the conduction bands depends on the d-electron count of
the M-site atom, while the position and width of these bands depends on the covalency of the
M–O interaction and, as we will learn, the linearity of the M–O–M bonds.
A generic band-structure diagram for a perovskite is shown in Figure 6.31. The symmetry of
the cubic MO3 network gives rise to several special features. At the Γ point, the crystal orbitals
at the top of the O 2p band are nonbonding while those at the bottom of the conduction band
are nonbonding t2g metal-based orbitals, as already discussed for ReO3. Therefore, the direct-
gap excitation, labeled CT in Figure 6.31, represents a charge-transfer excitation in the truest
sense; from crystal orbitals with pure oxygen 2p character to crystal orbitals with pure
transition-metal d-orbital character. Another point of interest is the separation between the
π* and σ* bands at the R point. All of the M nd–O 2p interactions at this point are completely
antibonding; consequently, the energy separation between π* and σ* bands at R is the solid
state equivalent of the ligand-field splitting of an octahedron, Δ.
To examine periodic trends in bonding, let’s take a quantitative look at how the band
structure evolves as we decrease the electronegativity of the transition metal: ReO3 → WO3
→ KTaO3 → BaHfO3. All four compounds have similar band structures, but the exact
energies and filling of the bands vary from one compound to the next. The calculated energies
of Eg, Δ, CT, and several other parameters are given in Table 6.1.

M dx2−y2, dz2
Δ (σ* bands)
energy (eV)

M dxy, dxz, dyz

(π* bands)

Eg Eg
CT

O 2p bands

-12
X R M Γ R
DOS, N(E)

Figure 6.31 A generic band-structure diagram (left) and a schematic DOS plot (right) for an AMO3
perovskite, where M is a d 0 transition-metal ion. Values of the marked parameters are given in Table 6.1
for several different perovskites.
 6.6 Band Structures of Three-Dimensional Materials 235

Table 6.1 Key features from band-structure calculations of cubic ReO3, WO3, KTaO3, BaHfO3,
and SrTiO3.

ReO3 WO3 KTaO3 BaHfO3 SrTiO3

M–O distance (Å) 1.87 1.95 1.99 2.09 1.95

O–O distance (Å) 2.65 2.76 2.81 2.96 2.76
d-electron configuration d1 d0 d0 d0 d0
M–O σ* bandwidth (eV) 7.2 6.4 6.2 5.1 4.1
M–O π* bandwidth (eV) 5.0 4.3 4.2 3.5 2.4
Δ (σ*–π* at R) (eV) 7.2 6.3 6.1 5.5 3.9
CT (O 2p–M π* at Γ) (eV) 2.3 3.4 3.8 5.3 4.1
Calc. band gap, Eg (eV)† Metal 2.4 3.5 5.0 3.4
Exp. band gap, Eg (eV) Metal 2.4 3.5 5.5 3.1
†
A scissors correction that increases the band gap by 2.1 eV has been applied to all semiconducting compounds
in this table.

On changing from ReO3 to WO3, a dramatic change arises from the fact that the π* bands
in WO3 are empty because W6+ has a d 0 configuration. This change in band filling makes
WO3 a semiconductor, in sharp contrast to the metallic conductivity of ReO3. As a rule,
semiconducting cubic perovskites have indirect band gaps because the valence-band max-
imum always falls on the line between M and R, while the conduction-band minimum always
falls at Γ. The measured band gap of WO3 is 2.4 eV, which is responsible for its yellow
color.28
Replacing Re with W has other subtle effects on the electronic structure. Because hexava-
lent tungsten is less electronegative than hexavalent rhenium,29 it takes more energy to excite
an electron from oxygen to the transition metal, hence CT increases from 2.3 eV to 3.4 eV.
The widths of the σ* and π* bands decrease because the antibonding interactions at the
R point, which define the maxima of these bands, are not quite as antibonding, due to longer
W–O distances and the fact that the W 5d and O 2p orbitals have a greater energetic
mismatch. The octahedral ligand-field splitting Δ decreases for the same reason. As we
continue to decrease the atomic number of the transition metal along WO3 → KTaO3 →
BaHfO3, the d 0 electron count is maintained but the nuclear charge steadily decreases. This
leads to a decrease in the electronegativity of the transition-metal ion, a decrease in the
covalency of the metal–oxygen bonds, and an increase in the bond length. These changes
increase CT and Eg and reduce the σ* and π* bandwidths.

28
It should be noted that the structure of WO3 is not actually cubic. It is distorted by out-of-center displacements of
tungsten atoms and rotations of the octahedra, both of which reduce the bandwidth and increase the band gap. The
tungsten displacements are driven by second-order Jahn–Teller distortions.
29
For a given oxidation state, electronegativities increase and orbital energies become more negative on moving
horizontally from left to right across the periodic table (see Table 5.2).
236 Electronic Band Structure

Perovskites containing 3d transition metals have narrower σ* and π* bands than their
4d and 5d analogues. Consider SrTiO3 and BaHfO3, where both transition metals come from
group 4. Table 6.1 shows that the σ* and π* bands are substantially narrower for SrTiO3 than
they are for BaHfO3. This effect stems from the fact that 3d orbitals are smaller than
5d orbitals (rmax = 0.53 Å for Ti 3d versus 0.88 Å for Hf 5d, Section 5.2.2). The contracted
nature of the 3d orbitals reduces their overlap with the O 2p orbitals, and, as a result, the
antibonding SrTiO3 crystal orbitals, particularly the σ* and π* crystal orbitals at R, are less
destabilized (have a lower energy) than the corresponding BaHfO3 crystal orbitals. The
decreased overlap results in the reduced bandwidth for SrTiO3.
Bandwidth in perovskites is an important parameter that can impact the properties,
sometimes in a dramatic fashion. Wider bands not only produce smaller band gaps, they
are more conducive to electron delocalization. We will see the effect of bandwidth on the
electrical conductivity of perovskites in Section 10.4.3. When we encounter such effects, it
will be very useful to remember the mechanisms for tuning the bandwidth in perovskites.
Table 6.2 contains a summary of those mechanisms.
Changing the identity of the element on the M site is not the only way to control the
electronic structure of a perovskite. While the A cation does not usually play a direct role in the
electronic structure near the Fermi level, its size controls the linearity of the M–O–M bonds.
When the A-site cation is too small for the cubic corner-sharing network, octahedral tilting
occurs (Section 1.5.3) leading to bending of the M–O–M bond angles. This introduces some
antibonding character into the π* bands at Γ (remember from Figure 6.30 they were strictly
nonbonding at Γ in a cubic perovskite) and reduces the antibonding overlap of the π* bands at
R. The net effect is a reduction in the widths of the π* bands, which leads to an increase in band
gap. Octahedral tilting has a similar effect on the width of the σ* band, which can have
important implications for conductivity and magnetism in some perovskites. Overlap consid-
erations predict that the width of the σ* bands, Wσ*, should scale proportionally to cos φ,
where φ is the deviation of the M–O–M bond angle from 180° [2]. As an example, consider
what happens when K+ in KTaO3 is replaced by Na+. The smaller Na+ drives an octahedral-
tilting distortion that reduces the Ta–O–Ta bond angle from 180° in KTaO3 to 159° (on
average) in NaTaO3. This increases the band gap from 3.5 eV in KTaO3 to 4.1 eV in NaTaO3.

Table 6.2 Factors that impact the width of the π* (Wπ*) and σ* (Wσ*) bands in
AMO3 perovskites.

Change in structure or composition Wσ* and Wπ*

M electronegativity increases Increase

M–O bond distance increases Decrease
M oxidation state increases Increase
3d transition metal substitutes for 4d/5d transition metal Decrease
Octahedral tilting bends M–O–M bond angle Decrease
 6.7 Problems 237

6.7 Problems
6.1 MO diagrams for cyclic HN molecules are shown in Figure 6.1. The AO phases are
shown for the lowest- and highest-energy MOs but not for the intermediate MOs. (a)
Sketch out the intermediate MOs indicating the phase of each AO for a cyclic H6
molecule. Determine the number of nodal planes for each of the six MOs. (b) Repeat
part (a) for a cyclic H10 molecule.
6.2 MO diagrams for cyclic HN molecules are shown in Figure 6.1. (a) Sketch out the MOs
and their relative energies for a square H4 molecule. Determine the number of nodal
planes for each MO. (b) This molecule is prone to a first-order Jahn–Teller distortion.
How will the distortion change the shape of the molecule? Redraw the MO diagram
after the distortion has taken place.
6.3 Show that the Bloch function given in Equation (6.1) meets the requirement that the
electron density, ρ(r) = ψ2(r) = ψ*(r)ψ(r), must be periodic according to Equation
(6.2).
6.4 The first Brillouin zone of an infinite chain of H atoms has an infinite number of crystal
orbitals, each with a different value of k. (a) State the limiting values of k within the first
Brillouin zone. (b) The figure below shows the real part of the wavefunction for
a crystal orbital ψ(r) with a specific value of k (the positions of the H atoms are denoted
by the black dots). Determine the value of k for this crystal orbital. (c) Calculate the
momentum of this crystal orbital.

6.5 State whether each of the following statements is true or false. (a) As the magnitude of
k increases, the momentum of the electron also increases. (b) As the magnitude of
k increases, the wavelength of the corresponding wavefunction also increases. (c) As the
size of the unit cell increases, the size of the first Brillouin zone also increases.
238 Electronic Band Structure

6.6 Consider an infinite 1D chain of equally spaced fluorine atoms. (a) Sketch the band
structure of this chain. Include all four valence orbitals (2s, 2px, 2py, 2pz) in the diagram
and indicate the position of the Fermi level (assume the chain propagates in the z direction).
(b) Sketch a DOS plot. (c) Would you expect this chain to be a metallic conductor?
6.7 Consider an infinite 1D chain of H2 molecules where the molecular axis is oriented
perpendicular to the chain direction, as shown below. (a) How many bands are there in
the band structure? (b) Sketch the crystal orbitals for each band at k = 0 and k = π/a. (c)
Sketch out the band structure of this chain and include the position of the Fermi level. (d)
Sketch a DOS plot. (e) Would you expect this chain to be a metallic conductor? Hint:
Develop your answer using the bonding and antibonding MOs of the H2 molecule.

H H H H H H

H H H H H H

6.8 Consider the infinite 1D chain formed by placing boron atoms between the H2 molecules
from Problem 6.7 to form an infinite chain (shown below) where the B–H distance is
1.27 Å while B–B and H–H are both 1.8 Å. The calculated band structure for this chain is
shown below. (a) Which of the boron orbitals can mix (form bonding/antibonding crystal
orbitals) with the H2 bonding orbital, ψ+, at k = 0? Which boron orbital can mix with ψ+ at
k = π/a? (b) Which of the boron orbitals can mix with the H2 antibonding orbital, ψ−, at
k = 0? Which boron orbital can mix with ψ− at k = π/a? (c) What is the AO character of the
two crystal orbitals that become degenerate at k = π/a? (d) At k = 0, only the lowest-energy
band has both B–H and H–H σ-bonding character. What is the AO character of this band
at k = 0? (e) At k = 0, the highest-energy band (E = +33.8 eV) has both B–H and H–H σ-
antibonding character. What orbitals contribute to this band at k = 0?

H H H H
x
B B B
H H H H
z
36

24
energy (eV)

12

0
EF
-12

-24
0 π/a
k
 6.7 Problems 239

6.9 Consider the band structure for an infinite 1D chain of equally spaced titanium and
oxygen atoms shown below, with an overall stoichiometry of TiO. In this calculation,
only the Ti 3d orbitals and the O 2p orbitals have been taken into account. (a) Identify
the orbital making the largest contribution to the band that spans the energy range
from approximately −12 eV to −6 eV. (b) Identify the orbitals that make the largest
contribution to the two flat bands located at roughly −11 eV. (c) Indicate the approxi-
mate location of the Fermi level. (d) Would you expect this chain to be a metal or
a semiconductor?

x
O Ti O Ti O Ti

-5
energy (eV)

-9

-13

-17
0 π/a
k

6.10 Consider the electronic structure of α-Po discussed in Section 6.6.1. (a) Sketch the
crystal orbitals for the py band at Γ, X, and R. (b) Sketch the crystal orbitals for the px,
py, and pz bands at M. (c) Order the px, py, and pz bands from lowest to highest energy at
Γ, X, M, and R.
6.11 NbN crystallizes with the NaCl type structure. Sketch out an MO diagram for octa-
hedrally coordinated niobium and use it to sketch a DOS plot for NbN. Indicate the
relative area of the σ*, π*, and N 2p sets of bands, and mark the approximate position
of the Fermi level in your DOS sketch. Would you expect NbN to be a metal or
a semiconductor?
6.12 GaAs (Eg = 1.4 eV) and ZnSe (Eg = 2.6 eV) are isoelectronic with Ge (Section 6.6.3).
Their band structures are shown below. Are they direct- or indirect-gap
semiconductors?
240 Electronic Band Structure

GaAs ZnSe

energy (eV)

energy (eV)
EF EF

W L Γ X W W L Γ X W

6.13 Consider LaCrSb3, whose structure is shown below (left). To a reasonable approximation,
this structure can be described as independent CrSb22− layers and Sb− layers separated by
La3+ ions [3]. If we neglect subtle distortions, the Sb− layer can be approximated as a 2D
square lattice of Sb− ions. There is one atom per unit cell, and the Sb–Sb separation is
3.1 Å. The calculated band structure for the idealized Sb− layer is shown below, on the
right. (a) Show the orbital overlap for the Sb 5s and 5p orbitals at Γ, X, and
M. Characterize the nearest-neighbor interactions for each band as (σ or π) bonding,
antibonding, or nonbonding at each of these k points. (b) Determine the orbital character
of each band, numbered 1–4 in the diagram. (c) LaCrSb3 is a metallic conductor. From the
band structure above, would you expect the Sb− layers to contribute to the conductivity?

12

CrSb22− 6 4
layer
0
energy (eV)

3
Sb− layer -6
EF
La3+ -12
2
z -18
1
y
x -24
Γ X M

6.14 A 2D square lattice of 3d transition-metal atoms 2.3 Å apart has the band structure shown
below (the contributions of the 4s and 4p orbitals have been omitted to simplify the
analysis). (a) Considering orbital overlap for each of the five 3d orbitals at Γ, X, and M,
characterize the nearest-neighbor interactions for each band as (σ, π, or δ) bonding,
antibonding, or nonbonding at each of these k points. (b) Use your answers from part (a)
 6.8 Further Reading 241

to associate bands numbered 1–3 in the diagram above with a dxy, dxz, or dyz orbital (due to
similar behavior at Γ, X, and M, bands 4–5 have contributions from both dx2−y2 and dz2). (c)
Which d-electron count will provide the strongest metal–metal bonding: d 1, d 4, or d 8?

y
-6
5
x
-8 1

energy (eV)
3

-10 4
2 5
3
-12 4
2

1
-14
Γ X M

6.15 The perovskite LaRhO3 is a semiconductor. (a) What is the orbital character of the
valence band? (b) What is the orbital character of the conduction band? (c) Will
a hypothetical cubic LaRhO3 be a direct- or indirect-gap semiconductor?
6.16 BaZrO3 and CaZrO3 are perovskites with similar Zr–O distances. While BaZrO3 is
cubic, CaZrO3 is orthorhombically distorted by octahedral tilting. Which of these two
compounds will have a larger band gap? Explain your reasoning. Hint: The Zr–O–Zr
angle is 180° in BaZrO3 and 146° (on average) in CaZrO3.

6.8 Further Reading

J.K. Burdett, “Chemical Bonding in Solids” (1995) Oxford University Press.
P.A. Cox, “The Electronic Structure and Chemistry of Solids” (1987) Oxford University Press.
R. Dronskowski, “Computational Chemistry of Solid State Materials” (2005) Wiley−VCH, Weinheim.
R. Hoffmann, “Solids and Surfaces: A Chemist’s View of Bonding in Extended Structures” (1989)
Wiley, New York.
E. Kaxiras, “Atomic and Electronic Structure of Solids” (2003) Cambridge University Press.
T. Wolfram, S. Ellialtioglu, “Electronic and Optical Properties of d-Band Perovskites” (2006)
Cambridge University Press.
242 Electronic Band Structure

6.9 References
[1] K.S. Novoselov, A.K. Geim, S.V. Morozov, D. Jiang, Y. Zhang, S.V. Dubonos, V. Grigorieva,
A.A. Firsov, “Electric field effect in atomically thin carbon films” Science 306 (2004), 666–669.
[2] J.B. Goodenough, J.-S. Zhou, “Localized to itinerant electronic transitions in transition-metal
oxides with the perovskite structure” Chem. Mater. 10 (1998), 2980–2993.
[3] N.P. Raju, J.E. Greedan, M.J. Ferguson, A. Mar, “LaCrSb3: A new itinerant electron ferromagnet
with a layered structure” Chem. Mater. 10 (1998), 3630–3635.
7 Optical Materials

We transition into the second part of the book with an in-depth look at materials used for
their optical properties. In the first half of the chapter, we consider materials that are valued
for the way they absorb light: pigments, dyes, and gemstones. In the second half, we turn our
attention to materials that emit light. Materials such as phosphors and light-emitting diodes
play a key role in devices that we encounter in our daily lives, including fluorescent and solid
state lighting.

7.1 Light, Color, and Electronic Excitations

All forms of electromagnetic radiation travel as self-propagating waves, moving through
vacuum at a speed of c = 2.99792458×108 m/s. The wavelength, λ, and frequency, ν, of the
wave are related to its speed through the relationship:

c ¼ λν (7.1)

Electromagnetic radiation can also be treated as a particle called a photon. The energy of
each photon is determined by its frequency (or wavelength) through Planck’s equation:

E ¼ hν ¼ hc=λ (7.2)

where h is the Planck constant of 6.62607015×10−34 J s for E in joules of the SI system. The
photon energy can also be expressed in electron volts (eV), which is the amount of potential
energy gained by moving an electron across an electric-potential difference of 1 V (1 eV =
1.602176634×10−19 J). Instead of wavelength, a spectroscopy unit called wavenumber can be
used (in units cm−1, the number of wavelengths that fit in one centimeter). A photon
of energy 1 eV represents radiation with a wavenumber of 8065.544 cm−1. As shown in
Figure 7.1, visible light makes up only a small slice of the electromagnetic spectrum.

243
244 Optical Materials

frequency, ν (s−1)
1024 1020 1016 1012 108 104

γ-rays X-rays UV IR micro- radiowaves

wave

10−16 10−12 10−8 10−4 100 104

wavelength, λ (m)

magenta

blue cyan green yellow red

400 450 500 550 600 650 700

wavelength, λ (nm)

Figure 7.1 The electromagnetic spectrum (upper), and the visible portion of the spectrum (lower),
showing the regions corresponding to the primary (RGB) and secondary (CMY) colors.

Colors are perceptions of the human eye. We have three different color receptors with
maximum sensitivity at three different wavelengths in the visible spectrum. To understand
colors, it is therefore useful to choose three colors of the rainbow as the primary colors of
light: red, green, and blue (RGB); and then define three (secondary) colors that arise by
mixing neighboring primary colors, as shown in the lower half of Figure 7.1. Mixing blue and
green light yields cyan (C), while red plus green gives yellow (Y). The combination of blue
plus red we see as a magenta (M). We could of course name many more colors than these six.
For example, violet contains more blue light than magenta (i.e. an unequal mixing of blue
and red), and orange is more reddish than yellow. Unlike the other primary and secondary
colors, magenta cannot be represented by a narrow slice of the visible spectrum; it results
from mixing red and blue. Hence, we perceive it as being more reddish than the violet end of
the spectrum.
Now let’s think about the colors that arise when substances absorb light. When
a substance absorbs blue light, both red and green are either transmitted or reflected,
and the substance appears yellow to us (remaining R + G = Y, termed the comple-
mentary color to B). If a substance absorbs red light, we perceive it as having a cyan
color (remaining B + G = C, complementary to R), and one that absorbs green
takes on a magenta color (remaining B + R = M, complementary to G). For example,
a Cu2+(aq) solution absorbs red light which leads to its familiar cyan color.
A substance that absorbs all three RGB primary colors is black, whereas a light source
that mixes the three primary colors creates white light. This is why screens of electronic
devices are based on pixels, each containing sources that can emit RGB colors, while
 7.2 Pigments, Dyes, and Gemstones 245

printers use the CMYK mixing of dyes. The K stands for blacK, formed by mixing
C + M + Y of highest saturation. Consider the color of a C + M dye. The C dye
absorbs R, while the M dye absorbs G, and the remaining color is B.
At the atomic level, the colors of chemical compounds arise from absorption of a visible
light of certain energy hν that leads to an electronic excitation.1 Such absorptions can be
broadly grouped into five categories: (a) d-to-d transitions; (b) charge-transfer transitions; (c)
band-to-band transitions in semiconductors; (d) transitions between molecular orbitals
(MOs), most commonly π-to-π* transitions in conjugated organic molecules; (e) f-to-f; and
(f) f-to-d transitions.2 In the following sections, we explore examples of the first four causes of
color. Transitions involving f orbitals are explored in Section 7.8 because of their importance
in luminescent materials.

7.2 Pigments, Dyes, and Gemstones

A pigment is a colored material that is dispersed in a medium in which it is insoluble.3 The
medium might be any number of substances ranging from oil to water to plastic. If a colored
material is soluble in the medium in which it is dispersed it is called a dye.
Pigments are among the oldest functional materials. The prehistoric artist’s palette
was confined largely to red, yellow, brown, and black. The reds, yellows, and browns
came from iron containing minerals such as Fe 2O3 (hematite or red ochre) and FeOOH
(goethite or yellow ochre), while manganese oxides and carbon were used for black.
Reliance on colored materials obtained from natural sources continued until the
eighteenth century, when synthetic pigments started to become widely available.
A representative list of pigments, dyes, and gemstones, arranged by color, is given in
Table 7.1. Minerals that are colored when chemically pure, like eskolaite Cr2O3 or
malachite Cu2CO3(OH)2, are said to be idiochromatic. Minerals that are colored due to
the presence of dopants, like ruby (Al2O3:Cr3+) and spinel (MgAl2O4:Cr3+) are said to
be allochromatic. This notation, where the dopant responsible for the color follows
a colon after the formula of the undoped host, will be used throughout this chapter. In
some instances, the site on which substitutional disorder occurs is obvious. For
example, in ruby and spinel the Cr3+ substitutes for Al3+ as Al2−xCrxO3 and MgAl2−
xCrxO4, respectively. In other cases, such as at low doping levels of aliovalent substitu-
tions, the distribution may be more complex or uncertain.

1
In most substances, the electron’s absorbed optical energy is subsequently released in several vibrational steps as
thermal energy. Photoluminescent materials are an exception.
2
One category of electronic transition that doesn’t fit neatly into any of these categories are the F centers or color
centers discussed in Chapter 2. The violet color of the mineral fluorite, CaF2, arises from F centers (from Farbe, the
German word for color).
3
Pigments can also be black or white.
246 Optical Materials

Table 7.1 Representative pigments, dyes, and gemstones.

Name Composition Use Electronic excitation

Blues
Azurite Cu3(CO3)2(OH)2 Pigment d-to-d (Cu2+)
Cobalt blue CoAl2O4 Pigment d-to-d (Co2+)
Egyptian blue CaCuSi4O10 Pigment d-to-d (Cu2+)
Indigo C15H9N2O Dye MO (π-to-π*)
Phthalocyanine blue Cu(C32N8H16) Pigment MO (π-to-π*)
Prussian blue Fe4[Fe(CN)6]3·xH2O Pigment MMCT Fe2+→Fe3+
Sapphire Al2O3:Fe2+,Ti4+ Gemstone MMCT Fe2+→Ti4+
Ultramarine Na8−x[(Si,Al)12O24](S3,Cl)1–2 Pigment MO (S3−)
Greens
Chrome green Cr2O3 Pigment d-to-d (Cr3+)
Emerald Be3Al2(SiO3)6:Cr3+ Gemstone d-to-d (Cr3+)
Phthalocyanine green Cu(C32N8Cl16−xHx) Pigment MO (π-to-π*)
Malachite Cu2CO3(OH)2 Pigment d-to-d (Cu2+)
Yellows
Bismuth vanadate BiVO4 Pigment LMCT O2−→V5+
Cadmium yellow CdS Pigment Band-to-band
Chrome yellow PbCrO4 Pigment LMCT O2−→Cr6+
Orpiment As2S3 Pigment Band-to-band
Reds
Alizarin C14H8O4 Dye MO (π-to-π*)
Cadmium red CdS1−xSex Pigment Band-to-band
Pyrope Mg3Al2Si3O12:Cr3+ Gemstone d-to-d (Cr3+)
Red lead Pb3O4 Pigment Band-to-band
Ruby Al2O3:Cr3+ Gemstone d-to-d (Cr3+)
Spinel MgAl2O4:Cr3+ Gemstone d-to-d (Cr3+)
Vermillion HgS Pigment Band-to-band

7.3 Transitions between d Orbitals (d-to-d Excitations)

Transition-metal compounds make up the largest family of colored substances. In most
cases, their color arises from electronic transitions between different d orbitals on the same
transition-metal ion. In this section, we explore the factors that determine the energy,
number, and intensity of the electronic absorption peaks arising from d-to-d transitions.

7.3.1 Ligand- and Crystal-Field Theory

Understanding how the energetic degeneracy of the d orbitals is removed by the surrounding
ligands is the first step in understanding the optical properties of transition-metal
 7.3 Transitions between d Orbitals (d-to-d Excitations) 247

compounds. Ligand-field theory relies on MO calculations (Chapter 5) to establish the

d-orbital energies. For example, in the MO diagram of an octahedral complex, the dz2 and
dx2−y2 orbitals form σ* interactions with the ligands, and hence these orbitals are more
antibonding than the dxy, dxz, and dyz orbitals that form π* interactions with the ligands
(Section 5.3.8). The σ* and π* MOs are usually referred to as eg and t2g orbitals,
respectively, after their symmetry labels. The energy separation between these two sets
of orbitals is the ligand-field splitting, Δoct, where the subscript “oct” denotes octahedral
coordination.
Crystal-field theory is a heuristic approach that qualitatively predicts splitting of the
d-orbital energies due to electrostatic repulsion from the ligands. For example, in an
octahedral field, the dz2 and dx2−y2 orbitals point directly at the ligands and the electrons in
them experience more repulsion than those in the dxy, dxz, and dyz orbitals that point between
the ligands and thus lie at a lower energy. Both crystal-field and ligand-field theory predict
the same patterns of d-orbital splitting, which are given for various ligand environments in
Figure 7.2. To make quantitative predictions of orbital energies, ligand-field theory is
needed, but for deducing the gross features of d-orbital splitting in a variety of coordination
environments, the simpler crystal-field approach is useful.
Values of Δoct for various combinations of transition metal ions and ligands are given in
Table 7.2. Examination of these values reveals several trends. Firstly, we see that, for
a given d-electron count, Δoct increases as the oxidation state of the metal cation increases,
as illustrated by the complex ions containing transition metals with a d 3 configuration: ([Cr(H2
O)6]3+ > [V(H2O)6]2+ and [MnF6]2− > [CrF6]3−). This trend results from the trend toward

dx2−y2

dx2−y2

dz2
dz2 dx2−y2
energy

dxy
dxy dxz dyz dz2

dxy dxy dx2−y2

dz2 dx2−y2 dz2 dxz dyz

dxy dxz dyz
dxz dyz
dxz dyz

octahedral tetrahedral square square trigonal

planar pyramidal bipyramidal

Figure 7.2 Examples of ligand-field splitting diagrams for various coordination environments. The
dashed line is the energy of the d orbitals in a spherically symmetric environment.
248 Optical Materials

Table 7.2 Ligand-field splitting, Δoct, for complex ions of various transition
metals.

Metal ion d count [MF6]n− [M(H2O)6]m+

Ti3+ d1 2.17 eV 2.52 eV

V3+ d2 2.00 eV 2.21 eV
V2+ d3 Not formed 1.54 eV
Cr3+ d3 1.87 eV 2.16 eV
Mn4+ d3 2.70 eV Not Formed
Tc4+ d3 3.52 eV Not Formed
Re4+ d3 4.07 eV Not Formed

shorter metal–ligand bonds as the oxidation state of the metal ion increases. The enhanced
overlap raises the energy of the σ* orbitals more than π* orbitals, thereby increasing Δoct.
Secondly, we see that Δoct increases as the valence orbitals on the central metal ion go from
3d to 4d to 5d (Mn4+ → Tc4+ → Re4+). The larger 4d and 5d orbitals (Section 5.2.2) experience
a stronger overlap with the ligand orbitals, which again leads to greater destabilization of the
σ* orbitals than the π* orbitals. Finally, for a given ion, Δoct also depends upon the ligand. For
example, the ligand-field splitting increases slightly when fluoride ions are replaced with water
molecules.

7.3.2 Absorption Spectra and Spectroscopic Terms

To explore the connection between electronic structure and absorption spectra, consider the
octahedral complex [Ti(H2O)6]3+ containing the d1 ion Ti3+. The octahedral ligand field splits
the d orbitals into π*(t2g) and σ*(eg) sets separated by 2.52 eV (Table 7.2). With Equation
(7.2), we calculate a wavelength of 492 nm for a photon of energy equal to the separation
between the two sets of orbitals. The UV–visible absorption spectrum of a [Ti(H2O)6]3+
solution shows a broad peak centered at ~490 nm (Figure 7.3). The peak is split by a Jahn–
Teller distortion (Section 5.3.10) of the complex, driven by the 3d1 configuration of Ti3+,
which lifts the degeneracy of both the eg and t2g orbitals.
The situation becomes more complicated when there are multiple electrons in the
d orbitals of the transition-metal ion. Consider [V(H2O)6]3+, containing the d 2 ion V3+.
Table 7.2 tells us that Δoct = 2.21 eV, which suggests that 561 nm light should excite
a t2g2eg0 → t2g1eg1 transition. As we see in Figure 7.3, there is an absorption peak that
reaches a maximum at ~570 nm, near the expected position, but, somewhat unexpect-
edly, there is another peak at ~390 nm, both arising from d-to-d transitions. The reason
for the increased complexity of the electronic absorption spectrum stems from electron–
electron interactions.
 7.3 Transitions between d Orbitals (d-to-d Excitations) 249

10

molar extinction coefficient, ε [l/(mol cm)]

[Ti(H2O)6]3+
8

6 [V(H2O)6]3+

0
300 400 500 600 700 800 900 1000
photon energy (eV)

Figure 7.3 The UV–visible absorption spectrum of aqueous solutions of [Ti(H2O)6]3+ (black) and
[V(H2O)6]3+ (gray).

To understand the [V(H2O)6]3+ spectrum, we must take a step back and consider the
electron–electron interactions in a free V3+ ion. Its electron configuration, [Ar] 3d 2, only
gives information on the first two quantum numbers, n and ℓ (Section 5.2). It neither tells us
whether the electrons occupy the same d-orbital or different d-orbitals, nor whether they
have the same or opposite spins. Yet these factors have a bearing on the collective energy of
the electrons. Coupling between the spin and orbital angular momenta of the electrons also
plays a role. To understand absorption spectra of transition-metal ions, we must take these
factors into account.
There are two approaches to incorporate electron–electron interactions and spin–
orbit coupling into the analysis. Within the Russell–Saunders coupling scheme, the spin
and orbital momenta are summed separately and then combined. This approach
assumes electron–electron repulsions are dominant and treats spin–orbit coupling as
a perturbation. Such an approximation is reasonable when the spin–orbit coupling is
relatively weak, which is generally applicable for 3d transition metals like V3+. The
j–j coupling scheme represents the opposite approach where spin–orbit coupling is first
applied to split the different orbital occupations into states, and then electron–electron
repulsions act as a perturbation on those energy levels. For elements with intermediate
levels of spin–orbit coupling, like the 4d and 5d transition-metal ions and the lanthan-
oids, the coupling is intermediate between these two limiting cases. We will adopt
Russell–Saunders coupling throughout this chapter, but the interested reader can find
further details on j–j coupling in Appendix F.
250 Optical Materials

mℓ 2 1 0 −1 −2 mℓ 2 1 0 −1 −2

L = 4, S = 0 L = 3, S = 1

Figure 7.4 Two microstates for a free ion with a d 2 configuration.

The different ways electrons can occupy the available atomic orbitals (AOs) are called
microstates of the parent electron configuration. For example, there are ten available
microstates for a free d1 ion such as Ti3+. The electron can reside in any one of the five
d orbitals with mℓ values ranging from 2 to −2, and its spin quantum number ms can be either
+½ or −½. Because there is only one electron, and hence no electron–electron repulsions to
consider, all ten microstates have the same energy. Moving to d 2, there are now 45 possible
microstates, not all of which lie at the same energy. Two such microstates are illustrated in
Figure 7.4. It can be shown that these 45 microstates separate into five sets of different energy
as discussed below. The label given to each set is called a spectroscopic term or term symbol.
To derive the term symbols for a d 2 ion, we begin by identifying the values of the total
angular-momentum quantum number L and the total spin quantum number S that can arise
from the orbital and spin angular momenta of the individual electrons. There are two
electrons with spin angular-momentum quantum numbers s1 and s2 and orbital angular-
momentum quantum numbers ℓ1 and ℓ2. According to the Clebsch–Gordon series, we find
that S and L can take the following values:

S ¼ s1 þ s2 ; s1 þ s2  1; …; js1  s2 j (7.3)
L ¼ ℓ1 þ ℓ2 ; ℓ1 þ ℓ2  1; . . . ; jℓ1  ℓ2 j (7.4)

For a d 2 ion (ℓ1 = 2, s1 = ½; ℓ2 = 2, s2 = ½), the spin angular-momentum quantum number

S can have values of 1 and 0, while the orbital angular-momentum quantum number L can
have values of 4, 3, 2, 1, and 0. The total orbital angular momentum of a spectroscopic term is
denoted by the uppercase letters given in Table 7.3. The total spin is normally reported as the
value 2S + 1, which is called the multiplicity of the term. The different values of multiplicity
are given the names singlet (S = 0 with 2S + 1 = 1), doublet (S = ½ with 2S + 1 = 2), triplet
(S = 1 with 2S + 1 = 3), and so on.

Table 7.3 Term-symbol notation for the total orbital angular-momentum

quantum number L.

L= 0 1 2 3 4 5 6 ...
S P D F G H I ..
 7.3 Transitions between d Orbitals (d-to-d Excitations) 251

For a d 2 ion, the 45 microstates divide into the following five sets: nine microstates make up the
1
G (L = 4, S = 0) term, 21 belong to 3F (L = 3, S = 1), five to 1D (L = 2, S = 0), nine to 3P (L = 1,
S = 1), and one to 1S (L = 0, S = 0).4 The electron spins are parallel for the microstates associated
with the triplet terms, 3F and 3P, and antiparallel for the microstates associated with the singlet
terms, 1G, 1D, and 1S. Within the Russell–Saunders scheme, the total angular-momentum quantum
number J can take values of J = L + S, L + S − 1, |L − S|. For the 1G term, L = 4 and S = 0, which
means there is only a single value of J = 4, while for the 3F term, L = 3 and S = 1 give J = 4, 3, and 2.
Once the allowable values of L and S are known, it’s possible to identify the lowest-energy
term using Hund’s rules. Hund’s first rule tells us to maximize S as this minimizes electron
pairing that costs energy. This rule tells us that the triplet states, 3F and 3P, should be lower
in energy than the singlet states, 1G, 1D, and 1S. Hund’s second rule is to maximize L. This
rule tells us that the 3F state is lower in energy than the 3P state. Hund’s third rule states that
for orbitals which are less than half filled, the ground state J is the smallest possible sum of
L and S: |L − S|; whereas for orbitals more than half filled, it is the largest possible sum:
|L + S|. The basis of this rule is in the spin–orbit coupling itself, and it may be violated when
the coupling is weak. For the 3F state, the lowest-energy term has J = |L − S| = 2, and the
term symbol is 2S+1LJ = 3F2 (“triplet-F-two”). The ground-state terms for free transition-
metal ions with partially filled d-orbitals as predicted by Hund’s rules are given in Figure 7.5.
The splitting of the five different sets of degenerate microstates for the V3+ ion discussed above
is illustrated on a phenomenological (not to scale) energy diagram in Figure 7.6. Note that
Hund’s rules reliably give the ground-state (lowest-energy) term, but among the excited-state
terms the experimentally observed order does not necessarily match the order they predict.

mℓ 2 1 0 −1 −2 dn L S J term

d1 2 ½ 3 2D
3/2

d2 3 1 2 3F
2

d3 3 3 3 4F
3/2

d4 2 2 0 5D
0

d5 0 5 5 6S
5/2

d6 2 2 4 5D
4

d7 3 3 9 4F
9/2

d8 3 1 4 3F
4

d9 2 ½ 5 2D
5/2

Figure 7.5 The ground-state term symbols for free d n ions with n = 1–9. A representative microstate for
each term symbol is shown on the left.

4
The number of microstates in each term is equal to (2L + 1)(2S + 1).
252 Optical Materials

1S 1S
0

1D 1D
2
singlets 1G 1G
4

3P
2

energy
3P 3P
1
d2
3P
0

3F
4

3F
triplets 3F
3

3F
2

Hund’s first Hund’s second Hund’s third

rule rule rule

Figure 7.6 Interactions that split the degeneracy of a free-ion d 2 configuration (energies not to scale).

7.3.3 Correlation Diagrams

The relative energies of the spectroscopic terms developed in the preceding section corres-
pond to the free-ion case where all five d orbitals are degenerate. Once ligands are introduced,
the individual d orbitals no longer have the same energy, as discussed in Section 7.3.1.
Consequently, the free-ion spectroscopic terms discussed above are further split by ligand-
field effects, as shown for the octahedral case in Table 7.4. The labels found in the bottom
row describe the effects of both electron–electron correlations and ligand-field splitting.
We are now ready to properly explain the UV–visible spectrum of [Ti(H2O)6]3+ (assuming
perfect octahedral symmetry). There is only one spectroscopic term for a d1 ion, the 2D term
(Figure 7.5).5 Table 7.4 tells us that a free-ion 2D term splits into two terms, 2T2g and 2Eg, in
octahedral coordination. These terms correspond to an electron in one of the t2g orbitals or one of
the eg orbitals, respectively. The T2g → Eg transition energy depends on the ligand-field splitting
Δoct, as shown in the correlation diagram6 in Figure 7.7a. The vertical arrow in this figure shows
the electronic transition responsible for the absorption peak at ~490 nm in Figure 7.3.
The situation becomes more complicated for the d 2 configuration with a free-ion ground-state
term of 3F. As we will see in the next section, transitions between states with different spin
multiplicities are very weak and to a first approximation can be neglected. Therefore, we will
only consider the terms that have the same spin multiplicity as the ground state. In the d 2 case,
the only excited-state term that is also a triplet is 3P (Figure 7.6). These two states are shown in
Figure 7.7b. Introduction of the octahedral ligand field splits the 3F term into terms 3T1g, 3T2g,
and 3A2g, each with a different occupation of the orbitals that are split by the ligand field. The

5
We ignore splitting due to spin–orbit coupling for the sake of simplicity.
6
Correlation diagrams of this sort are also called Orgel diagrams.
 7.3 Transitions between d Orbitals (d-to-d Excitations) 253

Table 7.4 The splitting of spectroscopic terms of a free ion upon the introduction of an
octahedral ligand field.

Free-ion term S P D F G
Terms in octahedral field A1g T1g T2g + Eg T1g + T2g+A2g A1g + Eg+T1g + T2g

(a) d 1 ion t20g e1g

2E
g
energy

2D

2T
2g
t12 g eg0

Δoct
(b) d 2 ion t20g eg2
3A
2g

3T
1g(P)
energy

3P
t12 g e1g

3F 3T
2g

t22g eg0
3T
1g(F)

Δoct

Figure 7.7 Correlation diagram for (a) a d 1 ion and (b) a d 2 ion in octahedral coordination as a function
of Δoct. The vertical arrows show the transitions responsible for the absorption peaks in Figure 7.3. The
high-spin d 6 and d 7 correlation diagrams are analogous to the d 1 and d 2 diagrams, respectively, but with
different values of S.

3
T1g term corresponds to a configuration with two electrons in the t2g orbitals; the 3T2g term
corresponds to one electron in the t2g orbitals and one in the eg orbitals; and the 3A2g term
corresponds to two electrons in the eg orbitals. The 3P free-ion term becomes a 3T1g term in an
octahedral environment (electron configuration t2g1eg1) but is not split by the octahedral crystal
field. These features are captured in the correlation diagram shown in Figure 7.7b. More
comprehensive diagrams called Tanabe–Sugano diagrams must be used to understand transi-
tions between states with different spin multiplicities. We neglect such transitions for now but
will return to them in Section 7.8.5 when we consider the luminescence of d 5 ions like Mn2+.
254 Optical Materials

Returning to the spectrum of [V(H2O)6]3+ (Figure 7.3), we can now assign the absorption
peak at ~570 nm to the 3T1g(F) → 3T2g(F) transition, while the absorption at 390 nm arises
from the 3T1g(F) → 3T1g(P) transition. Both are marked with vertical arrows in Figure 7.7b.
The 3T1g(F) → 3A2g(F) transition not only requires UV photons of an even higher energy, it
involves simultaneous excitation of two electrons from the t2g to eg orbitals and would
therefore be exceedingly weak. Experimentally, it is not observed.
It is laborious to work out the correlation diagrams for all d-electron counts. Fortunately,
there are several relationships that simplify the analysis. Firstly, for high-spin ions (Section
5.3.9), the correlation diagram for a d n+5 ion is analogous to a d n ion. Thus, the d 6 configuration
diagram is identical to the d 1 case, and the d 7 diagram is analogous to d 2. Secondly, it can be
shown that a d 10−n ion has the same correlation diagram as a d n ion except that all energies of
interaction have the opposite sign. Thus, the d 9 correlation diagram is just the inverse of the d 1
diagram, and the d 8 diagram is the inverse of the d 2 diagram, etc.
Applying these two relationships, we find that the d 1, d 4, d 6, and d 9 configurations all
exhibit a single spin-allowed d-to-d transition, whereas the d 2, d 3, d 7, and d 8 configurations
have three spin-allowed d-to-d transitions (though transitions involving promotion of two
electrons between t2g and eg orbitals are generally not observed). The correlation diagrams
for high-spin d 3 and d 4 ions are given in Figure 7.8.
There is one feature in the correlation diagram of a d 3 ion that has not yet been explained.
Because the T1g(F) and T1g(P) terms have the same symmetry, they mix as Δoct increases, and
their energies bend away from each other. Table 7.5 gives the absorption maxima and corres-
ponding electronic transition for representative [M(H2O)6] n+ ions.

Table 7.5 d-to-d transitions for representative [M(H2O)6]n+ ions.

Absorption Transition Transition

Complex ion max (nm) energy (eV) assignment*

d1 [Ti(H2O)6]3+ 492 2.52 2

T2g → 2Eg
d2 [V(H2O)6]3+ 561 2.21 3
T1g → 3T2g
389 3.19 3
T1g → 3T1g(P)
d3 [Cr(H2O)6]3+ 575 2.16 4
A2g → 4T2g
407 3.05 4
A2g → 4T1g(F)
265 4.69 4
A2g → 4T1g(P)
d4 [Mn(H2O)6]3+ 476 2.60 5
Eg → 5T2g
d6 [Fe(H2O)6]2+ 962 1.29 5
T2g → 5Eg
d7 [Co(H2O)6]2+ 1230 1.00 4
T1g → 4T2g
515 2.40 4
T1g → 4T1g(P)
d8 [Ni(H2O)6]2+ 1180 1.05 3
A2g → 3T2g
725 1.71 3
A2g → 3T1g(F)
395 3.14 3
A2g → 3T1g(P)
d9 [Cu(H2O)6]2+ 794 1.56 2
Eg → 2T2g
*Assuming octahedral coordination. Jahn–Teller distortions are neglected.
 7.3 Transitions between d Orbitals (d-to-d Excitations) 255

(a) d 3 ion 4T
1g(P)

4P

energy
4T
1g(F)
4F

4T
2g

4A
2g

Δoct
(b) d4 ion 5T
2g
energy

5D

5E
g

Δoct

Figure 7.8 Correlation diagram for (a) a d 3 ion and (b) a high-spin d 4 ion in an octahedral environment.
The d 8 and d 9 correlation diagrams are analogous to the d 3 and d 4 diagrams, respectively, but with
different values of S.

Correlation diagrams for tetrahedral coordination can be generated by using the

diagram for an octahedrally coordinated d n ion for a tetrahedrally coordinated d 10−n
ion. Consequently, the d 1 octahedral diagram is equivalent to the d 9 tetrahedral
diagram and vice versa. The d 2 octahedral diagram is equivalent to the d 8 tetrahedral
diagram and so on.

7.3.4 Selection Rules and Absorption Intensity

The intensity of an optical absorption is an important factor for applications. The strength
with which an electronic transition absorbs light is expressed by its molar extinction coeffi-
cient, εmax [in L/(mol cm)],7 composed of the absorption maximum Amax (dimensionless), the
concentration c of the solution (in mol/dm3 = mol/L), and the path length of the light,
l (in cm), according to Beer’s law:

7
The molar extinction coefficient is also referred to as molar absorptivity and molar attenuation coefficient, which is
the term preferred by IUPAC. Reduced to base SI units, L/(mol cm) is equivalent to m2/mol.
256 Optical Materials

Table 7.6 Typical molar absorption coefficients, εmax, for localized transitions
in transition-metal complexes.

Transition εmax, L/(mol cm)

Spin-forbidden <1
Laporte-forbidden, d-to-d centrosymmetric complex 1–20
Laporte-allowed, d-to-d non-centrosymmetric complex 10–1000
Symmetry-allowed charge transfer 1000–50000

Amax
εmax ¼ (7.5)
cl

We can estimate the strength of an electronic transition by considering the following

selection rules. The spin selection rule states that transitions between states of different
spin, S, are forbidden. The basis of this rule lies in the fact that the incident radiation
cannot change the relative orientations of the spins of an electron. Spin–orbit coupling
can relax the spin selection rule, but nonetheless spin-forbidden transitions (ΔS ≠ 0) are
always much weaker than spin-allowed transitions (ΔS = 0). This explains the very
weak absorptions and pale colors of complexes and compounds containing high-spin
d 5 ions.8
The Laporte selection rule states that in a centrosymmetric environment the only allowed
transitions are those between states that have a different parity with respect to the inversion
operation.9 Practically speaking, this means that transitions s → s, p → p, d → d, and f → f are
forbidden for ions in a centrosymmetric environment, whereas s → p, p → d, and d →
f transitions are allowed. Although ML6 octahedral complexes are centrosymmetric, weak
d-to-d transitions are observed because vibrations temporarily remove the inversion center
and relax the Laporte selection rule. However, these d-to-d transitions are much weaker than
fully allowed transitions such as charge-transfer excitations. Approximate values of molar
extinction coefficients for several types of transitions are given in Table 7.6. Revisiting
Table 7.1, we see that few pigments derive their color from d-to-d transitions because these
are relatively inefficient at absorbing light. The few exceptions have transition-metal ions in
non-centrosymmetric environments, like tetrahedral and trigonal-bipyramidal geometries,
where d-to-d transitions do not violate the Laporte selection rule. For gemstones, it is
generally desirable that some light pass through the crystal, hence the light absorption
need not be so efficient and d-to-d transitions are often the source of color.

8
You may be wondering how Fe2O3 (red ochre) and FeOOH (yellow ochre) could be used as pigments given the d 5
configuration of Fe3+. It is because the color comes from charge-transfer transitions (Section 7.4.1) rather than from
d-to-d transitions.
9
More precisely, transitions between states with a gerade (German for even), g, and an ungerade (German for odd), u,
label are permitted, but g → g and u → u transitions are forbidden.
 7.3 Transitions between d Orbitals (d-to-d Excitations) 257

Box 7.1 Materials Spotlight: Blue pigments through the ages

The evolution of blue pigments provides an interesting illustration of the intersection
between art, commerce, and science. Until the eighteenth century, ultramarine was the
most highly prized blue pigment. During the Renaissance, it is estimated to have been five
times more expensive than gold, which limited its use to artists with wealthy patrons.
Michelangelo famously used large quantities of ultramarine in his depiction of the Last
Judgment on the altar wall of the Sistine Chapel. Natural ultramarine is obtained by
grinding the semi-precious stone, lapis lazuli, of which the dominant component is the
mineral lazurite, an aluminosilicate with the sodalite structure (Section 1.5.5). The name
ultramarine means “beyond the sea” because the pigment was imported to Europe from
mines in what is now Afghanistan. Pure sodalite, which is colorless, has the composition
Na4Al3Si3O12Cl. In lazurite, the Cl− ions, which sit near the center of the sodalite cages
(see Figure 1.57), are partially replaced by S3−, analogous to the ozonide anion O3−,
together with smaller concentrations of S2− and S4−. Electronic transitions between MOs
belonging to S3− give rise to the blue color, while contributions from S2− and S4− can
shift the color towards yellow or red, respectively [1, 2]. The high cost of ultramarine
made alternative blue pigments, such as the copper-containing mineral azurite, popular
during the Renaissance.
In the early years of the eighteenth century, Prussian blue, Fe4[Fe(CN)6]3·xH2O, was
discovered in Berlin. It can be found in paintings from the Prussian court that date back
to 1710. An Fe2+→ Fe3+ charge-transfer excitation gives rise to the intense blue color
(Section 7.4.2). Given the scarcity and cost of ultramarine, Prussian blue quickly gained
popularity, becoming the first widely used synthetic pigment. A century later, manufac-
ture of cobalt blue, CoAl2O4, was initiated. Cobalt blue is a normal spinel (Section 1.5.1)
that gets its blue color from d-to-d transitions associated with tetrahedrally coordinated
Co2+. Cobalt blue and other synthetic pigments were instrumental to artists from the
impressionist era, such as van Gogh, Monet, and Renoir. In modern times, copper
pthalocyanine, Cu(C32N8H16), has become the dominant blue pigment. Interestingly,
the blue color comes largely from π-to-π* transitions associated with the pthalocyanine
ring rather than from Cu2+ d-to-d transitions.
In 2009, researchers at Oregon State University discovered a new blue pigment with
composition YIn1−xMnxO3, where the blue color comes from d-to-d transitions of
isolated Mn3+ in a trigonal-bipyramidal coordination [3]. The spectra of CoAl2O4 and
YIn0.8Mn0.2O3 (shown in Figure B7.1.1) are similar in the visible range, resulting in
a vibrant blue color due to absorbed R + G (note the figure shows reflectance spectra,
the opposite of absorbance). In YIn1−xMnxO3, the trigonal-bipyramidal coordination
splits the Mn3+ d orbitals into a 2 + 2 + 1 pattern. The visible absorption peak is due to
a transition from the half-filled and doubly degenerate dxy/dx2−y2 orbitals into the empty
dz2 orbital. In CoAl2O4, visible light is absorbed by a 4A2 → 4T1(P) transition on the
tetrahedrally coordinated Co2+. Crucially, the transition metal in both materials sits on
a site that lacks inversion symmetry, so the transitions are Laporte allowed.
258 Optical Materials

Box 7.1 (cont.)

1.0

YIn0.8Mn0.2O3
0.8
percent reflectance

0.6
dz2 2.0 eV

0.4 CoAl2O4

0.2 dxy dx2−y2

0.0
200 400 600 800 1000 1200 1400 1600 1800 2000 dxz dyz
wavelength (nm)

Figure B7.1.1 The reflectance spectra for YIn0.8Mn0.2O3 and CoAl2O4 and the d-orbital energies and occu-
pancies for the Mn3+ ion in a trigonal-bipyramidal coordination. The absorption (where the reflectance is
minimal) between 500 and 700 nm in YIn0.8Mn0.2O3 arises from an electronic transition from the dxy/dx2−y2
orbitals into the empty dz2 orbital. Data taken from [4].

In both materials, there are also d-to-d transitions that fall outside the visible range. In cobalt
blue there is a broad 4A2 → 4T1(F) transition centered in the infrared (IR) near 1350 nm, while in
YIn0.8Mn0.2O3 the dxz/dyz to dz2 transition lies in the near UV. While the strong IR absorbance
of CoAl2O4 has little impact on the color, it does lead to absorption of IR light, which can
produce unwanted heating. The reduced IR absorbance in YIn0.8Mn0.2O3 is advantageous for
use on roofs and other exterior architectural applications. Unfortunately, the high cost of
indium makes YIn1−xMnxO3 pigments costlier than conventional blue pigments.

7.4 Charge-Transfer Excitations

In a charge-transfer excitation an electron is transferred from an MO whose wavefunction is
predominantly associated with a given atom (or group of like atoms) to an MO whose
wavefunction is largely associated with a different atom (or group of like atoms). Transitions
where an electron is excited from a ligand-based MO to a metal-based MO are called ligand-
to-metal charge transfer (LMCT) transitions, while a metal-to-ligand charge transfer (MLCT)
 7.4 Charge-Transfer Excitations 259

is in the opposite direction. A metal-to-metal charge transfer (MMCT) refers to the transfer
of an electron from one metal center to a different metal center. Charge-transfer transitions
are fully allowed, with molar extinction coefficients several orders of magnitude larger than
d-to-d transitions (Table 7.6), an attribute that is ideal for a pigment because bold colors can
be realized with relatively small amounts of pigment.
In this section, we take a closer look at LMCT and MMCT transitions. MLCT transitions
that fall in the visible region of the spectrum generally require ligands with low-lying
unoccupied π* orbitals, such as bipyridine or phenanthroline. These transitions are rare
among pigments and unknown for gemstones, and thus will not be considered here.

7.4.1 Ligand-to-Metal Charge Transfer

A variety of pigments rely on LMCT transitions for their color. Perhaps the best-known
examples are salts of the chromate ion, CrO42−, such as PbCrO4 (chrome yellow). The MO
diagram for a tetrahedral CrO42− anion is shown together with the UV–visible absorption
spectrum of its aqueous solution in Figure 7.9. A single absorption peak is centered at
375 nm. Even though it reaches a maximum in the UV, it tails into the visible region,
absorbing the short-wavelength blue light. The remaining green and red light are transmitted
producing the yellow color of chromate (G + R = Y).
The highest occupied molecular orbital (HOMO)-to-lowest unoccupied molecular
orbital (LUMO) transition from the triply degenerate t1 set of orbitals to the doubly
Cr 3d−O 2p MOs

t2
antibonding

5000
dxy dxz dyz t1 → e
molar extinction coefficient, ε [L/(mol cm)]

t1 → t2
4000
e t2 → e
dz2 dx2−y2
3.3 eV

3000

2000

t1
nonbonding
O 2p MOs

1000

t2
0
250 300 350 400 450 500
a1 wavelength (nm)

Figure 7.9 A portion of the MO diagram for tetrahedral CrO42− (left) along with the UV–visible spectrum
of a dilute (0.001 M) aqueous solution of Na2CrO4 (right).
260 Optical Materials

Table 7.7 Calculated values of the LMCT gap between the t1 nonbonding O 2p HOMO and the
e antibonding LUMO for a series of tetrahedral MO4n− species where M is a d 0 transition-metal ion [6].

t1 → e t1 → e t1 → e t1 → e
Group 5 (eV) Group 6 (eV) Group 7 (eV) Group 8 (eV)

VO43− 4.5 CrO42− 3.3 MnO4− 2.2

MoO42− 5.3 TcO4− 4.3 RuO4 3.1
WO42− 6.2 ReO4− 5.3 OsO4 4.0

degenerate e set of orbitals is responsible for the 375 nm absorption peak. Recall from
our previous discussion of the MO diagram of a tetrahedrally coordinated transition-metal
ion (Section 5.3.8), that the unoccupied e orbitals are antibonding MOs with significant
Cr 3dx2−y2 and 3dz2 orbital character, whereas the t1 orbitals are nonbonding O 2p orbitals. To
a first approximation, the transition is a transfer of an electron from oxygen to chromium (the
peak at 270 nm is a combination of transitions from lower-lying occupied O 2p t2 MOs into the
LUMO and transitions from the t1 HOMO into the unoccupied Cr-based t2 orbitals [5]).
The energy of the LMCT transition for a 3d tetrahedral MO4n− anion decreases with increas-
ing oxidation number of the transition metal, which lowers the e set of orbitals towards the
nonbonding oxygen MOs, reducing the LMCT gap (Table 7.7). The CrO42− and RuO42− ions
absorb blue and are thus yellow (G + R = Y), whereas MnO4− absorbs strongly in the green
range and the remaining B + R = M gives a magenta color. The other MO4n− anions that contain
4d and 5d metals are colorless because their LMCT transitions fall well into the UV.
The underlying causes of these periodic trends can be understood as follows. The increase in
oxidation state of the central atom, V(V) → Cr(VI) → Mn(VII), increases the effective nuclear
charge felt by the valence electrons, lowering the energy of the metal-based e and t2 orbitals,
thereby reducing the HOMO–LUMO gap. The second trend is an increase in the LMCT energy
on moving down a group (CrO42− → MoO42− → WO42−). This trend is due to the increase in the
relative sizes of the d orbitals, r3d < r4d < r5d (Section 5.2.3), increasing the overlap of the d orbitals
with the orbitals of the ligands and raising the energy of the antibonding e and t2 MOs. Another
contributing factor is the upward shift in the energies of the d orbitals that occurs on moving
down a group (Section 5.2.1). It is useful to remember these trends in d-orbital energies and
metal–ligand mixing as we explore various properties of transition-metal compounds.

7.4.2 Metal-to-Metal Charge Transfer

Blue sapphires and rubies are doped forms of the mineral corundum, Al2O3 (Figure 1.30).
Whereas d-to-d transitions on Cr3+ dopants are responsible for the red color of a ruby, iron
and titanium dopants are responsible for the blue color of a sapphire. Corundum doped with
small amounts of titanium is colorless, whereas similar amounts of iron lead to a pale-yellow
color. When both are present, the result is the magnificent deep-blue color of a sapphire. In
 7.5 Compound Semiconductors 261

e−
Ti4+

C N
e− O

Fe2+ Fe3+
Fe2+

MMCT in sapphire MMCT in Prussian blue

Figure 7.10 A localized view of the MMCT transitions in sapphire, Al2O3:Ti4+,Fe2+ (left), and Prussian
blue, (Fe3+)4[(Fe2+)(CN)6]3·xH2O (x ≈ 14) (right).

sapphires, co-doping by equal amounts of Ti4+ and Fe2+ maintains charge balance. When they
occupy adjacent six-coordinate sites in the corundum structure (Figure 7.10), it is possible for
the absorbed photon to excite an Fe2+ + Ti4+ → Fe3+ + Ti3+ MMCT transition.
Because of the differences in molar extinction coefficients, doping levels in sapphires and
rubies are different. At least 1% chromium must be present in corundum before the deep-
ruby-red color appears, whereas the blue color of a sapphire is observed with titanium and
iron concentrations as low as 0.01%. In fact, complete substitution leads to the mineral
ilmenite, FeTiO3 (Figure 5.28), where the charge transfer band is so intense that it absorbs
across the visible spectrum giving FeTiO3 its black color.
Prussian blue is another blue compound whose color originates from an MMCT excita-
tion. It is readily prepared by combining aqueous solutions of Fe3+ and ferrocyanide through
the following reaction:

4½FeðH2 OÞ6 3þ þ 3½FeðCNÞ6 4– → Fe4 ½FeðCNÞ6 3 xH2 O (7.6)

Prussian blue contains linear Fe2+–C≡N–Fe3+ linkages (see Figure 7.10) that make up
an infinite 3D cubic network. The carbon end of the cyanide group coordinates to a low-
spin d 6 Fe2+ center, while the nitrogen end coordinates to a high-spin d 5 Fe3+ center. An
intense Fe2+ → Fe3+ MMCT band centered near 705 nm absorbs visible light with λ > 500 nm.
The reflected/transmitted blue and violet light gives the distinctive color of Prussian blue.

7.5 Compound Semiconductors

Thus far, our discussions have been limited to electronic transitions that can be described
using a localized picture of bonding. Color can also be realized in systems with delocalized
bond networks, such as semiconductors. Historically, many red, orange, and yellow inor-
ganic pigments have been semiconductors with band gaps that selectively absorb a portion of
262 Optical Materials

the visible spectrum. Examples include orpiment (As2S3), cadmium yellow (CdS), and
vermillion (HgS). To understand the color in these compounds, we need to take a closer
look at the relationships between color, band gap, and composition.

7.5.1 Optical Absorbance, Band Gap, and Color

As discussed in Section 6.3, in semiconductors a band gap (Eg) separates the occupied valence
bands from the empty conduction bands. Semiconductors cannot absorb photons with ener-
gies less than Eg, but the presence of continuous bands means that they absorb photons more
energetic than Eg over a broad range of wavelengths. Consequently, longer-wavelength light
(hν < Eg) is either transmitted or reflected, while shorter-wavelength light (hν > Eg) is absorbed.
The fact that semiconductors absorb light with photon energies exceeding Eg limits their
colors (Figure 7.11). The most energetic visible photons have wavelengths of ~400 nm and
energies of 3.1 eV. Semiconductors that have band gaps larger than ~3.1 eV, like ZnS, do not
absorb visible light and are white in color. As the band gap decreases, B is absorbed, and the
reflected G + R = Y. A vibrant yellow is realized in CdS, where Eg = 2.4 eV. Further
reduction of the band gap gradually leads to the absorption of green in addition to blue, and
the reflected light changes first to orange and then to red (when both B and G are absorbed
equally), as exemplified by HgS with a band gap of 2.0 eV. Further decrease in the band gap
darkens the red color until it becomes black for Eg ≤ 1.7 eV, and all visible light is absorbed.
By forming solid solutions, it is possible to precisely control the band gap and tune the color.
This strategy has been effectively pursued to make an entire family of pigments amongst solid

wavelength (nm)
350 400 500 600 700

UV blue green red IR

ZnS (white)
absorbance (arbitrary units)

Eg = 3.6 eV

CdS (yellow)
Eg = 2.4 eV

HgS (red)
Eg = 2.0 eV

CdSe (black)
Eg = 1.7 eV

4.0 3.5 3.0 2.5 2.0 1.5

hν (eV)

Figure 7.11 Simulated absorbance profiles of four different semiconductors. The curves are offset along
the y axis for clarity. The vertical gray lines bracket the range of the visible portion of the spectrum.
 7.5 Compound Semiconductors 263

solutions formed between CdS (Eg = 2.42 eV) and CdSe (Eg = 1.73 eV). The colors of these
pigments range from yellow over orange to red and finally black. Although cadmium-based
pigments have many desirable characteristics, their use has declined due to concerns surround-
ing the toxicity of cadmium. This has spurred efforts to find non-toxic inorganic red, orange,
and yellow pigments to replace the lead and cadmium compounds. Candidates that have been
proposed include the orange Ca0.5La0.5TaO1.5N1.5 [7] and the red Ce2S3 [8].
The abruptness of the upturn in absorbance once hν > Eg depends in large part on whether
a semiconductor has a direct or an indirect band gap (Section 6.3.2). The absorption coefficient,
α, for a direct band-gap semiconductor is proportional to (hν − Eg)1/2, whereas for an indirect
band-gap semiconductor α is proportional to (hν − Eg)2. Because hν − Eg ≪ 1 near the band gap,
its square root rises sharply upon increasing hν, and a direct band-gap semiconductor has a much
sharper absorption edge than a comparable indirect band-gap semiconductor.

7.5.2 Electronegativity, Orbital Overlap, and Band Gap

While silicon remains the dominant material for electronic applications, elemental semicon-
ductors have significant limitations when it comes to optical and optoelectronic applications.
In addition to covering a limited range of band-gap energies, silicon and germanium both
possess indirect band gaps. Therefore, it is important to understand how the electronic band
structures of compound semiconductors differ from those of elemental semiconductors that
were previously discussed in Section 6.6.3.
The sphalerite-type (zinc blende) structure (Figure 1.32) of GaAs is an ordered variant of
the diamond network of Ge (Figure 1.37). The band structure of GaAs (Figure 7.12) has

10
conduction
bands

Γ p*
5

Γ s*
energy (eV)

Eg = 1.4 eV
0
Γp
valence bands

-5

-10
Γs Ga PDOS
As PDOS
-15
W L Γ X W
density of states

Figure 7.12 The band structure of GaAs (left) together with the partial density-of-states (PDOS) plot
showing the individual contributions of Ga and As.
264 Optical Materials

much in common with that of Ge (Figure 6.28), but there are also differences important
for applications. GaAs has a larger band gap (1.4 eV versus 0.7 eV), and the conduction-
band minimum is located at the Γ point. This means GaAs is a direct-gap semiconductor,
unlike Ge.
The increase in band gap comes from the difference in electronegativity between Ga and
As, which introduces ionic character into the bonding. This can be seen in the PDOS plot
shown on the right-hand side of Figure 7.12. The more electronegative arsenic makes a larger
contribution to the valence bands, while gallium makes a larger contribution to the conduc-
tion bands. If we further increase the electronegativity difference by going to ZnSe, the band
gap increases to 2.6 eV.
The band gaps of several semiconductors are shown in Table 7.8. As we move down the
periodic table (e.g. AlP → GaAs → InSb), the interatomic distance increases and the band
gap decreases. This effect arises from decreased orbital overlap, which makes the valence
bands less bonding and the conduction bands less antibonding. When we increase the
horizontal spacing of the two main-group elements (e.g. Ge → GaAs → ZnSe), the bond
distance remains reasonably constant, but the bond becomes more ionic, and, because the
more electronegative element makes a larger contribution to the valence band, its energy is
lowered. The opposite occurs for the conduction band, and the band gap increases. This type
of manipulation, sometimes referred to as band-gap engineering, plays an important role in
designing materials for many electrical and optical devices, including light-emitting diodes
(Section 7.9.1).

Table 7.8 Bond distances and optical band gaps, both direct (d) and indirect (i), for some sp3
semiconductors.

Elemental semiconductors III–V Semiconductors II–VI Semiconductors

Bond Band gap Bond Band gap Bond Band gap
distance (Å) (eV) distance (Å) (eV) distance (Å) (eV)

AlP 2.37 2.43 (i)

Si 2.35 1.11 (i) AlAs 2.45 2.16 (i)
AlSb 2.66 1.52 (i)
GaP 2.36 2.26 (i) ZnS 2.34 3.6 (d)
Ge 2.44 0.67 (i) GaAs 2.45 1.43 (d) ZnSe 2.45 2.58 (d)
GaSb 2.64 0.72 (d) ZnTe 2.64 2.25 (d)
InP 2.54 1.35 (d) CdS* 2.52 2.42 (d)
Sn† 2.81 ~0.0 InAs 2.62 0.36 (d) CdSe* 2.63 1.73 (d)
InSb 2.80 0.18 (d) CdTe 2.81 1.50 (d)
*Adopts the wurtzite structure, the bond distance given here is an average. †This refers to the α-Sn allotrope (also
called gray tin) that is isostructural with diamond and whose band gap is very small. Most sources describe α-Sn as a zero
band-gap semiconductor (a semimetal) [9].
 7.6 Conjugated Organic Molecules 265

7.6 Conjugated Organic Molecules

While many organic substances are colorless, those containing a conjugated network of
π bonds are an important exception. The π-to-π* transitions responsible for absorption of
visible light are allowed, and their molar absorption coefficients are high. Another attractive
feature is the ability to tune the energy of the π-to-π* transitions by changing the functional
groups on the periphery of the molecule. As we move through the book, we will see that
conjugated organic molecules make a disproportionately large contribution to the field of
functional organic materials.
Two historically important organic molecules used in dyes and pigments are shown in
Figure 7.13. Alizarin is the molecule that gives a red color to the dye extracted from the root
of a madder plant. This dye was widely used for centuries and is responsible for the color of
the “redcoats” worn by British soldiers until the early twentieth century. The color of this dye
can be captured in a pigment called madder lake by grinding it with an insoluble inorganic
substance, such as alumina. Indigo is another molecular substance from plants that has long
been used as a dye. It is most closely associated with the color of blue jeans. Today,
synthetically manufactured alizarin and indigo dyes have largely replaced dyes extracted
from plants.
Among the simplest conjugated aromatic molecules are the acenes, which are linearly
fused benzene rings. Benzene and the first four acenes are shown in Figure 7.14. The
electronic structure and MOs associated with the delocalized π network in benzene was
described in Section 5.3.7. Consider the six benzene π/π* MOs in Figure 5.19. As the
energies of the MOs increase, there is an increase in the number of nodal planes and
a progressive shift from nearest-neighbor interactions that are bonding to those that
are antibonding. In benzene and many other conjugated organic molecules, the MOs
with net bonding character are filled, while those with net antibonding character are
empty.
We can extend the same principles to the larger acenes. As with benzene, each
carbon atom contributes one 2p orbital to the π network, consequently, the number
of π MOs is equal to the number of carbon atoms. The number of nodal planes in the

O OH H
O
OH N

N
O
O H

alizarin, C14H8O4 (red) indigo, C15H9N2O (blue)

Figure 7.13 The molecular structures of alizarin and indigo.

266 Optical Materials

benzene napthalene anthracene tetracene pentacene

HOMO−LUMO HOMO−LUMO HOMO−LUMO HOMO−LUMO HOMO−LUMO
Δ = 4.69 eV Δ = 3.90 eV Δ = 3.12 eV Δ = 2.37 eV Δ = 1.85 eV

Figure 7.14 The HOMO-LUMO gaps, Δ, for benzene and the first four acenes.

HOMO increases from two in benzene (see Figure 5.19), to three in naphthalene, four
in anthracene, five in tetracene, and so on.10 Although benzene is colorless with a large
HOMO–LUMO gap (Δ = 4.69 eV), the gap decreases as the size of the π network
increases. When we reach tetracene, the lowest-energy π-to-π* transition has shifted
into the visible range, absorbing much of the blue and some green light, leading to an
orange color. The HOMO–LUMO gap shifts even further into the visible in pentacene,
leading to a dark-red coloration. This trend continues as the size of the π network
increases, until we reach the infinite network found in graphene where the band gap
goes to zero (Section 6.5.1). While the acenes are not used as pigments or dyes, they
form an important class of organic conductors.
The planar porphyrin macrocycle (Figure 7.15, left) is the chromophore responsible
for the color of important biological molecules such as chlorophyll. Of the related
phthalocyanines (Figure 7.15, right), copper phthalocyanine-based pigments make up
the largest class of commercial organic pigments. They are non-toxic, inexpensive,
strongly absorbing, thermally stable up to 300 °C, and do not fade appreciably after
extended exposure to light. Although we often associate the color blue with Cu2+ salts,
here it is the π-to-π* transitions of the phthalocyanine rather than Cu d-to-d transitions
that are largely responsible for the color, as demonstrated by the fact that zinc and
magnesium phthalocyanine are also blue.

10
Here we include the nodal plane that lies in the plane of the molecule.
 7.7 Luminescence 267

R
N

N N N N
N N
R M R Cu
N N N N

R
metal porphyrin phthalocyanine blue, Cu(C32N8H16)

Figure 7.15 A generic metal porphyrin molecule (left) and the copper phthalocyanine molecule (right).

7.7 Luminescence
Luminescence describes processes in which materials called phosphors11 are used to convert
various forms of energy into electromagnetic radiation, typically in the UV, visible, and IR
regions of the spectrum. Luminescent materials appear in a wide range of applications
including lighting, display technology, medical imaging, and radiation detection. The most
familiar form of luminescence is photoluminescence, where electrons are excited by absorp-
tion of light that is subsequently reemitted at a different wavelength. Other forms of
luminescence exist, depending on the source of energy used to excite electrons.
Electroluminescence is the direct conversion of electrical energy into light. This type of
luminescence, which is the basis for light-emitting diodes, is discussed in Section 7.9.
Cathodoluminescence occurs when a phosphor is exposed to a beam of electrons accelerated
by an electric field. Mechanical energy can also act as the input that leads to luminescence.
Examples include: triboluminescence, where fracturing materials leads to the emission of
light; piezoluminescence, which is triggered by the deformation of matter; and sonolumines-
cence, where ultrasonic waves are converted to light. Luminescence can originate from
chemical (chemiluminescence) or biochemical (bioluminescence) reactions when the products
formed are in electronically excited states.12 Thermoluminescence occurs when heat activates
electrons trapped in excited states, allowing them to relax in a radiative manner to the ground
state. Thermal stimulation of luminescence is a more accurate description of this process, since
the initial excitation and trapping of electrons is caused by interaction with either visible or
UV light (afterglow phosphors) or high-energy photons like X-rays and/or γ-rays (storage
phosphors).

11
Phosphor is Greek for “light bearer”. The element phosphorus shares the same root because white phosphorus is
chemiluminescent when slowly oxidized.
12
One of nature’s best-known examples of bioluminescence is the firefly (Photinus pyralis), which emits 560 nm light
through an enzyme-catalyzed oxidation of luciferin to oxyluciferin.
268 Optical Materials

7.8 Photoluminescence
We begin our treatment of photoluminescence with the definitions of some commonly
encountered terms. In most instances, the absorbed photons have a higher energy than
those emitted, and the overall process is called down-conversion photoluminescence. In
some phosphors, multiple photons of light are absorbed and a higher-energy photon emitted
through a process called up-conversion photoluminescence. The quantum efficiency of
a phosphor is defined as the ratio of the number of emitted photons to the number of
absorbed photons. If the electronic transition that leads to emission is a spin-allowed
transition (ΔS = 0), the process is called fluorescence. If it is a spin-forbidden transition
(typically ΔS = 1), it is called phosphorescence. The decay times13 associated with fluorescence
range from 10−11 s to 10−8 s, while phosphorescence has much slower decay times that range
from 10−6 s to 10−2 s. Persistent phosphors, such as SrAl2O4 doped with Eu2+ and Dy3+, emit
light for hundreds of seconds after the excitation source is turned off, but this does not
involve spin-forbidden transitions. Instead, this afterglow is the result of photoionization,
where incoming photons ionize a site (typically a cation) in the lattice, and the ionized
electron is trapped by anion vacancies from which it later escapes via thermal stimulation
(i.e. thermoluminescence). Nevertheless, this process is sometimes described in the literature
as phosphorescence.
The different electronic transitions that accompany luminescence can be summarized in
a Jablonski diagram (Figure 7.16). Following absorption of a photon, an electron is pro-
moted from the singlet ground state to a singlet excited state. Non-radiative internal conver-
sions or relaxations into the lowest-energy singlet excited state occur at rates faster than 1012
per second. The next step is often emission of fluorescent photons as the electron returns to
the ground state at rates of 1011 to 108 per second. An alternative pathway is non-radiative
intersystem crossing (ISC) from an excited singlet state to a triplet state. From here, phos-
phorescence back to the singlet ground state will occur at rates of 106 to 102 per second, five
to six orders of magnitude slower than fluorescence. Yet another pathway, not shown in
Figure 7.16, is a non-radiative return to the ground state, an undesirable process that
competes with luminescence.

7.8.1 Components of a Phosphor

In the most general sense, a phosphor consists of three components: sensitizers, which are
sites where incoming photons are absorbed; activators, which are sites where photolumines-
cence occurs through radiative relaxation of electrons; and a host, in which both sensitizers
and activators are embedded, as illustrated in Figure 7.17. In some phosphors, the sensitizer
and the activator are the same ion; when this is not the case, an efficient mechanism for

13
The decay time is the time for a steady-state luminescence intensity to decay to 1/e ≈ 36.8% of its original value.
 7.8 Photoluminescence 269

singlet
internal conversion
excited states

intersystem crossing (ISC)

triplet excited

fluorescence
absorption
state

phosphorescence

singlet ground state

Figure 7.16 A Jablonski diagram illustrating the electronic transitions that can occur in a phosphor after
absorption of a photon.

absorbed
photon
sensitizer

energy
transfer

activator

emitted
photon
host

Figure 7.17 The components of a phosphor.

energy transfer from the sensitizer to the activator is needed. Phosphors where the host acts
as sensitizer and activator are called self-activating phosphors.
The host is important for a variety of reasons. It determines the local coordination
environment of the sensitizer and the activator, which play an important role in determining
the wavelengths of the absorbed and emitted light. The energy and nature of the vibrations in
the host impact the probability that excited-state electrons will return to the ground state
radiatively. Finally, the host must possess a band gap large enough to allow the incident light
to reach the sensitizers and the emitted light to escape the phosphor.
270 Optical Materials

7.8.2 Radiative Return to the Ground State

The energy difference between absorbed and emitted photons with frequencies νexc and νem,
respectively, is referred to as the Stokes shift:

EStokes ¼ Eexc  Eem ¼ hνexc  hνem (7.7)

In down-conversion phosphors, the absorbed photon has a higher energy than the emitted
photon, and the Stokes shift is positive. In most cases, the energy difference is converted to
thermal energy via lattice vibrations. Coupling between electronic and vibrational energy,
which we will hereafter refer to as vibronic coupling or electron–phonon coupling, is an integral
part of the process and must be included in any treatment of luminescence.
Parabolic potential-energy curves are used to describe the ground and excited electronic
states in the configurational coordinate model shown in Figure 7.18. In this depiction, Q is
a configurational coordinate used to describe a specific vibrational mode of the luminescent
center. In an approximate sense, we can think of Q as representing a bond distance between the
activator and the ligands that surround it. The horizontal lines or “rungs” represent vibrational
states. They are quantized and span a range of values on the horizontal axis due to the dynamic
expansion and contraction of the bonds associated with the relevant vibrational mode.
The coordinates Qg and Qe represent the equilibrium metal–ligand distances in the ground
and excited state, respectively. Absorption of a photon excites an electron into a higher-lying

ground excited
state state
2000

1800

1600 nʹ = 4
nʹ = 3
1400
nʹ = 2
nʹ = 1
energy

1200
nʹ = 0
1000

Eexc Eem
800

n=4
600
n=3
400 n=2
n=1
200 n=0
0
0 2 4 6 8 ∆Q 10 12 14
Q 16
Qg Qe

Figure 7.18 Configurational coordinate diagram where Qg and Qe represent the equilibrium bond
distance of the ground and excited states, respectively. The accessible absorption and emission transitions
are shown with vertical arrows, bold arrows signify the most intense transitions.
 7.8 Photoluminescence 271

electronic state that will, in general, have more antibonding character than the ground state,
leading to a weakening of the chemical bond. Consequently, the equilibrium bond length of
the excited state will be larger than in the ground state (Qe > Qg). As we will see, the resultant
shift in equilibrium bond distance, ΔQ = Qe − Qg, is closely coupled to many important
characteristics of the luminescence.
If we assume the vibrational motion to be harmonic,14 the energies of the allowed
vibrational states are those of a quantum harmonic oscillator:
 
1
En ¼ n þ ℏω (7.8)
2

where n = 0, 1, 2, 3, . . ., and ω is the angular frequency of the oscillator.15 At room

temperature (and below), the lowest-energy vibrational state is typically the most highly
populated state and is centered at Qg (the center of the n = 0 rung). Therefore, most electronic
excitations originate from this position. The wavefunctions for the excited vibrational states
tend to peak at values of Q that are close to where the rung meets the parabola, not unlike the
way a pendulum spends more time at the turning points than it does at the bottom of its arc.
According to the Franck–Condon principle, electronic excitations occur on timescales much
faster than vibrations. Optical transitions are therefore depicted as vertical lines in Figure 7.18,
representing no change in bond length upon absorption or emission of a photon. The bold
vertical arrow labeled Eexc corresponds to the optical excitation that has the maximum
intensity. It goes from the center of the n = 0 rung on the ground-state parabola, where the
ground-state wavefunction has maximum probability, to a point where the nʹ = 2 vibrational-
state wavefunction has a high probability (i.e. near the intersection of the nʹ = 2 rung and the
excited-state parabola). This transition will be more intense than transitions to other excited
vibrational states, because their wavefunctions reach their highest probabilities at more distant
values of Q. Although the n = 0 to nʹ = 2 transition has maximum intensity, less-intense
transitions to other states, shown with gray vertical arrows in Figure 7.18, can also occur.
Once in the excited state, the electron rapidly relaxes to the ground vibrational state
(nʹ = 0) via electron–phonon coupling. From this state, it can return to the ground electronic
state through emission of a photon, following the same principles that govern absorption.
The vertical arrow labeled Eem corresponds to the optical emission that has the maximum
intensity. Following emission of a photon, the electron returns to the original ground state
through further coupling to lattice vibrations. Thus, we see that electron–phonon coupling in
both the excited- and ground-state parabolas is the origin of the Stokes shift.
The magnitude of ΔQ in Figure 7.18 gives a measure of the changes in chemical bonding
that accompany promotion of an electron into an excited state. This change is captured by

14
The harmonic approximation requires that the restoring force, F, is proportional to displacement, ΔQ; F = −kF ΔQ,
where kF is a force constant, resulting in a parabolic potential-energy curve, E = ½kF ΔQ2.
15
The angular frequency of a simple two-body harmonic oscillator is defined as ω = (kF /μr)1/2 where kF is the force
constant and μr = (m1m2)/(m1 + m2) is the reduced mass of the oscillator.
272 Optical Materials

the dimensionless Huang–Rhys parameter, S, which is proportional to (ΔQ). Under weak

electron–phonon coupling (say, S < 1), the shift ΔQ of equilibrium bond distances between
the ground and excited states is small. That is the case for electronic transitions between
4f orbitals, because the 4f orbitals have minimal interactions with the ligands and therefore
little impact on the bond distances. In the ΔQ = 0 limit, the ground-state vibrational
wavefunctions for both parabolas have maxima at the same value of Q. In this limit, the
absorption and emission spectra will consist of single lines called zero-phonon lines, and the
Stokes shift will be zero. For small, yet non-zero ΔQ, a small Stokes shift and narrow
absorption/emission lines occur. Large ΔQ indicates strong electron–phonon coupling
(S > 5). This happens when the excited electronic state has very different bonding character
than the ground electronic state. Activators from the p block of the periodic table (e.g. Pb2+,
Bi3+) or oxyanions (e.g. VO43−, WO42−) often fall in the strong-coupling regime. Because of
the large ΔQ, the ground vibrational state of the lower parabola (n = 0) has substantial
overlap with several excited vibrational states on the upper parabola, leading to a broad
absorption band. For similar reasons, the emission bands will also be broad. Large ΔQ also
leads to considerable vibrational relaxation following both absorption and emission, which
makes for a large Stokes shift.
Not surprisingly, absorption and emission spectra change with temperature due to
changes in the thermal population of different vibrational levels on both parabolas.
Whereas at very low temperatures the fine structure of absorption and emission spectra
can be resolved, broadening occurs at higher temperatures, which can result in unresolved
absorption and emission bands.

7.8.3 Thermal Quenching

We now consider undesirable non-radiative processes that offer the excited electron alterna-
tive paths to the ground state. One of the most important is thermal quenching, which refers
to the process of electrons returning to the ground state by dissipating energy through lattice
vibrations. Consider the configurational coordinate diagrams in Figure 7.19, which depict
small and large ΔQ. The energy difference between the ground vibrational level in the excited
state and the energy where the two parabolas cross is denoted as ΔE. When ΔQ is large, the
two parabolas cross at a relatively low energy and ΔE is small. In this case, vibrational levels
close to the crossing point will have non-negligible populations at modest temperatures,16
allowing electrons to cross over from the excited-state parabola to the ground-state parabola
where they can return to the ground state non-radiatively through coupling with lattice
vibrations. When ΔQ is small, however, the curves cross at much higher energies, and the
vibrational states that lie at or above the crossing point only acquire non-negligible

16
The population of vibrational states is given by a Boltzmann distribution, where the probability of being in an
excited state whose energy is E above the ground state is proportional to e−E/kT, where the thermal energy is given by
kT, with k = 1.380649×10−23 J/K. For a temperature of 300 K, kT = 0.0258 eV.
 7.8 Photoluminescence 273

Small ∆Q Large ∆Q
minimal thermal quenching significant thermal quenching

2000 2000

1500 1500
∆E
energy

energy
∆E
1000 1000

500 500

0 0
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Q Q
∆Q ∆Q

Figure 7.19 Configurational coordinate diagrams for small (left) and large (right) ΔQ. The parameter
ΔE represents the energy between the ground vibrational state of the excited electronic state and the
energy where the parabolas cross.

populations at high temperatures. In such cases, thermal quenching at room temperature or

below tends to be minimal.
Even in cases where ΔQ ≈ 0, thermal quenching can still occur if the energy difference
between ground- and excited-state parabolas is small. In such cases, the electron can return
directly to the ground-state parabola through a process called multi-phonon emission, where
the lost electronic energy generates several high-energy phonons in the surrounding lattice.
As a rule of thumb, multi-phonon emission becomes significant when the energy of the
highest-energy phonon mode, E = hνmax, exceeds roughly 20% of the electronic energy
difference between the ground vibrational states of the two parabolas. This is a common
non-radiative decay pathway for many rare-earth ions,
The thermal-quenching temperature, T1/2, is the temperature at which a phosphor loses 50% of
its emission intensity with respect to an arbitrarily defined base temperature. High values of T1/2
are desirable, and in many industrially relevant phosphors it is greater than 100 °C (for base
temperature = room temperature). Phosphors with high T1/2 usually have stiff hosts to limit
expansion of bond distances in the excited state and keep ΔQ small. If the lattice softens,
ΔQ increases and ΔE decreases, making non-radiative return to the ground state more probable.
To illustrate the importance of the “stiff” host on thermal-quenching behavior, consider
the ordered double perovskites, Ba2M(W1−xUx)O6, with M = Mg2+, Ca2+, Sr2+, and Ba2+.
The UO6/2 entity acts as both sensitizer and activator. An O 2p → U 6d LMCT transition is
responsible for light absorption, and ΔQ is large since the uranium-centered octahedra will
expand and distort when the excited-state antibonding orbitals are populated. This expan-
sion will compress the M–O bonds of the coordination octahedra that alternate with the
UO6/2 octahedra. As the size of the M2+ cation increases, the M–O bonds lengthen and
274 Optical Materials

become softer, and this leads to an increase in ΔQ that is estimated to be 2% for M = Ca2+,
6% for Sr2+, and 9% for Ba2+, with respect to the ΔQ of Ba2Mg(W1−xUx)O6. This softening
leads to a decrease in T1/2 values (referenced to a base temperature of 4 K) from 350 K in
Ba2Mg(W1−xUx)O6 to 310 K (M = Ca2+), 240 K (Sr2+), and 180 K (Ba2+) [10].

7.8.4 Lanthanoid Activators

Lanthanoid ions are an important class of activators and can be divided into two categories.
When the optical transitions are between different 4f states, the absorption and emission
lines are very narrow, as expected in the weak coupling limit, and these activators are called
line emitters. When they involve 4f-to-5d transitions, the excited-state 5d wavefunctions have
substantial ligand character, leading to an increase in ΔQ that broadens the excitation and
emission transitions into bands. Lanthanoid activators of this type are called band emitters.
The absorption and emission energies of line emitters are relatively insensitive to their
surroundings, while those of band emitters can be altered by modifying the chemical
surroundings of the activator ion.
We start with line emitters. The lanthanoid elements have a strong preference for the 3+
oxidation state with a [Xe]4f n electron configuration (Table 7.9), but some can also take
either a 2+ or a 4+ oxidation state, particularly when it leads to an empty (Ce4+), half filled
(Eu2+, Tb4+), or completely filled (Yb2+) 4f subshell. To understand the energy levels of partially

Table 7.9 Ground-state electron configurations of lanthanoid ions.

Electron Ground-state Electron Ground-state

Z Element Ion configuration term Ion configuration term

57 Lanthanum La3+ [Xe]4f 0 1

S0
58 Cerium Ce3+ [Xe]4f 1 2
F5/2 Ce4+ [Xe]4f 0 1
S0
59 Praseodymium Pr3+ [Xe]4f 2 3
H4 Pr4+ [Xe]4f 1 2
F5/2
60 Neodymium Nd3+ [Xe]4f 3 4
I9/2
61 Promethium* Pm3+ [Xe]4f 4 5
I4
62 Samarium Sm3+ [Xe]4f 5 6
H5/2 Sm2+ [Xe]4f 6 7
F0
63 Europium Eu3+ [Xe]4f 6 7
F0 Eu2+ [Xe]4f 7 8
S7/2
64 Gadolinium Gd3+ [Xe]4f 7 8
S7/2
65 Terbium Tb3+ [Xe]4f 8 7
F6 Tb4+ [Xe]4f 7 8
S7/2
66 Dysprosium Dy3+ [Xe]4f 9 6
H15/2
67 Holmium Ho3+ [Xe]4f 10 5
I8
68 Erbium Er3+ [Xe]4f 11 4
I15/2
69 Thulium Tm3+ [Xe]4f 12 3
H6
70 Ytterbium Yb3+ [Xe]4f 13 2
F7/2 Yb2+ [Xe]4f 14 1
S0
71 Lutetium Lu3+ [Xe]4f 14 1
S0
*Promethium does not occur naturally. It is radioactive and its longest-lived isotope, 145Pm, has a half-life of 17.7 years.
 7.8 Photoluminescence 275

filled 4f orbitals, we need to return to the microstates described by term symbols first introduced
in Section 7.3.2. The orbital angular-momentum quantum numbers L can vary from 0 to 6, while
the spin angular-momentum quantum numbers can vary from 0 to 7=2. The possible number of
microstates is large and can be calculated with the expression 14!/[nf !(14−nf)!], with nf being the
number of f-electrons present. Using this expression, we find 3432, 364, and 14 possible micro-
states for the Gd3+ (f 7), Nd3+ (f 3), and Ce3+ (f 1) ions, respectively.
The interactions that split the 4f levels are spin–spin, spin–orbit, and orbit–orbit coupling,
which are considerably stronger for the lanthanoids than for the 3d transition-metal ions.
Conversely, the crystal-field splitting is much smaller than it is for the 3d ions due to the
limited radial extension of the 4f orbitals. We therefore concentrate on the effects of
interelectron coupling, and only afterwards allow energies to be shifted by crystal-/ligand-
field perturbations. The close similarity of the optical spectra of free lanthanoid ions and
those in compounds supports the validity of this approach.
In the 1960s, Dieke and co-workers analyzed optical spectra for Ln3+ ions in LaCl3 single
crystals and determined energy levels of the various terms, producing what have come to be
called Dieke diagrams. The term symbols used in these diagrams are derived using
the Russell–Saunders coupling scheme (Section 7.3.2). It is now generally accepted that
the coupling is intermediate between the two limiting cases of Russell–Saunders and
j–j coupling, and high-level calculations are needed to determine the relative order of the
terms [11]. Nonetheless, the Russell–Saunders coupling scheme provides a reasonable
approximation of the multielectron energy levels. For example, the ground state of any
lanthanoid ion can be correctly predicted using Hund’s rules.
As an example, consider the excitation and emission spectra of Eu3+. The electronic ground-
state configuration of Eu3+ is [Xe]4f 6. Applying Hund’s rules (Section 7.3.2), we determine that
the ground state has S = 3 and L = 3, which combine into a 7F term. The possible J values for
the 7F term are the integers between L + S = 6 and |L − S| = 0, and Hund’s third rule tells us that
the energy increases as J increases, as shown in Figure 7.20. An f 6 ion has 3003 microstates that
can be grouped into 295 distinct 2S+1LJ terms, which makes determining the excited-state
energies challenging. Fortunately, the photoluminescence of Eu3+ activators can be understood
from a relatively small subset of the total number of excited states. The relevant transitions are
shown on the left-hand side of Figure 7.20. Optical excitation is largely through spin-forbidden
transitions from the 7F0 ground state into various low-lying quintuplet states (S = 2), as well as
LMCT transitions from the surrounding anions to empty Eu 5d orbitals, which are typically
excited by photons with wavelengths in the 200–300 nm range. The excited-state electrons
rapidly relax to the lowest-energy quintuplet state 5D0, then undergo phosphorescence to return
to one of the 7FJ states. Because the spin–orbit coupling is relatively strong, the energies of the
various 7FJ states are well resolved. The energies of the 5D0 → 7FJ transitions in Eu3+
phosphors are relatively insensitive to their local environment. However, you can see a series
of closely spaced sharp lines associated with each 5D0 → 7FJ transition caused by subtle crystal-
field splitting effects (Figure 7.20).
276 Optical Materials

5D
4
5L
7 5D
0 → 7F2 Y2O3:Eu3+
5L
6
5D
3

intensity (arb. units)

5D
2
5D
1
5D
0
468 nm

702 nm

→ 7F0
414 nm

653 nm
394 nm

0
614 nm

5D
382 nm

592 nm
363 nm

578 nm

5D → 7F1
0 5D 5D → 7F4
0 → 7F3 0

7F
6
7F
7F
5 575 600 625 650 675 700 725
4
7F
7F 3 wavelength (nm)
2
7F
7F 1
0

Figure 7.20 The term scheme and separation of energy levels for a free Eu3+ ion (left), and the emission
spectrum of Y2O3:Eu3+, a commercial red phosphor used in fluorescent lights (right). Data are taken from
ref [12].

The color of Eu3+ emission can vary from orange to red, depending upon the host. This
happens because the intensities of the different 5D0 → 7FJ transitions are highly sensitive to
the local symmetry of the Eu3+ ion. When Eu3+ is located on a site with inversion symmetry,
optical transitions other than those where ΔJ = 0, ±1 violate the parity selection rule17 and
are very weak. Only the 5D0 → 7F1 (with ΔJ = 1) at λ ≈ 592 nm does not violate the
parity selection rule, hence the emitted light takes on a reddish-orange color. In hosts
where the Eu3+ sits on a site without inversion symmetry, such as Y2O3:Eu3+, crystal-field
components mix states of opposite parity into the 4f n configurational levels and the 5D0 → 7F2
(~614 nm) emission gains significant intensity (see Figure 7.20).18 This results in a deeper-red
emission, which is generally a desirable attribute in commercial phosphors. This example
shows that even though the energies of individual f-to-f transitions are only weakly perturbed
by the environment of the activator ion, their relative intensities can be greatly influenced by
the local structure imposed by the host.

17
The J = 0 → J = 0 transition is also forbidden by the parity selection rule even though ΔJ = 0.
18
f-to-f transitions that don’t violate the parity selection rule, like 5D0 → 7F1, are referred to as magnetic-dipole
transitions. Those that do violate this rule, like 5D0 → 7F2, are called forced electric-dipole transitions. Y2O3 of
space group Ia 3, has two different cation sites: 8b of site symmetry −3 and 24d of site symmetry 2. Forced electric-
dipole transitions with significant intensity are only seen for the latter.
 7.8 Photoluminescence 277

Now to band emitters. The energies of the empty 5d orbitals are sensitive to changes in the
chemical environment of the activator ion, because the 5d orbitals form antibonding orbitals
with anions of the host. The presence of antibonding character in the excited state results in
an expansion of the metal–ligand bond lengths with respect to the ground state. The increase
in ΔQ leads to a variable Stokes shift, higher rates of thermal quenching, broadening of the
absorption and emission bands, and the ability to tune excitation and emission spectra
through the appropriate choice of host. The two most important lanthanoid band emitters
are Ce3+ and Eu2+, which feature [Xe]4f 1 ↔ [Xe]4f 05d1 and [Xe]4f 7 ↔ [Xe]4f 65d1 transi-
tions, respectively. These transitions are fully allowed and give rise to strong absorption and
emission bands, which can in turn lead to highly efficient phosphors. Here we concentrate on
Ce3+ because the 4f 1 electron configuration simplifies the analysis.
The 4f 1 ground-state term of Ce3+ is split by spin–orbit coupling into two levels, 2F5/2 and
2
F7/2, separated by 0.25 eV. The energy separation between the 4f 1 ground state and the
empty 5d orbitals is ~6.3 eV for a free Ce3+ ion. This energy separation can be significantly
reduced when the Ce3+ ion is embedded in phosphor hosts, through two effects—a centroid
shift and crystal/ligand-field splitting of the 5d orbitals—as illustrated schematically in
Figure 7.21. Given the importance of these two effects in designing new phosphors, we
examine each separately.
The centroid shift is the downward shift in the average energy of all five 5d orbitals relative
to a free Ce3+ ion. The decreased 5d-to-4f separation is attributed to a reduction in electron–
electron repulsions due to delocalization (spreading out) of the excited-state 5d Ce3+ orbital

free ion centroid crystal-field

7 shift splitting

5
Ce3+ 5d
energy (eV)

~6.3 eV
3

2
2.6 to 5.2 eV

1
2F 2F 2F

0 2F
7/2
2F
7/2
2F
7/2
Ce3+ 4f
5/2 5/2 5/2

Figure 7.21 Schematic energy-level diagram for Ce3+, showing the combined effects of the centroid shift
and crystal/ligand-field splitting in lowering the energy of 4f-to-5d transition.
278 Optical Materials

Table 7.10 The centroid shift for Ce3+ doped into various inorganic hosts grouped
according to anions that surround Ce3+ in the host. Data taken from ref. [13].

Centroid shift (eV)

Compound type Number Minimum Maximum Median

Fluorides 25 0.54 0.91 0.70

Chlorides 17 1.61 1.89 1.84
Bromides 9 1.92 2.21 2.11
Iodides 4 2.36 2.84 2.67
Oxides (polar covalent)* 75 0.88 1.99 1.30
Oxides (ionic)† 3 1.97 2.59 2.27
Sulfides 4 2.61 2.98 2.80
Selenides 2 3.00 3.21 3.11
*Oxides where the “anionic” portion of the host contains a p-block element that forms
polar-covalent bonds to oxygen, such as P, B, Si, or Al. †Oxide hosts with low-
electronegativity cations (CaO, La2O3, LaLuO3) comparable to Ce3+.

wavefunctions onto the ligand orbitals through its interaction with them.19 For Ce3+, the
magnitude of the centroid shift ranges from approximately 0.5 to 3.2 eV, depending on its
neighboring atoms. Dorenbos [13] has examined the centroid shift for Ce3+ doped into more
than 130 different inorganic hosts, and his results are summarized in Table 7.10. Various
factors come into play, including the anion polarizability, the coordination number, and the
bond distances. Of these, the anion polarizability (softness of its electron cloud) is one of the
most important factors. While there is a range of centroid shifts for hosts containing each
type of anion, Table 7.10 shows a clear increase in the centroid shift as anion polarizability
increases (hence electronegativity decreases): F− < Cl− < Br− < I− and O2− < S2− < Se2−.
Notice in Table 7.10 that ionic oxides differ from hosts where oxygen forms polar covalent
bonds to p-block elements, as found in phosphates, borates, silicates, and aluminates. Among
these, the centroid shift varies over a wide range. The spread occurs because the electron density
on oxygen is shifted toward the p-block element. This increasingly covalent bonding engagement
of oxygen reduces its polarizability towards the Ce3+. Accordingly, the centroid shift decreases as
the weighted average electronegativity of non-oxygen atoms in the undoped host, χav, increases:
X
ni zi χi
χav ¼ Xi (7.9)
nz
i i i

where zi and χi are the charge and electronegativity of the non-oxygen atoms i, and ni is their
stoichiometric quotient in the empirical formula of the host [13].

19
The origins of the centroid shift are similar to those of the nephelauxetic (cloud expanding) effect used to understand
the optical spectra of transition-metal complexes. For more details see C.K. Jørgensen, “Modern Aspects of Ligand
Field Theory” (1971) North-Holland.
 7.8 Photoluminescence 279

Consider the behavior of Ce3+ doped into YPO4 and YAlO3 hosts. In YPO4, the weighted
average Allred–Rochow electronegativity of the non-oxygen atoms is χav = (3χY + 5χP)/8 =
1.70, and the centroid shift is 1.19 eV. For YAlO3, χav = (3χY + 3χAl)/6 = 1.29 and the centroid
shift is 1.60 eV. The highly covalent P–O bonds of the phosphate groups in YPO4 reduce the
polarizability of the oxygen atoms towards Ce3+, leading to a smaller centroid shift.
Ligand-field splitting of the 5d orbitals is the second factor that affects the energies of the
4f-to-5d transitions. The magnitude of the ligand-field splitting depends primarily upon the
coordination geometry (Figure 7.2) and bond distances. Hosts with larger anions tend to have
longer bond distances and thus smaller ligand-field splitting.20 The effects of coordination are
illustrated by the differences between a Ce3+ ion in an octahedron and a cuboctahedron. Both
share the same symmetry (m3m), and the 5d orbitals split into the familiar t2g (dxy, dyz, dxz) and
eg (dz2, dx2−y2) sets in both environments. Within the assumptions of crystal-field theory (Section
7.3.1), the t2g–eg splitting in a cuboctahedron is only 50% of its value in an octahedron, Δco =
0.5Δo. A larger splitting increases the energy spread of the 5d orbitals (Figure 7.21), thereby
reducing the energy separation between the 4f orbitals and the lowest-energy 5d orbital(s).
Hence, all other things being equal, an octahedrally coordinated activator will absorb and emit
light at longer wavelengths than the same activator in a host where its environment is
a cuboctahedron. Empirically, it has been shown [14] that crystal-field splitting for eight-
coordinate cube and dodecahedron coordinations is ~80–90% that of an octahedron. For a
nine-coordinate tricapped trigonal prism it is much smaller, 40–50% of that seen in an octahe-
dron, similar to the cuboctahedron.

7.8.5 Non-Lanthanoid Activators

In this section, we survey three additional classes of activators: (a) ions of p-block elements
with (n − 1)d10ns2 configurations, (b) oxoanions of transition metals with a d 0 configuration,
and (c) transition-metal ions with partially filled 3d orbitals. We begin with the optical
transitions of (n − 1)d10ns2 activators like Sn2+, Sb3+, Pb2+, and Bi3+. The ns2 ground state
is represented by a singlet 1S0 term, and the ns1np1 excited state can be divided into a singlet 1
P1 state and three triplet states that are further split by spin–orbit coupling: 3P2, 3P1, and 3P0
(see Figure 7.22). Excitation can occur either through spin-allowed 1S0 → 1P1 transitions
followed by intersystem crossing into the 3P1 state, or directly from the ground state into the
lower-lying 3P1 state. The 1S0 → 3PJ transitions are spin-forbidden, but the selection rule is
relaxed by spin–orbit coupling. The spin–orbit coupling is high for ions from the fifth period
(e.g. Sn2+, Sb3+) and even larger for those from the sixth period (e.g. Pb2+, Bi3+). The 1S0 → 3
P1 transition is the strongest of the three possible 1S0 → 3PJ transitions, because the 1S0 → 3P0
and 1S0 → 3P2 transitions are forbidden by the parity selection rule. This can be seen in the

20
As the size of the anion increases, so does its polarizability. Hence, the anions that give the largest centroid shifts
tend to give smallest crystal-field splitting.
280 Optical Materials

1P
1
intersystem
crossing 1S 3
3P 0→ P1 3P 1
2 1→ S0

excitation

intensity (arb. units)

3P emission
1

3P
0

phosphorescence
1S 1P
0→ 1
absorption

240 270 300 330 360 390 420 450

1S
0
wavelength (nm)

Figure 7.22 The energy-level diagram for a (n−1)d10ns2 ion (left). The excitation and emission spectra of
Ba2Mg(BO3)2:Bi3+ (right). Data taken from ref. [15].

excitation spectrum of Ba2Mg(BO3)2:Bi3+ in Figure 7.22. Phosphorescence involving a 3P1 → 1S0

transition dominates the emission spectrum.21
Because the ns1np1 excited-state electron configuration is both more antibonding than the
ns2 ground-state configuration and prone to Jahn–Teller distortions (Section 5.3.10), ions like
Bi3+, Pb2+, and Sn2+ can experience a large reorganization of their coordination environment
in the excited state. When this occurs, a large ΔQ, large Stokes shift and broad emission bands
are expected. If we start from a symmetric environment like an octahedron, the coordination
environment of the ns1np1 excited state can undergo a symmetric expansion, a tetragonal
distortion (first-order Jahn–Teller distortion), and/or a trigonal distortion (second-order
Jahn–Teller distortion) [16]. The tetragonal distortion leads to ligand-field splitting of the
3
P1 excited state that can result in splitting of the excitation and emission bands.
The extent of the structural reorganization in the excited state, and hence the size of the
Stokes shift, is highly dependent on the structure of the host. When an activator ion is
placed on a site that is compressed with respect to its preferred environment, reorganiza-
tion of the excited state is suppressed, minimizing ΔQ and leading to a small Stokes shift.
Conversely, if it is placed on a large site, relaxation of the coordination sphere of the
activator ion in the excited state can be extensive, resulting in a large Stokes shift. Stokes
shifts for Bi3+ activators vary from 0.1 eV in Cs2NaYCl6:Bi3+, where Bi3+ substitutes for
the smaller octahedrally coordinated Y3+, to 2.5 eV in Bi2Ge3O9, where Bi3+ has a strong

21
The 3P0 → 1S0 tends to make little contribution to absorption or emission spectra. However, for 6s2 activators at
very low temperatures, the 3P0 → 1S0 emission can be observed in some cases.
 7.8 Photoluminescence 281

trigonal distortion in the ground state with three short Bi–O bonds (2.14 Å) on one side and
three long Bi–O bonds (2.74 Å) bonds on the other side of the octahedron. In the latter
case, Bi3+ is thought to adopt a much more symmetric environment in the excited state, and
the large ΔQ leads to a large Stokes shift [17].
Oxoanions of d 0 metals, such as WO42− and WO66−, are further examples of broad-band
emitters with large and highly tunable Stokes shifts. CaWO4 is a paradigmatic scintillator
material that was used for many decades as an X-ray phosphor. Its luminescence is due to
LMCT transitions on the WO42− groups, analogous to those discussed for CrO42− in Section
7.4.1. Because this excitation promotes electrons from nonbonding O 2p orbitals into
5d orbitals with significant antibonding character, tungstates exhibit large ΔQ and Stokes
shifts that range from 1.2 eV to 2.5 eV. Increasing the coordination number of the central
metal decreases the energy of the LMCT transition. This is illustrated by a comparison
between CaWO4, where the onset of absorption (the band gap) is ~4.8 eV, and Sr2MgWO6,
where the onset of absorption for the WO6/2 octahedron is ~3.6 eV. Oxoanions of other d 0
transition metals (e.g. VO43−, MoO42−) can also luminesce. The energies of the excitation
and emission bands follow the trends found in Table 7.7.
Transition-metal ions with partially filled d orbitals are the final class of activator
that we will consider. Perhaps the best-known example of transition-metal ion lumines-
cence is ruby, where Cr3+ is doped into Al2O3.22 Ruby possesses two sharp, closely
spaced emission lines near 700 nm. The transition responsible for Cr3+ emission is spin-
forbidden phosphorescence from a 2E excited state to the 4A2 ground state. Because
this transition is spin-forbidden, the correlation diagrams of Section 7.3.3 are inad-
equate. Instead, we must turn to the so-called Tanabe–Sugano diagrams used to
understand both spin-allowed and spin-forbidden transitions. The Tanabe–Sugano
diagram for an octahedrally coordinated d 3 ion is shown in Figure 7.23. Because
both the excited state (2E) and ground state (4A2) have a t2g3eg0 configuration, the
energy spacing between them has little dependence on the ligand-field splitting, leading
to sharp emission lines and small ΔQ. However, to get the sharp, red emission seen in
ruby, it’s critical that the ligand-field splitting Δo is large enough for the energy of the
2
E level to be lower than that of the 4T2 state whose energy separation from the ground
state increases linearly with Δo. To meet this criterion, hosts with large Δo are needed.

7.8.6 Energy Transfer

So far, we have considered two pathways for excited-state electrons to return to the ground
state: radiatively via emission of a photon or non-radiatively through coupling with lattice
vibrations. A third possibility is energy transfer, where the excited-state electron returns to
the ground state and transfers its energy to a nearby acceptor ion, where an electron is

22
Stimulated emission in ruby was the basis for the first laser, developed in 1960.
282 Optical Materials

4T
1
2A
2

2A
1

4T
1

energy 4T
2

2F
2T
2

2T
1

2G 2E
4P

4A
4F 2

crystal-field splitting, Δo

Figure 7.23 A simplified Tanabe–Sugano diagram for an octahedrally coordinated d 3 cation. The spin-
forbidden 2E → 4A2 transition gives rise to luminescence with activators such as Cr3+ and Mn4+. The
ground-state 4A2 energy is plotted as the x axis.

promoted into an excited state in order to maintain conservation of energy. We can describe
this process with the equation:

D þ A → D þ A (7.10)

where the ion that is originally in the excited state is called the donor (D), the ion that receives
the energy transfer is called the acceptor (A), and an asterisk is used to denote an excited
electronic state. Energy transfer underpins the action of sensitizers.
Energy transfer generally occurs by one of two different mechanisms. The first is Förster
resonant-energy transfer (FRET), which is based on electromagnetic multipole interactions,
predominantly dipole–dipole interactions. In FRET, the electromagnetic field associated
with D* interacts with A, leading to a transfer of energy but not electrons between the two
sites. The efficiency of FRET scales with d −6 for dipole–dipole interactions, where d is the
distance between donor and acceptor. The range over which FRET is operative generally
doesn’t exceed 2–5 nm.23 To be efficient, the optical transitions should be allowed electric-

23
The strong distance dependence of FRET led to the development of FRET spectroscopy, which allows conform-
ational changes of biomolecules to be followed.
 7.8 Photoluminescence 283

D* A D A*

Förster resonance energy transfer

D* A D A*
Dexter electron transfer

Figure 7.24 FRET and Dexter electron-transfer mechanisms.

dipole transitions, and the emission spectrum of D* should have significant overlap with the
absorption spectrum of A.24
An alternative is Dexter electron transfer where an excited-state electron is transferred
from donor to acceptor while a ground-state electron is transferred in the opposite direction,
as shown in the lower half of Figure 7.24. This mechanism does not require overlap between
emission and absorption spectra of donor and acceptor, nor does it depend on the selection
rules for either transition. It does, however, require a significant overlap of the molecular
orbitals on the two sites, which limits its operability to distances ≤ 1 nm.
The dominant type of energy transfer depends on the nature of the donor and acceptor
sites. FRET from a broad-band emitter to a line absorber (e.g. Ce3+ → Tb3+) is highly
inefficient and energy transfer between these species must rely upon the Dexter mechanism,
limiting transfer to near neighbors in the host. The opposite combination, energy transfer
from a line emitter to a band absorber, can occur with reasonable efficiency over longer
distances via FRET.

7.8.7 Sensitizers
Several otherwise useful activators do not effectively absorb light at practical wavelengths,
and therefore can only be used in combination with an appropriate sensitizer. This is
particularly true for activators that rely upon spin-forbidden transitions to absorb light,
like Tb3+ and Mn2+.
Tb3+ is an efficient emitter of green light, but it only absorbs strongly at excitation
wavelengths smaller than 230 nm, where the 4f 8 → 4f 75d 1 transition can be excited (the
24
Though FRET is a non-radiative process that occurs through electric fields, conceptually one can think of the donor
emitting a virtual photon that is instantly absorbed by the acceptor.
284 Optical Materials

lower-energy f-to-f transitions are spin-forbidden). In most fluorescent lamps, a low-pressure

Hg-plasma discharge of dominant 254 nm emission acts as the excitation source. These
photons are not sufficiently energetic to excite Tb3+ activators, but they can effectively excite
the 4f 1 → 5d1 transition of Ce3+ ions in an appropriate host, which can then efficiently
transfer their energy to Tb3+ activators. Nearly all commercial green-emitting fluorescent-
lamp phosphors (LaPO4:Ce3+,Tb3+; CeMgAl11O19:Tb3+; GdMgB5O10:Ce3+,Tb3+) rely on
absorption by Ce3+ sensitizers and emission from Tb3+ activators. The spin-forbidden nature
of the electronic transitions of Tb3+ means that effective energy transfer only occurs when the
sensitizer and activator sites are in close proximity, where the Dexter electron-transfer
mechanism is operative (< 1 nm).

7.8.8 Concentration Quenching and Cross Relaxation

Energy transfer can also occur between ions of the same type over longer distances, particu-
larly for line emitters whose absorption and emission lines have near-perfect spectral
overlap. Multiple transfers are common and can lead to energy migration over significant
distances25 in the host crystal, until an impurity or defect is encountered where non-radiative
return to the ground state can occur (so-called killer sites). This type of quenching is called
concentration quenching. It does not occur at low activator concentrations, where energy
transfer is inhibited by the large distances between luminescent centers, but can become
significant at higher concentrations. Phosphors that contain Eu3+, Tb3+, and Gd3+ acti-
vators often show maximum photoluminescence when these ions are present as substitu-
tional dopants in low concentrations. Levels of substitution beyond a few atomic percent
lead to a decrease in photoluminescence due to concentration quenching.
In phosphors where the activator has a large Stokes shift, there is minimal overlap between
the absorption and emission spectra, limiting energy transfer and the effects of concentration
quenching. This explains why self-activating phosphors like CaWO4 and Bi4Ge3O12, where
the activator ion is present as a stoichiometric component of the host, nearly always contain
activators that exhibit a large EStokes.
If the donor only transfers part of its energy, relaxing to a lower energy state but not all the
way to the ground state, the energy-transfer process is called cross-relaxation. This process
can quench certain emission lines while leaving others intact. In the case of Tb3+ pairs, the
energy difference between 5D3 and 5D4 excited states approximately matches the energy
difference between the 7F6 ground state and the higher-lying 7F0 state. At concentrations
above 5%, cross-relaxation quenches emission from the 5D3 level in favor of emission from
the 5D4 level (see Figure 7.25). The Tb3+ ion that is excited into the 7F0 excited state via cross-
relaxation can return to the 7F6 ground state non-radiatively through internal conversion.
Cross-relaxation is why blue emissions that originate from 5D3 → 7FJ transitions are
suppressed and green emissions from 5D4 → 7FJ transitions are enhanced in phosphors

25
In some instances, the number of energy transfers can exceed 10000 before decay.
 7.8 Photoluminescence 285

ion 1 ion 2 ion 1 ion 2

5D 5D
3 3

5D 5D
4 4

green
luminescence

7F 7F
0 0
non-radiative
decay
7F 7F
6 6

Tb3+(5D3) + Tb3+(7F6) Tb3+(5D4) + Tb3+(7F0)

Figure 7.25 Cross-relaxation of Tb3+ through an energy transfer from ion 1, the donor, to ion 2, the
acceptor. Following the energy transfer, ion 1 emits a green photon while ion 2 undergoes non-radiative
relaxation to the ground state.

containing higher Tb3+ concentrations. Cross-relaxation is also responsible for quenching

blue emissions of Eu3+ and the visible emissions of Sm3+ and Dy3+ at concentrations as low
as ~1%.

7.8.9 Up-Conversion Photoluminescence

Up-conversion photoluminescence occurs when the energies of absorbed photons are lower
than those of the subsequently emitted photons.26 The most common up-conversion phos-
phors convert near-IR radiation to visible light. A typical up-conversion process is repre-
sented schematically in Figure 7.26 for Y2O3:Er3+,Yb3+. The process begins with the
absorption of a near-IR photon (λ = 980 nm) at Yb3+, via the spin- and parity-allowed
2
F7/2 → 2F5/2 transition. The Yb3+ then transfers its energy to Er3+ triggering a 4I15/2 → 4I11/2
transition. Absorption of a second near-IR photon at Er3+ further promotes the electron into
the 4F7/2 level, from which non-radiative relaxation to the 2H11/2, 4S3/2, or 4F9/2 levels can
occur. The final step is radiative return of the Er3+ ion to the ground state, leading to
emission of 662 nm (4F9/2 → 4I15/2) red light and/or green light with wavelengths of 525 nm
(2H11/2 →4I15/2) and 550 nm (4S3/2 → 4I15/2).
Yb3+ is the most widely used sensitizer in up-conversion phosphors because the 2F7/2 →
2
F5/2 transition has a high absorption cross-section, and there are no accessible higher-energy
excited states that permit up-conversion at the Yb3+ site. Blue emission can be obtained if
Yb3+ is paired with a Tm3+ activator instead of Er3+. The best host materials have minimal

26
Because Eexc < Eem up-conversion photoluminescence can be described as an anti-Stokes process.
286 Optical Materials

4F 4F
7/2 7/2
2H 2H
11/2 11/2
4S 4S
3/2 3/2
4F 4F
9/2 9/2

4I 4I
9/2 9/2
2F 4I 2F 4I
5/2 11/2 5/2 11/2

980 nm
4I 4I
13/2 13/2

2F 4I 2F 4I
7/2 15/2 7/2 15/2

Yb3+ Er3+ Yb3+ Er3+

(1) Yb3+ absorbs IR photon (2) Yb3+ to Er3+ energy transfer

4F 4F 4F
7/2 7/2 7/2
2H 2H 2H
11/2 11/2 11/2
4S 4S 4S
3/2 3/2 3/2
4F 4F 4F
9/2 9/2 9/2

4I 4I 4I
9/2 9/2 9/2
4I 4I 4I
11/2 11/2 11/2

4I 4I 4I
13/2 13/2 13/2

622 nm
525 nm
550 nm
4I 4I 4I
15/2 15/2 15/2

Er3+ Er3+ Er3+

(3) Er3+ absorbs (4) non -radiave (5) visible
second IR photon relaxaon luminescence

Figure 7.26 The five-step process of up-conversion photoluminescence in Y2O3:Yb3+,Er3+.

electron–phonon coupling (Section 7.8.2) to reduce thermal relaxation and achieve long-
lived excited states. Only then can the excited state persist long enough to allow a second
photon to be absorbed before relaxing to the ground state. For example, NaYF4 is a good
up-conversion host, in part because it possesses phonon modes whose energies are much
smaller than the energy separation of the 2F7/2 and 2F5/2 states on Yb3+.27 In hosts with
higher-energy phonons, non-radiative decay of the Yb3+ 2F5/2 excited state competes with
absorption of a second photon.
Because up-conversion materials rely on a multiple-photon absorption process rarely
found in nature, they are attractive as security markers to protect financial and government
documents. Up-conversion photoluminescence is also being explored for applications in

27
The dominant phonon mode in NaYF4 has an energy of 0.044 eV (~350 cm−1) which is ~29 times smaller than the
1.26 eV separation of the 2F7/2 and 2F5/2 states.
 7.9 Electroluminescence 287

lasers, next-generation lighting, near-IR photon detectors, nanometer-sized biological

labels, and night-vision goggles.

7.9 Electroluminescence
Electroluminescence is the direct conversion of electrical energy into optical energy. The basis of
electroluminescence is the radiative recombination of electrons and holes, driven by an electric
field. Although electroluminescence can take various forms, here we limit our discussion to
inorganic light-emitting diodes (LEDs) where electron–hole recombination occurs at the inter-
face between p- and n-type semiconductors, and organic light-emitting diodes (OLEDs) where
recombination is driven by injecting current into films of organic semiconductors.

7.9.1 Inorganic Light-Emitting Diodes (LEDs)

The electrical properties of semiconductors are highly sensitive to the presence of impurities,
particularly aliovalent substitutional impurities (Section 2.4). This topic is covered in detail in
Chapter 10, but to understand the operation of LEDs a few basic concepts are touched upon
here. If the substitutional impurity has more valence electrons than the atom for which it
substitutes, the “extra” electrons are donated to the conduction band, and the semiconductor
is said to be doped n-type. If the substitutional impurity has fewer electrons than the atom it
replaces, it accepts electrons from the valence band, and the semiconductor is said to be doped
p-type. The missing electron in the valence band carries a positive charge and is called a hole.
If p- and n-type semiconductors are joined together, the region where they meet is called
a p–n junction. We’ll discuss the fabrication, physics, and operation of p–n junctions in
Section 10.3.5. For now, we only need to know that when an appropriate voltage is applied to
a p–n junction, electrons and holes are driven to the interface and can recombine radiatively
to generate light. The host semiconductor is typically the same on either side of the junction,
and the energy of the emitted photons is determined by the band gap of the semiconductor.
To favor radiative recombination, direct band-gap semiconductors are preferred for use in
LEDs, though lower-efficiency LEDs can be made from indirect band-gap materials.
The first practical LEDs were made in the early 1960s from GaAs, which possesses
a direct band gap of 1.43 eV, and therefore emits in the near-IR. LEDs made from pure
GaAs are still used today as IR sources in fiber-optic communications. Another
semiconductor used in LEDs is GaP, which has an indirect band gap of 2.26 eV (see
Table 7.8) and emits green light. By forming GaAs1−xPx solid solutions, it is possible to
make LEDs that emit photons with energies intermediate between the two end members.
Compositions close to GaAs0.6P0.4 are used in red LEDs, while orange and yellow LEDs
can be made from compositions with higher GaP content. Unfortunately, the band gap
becomes indirect and the efficiency goes down sharply when x > 0.45. The emission of
GaAs LEDs can also be shifted to shorter wavelengths by forming solid solutions with
288 Optical Materials

AlAs, a semiconductor with almost the same lattice parameter but an indirect band gap
of 2.16 eV. The crossover from direct to indirect band gap occurs in the Ga1−xAlxAs
system for compositions with x > 0.4. By alloying InP (Eg = 1.35 eV, direct), AlP (Eg =
2.43 eV, indirect), and GaP it is possible to make (Ga1−xAlx)1−yInyP LEDs that emit
colors from green to the near-IR. In part because of the absence of arsenic, these have
become the preferred semiconductors for yellow, orange, and red LEDs.
Fabricating blue, violet, and UV LEDs challenged researchers for many decades. Solid
solutions between ZnS (Eg = 3.6 eV) and ZnSe (Eg = 2.58 eV) were investigated, but the
high concentration of defects and difficulties in obtaining high-quality p–n junctions
limited progress. In the late 1980s, Cree introduced a commercial blue LED based on
silicon carbide, but the device efficiency was so low that it never gained popularity. In the
1990s, Ga1−xInxN emerged as the material of choice for blue LEDs. Its wavelength can be
tuned from 370 nm (pure GaN) to 470 nm by increasing the indium content. Longer-
wavelength emission can be realized, but the efficiency drops as the indium content
increases. This is due to compositional segregation upon cooling, caused by the limited
solubility of InN in GaN. In Section 7.10.2, we will see that blue LEDs play a key role in modern
solid state lighting. While green LEDs can be made in either the Ga1−xInxN or (Ga1−xAlx)1−yInyP
systems, a high-efficiency green LED remains elusive. This challenge is sometimes referred to as
the “green gap”.

Box 7.2 Synthetic Methods: Synthesis and p-doping of GaN

GaN has the hexagonal wurtzite structure and a direct band gap of 3.4 eV. The first demon-
stration of a GaN LED was in 1972, but the device had an efficiency too low for practical
applications. Two major hurdles prevented further progress. Firstly, the lack of a good lattice-
matched substrate led to high defect densities in the films from which LEDs are made.
Secondly, while n-type samples can be made by replacing some gallium with silicon, it proved
difficult to reproducibly prepare p-type GaN.
In the mid 1980s, Akasaki and Amano used metalorganic vapor-phase epitaxy (MOVPE) to
grow GaN films on sapphire substrates with lower defect concentrations than achieved previ-
ously [18]. Earlier attempts to use sapphire substrates had not produced high-quality films due
to the 16% lattice mismatch with GaN. Their breakthrough was the deposition of a 30 nm
buffer layer of polycrystalline AlN onto the sapphire substrate at 500 °C, followed by an
annealing step at 1000 °C. This approach promotes the growth of small crystallites with
preferred orientation upon which the GaN can subsequently nucleate and grow. While the
GaN in close proximity to the AlN layer has a high concentration of dislocations, after a few
microns the defect concentration is low enough for use in LED applications. Later, Nakamura
[19] simplified the process by covering the sapphire at 600 °C with a 20 nm buffer layer of nearly
amorphous GaN, before subsequently depositing a highly crystalline GaN film. Without this
intermediate buffer layer, hexagonal columns of GaN grow that produce a rough surface and
result in poor electrical properties.
 7.9 Electroluminescence 289

Box 7.2 (cont.)

In the late 1980s, Akasaki and Amano observed that Zn-doped GaN emitted more blue light
when the device was placed inside a scanning tunneling microscope. Subsequently, it was
shown that the p-type behavior of Mg-doped GaN was significantly enhanced when the device
was irradiated with low-energy electrons. Nakamura and coworkers [20] showed that these
effects were caused by formation of hydride complexes (originating from trimethyl gallium and
ammonia used in the epitaxial growth), which passivated the acceptor sites and limited the
formation of holes. By irradiating with an electron beam, the unwanted hydrogen is expelled
from the sample, activating the acceptors and improving device performance. Annealing at
temperatures above 700 °C has a similar effect.
Building on these advances, Nakamura and co-workers produced a blue LED with
a quantum efficiency of 2.7% from a Ga1−xInxN/Ga1−xAlxN heterostructure in 1994. This
demonstration revolutionized the compact-disc industry and triggered a massive surge in
research and development activity. Quantum efficiencies have steadily increased over the
intervening years and now exceed 80% in state-of-the-art GaN-based LEDs. In 2014,
Amano, Akasaki, and Nakamura shared the Nobel Prize in Physics for their work.

7.9.2 Organic Light-Emitting Diodes (OLEDs)

OLEDs convert electrical energy to light through electron–hole recombination. While the
overall process has many similarities with the inorganic LEDs just discussed, OLEDs offer
distinct advantages. They can be very thin, and the methods of deposition (spin coating, vacuum
deposition) are simpler, cheaper, and less energy-intensive than the methods used to deposit
films of inorganic semiconductors. They can be made in almost any shape and deposited on
flexible materials. Initially, OLED device performance was limited by the poor electrical
conductivity of organic materials. However, the emergence of highly conductive polymers
such as poly(N-vinylcarbazole) and poly(p-phenylene vinylene) reignited activities in this field.
OLEDs now find widespread use in mobile phones, digital cameras, and flat-panel displays,
where they compete with liquid-crystal displays (LCDs). Compared with LCDs, OLEDs are
thinner, lighter, brighter, produce truer colors, refresh much faster, and consume less power.
A typical OLED is made up of several semiconducting organic materials, sandwiched
between two electrodes, one of which must be transparent. A schematic of a relatively simple
OLED made of three organic layers is shown in Figure 7.27. The organic semiconducting
materials are invariably π-conjugated systems, either small organic molecules or conducting
polymers (Section 10.5). When a voltage is applied, electrons are removed at the anode, which
is equivalent to injecting holes into the HOMO of the hole-transport layer. At the same time,
electrons are injected from the cathode into the LUMO of the electron-transport layer. To
facilitate charge injection, the Fermi level of the anode should have a reasonably good
290 Optical Materials

transparent hole- electron-

glass emissive cathode
anode transport transport
substrate layer (Ca) seal
(ITO) layer layer

LUMO e− e− e− e−
e−
photon LUMO
EF
out
exciton
HOMO
h+
EF
h+ h+ h+ HOMO
h+

Figure 7.27 A schematic of an OLED with three organic layers: a hole-transport layer, electron-transport
layer, and emissive layer.

energetic alignment with the HOMO of the hole-transport layer. This can be achieved by using
a transparent conductor, such as In2−xSnxO3 (ITO), deposited onto a glass substrate that
offers mechanical support and protects the active layers of the OLED from the environment.
Similarly, the Fermi level of the cathode should be sufficiently high in energy that it lies close to
the LUMO of the electron-transport layer. This necessitates using active metals like Ca, Ba, or
alloys like Mg1−xAgx, which must again be encapsulated due to their moisture sensitivity.
After charge injection, holes and electrons move in opposite directions under the external
electrical field, hopping from molecule to molecule. Holes from the hole-transport layer
migrate to the emissive layer where they encounter electrons that have migrated from the
electron-transport layer. When they meet, the electrostatic attraction between the two
oppositely charged particles leads to the formation of a bound electron–hole pair called an
exciton. The exciton is a neutral quasi-particle that can radiatively decay through electron–
hole recombination. The color of the emitted photon is determined by the HOMO–LUMO
gap of the emissive layer minus the exciton binding energy, the attractive potential energy
that holds the electron and hole together. Exciton binding energies in organic semiconduct-
ors are typically on the order of 0.3–0.5 eV.28
When an electron and hole meet, they may possess the same or the opposite spin, leading
to the formation of both singlet (S = 0) and triplet (S = 1) excitons. Radiative decay from the
triplet state (i.e. phosphorescence) is spin-forbidden, so triplet excitons decay predominantly
through non-radiative pathways. Quantum-mechanical momentum conservation tells us
that only 25% of all excitons are singlets. This means only one in four excitons decays
radiatively, which limits OLED efficiencies. To circumvent this limitation, neutral organo-
metallic complexes containing heavy metals like Ir or Pt can be incorporated into the

28
Excitons can also form in inorganic materials, but the binding energies are an order of magnitude smaller than in
organic materials.
 7.10 Materials for Lighting 291

emissive layer [21]. The presence of a heavy metal leads to strong spin–orbit coupling, which
facilitates intersystem crossing and radiative decay of triplet excitons. The triplet state of the
organometallic molecule is chosen to lie at a lower energy than that of the semiconducting
emissive layer, so that triplet excitons migrate to the organometallic molecules. The incorp-
oration of molecules containing heavy metals dramatically increases the brightness of
OLEDs, facilitating their commercialization. In addition to boosting efficiency, the wave-
length of photons emitted via triplet-exciton phosphorescence will in general differ from the
photon emitted by singlet-exciton fluorescence, because phosphorescence is governed by the
HOMO–LUMO gap of the organometallic complex, while fluorescence is predominantly
governed by HOMO–LUMO gap of the host organic layer. The flexibility to engineer
materials that emit at multiple wavelengths can be useful for applications.

7.10 Materials for Lighting

It’s hard to conceive modern life without abundant, inexpensive electric lighting, but until
the late nineteenth century cities and homes were still largely illuminated by flame. In the
twentieth century, incandescent lighting became ubiquitous, but this revolutionary technol-
ogy has since largely been replaced by more energy efficient methods of producing white
light. Fluorescent lights are still widely used, but they are increasingly being displaced by
high-efficiency blue LEDs coupled with down-conversion phosphors to produce white light.
Before discussing the phosphors used in both fluorescent and solid state LED lighting, we
must understand the metrics used to evaluate intensity and color. The amount of visible light
emitted by a light source is called the luminous flux. It is measured in lumens (lm) and is
defined as the product of 1 candela (cd) times the solid angle in steradians (sr). A candela is
roughly equivalent to the light given off by a single candle. More precisely, the candela is the
luminous intensity per unit solid angle weighted by a luminosity function that models the
sensitivity of the human eye.29 Luminous efficacy is the ratio of luminous intensity out to
electrical power consumed and is measured in units of lumens per watt (W). A typical 100 W
incandescent light bulb gives off 15–17 lm/W. By comparison, fluorescent lights produce
50–100 lm/W, and phosphor-converted LEDs can achieve a luminous efficacy of 200 lm/W.
Although hot objects like the filament of an incandescent bulb give off a broad spectrum of
light that spans the visible range, it is possible to mimic white light by mixing discrete colors.
Mixing blue and yellow light, leads to a “cold” white light, while mixing red, blue, and green
can produce a more natural white light, as described in Section 7.1. The color of a white light
source is an important parameter and there are various metrics for quantifying color, two of
which are touched upon below.
Color temperature is defined by comparing the output of a light source with the light
emitted by a black-body radiator, which changes as the temperature increases in the

29
The human eye is most sensitive to green light with a wavelength of 555 nm. A monochromatic source that emits
555 nm light with a radiant intensity of 1/683 W has a luminous intensity of 1 cd.
292 Optical Materials

sequence: red, orange, yellowish-white, white, and ultimately bluish-white. Light sources
with color temperatures >5000 K give off bluish-white light and are described as cool,
whereas those with color temperatures ranging from 2500 to 3500 K are described as
warm.30 The color-rendering index (CRI) is an alternative metric. The CRI is a measure of
the ability of a light source to reproduce the colors of various objects faithfully in comparison
with an ideal natural light source. The CRI is calculated by comparing the reflection spectra
of test colors to the spectra obtained when the same colors are irradiated with a source that
simulates sunlight. The CRI for a true black-body radiator, like an incandescent lamp,
would be 100. At the other extreme, a white object irradiated with a monochromatic light
source, such as a laser, can only reflect a single color (the color of the source) and has a CRI
of 0. In general, a CRI in the 70s would be considered acceptable for interior lighting
applications; a score in the 80s, good; and a CRI in the 90s, excellent.

7.10.1 Fluorescent-Lamp Phosphors

Fluorescent lights rely upon phosphors to convert the UV light from a mercury discharge lamp
to white light. The mercury atoms in the discharge emit about 85% of their radiation at 254 nm,
and 12% at 185 nm, so fluorescent-lamp phosphors should absorb efficiently at these wave-
lengths. Discovery of the halo-apatite phosphor Ca5(PO4)3(F,Cl):Sb3+,Mn2+ in 1949 was
a major turning point for the lighting industry. The Sb3+ ions have strong absorption peaks
at 255 nm (1S0 → 3P1) and 205 nm (1S0 → 1P1). The Sb3+ ions emit a broad band centered in the
blue region of the spectrum near 480 nm via a 3P1 → 1S0 transition (Section 7.8.5). They also
function as a sensitizer for the Mn2+ ions, which are not able to efficiently absorb light from the
plasma discharge because all electronic transitions in this high-spin d 5 ion are spin-forbidden.
Following energy transfer from Sb3+, the Mn2+ activators emit a broad band of orange light
near 580 nm via a 4G5/2 → 6S5/2 transition. The blue emission originating from Sb3+ and the
orange emission of Mn2+ combine to create a “whitish” light. Increasing the Mn2+ concentra-
tion suppresses the blue emission and enhances the orange emission. Increasing the chloride
content shifts the Mn2+ emission band to shorter wavelengths. In this way, color temperatures
ranging from 2700 to 6500 K can be obtained. Unfortunately, there is a trade-off between
optimizing the luminous efficacy and the CRI. If the brightness is high (efficacy ~80 lm/W) the
CRI is on the order of 60. It is possible to increase the CRI to 90 through appropriate
compositional tuning, but luminous efficacy drops to ~50 lm/W [16].
By combining phosphors that emit in narrow wavelength intervals in the red, green, and
blue regions of the spectrum it is possible to achieve high luminous efficacy (~100 lm/W) and
good color rendering (CRI ≈ 80). This type of fluorescent light, known as a tricolor lamp,
might include the following phosphors: BaMgAl10O17:Eu2+ that emits in the blue, near
450 nm; (Ce1−xGdx)MgB5O10:Tb3+ or LaPO4:Ce3+,Tb3+ that emit in the green, near
540 nm; and Y2O3:Eu3+ that emits in the red, near 610 nm. All three phosphors have

30
It’s somewhat paradoxical that “cool” light has a higher color temperature than “warm” light.
 7.10 Materials for Lighting 293

individual quantum efficiencies near 90%. To achieve even better color rendering, needed in
museums and store displays, alternative phosphors are chosen that shift the emission
maximum of the blue phosphor to longer wavelengths.

7.10.2 Phosphor-Converted LEDs for White Light

Phosphor-converted LEDs are being pursued as the light source of the future, due to their
combination of high luminous efficacy and long lifetime. In the most common configuration,
a Ga1−xAlxN LED acts as a source of blue light and is combined with phosphors that absorb
a fraction of that blue light, then emit the longer-wavelength photons needed for white light.
Phosphors for this application should: (a) possess high quantum efficiencies on the order of
90–95%, (b) have excitation maxima that are well matched to the light provided by the LED
(emission maxima typically fall between 370 and 470 nm), (c) emit light at wavelengths that
provide an optimal CRI, and (d) show minimal loss of efficiency at operating temperatures.
To be commercially viable, they should also be non-toxic, inexpensive, and possess excellent
thermal, chemical, and photochemical stabilities. Both the temperature and photochemical
stability requirements for an LED phosphor are more stringent than those for a fluorescent-
lamp phosphor, because the excitation densities of LED-based lighting systems (~30 W/cm2)
are about three times higher than those in fluorescent lights.
The simplest white LEDs combine blue light with a wavelength of ~450 nm from
a Ga1−xInxN LED and yellow light from a phosphor where Ce3+ is doped into an appropri-
ate host, usually Y3Al5O12 (YAG). The Y3Al5O12:Ce3+ phosphor has broad absorption and
emission bands that peak near 460 nm and 560 nm, respectively (Figure 7.28). By turning to
hosts that are complex solid solutions, such as (Y1−xGdx)(Al1−yGay)5O12:Ce3+, the emission
maximum can be tuned between 510 nm and 580 nm, thereby adjusting color temperature
between 3000 K and 8000 K. The CRI of this type of phosphor-converted LED ranges from
just below 75 to slightly above 80.

blue Y3Al5O12:Ce3+
Y3Al5O12:Ce3+
light yellow emission
intensity (arb. units)

excitation
light
blue LED
emission

Y3Al5O12:Ce3+
glass composite

blue
Ga1−xInxN
LED electrodes
300 400 500 600 700
wavelength, λ (nm)

Figure 7.28 A schematic of a phosphor-converted white LED (left). The excitation and emission spectra
of a Y3Al5O12:Ce3+ phosphor superimposed on the emission from a blue Ga1−xInxN LED (right).
294 Optical Materials

Trichromatic LEDs provide superior CRIs (>90) by combining green and red phosphors
with the blue light emitted by the Ga1−xInxN LED. The phosphors developed for fluorescent
lamps are unsuitable for white LEDs because their excitation bands are not well matched to
the output of the blue LED. Instead, researchers have largely concentrated on phosphors
containing the band emitter Eu2+, particularly for the red phosphor. The [Xe]4f 65d 1 → [Xe]
4f 7 transition of Eu2+ usually emits blue or green light when Eu2+ is incorporated into an
oxide, but in sulfide, nitride, and oxynitride hosts the Eu2+ experiences a larger centroid shift,
sinking the barycenter of the d-orbital set, and shifting the absorption and emission maxima
to longer wavelengths (Figure 7.21).
Binary sulfides with the rock-salt structure like SrS:Eu2+ and CaS:Eu2+ emit in the red
with maxima of 610 nm and 660 nm, respectively. The emission spectrum can be tuned by
forming solid solutions or turning to more complex sulfides. Unfortunately, sulfides are
prone to corrosion in the presence of minute traces of water, necessitating encapsulation.
They also tend to suffer from strong thermal-quenching effects, as illustrated by the green
phosphor SrGa2S4:Eu2+ where the quantum efficiency drops from 75–80% at room tempera-
ture to 50% at 170 °C [22]. Nitridosilicates, like Sr2Si5N8:Eu2+ (emission maximum 630 nm)
and nitridoaluminates, like SrLiAl3N4:Eu2+ (emission maximum 650 nm), offer much better
stability and excellent photoluminescent properties. Further improvements in luminous
efficacy can be made by reducing the spectral overlap between the excitation band of the
red phosphor and the emission band of the green/yellow phosphor, as well as limiting
spillover of the red emission into the near-IR where the human eye cannot detect it [23, 24].

7.11 Problems
7.1 What is the wavelength (in nm), frequency (in s−1), and color of a photon with an
energy of 3.60×10−19 J? What is its energy in eV?
7.2 What color would a material be if it absorbed (a) red light, (b) blue and green light, (c)
yellow light, (d) green light, (e) green and yellow light?
7.3 Classify each of the following pigments or gemstones as idiochromatic or allochromatic:
(a) cobalt blue, CoAl2O4, (b) emerald, Be3Al2(SiO3)6:Cr3+, (c) yellow ochre, FeOOH.
7.4 Use crystal-field theory to qualitatively predict the energy splitting of the d orbitals for
a transition-metal ion linearly coordinated along the z axis.
7.5 Predict which of the following pairs will have larger ligand-field splitting (Δoct): (a)
[Fe(H2O)6]2+ or [Co(H2O)6]3+, (b) [Co(H2O)6]3+ or [Rh(H2O)6]3+.
7.6 A carbon atom has a 1s22s22p2 electron configuration. The various ways of filling the
2p orbitals lead to 15 microstates that can be grouped into three terms: 1D, 3P, and
1
S. (a) Determine the value of L and S for each term and the number of microstates
associated with each term. (b) What are the allowed values of the total angular
momentum quantum number J for each term? (c) What is the energy order predicted
by Hund’s rules for the various (2S+1)LJ terms?
 7.11 Problems 295

7.7 The UV–visible spectrum of the octahedral complex [Ni(NH2CH2CH2NH2)3]2+ is

shown below. (a) Given that ethylenediamine ligand, NH2CH2CH2NH2, is a neutral
bidentate ligand (both nitrogen atoms coordinate to the metal), determine the electron
configuration of the nickel ion. (b) Use a correlation diagram to assign each of the three
d-to-d transitions labeled as 1, 2, and 3 below. (c) What color would you predict for
a solution of this complex ion from the spectrum below? (d) Compare these transitions
to those observed for [Ni(H2O)6]2+ (Table 7.5) and determine which complex ion has
a larger Δoct.

3
8
molar extinction coefficient, ε

2
6
1

0
200 300 400 500 600 700 800 900 1000 1100 1200

wavelength (nm)

7.8 Which of the two correlation diagrams shown below would be appropriate for tetrahe-
dral [CoCl4]2−? For the applicable correlation diagram, what are the correct term
symbols for the lines labeled W, X, Y, and Z?

(a) Z (b) Z

Y Y
energy
energy

X X

W
W

Δtet Δtet

7.9 For which of the following ions is the d-to-d transition spin-forbidden, Laporte-
forbidden, both or neither: (a) tetrahedral Co2+, (b) square-planar Cu2+, (c) octahedral
Mn2+ (high spin), (d) trigonal-bipyramidal Mn3+?
296 Optical Materials

7.10 Tetrathiomolybdate MoS42− is tetrahedral and features a strong t1 → e LMCT

absorption at 470 nm. (a) What color would you predict for (NH 4)2MoS4? (b)
What is the energy in eV of the LMCT transition and how does it compare with
MoO42−? Why does the LMCT energy shift in the way that it does? (c) The
compound Li3NbS4∙2TMEDA (TMEDA = tetramethylenediamine) features iso-
lated NbS43− tetrahedra. Do you expect the lowest-energy LMCT absorption peak
of NbS43− to occur at shorter or longer wavelengths than MoS42−? Explain your
reasoning.
7.11 Which semiconductors in Table 7.8 will have colors other than white or black? For each
compound in your list predict the approximate color.
7.12 The red pigment vermillion (HgS) is a semiconductor with a band gap of 2.0 eV. It was
replaced in the late nineteenth and early twentieth century by CdS1−xSex pigments that
are less toxic and more stable. Assuming the band gap follows a linear Vegard’s law-
type relationship, what composition of CdS1−xSex will have a band gap of 2.00 eV,
roughly the equivalent of HgS? See Table 7.8 for the band gaps of CdS and CdSe.
7.13 How do fluorescence and phosphorescence differ from each other in theory and
experiment?
7.14 The Pr3+ ion has 91 microstates that can be grouped into six terms before spin–orbit
coupling is included: 1D, 3F, 1G, 3H, 3P, 1S. (a) What is the electron configuration of Pr3+?
(b) What are the values of L and S for each of the six terms? (b) Use Hund’s first two rules
to arrange these six terms in order of increasing energy. (c) Determine the allowed values
of the total angular-momentum quantum number J for each of the six terms, and give the
lowest-energy (2S+1)LJ term for each.
7.15 For each of the following phosphors, identify the type of electronic transition respon-
sible for luminescence and predict whether it will show weak, moderate, or strong
electron–phonon coupling: (a) YVO4, (b) Y2O3:Eu3+, (c) SrGa2S4:Eu2+, (d) Y3Al5O12:
Nd3+, (e) Bi4Ge3O12.
7.16 As the difference ΔQ between the equilibrium bond distances of the ground and excited
states increases, would you expect the following phosphor characteristics to increase,
decrease, or be unaffected: (a) Stokes shift, (b) energy transfer between luminescent
centers of the same type, (c) thermal-quenching temperature, (d) Huang–Rhys param-
eter, (e) width of the emission line(s)?
7.17 The luminescence of AWO4 scheelites (A = Ba, Sr, Ca) shows large variations in thermal-
quenching behavior. The photoluminescence of CaWO4 only drops by a few percent
when heated from 90 K to 270 K, while that of SrWO4 decreases by ~90% over the same
temperature interval. The luminescence of BaWO4 is completely quenched by 90 K. How
do you rationalize the observed increase in thermal quenching as the size of the alkaline-
earth ion increases?
7.18 Orange, red, and near-IR LEDs are often made from GaAs1−xPx solid solutions. For
compositions with x < 0.45, a direct band gap is observed, while for larger x the band
gap is indirect. Assuming a linear Vegard’s law relationship between the end members
 7.11 Problems 297

GaAs (Eg = 1.43 eV) and GaP (Eg = 2.25 eV), determine the shortest wavelength
emitted from a composition with a direct band gap. What color of light would such an
LED emit?
7.19 The phosphor BaMgAl10O17:Eu2+,Mn2+, whose excitation and emission spectra are
shown below, is of interest as a combined blue and green phosphor in plasma-display
panels. The Eu2+ ions absorb at 336 nm and emit in the blue with a maximum of
450 nm. There is also energy transfer from Eu2+ to Mn2+ that leads to the green
emission at 512 nm. (a) What are the electronic transitions responsible for emission
on Eu2+ and Mn2+? (b) What is the Stokes shift for Eu2+? (c) What is the most likely
mechanism of energy transfer from Eu2+ to Mn2+, Förster (FRET) or Dexter?

Eu2+ Eu2+ emission

excitation
intensity (arb. units)

Mn2+

200 300 400 500 600

wavelength, λ (nm)

7.20 CeMgAl11O19:Tb3+ is a green phosphor used in tricolor fluorescent lights. Ce3+ ions
in the host absorb photons from the Hg-plasma discharge with an excitation max-
imum near 270 nm and emit via a [Xe]4f 05d1 → [Xe]4f 1 transition with a Stokes shift
of 0.9 eV. The Tb3+ dopant emits via f-to-f transitions from a 5D4 excited state to
various 7FJ levels (see Figure 7.25), with a maximum near 540 nm. (a) At what
wavelength will the emission maximum of the Ce3+ ion fall? (b) Which lanthanoid
ion is responsible for the green light? Why is the other lanthanoid ion needed? (c)
What mechanism is responsible for energy transfer between Ce3+ and Tb3+?
7.21 When Ce3+ ions substitute for La3+ in a LaCl3 host, the lowest-energy peak in the
photoluminescence excitation spectrum falls at 281 nm, whereas a similar substi-
tution in the double-perovskite host Cs2NaLaCl6 shifts the lowest-energy peak in
298 Optical Materials

the excitation spectrum to 342 nm. The La3+ ion in LaCl3 sits in a nine-coordinate
tricapped trigonal-prismatic site, while in Cs2NaLaCl6 the La3+ ion is octahedrally
coordinated. (a) Would you predict a larger, smaller, or similar centroid shift in
Cs2NaLaCl6? (b) How will differences in the coordination environment shift the
lowest-energy Ce3+ excitation peak in Cs2NaLaCl6? (c) Is the observed shift in the
Ce3+ excitation spectrum due to the centroid shift, the change in coordination, or
both?
7.22 In thiogallate phosphor, SrGa2S4:Eu2+, Eu2+ substitutes for Sr2+ and is sur-
rounded by a square antiprism of eight sulfide ions. The lowest-energy 5f 7 →
4f 65d1 excitation is centered at ~480 nm, and the relatively narrow (full width at
half maximum ≈ 50 nm at 300 K) 4f 65d1 → 4f 7 emission at 540 nm, which makes
this phosphor a potentially useful green emitter in phosphor-converted tricolor
LEDs. (a) What is the magnitude of the Stokes shift in eV? (b) The binary rock-
salt sulfide SrS:Eu2+, with octahedrally coordinated Sr2+ sites, emits in the red
with a maximum at ~620 nm. What factor(s) is responsible for the longer-
wavelength emission in SrS:Eu2+?
7.23 SrS, CaS, and SrSe all crystallize with the cubic rock-salt structure. The emission for
SrS:Eu2+ reaches a maximum near 620 nm. By comparison, the emission maximum for
CaS:Eu2+ is ~660 nm. (a) What factor(s) is responsible for the red shift of the CaS:Eu2+
emission with respect to SrS:Eu2+? (b) The emission maximum for SrSe:Eu2+ is
~570 nm; what factor(s) is responsible for the blue shift of its emission with respect to
SrS:Eu2+?

7.12 Further Reading

P. Atkins, T. Overton, J. Rourke, M. Weller, F. Armstrong, “Inorganic Chemistry” 6th edition (2014)
W.H. Freeman.
G. Blasse, B.C. Grabmaier, “Luminescent Materials” (1994) Springer Verlag.
J.A. Duffy, “Bonding, Energy Levels and Bands in Inorganic Solids (1990) Longman Group Essex.
J.E. House, “Inorganic Chemistry” 2nd edition (2013) Academic Press.
K. Nassau, “The Physics and Chemistry of Color: The Fifteen Causes of Color” 2nd edition (2001) John
Wiley and Sons.
J.C. Phillips, J.A. Van Vechten, “Dielectric classification of crystal structures, ionization potentials
and band structures” Phys. Rev. Lett. 22 (1969), 705–708.
C. Ronda (editor), “Luminescence: From Theory to Applications” (2008) Wiley-VCH.
 7.13 References 299

7.13 References
[1] E. Climent-Pascual, J. Romero de Paz, J. Rodriguez-Carvajal, E. Suard, R. Saez-Puche,
“Synthesis and characterization of the ultramarine-type analog Na8–x[Si6Al6O24](S2,S3,CO3)1–2”
Inorg. Chem. 48 (2009), 6526–6533.
[2] M. Ganio, E.S. Pouyet, S.M. Webb, C.M. Schmidt Patterson, M.S. Walton, “From lapis lazuli to
ultramarine blue: Investigating Cennino Cennini’s recipe using sulfur K-edge XANES” Pure
Appl. Chem. 90 (2018), 463–475.
[3] A.E. Smith, H. Mizoguchi, K. Delaney, N.A. Spaldin, A.W. Sleight, M.A. Subramanian, “Mn3+
in trigonal bipyramidal coordination: A new blue chromophore” J. Am. Chem. Soc. 131 (2009),
17084–17086.
[4] A.E. Smith, M.C. Comstock, M.A. Subramanian, “Spectral properties of the UV absorbing and
near-IR reflecting blue pigment YIn1−xMnxO3” Dyes Pigm. 133 (2016), 214–221.
[5] S. Jitsuhiro, H. Nakai, M. Hada, H. Nakatsuji, “Theoretical study on the ground and excited
states of the chromate anion CrO42−” J. Chem. Phys. 101 (1994), 1029–1036.
[6] A.C. Stückl, C.A. Daul, H.U. Güdel, “Excited-state energies and distortions of d0 transition
metal tetraoxo complexes: A density functional study” J. Chem. Phys. 107 (1997), 4606–4617.
[7] M. Jansen, H.P. Letschert, “Inorganic yellow-red pigments without toxic metals” Nature 404
(2000), 980–982.
[8] S. Romero, A. Mosset, J.C. Trombe, “Study of some ternary and quaternary systems based on γ-
Ce2S3 using oxalate complexes: Stabilization and coloration” J. Alloys Compd. 269 (1998),
98–106.
[9] S. Kufner, J. Furthmüller, L. Matthes, M. Fitzner, F. Bechstedt, “Structural and electronic
properties of α-tin nanocrystals from first principles” Phys. Rev. B 87 (2013), 235307.
[10] J.T.W. de Hair, G. Blasse, “Luminescence of octahedral uranate group” J. Lumin. 14 (1976),
307–323.
[11] P.S. Peijzel, A. Meijerink, R.T. Wegh, M.F. Reid, G.W. Burdick, “A complete 4fn energy level
diagram for all trivalent lanthanide ions” J. Solid State Chem. 178 (2005), 448–453.
[12] A.R. Sharits, J.F. Khoury, P.M. Woodward, “Evaluating NaREMgWO6 (RE = La, Gd, Y)
doubly ordered double perovskites as Eu3+ phosphor hosts” Inorg. Chem. 55 (2016),
12383–12390.
[13] P. Dorenbos, “Ce3+ 5d-centroid shift and vacuum referred 4f-electron binding energies of all
lanthanide impurities in 150 different compounds” J. Lumin. 135 (2013), 93–104.
[14] P. Dorenbos, “Crystal field splitting of lanthanide 4fn−1 5d levels in inorganic compounds”
J. Alloys Compd. 341 (2002), 156–159.
[15] N. Lakshminarasimhan, S. Jayakiruba, K. Prabhavathi, “Ba2Mg(BO3)2:Bi3+– A new phosphor
with ultraviolet light emission” Solid State Sci. 72 (2017), 1–4.
[16] G. Blasse, B.C. Grabmaier, “Luminescent Materials” (1994) Springer Verlag, Berlin.
[17] C.W.M. Timmermans, O.B. He, G. Blasse, “The luminescence of Bi2Ge3O9” Solid State
Commun. 42 (1982), 505–507.
[18] H. Amano, N. Sawaki, I. Akasaki, Y. Toyoda, “Metalorganic vapor phase epitaxial growth of
a high quality GaN film using an AlN buffer layer” Appl. Phys. Lett. 48 (1986), 353–355.
[19] S. Nakamura, “GaN growth using GaN buffer layer” Jpn. J. Appl. Phys. 30 (1991), L1705–1707.
[20] S. Nakamura, N. Iwasa, M. Senoh, T. Mukai, “Hole compensation mechanism of p-type GaN
films” Jpn. J. Appl. Phys. 31 (1992), 1258–1266.
300 Optical Materials

[21] H. Sasabe, J. Kido, “Multifunctional materials in high-performance OLEDs: Challenges for

solid-state lighting” Chem. Mater. 23 (2011), 621–630.
[22] Z. Xinmin, W. Hao, Z. Heping, S. Qiang, “Luminescent properties of SrGa2S4:Eu2+and its
application in green-LEDs” J. Rare Earths 25 (2007), 701–705.
[23] L. Wang, R.-J. Xie, T. Suehiro, T. Takeda, N. Hirosaki, “Down-conversion nitride materials for
solid state lighting: Recent advances and perspectives” Chem. Rev. 118 (2018), 1951–2009.
[24] Z. Xia, Q. Liu, “Progress in discovery and structural design of color conversion phosphors for
LEDs” Prog. Mater. Sci. 84 (2016), 59–117.
8 Dielectrics and Nonlinear Optical Materials

Materials can be broadly classified as either conductors or insulators. Conductors can

further be divided into electronic and ionic conductors, which are covered in Chapters 10
and 13, respectively. Insulating materials, which are also referred to as dielectric materials,
are the subject of this chapter. Dielectric materials find widespread application in electronics.
In some cases, the role of the dielectric is simply to insulate active circuit components from
each other. In other instances, the dielectric plays an active role as a capacitor, antenna, or
filter. In the latter case, the response of the material to external electric fields is of critical
importance.
Nonlinear optical (NLO) materials are a subset of the broader class of dielectric
materials. When electromagnetic radiation passes through a NLO material new frequen-
cies of radiation are generated. A familiar example is found in green laser pointers, where
an NLO crystal is used to convert infrared (IR) light into green light. As we will see
NLO effects are only observed in materials that meet specific symmetry criteria. In the
later sections of this chapter we’ll look at the characteristics of several important NLO
materials.

8.1 Dielectric Properties

When an electric field of intensity1 E is applied to a dielectric material, an electrical
polarization develops in response to the applied field. In the following sections, we explore
this response, starting with a macroscopic description followed by a closer look at the
microscopic origins of the dielectric response.

1
An electric field is a vector field, but when homogeneous or when speaking of a local value of unambiguous
direction, only the magnitude of its intensity, E, is sufficient. The unit of E or E is volts per meter. 1 V/m exerts
a force of 1 N onto a point charge of 1 C.

301
302 Dielectrics and Nonlinear Optical Materials

8.1.1 Dielectric Permittivity and Susceptibility

A parallel-plate capacitor consists of two conducting plates, each with an area, A, separated
by a distance, d, as shown in Figure 8.1. In a vacuum, its capacitance2 is C0:

A
C 0 ¼ ε0 (8.1)
d

where ε0 is the electric constant.3 The quantity of charge, Q, that can be stored4 is a product
of the capacitance and voltage drop across the dielectric, V = Ed:

Q ¼ CV (8.2)

For a parallel-plate capacitor in a vacuum, the stored charge, Q0, is given by:
 
ε0 A
Q0 ¼ C0 V ¼ Ed ¼ ε0 AE (8.3)
d

If we place a dielectric material between the plates while maintaining the voltage drop V,
the amount of stored charge will increase εr times, from Q0 to Q, due to the polarization of
the material.5 The polarization results in a separation of positive and negative charges, by
which the dielectric material acquires an internal field that opposes the applied field E. More
charge is then needed to compensate this dielectric polarization while maintaining the

+ +
+
+ +
+ + + + + + + + +
–
– – +
d + –
− + + −
− −
– – – – – – – – – −

(a) Stored charge = Q0 (b) Stored charge = Q

Figure 8.1 A parallel-plate capacitor with (a) a vacuum (b) a dielectric material.

2
Its SI unit is one farad, 1 F = 1 C/1 V; it is the charge (in coulombs) a capacitor will accept for the potential across it to
change by 1 V.
3
The electric constant is a fundamental constant whose value (to five significant figures) is 8.8542×10−12 F/m. It is also
referred to as the vacuum permittivity or the permittivity of free space.
4
The stored charge is the charge at one of the plates, typically the + charge is considered.
5
Alternatively, we could also keep the charge constant in our thought experiment. Then the voltage and E would
decrease due to the presence of the dielectric material.
 8.1 Dielectric Properties 303

Table 8.1 Relative permittivity εr for a variety of materials at room

temperature.

Substance εr Substance εr

Air 1.0006 SnO2 13

Teflon 2 ZrO2 22
C (diamond) 6 TiO2 94
Si 12 Perovskite dielectrics
NaF 5 BaSnO3 18
MgO 10 BaZrO3 43
SiO2 4 CaTiO3 165
CaF2 7 SrTiO3 330
PbI2 21 KTaO3 242
H2O 80 Ba3ZnTa2O9 30

original voltage of the circuit. Consequently, the capacitance will increase εr times, from C0
to C. The factor εr is termed the relative dielectric permittivity6:

εr ¼ C=C0 (8.4)

Relative dielectric permittivities of substances vary across a wide range, as shown in Table
8.1, but most insulating solids have εr < 30. The reasons for the high permittivity values of
water and certain metal oxides, such as the titanates, will become clear later in the chapter. In
Section 8.4, we will discuss a special class of dielectric materials called ferroelectrics whose
relative dielectric permittivity can be in the tens of thousands.
Upon subtracting 1 from the factor εr (i.e. subtracting the contribution of the
vacuum from the total capacitance), we obtain a dimensionless value that quantifies
the ability of a dielectric material to become polarized in an external field, its electric
susceptibility, χe:
χ e ¼ εr  1 (8.5)

8.1.2 Polarization and the Clausius–Mossotti Equation

The capacitance of our parallel-plate capacitor increases when a material occupies the space
between the plates, because the electric field creates dipoles within the material that oppose
the applied field. Thus, polarization is the material property of interest to us. Quantitatively
we define bulk polarization, P, of a dielectric substance with the following equation (magni-
tudes-only for simplicity):

6
Terms relative dielectric permittivity and dielectric constant are used interchangeably.
304 Dielectrics and Nonlinear Optical Materials

P ¼ ε0 χ e E (8.6)
where the polarization P is expressed in units7 of F/m. Substituting for χe from Equation
(8.5), we obtain:

P ¼ εr ε0 E  ε0 E (8.7)

In this expression, we then define the electric displacement,8 which describes the electric
field inside the dielectric, as D = εrε0E. We see that D has two components; the electric
displacement in a vacuum and the polarization of the material:

D ¼ ε0 E þ P (8.8)

Because (as noted above) D, E, and P are all vectors, the polarization of a crystal in an
electric field is anisotropic.
Let’s consider polarization from induced local dipole moments within the dielectric. The
bulk polarization P [C/m2] can be obtained by summing up the individual induced dipole
moments, p [C m], over the N atoms, ions, or molecules that occupy a cubic meter. For the
sake of simplicity, we will speak of polarization of neutral atoms (e.g. Si) for the remainder of
this section, but the expressions that are derived here are equally valid for ionic and
molecular solids.
The size of the local induced dipole moment p is proportional to the local electric field,
Eloc, experienced by the atom, with polarizability, α, of that atom acting as the proportional-
ity constant:

p ¼ αEloc (8.9)

The bulk polarization P (the induced dipole moment per m3) of a crystal containing N atoms
per cubic meter is then:

P ¼ NαEloc (8.10)

Unfortunately, Eloc differs from the applied external field E because the induced dipole
moments of neighboring atoms alter the field strength within the dielectric, hence Eloc
depends on the atom’s location in the crystal. Equation (8.10) is therefore of limited utility
as we cannot directly measure Eloc. Fortunately, it is possible to derive a relationship between
the local and external fields:

7
Because susceptibility is dimensionless, the polarization P has units of C/m2 since ε0 [F/m] × E [V/m] × χe = P [F V/m2)]
and 1 F = 1 C/V. P is also called the polarization density, because it represents an electric dipole moment per cubic meter
[C m/m3].
8
Electric displacement is also referred to as electric-flux density or electric induction (the latter in analogy with
magnetic induction, see Chapter 9) and is generally a vector.
 8.1 Dielectric Properties 305

1
Eloc ¼ ðεr þ 2ÞE (8.11)
3

Substituting Equation (8.11) into Equation (8.10), we obtain an expression for P, which we
can equate to polarization as expressed in Equation (8.7). After simplifying this expression,
we obtain the Clausius–Mossotti equation, which relates the atomic polarizability α,
a microscopic quantity, to the dielectric permittivity εr, a macroscopic quantity:
 
3ε0 εr  1  
α¼ SI: α in C m2 =V; ε0 in C=ðV mÞ; N in m3 (8.12)
N εr þ 2

Because an average Eloc was assumed, Equation (8.12) is strictly speaking only valid for crystal
structures of certain symmetry that makes them isotropic with respect to the dielectric-constant
tensor. An involved evaluation shows that only cubic crystals meet this criterion, but in practice
it works well for other crystal systems if the structure is not too anisotropic.
It is often convenient to work in CGS (centimeter–gram–second) units where ε0 = 1 and
hence α has units of volume. Because εr = 1 + 4πχe in the CGSes (electrostatic) system, the
Clausius–Mossotti equation takes on a somewhat different form:
 
3 εr  1 83 
83 ; Va in A
α ¼ Va ½CGSes : α in A (8.13)
4π εr þ 2

where Va is the volume per atom or formula unit, determined by dividing the unit-cell volume
by the number of the formula units it contains. For example, the sphalerite form of ZnS
(F43m, a = 5.32 Å, Z = 4, Figure 1.33) has a Va = (5.32 Å)3/4 = 37.6 Å3.

8.1.3 Microscopic Mechanisms of Polarizability

Polarizability is a key parameter for understanding dielectric permittivity of materials
because it directly relates to properties of the atoms that make up the solid. In an ideal
dielectric with no electronic or ionic conductivity,9 there are three microscopic polarization
mechanisms that can contribute to the dielectric response of a material: the electronic
polarizability, αe; the ionic polarizability, αi; and the dipolar polarizability, αd. These mechan-
isms are illustrated in Figure 8.2.
The electronic polarizability αe, which is present in all substances, arises from polarization
of the negatively charged electron cloud surrounding the nucleus. The ionic polarizability αi
arises from field-induced displacements of cations and anions in opposite directions. In ionic
solids, including most technologically important dielectric materials, ionic polarizability is
the principal source of polarization. The dipolar polarizability αd arises from reorientations

9
In ionic conductors, very high polarizability can occur due to migration of ions. This type of polarization is called
space-charge polarizability.
306 Dielectrics and Nonlinear Optical Materials

electron –
electronic cloud
polarizability, αe atoms
nucleus
+
Na+ –
ionic ions
polarizability, αi

Cl– +
–
dipolar polar
polarizability, αd H
molecules
Cl
+

Figure 8.2 Three microscopic polarizability mechanisms. The change from left to right shows the
polarization in response to an applied electric field.

of polar molecules in response to the applied field. It is normally only relevant in molecular
substances containing polar molecules, such as H2O and HCl. The αd contribution tends to
be very temperature-dependent because the reorientations tend to freeze out at low temper-
atures. For most technologically important dielectric materials, the dipolar polarizability is
not relevant.
The magnitude of these three contributions typically follows the order αe < αi < αd. The
relative magnitude of these polarizabilities is reflected in the relative permittivity values given
in Table 8.1. In covalent network solids like diamond and silicon, αe is the only source of
polarization and εr is fairly small. The εr values for “ionic” materials, where αi plays an
important role, vary over a wide range. Large values are often seen in compounds containing
transition metals with a d 0 electron configuration. We will explore the reasons for this
behavior in Section 8.6. The high permittivity of water stems from its being the only
substance in Table 8.1 where αd contributes.

8.1.4 Frequency Dependence of the Dielectric Response

Polarization involves reorganization of charge. Electronic polarization depends on deform-
ation of electron density, ionic polarization depends on displacements of ions, and dipolar
polarization involves reorientations of molecules. All three mechanisms can respond to
a static electric field, but when subjected to an alternating field of increasing frequency, ν,
a limit is eventually reached where the field direction changes faster than the charged entity
 8.1 Dielectric Properties 307

can follow. When this occurs, a given polarization mechanism no longer contributes to the
dielectric permittivity. For dipolar polarization this occurs when ν > 109 Hz (GHz or
microwave frequencies), for ionic polarization when ν > 1013 Hz (THz or IR frequencies),
and for electronic polarization when ν > 1017 Hz (X-rays). This response is illustrated in
Figure 8.3, where εr is plotted as a function of ν for a material where at low frequencies all
three polarization mechanisms contribute.10
For most useful dielectrics, αd = 0, and the relative permittivity at low frequencies is
determined by αi and αe. The sum of these two contributions can be determined by measuring
the static dielectric constant, denoted εr (ν → 0) or εstat, via capacitance measurements
described in Section 8.1.1. To separate the contributions of αi and αe, one then measures
the response of the dielectric to visible light where αi is frozen out. The high-frequency or
optical dielectric constant, εopt, is related to the refractive index n (see Section 8.7), through
the relationship:
εopt ¼ n2 (8.14)

For a covalent material like diamond, εstat and εopt are very similar (εstat = 5.68, εopt = 5.66),
which tells us that αi is essentially zero. For ionic materials like NaCl (εstat = 5.90, εopt = 2.34)
and LiCl (εstat = 11.95, εopt = 2.78), the ionic and electronic contributions are of comparable
magnitude. For compounds of unusually high permittivities, like TiO2 (εstat = 94, εopt = 7),
the ionic contribution is dominant for reasons that will be discussed in Section 8.6.

Figure 8.3 Dielectric permittivity εr as a function of the frequency ν of an alternating applied field for
a hypothetical dielectric material where all three polarization mechanisms contribute at low frequencies.

10
If a material exhibits ionic conductivity, the space-charge polarizability relaxes out when ν > 106 Hz.
308 Dielectrics and Nonlinear Optical Materials

8.1.5 Dielectric Loss

An ideal capacitor stores energy without loss or dissipation of energy. In theory, a dielectric
material can act as an ideal capacitor, but, in practice, some energy is always lost. Resistive
heating due to non-zero conductivity in the dielectric material is one obvious source of loss.
However, losses can also occur in an alternating electric field even in the absence of
conductivity.
When an alternating field is applied, the capacitor electrodes alternate between positive
and negative charges, and the polarization within the capacitor’s dielectric tries to follow
suit. Some energy is required to drive the movement needed to redistribute charge in the
dielectric: molecular reorientations for dipolar polarization, or lattice vibrations (phonons)
for ionic polarization. This energy eventually converts to heat and is therefore called dielec-
tric loss. In a simplified approach, the dielectric permittivity in an alternating field is
a complex number, εr*:

εr ¼ ε0 þ iε00 (8.15)

where ε 0 and ε 00 are the real and imaginary parts of the dielectric permittivity, respectively.
The real part is a measure of the polarization that is in phase with the external field (i.e. it
keeps up with the field), while the imaginary part corresponds to polarization that is out of
phase with the field (i.e. it lags behind the field). The real component represents the energy
that is stored while the imaginary component is a measure of dielectric loss.
Quantitatively, the dielectric loss is defined as tan δ = ε 00 /ε 0 . The parameter δ is a measure of
the phase difference between the instantaneous current, i, and the instantaneous voltage, v.
The phase of the alternating current leads that of the voltage11 by an angle of 90°−δ. When
δ = 0, the phase of the current is 90° ahead of the phase of the voltage, so that i × v = 0 and
there are no losses. When δ ≠ 0, a component of the current has the same phase as the voltage,
and energy dissipates as heat.
If we assume a so-called Debye model, which consists of an ideal non-interacting system of
dipoles,12 the real and imaginary parts of the dielectric permittivity can be approximated by
the following equations:
εstat  εopt
ε0 ¼ εopt þ (8.16)
1 þ ðωτÞ2
ðεstat  εopt Þ ωτ
ε″ ¼ (8.17)
1 þ ðωτÞ2

11
In a capacitor, the current precedes the voltage; current must flow into a capacitor to establish the voltage drop.
12
The Debye model most accurately approximates dipolar polarization. For ionic and electronic polarization,
a harmonic-oscillator model is a better approximation, but leads to a slightly more complicated expression.
Qualitatively, both models give a similar dependence on frequency.
 8.2 Dielectric Polarizabilities and the Additivity Rule 309

ωτ = 1
εstat

ε’ ε’’

εopt ωτ = 1

109 1011 1013 1015 109 1011 1013 1015

ν (Hz) ν (Hz)

Figure 8.4 The frequency dependence of ε 0 and ε 00 for a material where only ionic and electronic
polarization contribute (αd = 0).

where εstat and εopt represent the static and optical dielectric constants, the angular fre-
quency, ω = 2πν, is calculated from the frequency ν of the alternating electric field and from
the characteristic relaxation time τ of the polarization. Typical values of τ are on the order of
~10−11 s for dipolar polarizability and ~10−14 s for ionic polarizability. The behaviors of ε 0
and ε″ as a function of frequency are plotted in Figure 8.4.
The imaginary part of the dielectric permittivity ε 00 and the dielectric loss, tan δ = (ε 00 /ε 0 ), reach
a maximum when the frequency of the alternating external field matches the resonant frequency
of the material, ωτ = 1. For such frequencies, vibrational modes are excited, and the absorbed
energy is dissipated as heat. At frequencies sufficiently low that ωτ ≪ 1, each of the polarization
mechanisms saturates before the field reverses; Equation (8.16) reduces to ε 0 ≈ εstat and Equation
(8.17) to ε 00 ≈ 0. At frequencies high enough to make ωτ ≫ 1, the above equations reduce to
ε 0 ≈ εopt and ε 00 ≈ 0. Losses do not occur in the low-frequency limit because there is no coupling
between the applied field and lattice vibrations, while in the high-frequency limit the lattice
vibrations cannot respond to the rapidly oscillating electric field.

8.2 Dielectric Polarizabilities and the Additivity Rule

From a materials-design perspective, it would be attractive to predict the dielectric permittivity
from composition and structure. This is possible when the Clausius–Mossotti equation (8.13) is
rearranged to give εr as a function of the polarizability α and volume per polarized atom Va:

Va þ 2αð4π=3Þ
εr ¼ (8.18)
Va  αð4π=3Þ
310 Dielectrics and Nonlinear Optical Materials

The polarizability of a compound AxByCd . . . can be approximated by summing the polariz-

abilities of its constituent “ions” α(A), α(B), α(C), . . .:

αðAx By Cz : : :Þ ¼ xαðAÞ þ yαðBÞ þ zαðCÞ þ : : : (8.19)

This relationship is called the additivity rule. Working from Equations (8.18) and (8.19),
Shannon [1] refined values of α for individual ions to reproduce the measured values of εr for
129 oxides and 25 fluorides. Selected α values are given in Table 8.2, in the CGSes units Å3 to
go with Equation (8.13).
Intuitively, we might expect that polarizability would increase as the radius of the ion increases
and its charge decreases. Atoms with outer electrons that are well screened from the nucleus will
be the most polarizable. For the most part, the trends in polarizability follow our expectations by
increasing down a group and decreasing across a period. There are two important exceptions,
each associated with a specific electron configuration. Firstly, ions with an s2 lone-pair configur-
ation (Tl+, Pb2+, Sb3+, Bi3+, Se4+, and Te4+) have larger polarizabilities than expected. Secondly,
the polarizabilities of transition-metal “ions” do not continue to decrease when the oxidation
state climbs above +3. Both effects result from mixing of empty cation orbitals with occupied
anion orbitals and will be examined more closely in Section 8.6.
To illustrate how the ionic polarizabilities and the additivity rule can be used to estimate the
dielectric permittivity of a compound, consider the cubic garnet (Section 1.5.2), Y3Al5O12, with
a = 12.01 Å. There are eight formula units per unit cell (Z = 8), so the volume per formula unit
is (12.01 Å)3/8 = 216.4 Å3. Ionic polarizabilities from Table 8.2 and Equation (8.19) give the

Table 8.2 Dielectric polarizabilities α (in Å3) for selected ions from ref. [1].

Li+ Be2+ B3+ O2− F−

1.20 0.19 0.05 2.01 1.62
Na+ Mg2+ Al3+ Si4+ P5+
1.80 1.32 0.79 0.87 0.27
K+ Ca2+ Sc3+ Ti4+ V5+ Zn2+ Ga3+ Ge4+ As5+
3.83 3.16 2.81 2.93 2.92 2.04 1.50 1.63 1.72
Rb+ Sr2+ Y3+ Zr4+ Nb5+ Cd2+ In3+ Sn4+ Sb3+ Te4+
5.29 4.24 3.81 3.25 3.97 3.40 2.62 2.83 4.27 5.23
Cs+ Ba2+ La3+ Hf4+ Ta5+ Hg2+ Tl+ Pb2+ Bi3+
7.43 6.40 6.07 --- 4.73 --- 7.28 6.58 6.12

Ce4+ Pr3+ Nd3+ Pm3+ Sm3+ Eu3+ Gd3+ Tb3+ Dy3+ Ho3+ Er3+ Tm3+ Yb3+ Lu3+
3.94 5.32 5.01 --- 4.74 4.53 4.37 4.25 4.07 3.97 3.81 3.82 3.58 3.64
 8.2 Dielectric Polarizabilities and the Additivity Rule 311

polarizability estimate per formula unit: α(Y3Al5O12) = 3α(Y3+) + 5α(Al3+) + 12α(O2−)

= 3(3.81 Å3) + 5(0.79 Å3) + 12(2.01 Å3) = 39.5 Å3. Plugging α = 39.5 Å3 and Va =
216.4 Å3 back into Equation (8.18) yields εr = 10.7, in good agreement with the
measured value of 10.6.
It is important to note that this approach is only valid for “normal” dielectric materials.
Deviations between the calculated and observed values may be attributed to ferroelectricity
(Section 8.4), piezoelectricity (Section 8.5), conductivity (ionic or electronic), the presence of
rattling ions, compressed ions, and dipolar impurities (such as H2O).
The polarizabilities in Table 8.2 account for both ionic and electronic contributions to
polarization. Shannon and Fischer [2] developed a set of polarizabilities that contain only the
electronic contribution by substituting εopt for εr in Equation (8.18). These values are of use for
estimating refractive indices, simulating spectra, and various approaches to modeling extended
solids. For more details on the relationship between electronic polarizabilities and refractive
index, see Section 15.4. As one would expect, the electronic polarizabilities are smaller than the
dielectric polarizabilities, because the ionic polarization no longer contributes.

Box 8.1 Materials Spotlight: Microwave dielectrics

Many technologies, including cellular phones, use electromagnetic radiation with wavelengths in
the microwave region (frequencies in the GHz range) to transmit information. Several key
components in a microwave communication network, such as antennas, transmitters, and filters,
are built from dielectric ceramics. One such component is the microwave resonator used in a cell-
phone base station, which links the radio signals that cellular phones send and receive with the
network switching subsystem. The resonator is typically a hollow ceramic cylinder, called a puck.
The dimensions of the puck (on the order of centimeters) are chosen so that it can sustain a standing
wave within its body when exposed to microwaves of a specific resonant frequency. This allows it to
transmit microwaves that fall within a narrow frequency range and filter out other frequencies.

ν0
relative transmitted power (dB)

Δν0

ceramic pucks used in

microwave resonators

frequency
312 Dielectrics and Nonlinear Optical Materials

Box 8.1 (cont.)

A dielectric material useful for this application should meet three key requirements. Firstly,
because the size of the resonator is proportional to 1/√εr, a high dielectric permittivity is desirable
because it allows for a reduction in the size of the resonator. In practice, the optimal range is
20 < εr <50. Secondly, the selectivity of the resonator increases as the dielectric loss decreases.
The selectivity is defined by a parameter called the quality factor, Q, which is approximately
equal to 1/(tan δ). If the transmitted power is plotted as a function of the microwave
frequency, a peak is observed at the resonant frequency, as shown above. The quality factor
is equal to the resonant frequency, ν0, divided by the width of the peak, Δν0, as measured at
a transmitted power 3 dB below the peak. Higher Q reduces the crosstalk within the specified
frequency range. The value of Q is dependent on the resonant frequency of the puck ν0, but
the product Q × ν0 should in theory be constant and is often used as a figure of merit for
comparing different materials. For use in cell-phone base stations, Q × ν0 should be equal to
or larger than 4×104. Thirdly, the resonant frequency should be nearly independent of
temperature. The temperature dependence is given by the temperature coefficient of resonant
frequency, τf, which quantifies the change in the resonant frequency Δν0 with a change in
temperature ΔT. It is typically expressed in units of ppm/K, (106Δν0/ν0)/ΔT. For use as
a microwave resonator, τf should be smaller than ±3 ppm/K.
Most ceramics have εr too small for microwave resonators. Ferroelectrics are not suitable
because their losses are much too high, which drives down Q. Incipient ferroelectrics like
SrTiO3 are not suitable because their permittivity changes too much with temperature. The
best microwave resonators are insulating materials with relatively high, temperature-stable,
permittivities and very low losses. Many of the best microwave dielectrics come from the
perovskite family as seen in the table below. Note the τf values for all materials listed below
are nearly zero [3].

Material εr Q × ν0 (GHz) Structure

Ba3MgTa2O9 24 2.5 × 105 2:1 ordered perovskite

Ba3ZnTa2O9 29 1.5 × 105 2:1 ordered perovskite
Ba3(Co1−xZnx)Nb2O9 34 9.0 × 104 2:1 ordered perovskite
(Sr1−xLax)(Ti1−xAlx)O3 39 6.0 × 104 Simple perovskite
(Ca1−xNdx)(Ti1−xAlx)O3 45 4.8 × 104 Simple perovskite

These perovskites contain a mixture of d 0 cations (Ta5+, Nb5+, Ti4+) and non d 0 cations
(Mg2+, Zn2+, Co2+, Al3+) on the octahedral sites. The large εr is due to the presence of d 0
cations, while dilution with non d 0 cations prevents phase transitions into a ferroelectric state
and reduces τf. The perovskites can be divided into two categories, those with a 2:1 ordering of
octahedral cations, and those that are solid solutions of high-εr materials (SrTiO3 or CaTiO3)
and low-εr materials (LaAlO3 or NdAlO3).
 8.3 Crystallographic Symmetry and Dielectric Properties 313

8.3 Crystallographic Symmetry and Dielectric Properties

All dielectric materials are polarized by their interaction with an external electromag-
netic field, but certain properties we encounter later in this chapter can only arise in
crystals with specific types of symmetry. In this section, we briefly explore the sym-
metry restrictions on pyroelectricity, ferroelectricity, piezoelectricity, and second-
harmonic generation.
Crystals are classified as either centrosymmetric if they possess an inversion center, or non-
centrosymmetric if they do not. Of the 32 crystallographic point groups (Section 1.1.3), 21
are non-centrosymmetric (Table 8.3). Piezoelectricity and second-harmonic generation are
permitted in all non-centrosymmetric crystals, except cubic crystals with 432 point-group
symmetry, which includes space groups with primitive cubic (P432, P4132, P4232, P4332),
body-centered cubic (I432, I4132), and face-centered cubic (F432, F4132) Bravais lattices,
illustrating that it is the point-group symmetry, not the Bravais lattice, that dictates which of
the above dielectric phenomena are permitted.
Materials that form crystals with a macroscopic electric dipole moment are called polar
materials.13 Such a crystal is said to be spontaneously polarized, because the non-zero
electrical polarization forms spontaneously, even in the absence of an external electric
field. To be polar, a material must not only lack an inversion center, it must also possess
a polar axis. A rotational axis 1, 2, 3, 4, or 6 is polar if its positive and negative ends are
not equivalent, hence the axis must not be perpendicular to a twofold axis or a mirror

Table 8.3 Non-centrosymmetric point-group symmetries sorted according to whether they

allow piezoelectricity (20 piezoelectric crystal classes), both piezoelectricity and
pyroelectricity (10 polar crystal classes), or neither.

Piezoelectric crystal classes

Crystal system Polar crystal classes Neither

Triclinic 1 – –
Monoclinic 2, m – –
Orthorhombic mm2 222 –
Trigonal 3m, 3 32 –
Tetragonal 4mm, 4 422, 42m, 4 –
Hexagonal 6mm, 6 622, 62m, 6 –
Cubic – 23, 43m 432

13
Somewhat confusingly, in the crystallographic literature the terms polar material and polar crystal class are used to
describe any material that crystallizes in a non-centrosymmetric space group. Throughout this book, we use the
more restrictive condition that a crystal must contain a polar axis to be considered a polar material.
314 Dielectrics and Nonlinear Optical Materials

plane.14 This narrows the list to 10 polar crystal classes given in Table 8.3.15 Pyroelectricity
and ferroelectricity are only permitted in polar crystals.

8.4 Pyroelectricity and Ferroelectricity

In all polar materials, the magnitude of the bulk polarization changes with temperature;
a property referred to as pyroelectricity. Ferroelectrics are a special class of polar materials
where the permanent electric dipole moment of a crystal can be reversed through application
of an external electric field. In ferroelectric materials there is a temperature called the Curie
temperature, TC, above which thermally induced vibrations of the ions lead to a loss of spontan-
eous polarization16, and hence a loss of ferroelectricity. Above TC, a ferroelectric material enters
a paraelectric state. Ferroelectric materials are used in capacitors due to their (often) large
dielectric permittivity. Because the direction of the electrical polarization switches permanently
after application of an electrical field, they also find use in data storage (ferroelectric random-
access memory).
Although the term pyroelectric is general, it is often used in a narrower sense to describe those
polar materials that are not ferroelectric. Compounds with the wurtzite structure (P63mc of
crystal class 6mm, Figure 1.33), such as GaN, provide one such example. Alternating cation and
anion layers stack perpendicularly to the c axis in the wurtzite structure, creating a net dipole
moment parallel to the c axis. One would have to reverse the layer-stacking sequence from Ga–
N–Ga–N– . . . to N–Ga–N–Ga– . . . to switch the polarity of the crystal. The energy barrier to
this reorganization of the crystal is too large to be overcome with an external electric field, hence
GaN is a pyroelectric but not a ferroelectric. Gallium nitride and other pyroelectric materials
find application as IR (heat) sensors.

8.4.1 Ferroelectricity in BaTiO3

To understand the origins of ferroelectricity, we take a closer look at the archetypal
perovskite ferroelectric, BaTiO3. We first relate the dielectric properties of BaTiO3 to
changes in its average crystal structure that occur at the Curie temperature, a model that
provides a good entry point for understanding structure–property relationships in ferroelec-
trics. Subsequently, we investigate the phase transitions of BaTiO3 and see how the local
structure may differ from the average crystal structure.
Above TC ≈ 400 K, BaTiO3 has the cubic perovskite structure (Section 1.5.3). On cooling
below TC, the symmetry lowers from cubic (space group Fm3m) to tetragonal (P4mm). This
14
Neumann’s principle states that the physical property of a crystal must have at least the symmetry of the crystal’s
point group. The polarization arrow does not have a twofold axis or m perpendicular to it.
15
The oft-used term “unique axis” can be misleading because one may not realize which axis is meant and that it includes
the onefold axes in crystal classes 1 and m ≡ 1m1 (the extended Hermann–Maugin symbol; see Appendix B).
16
A crystal that has a net electric dipole moment is said to be spontaneously polarized, because the non-zero electrical
polarization persists in the absence of an external electric field.
 8.4 Pyroelectricity and Ferroelectricity 315

dTi–O = 2.00 Å dTi–O = 1.83 Å

Ba

Ti

b
a dTi–O = 2.00 Å dTi–O = 2.20 Å

cubic BaTiO 3 (T = 474 K) tetragonal BaTiO 3 (T = 280 K)

Figure 8.5 The average structure of BaTiO3 above (left) and below (right) TC. The four equatorial Ti–O
bond lengths remain ~2.00 Å in the tetragonal structure.

distortion of the average structure involves a displacement of the titanium ions along the
tetragonal fourfold axes, accompanied by a shift of the oxide ions in the opposite direction,
resulting in the formation of one short and one long Ti–O bond per octahedron, as shown in
Figure 8.5. These cooperative ionic displacements cause the crystal to develop a net polar-
ization, whereby the crystal face the titanium ions approach develops a positive charge and
the opposite face a negative charge.
While this simple picture captures the microscopic origins that lead to the formation of local
dipole moments in BaTiO3, the macroscopic properties are strongly impacted by the formation
of domains. To understand what domains are and how they affect the macroscopic properties of
a ferroelectric crystal, we need to investigate the crystal over a much larger length scale.
The high-temperature cubic structure possesses fourfold axes running parallel to each of the
three Cartesian directions. Consequently, each titanium atom has six equivalent directions (see
Figure 8.6) along which it can displace and still attain the tetragonal symmetry of the average
structure below TC. Over the length scale of hundreds or thousands of unit cells, the average
titanium displacements are likely to be in the same direction, but, when viewed over a larger
length scale, the crystal actually consists of individual domains, each with one of the six different
polarization directions. Domains in ferroelectric materials typically have dimensions on the
micron length scale, with domain walls a few unit cells thick. The formation of ferroelectric
domains helps to minimize the electrostatic and elastic energy of the crystal.17 If the displace-
ments in neighboring domains are in opposite directions, the boundary is said to be a 180°

17
Spontaneous polarization within a crystal leads to the formation of a surface charge that creates a depolarizing field
oriented oppositely to the bulk polarization. The electrostatic energy associated with the depolarizing field may be
minimized if the ferroelectric splits into domains with oppositely oriented polarization. Domain formation also
helps to offset strains that arise due to changes in shape of the crystal that occur as it is cooled through the
paraelectric–ferroelectric phase transition.
316 Dielectrics and Nonlinear Optical Materials

Figure 8.6 Six possible average polarization directions at the cubic-to-tetragonal phase transition in bulk
BaTiO3 (left), and a cross-section of a typical domain structure with these polarization directions
represented by arrows (right).

domain wall. If the displacements are at right angles to each other, the boundary is a 90° domain
wall. Both types of domains are illustrated schematically in Figure 8.6.
The polarization of a ferroelectric crystal as a function of the applied electric field is illustrated
in Figure 8.7. This curve is called a hysteresis loop, because the polarization that occurs on
increasing the field is not reproduced on subsequently decreasing the field. Due to the formation
of oppositely polarized domains, the bulk polarization P of a ferroelectric crystal that has never
been exposed to an electric field will be nearly zero. As the applied field increases (pathway 1 → 2
in Figure 8.7), the polarization increases until it reaches the saturation polarization, Psaturation
(point 2 in Figure 8.7). This corresponds to the state where all domains are oriented parallel to
the applied field. The saturation polarization of a good ferroelectric typically ranges from
0.1–1 C/m2. Upon removing the applied field, the polarization decreases, but does not go to
zero. The value of the polarization that remains is called the remanent polarization, Premanent
(point 3 in Figure 8.7). If a field is now applied in the opposite direction, the polarization
continues to decrease until it reaches zero. The magnitude of this field is the (negative) coercive
field, −Ecoercive (point 4 in Figure 8.7). Upon increasing this reverse field, the polarization
saturates at −Psaturation (under the switching field, point 5). From there, the polarization follows
the hysteresis loop marked with a solid line in a counterclockwise direction for further changes in
applied field. We will see in Chapter 9 that ferromagnets behave similarly, with magnetic dipole
moments taking the place of electric dipoles.
The relative permittivity of a BaTiO3 single crystal is plotted as a function of temperature
in Figure 8.8. The peak at TC ≈ 400 K separates the paraelectric cubic phase from the
tetragonal ferroelectric phase. Such a peak at TC is a characteristic of ferroelectric materials.
The peak near 270 K is associated with a transition to an orthorhombic structure, while the
peak near 180 K corresponds to a transition to a rhombohedral structure. The tetragonal,
orthorhombic, and rhombohedral forms are all ferroelectric.
 8.4 Pyroelectricity and Ferroelectricity 317

+ + +
P 1 2
Psaturation
2
Premanent 3
– – –

6 3
−Ecoercive
1 7
4 E
Ecoercive – – – – – –
5 4

5 6
+ + + + + +

Figure 8.7 Hysteresis loop for a ferroelectric material and schematic evolution of a vertical domain pair
(for simplicity) at several points on the loop. The domain structure at point 7 is identical to point 4, but
with an oppositely polarized external field.
rhombohedral

orthorhombic

10000
tetragonal

cubic
8000

6000
εr
a axis
4000

2000
c axis
0
100 200 300 400

T (K)

Figure 8.8 The dielectric permittivity εr of a single crystal of BaTiO3, measured parallel to the a and c axes
at a non-zero electric field, as a function of temperature. Data taken from reference [4].

Naively, one might expect εr to be quite small in the cubic structure because there are no
permanent local dipole moments, yet for temperatures just above TC the relative permit-
tivity is higher than it is at most temperatures below TC. The reason for this behavior is
318 Dielectrics and Nonlinear Optical Materials

that the ionic displacements, and the local dipoles that accompany them, do not vanish
above TC; they become dynamic. In the paraelectric state, lattice vibrations have sufficient
energy to move the titanium ions back and forth from one side of the octahedron to
another. While on average each titanium ion appears to sit at the center of its octahedron,
at any given instant in time it is almost certainly shifted away from the center of its
octahedron. These thermally induced vibrations make it impossible to build up
a spontaneous permanent polarization of the crystal along a specific direction.
Nonetheless, there is still a large response to an applied electric field, giving rise to the high
εr seen above TC in Figure 8.8. The fact that εr reaches a maximum can be largely
attributed to weakening of the domain structure as the temperature approaches TC from
below. As we have already learned, the formation of domains tends to reduce the net
polarization of the crystal, so it should not be surprising to find that εr increases as the
domain structure disappears.
The tetragonal, orthorhombic, and rhombohedral forms of BaTiO3 exhibit average
displacements of Ti towards the corner, edge, and face of the octahedra, respectively, as
shown in the lower half of Figure 8.9, each slightly distorting the original cubic cell. Every
time the symmetry changes, the domain structure also changes, and this explains the
permittivity maximum at each phase transition (Figure 8.8). Interestingly, probes of local
structure, such as pair distribution function (PDF) analysis and X-ray absorption fine
structure (EXAFS), reveal that the local structure deviates from the average crystallographic
structure. In all four modifications of BaTiO3, local displacements of titanium toward a face

local
structure

average
structure

(a) Tetragonal (b) Orthorhombic (c) Rhombohedral

Figure 8.9 The Ti-displacement disorder in BaTiO3 (top), and the resulting average structure (bottom).
The small spheres shown in the upper half of the figure represent disordered Ti positions along the body
diagonals of the high-temperature cubic unit cell. They are slightly exaggerated for illustrative purposes.
The degree of disorder gradually decreases as the symmetry changes from (a) tetragonal, to (b) ortho-
rhombic, to (c) rhombohedral. The darkly shaded bonds are the shortest average Ti–O distances. See ref.
[5] for more details.
 8.4 Pyroelectricity and Ferroelectricity 319

of its coordination octahedron are observed, giving three short and three long Ti–O bonds
[5]. In other words, the local structure around each Ti looks very much like the low-
temperature rhombohedral structure shown in Figure 8.9c, even at temperatures where the
average structure is tetragonal, orthorhombic, or cubic.

8.4.2 Antiferroelectricity
Not all compounds that have local dipole moments exhibit a spontaneous polarization.
Antiferroelectric materials possess local dipoles, but their arrangement is such that for every
dipole there is an adjacent dipole that is oriented in the opposite direction (antiparallel).
Because the dipoles cancel, the antiferroelectric state is nonpolar and does not exhibit
a hysteresis loop. Nevertheless, the local dipoles still become dynamic above a certain
temperature, called the antiferroelectric Curie temperature, where the material crosses
over to a paraelectric state.
Subtle changes in bonding can trigger a crossover between ferroelectric and antiferro-
electric behavior. For example, the ferroelectric perovskites PbTiO3 (TC = 763 K) and
KNbO3 (TC = 707 K) are closely related to antiferroelectric PbZrO3 (TC = 606 K) and
NaNbO3 (TC = 911 K). In these materials, changes in the perovskite tolerance factor (Section
1.5.3) modify the long-range coupling of the local distortions sufficiently to alter the
competition between competing ferroelectric and antiferroelectric ground states. The ferro-
electric KH2PO4 (TC = 123 K) and antiferroelectric (NH4)H2PO4 (TC = 148 K) are another
such pair; in this instance differences in hydrogen bonding are responsible for their different
dielectric properties.
Despite the cancellation of dipoles, antiferroelectrics can exhibit high dielectric permittiv-
ity, especially near the Curie temperature. For example, the relative permittivity of PbZrO3 is
~3000 just a few degrees below its Curie temperature. Because the bonding interactions that
differentiate ferroelectrics from antiferroelectrics can be subtle, it is sometimes possible to
drive a reversible transition from the antiferroelectric state to a ferroelectric state with an
applied field, as shown in Figure 8.10. At low fields, the response is linear, but at higher fields
a hysteresis loop is observed, as in PbZrO3.

P P P

E E E

ferroelectric paraelectric antiferroelectric

Figure 8.10 Polarization as a function of electric field for a ferroelectric, a paraelectric, and an antiferro-
electric material that reversibly transforms to a ferroelectric state when a sufficiently large field is applied.
320 Dielectrics and Nonlinear Optical Materials

Box 8.2 Nanoscale Concepts: Relaxor ferroelectrics

Relaxor ferroelectrics, often referred to simply as relaxors, are frustrated ferroelectrics, typically via
substitution-induced structural disorder. Many relaxors are perovskites with the generic formula
A(M0 1−xM00 x)O3 where some degree of disorder exists among the octahedral-site metals, M0 and
M 00 . In such compounds it is common that a lone-pair cation, often Pb2+, resides on the A-site and
a high-valent d 0 transition metal occupies the M 00 site. Importantly, both are prone to off-center
displacements that create local dipole moments, as discussed in Section 8.6.
The properties of relaxors and normal ferroelectrics are similar, but the cooperative phase
transitions that lead to the formation of large, spontaneously polarized domains are suppressed
in relaxors. Like normal ferroelectrics, the dielectric permittivity of a relaxor goes through
a maximum as a function of temperature. Unlike normal ferroelectrics where the permittivity
peaks sharply near TC and shows little frequency dependence, relaxors exhibit a broad and
frequency-dependent εr peak, as shown below for one of the most important relaxors, Pb(Mg1/3
Nb2/3)O3 (at alternating-current frequencies of 10−2, 100, 102, 104, and 105 Hz) [6]. The lack of
an abrupt change in crystal structure upon cooling through the permittivity maximum is
another feature of relaxors that distinguishes them from normal ferroelectrics. Instead, the
polar distortions are confined to nanoscale islands called polar nanoregions (PNRs). Whereas in
normal ferroelectrics such regions would grow and coalesce upon cooling into the ferroelectric
state, in relaxor ferroelectrics the sizes of the PNRs change with temperature yet remain
separated from each other by a nonpolar matrix. Although relaxors do not undergo coopera-
tive phase transitions, above the Burns temperature the PNRs disappear, and the relaxor enters
a paraelectric state. The Burns temperature of Pb(Mg1/3Nb2/3)O3 is ~630 K.

25000

20000

15000
ν = 10−2 Hz
εr
10000
ν = 105 Hz

5000

0
150 200 250 300 350 400

T (K)

Each PNR possesses a large dipole moment that does not strongly couple to other PNRs.
 8.5 Piezoelectricity 321

Box 8.2 (cont.)

The presence of large, uncoupled, dipole moments can produce very large dielectric
permittivities, as high as 20000 to 35000 in some relaxors. This makes them attractive for
applications in multilayer capacitors. The large permittivities also lead to large electrostric-
tion effects (changes in volume in response to an applied electric field). Furthermore, the
absence of a conventional domain structure allows for a fast electrostrictive response to
applied fields that makes relaxors ideal for micropositioners in optical devices, low-
frequency transducers in sonar systems, and high-frequency transducers in biomedical
devices.

8.5 Piezoelectricity
When a suitably oriented mechanical stress is applied to a polar crystal, charges will develop
on opposite faces of the crystal. This phenomenon is known as the piezoelectric effect. When
a suitably oriented electric field is applied to a polar crystal, it will change the dimensions of
the crystal, a phenomenon known as the converse piezoelectric effect. Both effects are
illustrated schematically in Figure 8.11.
The piezoelectric effect originates from distortions of coordination polyhedra that create
or modify local dipoles. In polar materials like tetragonal BaTiO3, compression along certain
directions amplifies the polarization while in other directions compression diminishes the
polarization, as shown in Figure 8.12. Application of an external stress can also induce a net
polarization in a nonpolar crystal if its point group belongs to one of the piezoelectric crystal
classes listed in Table 8.3. To see how a nonpolar entity with no inversion center can develop
a dipole moment, consider the nonpolar trigonal-planar coordination polyhedron. As illus-
trated in Figure 8.12c, the appropriate external stresses can deform the equilateral triangle in
a manner that creates a local dipole.
Given the geometric considerations discussed in the preceding paragraph, it should not
come as a surprise that many piezoelectrics are either ferroelectric materials (all ferroelectrics
are piezoelectric, but the converse is not true) or structures containing nonpolar building
units that lack an inversion center, like trigonal planes or tetrahedra. The applications of
piezoelectrics are numerous and amongst the most important of the materials encountered in
this chapter. One of the oldest is in sonar, where piezoelectric transducers convert pressure
changes associated with sound waves into electrical signals. Piezoelectrics are also used to
generate and detect ultrasonic pulses used in medical imaging technologies like ultrasound.
Because electrical charges build up when a pressure is applied to a piezoelectric crystal, they
can be used to create sparks that ignite flammable gases. On the other hand, the converse
322 Dielectrics and Nonlinear Optical Materials

compressive
stress

– – – – –

+ + + + +

Piezoelectric effect

Converse piezoelectric effect

Figure 8.11 The piezoelectric effect where compressive stress leads to polarization of the crystal (top), and
the converse piezoelectric effect where an applied electric field changes the shape of the crystal (bottom).

δ−
c c

δ+

(a) Polarization increases (b) Polarization decreases (c) Dipole created

Figure 8.12 Piezoelectric materials: changes in local polarization in response to stress. (a) A compression
in the ab plane of tetragonal BaTiO3 amplifies the local polarization, (b) a compression along c diminishes
it. (c) External stress induces local polarization (fractional charges δ) in a trigonal-planar polyhedron.

piezoelectric effect is used in devices to reproducibly control minute movements by applying

a voltage, for example to position the tip of an atomic force microscope.
 8.5 Piezoelectricity 323

800
cubic
orthorhombic Pm3m
700 Pbam paraelectric
antiferroelectric

600 tetragonal
P4mm
ferroelectric
T (K)

rhombohedral I MPB
500
R3m
ferroelectric
400 monoclinic
Cm and/or Pm
ferroelectric
300 rhombohedral II
R3c
ferroelectric
200
0 0.2 0.4 0.6 0.8 1

x 3
PbZrO3 PbTiO3

Figure 8.13 The PbZr1−xTixO3 phase diagram.

While tetrahedral-network solids find use in some applications, such as quartz oscillators for
clocks, ferroelectric perovskite oxides dominate the commercial market for piezoelectrics. The
most widely used piezoelectrics are based on the Pb(Zr1−xTix)O3 (PZT) system, the phase
diagram of which is shown in Figure 8.13. Pure PbZrO3 becomes antiferroelectric below 506 K.
Its structure is distorted from cubic to orthorhombic symmetry by antiparallel displacements of
Pb and rotations of the zirconium-centered octahedra (a0b−b− tilting, Section 1.5.3). Upon
introducing a small fraction of Ti at the Zr site, the paraelectric cubic state transforms18 to
a ferroelectric state upon cooling [7]. This phase, labeled rhombohedral I in Figure 8.13,
possesses a slight rhombohedral distortion of the cubic cell and space-group symmetry R3m.
In this structure, Pb2+ as well as (Ti/Zr)4+ displace along one of the eight ⟨ 111 ⟩ directions of
the parent cubic cell. A second rhombohedral phase, labeled rhombohedral II in Figure 8.13,
appears at low temperatures. This phase of R3c symmetry features a−a−a− octahedral tilting, in
addition to the polar displacements of cations that were already present in the high-
temperature R3m phase. For Ti-rich (x > 0.5) compositions, Pb(Zr1−xTix)O3 is isostructural
with the average structure of tetragonal BaTiO3 (Figure 8.5) [8].19
Near the PbZr0.5Ti0.5O3 composition, the ferroelectric solid solution changes sym-
metry from rhombohedral on the Zr-rich side to tetragonal on the Ti-rich side. The solid
line separating these two regions in Figure 8.13 is called the morphotropic phase

18
This transformation is sluggish and can have a large thermal hysteresis.
19
The Pb2+ displacements in PbTiO3 influence the Ti4+ displacements in such a way that both the local and average
structure of tetragonal PbTiO3 can be described by displacements along one of the equivalent ⟨ 001 ⟩ directions,
unlike BaTiO3 where the local displacements are along ⟨ 111 ⟩ .
324 Dielectrics and Nonlinear Optical Materials

boundary (MPB). An abrupt structural change upon changing the composition of

a solid solution is the defining characteristic of an MPB. For solid-solution compos-
itions close to the MPB, a polarization rotation occurs where the polar vector
defining each domain is easily reoriented by an electric field, leading to a very strong
piezoelectric response.
The structure of Pb(Zr1−xTix)O3 becomes quite complicated near the MPB, with
some details still being debated. In the late 1990s, the 0.48 ≤ x ≤ 0.50 region, which
had previously been thought to be a two-phase mixture of rhombohedral and
tetragonal phases, was shown to be more accurately described as a monoclinic
phase [9]. A more recent study provides evidence to suggest there may be two
different phases with monoclinic symmetry in the MPB region [10]. Upon increasing
the Zr content, the average structure becomes rhombohedral, but vestiges of the
monoclinic structure can still be found in the local structure. The subtle energetic
differences between competing phases gives rise to a complex fine-scale domain
structure that is at least partially responsible for the exceptional piezoelectric proper-
ties of PZT.
In recent years, environmental concerns over the detrimental impacts of lead have
motivated a search for Pb-free piezoelectrics. Perovskites such as (K 1−xNax)NbO3,
Ba(Zr1−xTix)O3, and (Ba1−xCax)TiO3, have emerged as possible replacement candi-
dates. Nevertheless, piezoelectrics with properties that match those of PZT remain
elusive.

8.6 Local Bonding Considerations in Non-Centrosymmetric Materials

As a general rule, non-centrosymmetric structures are more likely to form when their
coordination polyhedra lack an inversion center. Apart from polyhedra that are inherently
non-centrosymmetric (triangles, tetrahedra), polyhedra that would otherwise possess an
inversion center can become non-centrosymmetric by distorting. For example, when an
octahedrally coordinated cation undergoes a second-order Jahn–Teller (SOJT) distortion
as in the ferroelectric phases of BaTiO3.
Halasyamani and Poeppelmeier [11] surveyed the ICSD and found ~580 oxides that
adopt non-centrosymmetric crystal structures. Among these, two-thirds belonged to
polar crystal classes, while the remaining third were nonpolar. Roughly 75% of all
entries contain one or more of the following coordination environments: (a) a d 0
cation that undergoes an SOJT distortion (e.g. Ti4+, Nb5+), (b) a p-block cation with
an s2p0 electron configuration that undergoes an SOJT distortion leading to
a stereochemically active lone-pair distortion (e.g. Pb2+, Sb3+), or (c) a tetrahedrally
coordinated cation. In the following section, we take a closer look at the bonding that
drives SOJT distortions.
 8.6 Local Bonding Considerations in Non-Centrosymmetric Materials 325

8.6.1 Second-Order Jahn–Teller Distortions with d 0 Cations

SOJT distortions were introduced in Section 5.3.10. In extended solids, the most common
SOJT distortions are those involving a transition-metal cation with a d 0 configuration and
those involving a main-group cation with an s2p0 configuration. We consider d 0 cations in
this section and s2p0 cations in the next.
The molecular-orbital (MO) diagram for an octahedrally coordinated transition-metal
atom was considered in detail in Section 5.3.8. When the transition metal has a d 0 configur-
ation, the empty triply degenerate set of MOs with t2g symmetry is the LUMO (lowest
unoccupied molecular orbital), while the HOMO (highest occupied molecular orbital) has
nonbonding anion character (see Figures 5.21 and 5.22). SOJT distortions are characterized
by a shift of the cation out of the center of the octahedron that lowers the symmetry, enabling
the unoccupied t2g orbitals to mix with filled anion states, as described below. A thorough
treatment of SOJT distortions involving d 0 cations in both molecular and extended solids
can be found in the literature [12].
Because perovskite oxides account for a large fraction of ferroelectric materials, let’s
take a closer look at the bonding that drives SOJT distortions in these materials. Their
electronic band structure was covered in detail in Sections 6.6.4 and 6.6.5. For a cubic
perovskite with a d 0 cation on the octahedral site (e.g. SrTiO3 or KTaO3), the conduction-
band minimum occurs at the Γ point (see Figure 6.31). At Γ, the orbital character of six
out of the nine valence bands becomes strictly oxygen 2p nonbonding, while the three
lowest-energy conduction bands are strictly metal d nonbonding orbitals (dxy, dyz, dxz) as
illustrated in Figure 6.30. The key point is that mixing (overlap) of the metal d orbitals and
the oxygen 2p orbitals to form π-bonding and -antibonding states is symmetry-forbidden
at the Γ point.
Now consider how the band structure changes in response to a SOJT distortion. The
essential features of this analysis are captured by considering the π interactions and
the displacements of the M cation within the xy plane towards the edge of the octahedron.
The crystal orbitals at Γ representing the empty dxy band and one of the filled O 2p bands
with t1g symmetry are shown in the upper half of Figure 8.14. These are representative of the
crystal orbitals from conduction and valence bands, respectively, separated by the charge
transfer gap CT in Figure 6.31.
If we lower the symmetry by moving the M cations along [110], the symmetry
constraints are relaxed, and mixing (overlap) between the two formerly nonbonding
crystal orbitals is allowed. Because each oxygen ion now makes one short and one
long bond to the neighboring metal atoms, the bonding and antibonding contributions
do not cancel out. This introduces π-bonding interactions that stabilize the occupied
nonbonding O 2p states (see lower middle panel of Figure 8.14), while π*-antibonding
interactions destabilize the empty metal d states (see lower right panel of Figure 8.14).
The amount of mixing increases as the energy gap (prior to the distortion) between the
O 2p (t1u) and M nd (t2g) crystal orbitals decreases. Because the destabilization of
326 Dielectrics and Nonlinear Optical Materials

cubic O 2p t1g band (at Γ ) M dxy band (at Γ)

orthorhombic O 2p−M dxy π (at Γ ) M dxy−O 2p π* (at Γ )

Figure 8.14 Top: Metal–oxygen interactions within an MO2 plane of a perovskite (left), and crystal
orbitals at Γ for the valence-band maximum (middle) and conduction-band minimum (right). Bottom:
Changes that occur when the octahedral cation displaces within the plane toward an edge of the
octahedron.

the conduction-band states is larger than the stabilization of the valence-band states, the
SOJT distortion quickly becomes unfavorable as the d orbitals are populated. This is
seen in the crystal chemistry of MO3 compositions with ReO3 topologies. SOJT distor-
tions lead to large cation displacements in the d 0 oxides WO3 and β-MoO3, whereas the
d 1 oxide ReO3 is cubic.
A survey of AMO3 perovskites shows that not all compounds that contain a d 0 cation
undergo SOJT distortions. SOJT distortions are observed for KNbO3, NaNbO3, and
BaTiO3, but not for KTaO3, NaTaO3, SrTiO3, CaTiO3, AZrO3, or AHfO3 (A is Ba, Sr,
Ca). This can be understood by realizing that the driving force for the SOJT distortion
increases as the energy separation between the metal nd orbitals and the nonbonding oxygen
2p states decreases.20 This energy separation reaches a minimum at the Γ point (Section
6.6.5). SOJT distortions are therefore seen in KNbO3, where the narrow charge-transfer gap
facilitates mixing of M nd and O 2p orbitals on distortion, but not in BaZrO3 where the gap is
larger. SOJT distortions occur in ferroelectric KNbO3 and antiferroelectric NaNbO3, but
not in KTaO3 and NaTaO3 for the same reason. Using a variety of approaches, several
different studies have reached similar conclusions regarding the tendency for various d 0
“cations” to undergo SOJT distortions [13, 14, 15, 16]: Hf4+ < Zr4+ < Ta5+ < Ti4+ < Nb5+ <

20
In Figure 6.31, the HOMO–LUMO gap at Γ is labeled a charge-transfer excitation: the smallest direct band gap in
a cubic perovskite that allows electromagnetic radiation to excite electrons from nonbonding anion crystal orbitals
to empty d orbitals of the central metal atom.
 8.6 Local Bonding Considerations in Non-Centrosymmetric Materials 327

W6+ < V5+ < Mo6+. Notice the similarity of elements that are diagonal neighbors on the
periodic table (e.g. Ti4+, Nb5+, W6+).
The charge-transfer gaps of most A2+TiO3 and A+TaO3 perovskites make them nearly
ferroelectric; close to the border between ferroelectric and paraelectric behavior (except
BaTiO3 and PbTiO3, which are ferroelectric). Although SrTiO3 and KTaO3 are cubic and
therefore nonpolar at room temperature, they have dielectric permittivities much higher than
predicted by the Clausius–Mossotti equation (330 and 242 respectively, see Table 8.1). These
permittivities increase further with decreasing temperature as though approaching
a transition into a ferroelectric state, but the transition is never realized. Consequently,
these phases are referred to as incipient ferroelectrics. The close proximity of the ferroelectric
state means that only a small perturbation is needed to stabilize ferroelectricity. In the case of
SrTiO3, ferroelectricity can be realized by stretching the lattice, which occurs when Sr2+ ions
are replaced with the larger Ba2+ ions to form BaTiO3.21 In a similar vein, the dielectric
properties of thin epitaxial SrTiO3 films are very sensitive to stresses that result from lattice
mismatch with the underlying substrate. For example, the dielectric permittivity of SrTiO3
films grown on DyScO3 substrates, which impose a 1% in-plane tensile strain, is increased
nearly 20 times compared with bulk SrTiO3 [17]. It has even been shown that isotopic
substitution of 18O for 16O can stabilize ferroelectricity in SrTiO3 below 23 K [18].

8.6.2 Second-Order Jahn–Teller Distortions with s 2p0 Cations

Cations that have an s2p0 configuration are also prone to distortions. These distortions are
called lone-pair distortions, or more accurately, stereochemically active lone-pair distor-
tions, because their presence is normally inferred from the distortion of the ligands sur-
rounding the s2p0 central atom. To understand the driving forces behind these distortions, we
again turn to MO theory to show that “lone-pair distortions” are just another type of SOJT
distortion as discussed previously in Section 5.3.10.
Consider a main-group cation located on a site with m3m (Oh) symmetry. This includes
three important coordination environments found in extended solids: the octahedron, the
cube, and the cuboctahedron. Once again, the perovskite structure provides us with good
examples, as both the smaller six-coordinated cation in the octahedron and the larger 12-
coordinated cation in the cuboctahedron have m3m site symmetry.
Figure 8.15 shows two perovskites where an s2p0 cation drives a SOJT distortion: CsGeCl3
and BiFeO3. The distortion is driven by Ge2+ on the six-coordinate site in the former, and by
Bi3+ on the 12-coordinate site in the latter. In both, the lone-pair cation displaces along

21
According to the distortion theorem (Section 5.4), off-center shifts of an atom within its coordination polyhedron
increase the atom’s bond-valence sum. In those AMO3 perovskites where the coordination octahedron defined by
AO3 packing becomes too large (tolerance factor > 1), the M of d 0 configuration and high oxidation state becomes
underbonded. The SOJT distortion leads to an increase in the M-atom bond-valence sum, because bonds that
shorten gain more bond valence than bonds that lengthen lose.
328 Dielectrics and Nonlinear Optical Materials

Cl Fe

Ge
O
Bi

Cs

bond distances (Å) bond distances (Å)

Ge–Cl 2.348 (x 3) Fe–O 1.951 (x 3)
3.092 (x 3) 2.113 ( x3)
Cs–Cl 3.842 (x 3) Bi–O 2.273 (x 3)
3.844 (x 6) 2.526 (x 3)
3.883 (x 3) 3.216 (x 3)
3.446 (x 3)

Figure 8.15 The crystal structures of the rhombohedrally distorted perovskites CsGeCl3 (left) and
BiFeO3 (right). Only the shortest Ge–Cl and Bi–O bonds are shown.

a threefold axis, lowering the site symmetry to 3m and the crystal symmetry to
rhombohedral.
Consider the MO diagram for an isolated GeCl64− octahedron (Figure 8.16). Because the
Ge–Cl π overlap is minimal, we can concentrate on the interactions between the six Cl− σ
donors and the 4s and 4p valence orbitals of germanium. In a perfect octahedron, these
orbitals are orthogonal and do not mix. The HOMO is the strongly antibonding interaction
between the Ge 4s orbital and the ligand symmetry-adapted linear combination (SALC) with
a1g symmetry. The LUMO is the triply degenerate t1u set of orbitals formed from antibond-
ing Ge 4p–Cl interactions. The relatively small HOMO–LUMO gap provides the necessary
driving force for an SOJT distortion.
Figure 8.17 shows the changes to the frontier orbitals (HOMO and LUMO) that result
from a shift of Ge2+ along the along threefold axis toward the face of the coordination
octahedron, as seen in CsGeCl3. In the distorted octahedron with 3m point-group symmetry,
the Ge 4s and 4pz orbitals both have a1 symmetry and can mix with each other, lowering the
energy of the HOMO. In this way, the electron density of the spherically symmetric a1g
orbital of the undistorted octahedron is redistributed to a lobe located on the “more open
side” of the Ge ion, forming the stereochemically active electron lone pair. At the same time,
the energy of the Ge 4pz–Cl σ* MO is raised by the s–p interaction, but, because this orbital is
not occupied, there is no energy penalty for this destabilization. Similar symmetry arguments
can be made for Bi3+ that sits on the 12-coordinate site in BiFeO3.
In oxides where the metal M comes from the fourth period, the M 4s–O 2p interaction is
very strong. Consequently, the M 4s–O 2p σ* level is so antibonding that it is difficult to
stabilize the 4s24p0 configuration in any geometry. Hence, Ga+ and Ge2+ are rarely observed
 8.6 Local Bonding Considerations in Non-Centrosymmetric Materials 329

2t1u σ*
Cl 4−
Cl
Cl Ge Cl
2a1g σ*
Cl
Cl

eg
Ge 4p
eg (nonbonding)
t1u
a1g

Cl 3p σ
SALCs
1tu σ Ge 4s

1a1g σ

Figure 8.16 The MO diagram for an isolated GeCl64− octahedron. Only the valence-shell Ge 4s and
4p orbitals, and the Cl σ-donor orbitals, are considered.

a1 σ*
energy

t1u σ* e σ*

a1g σ*
a1 “nonbonding
lone pair”

Figure 8.17 HOMO and LUMO of an s2p0 cation in a regular octahedral coordination (left) and their
response when the s2p0 cation shifts toward a face of the octahedron (right) as in CsGeCl3.

in oxides and fluorides. In contrast, many oxides of Sn2+, Sb3+, and Te4+ exist because the
M 5s–O 2p interaction that gives rise to the 2a1g σ* MO is not as strongly antibonding.
However, the interaction is still strong enough that a pronounced stereochemically active
electron lone-pair distortion is common. While such distortions produce local dipole
330 Dielectrics and Nonlinear Optical Materials

moments, the energy required to invert the orientation of the stereochemically active lone
pair is often quite large. Hence, many oxides containing Sn2+, Sb3+, and Te4+ are pyroelectric
but few are ferroelectric.
Moving to the sixth period, the spatial and energetic overlap between metal 6s and
O 2p orbitals is significantly diminished, due to the contraction of the 6s orbital driven by
relativistic effects (Section 5.2.2). Consequently, sixth-period cations like Tl+, Pb2+, and Bi3+
show a wide variety of coordination environments. In some compounds (e.g. PbWO4,
BiVO4) the lone pair is not stereochemically active and a symmetric coordination environ-
ment results. In those cases, long cation–oxygen bonds minimize the M 6s–O 2p overlap and
the 6s2 electron pair remains essentially the nonbonding, non-stereochemical, inert pair of
the isolated atom, having its original spherical symmetry. In other compounds, like PbTiO3
and PbZrO3, the SOJT distortion is active [19]. The occurrence of the SOJT distortion and
the accompanying stereochemical activity depends on the details of the cation–anion inter-
action. Brown [20] has shown that the local environment of Tl+ can vary from completely
symmetric to highly distorted when in the presence of anions that require strong, short,
bonds with Tl+. When a SOJT distortion of a sixth-period cation does occur, the direction of
the displacement can often be inverted by an applied field, hence these cations, particularly
Pb2+, are often found in ferroelectrics.

8.7 Nonlinear Optical Materials

When electromagnetic radiation travels through a dielectric material, the electric-field
component of the radiation induces an oscillating polarization of the charged species in
the material. This slows the wave’s velocity as it passes through the material, as though the
oscillating charges emit their own electromagnetic wave at the same frequency as
the propagating electromagnetic radiation, but with a phase delay. The ratio between the
speed of light, c, in a vacuum and its velocity, v, inside the material is defined as its refractive
index, n = c/v.22 For ultraviolet (UV), visible, and near-IR radiation, only electron clouds can
respond to the high-frequency electric field of the propagating light wave.
The phenomenon whereby light is slowed as it passes through a material is a linear optical
effect. Much weaker nonlinear optical (NLO) effects can occur, where the propagating
electromagnetic wave and the light emitted by the material have a different frequency.
Because NLO effects are weak, they are only observable with intense light sources such as
lasers. A full treatment of NLO effects and applications is beyond the scope of this text.
Instead, we will concentrate on what is arguably the most important NLO effect, second-
harmonic generation (SHG).
SHG is a process where two photons with frequency ν combine to produce a new
photon with twice the frequency, 2ν. Using relationships introduced in Section 7.1, it

22
Be careful not to confuse the italic v for velocity with the Greek letter ν for frequency.
 8.8 Nonlinear Susceptibility and Phase Matching 331

Green laser pointer

808 nm 1064 nm 1064 nm, 532 nm
light light 532 nm light

Diode laser Nd-doped KTiOPO4 (KTP) IR filter

YVO4 crystal SHG crystal

Frequency-tripled UV laser
1064 nm 1064 nm, 1064 nm, 532 nm,
light 532 nm 355 nm

Nd-doped LiB3O5 (LBO) LiB3O5 (LBO)

Y3Al5O12 (YAG) SHG crystal sum frequency
laser mixing crystal

Figure 8.18 The main optical components of a green laser pointer (top) and a laser that uses two NLO
crystals to generate UV light (bottom).

can be shown that the new photon has twice the energy and half the wavelength of
the incoming photons. Perhaps the most familiar example is the green laser pointer
(Figure 8.18), which uses a (Y1−xNdx)VO4 laser crystal to emit IR radiation with
a wavelength, λ = 1062 nm, that is subsequently halved to 532 nm by a KTiOPO4
SHG crystal.23 SHG crystals are used in many types of lasers to convert IR light to
visible and/or UV light.

8.8 Nonlinear Susceptibility and Phase Matching

When polarization was introduced in Section 8.1.2, nonlinear effects were neglected. We can
expand Equation (8.6) to take nonlinear effects into account:
 
P ¼ ε0 χð1Þ
e E þ χ ð2Þ 2
e E þ χ ð3Þ 3
e E þ … (8.20)

23
One drawback of this green laser pointer design is the low SHG efficiency associated with the low power levels
employed in a laser pointer. Hundreds of milliwatts of IR light are required for generating the standardized 1 mW of
green light. Accordingly, battery life is relatively short and filters must be used to take out the IR light that poses
a serious safety hazard to the eye.
332 Dielectrics and Nonlinear Optical Materials

ð1Þ ð2Þ ð3Þ

where χe is the linear electric susceptibility and the higher-order terms χe , χe , . . . are the
nonlinear electric susceptibilities of the second, third, and higher orders, respectively.
Because each successive term in Equation (8.20) is much smaller than the preceding one,
ð2Þ ð2Þ ð4Þ
we neglect terms past the second-order χe . The even terms (χe , χe , . . .) are zero for
centrosymmetric crystals, so only non-centrosymmetric crystals are capable of SHG activity,
as discussed in Section 8.3. Although frequency tripling is allowed in centrosymmetric
crystals, the effect is generally so small that it is not of practical importance. Instead,
frequency tripling is usually achieved by combining frequency doubled light with the primary
beam in a second crystal, where sum frequency generation occurs, as schematically illus-
trated in Figure 8.18.
Each of the susceptibility terms in Equation (8.20) is a tensor, a proportionality constant
between properties described by a matrix or a vector. The second-order electric susceptibility
ð2Þ
χe has 27 individual χijk terms. For practical purposes, these are often converted into the so-
called NLO coefficients, dijk. For applications, it is desirable that the NLO coefficients be
relatively large, but, as we will see, this is not the only important materials consideration.
NLO coefficients are often reported relative to potassium dihydrogen phosphate, KH2PO4
(KDP), whose d321 = 0.44 pm/V is the standard coefficient against which other NLO
materials are measured.24
The refractive index of a material is not a constant; it depends on both the frequency of the
electromagnetic radiation and the temperature of the material. The frequency dependence is
important for SHG because in general the fundamental beam of frequency ν and the second-
harmonic beam of frequency 2ν travel at different speeds, which leads to destructive inter-
ference that can dramatically reduce SHG efficiency. Fortunately, it is possible to match the
refractive indices of the fundamental and second-harmonic beam in birefringent crystals
through a technique known as phase matching.
To understand phase matching, one must first be familiar with the optical properties of
birefringent crystals. In such crystals, the refractive index is not isotropic, it varies with the
direction and polarization of the light beam in the crystal. In a birefringent crystal that is
uniaxial,25 the component of light that is polarized in the direction of the unique axis is called
the ordinary beam, while the component polarized orthogonal to the unique axis is called the
extraordinary beam. The ordinary beam will experience the same refractive index (no)
regardless of its direction of propagation, whereas the extraordinary beam will experience
a refractive index that depends upon the direction it travels through the crystal (ne). In every

24
The electric field and the polarization response of a crystal are both vectors, and the tensor describing the
relationship between the two is a second rank tensor with nine coefficients. Because the SHG process ν + ν → 2ν
involves three photons, 27 dijk terms are needed for a general description of second-order NLO effects. When the
susceptibilities are independent of frequency, the number of coefficients can be reduced from 27 to 18 terms. For
SHG effects, it’s common to use a condensed notation because the j and k terms can be permuted, for example, d36 ≡
d321 = d312. For more details, see R. Boyd in Further Reading.
25
Crystals with a unique axis of symmetry, namely those that belong to the trigonal, tetragonal, or hexagonal crystal
systems.
 8.8 Nonlinear Susceptibility and Phase Matching 333

beam
nǁ nǁ direction nǁ
ne(ν) no(ν) ne(2ν)
θ

n⊥ n⊥ n⊥

no(2ν)

Refractive index of Critical phase Refractive index of second-

fundamental beam matching n(ν) = n(2ν) harmonic beam

Figure 8.19 Cross-sectional cuts of the optical indicatrix showing the refractive indices of the fundamen-
tal (left) and second-harmonic (right) beams in a birefringent crystal. The vertical and horizontal axes
represent propagation parallel and orthogonal to the optic axis, respectively. The circle of the refractive
index of the ordinary beam, no, is solid, the ellipse of the extraordinary beam, ne, is dashed. The middle
panel shows the critical angle, θ, of the propagation direction for which the phase matching occurs.

birefringent crystal, there is at least one direction where no = ne, and that direction is defined
as the optic axis. In uniaxial crystals, the optic axis is parallel with the unique crystallo-
graphic axis, whereas biaxial crystals have two optic axes. Double refraction, a process
whereby an electromagnetic wave is split into two rays that take slightly different paths
because of differences in the refractive indices of the ordinary and extraordinary beams, is
a familiar property of birefringent crystals.26
The variation of the refractive index as a function of the light-beam direction in a crystal is
represented by an ellipsoid, called an optical indicatrix. The shape of the ellipsoid is such that
its radius along any given direction is directly proportional to the refractive index in that
direction. Because the refractive index of the ordinary beam is isotropic, its optical indicatrix
is a sphere. When the beam direction is parallel with the optic axis, the ordinary and
extraordinary beams have the same refractive index, as shown on the left- and right-hand
sides of Figure 8.19.
The most common method of phase matching, critical phase matching, relies on the optical
properties of birefringent crystals. The incoming fundamental beam with frequency ν is
polarized in a plane parallel to the optic axis of the SHG crystal; it acts as an ordinary beam
with an index of refraction that is independent of the direction it travels through the crystal.
When the second-harmonic beam is generated, its polarization is perpendicular to the
fundamental beam and thus it travels through the crystal as an extraordinary beam.
Hence, the refractive index of the second-harmonic beam can be tuned by changing the
orientation of the propagating beam with respect to the optic axis of the crystal. There is
a critical angle, θ, where no(ν) = ne(2ν), and critical phase matching occurs, as illustrated in

26
The calcite form of CaCO3 is one of the most familiar examples of a birefringent crystal. The only orientation when
it does not split light into two beams is when the light travels parallel to the optic axis.
334 Dielectrics and Nonlinear Optical Materials

the center panel of Figure 8.19. One downside of critical phase matching is that the strength
of the SHG signal is highly sensitive to misalignment of the beam from θ.
If the angle at which phase matching occurs corresponds to one of the principal axes of the
birefringent crystal, for example the a axis of a tetragonal crystal, the SHG efficiency becomes
less sensitive to small deviations from the critical angle. The odds that this random coinci-
dence will occur can be improved by changing the temperature of the crystal, because the
refractive index of the extraordinary beam ne is typically more sensitive to changes in
temperature than that of the ordinary beam no. Hence the angle θ at which phase matching
occurs can sometimes be tuned by temperature to align with one of the principle axes of the
crystal. This method is called noncritical phase matching. The advantage of this approach is
decreased sensitivity to small misalignments of the crystal, but it requires precise control of
the temperature of the SHG crystal.
A third approach, called quasi-phase matching, is used in materials where the two
former approaches to phase matching are not practical. The idea is to allow a degree of
phase mismatch between the fundamental and second-harmonic beams (due to different
refractive indices experienced by the two beams), but before the phase mismatch and
destructive interference become too large, a specially engineered domain structure of the
crystal inverts the two beams. Every few microns, there is a 180° ferroelectric domain
wall that changes the refractive indices in such a way that the relative speeds of
the second-harmonic and fundamental beams are inverted. As the fundamental beam is
alternatively going faster and slower than the second-harmonic beam, the two never get
very far out of phase; hence phase matching.

8.9 Important SHG Materials

While symmetry restrictions narrow the scope of possible SHG materials, there are still
thousands of non-centrosymmetric materials from which to choose. What other criteria can
be used to select and/or design SHG materials? Becker [21] has proposed seven criteria for
selecting good SHG materials: (1) relatively large nonlinear optical coefficients, (2) moderate
birefringence, (3) wide transparency range, (4) wide phase-matching range, (5) high light-
induced damage threshold (hereafter referred to simply as damage threshold), (6) good
chemical and mechanical stability, and (7) ease of crystal growth.
Some of the most important inorganic SHG materials are listed in Table 8.4. They can
be divided into four different families: (a) KH2PO4 (KDP) and related compounds, (b)
KTiOPO4 (KTP) and related compounds, (c) niobates and tantalates, and (d) borates.
KDP was the first commercial SHG material, but it has largely been supplanted as new
materials have been discovered and commercialized. KTP and related materials are
attractive because of the ease with which very large, high-quality crystals can be
grown. The large NLO coefficients of the niobates and tantalates make them attractive
with low-power laser sources, but applications are limited by low damage thresholds and
Table 8.4 Properties of commercially important SHG materials.

KH 2PO 4 KTiOPO 4
(KDP) (KTP) LiB 3O 5 BiB 3O 6 β-BaB 2O 4 LiNbO 3

Space group I42d Pna2 1 Pna2 1 C2 R3c R3c

Point group 42m mm2 mm2 2 3m 3m
Melting point 526 K 1445 K 1107 K 999 K 1368 K 1526 K
Refractive indices* n o = 1.494 n x = 1.738 n x = 1.566 n 1 = 1.917 n o = 1.655 n o = 2.232
n e = 1.460 n y = 1.745 n y =1.590 n 2 = 1.757 n e = 1.542 n e = 2.156
n z = 1.830 n z = 1.606 n 3 = 1.784
Transparency range 200–1500 350–3500 160–2600 290–2500 190–3500 420–5200
(nm)
Effective NLO 1.0 8.4 2.7 ~9 5.3 40
coefficient†
Damage threshold 0.25 1.0 9 ** 5 0.3
(GW/cm2 )‡
Comments Crystals can be High NLO Highest Higher NLO Broad High NLO
grown from coefficient, damage coefficients transmission coefficients,
solution, inexpensive threshold than most and phase used in
hygroscopic compared to among borates matching electro-optic
borates commercial ranges, high applications
materials damage
threshold
*As measured at 1064 nm and room temperature. For uniaxial rhombohedral and tetragonal crystals, the refractive index of the ordinary beam, n o,
and the extraordinary beam, n e, are given. For biaxial orthorhombic and monoclinic crystals, three refractive indices are needed.† The effective
NLO coefficients are versus the d 321 ≡ d 36 NLO coefficient of KDP.‡ The damage thresholds are for a 10 ns pulse of 1064 nm radiation. **The
damage threshold for BiB 3O 6 is reported to be similar to LiB 3O 5.
336 Dielectrics and Nonlinear Optical Materials

lack of transparency in the UV. The borates are versatile SHG materials due to their
broad transparency range, high damage threshold, and ease of phase matching. The
structures and properties of the key members of each family are discussed in more detail
in the subsequent sections.

8.9.1 KH2PO4
Potassium dihydrogen phosphate, KH2PO4 (KDP), crystals were among the first to produce
useful levels of frequency-doubled light. KDP is tetragonal, and its anionic network, linked
by hydrogen bonds along c, is shown in Figure 8.20. Since two of four oxygens of the
phosphate anion form covalent bonds to hydrogen, their bonds to P become single bonds,
and the phosphorus atom shifts toward the remaining two oxygens to compensate for this
loss of bonding. This shift generates a net dipole moment parallel to the c axis (see Figure
8.20). The most attractive aspect of KDP as an NLO material is the possibility to obtain
large, high-quality crystals. However, KDP has several limitations, including a relatively
small NLO susceptibility, a low damage threshold, and hygroscopicity.

8.9.2 KTiOPO4
KTiOPO4 (KTP) has several advantages over KDP. Not only does KTP have an NLO
coefficient eight times larger than KDP, it has a higher damage threshold, and is not
hygroscopic. Good crystals of KTP can be grown either hydrothermally or from a flux.

z=0

O
c
z=¼ P
z=¼

H
z=½
polar H2PO4−

Figure 8.20 The structure of KH2PO4 viewed against the c axis. Covalent O–H bonds are solid, hydrogen
bonds are dashed. K+ omitted for clarity. An individual H2PO4− anion is shown rotated on the right. The
white arrow indicates the shift of the phosphorus parallel to c.
 8.9 Important SHG Materials 337

TiO6

b
PO4

b
K
spiraling chain of
a corner-connected
TiO6 octahedra

Figure 8.21 The structure of KTiOPO4. A closer view of one of the spiraling corner-connected TiO4+2/2
chains is shown on the right. The short Ti–O bonds (d < 1.75 Å) are drawn to indicate the displacements of Ti.

These factors make KTP one of the most popular SHG materials, particularly for low- and
medium-power lasers.
While KH2PO4 and KTiOPO4 are both phosphates, the structural origins of the non-
centrosymmetry are not the same. In KTiOPO4, the non-centrosymmetry arises from SOJT
distortions of Ti ions with a d 0 configuration, which yield larger NLO coefficients than in
KDP. The structure of KTiOPO4 can be described as spiraling chains of corner-shared Ti-
centered octahedra held together by phosphate groups. The K+ ions sit in the channels within
this framework. Four of six oxygens around each titanium come from the tetrahedral
phosphate groups, while the other two oxygens connect the octahedra into the spiral chains
seen in Figure 8.21.
A closer look at the bonding within the octahedron shows that the bonds are not
symmetric. Titanium forms four intermediate-length bonds (1.95–2.05 Å) with oxy-
gens of the surrounding phosphate groups. It completes its octahedron by forming
one short bond (1.71–1.74 Å) and one long bond (2.10–2.15 Å) to the bridging
oxygens that are not bonded to phosphorus. The net effect is a long–short . . . Ti–
O–Ti–O– . . . bond alternation along the spiraling chain (Figure 8.21), creating a polar
axis along the chain direction. Other members of the KTP family include KTiOAsO4
(KTA), RbTiOPO4 (RTP), and RbTiOAsO4 (RTA). By replacing phosphate with
arsenate, NLO coefficients increase. More importantly, the long-wavelength transpar-
ency increases from 3500 nm to 5000 nm.
338 Dielectrics and Nonlinear Optical Materials

8.9.3 Niobates and Tantalates

Niobates and tantalates are widely used in optical communication networks as waveguides
and electro-optic modulators. Hence, the optical properties and methods for growing high-
quality crystals have been extensively studied. KNbO3 is a perovskite ferroelectric with
structures and phase transitions analogous to BaTiO3. It has large NLO coefficients, which
makes it attractive for low-power lasers. Between 263 K and 498 K, KNbO3 adopts an
orthorhombically distorted perovskite structure with Amm2 space-group symmetry. To attain
phase matching, periodic poling of the ferroelectric domains is necessary (quasi-phase match-
ing, see Section 8.8). Phase transitions into either the rhombohedral structure (T < 263 K) or
the tetragonal structure (T > 498 K) limit the range over which the crystal can operate. If the
temperature drifts outside of these limits, the periodically poled domain structure is lost.
LiNbO3 is the most important and widely used material in this family. Like KNbO3, it has
large NLO coefficients and relies on quasi-phase matching. However, the phase transitions
that complicate growth and use of KNbO3 crystals are not an issue because LiNbO3 has no
phase transitions below its Curie temperature (TC = 1483 K). While the robust polar state in
LiNbO3 prevents the switching that would be required for ferroelectric applications, this
aspect of its crystal chemistry is an advantage for NLO applications. LiNbO3 is also used as
a pyroelectric and a piezoelectric.
The structure of LiNbO3 (Figure 8.22), and the isostructural LiTaO3, can be described as an
ordered variant of the corundum structure (Figure 1.30). In Al2O3 and other oxides with the
corundum structure, electrostatic repulsions lead to a displacement of the cations away from the
shared octahedral face. In LiNbO3, the cations are ordered so that each pair of face-sharing
octahedra contains one lithium and one niobium. This destroys the inversion center in the

Nb

O
Li

face-sharing pair of Li- and

Nb-centered octahedra

Figure 8.22 Left: The unit cell of LiNbO3 with Nb-centered octahedra shaded. Right: A closer view at the
bonding within each pair of face-sharing octahedra.
 8.9 Important SHG Materials 339

structure, lowering the space-group symmetry from R3c to R3c. Just like in corundum, the
central atoms (Li+ and Nb5+) repel each other leading to highly distorted trigonal-antiprismatic
environments. The Nb–O distances are 1.88 Å (×3) and 2.13 Å (×3), while the Li–O distances
are 2.05 Å (×3) and 2.27 Å (×3). Although Li and Nb move in opposite directions, the niobium
ion carries a higher charge so that the distortion creates a local dipole moment. Because all
niobium atoms shift in the same direction, a net dipole moment develops parallel to the c axis.27
While the niobates and tantalates have the largest NLO coefficients among commercial
inorganic materials, their properties are not well suited for many applications. Their low
damage threshold rules them out for high-power lasers. Because they have very limited
transparency in the UV (the absorption edge for LiNbO3 is 400 nm), they cannot be used if
the SHG light is in the UV. Finally, the need to engineer the domain structure to achieve
quasi-phase matching adds a level of complexity to the crystal-preparation process.

8.9.4 Organic and Polymer NLO Materials

The NLO materials discussed thus far depend on displacements of ions to form local dipole
moments that are responsible for the NLO response. Another approach to NLO materials
design is to employ molecular fragments that have
N(CH3)2 highly polarizable bond networks. Molecules that
O2N exhibit conjugated π bonding tend to show high
polarizability, particularly those with an electron-
Figure 8.23 The structure of 4-(N,
donating group on one end and an electron acceptor
N-dimethylamino)-4 0 -nitrostilbene. on the other. A prototypical example is 4-(N,
N-dimethylamino)-4 0 -nitrostilbene (Figure 8.23),
where two benzene rings, connected by an ethylene group (stilbene), make up the conjugated
π system; the dimethylamino group, –N(CH3)2, acts as the donor, and the nitro group, –NO2,
on the opposite end of the molecule acts as the acceptor [22].
In such molecules, the electromagnetic field of the propagating light wave can induce
significant reorganization of the delocalized π-bonding electron density. The effective
NLO coefficients of the best molecular species can be orders of magnitude larger than
the classic inorganic NLO materials, but several practical considerations have limited
their use. Firstly, it is difficult to control the crystallization to obtain a non-
centrosymmetric crystal. This is especially true for conjugated polymers that have
attracted interest for NLO applications. Secondly, the optical quality of organic crystals
tends to be poor and the damage threshold low. Finally, conjugated organic molecules
tend to absorb strongly in the near-UV and visible regions of the spectrum.

27
The LiNbO3 and ilmenite (FeTiO3) structures are closely related ordered variants of the corundum structure, with
one important difference. In LiNbO3, the Li–Nb pairs all point in the same direction (e.g. Li up and Nb down),
which produces a polar structure, while in ilmenite the orientation of the Fe–Ti pairs alternates, leading to
a centrosymmetric structure.
340 Dielectrics and Nonlinear Optical Materials

8.9.5 Borates
Conjugated organic molecules are not the only species that possess delocalized π bonds.
Trigonal-planar polyatomic anions like NO3−, CO32−, and BO33− all have polarizable
delocalized π-bonding MOs. Of these non-centrosymmetric building blocks, borates are by
far the most important for NLO applications. About 36% of borates crystallize in non-
centrosymmetric space groups, as compared to ~15% of all inorganic solids [23]. Another
attractive feature of the borates is a large HOMO–LUMO gap that makes them transparent
to near-UV light.
Two important borates are β-BaB2O4 and LiB3O5. These materials were developed in the
1980s, and they have subsequently become among the most important NLO materials. The
appeal of the borates stems from three factors: (a) damage thresholds are much higher than
KTP or LiNbO3, (b) many borates are transparent to wavelengths that extend below 200 nm,
and (c) relative ease of phase matching.
The NLO coefficients and birefringence of borate crystals largely depend on the
concentration and orientation of trigonal-planar BO3 groups, containing sp2-
hybridized boron. The structure of β-BaB2O4 is shown in Figure 8.24. A key feature
is the presence of planar B3O63− anions oriented perpendicular to the polar threefold
axis. This structural arrangement leads to large NLO coefficients. Furthermore, the
coplanar orientation results in high birefringence, a useful trait for critical phase
matching.
In LiB3O5, triangles and tetrahedra of borate anions link together to form a complex
3D network. There are two important changes with respect to β-BaB2O4. Firstly, only

B3O63– group

Ba
b

Figure 8.24 The structure of β-BaB2O4. The planar B3O63− anion is shown on the right.
 8.9 Important SHG Materials 341

c
Bi
b a

Figure 8.25 The structure of BiB3O6 showing the chains of corner-connected BO4 tetrahedra and BO3
triangles.

two-thirds of the boron atoms are in a trigonal-planar coordination environment.

Secondly, the coplanar alignment of these triangles is lost. Consequently, LiB3O5 has
both smaller NLO coefficients and a smaller birefringence than β-BaB2O4. Its advantage
is a higher damage threshold, among the highest for commercial SHG materials. Phase
matching with LiB3O5 crystals is often achieved via noncritical phase matching (“tem-
perature tuning”).
An attractive design strategy is to combine the polarizability of the borate anions with the
high nonlinearities that come with SOJT distortions of cations. Unfortunately, compounds
that contain borate anions with cations prone to SOJT distortions are relatively rare, and in
those that do exist, the orientations of the cations and anions are not well aligned. An
exception is BiB3O6 (Figure 8.25) where Bi3+ with a pronounced stereochemically active lone
pair is present. Calculations of the NLO coefficients show that both the borate anions and the
Bi3+ cations make important contributions to the NLO response, which is much higher than
LiB3O5, even though both compounds contain similar concentrations of tetrahedral and
trigonal-planar boron.
There are many other borate materials that have been investigated for SHG applications.
CsLiB6O10 is a congruently melting structural derivative of LiB3O5, which makes crystal
growth easier. Sr2Be2B2O7 and KBe2BO3F2 are transparent down to 150–160 nm and thus
are well suited for applications involving short-wavelength UV light. The isostructural
compounds YCa4(BO3)3O and GdCa4(BO3)3O can be doped with luminescent rare-earth
ions (e.g. Nd3+) to use in self-frequency-doubling lasers, where the same material is used for
lasing and nonlinear frequency conversion. This results in compact devices capable of
producing UV, blue, and green light.
342 Dielectrics and Nonlinear Optical Materials

Box 8.3 Synthetic Methods: Crystal growth for NLO applications

For practical applications, high-quality single crystals of nonlinear optical materials are
a necessity. The preferred method of growth depends on the solubility and melting
characteristics of the material in question. The most familiar method of crystal growth
is from aqueous solution. Large KH2PO4 crystals are grown by this technique.
Crystallization involves two steps, nucleation and growth. To obtain large single crys-
tals one needs to avoid multiple nucleation sites. Therefore, a seed crystal is used to
introduce a single nucleation site. The next step is to gently oversaturate the solution
either through evaporation of the solvent or gradual cooling. The seed crystal is slowly
rotated to encourage uniformity in the growth. If the material is sparingly soluble, it
may be possible to increase its solubility using hydrothermal conditions. This is the case
for KTiOPO4, where seed crystals and the source material, often polycrystalline
KTiOPO4 in the presence of mineralizers like KH2PO4 and KNO3, are placed in
a hydrothermal autoclave [24]. Once the desired high temperature and high pressure
have been reached, a temperature gradient is used to grow the seeds at the cooler end
into large crystals.
Many commercial SHG materials are not water-soluble, which means that other methods of
growth must be employed. The best strategy depends on the melting characteristics of the
material. Congruently melting compounds (Section 4.2.2) are often grown by the Czochralski
method: A seed crystal is placed in contact with the surface of the melt whose temperature is
kept slightly above the melting point. The seed is then slowly pulled out of the melt, setting up
a temperature gradient that causes the melt to crystallize on the tip of the seed crystal. The seed
is usually rotated as it is pulled out to maintain uniformity of the temperature and composition
of the melt.
Surprisingly, many commercially important SHG materials do not melt congruently.
Incongruently melting compounds decompose upon melting, forming a solid and a melt
of differing compositions. Their crystal growth requires knowledge of the phase diagram to
properly select the composition of the starting mixture such that the melt formed will be in
equilibrium with the composition of the desired crystal. Another possibility is a “flux”
method. A flux is an additive, typically a low-melting salt, whose molten state acts as
a solvent. Growth of NLO crystals from a flux generally involves top-seeded solution
growth, where a rotating seed crystal is slowly pulled out of the flux, similar to the
Czochralski method. β-BaB2O4 is often grown from NaF fluxes. KTiOPO4 can be grown
from fluxes such as K4P2O7 and K6P4O13. LiB3O5 and LiNbO3 are grown by “self fluxing”,
that is by using an excess of one of the reagents, B2O3 and Li2O, respectively. The
disadvantage of a flux is that it can introduce undesired impurities and/or lead to
nonstoichiometry.
 8.10 Problems 343

8.10 Problems
8.1 Derive the Clausius–Mossotti equation for the CGSes system of units.
8.2 Use the Clausius–Mossotti expression in Equation (8.13) and Table 8.2 to estimate
permittivities of SnO2 (rutile-type, P42/mnm, Z = 2, a = 4.74 Å, c = 3.19 Å), TiO2
(rutile-type, a = 4.59 Å, c = 2.96 Å) and ZrO2 (baddeleyite-type, P21/c, Z = 4, unit-cell
volume = 141 Å). Does the Clausius–Mossotti equation give a reasonably accurate
estimate for each compound when compared to the experimental values of εr given in
Table 8.1? If not, what is the origin of the discrepancy?
8.3 Why does the polarizability of the lanthanoid ions decrease as the atomic number increases?
8.4 Lanthanoid zirconates Ln2Zr2O7 (Ln = La, Pr, Nd, Sm, Eu) adopt the cubic pyrochlore
structure (Fd3m, Z = 8), with cubic unit-cell edge a = 10.80 Å (La), 10.69 Å (Pr), 10.67 Å
(Nd), 10.59 Å (Sm), 10.55 Å (Eu). Use this information, Equation (8.13) and Table 8.2 to
estimate the dielectric permittivity of these phases. Does the Clausius–Mossotti equation
predict that the dielectric constant will increase or decrease as the radius of the rare-earth ion
decreases?
8.5 Use the Clausius–Mossotti expression to estimate the dielectric permittivities of the
cubic perovskites BaZrO3 (a = 4.19 Å), KTaO3 (a = 3.99 Å), and SrTiO3 (a = 3.90 Å).
Compare your estimates to the observed values in Table 8.1. How does the divergence
between calculated and observed values correlate with the tendency for the octahedral
cations to undergo SOJT distortions? What does this tell you about compositions for
which the Clausius–Mossotti equation can reliably be used?
8.6 Identify three characteristics of a relaxor ferroelectric that distinguish it from a normal
ferroelectric.
8.7 Why don’t first-order Jahn–Teller distortions, such as those seen with six-coordinate Cu2+
or Mn3+, typically lead to polar materials?
8.8 In each of the following pairs, identify the cation that is more likely to undergo an SOJT
distortion. Briefly explain your reasoning. (a) Ti4+ or V4+, (b) Mn3+ or Ta5+, (c) Mo6+ or
Zr4+, (d) Sn4+ or Te4+.
8.9 In a simple cubic perovskite like SrTiO3, all oxygens are equivalent. Each makes two
bonds to Ti4+ and four to Sr2+, as shown in a bond graph below.
O

Ti O Sr

(a) Draw a comparable bond graph for a 1:1 ordered perovskite Ba2ScTaO6 that obeys
Pauling’s rule of parsimony (Section 1.4.4). (b) Use the bond graph to estimate the Ba–O,
Sc–O, and Ta–O bond valences. (c) Draw a comparable bond graph for the 1:2 ordered
perovskite Ba3ZnTa2O9 and determine the minimum number of chemical environments
344 Dielectrics and Nonlinear Optical Materials

needed for oxygen is this structure. For each such type of oxygen, determine the number
of bonds it makes to each cation. (d) What are the bond valences for Ba–O, Zn–O, and
Ta–O? What are the bond-valence sums for each type of oxygen? (e) Based on your
answers to parts (b) and (d), do the oxygen bond valences suggest a bonding instability
that might trigger an SOJT distortion of the Ta5+ ions? Explain your reasoning.
8.10 Is it possible for a cubic material structure to be (a) non-centrosymmetric, (b) piezoelectric,
(c) pyroelectric?
8.11 The resonant frequency of a microwave resonator can be approximated by the relationship:
c
ν0 ≈
dcavity εr

where c is the speed of light and dcavity and εr are the diameter and permittivity of the
dielectric puck, respectively. (a) What diameter would give a resonant frequency, ν0, of
850 MHz for a puck made of Ba3(Co1−xZnx)Nb2O9 (εr = 34)? (b) How would the
diameter change for a resonator made from (Ca1−xNdx)(Ti1−xAlx)O3 (εr = 45)?
8.12 The perovskite BiInO3 can be prepared using high-pressure synthesis. At room tempera-
ture, it has Pna21 space-group symmetry, with an SHG signal 120–140 times that of α-
quartz. Upon heating, no phase transitions occur until it decomposes into In2O3 and Bi25
InO39 at 873 K. Based on this information, what can you say about the potential of BiInO3
for application as (a) a pyroelectric, (b) a ferroelectric, (c) a piezoelectric? (d) How would you
expect the dielectric constant to change upon heating?
8.13 BiAlO3 and BiGaO3 can be prepared by high-pressure synthesis. The space group of BiAlO3
is R3c while that of BiGaO3 is Pcca. Based on the symmetry alone, what can you say about
the possibility for (a) ferroelectric, and (b) piezoelectric behavior in these two phases?
8.14 PbO adopts the tetragonal litharge structure that can be described as a distorted variant
of the CsCl structure:

SOJT
distortion

c c

b
a b
a
CsCl structure litharge structure

The SOJT distortion driven by Pb2+ leads to a large displacement of the cation toward
a square face of the original cubic coordination environment and a corresponding
elongation of the c axis of the unit cell. (a) The bond-valence parameters for Pb2+–O
bonds are R0 = 2.11 Å and B = 0.37 Å. Use Equation (5.18) (dij ¼ R0ij  Blnvij ) to
estimate the Pb–O bond length in the hypothetical cubic CsCl-type structure of PbO.
 8.10 Problems 345

Use this distance to calculate the unit-cell edge and volume. (b) The actual structure of
PbO has space group P4/nmm and Z = 2, with a = 3.974 Å, c = 5.022 Å. Calculate the
unit-cell volume per formula unit and compare with your prediction for PbO with the
CsCl structure. What volume expansion (in %) is thus needed to make room for
the stereochemically active electron lone pair? (c) If we take Pb2+ and O2− to be equal in
size (eight-coordinate radii are 1.43 Å and 1.28 Å, respectively, for Pb2+ and O2−), how
does the “volume” of a lone pair compare with the “volume” of an oxide ion?
8.15 The black modification of SnO is isostructural with PbO litharge of the previous
problem. (a) The bond-valence parameters for Sn2+–O are R0 = 1.98 Å and B = 0.37 Å.
Use Equation (5.18) (dij ¼ R0ij  Blnvij ) to predict the Sn–O bond length in the hypo-
thetical CsCl-type SnO. Use this distance to calculate the unit-cell edge and volume. (b)
The crystal structure of SnO has space group P4/nmm and Z = 2, with a = 3.803 Å, c =
4.838 Å. Calculate the unit-cell volume per formula unit and compare with your
prediction of the volume for “cubic” SnO with the CsCl structure. What percent
expansion is needed to make room for the stereochemically active electron lone pair?
8.16 We can approximate the Sn2+ coordination in the hypothetical CsCl-type SnO of the
previous problem by an SnH86− cube. The point-group symmetry for a cube is m3m (Oh).
In SnH86−, the Sn 5s and 5p orbitals have a1g and t1u symmetry, respectively. The hydride
ligands form two triply degenerate SALCs with t1u and t2g symmetry, as well as two singly
degenerate SALCs with a1g and a2u symmetry. Use this information to construct an
approximate MO diagram for a cubic SnH86− molecule. What is the degeneracy, bonding
character, and symmetry of the HOMO and LUMO?
8.17 The structure of NH4H2PO4 is closely related to KH2PO4. A projection of the NH4H2PO4
structure (comparable to KH2PO4 in Figure 8.20) is shown below. The lightly shaded
phosphate tetrahedra are at z = ½, the darker ones at z = 0. The hydrogen atoms are
represented by black spheres. Based on the pattern of O–H bonds in this figure, predict the
displacements of the phosphorus atoms. Would you expect NH4H2PO4 to be an SHG
material like KDP?

z= 0

z= 0 z= ½

x z= ½
346 Dielectrics and Nonlinear Optical Materials

8.18 In 2005, two new polymorphs of BiB3O6 (β- and γ-BiB3O6) were synthesized using boric
acid as a flux [25]. All three polymorphs adopt closely related monoclinic structures.
While α-BiB3O6 has C2 space-group symmetry, β- and γ-BiB3O6 both crystallize with the
P21/n space group. Would you expect β- and γ-BiB3O6 to show NLO properties compar-
able to α-BiB3O6?
8.19 Determine the crystal system and point-group symmetry from the space-group symbol for
each of the following borates. In each case determine the point group and state whether the
symmetry permits SHG activity. (a) I42d for CsLiB6O10, (b) Ia3d for Sr4Li(BO3)3,
(c) Ama2 for Ca4Na(BO3)3, (d) P212121 for CsB3O5, (e) P21/n for BaLiBO3, (f) C2 for
CsBe2BO3F2.
8.20 For borates with similar orientation of the BO3 groups, the NLO coefficients should
roughly scale with the coplanar character and density of the BO3 groups. The following
borates have highly coplanar anions: β-BaB2O4 (R3c, unit-cell volume 1731 Å3, Z = 18),
Sr2Be2(BO3)2O (P6c2, unit-cell volume 290.8 Å3, Z = 2), KBe2BO3F (R32, unit-cell
volume 318.0 Å3, Z = 3). (a) Calculate the concentration of BO3 groups per unit volume
for each compound. (b) The largest NLO coefficients for each compound are the d22 =
2.3 pm/V for β-BaB2O4, d11 = 1.52 pm/V for Sr2Be2(BO3)2O and d11 = 0.8 pm/V for
KBe2BO3F. Do the NLO coefficients scale with the concentration of BO3 groups? (c)
CsBe2BO3F2 (C2, unit-cell volume 243 Å3, Z = 2) also has nearly coplanar BO3 groups.
How would you expect its NLO coefficients to compare with the other three compounds?

8.11 Further Reading

S. Elliot, “The Physics and Chemistry of Solids” (1998) Wiley.
A.R. West, “Solid State Chemistry and its Applications” 2nd edition (2014) Wiley.
R.W. Boyd, “Nonlinear Optics” 3rd edition (2008) Academic Press.
P.S. Halasyamani, K.R. Poeppelmeier, “Noncentrosymmetric oxides” Chem. Mater. 10 (1998),
2753–2769.
M.T. Sebastian, “Dielectric Materials for Wireless Communication” (2008) Elsevier.
I.M. Reaney, D. Iddles, “Microwave dielectric ceramics for resonators and filters in mobile phone
networks” J. Am. Ceram. Soc. 89 (2006), 2063–2072.
A.A. Bokov, Z.G. Ye, “Recent progress in relaxor ferroelectrics with the perovskite structure”
J. Mater. Sci. 41 (2006), 31–52.

8.12 References
[1] R.D. Shannon, “Dielectric polarizabilities of ions in oxides and fluorides” J. Appl. Phys. 73
(1993), 348–366.
 8.12 References 347

[2] R.D. Shannon, R.X. Fischer, “Empirical electronic polarizabilities in oxides, hydroxides, oxy-
fluorides, and oxychlorides” Phys. Rev. B 73 (2006), 235111.
[3] I.M. Reaney, D. Iddles, “Microwave dielectric ceramics for resonators and filters in mobile phone
networks” J. Am. Ceram. Soc. 89 (2006), 2063–2072.
[4] W.J. Merz, “The electrical and optical behavior of BaTiO3 single-domain crystals” Phys. Rev. 76
(1949), 1221–1225.
[5] M.S. Senn, D.A. Keen, T.C.A. Lucas, J.A. Hriljac, A.L. Goodwin, “Emergence of long-range
order in BaTiO3 from local symmetry-breaking distortions” Phys. Rev. Lett. 116 (2016), 207602.
[6] A.A. Bokov, Z.G. Ye, “Recent progress in relaxor ferroelectrics with the perovskite structure”
J. Mater. Sci. 41 (2006), 31–52.
[7] F. Cordero, F. Craciun, F. Trequattrini, C. Galassi, P.A. Thomas, D.S. Keeble, A.M. Glazer,
“Splitting of the transition to the antiferroelectric state in PbZr0.95Ti0.05O3 into polar and
antiferrodistortive components” Phys. Rev. B 88 (2013), 094107.
[8] A. Yoshiasa, T. Nakatani, A. Nakatsuka, M. Okube, K. Sugiyama, T. Mashimoto, “High-
temperature single-crystal X-ray diffraction study of tetragonal and cubic perovskite-type
PbTiO3 phases” Acta Crystallogr. Sect. B 72 (2016), 381–388.
[9] B. Noheda, D.E. Cox, G. Shirane, J.A. Gonzalo, L.E. Cross, S.-E. Park, “A monoclinic ferro-
electric phase in the Pb(Zr1−xTix)O3 solid solution” Appl. Phys. Lett. 74 (1999), 2059–2061.
[10] N. Zhang, H. Yokota, A.M. Glazer, D.A. Keen, S. Gorfman, P.A. Thomas, W. Rena, Z.G. Ye,
“Local-scale structures across the morphotropic phase boundary in PbZr1−xTixO3” IUCrJ 5
(2018), 73–81.
[11] P.S. Halasyamani, K.R. Poeppelmeier, “Noncentrosymmetric oxides” Chem. Mater. 10 (1998),
2753–2769.
[12] R.A. Wheeler, M.H. Whangbo, T. Hughbanks, R. Hoffmann, J.K. Burdett, T.A. Albright,
“Symmetric vs. asymmetric linear M–X–M linkages in molecules, polymers, and extended
networks” J. Am. Chem. Soc. 108 (1986), 2222–2236.
[13] H.W. Eng, P.W. Barnes, B.M. Auer, P.M. Woodward, “Investigations of the electronic structure
of d 0 transition metal oxides belonging to the perovskite family” J. Solid State Chem. 96 (2003),
535–546.
[14] F. Cora, C.R.A. Catlow, “QM investigations on perovskite-structured transition metal oxides:
Bulk, surfaces and interfaces” Faraday Trans. 114 (1999), 421–442.
[15] M. Kunz, I.D. Brown, “Out-of-center distortions around octahedrally coordinated d 0 transition
metals” J. Solid State Chem. 115 (1995), 395–406.
[16] K.M. Ok, P.S. Halasyamani, C.D. Casanova, M. Llunell, A.P. Alemany, S. Alvarez, “Distortions
in octahedrally coordinated d 0 transition metal oxides: A continuous symmetry measures
approach” Chem. Mater. 14 (2006), 3176–3183.
[17] W. Chang, S.W. Kirchoefer, J.M. Pond, J.A. Belloti, S.B. Qadri, “Room-temperature tunable
microwave properties of strained SrTiO3 films” J. Appl. Phys. 96 (2004), 6629–6633.
[18] M. Itoh, R. Wang, Y. Inaguma, T. Yamaguchi, Y.J. Shan, T. Nakamura, “Ferroelectricity
induced by oxygen isotope exchange in strontium titanate perovskite” Phys. Rev. Lett. 82
(1999), 3540–3543.
[19] M.W. Stoltzfus, P.M. Woodward, R. Seshadri, J-H. Klepeis, B.E. Bursten, “Structure and
bonding in SnWO4, PbWO4 and BiVO4: Lone pairs vs. inert pairs” Inorg. Chem. 46 (2007),
3839–3850.
[20] I.D. Brown, “The Chemical Bond in Inorganic Chemistry: The Bond Valence Model” (2002),
Chapter 8, Oxford University Press.
348 Dielectrics and Nonlinear Optical Materials

[21] P. Becker, “Borate materials in nonlinear optics” Adv. Mater. 10 (1998), 979–992.
[22] T. Vijayakumar, I.H. Joe, C.P.R. Nair, V.S. Jayakumar, “Efficient electron delocalization in
prospective push–pull non-linear optical chromophore 4-[N, N-dimethylamino]-4’-nitro stilbene
(DANS): A vibrational spectroscopic study” Chem. Phys. 343 (2008), 83–99.
[23] D. Xue, K. Betzler, H. Hesse, D. Lammers, “Nonlinear optical properties of borate crystals”
Solid State Commun. 114 (2000), 21–25.
[24] R.A. Laudise, W.A. Sunder, R.F. Belt, G. Gashurov, “Solubility and P–V–T relations and the
growth of potassium titanyl phosphate” J. Cryst. Growth 102 (1990), 427–433.
[25] L.Y. Li, G.B. Li, Y.X. Wang, F.H. Liao, J.H. Lin, “Bismuth borates: Two new polymorphs of
BiB3O6” Inorg. Chem. 44 (2005), 8243–8248.
9 Magnetic Materials

9.1 Magnetic Materials and Their Applications

While most people are familiar with the concept of a magnet, many do not realize the
ubiquity of magnetic materials in our everyday lives. Every electric motor contains a ferro-
or ferrimagnet. So does every headphone, loudspeaker, and power-supply transformer.
Magnetic card strips contain ferrimagnetic γ-Fe2O3. The old technology of tape recording
used γ-Fe2O3 or ferromagnetic CrO2. Hard disks in computers have recording platters
coated with ferromagnetic alloys patterned on a nanometer scale.
The fundamental origin of magnetism can be traced back to the movement of an electrical
charge. This can be the flow of current in an electrical circuit or, as in the solids we’ll discuss,
due to the quantum-mechanical properties of electrons in atoms. In this chapter we first
introduce some of the key physical concepts of magnetism and define some of the quantities
involved. We will discuss how to understand concepts such as diamagnetism and paramagnet-
ism of isolated atoms and their assemblies. We will then move on to study the origins of
cooperative phenomena such as antiferromagnetism, ferromagnetism, and ferrimagnetism and
how these can be controlled and exploited in functional materials. It is perhaps worth noting at
the outset that magnetism is an area where we’ll encounter unfamiliar units and where it’s often
more convenient to work in non-standard units, the so-called CGSem units, than the standard
SI system. We’ll generally adopt SI but will choose to list the alternative CGSem units
in situations where they’re most commonly encountered in the literature.

9.2 Physics of Magnetism

9.2.1 Bar Magnets and Atomic Magnets
Most people are familiar with the everyday properties of bar magnets from childhood toys
and school science experiments. We know that they send magnetic field as though emanating

349
350 Magnetic Materials

Figure 9.1 Magnetic field generated by a bar magnet and by an electric solenoid in a 2D rendering.

from one end, through the surrounding space, and back at the opposite end, forming
a magnetic loop. When suspended, the bar aligns itself with the magnetic field of the
Earth. By convention, the end that points towards the Earth’s magnetic pole in the north
is called the magnet’s north pole, N, and the other end the south pole, S. The N and S poles of
two bar magnets would attract each other until they unite in a longer bar of just two poles.
Accordingly, no division of a bar magnet isolates the poles. However small, a magnet always
behaves as an N, S magnetic dipole.
Amazingly, the same magnetic field can be generated by sending a direct electric current
through a solenoid (Figure 9.1). This not only facilitates the physical description of magnets
via moving electrical charges, but also suggests that magnetism has atomic origins, being
caused by the movement of charges in atoms.
Let’s extract a single current loop from the solenoid in Figure 9.1. The magnetic field
lines around such a circular loop are drawn in Figure 9.2. The product of the loop area
A and the loop current I has the magnitude μ of a vector μ, called the magnetic dipole
moment:1

μ ¼ IA (9.1)

This defines the SI unit2 of the magnetic dipole moment as A m2.

Let’s now consider, on Figure 9.3, just one electron orbiting with velocity v at radius r.3 As
a rotating particle, it has an angular momentum that is a vector,4 as shown in Figure 9.3. As

1
All magnetic moments are magnetic dipole moments, and we typically omit the word “dipole”. For convenience, we
often choose to deal only with the magnitude of this vector, as will be done for most of this chapter, and omit the
“magnitude” by speaking of its absolute value as a “magnetic moment”.
2
The unit A m2 is equivalent to joule per tesla, J/T = kg m2 s−2/(kg s−2 A−1), in which the joule is the unit of work or
energy and (as you’ll see shortly) the tesla is the unit of magnetic induction.
3
Be careful not to confuse the italic v for velocity with the Greek letter ν for frequency.
4
The direction of the angular-momentum vector follows the right-hand rule of fingers indicating the circular motion
of the particle.
 9.2 Physics of Magnetism 351

Figure 9.2 Left: Electric current generates a magnetic field with a specific orientation. Right:
Magnetic-moment vector μ due to electric current I in a circular loop of area A whose vector
is not drawn but is parallel with μ (as it follows the thumb of the right hand with fingers along
the current direction). The current direction is by definition the opposite of the movement of the
electrons.

Figure 9.3 Electron as an orbiting and charged particle has an angular momentum and gives rise to
a magnetic moment.

a rotating charge, it produces magnetic moment. The current I is the flow of the electron’s
charge per time τ = 2πr/v to complete one circuit,
e ev
I¼ ¼ (9.2)
τ 2πr

where e is the elementary charge, 1.602176634×10−19 C (Appendix J). Accordingly, the

current direction is the opposite of the electron’s direction, turning the area vector A as
well as the magnetic moment downwards in Figure 9.3, in agreement with the right-hand rule
in the caption of Figure 9.2. We see that the magnetic-moment vector has the opposite
direction of the angular-momentum vector. From now on, however, let’s consider only
magnitudes. The magnitude of the magnetic moment is:
352 Magnetic Materials

ev 2 evr
μ ¼ IA ¼ πr ¼ (9.3)
2πr 2

To express the velocity v, we turn to angular momentum. From classical mechanics we know
that the magnitude of angular momentum of the rotating electron would be given by mevr,
where me is the mass of the electron. However, we also know that this value is quantized5 and
restricted to mℓħ. Equating these two expressions for the momentum, we obtain for v:

mℓ ħ
v¼ (9.4)
me r

The magnitude of the magnetic moment is then found by substituting into Equation (9.3):

emℓ ħr eħ
μ¼ ¼ mℓ ¼ μB mℓ (9.5)
2me r 2me

Here we have introduced the Bohr magneton μB = eħ/2me as a convenient unit for the very
small atomic magnetic moment, μB = 9.27401×10−24 A m2.

9.2.2 Magnetic Intensity, Induction, Energy, Susceptibility, and Permeability

Let’s consider the solenoid in Figure 9.1 and imagine that it is much longer than its diameter.
The intensity of the magnetic field, H,6 generated in the center of such a solenoid, along its
axis, is one ampere per meter (A/m) when the product of the current (in A) and of the density
of turns (in m−1) along the solenoid length is unity.7 As noted earlier, we use the moving
electrical charge to define the magnetic field. The Earth’s magnetic field at sea level is in tens
of A/m and a refrigerator magnet might be ~105 A/m. Remember that H is actually a vector,
and so are B and M, introduced below.8
When a medium is immersed into a magnetic field, it is penetrated by it. The density of the
magnetic force lines in the medium is called the magnetic induction, B.9 The magnetic induction is
defined as the force that a homogeneous magnetic field exerts on a unit length of a straight wire
carrying unity current (Appendix E). The unit of B, one N/(A m) = J/(A m2), is named the

5
It is quantized with the orbital magnetic quantum number, mℓ (Chapter 5). For our rotating electron, the quantiza-
tion follows already from the condition of having an integer number mℓ of wavelengths λ = h/mev around the orbit
when the electron is taken as a standing wave.
6
Also called magnetic-field strength. When used to magnetize objects, it is termed magnetizing field. The electric
analogy of H is the electric-field intensity E introduced in Section 8.1.
7
A current of 0.001 A in a solenoid of 1000 turns per meter of its length will generate H = 1 A/m.
8
We choose to deal only with the magnitude because magnetic measurements are typically performed on powder
samples of isometric shape, where magnetic susceptibility (see next page) is measured as a scalar. In a single crystal, it
would generally be a tensor, a correlation scheme between two vectors.
9
B is therefore also called magnetic-flux density, but physicists often prefer just magnetic field. The electric analogy of
B is the electric-flux density D (the electric displacement introduced in Section 8.1.2).
 9.2 Physics of Magnetism 353

tesla (T). The relationship between the field intensity H and the induction B0 in a vacuum
(absence of any medium, hence the zero) is:

B0 ¼ μ0 H; (9.6)

where μ0 (4π×10−7 N/A2) is the magnetic constant.10 Inside any other medium, the induction
B will be the sum of the contribution from the external field (µ0H) and from the medium itself
(µ0M):

B ¼ μ0 ðH þ MÞ (9.7)

where M is the magnetization of the medium, its magnetic moment per unit volume
(A m2/m3 = A/m).11 The dependence of M on H characterizes a material’s magnetic
properties at a given temperature. The magnetization M is the key quantity because it
represents a material-specific response to an external magnetic field.
Materials that are less penetrable for a magnetic field than a vacuum have M < 0 and are
said to be diamagnetic, while others, such as paramagnetic materials, concentrate the mag-
netic field in their volume.12 As shown in Figure 9.4, in an uneven magnetic field (under
a gradient), paramagnets will move towards regions of highest magnetic intensity (be
attracted), whereas diamagnets will move away (be repelled).
Magnetization M is conveniently described relative to the field intensity H that caused it—
by volume (magnetic) susceptibility, χv; the susceptibility per unit volume of the material,

Figure 9.4 A paramagnet is attracted into the magnetic field, diamagnet is repelled; the paramagnet will
appear to weigh more in the experiment sketched.

10
Also known as the permeability of free space (or vacuum permeability), it is defined in SI units via the force of
2×10−7 N exerted by a current of 1 A flowing in two 1-m-long and 1-m-distant parallel wires in absolute vacuum.
The Laplace–Biot–Savart law then gives μ0. Do not confuse μ0 with the symbol μ for the magnetic moment and μB
for Bohr magneton. The electric analogy of μ0 is the electric constant ε0 introduced in Section 8.1.1.
11
The electric polarization P (Section 8.1.2) is not analogous to this magnetization M; it is analogous to the rarely used
term magnetic polarization, J = μ0M, so that B = μ0H + J corresponds to Equation (8.8) for electric displacement.
The use of M and not J in this book follows from the definition of the magnetic moment μ via electric current in
a loop, as opposed to the alternative definition via the mechanical force moment of the magnetic pole of a magnet.
The pole’s exact coordinate is not well defined for an atom, whereas electric charges are separate and can be
approximated as points.
12
As do the cooperative antiferromagnetic, ferromagnetic, and ferrimagnetic materials we’ll meet later.
354 Magnetic Materials

M
χv ¼ (9.8)
H

which is dimensionless in the SI system. Susceptibility is an important parameter of

a magnetic material. It’s often convenient to divide χv with the mass density (in kg/m3) to
obtain the mass susceptibility, χm (in m3/kg). Multiplying χm with the molar mass (in kg/
mol)13 yields the molar susceptibility, χmol (in m3/mol).
If we substitute Equation (9.8) for χv back into Equation (9.7), we obtain

B ¼ μ0 Hð1 þ χv Þ (9.9)

from which we see that diamagnetic materials (for which induction B is less than in
a vacuum) have negative χv, and paramagnetic materials have positive χv. Typical
values of χmol at room temperature might range from −16×10−10 m3/mol for a diamag-
netic compound like H2O to a few 1000×10−10 m3/mol for a paramagnetic transition-
metal compound.
Figure 9.5 summarizes the relationships of B, H, and M. It shows that the volume
susceptibility arises from introduction of dimensionless variables when Equation (9.7) for
magnetization in a medium is divided by Equation (9.6) for magnetization in a vacuum.
The second dimensionless variable thus obtained is the relative magnetic permeability, μr,
defined as μr = B/(μ0H) = χv + 1. The value of μr tells us how many times the magnetic
induction is increased by the given material as opposed to a vacuum.14 Materials with a high
induction in their interior have permeability much larger than 1; vacuum has μr = 1.

B µ H+M B M
in A m
B H
B µ H
µ

H
B kg A−1 s−2
μ kg m s−2 A−2

Figure 9.5 Dimensionless magnetic susceptibility in SI units.

13
Or χv with the molar volume in m3/mol.
14
Just like the relative dielectric permittivity εr in Section 8.1.1 told us how many times the stored electric charge in
a capacitor increased upon inserting a material into it instead of a vacuum.
 9.2 Physics of Magnetism 355

Up to now, we assumed that the moment of the immersed magnetic body is aligned with
the external field. If it is not, the magnet will experience a torque to align it. It will have
a potential energy, an ability to do work. The potential energy of a magnetic moment cross
aligned in a homogeneous external field is a dot product of the vectors of the moment μ and
of the magnetic induction B. It’s a projection of μ on B times |B|, a scalar. When the
magnetized object is aligned within the applied field, we can consider it to be at the bottom
of an energy well of depth μB.

9.2.3 Unit Systems in Magnetism

We’ve mentioned in the introduction that there’s an alternative set of units to SI units
in magnetism. For various reasons, this centimeter–gram–second electromagnetic
(CGSem) system of units is still sometimes used when working with magnetism. In
this system, the μ0 terms disappear from the equations given above. The variety of
units can be very confusing when one starts working in this area. Figure 9.5 is recast
into CGSem units in Figure 9.6.
The conversions are not purely decimal, because the two systems build on different
system of units and their relations. For example, the law in Appendix E about the
force F acting on a unit segment l of a unit-current-carrying wire in a field of unit
magnetic induction B is dimensionally correct in CGSem only when the unit of current
is 10 A. The CGSem unit for the magnetic moment caused by a loop of electric
current, the “electromagnetic moment unit” (emu) is then realized by 10 A cm2,
whereas the SI unit is 1 A m2. Therefore, 1 emu = 10−3 A m2. As a unit of magnetic
moment, the emu also equals erg/G in analogy to J/T = A m2 in SI. The conversions
are summarized in Table 9.1.

B H+ M B M
B H
B H
μ

H
B
conversion to SI: 10 Oe =1000/4 A/m 1 G = 10–4 T

Figure 9.6 Dimensionless magnetic susceptibility in CGSem units.

356 Magnetic Materials

Table 9.1 CGSem–SI conversions for quantities in Figure 9.5 and Figure 9.6.

Quantity SI units CGSem units

H Field intensity 1000/4π A/m =1 Oe

B Induction 10−4 T =1 G
M Volume magnetization 1000 A/m =1 emu/cm3
Mm Mass magnetization 1 A m2/kg =1 emu/g
χv Volume susceptibility 4π dimensionless =1 emu/(cm3 Oe)
χm Mass susceptibility 4π×10−3 m3/kg =1 emu/(g Oe)
χmol Molar susceptibility 4π×10−6 m3/mol =1 emu/(mol Oe)
μ Magnetic moment 10−3 A m2 =1 emu

9.3 Types of Magnetic Materials

When we come to look at collective properties of magnetic moments and the technologically
important properties they impart, we’ll categorize magnetic materials into six basic types.
The magnetic behaviors of the five most important are summarized in a cartoon form in
Figure 9.7, together with dependences of key quantities on field and temperature.
Diamagnetic materials repel a magnetic field (Figure 9.4) and thus have a negative suscepti-
bility χ (a vacuum has χ = 0). Their relative permeability is less than one. Paramagnetic
materials contain unpaired electrons, have magnetic dipoles, concentrate a magnetic field
and thus have a positive χ. The defining property of paramagnets is that their magnetic atoms
or ions act independently of each other. In weak magnetic fields, plots of M versus H are
linear for diamagnets/paramagnets; having negative/positive slopes, respectively, because
the gradient of M versus H gives χv.
We will discuss four types of cooperative magnetic phenomena that can occur when
paramagnetic materials are cooled through a critical ordering temperature. In an anti-
ferromagnet, magnetic moments of equal magnitude align in an antiparallel fashion
below the Néel temperature, TN, and compensate each other. In a ferromagnet, they
align in a parallel fashion below the Curie temperature, TC, and sum with each other.
In a ferrimagnet, two or more unequal magnetic moments align antiparallel below TC
without exact compensation, giving a non-zero overall magnetic moment resembling
a ferromagnet. Ferro- and ferrimagnets concentrate a magnetic field to a much greater
extent than the other clases of materials. The M versus H plots for these materials
show a much larger magnetization at lower fields, they are nonlinear (χ depends on H)
and saturate at lower fields (M becomes independent of H). Finally, like ferroelectric
materials that we encountered in Chapter 8, they show hysteresis, such that when the
field is removed they retain a portion of their magnetization. This is what allows them
 9.4 Atomic Origins of Magnetism 357

M M M M M

H H H H H

T T T T T T T T

T T T T T T T

Figure 9.7 Types of magnetic materials and their typical dependences of M versus H and of χ and χ−1
versus T.

to be made into permanent magnets. A final category of magnetic materials is the spin
glass that contains moments that are “frozen” in a disordered arrangement
(Section 9.11).15

9.4 Atomic Origins of Magnetism

9.4.1 Electron Movements Contributing to Magnetism and Their Quantization
The movement of an electron in an atom can be imagined in terms of the electron orbit
around the nucleus (hence having its orbital angular momentum16) and of the electron spin
around its own axis (hence having its spin angular momentum). Both contribute to magnet-
ism. In Section 9.2.2 we saw how the orbiting electron’s angular momentum yields
a magnetic moment. The spin angular momentum does not lend itself to a macroscopic
interpretation of a “spinning electron”, this is a quantum-mechanical effect. When both
orbit- and spin-magnetic moments are present, they will combine into a total moment, the

15
Having read “spin glass” and “frozen”, one might rename a paramagnet as a “spin fluid” of spin orientations. In
ferro-, ferri-, and antiferromagnets, the ordered spin orientations would be a “spin crystal”.
16
For electrons in atoms, the term orbital angular momentum was adopted because electrons occur in orbitals. As the
momentum originates from the motion of an electron orbiting the nucleus, the term is not meant to imply that this
electron is confined to a single orbital.
358 Magnetic Materials

Table 9.2 Angular momenta in one-electron atom: Quantization, vector lengths, and
projections onto the quantization axis z.

Quantum number (QN) Permitted values Magnitude

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Orbital angular momentum QN: ℓ 0, 1, 2, . . . (≡ s, p, d, . . .) |ℓ|=ℏ ℓðℓ þ 1Þ
(Orbital) magnetic QN: mℓ ±ℓ, ±(ℓ − 1), . . ., 0 ℓz = ħp· ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
mℓ
Spin (angular momentum) QN: s ½ |s|=ℏ sðs þ 1Þ
Spin-magnetic QN: ms ±½ sz = ħp ·mffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s
Combined angular momentum QN: j (ℓ ≠ 0 only) ℓ±½ | j |=ℏ jðj þ 1Þ
Combined magnetic QN: mj ±j, ±(j − 1), . . ., ±½ jz = ħ · mj

angular momentum behind which will also be quantized, as the pure orbital or spin angular
momenta are.
Let’s review the quantization in a one-electron atom. From Chapter 5 we know that the
identity of an electron in an atom is given by four quantum numbers; n, ℓ, mℓ, and ms.17 Of
these, ℓ, mℓ, and ms relate to an electron’s rotational movements. They are a minimum subset
of the in-total six quantum numbers defined in Table 9.2: two for orbit, two for spin, and two
for their total. Of these six, ℓ, s, j quantize the respective angular-momentum vectors ℓ
(orbital), s (spin), and j (orbital and spin combined), and mℓ, ms, mj quantize the behavior
in magnetic field; they represent the respective projections ℓz, sz, jz of the allowed orientations
of those three vectors onto the axis of quantization, which is the axis of the external magnetic
field. For each angular-momentum quantum number ℓ, or s, or j, there are 2 × (ℓ or s or j) + 1
such possible orientations. Each orientation has a different z-projection magnetic moment,
μz, and the electron’s potential energy is therefore split into a multiplet when the external
magnetic field is applied—Zeeman effect (also termed Zeeman splitting).
The magnitudes µ (see also Footnote 1 in this chapter) of the absolute and z-projection
magnetic moments are calculated from the Magnitude expressions for angular momenta in
Table 9.2 and

Magnitude
μ ¼ gi μB (9.10)
ℏ

where gi (i = ℓ or s or j) is one of three different g factors: For orbital moment, gℓ = 1

classically. For spin, gs = 2.002319 for a single18 electron, known as the electron-spin g factor
(in many-electron systems denoted ge). For combinations of spin and orbit, the gj factor must
be calculated as shown in Section 9.4.2. The energy split of the multiplet under a magnetic
field applied along z is by Um,i (for i = ℓ or s or j, see Figure 9.10) calculated from the

17
For example, the value of ℓ tells us what type of orbital the electron occupies (s, p, d, or f) . . .
18
For some atoms, the exact value varies slightly with environment.
 9.4 Atomic Origins of Magnetism 359

s ½ ms s j mj

s ss
½
–½
m

j j j

mj
s j ½

Figure 9.8 Single electron’s spin angular momenta (left top) combined with orbital angular momenta (left
bottom) yield two choices for the total angular momentum j (right), quantized in units of ħ. The sizes of
the two spheres on the right correspond roughly to superposition of the spin cones over the orbital
angular momentum vectors at left bottom (precisely in the classical limit of ℓ→∞).

magnitudes of z-projection moments, μz, onto the field axis (Table 9.2) and mirrored by ±
values of mℓ, ms, or mj (Table 9.2) around the initial energy taken as zero:

Um;i ¼ B  μz ¼ B  gi μB mi (9.11)

Only mℓ and ms follow directly from the electron’s identity in the atom. The mj is obtained by spin–
orbit coupling. The one-electron angular momentum j is a vector sum of the orbital and spin
angular momenta, j = ℓ + s, which are quantized. As shown in Figure 9.8, top left, the spin
angular-momentum vector can lie anywhere on the surface of a cone around the field direction
and has two possible orientations (ms = +½, −½). Its components along the field direction are
+½ħ and −½ħ (Table 9.2). This spin angular momentum is brought into the picture of the orbital
angular momentum shown for a p electron (ℓ = 1) in Figure 9.8, bottom left, where the angular-
momentum vector quantizes in an external magnetic field into three cones with respective
projections −1ħ, 0, +1ħ onto the field direction. Since the spin (vector) has two possible orienta-
tions (ms = ½, −½), two spheres containing the mj-quantized cones are obtained, a smaller one for
j = ℓ − ½ (Figure 9.8, bottom right) and a larger one for j = ℓ + ½ (Figure 9.8, top right). The two
j states differ in energy. The difference is a measure of the spin–orbit coupling strength, which is
proportional to Z 4 and hence increases rapidly with the atomic number Z.

9.4.2 Atomic Magnetic Moments

When looking at the magnetic properties of a material, we’ll typically be interested in
the most stable or ground state of its magnetic atoms. The Russell–Saunders coupling
under Hund’s rules (Section 7.3.2) yields correct ground-state magnetic moments for
360 Magnetic Materials

most free atoms of transition metals and the lanthanoids. The j–j coupling scheme
(Appendix F) is needed for the actinoids. Figure 9.9 shows a useful summation scheme
for the Russell–Saunders coupling under Hund’s rules. The ground state for the Fe2+
example in Figure 9.9 is 2S+1LJ = 5D4 (“quintuplet-D-four”) because S = 2, L = 2
(hence D; Table 7.3), and J = 4 as the orbital shell is more than half filled. The S in the
ground-state term 2S+1LJ equals the largest value of the total MS that can be obtained
by summing ms, and the L equals the largest ML that can be obtained by summing mℓ.
In Figure 9.10, the ground state for our Fe2+ ion is illustrated on an energy
diagram. The many-electron state 5D is a quintuplet; the spin–orbit coupling produces
2S + 1 = 5 states of J between |L − S| = 0 and |L + S| = 4. In an applied magnetic

Electrons:
First, maximum S ms ½ ½ ½ ½ ½ ½ ½ ½ ½ ½

Second, maximum L m 2 1 0 1 2 2 1 0 1 2

Third Hund rule J= |L S| |L+S|

Figure 9.9 Hund’s rules in a summation scheme to determine the ground state of a free atom or ion with
d-valence electrons via Russell–Saunders coupling, illustrated on Fe2+ having 2S+1LJ = 5D4.

J L – S| MJ

L ,S

U J
gJ B
gJ B
gJ B
J L + S| gJ B
gJ B
gJ B
gJ B
B gJ B

Figure 9.10 Energy diagram for 5D sub-states of an isolated Fe2+ under spin–orbit coupling and
a magnetic field. For the ground state J = 4, magnetic potential energies Um,J are listed to illustrate the
meaning of the split in magnetic field.
 9.4 Atomic Origins of Magnetism 361

U S
MS
gs B
gs B
S
gs B
gs B

Figure 9.11 Energy diagram for isolated Fe2+ under spin-only approximation.

field, each of these states splits in energy into a multiplet of MJ values from −J to +J,
differing by 1.19
Note that the free-ion scheme will not necessarily be valid in a compound of a magnetic
atom. We know from Chapter 7 that the ligand field splits the degeneracy of d-electron levels.
It is then a question of the mutual symmetry and occupation of the remaining degenerate
orbitals whether or not there is a contribution from the orbital angular momentum. If not, as
is the case of some 3d ions in compounds, we say that the orbital moment is quenched, and
the “spin-only” splitting of the ground state in the applied field is much simpler (Figure 9.11).
As f electrons are much less sensitive to ligand-field effects, a scheme (see Figure 9.13 later)
analogous to Figure 9.9 applies to the magnetic ground state of both free and bonded
lanthanoid ions (except for Sm3+ and Eu3+). This is discussed in more detail in Section 9.4.3.
Magnetic moments are measured at non-zero fields and non-zero temperatures. The
measured value then depends on how the states exemplified in Figure 9.10 are populated.
This is controlled by the relative size of the thermal energy kT versus both the spin–orbit
J-splitting energy and the MJ-splitting energy that increases with increasing applied field B,
Equation (9.11). Two simple cases emerge, depending on the field and temperature:

1. In very strong fields of tens of teslas at very low temperatures of a few kelvins, a saturation
moment, μsat, is achieved for paramagnetic materials. It corresponds to the z-projection
value of the magnetic moment as calculated with Equation (9.10) and the appropriate
z-projection Magnitude listed in Table 9.2, in which mℓ, ms, mj are replaced by ML, MS,
MJ. In magnetically ordered materials, μsat is obtained via neutron diffraction.
2. In weak fields at high temperatures, an effective moment, μeff, is measured. It corresponds
to the absolute magnetic moment, the value of which is calculated with Equation (9.10)
and the appropriate Magnitude listed in Table 9.2, in which ℓ, s, j are replaced by L, S, J.

Four cases of these two types of moment are discussed in more detail in the following.

19
It’s worth stating that for a full electron shell, L, S, and J must all be zero, so the total angular momentum will be
zero and there will be no magnetic moment; the atom will be diamagnetic. This also means that we only need to
consider valence electrons when calculating magnetic moments.
362 Magnetic Materials

The weak-field limit for spin-only moment. In the weak-field limit, the magnetic MS split is
weak, kT is much larger than its splitting energy, and all levels of the S multiplet in Figure
9.11 are equally occupied. The absolute value of the moment in Table 9.2 is applicable, not its
projection onto the z axis, and the spin-only effective moment μeff is:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
μeff ¼ ge μB SðS þ 1Þ (9.12)

Since ge ≈ 2 and S = ½ nu, where nu is the number of unpaired electrons, the approximate
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
value of μeff in Bohr magnetons is given by nu ðnu þ 2Þ.
The strong-field limit for spin-only moment. Under large B, the difference in the MS sub-
state energies is large, and, at low temperatures, only the lowest MS is occupied (for example
the MS = −2 level in Figure 9.11). The saturated magnetic moment of the magnetized
substance is then calculated from the z-component magnitude in Table 9.2:

μsat ¼ ge μB S (9.13)

Setting ge ≈ 2 makes μsat in Bohr magnetons equal to the number of unpaired electrons.
The weak-field limit for spin–orbit coupling. Let’s consider the case where kT is less than the
J splitting (spin–orbit) but larger than the MJ splitting (magnetic). This means that the lowest
J multiplet is rather evenly populated, as exemplified by rare-earth ions at ambient temper-
atures. The grand total angular momentum J produces the magnetic moment. Because J is
composed of spin and orbital contributions, vector summation needs to be applied.
Figure 9.12 shows the summation graphically. Because of the “doubled” contribution
from S (ge ≈ 2), the instantaneous moment is not aligned with the angular momentum J, but

B
g e S cos( JS ) L cos( JL)
ħ
eff

J L
L

geS

Figure 9.12 Vector addition of the orbital and spin angular momenta (in black), and of their contribu-
tions towards the magnetic moment (in gray), valid in the weak-field case described in text. The effective
magnetic moment, μeff, is proportional to the sum of projections of geS and L onto J. For visual clarity of
the summation, the angular and magnetic moments are drawn parallel to each other instead of
antiparallel.
 9.4 Atomic Origins of Magnetism 363

rapidly sweeps a cone around it (Larmor precession), so that only the projection onto J over
time should be considered. The contributions towards the effective magnetic moment μeff
from the two angular momenta will also be projections: |L|cos(JL) and ge| S |cos(JS). The two
angles can be expressed from the cosine law, which says that |L|2 = | J |2+| S|2−2|J || S |cos(JS)
and that |S |2 = |J |2+|L|2−2|J ||L|cos(JL) (Figure 9.12). The magnitudes of the angular
momentum vectors (|S |, |L|, and | J |) are then expressed in terms of quantum numbers.
Hence, for J,
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
μeff ¼ gJ  μB JðJ þ 1Þ; (9.14)

where:
JðJ þ 1Þ  LðL þ 1Þ þ SðS þ 1Þ JðJ þ 1Þ þ LðL þ 1Þ  SðS þ 1Þ
gJ ¼ ge þ (9.15)
2JðJ þ 1Þ 2JðJ þ 1Þ

For ge approximated as 2, this simplifies into the Landé factor:

JðJ þ 1Þ  LðL þ 1Þ þ SðS þ 1Þ

gLandé ¼ 1 þ ≈ gJ (9.16)
2JðJ þ 1Þ

The strong-field limit for spin–orbit coupling. Under large B at low temperatures, only the
lowest MJ is occupied, and the saturated magnetic moment of the magnetized substance is
then calculated as μsat ¼ gJ μB J from the z-component magnitude in Table 9.2 but using gJ of
Equation (9.15).

9.4.3 Magnetic Moments for 3d Ions in Compounds

Values of μeff for 3d transition-metal ions are given in Table 9.3. The spin-only values are
much closer to those observed experimentally than to those calculated using the total angular
momentum. The reason for this is the well-known ligand-field effect, also discussed in
Section 5.3.8 and Section 7.3.1. As an example, a d1 Ti3+ free ion has a 10-fold degenerate
2
D ground state (the electron can be spin up or spin down in any one of five d orbitals, the
multiplicity 2S + 1 = 2, Figure 7.5). In an octahedral ligand field, the degeneracy is lifted,
resulting in t2g (dxy, dxz, dyz) and eg (dz2, dx2−y2) orbital sets.20 The eg orbitals are ~2.5 eV
above the t2g and will not be populated under normal conditions, so the Ti3+ electron is in t2g.
One therefore has a sixfold degenerate ground state. Whether the orbital angular momentum
contributes or is quenched then depends on the symmetry of these degenerate orbitals. We
can understand this by returning to our picture of the orbit angular momentum resulting

20
The ground-state term under a ligand field is found in correlation diagrams, dealt with in Section 7.3.3. In short, for
d n in an octahedral field, the terms can be A1g, A2g, Eg, T1g, or T2g, in a tetrahedral field the subscript g falls off. In
a weak octahedral/tetrahedral field (≡ high-spin case), free-ion terms D (d 1, d 9, d 4, d 6) split to E + T2, the F terms
(d 2, d 8, d 3, d 7) split to A2 + T1 + T2, the S term (d 5) is not split but renamed A1. Note that the subscripts are not J as
in the magnetic-field-induced R–S term, but relate to symmetry.
Table 9.3 Electronic configurations, free-ion terms, configurations, and ground terms for high-spin magnetic moments in Bohr
magnetons for selected 3d transition-metal salts featuring an octahedral ligand field. Note that different ground terms arise for
low-spin cases. Experimental data taken from Nicholls [1].

HS μ eff μ eff μ eff μ eff

ground Orbital calculated*
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi calculated
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi expt expt
dn Free ion HS config. term contribution? g J JðJ þ 1Þ g e SðS þ 1Þ 80 K 300 K Example
1 2 1 2
3d D 3/2 t 2g T 2g Yes 1.55 1.73 1.4 1.8 Cs 2VCl 6
3d2 3
F2 2
t 2g 3
T 1g Yes 1.63 2.83 2.7 2.7 (NH 4)V(SO 4) 2·12H 2O
3d3 4
F 3/2 3
t 2g 4
A 2g No 0.77 3.88 3.8 3.8 KCr(SO 4) 2·12H 2O
3d4 5
D0 3 1
t 2g eg 5
Eg No 0.00 4.90 4.8 4.8 CrSO 4·6H 2O
3d5 6
S 5/2 3 2
t 2g eg 6
A1 g No 5.92 5.92 5.9 5.9 K 2Mn(SO 4) 2·6H 2O
3d6 5
D4 4 2
t 2g eg 5
T 2g Yes 6.70 4.90 5.4 5.5 (NH 4) 2Fe(SO 4) 2·6H 2O
3d7 4
F 9/2 5 2
t 2g eg 4
T 1g Yes 6.63 3.88 4.6 5.1 (NH 4) 2Co(SO 4) 2·6H 2O
3d8 3
F 4, 6 2
t 2g eg 3
A 2g No 5.59 2.83 3.3 3.3 (NH 4) 2Ni(SO 4) 2·6H 2O
3d9 2
D 5/2 6 3
t 2g eg 2
Eg No 3.55 1.73 1.9 1.9 (NH 4) 2Cu(SO 4) 2·6H 2O
* Notice that, due to Hund’s third rule, spin–orbit coupling makes the experimental moment smaller than the spin-only moment when d orbitals
are less than half filled and bigger when more than half filled.
 9.4 Atomic Origins of Magnetism 365

from the circular motion of a charged particle around an axis (Figure 9.3). The orbital
angular momentum occurs when an electron has two orbitals available at about the same
energy level, that are rotationally equivalent around the axis of the electron’s conceived
circular motion. Around z, such symmetry-related orbitals would be dxz and dyz. Hence, in
our example of Ti3+ in an octahedral ligand field, an orbital contribution to the magnetic
moment is possible. In a d 3 configuration of octahedral Cr3+, on the other hand, the angular
momenta of the dxy, dxz, and dyz sphere cancel.
Symmetry being the key, the rule is such that a T ground term21 allows the orbital
angular-momentum contribution (Table 9.3). The overall picture is unfortunately more
complex still in that spin–orbit coupling leads to a further splitting of the t2g orbitals of
Ti3+ even before applying the external magnetic field. The energy separation is of the
order of kT, leading to the population, hence magnetic moment, changing with tem-
perature. For d1 Ti3+ in an octahedral ligand field, calculations22 of thermally induced
pffiffiffi
populations of the microstates predict that μeff would vary from 0 to 5 μB as the
temperature changes from 0 to ∞ K. The experimental values are far less extreme and
somewhat fortuitously end up being close to spin-only predictions at room tempera-
ture. The less extreme temperature dependence of such moments is partly due to
covalency (electrons shared with ligands) and partly because the true local environment
of d1 Ti3+ will not be a regular octahedron but will undergo Jahn–Teller splitting
(Section 5.3.10) into levels where the moment would be temperature-independent.
Similar arguments hold for the other T ground states, such as those for d 2, d 6, and
d7 in an octahedral field. Compounds of these ions therefore show moments that depart
from spin-only values due to spin–orbit coupling, and have contributions from ther-
mally excited states that bring a change with temperature.
Things are more straightforward for ions with ground terms A2g and Eg that allow
no orbital contribution. In an octahedral field, such d 3, d 4, d 8, and d 9 ions are found
experimentally to have μeff values close to spin-only predictions, and temperature
effects due to thermally excited states23 are weak. For d 5, spin-only μeff values are
observed.24
A peculiarity occurs for the octahedral d 3/d 8 ions (A2g terms), called zero-field splitting.
Even before magnetic field splits the spins states (into an array from −MS to +MS, see Figure
9.11), the ligand field splits them a little, yet differently; according to their absolute value,
into larger and smaller |MS |. This leads to a paramagnetic anisotropy; the split-state

21
Adopted in an octahedral field for t2g1, t2g2, t2g4eg2, and t2g5eg2; tetrahedral field e2t21, e2t22, e4t24, and e4t25.
22
According to Kotani theory.
23
Minor deviations from spin-only values arise due to mixing with higher-energy T terms of the same multiplicity as
the ground term. These are always present for d 3/d 8 A2g and d 4/d 9 Eg terms and can lead to μeff values lower (for a d
shell less than half filled) or higher (for a d shell more than half filled) than expected. In other words, an admixture of
excited states introduces some degree of spin–orbit coupling.
24
There are no higher T terms of the same multiplicity to induce even minor deviations.
366 Magnetic Materials

occupancies slightly change with the direction of the magnetic field and so do the suscepti-
bilities if measured on a single crystal. A weak anisotropy is also caused in the octahedral d 4
and d 9 ions (Eg terms) by Jahn–Teller distortion.

9.4.4 Magnetic Moments for 4f Ions in Compounds

For the lanthanoids, the situation is much simpler than above (see Table 9.4). The Russell–
Saunders scheme can still be applied, and the contracted nature of the 4f orbitals means that
ligand-field effects are far less important. Hund’s third rule (Section 9.4.2) is obeyed, and the
agreement between μeff(calc) using J and μeff(expt) is very good in most cases. For Sm3+ and
Eu3+, however, the experimental values are much higher than expected for the ground state.
This is because there are low-energy excited states partially occupied due to thermal energy,
and the moment rapidly falls on cooling.
As an example of the calculation, the Nd3+ ( f 3 ) Russell–Saunders 2S+1LJ term is 4I9/2
where S = ³⁄₂ yields 2S + 1 = 4, L = 6 yields the I (Table 7.3), and J = |L−S| is 6 − ³⁄₂ = ⁹ ⁄₂ .
Equation (9.16) then gives gLandé = 0.727, with which Equation (9.14) yields μeff = 3.61 μB.

9.4.5 Note on Magnetic Moments of 4d and 5d Metals in Compounds

When comparing the 3d and 4f ions just discussed, we see that relatively simple formulas can
be applied to estimate magnetic moments for most of them (except Sm3+ and Eu3+).
Estimates became rather complex for 4d and 5d metals since their larger spin–orbit coupling
( ∝ Z4) is comparable to ligand-field effects, and the situation is intermediate between the
relatively clear-cut 3d and 4f cases.

Table 9.4 Calculated μeff in Bohr magnetons for rare-earth ions.

Ion fn R–S term μeff(calc) μeff(expt)

Ce3+ 4f 1 2
F5/2 2.53 2.4
Pr3+ 4f 2 3
H4 3.58 3.5
Nd3+ 4f 3 4
I9/2 3.61 3.5
Pm3+ 4f 4 5
I4 2.68 not available
Sm3+ 4f 5 6
H5/2 0.84 1.5
Eu3+ 4f 6 7
F0 0.00 3.4
Gd3+ 4f 7 8
S7/2 7.95 8.0
Tb3+ 4f 8 7
F6 9.73 9.5
Dy3+ 4f 9 6
H15/2 10.65 10.6
Ho3+ 4f 10 5
I8 10.61 10.4
Er3+ 4f 11 4
I15/2 9.58 9.5
Tm3+ 4f 12 3
H6 7.56 7.3
Yb3+ 4f 13 2
F7/2 4.54 4.5
 9.6 Paramagnetism 367

Electrons:
First, maximum S ms ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½ ½

Second, maximum L m 3 2 1 0 1 2 3 3 2 1 0 1 2 3

Third Hund rule J= |L S| |L+S|

Figure 9.13 Hund’s rules summation to determine the ground state of a 4f atom or ion via Russell–
Saunders coupling, illustrated on Nd3+ of 2S+1LJ = 4I9/2.

9.5 Diamagnetism
In addition to the orbital and spin contributions to the magnetic moment of an atom with an
unpaired electron, there is a diamagnetic contribution. Diamagnetism is caused by the
change in the motion of all electrons in an applied field, and that includes the paired electrons
that we did not have to consider so far. We can imagine this effect as occurring due to
induced currents. Lenz’s law of induction states that the “magnetic field of an induced loop
current is oriented against the applied field”. Diamagnetism will therefore make χ negative by
expelling the field from the material. In a field with a gradient, the diamagnet will be driven
out to regions of weaker field. This is the origin of the famous levitating-frog experiment that
abounds on the internet. Diamagnetism occurs in all materials, though its effect is eclipsed by
the positive susceptibility arising from any unpaired electrons.25 Of the elements with no
unpaired electrons, Bi is the strongest diamagnet26 (it has the largest number of electrons and
J = 0) with χv = −166×10−6. In all metals, an additional weak diamagnetic effect due to
delocalized electrons occurs, termed Landau diamagnetism.27 Normal diamagnetism is not
a desperately interesting phenomenon; however, in superconductors (Chapter 12) the inner
currents ideally expel all the magnetic field, and one has a perfect diamagnet of χv = −1. Such
an extreme diamagnetic susceptibility of a superconductor is able to levitate trains.

9.6 Paramagnetism
Paramagnetism is a situation when atomic magnetic moments in the given substance are
randomly oriented. Thermal agitation disorders their orientations, whereas an applied
magnetic field tends to align them. A paramagnetic material will concentrate the magnetic

25
Nevertheless it’s important to correct for diamagnetic contributions when measuring and interpreting properties of
paramagnetic substances. Such diamagnetic corrections are often made from tabulated values of Pascal’s constants
as mentioned in Appendix I.
26
A compass with a Bi needle points east–west; true to the prefix “dia” meaning across.
27
Also conjugation electrons of a graphite plane (graphene) cause a strong anisotropic diamagnetism with χ =
−400×10−6 for the field perpendicular to the plane. Similarly for aromatic conjugated molecules.
368 Magnetic Materials

field in its volume (χ is positive) and be attracted to regions of high field. Typical values of χ
are 10−3 to 10−5. Paramagnets can be split into two categories according to the localized or
delocalized nature of the contributing electrons, which in turn determines how susceptibility
depends on temperature. Substances with localized electrons are classified as Curie or Curie–
Weiss paramagnets, while those with delocalized unpaired electrons are called Pauli
paramagnets.

9.6.1 Curie and Curie–Weiss Paramagnetism

Substances that contain non-interacting paramagnetic moments have a susceptibility that
varies inversely with temperature and obeys the Curie law,

C
χ¼ (9.17)
T

where C is the Curie constant. This hyperbolic temperature dependence arises from the
competition between the increasingly stronger thermal agitation of the independent moments
and the tendency of the applied field to align them (Figure 9.14). Examples of Curie para-
magnets include compounds with well-separated transition-metal ions, such as the Tutton’s
salt (NH4)2Mn(SO4)2·6H2O and the iron/chromium alums NH4M(SO4)2·12H2O (M = Cr, Fe),
rare-earth compounds where the magnetism arises from localized f electrons, and O2(g). The
Curie law is valid unless the applied field is very strong (>1 T) and the temperature very low
(<20 K). Deviations from the Curie law can be checked visually by plotting the reciprocal
susceptibility versus temperature, which provides a straight line, χ−1 = C −1T, when the Curie
law is obeyed (Figure 9.14).

Curie paramagnet

= C/T

0 T (K)
1
magnetic moments
randomly oriented
and dynamic 1 = C 1·T

0 T (K)
0

Figure 9.14 Curie paramagnet and Curie law: Temperature dependence of the magnetic susceptibility χ
(top), and linearization via its inverse (bottom).
 9.6 Paramagnetism 369

Materials that exhibit interactions between individual magnetic moments will

only behave as paramagnets above the temperature at which the thermal energy
overcomes those interactions and effectively randomizes the moments. That is above
TC for ferromagnets and above TN for antiferromagnets. The Curie–Weiss law is then
valid,

C
χ¼ (9.18)
T θ

where θ is the Weiss constant. Figure 9.15 shows χ and χ−1 versus absolute temperature of
these two types of Curie–Weiss paramagnets. As magnetization in general aligns magnetic
moments in one direction, a negative Weiss constant means a negative propensity to the
alignment, as in antiferromagnets, whereas a positive θ means a positive propensity to align
the moments, as in ferromagnets. The ferromagnetic order becomes thermally randomized
when a positive θ ≈ TC is reached upon warming, at which point it starts to resemble a Curie
paramagnet at 0 K. Since an antiferromagnet’s θ ≈ −TN, the antiferromagnet is as weak
a magnet at 0 K as a Curie paramagnet would be at a temperature TN. It should be noted that
when |θ| and the ordering temperature are very different, it indicates some type of frustration
working against the order (Section 9.11).
The derivation of the Curie law is in principle quantum mechanical because susceptibility
depends on thermally induced populations of the levels that are split from MJ in an applied
magnetic field (Figure 9.10). For Curie paramagnets at temperatures larger than ~20 K, in

T
T

C T

T T T

C T

Figure 9.15 Comparison of the two opposite Curie–Weiss paramagnets: A ferromagnet at high temperat-
ures and antiferromagnet at lower temperatures. Their temperature dependence of the magnetic suscepti-
bility χ is on top, and its linearization on the bottom shows the difference in their Weiss constants θ.
370 Magnetic Materials

fields that are not exceptionally strong, the quantum-mechanical result derived in
Appendix H is well approximated by the Curie law,

μeff 2 μ0
χN ¼ N (9.19)
3kT

where N is the number of magnetic atoms per which the susceptibility is expressed28 and μeff
is the effective magnetic moment (in A m2) from which S or J is calculated, leading to the
number of unpaired electrons per magnetic atom. As an example, for a spin-only moment,
Equation (9.12) is introduced into Equation (9.19) for μeff. With NA replacing N, χmol per
mole of magnetic atoms is obtained:

NA ge 2 μB 2 μ0 SðS þ 1Þ 3
χmol ¼ SðS þ 1Þ ¼ 6:3003106 m mol1 (9.20)
3kT T

Note that since χ is proportional to μeff2 in Equation (9.19), in a compound of two magnetic
atoms their μeff values aren’t additive. Susceptibility χ and therefore μeff2 values are. For
example, consider the perovskite-related LaSrMnMoO6. It contains two atoms with
unpaired electrons. High-spin Mn2+ has five unpaired electrons hence S = 2.5, from which
a spin-only μeff = 5.916 μB is calculated with Equation (9.12). The other magnetic atom is d1
Mo5+ of a spin-only μeff = 1.732 μB calculated from S = ½. The average moment per magnetic
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
atom is μeff ¼ 0:5  5:9162 þ 0:5  1:7322 ¼ 4:36 μB. It compares well with the average
moment of 4.33 μB obtained [2] by least-squares fitting with Equation (9.19) of the experi-
mental χ versus 1/T data for LaSrMnMoO6.
As noted earlier, when an extremely strong field is applied to a paramagnet at low
temperatures, the atomic moments saturate at the value μsat, Equation (9.13), and their
sum per unit volume saturates at Msat. This is described by the Brillouin function for the
relative magnetization Mr = M/Msat versus B/T, derived in Appendix H,
     
1 1 1 1 1 gJ μB B
Mr ¼ J þ coth J þ ξ  coth ξ ξ¼ (9.21)
J 2 2 2 2 kT

in which ξ is a ratio of the energy separation of the MJ levels and the thermal energy. As an
example, the Brillouin function for a spin-only case (J = S, gJ ≈ 2) is plotted in Figure 9.16 for
several values of S at temperatures of 1 K and 300 K. For curiosity it may be noted that for
S larger than ~10, the curve at 1 K would become indistinguishable from the Langevin
function (Appendix G) derived without quantization as if the levels were continuous.
Figure 9.16 also illustrates that at 300 K the relative magnetization is very small even at
very strong fields and follows a linear function of B.

28
If N is per unit volume, the dimensionless volume susceptibility χv is obtained, if per mol, χmol is obtained, etc.
 9.6 Paramagnetism 371

Figure 9.16 Relative magnetization Mr = M/Msat versus an applied field B at 1 K and 300 K, calculated
with the Brillouin function for a spin-only case and several values of S.

9.6.2 Pauli Paramagnetism

Curie–Weiss paramagnetism is applicable to substances where the unpaired electrons are
localized. When this is not the case, the temperature-independent Pauli paramagnetism can be
observed. It occurs when the electrons can be reasonably well described as a gas of non-
interacting particles. This picture is only appropriate in metallic conductors where the
valence electrons are highly delocalized, such as in alkali-metals and Mg, Ca, Sr, Ba, and
Al. Other metallic elements are either diamagnetic, like Cu and Bi, or have more complicated
magnetism due to interactions between electrons.
We can understand the origins of Pauli paramagnetism through the schematic density-of-
states (DOS) picture in Figure 9.17, where we draw separate bands for electrons with
moments29 up (↑) and down (↓). In the absence of a magnetic field, the energy of electrons
in each of these sub-bands will be identical and there will be the same number of ↑ electrons
as ↓. In the presence of a magnetic field, the energy of the spin moments parallel to it will sink
by μBB, the energy of the single-electron moment, and the energy of those that are antiparal-
lel will increase by the same amount.
The Pauli paramagnetic moment for the electron-gas model is small, but still three times
larger than the opposite Landau diamagnetic moment of those electrons. It is much smaller,
though, than the moment of Curie paramagnets.30 In many transition metals, an enhanced
Pauli paramagnetism occurs, which varies slightly with temperature. It is associated with

29
The spin magnetic moment is opposite to the spin angular momentum (Figure 9.3).
30
The reason is as follows: Starting with Equation (9.19) for susceptibility under Curie law, for a free electron gas, one
calculates a volume susceptibility with an NV of the conducting electrons per unit volume, replaces the thermal
energy kT with the much higher Fermi energy EF (a different statistics also applies—one of Fermi–Dirac, not that of
Boltzman), and the factor ⅓ with ⅔. The high EF is also the reason why Pauli paramagnetism is temperature
independent; the kT energy is very small compared to EF.
372 Magnetic Materials

E E E
B
EF

B
N(E) N(E) B N(E)

Figure 9.17 A schematic DOS plot (on the left, with Fermi level EF) can be separated into sub-bands of
spin-up and spin-down moments (center). The magnetic field B will change the relative energies and
populations of these sub-bands (right).

weak interaction between magnetic moments of the conducting electrons, and this slightly
favors their parallel arrangement. More or less enhanced Pauli paramagnetism also occurs in
many metallic compounds as an alternative to an outright ferromagnetism, for example in
RuO2.

9.7 Antiferromagnetism
Antiferromagnetic materials obey the Curie–Weiss law, Equation (9.18), at high temperature
where they behave as paramagnets with a negative value of θ. On cooling through the Néel
temperature, TN, their magnetic moments align in an antiparallel fashion. Because of this
interaction among the moments, the effect of an external field is smaller on an antiferro-
magnet than a paramagnet, and its susceptibility, while positive, is smaller than in the
paramagnetic state close to TN. Plots of χ and χ−1 versus temperature are shown in
Figure 9.18.
Antiferromagnetic materials also show a significant anisotropy below TN. The suscepti-
bility when the external field is parallel to the moments (χ||) is considerably smaller than when
the field is perpendicular to them (χ⊥), Figure 9.18.31 A powdered sample will show
a weighted average ⅓χ|| + ⅔χ⊥ of those two lines. We can understand this anisotropy by
considering how the thermal agitation of the spins will be affected when the field is applied
either parallel or perpendicular to the antiferromagnetic alignment. Let’s start at 0 K, assume
no thermal agitation and hence a fixed direction of the opposing moments. If a weak external
magnetic field is applied along these moments, it cannot change the already saturated and
compensating moments, and χ|| is zero. Above 0 K, thermal agitation decreases the magnet-
ization in both sets of opposing moments equally in zero field. When an external field is
applied, it strengthens the set of moments parallel to it (by decreasing the amplitude of their
thermal agitation) and weakens the set of antiparallel moments (by increasing the amplitude
of their thermal agitation). This means that χ|| increases with increasing T. If the weak
external field is applied at 0 K across the antiferomagnetic moments, it tilts both sets of

31
Ferromagnets also possess anisotropy of magnetization, but its manifestation is hidden by ferromagnetic domain
formation.
 9.7 Antiferromagnetism 373

χ
χ⊥ χ= C
T–θ

χ θ

0 TN T
χ –1

χ –1

χ –1⊥

antiferromagnet paramagnet
θ 0 TN T

Figure 9.18 Susceptibility of an antiferromagnet as a function of temperature.

moments somewhat into its direction, producing a non-zero χ⊥. On increasing temperature,
the external field influences the increasing thermal amplitudes of both sets equally and the
total tilting is not affected. The magnitude of χ⊥ remains constant up to TN. The behavior
shown in Figure 9.18 is valid for a weak external field. In the presence of a very strong
external field, stronger than the antiferromagnetic interactions, the behavior is different. In
the perpendicular direction, the tilt increases until all moments are aligned with the field,
whereas in the parallel direction, the opposing moments at some point will flip to yield
a ferromagnetic state.
There are numerous ways in which magnetic moments can be arranged to give an
antiferromagnetic pattern, even for simple structure types [3]. Figure 9.19 shows the possible
ordering patterns for a primitive cubic arrangement of magnetic atoms.32 It’s worth noting
that ordering of antiferromagnetic moments leads to a magnetic unit cell that is larger than
the chemical unit cell. This is probed by neutron diffraction as explained in Box 9.1.
In addition to simple spin arrangements of the type shown in Figure 9.19, more complex
possibilities exist. In some materials, spin alignments follow a helical arrangement as one
moves through the structure, like the steps of a spiral staircase. The pitch of the helix may be
such that the spin arrangement repeats after a few unit cells (e.g. with steps at a 120° angle, the
spin would repeat after three chemical unit cells) or it may be that there is no simple registry
such that the magnetic structure is incommensurate (Section 2.10) with the chemical structure.

32
We’ll for example see in Chapter 11 that CaMnO3 adopts type G while LaMnO3 the type A arrangement, and an
important group of materials is obtained when these two form a solid solution.
374 Magnetic Materials

G type C type A type

D type E type F type

Figure 9.19 Types of antiferromagnetic order on a P cubic lattice of magnetic atoms. One of many
possible moment directions is drawn; more generally, a + and − would replace the arrows. The chemical
cell is in a thick dotted line. The magnetic unit cell is drawn for the G type in bold as an example.

9.8 Superexchange Interactions

Antiferromagnetic ordering turns out to be very common in insulating transition-metal oxides
and fluorides. For example, monoxides of Mn to Ni order antiferromagnetically below the
Néel temperatures given in Table 9.5. The moments of metal atoms alternate their up/down
orientation in all three directions. The magnetic information leading to this order is transmit-
ted by the nonmagnetic linking atoms in a process called superexchange [4, 5, 6].
The chemically most intuitive description of superexchange in oxides and fluorides is the
partial covalent model. We’ll assume initially a d-metal M cation (Lewis acid) and a p6 anion
L (“ligands” O2− or F−, Lewis bases) in an M–L–M bonding segment. We’ve seen in Chapter 5
that eg orbitals on M have the correct symmetry for strong σ overlap33 with a p orbital on L. If
we consider the straight 180° interaction d 3–L–d 3 of Figure 9.20 (top) we can therefore
envisage a partial transfer of one of the ligand p electrons to an eg orbital on the metal on
the left (a polar covalent bond). Hund’s first rule suggests that the M t2g electrons will have the
same spin as that shared with L, here ↑. Coming to the M on the right, the same p orbital of
33
For σ bond overlap, the overlapping portion must not change sign when coordinate axes of any of the two involved
orbitals are rotated around the line connecting the two atoms. Such σ overlaps are pz with dz2, and dx2−y2 with either
px or py. Such π overlaps along x are py with dxy, and pz with dxz. Along y, they are px with dxy, and pz with dyz. Along
z, they are px with dxz, and py with dyz. Orthogonal, non-overlapping, orbitals are px with either dxz or dxy along x, py
with either dxy or dyz along y, and pz with either dxz or dyz along z.
 9.8 Superexchange Interactions 375

Table 9.5 Antiferromagnetic NaCl-type oxides and KMF3 perovskites of M = Mn–Ni.

Phase TN (K) dM–O (Å) Phase TN (K) dM–F (Å)

MnO 122 2.22 KMnF3 88 2.09

FeO 198 2.14 KFeF3 115 2.06
CoO 293 2.13 KCoF3 135 2.04
NiO 523 2.09 KNiF3 275 2.01

P P

d d

P P

d d

P P

d d

Figure 9.20 Superexchange M–L–M 180° coupling with eg orbitals of M. L is a p6 anion, M is a d n cation
in a weak Δ-splitting octahedral L field. The corresponding spin-up and spin-down levels are separated by
the spin-pairing energy P. Electron-spin sharing (partially filled arrows on L) corresponds to partial
presence of the L p electron at the cation owing to partial covalency of the M–L bond.

L shares its ↓ electron, making the unpaired t2g electrons ↓ on that M. There is thus an overall
antiferromagnetic coupling of magnetic moments on the two metal sites, M(↑)(↑)L(↓)(↓)M.
Overlap arguments suggest that a decrease in the bond angle from 180° will reduce the strength
of this antiferromagnetic superexchange.
A similar situation occurs for a high-spin d 5–L–d 5 superexchange (Figure 9.20, middle)
though now the Pauli exclusion principle dictates that the spin on M is opposite to that on L,
which leads to M(↓)(↑)L(↓)(↑)M that is again antiferromagnetic. For a d 3–L–d 5 case,
376 Magnetic Materials

Figure 9.21 The right-angle superexchange of d 3 cations M in a weak octahedral field of p6 anions L. The
two orthogonal p orbitals on L are drawn one in black and one in gray.

a combination of these arguments predicts ferromagnetic coupling, M(↑)(↑)L(↓)(↑)M, as shown

in Figure 9.20 (bottom).
These simple ideas explain at least in part why TN increases along the MO and KMF3
series in Table 9.5. From M = Mn to Co, the metal electronegativity increases, causing
covalency to increase, and this in turn increases the exchange interaction. The same argu-
ment explains why higher oxidation states tend to have higher TN (FeO, 198 K; versus
α-Fe2O3, 953 K) and why TN is generally higher for oxides than for fluorides.
When one has a 90° M–L–M angle, a slightly different argument holds. For our d 3–L–d 3 case,
we can again consider the partial transfer of an L ↑ electron towards the metal ion on the left
(Figure 9.21). However, a different p orbital is overlapping with M on the right. We can therefore
use Hund’s rule to state that the spins of the emerging holes on L will be parallel such that an L ↑
electron in this other p orbital is again used, making the overall coupling ferromagnetic. It should
be noted that the 90° superexchange is typically weaker than the 180° one.

Box 9.1 Characterizations: Neutron diffraction of magnetic materials

X-ray diffraction is one of the most powerful probes of crystal structure. When a crystal is
illuminated with X-rays of wavelength similar to interatomic separations, interference of X-rays
scattered from the electron clouds of atoms leads to radiation being diffracted in certain directions.
The direction of the diffracted radiation carries information on the size and shape of the unit cell of
the material (via Bragg’s law stating that the radiation wavelength λ = 2dhkl sinθ, where θ is the
glancing angle between the beam and the atomic plane hkl). Intensities of the diffracted radiation
peaks carry information on where atoms lie within that cell. Neutrons can also be produced (either at
reactor or spallation sources) with wavelengths comparable to interatomic distances. Neutrons are
scattered by atomic nuclei giving rise to diffraction patterns sensitive to isotopes. Since neutrons are
fermions, they have half-integer spin and will also be scattered by the spin the atom may have due to
its unpaired electrons (paired electrons not contributing). In contrast, diffraction of X-rays as
bosons of zero spin is generally insensitive to magnetic moment. For parallel spins of all identical
atoms (inside a domain), the diffracted Bragg peak will be a simple sum of the contribution from the
so-called “nuclear” and “magnetic” scattering. This is shown in the bottom row of the figure below,
where the filled peaks represent nuclear scattering and the open peaks, overall scattering. In the case
 9.9 Ferromagnetism 377

Box 9.1 (cont.)

of anti-parallel magnetic moments, neutrons will “see” two different types of atoms, the spin-ups
and spin-downs. Diffraction from magnetic unit cells that are multiples of the “chemical cell” (also
known as the “nuclear cell”) will produce additional peaks solely due to spin ordering. These
“magnetic” peaks are marked below by the black hkl indices for three of the antiferromagnetic
ordering types shown in Figure 9.19 (while the nuclear peaks have gray indices referring to the unit
cell drawn). This contribution to the diffraction pattern from magnetic scattering means that
neutron diffraction can be used to determine magnetic structures of ordered moments.

9.9 Ferromagnetism
Ferromagnetism is the magnetic behavior of parallel orientations of atomic magnetic
moments (Figure 9.14). Ferromagnets order their moments below a critical temperature
called the Curie temperature, TC. The total atomic magnetic moment exhibited by such
materials is called spontaneous magnetization.
From this description we might expect that all ferromagnetic materials, such as all pieces of
iron, would have a permanent magnetization below TC and attract each other—which we know
isn’t the case. The reason for this discrepancy lies in the domain structure of ferromagnets. If all
atomic moments in a material point in the same direction, it behaves like the bar magnet of
Figure 9.1 and needs to send a powerful magnetic field through its neighboring space. This costs
378 Magnetic Materials

Figure 9.22 The reason why ferromagnetic materials adopt domain structure is to avoid sending the field
into the poorly magnetizable neighborhood.

M M
M M
M

H M
H
H H
H
H

Figure 9.23 Hysteresis loops for hard and soft ferromagnets.

energy (called the magnetostatic energy). This energy can be reduced by forming a structure of
closed magnetic loops inside the magnet, as shown in Figure 9.22, which essentially avoids
sending the field outside the body of the magnet. Some details of domain formation have been
dealt with in Chapter 8. In short, it represents a compromise between (a) the magnetostatic
energy, (b) the magnetocrystalline energy that arises because it is normally more favorable to
align spins in some crystallographic directions (the easy axes) than others, and (c) the magne-
tostrictive energy of strain caused by magnetostriction, the contraction along that favorable
direction of magnetization where the spins attract themselves. Boundaries between domains are
called Bloch walls. For Fe at room temperature, these might be around 650 Å thick.
From this description, one may again wonder why the domain-structured ferromagnets can
become magnetic at all. The domain structure of a material may not be very strong; it only needs
to be more stable than the alternative of sending the field through the material’s surroundings.
An external magnetic field can therefore reorient domains and change their size. Favorable
domains with easy axes of magnetization close to the field direction will therefore grow at the
expense of less favorable ones, and then eventually reorient closer to the field direction as the
magnetic saturation is approached. Such changes in domain structure are behind the character-
istic M versus H hysteresis loops of ferromagnetic materials shown in Figure 9.23.
From the origin at zero for zero H, the magnetization M follows the light curve of
Figure 9.23. Saturation is easy and achieved at fields far lower than for an isolated paramag-
netic moment. High magnetization occurs already at a low field. Although there is an
 9.9 Ferromagnetism 379

energetic cost associated with moving domain walls past dislocations or defects in a material,
the external field is able to overcome this energy. When the external field is removed, the new
domain structure in the material may become partially locked by defects and dislocations
rearranged by magnetostriction. On reducing the field to zero, the material then retains
remanent34 magnetization, Mremanent.35 This type of domain realignment is the experiment
one performs when “stroking” a needle with a permanent magnet to magnetize it. Permanent
magnets are manufactured commercially by cooling a glowing hot material in a strong
permanent field that keeps the forming domains aligned. A high Mremanent is obtained
when the field is removed from the cold product.
Figure 9.23 shows that it then takes an opposite field, the negatively taken intrinsic36
coercive field, Hci [7], to fully demagnetize the ferromagnet so that M = 0. Ferromagnets with
easily movable domains have a low Hci and are called soft magnets. Ferromagnets with
a large Hci are called hard ferromagnets and have important applications as permanent
magnets in motors. For data storage, a high coercive field is not desirable because even
higher fields are then needed to rewrite the information. On the other hand, high remanent
magnetization is essential. The hysteresis loop for a data-storage material should therefore
approach the form of a narrow and tall rectangle. In contrast, good permanent magnets have
hysteresis loops that are both tall and wide. One useful way of representing magnetic
measurements on such magnets is to plot B versus H. The maximum energy product,
(BH)max (in units of T A/m = J/m3), can then be defined, as shown in Figure 9.24, via the
point on the demagnetization line where the product of B and H is a maximum.
The temperature dependence of the magnetization of a ferromagnet is shown in
Figure 9.25. Below TC, it resembles the curve for the order–disorder transition37 of second
order (Figure 4.16). Above TC, the magnetization becomes so low that it would not be seen
on the given scale of M, and the plot in Figure 9.25 is continued with reciprocal
susceptibility38 instead, following the Curie–Weiss law of Equation (9.18).
Let’s take first the T > TC side of the plot in Figure 9.25, where the originally ferromagnetic
material is thermally disordered into a paramagnet and obeys the Curie–Weiss law. Weiss, in
his phenomenological description of ferromagnetism, assumed that the ferromagnetic align-
ment of atomic moments is transmitted by an internal field, HW, between these moments,39
the strength of which is proportional to the sample magnetization:

HW ¼ λM (9.22)

34
An older version of “remnant”. Now used only in this physics meaning. It is best not to confuse the two.
35
The demagnetizing field (the field lines external to the sample in Figure 9.22) is much weaker than an external field
and can’t move domain walls past all defects.
36
The intrinsic coercive field Hci (also called intrinsic coercivity) zeroes the magnetization M, whereas the coercive
field, Hc (coercivity), zeroes the magnetic induction B, B = μ0(M + H).
37
This approach to describing M(T) is illustrated in Problem 4.12.
38
Because χ−1 = H/M, Equation (9.8), it is as though M−1 were plotted for unity field.
39
Weiss called this field a molecular field. The field is constant throughout the material; it represents a mean-field
approximation of the quantum-mechanical exchange interactions between the atomic moments.
380 Magnetic Materials

MH BH

M B
M M M
BH

H H
H H

H B µM
B µ H M B H M

BH H M

BH µM B µ

Figure 9.24 Maximum energy product for an idealized ferromagnet in SI units. See Footnote 36 of this
chapter for the coercive-field symbols.

T T

Figure 9.25 Magnetization M of a ferromagnet as a function of temperature. Paramagnetic reciprocal

susceptibility is plotted after M drops to very low values at TC. Note that only when the applied field H →
0 will M fall sharply to zero at TC.

with λ being called the internal-field constant. The total field that magnetizes the moments is
Htotal = H + HW. Let’s recollect that the Curie law is derived on the assumption that magnetic
moments do not interact with each other. How would the law change when the magnetiza-
tion is almost entirely due to the field these very moments produce? Using Htotal in Equation
(9.8), we write for the inherent total paramagnetic volume susceptibility under Curie law:

M M C
χinherent ¼ ¼ ¼ (9.23)
Htotal H þ λM T

Then we take the latter identity, and express the measured susceptibility χ = M/H,

M C
χ¼ ¼ (9.24)
H T  Cλ
 9.9 Ferromagnetism 381

which has the same form as the Curie–Weiss law in Equation (9.18), only Cλ replaces the
temperature θ ≈ TC above which a ferromagnetic material magnetizes like a paramagnet. The
expression TC ≈ Cλ means that TC is a measure of the strength parameter λ of the field that
causes the moments to align parallel; it is a measure of the exchange field.
Let’s now turn to the T < TC side of the plot in Figure 9.25 and consider the temperature
dependence of the ferromagnetic magnetization. The descending curve originates in the
Brillouin function for the relative magnetization Mr, Equation (9.21), of paramagnets. In
order to apply it to a ferromagnet, B is replaced with μ0Htotal, where Htotal = H + HW, as
above. However, HW is a function of M, Equation (9.22), hence of Mr, and it turns out that
these two functions for Mr(T) of a ferromagnet (Appendix H) can only be solved numeric-
ally. The solution yields the simple curve in Figure 9.25.
We haven’t yet addressed the materials-chemistry question of what makes a good ferro-
magnet; that is, what are the forces that cause magnetic moments to align? Before discussing
this, it’s useful to separate ferromagnets into three different categories. One way of doing this
is via schematic DOS plots of the type we used in Figure 9.17, where we split the DOS to show
the two sub-bands for the majority/minority spins with magnetic moments parallel/antipar-
allel to the field. These are often called spin-polarized bands. This viewpoint allows us to split
materials into categories of ferromagnetic metals (with DOS non-zero at EF for both majority
and minority spins), ferromagnetic half-metals (with the majority spin only at EF, a gap
occurs for minority spins), and the rare case of ferromagnetic insulators (with no spin at EF).
as shown in Figure 9.26.

9.9.1 Ferromagnetic Insulators and Half-Metals

Ferromagnetic insulators (semiconductors) are rare. Examples are weak ferromagnets of low
TC such as EuO (TC = 69 K) and EuS (16 K) of rock-salt structure, a PbCl2-type EuH2
(18 K), EuLiH3 cubic perovskite (37 K), or the transparent K2NiF4-type salt Rb2CrCl4
(52 K). The nonmetallic ferromagnet La2MnIVNiIIO6 (TC = 280 K [8]) arises from the high-
spin d 3–L–d 8 superexchange combination (Section 9.8). An obtuse-angle superexchange
leads to weak ferromagnetism in highly tilted perovskites YTiO3 (TC = 30 K) and SeCuO3

E E

ferromagnetic ferromagnetic ferromagnetic

-

Figure 9.26 Types of ferromagnets according to schematic spin-polarized bands.

382 Magnetic Materials

(26 K) or in the CdCr2Se4 spinel (130 K). These materials are insulating or semiconducting
above TC (EuO is yellow) but their conductivity increases below TC by many orders of
magnitude. The reason is the easy hopping of electrons between sites of the same spin.
Half-metals, as illustrated in Figure 9.26 (center), are solids that for one spin orientation
have a gap, whereas the other orientation is gapless, like in a metal. Half-metals are
conductors of spin-polarized electrons. The spin orientation also means that all half-
metals are magnetically ordered. A typical half-metal is CrO2, with TC = 392 K and an
integer spin-only ferromagnetic moment of 2.0 µB per Cr at 0 K. This saturation moment
corresponds to complete polarization of the two nonbonding t2g electrons present in stoi-
chiometric CrO2, as required by the half-metal definition. More about this in Chapter 11.
There are several origins of ferromagnetism in nonmetallic materials. One is the ferromagnetic
superexchange detailed in Section 9.8. Another is an absence of antiferromagnetic super-
exchange; a weak ferromagnetism (of low TC) may then appear due to direct interaction of
the two not-so-distant magnetic atoms along the array M–L–M. Another factor contributing
to ferromagnetism (at least locally) is double exchange in valence-mixed compounds (Chapter
11), where an electron is present simultaneously at two or more neighboring sites of an element.
This requires a parallel orientation of spins at the sites involved. The term double exchange was
chosen [9] for cases where the imagined hopping happens over an L atom in between the two
sites. A typical example is magnetite, Fe3O4 (see also Section 9.10), in which minority-spin t2g1
electrons of octahedrally coordinated Fe2+ hop into the same yet empty orbital of Fe3+ at an
equivalent octahedral site. Another example is La2/3Sr1/3MnO3, where the eg1 electron of Mn3+
hops to Mn4+. With close enough energies, even orbitals of two different metals may give rise
to double exchange and ferromagnetism. An example is sharing of the t2g1 Fe2+ electron with
the d0 configuration of Mo6+ in the half-metallic ferromagnet Sr2FeMoO6 of the perovskite-
related structure [10]. Because electron hopping is involved in stabilizing ferromagnetic inter-
actions, double-exchange ferromagnets are close to being metallic.

9.9.2 Ferromagnetic Metals

Ferromagnetism well above room temperature is displayed by three elements: Fe, Co, and Ni
with TC = 1043 K, 1388 K, and 627 K, respectively. Their magnetic coupling is due to
conduction electrons, and the phenomenon is called itinerant ferromagnetism. The origin of
the parallel coupling of magnetic moments is the quantum-mechanical exchange interaction
that is also behind the Hund’s first rule (maximum spin multiplicity). We can rationalize why
Fe, Co, and Ni display spontaneous magnetization, whereas Cu and Zn don’t by using a simple
band-structure argument similar to that used to explain Pauli paramagnetism. In the absence
of exchange interactions, the ↑ and ↓ spin levels are at equal energies and therefore equally
populated. In the presence of an exchange, one sub-band is stabilized relative to the other, and
hence filled with more electrons. Overall, this means that the gain in exchange energy must be
balanced against the energy cost to move electrons from levels just below EF in the thus-
emerging minority-spin band to the top of the thus-emerging majority-spin band. The question
 9.9 Ferromagnetism 383

s d T d T

E
E

s s s

NE NE NE NE NE NE

Figure 9.27 Comparison of a spin-polarized DOS plot at non-zero T for diamagnetic Cu (idealized Cu-
atom configuration 4s13d10) with one for ferromagnetic Ni (idealized Ni-atom configuration 3d10),
showing the effect of the exchange interaction below TC.

of whether a metal will be magnetic depends in large part on this balance. When the number of
states N(EF) at the Fermi level is high, the electrons move to states that are only marginally
higher in energy, and the cost is low. This is the case for Co, Fe, and Ni, where the Fermi level
cuts through the d bands. When N(EF) is small, the electrons transferred to majority-spin
bands are forced to occupy states that are significantly more antibonding, and the cost is high.
This is the case for metals like Cu, where the Fermi level cuts through the s band. By way of
example, a schematic DOS plot is shown in Figure 9.27.
For Cu, EF lies in the broad 4s band of 1 electron per atom, where the DOS is low (Figure
9.27, left). The energy cost of flipping that electron and putting it onto the other sub-band is
too large to be offset by any exchange energy, and Cu therefore doesn’t show ferromagnet-
ism. For Ni, one has a total of 10 valence electrons. Of these, around 9.4 reside in the d band
and 0.6 in the s band. The exchange splitting is such that the majority-spin band is completely
filled40 and the 0.6 holes lie entirely in the minority band. The saturation magnetization per
Ni atom is therefore 0.6 μB. This simple band-structure picture helps rationalize why
magnetic moments recorded experimentally don’t suggest an integer number of electrons
contributing to the moment. Similar arguments hold for Co and Fe, which have fewer
valence electrons than Ni, increasing the excess of spin-polarized electrons and increasing
Msat. The picture is very similar to that used to explain Pauli paramagnetism, but the origin
of the energy splitting of sub-bands is the internal exchange energy rather than an external
magnetic field. An element that sits on the fence is Pd. It has N(EF) nearly high enough for
spontaneous magnetization, and this leads to high magnetization in applied fields (an
enhanced Pauli paramagnetism). Importantly, since high N(EF) is associated with narrow
bands, a decrease in dimensionality leads to ferromagnetism in Pd films and nanoparticles.
Our simple picture of itinerant ferromagnetism of elements ends around Fe. Mn adopts
a different crystal structure, has broad d bands of low N(EF) that increases the cost of moving
40
Chemists may appreciate that with 9.4 d electrons a sub-band forms with full d 5 configuration.
384 Magnetic Materials

between the ↑ and ↓ spin levels such that Mn is not ferromagnetic. Ferromagnetism of
conduction electrons requires a high density of states at the Fermi level combined with strong
ferromagnetic exchange interactions favored by densest packing, and occurs for electron
counts that allow one spin-polarized sub-band to become fully populated at the expense of
the other sub-band (Figure 9.27). That also occurs in some intermetallic compounds when
their average valence-electron count per atom (VEC) approaches 10. An example is the
Laves phase ZrZn2 of VEC = 9.33 and TC = 28 K.

9.9.3 Superferromagnets
A combination of d-shell and f-shell magnetism in superferromagnets SmCo5 and Nd2Fe14B has
produced the strongest permanent magnets so far, i.e. those with the highest maximum-energy
products (BH)max. Since their discovery in the 1960s, these materials have replaced
other permanent magnets in most applications. Table 9.6 shows that their coercivities exceed
those of traditional magnets. The strength of superferromagnets may be illustrated by the fact
that a cube-shaped magnet of 1 inch edge cannot be pulled straight from an iron plate by hand;
a task comparable to lifting a mass of 50 kg. The best ceramic magnet, BaFe12O19, of the same
size would behave like a 3 kg mass.
What are the origins of such exceptional properties? For SmCo5 they can be traced back to
the magnetic structure shown in Figure 9.28. The high remanence is because the Co moment
is high, 1.8 μB at 4 K. The Sm moment of 0.38 μB is parallel but decreases to about zero at
room temperature, while the ordered Co moments persist up to the very high TC of these
magnets.41 One can think of the Sm “stuffing” atom as causing just a slight reduction in TC

Table 9.6 Magnetic properties of some anonymized commercial polycrystalline hard

magnets: Bremanent (remanence), Hc (coercivity), energy product (BH)max, and TC.

Composition Bremanent (T) Hc (kA/m) (BH)max (kJ/m3) TC (K)

~SmCo5 1.05 730 200 1100

~Nd2Fe14B 1.23 900 280 550
Al7Ni15Co35Cu4Ti5Fe34 1.05 62 70 1150
BaFe12O19 0.40 190 9 742

41
The moment magnitudes can be accounted for by considering spin–orbit coupling and localized moments. For Sm2+
(4f65s2p6), the calculated saturated moment μsat = gJ (Table 9.3) is 0 μB, based on S = 3 and L = 3 from the Russell–
Saunders scheme in Figure 9.13 and J = |L − S|. However, J approaching zero partially uncouples L and S and
brings a second-order Zeeman effect (see Orchard in Further Reading) contribution to the magnetic moment, and
the moment becomes non-zero (even at low temperatures). Treating SmCo5 as a Zintl phase (Section 1.5.6), one can
speculate that formation of Sm2+ provides two electrons to five Co. The average saturation moment for the thus
formed 3d9.4 cobaltide configuration in this compound is gJ = 1.80 μB, calculated with S = 0.3 and L = 1.2,
interpolated via the Russell–Saunders scheme of Figure 9.9.
 9.10 Ferrimagnetism 385

Figure 9.28 Magnetic structure of SmCo5 at room temperature.

relative to the 1388 K of hexagonal close-packed Co. The key role of Sm is in the extremely
high magnetocrystalline anisotropy it induces. It triggers a combination of crystal-field
effects and spin–orbit coupling that makes magnetization very easy in one direction and
very difficult in others (high coercivity). When such magnets are manufactured commer-
cially, intentional phase admixtures and careful processing are used to orient grains, control
grain boundary effects, and help pin magnetic domains.

9.10 Ferrimagnetism
In Section 9.8 we’ve described how superexchange gives rise to antiferromagnetic coupling in
insulating materials. In Section 9.9 we learned that ferromagnetic insulators are extremely
rare. How then do we produce an insulating material with ferromagnet-like properties? One
solution is to use a ferrimagnet (Figure 9.29) where antiferromagnetic coupling of opposing
moments of unequal magnitude (e.g. due to two different metals being involved) leads to
a net overall magnetization. While bulk properties are similar to a ferromagnet, the atomic-
scale interactions are more complex. A ferrimagnet of two magnetic atoms may have up to
three types of couplings, one dominant antiferromagnetic and two ferromagnetic.42 This
takes away the simplicity of the Weiss approach, which would have to incorporate three
Weiss fields between two moments of unequal magnitude, producing an asymptotic
approach to Curie–Weiss behavior at high temperatures. The behavior below TC is also
more complicated, as each of the two relative saturation moments will follow their own
temperature dependence.43 Experimentally, a linear Curie–Weiss plot is obtained in the high-
temperature extrapolation, having negative θ indicative of the antiferromagnetic coupling.

42
The white–gray antiferromagnetic interaction and the white–white or gray–gray ferromagnetic interactions of
Figure 9.29 (left) cartoon.
43
At some point, the two moments may even cancel; the point is termed the compensation temperature.
386 Magnetic Materials

Tc T

Figure 9.29 Magnetization M of a ferrimagnet as a function of temperature T. Reciprocal susceptibility is

plotted after magnetization becomes very low above TC.

On cooling towards TC, there is a marked increase in χ, followed by an onset of spontaneous

magnetization at TC.
Ferrimagnetism is common among spinel-type materials; indeed the ferrimagnetic spinel
Fe3O4 (lodestone ≡ leading stone) with a TC of 860 K was the technological solution the
ancient Chinese either employed directly in compasses or used to magnetize needles. The Fe3
O4 spinel (Section 1.5.1) has antiferromagnetic coupling between the tetrahedral and octa-
hedral sites, [Fe3+(↓)]tet[Fe3+Fe2+(↑)]octO4. Since magnetic moments of Fe3+ at both sites
cancel, the saturation moment is solely due to the 4 unpaired electrons of high-spin d 6 Fe2+,
hence 4ge(½) = 4 µB. The ability of the spinel structure to accommodate various elements
yields a variety of magnetic properties as well as the ability to control the saturation moment
via substitutions.
Another group of important ferrimagnetic materials has the garnet-type structure
(Section 1.5.2). Its simplest representation is [cube]3[octahedron]2[tetrahedron]3O12.
Let’s choose Gd3Fe5O12 as one example of many. Its “magnetic” crystal-chemical
formula can be summarized as [Gd3+(↑)]3[8c][Fe3+(↑)]2[6o][Fe3+(↓)]3[4t]O12. Two of
the d 5 tetrahedral Fe3+ moments are cancelled by the d 5 (both are S = ⁵⁄₂ ) octahedral
Fe3+ such that the overall saturation moment is given by 3|μ(Gd3+)| − |μ(Fe3+)|.
Garnets can be prepared as large single crystals that are stable, colored, translucent
insulators with a bulk ferromagnetic-like behavior. This unique combination of prop-
erties is utilized in magneto-optical devices: disk memories, converters of magnetic field
into optical images, and sensors of movement or rotation. Garnets are also used in
optoelectronics and microwave technology as semi-permeable transmission media called
Faraday rotators.44

44
The Faraday effect is a usually tiny rotation of the plane of polarized light upon transmission through a transparent
magnet, the angle of which is proportional to the magnetic-field component in the direction of the light beam. The
angle achievable in garnets is rather high, around 45°. An analogous effect occurring upon reflection from surfaces
of magnetized crystals is called the Kerr magneto-optical effect.
 9.11 Frustrated Systems and Spin Glasses 387

9.11 Frustrated Systems and Spin Glasses

In earlier sections, we’ve looked at exchange interactions that lead to ferro- or antiferromag-
netic ordering. In some situations, it may not be possible to simultaneously satisfy the
energetic requirements for all components in the system. This is called frustration and can
lead to a variety of energetically equivalent ground states being present. One of the best-
known magnetic examples of this occurs on the so-called kagome lattice made up of corner-
sharing triangles.45 With antiferromagnetic coupling between moments on two corners of
the triangle, there is an immediate problem when placing a moment on the third corner
(Figure 9.30). Although the kagome frustration has been treated theoretically in consider-
able detail, only a few real examples exist. One is the hydronium jarosite, (H3O)Fe3(SO4)2
(OH)6 [11]. A similar frustration occurs for tetrahedral networks. The tetrahedral frustration
is often referred to as “spin ice” by topological analogy with the disordered arrangement of
protons in hydrogen bonds of H2O(s) described by Pauling [12]. In general, the amount of
frustration is often quantified by the ratio of the Weiss temperature θ obtained from the
Curie–Weiss high-temperature behavior and the freezing/ordering temperature such as Tg
(freezing to a glass of spins, see below) or TN (antiferromagnetic ordering). The
|θ/Tfreezing/ordering| will be large for a frustrated material (spin-ice freezing) and close to 1
for a perfect antiferromagnet (spin ordering).
A second way to achieve frustration is via random site occupation in alloys. In metallic
alloys containing a few percent of a magnetic dopant such as Fe in Au, Mn in Cu, or Er in Y,
exchange via the host electrons leads to random interactions between the magnetic
dopants.46 Similar effects occur in inherently disordered alloys such as amorphous GdAl2.
A spin glass is a material formed by cooling the multi-degenerate spin-frustrated ground
state through a “freezing” transition47 into a glass-like, fixed, residual spin disorder. The

Figure 9.30 Magnetic frustration. While it’s possible to have all nearest-neighbor antiferromagnetic
interactions satisfied on a square, this is not possible for a triangle. Materials based on the kagome lattice
(a segment is shown on right) then show spin-glass behavior.
45
Kago-me = “basket-eye” in Japanese, a light basket of bamboo strains woven such that large hexagonal openings,
“eyes”, and small triangles emerge.
46
Below a certain temperature, further cooling begins to decrease the conductivity of the alloy—against what is
common for metals—as the increasing strength of the coupling condenses more conducting electrons around the
magnetic impurity (the Kondo effect).
47
The term “freezing” is used in analogy with water-ice freezing upon intrinsic disorder of H-atom orientations
around O-atom nodes of the tetrahedral network.
388 Magnetic Materials

thermo- field-cooled
remanent

zero-field-cooled

0 10 102 103 0 Tg T

Figure 9.31 Right: Equilibrium magnetic susceptibility of a spin glass upon cooling through the spin-glass
temperature Tg with or without an external field. Left: Their isothermal relaxations as a function of time τ
after the external field was switched off or on, respectively.

temperature dependence of the susceptibility of a typical spin glass is shown in Figure 9.31
(right). At high temperature, the material is paramagnetic and follows the Curie–Weiss law.48
On cooling through Tg, the spins interlock into one of many possible ground states of the
different frustrated combinations. Below Tg, the susceptibility depends on the field under which
the material was cooled. The larger the field in which the material is cooled, the higher the
interlocked magnetization. The high-field-cooled susceptibility of the spin glass remains high as
a function of temperature. The zero-field-cooled susceptibility is low and decreases with
temperature as local antiferromagnetic interactions are less agitated. The conversion between
these two extreme interlocked results (the former created in zero field, the latter in a high field)
takes a long time (Figure 9.31). The relaxation times also depend on whether or not equilib-
rium has actually been reached upon the field- and no-field cooling below Tg. Magnetic
properties of a spin glass therefore depend strongly on its thermal history and are reminiscent
of physical properties of high-viscosity plastics or other amorphous materials (Chapter 15).
Spin glasses are characterized by a sharp peak in ac susceptibility measurements close to Tg.

9.12 Magnetoelectric Multiferroics

A magnetoelectric medium that becomes magnetized in an electric field, E, and electrically
polarized in magnetic field, H, is a largely unfulfilled idea that dates back to Pierre Curie at
the end of the nineteenth century. In particular, ferroelectric ferromagnets49 are of interest
[13]. Potential applications include data-storage devices, magnetic or electric modulation of

48
The paramagnetic state may or may not transform directly into the spin glass. If an intermediate ferromagnetic state
is formed upon cooling, the final spin glass at the lowest temperature is called reentrant spin glass, in reference to the
re-emerging magnetic disorder.
49
This is the most prominent pair of the so-called multiferroics, materials that combine two types of field-induced,
ferroic, alignment (of magnetic dipole moments due to a magnetic field, of electric dipole moments due to an electric
field, of strains by a stress field) thus achieving coveted cross-coupling phenomena when one field tunes the
alignment pertinent to the other field.
 9.13 Molecular and Organic Magnets 389

transducers, multifunctional sensors, and actuators. Yet making this happen in a single
chemical compound needs careful control of symmetry [14, 15] and is challenging compos-
itionally. Most ferromagnets have unpaired d electrons and are good electrical conductors
(we have already noted the rarity of insulating ferromagnets), while ferroelectrics need to be
insulators with non-centrosymmetric structures and are typically associated with the d0
configuration (Chapter 8). One possible solution is to mix dn and d0 ions in one network,
but this dilutes the ferroelectric and ferromagnetic subsets and weakens both the ferroelectric
coupling and magnetic superexchange. Loss of ferroelectricity ensues, with the dilute mag-
netic arrangement having a low ordering temperature, if any at all.
Another path is to introduce transition metals into the pyro- or ferroelectric salts of some
sp anions (typically non-centrosymmetric borates or phosphates). However, only weak
magnetic interactions are achieved, for example in ferroelectric, weakly ferromagnetic, and
ferroelastic boracites (M3B7O13X; M = transition metal, X = halogen).
The opposite approach is to stuff potentially ferromagnetic networks with cations that
have a stereochemically active lone pair, such as Pb2+, Bi3+, Sn2+, or Sb3+. Indeed, BiFeO3 is
a ferroelectric antiferromagnet with a weak ferromagnetic moment due to canting of spins.
A more robust ferromagnetism has been obtained by replacing La3+ in La2MnNiO6 with the
lone-pair cation Bi3+ to form Bi2MnNiO6 [16]. This leads to simultaneous ferroelectricity
and ferromagnetism below 140 K. A problem of this approach is that the ferromagnetism, or
its coupling with ferroelectricity, still is weak. How does one achieve the desired high
magnetoelectric coupling between properties?
A window of opportunity appears in geometrically frustrated networks where the bonding
compromise creates off-center positions for atoms in their coordination polyhedra. As an
example, in hexagonal YMnO3 (not a perovskite), off-center Y3+ cause ferroelectric polar-
ization that combines with a frustrated magnetic order of Mn3+ that is easily reoriented [17].
A related way of smuggling ferroelectricity into a magnetic material is to remove the
centrosymmetry of suitable valence-mixed magnetically ordered networks by ordering them
into integer valences. The loss of inversion symmetry that may occur upon such charge
ordering then gives rise to a weak ferroelectricity. The ferrimagnetic spinel magnetite, Fe3O4,
is reported to be ferroelectric below the Verwey transition at 125 K (see Chapter 11). An
analogous symmetry-breaking transition in the NdBaFe2O5 perovskite (Pmmm → P21ma)
removes inversion centers in this antiferromagnet [18]. A similar case occurs in LuFe2O4 [19].
However, a common problem with electrical polarization in these phases is their relatively
high electrical conductivity, which is undesirable for ferroelectric applications.

9.13 Molecular and Organic Magnets

The magnetic materials discussed so far were solid oxides or metals of high specific weight,
which have to be manufactured by high-temperature processing. The possibility of creating
molecular magnets that could be synthesized in solution and readily processed to a useful
390 Magnetic Materials

form is attractive. Unfortunately, only a small number of sp molecules have unpaired

electrons; examples include NO and O2 (in its ground state). Liquid oxygen is very strongly
attracted to magnetic fields, but as a liquid it cannot acquire any permanently oriented
moment on its own. Solid O2 is antiferromagnetic. Ozonides, such as [N(CH3)4]+ [O3]−
(stable up to 70 °C in the dark), contain one unpaired spin per anion, but their magnetic
properties have not been studied. Superoxides (with an O2− anion) also carry an unpaired
electron.
Other verified cases of sp magnetism involve free-radical species, such as NO2.
Unfortunately, molecules with unpaired electrons have a propensity for dimerizing and
thus losing their magnetism. Some organic free radicals are more stable, such as triphenyl-
methyl (C6H5)3C• and its high-spin polymer, but have not been shown to be ferromagnetic
[20]. The first organic ferromagnet was reported [21] in the orthorhombic β-modification of
para-nitrophenyl nitronyl nitroxide (p-NPNN) where the unpaired electron is carried by
a conjugated moiety (Figure 9.32), but the TC of 0.6 K is very low.
Some success in terms of bulk ferromagnetism has been achieved with transition-metal-
containing molecules [22]. As an example, decamethylferrocene FeII[C5(CH3)5]2, diamag-
netic due to low-spin iron, is oxidized by the electron acceptor tetracyanoethylene (TCNE,
Figure 9.33) into [FeIII(C5(CH3)5)2]+ and [TCNE]−. Here, the d 5 Fe of the cation carries one

Figure 9.32 The first ferromagnetic radical, p-NPNN.

Figure 9.33 Molecular structures: FeII(C5(CH3)5)2 and TCNE (left), MnTPP (middle), Mn12O48 cluster
of the “acetate-molecule” magnet described in text (right).
 9.14 Problems 391

unpaired electron, and the crystal becomes ferromagnetic below TC = 4.8 K [23]. It is also an
electrical insulator, and its ferromagnetism occurs despite no direct covalent bonding
between the metal and TCNE. Higher TC values do occur with covalent interactions. For
example, [MnTPP]+ [TCNE]− (MnTPP is a Mn3+ complex of tetraphenylporphyrin;
Figure 9.33) has TC = 16 K and a structure of planar cations having their Mn centers
covalently linked by nitrogens of TCNE into 1D chains, hence it is not a simple isolated
“molecule”.
There is considerable interest in single-molecule magnets (SMMs) containing one or
several metal centers with a fixed magnetization direction due to anisotropy within
a molecule, the environment of which also shields them from forming intermolecular
couplings. One way to achieve the anisotropy is a significant negative zero-field split-
ting (this occurs for the A2g-term d3/d 8; Table 9.3) such that the largest MS values
become lowest in energy, and S is as large as possible. Another way is through Jahn–
Teller distortion of d 4 or d 9 ions (Section 9.4.3). An example is the Mn12O48 cluster of
Mn4+ and Mn3+ coordination octahedra inside the Mn12O12(CH3COO)16(H2O)4 “acet-
ate molecule” (Figure 9.33, right), in which 90° O–Mn–O superexchange coupling of
the 8 Mn3+ ions is ferromagnetic, as is the coupling of 4 Mn4+ in the Mn4O4 central
cube. Their antiparallel orientation gives an S = 8∙2 − 4∙3/2 = 10 ground-state
ferrimagnet.
SMMs with a single metal center would be one more step towards the ultimately smallest
unit to store binary data written by an external magnetic field. The task is not trivial, as the
examples above suggest that these SMMs would keep their memory only at very low
temperatures. One of the most interesting candidates has therefore been based on Dy3+
with magnetism of its five unpaired electrons augmented by an orbital-momentum contribu-
tion (see Table 9.4). When Dy3+ is complexed with the large tris(tert-butyl)cyclopentadienyl
ligands that, together with bulky [B(C6F5)4]− anions, prevent intermolecular coupling, the
molecule shows magnetic hysteresis up to 60 K [24]. The magical border of the liquid-N2
temperature was crossed with hexakis(isopropyl)cyclopentadienyl ligands with magnetic
hysteresis up to 80 K [25].

9.14 Problems
9.1 Are penguins found at the south pole of the Earth’s magnetic field?
9.2 What is the orbital angular momentum for an s electron and why?
9.3 Convert the Earth’s magnetic field of 0.7 Oe to SI units.
9.4 What is the volume magnetization M and what is its SI unit?
9.5 Which 3d electron configurations (high- and low-spin where applicable) yield in octa-
hedral ligand field an orbital magnetic moment that is quenched or zero? State the
reason in each case.
392 Magnetic Materials

9.6 State whether or not you would expect an orbital contribution to the magnetic
moment of NiII in octahedral or tetrahedral coordination. Calculate the spin-only
moment μeff.
9.7 Determine the spin–orbit coupling ground-state terms 2S+1LJ for isolated Mn4+, Mn3+,
and Mn2+. Calculate μeff in Bohr magnetons, with and without an orbital contribution.
9.8 Show that, when the orbital momentum is quenched, Equation (9.14) simplifies to the
spin-only model.
9.9 Calculate effective and saturated spin-only magnetic moments (in μB) per Cr in
NH4Cr(SO4)2∙12H2O.
9.10 Calculate effective and saturated Russell–Saunders total moments (in μB) per high-spin
Fe in NH4Fe(SO4)2∙12H2O.
9.11 For NH4Fe(SO4)2∙12H2O: (a) Calculate the molar susceptibility at 298 K. (b) Given
that the cubic unit cell has a = 1.243 nm and contains four formula units, convert the
molar susceptibility to dimensionless χv. (c) Calculate the relative permeability.
9.12 Estimate the spin-only dimensionless susceptibility of O2 as an ideal gas at 273.15 K
and 100 kPa.
9.13 Superconductors expel a magnetic field from their bulk. In liquid N2, 102.5 mg of
a YBa2Cu3O6.9 powder of apparent density 3000 kg/m3 was placed into the center of
a miniature induction coil (a pick-up coil), and the induced voltage dropped by 450 mV.
At 26 °C, a 38.4 mg standard increased the (empty-coil) voltage by 5.73 V. Assuming
a linear voltage response to susceptibility, calculate the fraction of the magnetic field
expelled from the inside of the superconductor, given that the standard was a Curie–
Weiss paramagnet with Cm = 9.164×10−4 m3 K/kg and θ = −24 K. Hints: The expelled
field corresponds to the dimensionless (volume) diamagnetic susceptibility of the
superconductor. Because the samples were weighed, mass susceptibility must be calcu-
lated first.
9.14 Calculate the effective and saturated, spin-only, ge = 2, paramagnetic moments per
high-spin Mn in La0.5Sr0.5MnO3.
9.15 Calculate the effective paramagnetic moment per magnetic atom in DyCrO3.
9.16 Mass susceptibility of Sr2MnMoO6 (molar mass 422.114 g/mol) was measured as
a function of temperature, from which the Curie constant, Cm, was obtained by least-
squares fitting as Cm = 1.306×10−4 m3 K/kg. Determine the oxidation states of Mn and
Mo. Hints: Select an expression for the Curie constant C in Equation (9.19). Replace
μeff in it with μeff / μB so that μeff is now in Bohr magnetons. Assuming two magnetic
atoms per formula, Mn and Mo, calculate the number N of these atoms in one kilogram
of the sample.
9.17 The molar susceptibility χmol of the Prussian-blue CsFe[Cr(CN)6] Curie paramagnet is
0.0163 emu/mol at 300 K and 0.0094 emu/mol at 200 K (CGSem). Using the CGSem
formula μeff = 2.828√C in Bohr magnetons per formula unit of Appendix I, the Curie
law C = χT, and the spin-only as well as ge = 2 approximation, determine the spin state
of Fe2+.
 9.14 Problems 393

9.18 Calculate the value with which you divide molar magnetization Mmol in emu/mol in
order to get the amount of Bohr magnetons per mole.
9.19 Reciprocal mass susceptibility χm−1 for (NH4)2Fe(SO4)2∙6H2O (molar mass 392.139 g/
mol) at T = 100 K, 200 K, and 300 K was measured in a weak field as 1037000 kg/m3,
2075000 kg/m3, and 3112000 kg/m3, respectively. (a) Plot and fit T versus χm−1. What
type of magnetism do you see? Is it expected? (b) Assuming a spin-only moment, is iron
in a high- or low-spin state?
9.20 Reciprocal mass susceptibilities χm−1 for CoO at T = 400 K, 500 K, and 600 K in
a weak field are 2.000×106, 2.315×106, and 2.635×106 kg/m3, respectively. Plot
and fit T versus χm−1. Assuming a spin-only moment, is cobalt in a high- or low-
spin state? What type of magnetic order would you expect at low temperature
from the Weiss constant θ?
9.21 What is the percentage of magnetic saturation in an S = ½ paramagnet at 10 K under
a field of magnetic induction 1 T?
9.22 Estimate the magnetic induction BW and the intensity HW of the Weiss field for a spin-
only material of S = ½ that orders ferromagnetically at 300 K. How does BW (in teslas)
compare with fields generated in typical laboratory magnetometers?
9.23 Predict the type of magnetic coupling for the following materials assuming localized
electrons: (a) VO with the NaCl-type structure, (b) LaCrO3 with the perovskite
structure, (c) the hypothetical ordered perovskite La2CrIIIFeIIIO6, (d) cubic
CsNi[Cr(CN)6] with Ni and Cr like Na and Cl in rock salt and CN groups (as if one
p-group pseudoatom) in between each Ni and Cr.
9.24 Predict the type of cooperative magnetism in CrCl3 having 90° interactions.
9.25 α-MnS and MnO adopt the rock-salt structure. The Néel temperature of 130 K for α-
MnS is higher than 117 K observed for MnO despite the fact that the Mn2+ ions are
much farther apart in α-MnS. Suggest a factor responsible for this behavior.
9.26 Data suggest that the majority-spin 3d sub-band of Co is full. If so, how many electrons
are in the delocalized 4s band, given the saturated ferromagnetic moment of 1.751 μB
per Co and a g factor for Co of 2.17?
9.27 Transport measurements show Fe has 0.95 electrons in the delocalized s band. Given
a ferromagnetic moment of 2.22 μB per Fe and g factor 2.094, determine the filling in
the spin-polarized d sub-bands. Sketch an equivalent picture to Figure 9.27.
9.28 Would you expect Au4V to be ferromagnetic?
9.29 A sample of C60 has 1 mass ppm of Fe impurity. When heated at 700 °C under pressure,
C60 is converted to a carbon polymer and iron to Fe3C. Estimate the mass magnetiza-
tion in J/(T kg) (≡ A m2/kg) of the product first magnetized and then taken to the
Earth’s field. For Fe3C, assume Mremanent = 700 kA/m and the density ρ = 7000 kg/m3.
9.30 Derive the formula for the maximum energy product of an ideal ferromagnet in
Figure 9.24.
9.31 The spinel MnFe2O4 ≡ [(Mn2+)1−x(Fe3+)x]tet[(Fe3+)1−x/2(Mn2+)x/2]2octO4 has a saturated
moment independent of x. Explain why.
394 Magnetic Materials

9.32 Which of the following garnets is not diamagnetic: Ca3Te2Zn3O12, Y3Fe2Fe3O12,

Mg3Al2Si3O12, NaCa2Zn2V3O12?
9.33 What is the saturation magnetization, Msat, in Bohr magnetons per formula unit of
ferrimagnetic Y3Fe5O12 and Gd3Fe5O12 with Gd3+ in 4f75s2p6 configuration?
9.34 The so-called Cu6 ring (cluster) magnet has six Cu2+ coordination octahedra sharing
opposite edges of two oxygens. Which type of magnetism would you expect?

9.15 Further Reading

A.F. Orchard, “Magnetochemistry” (2003) Oxford University Press.
N.A. Spaldin, “Magnetic Materials: Fundamentals and Device Applications” (2003) Cambridge
University Press.
S. Blundell, “Magnetism in Condensed Matter” (2001) Oxford University Press.
R.L. Carlin, “Magnetochemistry” (1986) Springer-Verlag.
B.N. Figgis, M.A. Hitchman, “Ligand Field Theory and Its Applications” (2000) Wiley.
A.P. Guimarães, “Principles of Nanomagnetism” 2nd edition (2017) Springer.

9.16 References
[1] D. Nicholls, “Complexes and First-Row Transition Elements” (1979) Macmillan, p. 106–7.
[2] E.N. Caspi, J.D. Jorgensen, M.V. Lobanov, M. Greenblatt, “Structural disorder and magnetic
frustration in ALaMnMoO6 (A = Ba, Sr) double perovskites” Phys. Rev. B 67 (2003),
134431/1–11.
[3] E.O. Wollan, W.C. Koehler, “Neutron diffraction study of the magnetic properties of the series of
perovskite-type compounds La1−xCaxMnO3” Phys. Rev. 100 (1955), 545–563.
[4] J.B. Goodenough, “An interpretation of the magnetic properties of the perovskite-type mixed crystals
La1−xSrxCoO3−δ” J. Phys. Chem. Solids 6 (1958), 287–297.
[5] P.W. Anderson, “New approach to the theory of superexchange interactions” Phys. Rev. 115
(1959), 2–13.
[6] J. Kanamori, “Superexchange interaction and symmetry properties of electron orbitals” J. Phys.
Chem. Solids 10 (1959), 87–98.
[7] H.W.F. Sung, C. Rudowicz, “Physics behind the magnetic hysteresis loop: A survey of miscon-
ceptions in magnetism literature” J. Magn. Magn. Mater. 260 (2003), 250–260.
[8] N.S. Rogado, J. Li, A.W. Sleight, M.A. Subramanian, “Magnetocapacitance and magnetoresist-
ance near room temperature in a ferromagnetic semiconductor La2NiMnO6” Adv. Mater. 17
(2005), 2225–2227.
[9] C. Zener, “Interaction between the d shells in the transition metals” Phys. Rev. 81 (1951),
440–444.
[10] J. Lindén, T. Yamamoto, M. Karppinen, H. Yamauchi, T. Pietari, “Evidence for valence
fluctuation of Fe in Sr2FeMoO6−w double perovskite” Appl. Phys. Lett. 76 (2000), 2925–2927.
 9.16 References 395

[11] A.S. Wills, G.S. Oakley, D. Visser, J. Frunzke, A. Harrison, K.H. Andersen, “Short-range order
in the topological spin glass (D3O)Fe3(SO4)2(OD)6 using xyz polarized neutron diffraction”
Phys. Rev. B 64 (2001), 094436/1–8.
[12] L. Pauling, “The structure and entropy of ice and of other crystals with some randomness of
atomic arrangement” J. Am. Chem. Soc. 57 (1935), 2680–2684.
[13] D.I. Khomskii, “Multiferroics: Different ways to combine magnetism and ferroelectricity”
J. Magn. Magn. Mater. 306 (2006), 1–8.
[14] H. Schmid, “Some symmetry aspects of ferroics and single phase multiferroics” J. Phys.:
Condens. Matter 20 (2008), 434201/1–24.
[15] M.S. Senn, N.C. Bristowe, “A group-theoretical approach to enumerating magnetoelectric and
multiferroic couplings in perovskites” Acta Crystallogr. Sect. A 74 (2018), 308–321.
[16] M. Azuma, K. Takata, T. Saito, S. Ishiwata, Y. Shimakawa, M. Takano, “Designed ferromag-
netic, ferroelectric Bi2NiMnO6” J. Am. Chem. Soc. 127 (2005), 8889–8892.
[17] B.B. Van Aken, T.T.M. Palstra, A. Filippetti, N. Spaldin, “The origin of ferroelectricity in
magnetoelectric YMnO3” Nat. Mater. 3 (2004), 164–170.
[18] P.M. Woodward, E. Suard, P. Karen, “Structural tuning of charge, orbital, and spin ordering in
double-cell perovskite series between NdBaFe2O5 and HoBaFe2O5” J. Am. Chem. Soc. 125
(2003), 8889–8899.
[19] N. Ikeda, H. Ohsumi, K. Ohwada, K. Ishii, T. Inami, K. Kakurai, Y. Murakami, K. Yoshii,
S. Mori, Y. Horibe, H. Kito, “Ferroelectricity from iron valence ordering in the charge-frustrated
system LuFe2O4” Nature 436 (2005), 1136–1138.
[20] J.S. Miller, “Organic magnets: A history” Adv. Mater. 14 (2002), 1105–1110.
[21] M. Tamura, Y. Nakazawa, D. Shiomi, K. Nozawa, Y. Hosokoshi, M. Ishikawa, M. Takahashi,
M. Kinoshita, “Bulk ferromagnetism in the β-phase crystal of the p-nitrophenyl nitronyl nitr-
oxide radical” Chem. Phys. Lett. 186 (1991), 401–404.
[22] J.S. Miller, “Magnetically ordered molecule-based materials” MRS Bull. 32 (2007), 549–555.
[23] J.-H. Her, P.W. Stephens, J. Ribas-Arin, J.J. Novoa, W.W. Shum, J.S. Miller, “Structure and
magnetic interactions in the organic-based ferromagnet decamethylferrocenium tetracyanoethe-
nide, [FeCp*2]•+[TCNE]•−” Inorg. Chem. 48 (2009), 3296–3307.
[24] C.A.P. Goodwin, F. Ortu, D. Reta, N.F. Chilton, D.P. Mills, “Molecular magnetic hysteresis at
60 kelvin in dysprosocenium” Nature 548 (2017), 439–442.
[25] F.-S. Guo, B.M. Day, Y.-C. Chen, M.-L. Tong, A. Mansikkamäki, R.A. Layfield, “Magnetic
hysteresis up to 80 kelvin in a dysprosium metallocene single-molecule magnet” Science 362
(2018), 1400–1403.
10 Conducting Materials

10.1 Conducting Materials

An electrical current is a flow of charged particles, and electrical conductivity is a measure
of how easily a current can pass through a material. Materials where the current is carried
by electrons are called electronic conductors, while those where the current is carried by ions
are known as ionic conductors. In this chapter we take close look at the origins of electronic
conductivity in a wide variety of materials. Ionic conductors are covered in Chapter 13.
It is difficult to imagine a material property that spans a greater range than electrical
conductivity, nearly 30 orders of magnitude (Table 10.1). To put the conductivities of
various materials in context, we need to define the units associated with conductivity. We
begin with Ohm’s law:
V
I¼ (10.1)
R

where I is the current in amperes, V is the potential difference in volts, and R is the resistance
in ohms. Resistance is the material property of interest to us, but, as an extrinsic property, its
value depends on the size and shape of the sample. To express Ohm’s law in a form that is
independent of the sample dimensions, we replace current in Equation (10.1) with current
density, J = I/A (current per area),1 potential difference with electric-field intensity,
E = V/L (voltage per length), and resistance with resistivity, ρ = RA/L (resistance per length
and per inverse area). In doing so, Equation (10.1) becomes:

V EL EL E
J¼ ¼ ¼ ¼ ¼ σE (10.2)
AR AR AðρL=AÞ ρ

1
Because the electric current (in amperes) is the number of flowing charges (in coulombs) per second, the current
density J (in A/m2) is the same as the charge flux J [in [C/(m2 s)]. We’ve encountered the flux J already in Chapter 3.

396
 10.1 Conducting Materials 397

Table 10.1 Electrical conductivity, σ, of selected materials at

room temperature.

Substance σ (S/m) Substance σ (S/m)

Ag 6.2×107 Bi2Ru2O7 2×105

Cu 5.9×107 LaNiO3 1×105
Al 3.8×107 Doped polyacetylene 8×104
Na 2.1×107 Fe3O4 2×104
ReO3 1.1×107 YBa2Cu3O7* 1×102
Ti 2.5×106 Ge 2×100
La 1.6×106 Si 10−3
SrMoO3 1.0×106 NiO 10−8
Bi 7.7×105 Al2O3 10−12
Mn 6.2×105 S 10−15
NbN 4×105 SiO2 (quartz) 10−16
TiO 3×105 Teflon 10−22
*Conductivity in the ab plane of the crystal.

resistivity, ρ (Ω m)
10−8 10−4 100 104 108 1012 1016 1020

semiconductors
metals insulators

108 104 100 10−4 10−8 10−12 10−16 10−20

conductivity, σ (S/m)

Figure 10.1 Approximate ranges for the conductivity and resistivity of metals, semiconductors, and
insulators.

where σ is conductivity, the inverse of resistivity, σ = 1/ρ. Resistivity has units of ohms meter
[Ω m = (V/A) m] and conductivity has units of siemens per meter [S/m = (A/V)/m], where
1 S = 1 Ω−1. Conductivity and resistivity are intrinsic properties of materials. Approximate
ranges of conductivity and resistivity for metals, semiconductors, and insulators are illus-
trated in Figure 10.1.
Metals are the most familiar electronic conductors, yet we see from Table 10.1 that not all
metals are equally able to conduct electricity. For example, silver and copper have conduct-
ivities that are two orders of magnitude greater than manganese and bismuth. As the name
suggests, semiconductors, like silicon and germanium, are far less conductive than metals.
However, semiconductors differ from metals in ways other than their absolute conductivity.
398 Conducting Materials

Semiconductors become more conductive with increasing temperature, whereas temperature

has the opposite effect on a metal. The conductivity of a semiconductor can be increased
dramatically by adding small amounts of other elements, a process called doping, whereas
impurities typically decrease the conductivity of a metal. In this chapter we explore the
origins of these differences.
Metallic elements and alloys are not the only materials that are good conductors.
Transition-metal oxides and nitrides, like ReO3 and NbN, have conductivities that are
comparable to many metals. However, unlike elemental metals, the electrical conductivities
of transition-metal compounds span a very wide range. For example, TiO is only 10 times
less conductive than titanium metal, whereas stoichiometric TiO2 is highly insulating. The
differences between TiO and TiO2 can be attributed to differences in the occupation of the
d orbitals, but not all transition-metal oxides with partially filled d orbitals exhibit metallic
conductivity. NiO and TiO both crystallize with the NaCl structure and both have ions with
partially filled d orbitals, but NiO is less conductive than TiO by 13 orders of magnitude.
Clearly, many aspects of conductivity cannot easily be understood without a closer look at
the underlying mechanisms of conductivity in different classes of materials.

10.2 Metals
In most applications that require high electrical conductivity, metals are the materials of
choice. Copper is widely used for electrical wires, circuit boards, and electrical contacts
because of its high conductivity, ductility, chemical stability, and cost. Silver is not com-
monly used as a conductor because it is more expensive yet only marginally more conductive
than copper. Aluminum is used when material cost or weight is an important factor, while
gold is favored in electronic circuits because of its excellent chemical stability.
We learned in Chapter 6 that metals conduct electricity because the Fermi level cuts
through a band. In such materials, there is very little energy cost for electrons to move
from occupied to empty crystal orbitals (the infinite crystal equivalents of molecular orbitals,
see Chapter 6), allowing easy movement in response to an external electric field. In fact,
a partially filled band is a necessary but not sufficient condition for metallic conductivity.
Unfortunately, this simple picture does not give any clues as to why some metals are better
conductors than others. Nor does it make any prediction about the how conductivity will
change as a function of temperature or defect concentration.

10.2.1 Drude Model

One of the earliest attempts to explain the conductivity of metals was put forward by Paul
Drude in 1900. It predates the development of quantum mechanics and is therefore
a classical treatment. Although too simplistic in many ways, the Drude model serves as
a useful introduction to conduction. Many of the parameters used to describe conduction in
 10.2 Metals 399

metal
cation

e–

Figure 10.2 The trajectory of an electron in the Drude model in the absence of an applied field.

the Drude model, such as carrier concentration and mobility, have been retained in more
modern treatments.
We begin by assuming that a metal can be treated as an ordered array of positively charged
stationary ions surrounded by a negatively charged sea of delocalized valence electrons. The
valence electrons move throughout the crystal and behave as an ideal gas, albeit one where
the “gas molecules” have a very small mass. If we were to follow the path of a single electron,
we would see that it undergoes frequent collisions with the stationary ions, leading to
a random trajectory, as shown schematically in Figure 10.2.
In the kinetic theory of gases, the atoms or molecules of the gas are in constant random
motion, and all collisions are assumed to be elastic. Potential energy is neglected, so the energy
of each particle is simply its kinetic energy, E = ½mv2, where m and v are the mass and velocity
of particle, respectively. Here we can treat velocity as a scalar (a speed). Not all atoms/
molecules of the gas move with the same speed; at any given moment, some are moving
much faster than the ensemble average and others much slower. The spread of their energies
follows a Maxwell–Boltzmann probability distribution, with an average kinetic energy of:

3
E ¼ kT (10.3)
2

where k = 1.380649×10−23 J/K is the Boltzmann constant and T is the absolute temperature.
Since the particles, here electrons, all have the same mass, we can calculate the rms velocity2
of randomly moving electrons in the ensemble from the definition of kinetic energy and
Equation (10.3):

2
This value is called the root mean square (rms) velocity because it is obtained from the average kinetic energy E that is
proportional to velocity squared. The average (mean) velocity for the Maxwell–Boltzmann probability distribution
of kinetic energies is smaller, 92.1% of the rms velocity.
400 Conducting Materials

sffiffiffiffiffiffiffiffi
3kT
vrms ¼ (10.4)
me

where me is the mass of an electron. Using this relationship, we can estimate the rms velocity
of an electron at room temperature to be on the order of 105 m/s.
Next, we define two additional parameters that come directly from kinetic theory. The
mean free path, ℓ, is the average distance the electron travels between collisions, and the
relaxation time, τ, is the average time between collisions. These two quantities are directly
proportional to each other, with the inverse of velocity acting as the proportionality
constant:
ℓ
τ¼ (10.5)
v

Because metal structures are closely packed, we might assume that the electrons do not travel
far without being deflected by a collision with an ion. The Drude model yields qualitatively
reasonable estimates of the conductivity for main-group metals if we choose a mean free path
of a few nanometers. This distance amounts to a collision every few unit cells, which does not
seem unreasonable. Using a value of 1 nm for the mean free path and taking our earlier
velocity estimate of 105 m/s, we obtain a relaxation time τ on the order of 10−14 s, one
collision every 10 femtoseconds.
Even though electrons are moving quickly, in the absence of an electric field the motions
are random, and the net flow of charge is zero. The situation changes once an electric field
E is applied. The electrons are attracted to the positively charged end of the conductor by
a force F = −eE, where e is the elementary charge (1.602×10−19 C), and the negative sign
originates from the two vectors being oppositely aligned.3 The net velocity of the electron
along the direction of the electric field, called the drift velocity, vd, is equal to the product of
the acceleration, a, due to the field (F = ma) and the time τ that elapses before a collision
changes the path of the electron:
 
F Eeτ
vd ¼ aτ ¼ τ¼ (10.6)
me me

If we consider only the magnitudes (absolute values) of the vectors, we can calculate the drift
velocity. An electric potential of 1 V applied along a wire 1 cm long generates an electric field
of magnitude 100 V/m. Taken together with our earlier estimate of 1×10−14 s for τ, we can
restate Equation (10.6) with scalars to obtain vd = Eeτ/me ≈ 0.2 m/s for the drift velocity. This
net velocity of an electron is more than five orders of magnitude smaller than our earlier
estimate of its average thermal velocity. The fact that the electrons keep colliding with the

3
The direction of the electric-field vector is defined to be from positive to negative charge. The force attracts the
electron towards the positive end.
 10.2 Metals 401

stationary ions explains why the drift velocity is so much smaller than the thermal velocity.
We can think of these collisions as a source of resistance to the electrical current flow.
Before returning to macroscopic quantities, we need to define one more quantity, electron
mobility, μ, which is the electron’s drift velocity divided by the electric-field intensity acting
on it, has units of m2/(V s) and is a measure of the ease with which an electron moves through
a crystal:
vd
μ¼ (10.7)
E

Combining Equations (10.6) and (10.7), we obtain the following scalar expression for the
mobility of an electron:
eτ
μ¼ (10.8)
me

As the collisions become less frequent (the relaxation time τ increases), the electric field has
more time to accelerate the electron along its path and increase its mobility.
Returning to macroscopic quantities, it can be shown that the current density J (in A/m2) is
equal to the number of conduction electrons per unit volume, n, multiplied by their charge
and drift velocity:

J ¼ nevd ¼ neμE (10.9)

By comparison with Equation (10.2) (J = σE) we see that:

σ ¼ neμ (10.10)

This is a key equation for understanding the conductivity of a material. The electrical conductiv-
ity of a material depends upon the concentration of charge carriers n and their mobility μ.
It is quite difficult to directly measure the relaxation time in a metal, but, as a crude
approximation, we might assume that τ and hence μ do not change substantially from one
metal to the next. To the extent that this assumption is valid, the conductivity should scale with
n, the number of charge carriers per unit volume. In the Drude model, all valence electrons are
free to respond to the applied field and act as charge carriers. To calculate n within this model,
one simply needs to multiply the atomic density by the number of valence electrons per atom.
The valence-electron concentration and room-temperature conductivity of several metals
are shown in Table 10.2. Although the conductivities of the alkali metals do increase with
increasing valence-electron concentration, we see that any relationship between the two
variables quickly breaks down as we start moving around the periodic table.4 To better
understand these periodic trends, as well as many other aspects of metals, we need a more
realistic treatment of the conduction electrons.
4
There are many other experimental observations that reveal the shortcomings of the Drude model; the heat capacity
of metals and the conductivity of alloys are two examples.
402 Conducting Materials

Table 10.2 Conductivity σ does not scale with the valence-electron concentration n.

Metal n (m−3) σ (S/m) Metal n (m−3) σ (S/m)

28 7 28
Rb 1.1×10 0.8×10 Ag* 5.9×10 6.2×107
K 1.3×1028 1.4×107 Au* 5.9×1028 4.5×107
Na 2.5×1028 2.1×107 Cu* 8.4×1028 5.9×107
Ca 4.6×1028 2.9×107 Zn* 13.1×1028 1.7×107
Mg 8.6×1028 2.3×107 Sc 12.0×1028 0.18×107
Al 18.0×1028 3.8×107 Ti 20.5×1028 0.25×107
* The d subshells of Cu, Ag, Au, and Zn are assumed to be full and are not counted in the
valence-electron count.

10.2.2 Free-Electron Model

The failure of the Drude model to quantitatively explain the conductivities of metals should
not come as a surprise; after all, electrons follow the laws of quantum mechanics, where the
energy of each electron is dictated by the band it occupies and its wave vector k (Section
6.4.2). The simplest quantum-mechanical description of electrons in a metal is the free-
electron model. It assumes that the electric potential felt by an electron is constant and equal
to a constant potential V0 everywhere inside the crystal and effectively infinite outside the
crystal (Section 5.2). This potential for a 1D version of this idealized crystal is plotted in
Figure 10.3. You might think that such a simple model would provide little insight to real
materials, but, with some fine tuning, it works reasonably well for simple main-group metals.
Recall from Section 6.1.2 that in a 1D crystalline solid the electronic wavefunction takes
the form ψ(r) = eikru(r), where u(r) is a periodic function representing the potential felt by the
electron. For a given band, each electron must have a unique value of k. In this sense, k acts
as a quantum number specifying the crystal orbital that the electron occupies, but it is more
than just a label. The value of k is inversely proportional to the wavelength λ of the crystal-
orbital wavefunction k = 2π/λ; Equation (6.10). In a 3D crystal, it becomes a vector k (Section
6.4.2) that is parallel to the propagation direction of the wave (perpendicular to its wave
front), the wavelength is λ = 2π/|k| and the wavefunction becomes ψ(r) = eik·ru(r).5
In the free-electron model, the potential is uniform throughout the crystal, u(r) = 1, and
the wavefunction simplifies to ψ(r) = eik·r. When this wavefunction is inserted into the time-
independent Schrödinger equation, Hψ(r) = Eψ(r), the energies E of the allowed electronic
states are:

5
Perhaps the simplest illustration of this is the crystal orbitals for a 2D sheet of hydrogen atoms depicted in Figure 6.17. At
the X point (k = π/akx + 0ky), the wavefunction changes sign (propagates) upon moving through the crystal in the
x direction. The vector nature of k is essential to distinguish this crystal orbital from that at the Y point (k = 0kx + π/aky);
both have the same wavelength but the waves propagate in different directions.
 10.2 Metals 403

V effectively V V effectively
infinite infinite

V = V0
x

vacuum metal vacuum

Figure 10.3 The potential V (arbitrary scale, units V) experienced by an electron in the free-electron
model in a 1D metal of finite length.

ℏ2 k2
E ¼ V0 þ (10.11)
2me
To apply the free-electron model to real materials, the electron mass, me, is replaced with the
effective mass, m*:
ℏ2 k2
E ¼ V0 þ (10.12)
2m
The effective mass is an adjustable parameter that varies from one material to another. We
will return to its significance later.
Using Equation (10.12), we can plot energy as a function of k to obtain a band-structure
diagram. The band structure, which is a simple parabola with a minimum at k = 0, is plotted
in Figure 10.4, where it is compared with the band structure for a 1D chain of hydrogen
atoms discussed in Section 6.1.5. Even though they are derived using quite different assump-
tions, the tight-binding and free-electron band structures are surprisingly similar. To obtain
this level of agreement, two parameters in the free-electron model must be adjusted. The
constant potential V0 must be chosen so that the bands have the same energy at k = 0, and the
effective mass m* must be adjusted so that the bandwidths are similar.

10.2.3 Fermi–Dirac Distribution

In the electron-gas Drude model, the energy of each electron is taken to be its kinetic energy
that follows a Maxwell–Boltzmann distribution, whereby the average electron energy is
directly proportional to the absolute temperature, Equation (10.3). In any quantum-
mechanical treatment, including the free-electron model, the energy of each electron is
404 Conducting Materials

free

Figure 10.4 1D band structure for the free-electron model (left), and an infinite chain of hydrogen atoms
spaced by 1 Å, as calculated using a tight-binding model (right).

determined by the crystal orbital it occupies, which is dictated by its wave vector k. Because
the Pauli exclusion principle prevents two electrons of the same spin from occupying the
same crystal orbital, the imposition of quantum mechanics leads to a very different distribu-
tion of electron energies. That is not to say temperature has no effect in the more realistic
quantum-mechanical treatment, as we see below.
At finite temperatures, thermal energy excites some electrons into states with energies
greater than the Fermi energy, EF, while creating vacancies in states with energies less than
EF. The probability that a given electronic state (a crystal orbital) is occupied, f (E), is given
by the Fermi–Dirac distribution function:

1
f ðEÞ ¼ (10.13)
1 þ exp½ðE  EF Þ=kT

where E is the energy6 of the electronic state in question. For example, at 300 K, f (E) = 0.02
for a crystal orbital whose energy is 0.10 eV above EF, which means there is only a 2%
probability that this state will be occupied. The Fermi–Dirac distribution is plotted for
a metallic solid at T = 0 and 300 K in Figure 10.5. At 300 K, the electron distribution is
smeared out by a few tenths of an electronvolt on either side of the Fermi energy EF.
The imposition of Fermi–Dirac statistics changes our picture of conductivity in an
important way. For an electron to move through the crystal in response to an applied electric
field, there must be an unoccupied state of similar energy for the electron to move into.
Electrons at energies well below the Fermi level, where f (E) ≈ 1, do not meet this criterion. As
a result, only a small percentage of the total number of valence electrons contribute to the
conduction process.
6
Do not confuse the symbol E used for energy with the E used for electric-field intensity, or k representing the
Boltzmann constant with the symbol used for the wave vector.
 10.2 Metals 405

T=0K T = 300 K
empty states
1 1
below EF

states below states above

EF fully EF empty
occupied
f (E)

0.5 0.5

occupied states
above EF

0 0
−0.5 −0.25 0 0.25 0.5 −0.5 −0.25 0 0.25 0.5
E − EF (eV) E − EF (eV)

Figure 10.5 Fermi–Dirac distribution function, f(E), in a metal near EF at T = 0 (left) and 300 K (right).

10.2.4 Carrier Concentration

Conductivity is proportional to the carrier concentration n per volume and to the mobility of
those carriers μ, Equation (10.10). What do the free-electron model and Fermi–Dirac
statistics tell us about how the carrier concentration n changes as the valence-electron
concentration (per unit volume) varies from metal to metal? For the parabolic band that
comes out of the free-electron model in Section 10.2.2, it can be shown that the density of
states (DOS), N(E), is proportional to the square root of the energy, N(E) ∝ E1/2. This is
shown in Figure 10.6, along with the product f(E)·N(E), which emphasizes the occupied
states at a given temperature. The electrons that occupy states close to EF are the ones that
contribute to the conductivity, while those well below EF cannot access empty states needed
to move through the crystal. If we increase the concentration of valence electrons, band
filling, and the DOS at the Fermi level, N(EF) will also increase. Hence, the free-electron
model predicts that n will increase, even if in a nonlinear fashion, as the concentration of
valence electrons increases. This helps to explain why the conductivities of the alkali metals
increase as they become smaller (Rb → K → Na), and the valence-electron concentration
goes up (Table 10.2).
Closer inspection of Table 10.2 shows that the conductivity does not change uniformly as
the valence-electron density increases. The discrepancy between theory and observation
stems in part from the inability of the free-electron model to approximate the band structures
of most metals. It works best for the alkali metals where the valence electrons occupy one half
of a single band originating from overlap of s orbitals. For other elements, the Fermi level
cuts through multiple bands, and the free-electron model breaks down. In later sections, we
will take a more accurate look at the band structures of real metals.
406 Conducting Materials

f(E)·N(E) N(E)
5
EF
4

energy ((eV)
eV)
3

energy
2

0
0 0.5 1 1.5 2 2.5 3

DOS,
DOS, N(E)
N(E)

Figure 10.6 The DOS corresponding to the free-electron model, N(E), multiplied by the occupancy
probability at 500 K, N(E)·f (E), (shaded).

10.2.5 Carrier Mobility and Effective Mass

Carrier concentration is not the only variable that determines the conductivity of
a substance. Mobility plays an equally important role; but how are we to assess the changes
in mobility from one substance to another? Do the conduction electrons in copper have
a higher or lower mobility than those in potassium, and what about transition metals like
titanium? To answer these questions, we need to take a closer look at the relationship
between band structure and electron mobility.
Effective mass is a kind of “fudge factor” that allows many predictions of the simple free-
electron model to be applied more generally than would otherwise be justified. As a rule of
thumb, when the Fermi level cuts through a wide band, the electrons behave as high-mobility
“light” charge carriers with relatively small effective masses. Conversely, when the Fermi
level cuts through a narrow band, electrons have low mobility, and m* is large.7
Effective mass can be defined in a more quantitative way by differentiating Equation
(10.12) twice with respect to the wave vector k. In real materials, effective mass is a tensor
whose magnitude depends upon the directions of the wave vector and the applied electric
field. Fortunately, the parabolic band shape of the free-electron model evolves from its
minimum at k = 0 identically in any direction of k space, which allows us to obtain an
expression that gives a scalar value of the effective mass:

d2 E ℏ2
¼ (10.14)
dk2 m

Rearranging to solve for m* we get:

7
To help remember this guideline think of the bands as the tracks of a rollercoaster and the carriers as the cars on those
tracks. The speed of the car is a proxy for mobility.
 10.2 Metals 407

ℏ2
m ¼ (10.15)
ðd2 E=dk2 Þ

The effective mass of the conduction electrons therefore depends on the curvature of the
bands that cross the Fermi level. In Section 6.1.5, it was stated without full justification that
flat (narrow) bands are associated with localized electrons and wide bands with highly
delocalized electrons. Now we see the rationale; flat bands have very little curvature, which
makes the denominator of Equation (10.15) small and leads to a large effective mass. The
opposite relationship holds for wide bands.
Although the qualitative relationship between bandwidth and effective mass is quite
general, the quantitative use of Equation (10.15) is much more limited. For a parabolic
band, Equation (10.14) yields m* constant, independent of the wave vector k. For more
realistic band structures this is not the case; there are different values of m* at different values
of k. The usefulness of Equation (10.15) is generally limited to substances where the charge
carriers reside near the bottom or top of a band whose shape is approximately parabolic. As
we will see later, this approximation often works reasonably well for semiconductors.
When Equation (10.15) is used to describe charge carriers near the top of the band, we get
an unusual result. Consider the band structure of the 1D hydrogen atom chain shown on the
right-hand side of Figure 10.4. Because the band has an inflection point midway between k = 0
and k = π/a, the quantity d2E/dk2 will be negative when the band is more than half filled.
Consequently m* will also be negative. What does it mean to have a negative effective mass? It
means that nearly full bands conduct as though the charge carriers were positively charged. That
is to say, the charge carriers move in the opposite direction than expected for electrons. These
charge carriers, called holes, respond to an electric field just like an electron with a negative mass.
We will return to holes when discussing semiconductors where they play an important role.

10.2.6 Fermi Velocity

In the Drude model, a Maxwell–Boltzmann distribution was used to estimate the average
energy of the electrons, and, through Equation (10.4), their rms velocity. The free-electron
model produces a different distribution of energies, which means our earlier velocity estimate
of ~105 m/s from Section 10.2.1 is suspect. Let us revisit the question of carrier velocity, this
time within the more realistic quantum-mechanical picture.
The velocity of a particle is directly proportional to its momentum, p = mv, and, as we
learned in Chapter 6, the crystal momentum of an electron is related to the wave vector
k through Equation (6.12), p = ħk.8 Combining these expressions, we express the electron’s
velocity in terms of its wave vector:

8
Though not quite the same as true momentum, crystal momentum must be conserved when electrons are scattered by
collisions with lattice vibrations (phonons), impurities, or other electrons. Crystal momentum must also be con-
served when electrons absorb or emit photons.
408 Conducting Materials

applied field, E
no applied field
20

))
eenerggyy((eeVV 10

0
uncompensated
EF EF
electrons

-10

-20
−π/a 0 π/a −π/a 0 π/a
k k

Figure 10.7 The occupied states in a partially filled band of a 1D metal in the absence of an external
electric field (left), and once an external electric field is applied (right). Thermal excitations of the carriers
have been neglected.

ℏk
v¼ (10.16)
m

The band structure of an arbitrary 1D metal with a single band is plotted in Figure 10.7,
where a thick line is used to denote occupied states. In the 1D case, we can treat the wave
vector k as a scalar. The electrons near the center of the Brillouin zone have low velocities,
while those with larger values of k are moving much faster. In the absence of an external field,
the distribution is symmetric on either side of k = 0, which means that for every electron with
a momentum ħk there is another electron with the same absolute value of momentum
moving in the opposite direction. Consequently, there is no net migration of charge and no
electric current flowing. While thermal excitations will smear out the occupied states near EF
slightly (as governed by the Fermi–Dirac function), the distribution will remain symmetric
with respect to k regardless of the temperature.
If an electric field is applied in the positive k direction, the states with negative k are stabilized
and those with positive k are destabilized. This shifts the distribution of occupied states slightly
to the left, as shown in Figure 10.7.9 Because the displacement is small (the shift has been
exaggerated in Figure 10.7), most electrons with wave-vector length −k are still cancelled by an
electron with a wave vector of length +k. However, near the Fermi level, a few occupied states
with wave vectors of −k are not compensated by an electron moving in the opposite direction.
It is these uncompensated electrons that are responsible for the flow of current. We should also
note that if the band is completely full, as it is for an insulator, the applied field is not able to
shift the occupied states and there is no flow of current in response to the field.

9
The states with negative values of k are stabilized by an electric field because of the negative charge of an electron.
Remember that by convention electrons flow in the opposite direction of current.
 10.2 Metals 409

Because the uncompensated electrons are equally distributed around the Fermi level, their
average energy is EF. By setting EF equal to the average kinetic energy of the conduction
electrons (½mv2), we can calculate the rms velocity of the conduction electrons. This value,
called the Fermi velocity, vF, is:
rffiffiffiffiffiffiffiffi
2EF
vF ¼ (10.17)
m

Using experimentally measured values of EF and setting m* = me, we can estimate vF. For
copper, EF = 7.0 eV and vF = 1.6×106 m/s, whereas for aluminum these values are 12 eV and
2.0×106 m/s.10 The Fermi velocities for these metals are more than an order of magnitude
larger than the velocity calculated with Equation (10.4). Thus, we see that by treating the
electrons classically, the Drude model overestimates the number of carriers, but underestimates
their velocity. Because the two estimates err in opposite directions, the Drude estimates of the
conductivity are closer to experimental values than one might expect.

10.2.7 Scattering Mechanisms

By expressing the mobility term in Equation (10.10) with Equation (10.8) and replacing me
with m*, we can obtain an expression for conductivity in terms of relaxation time:

ne2 τ
σ ¼ neμ ¼ (10.18)
m

Using Equation (10.5) (τ = ℓ/v) we can write a similar expression that relates conductivity to
mean free path:

ne2
σ¼ ℓ (10.19)
m  vF

These equations make intuitive sense; as the relaxation time and mean free path increase, the
electrons are better able to respond to an applied field and the conductivity increases. From
conductivities measured at room temperature, Equations (10.17) and (10.19) provide mean
free paths of 42 nm and 29 nm for Cu and Al, respectively. This is considerably larger than
our earlier estimate of a few nanometers (Section 10.2.1). With mean free paths that are tens
of nanometers, the conduction electrons are able to pass by hundreds of atoms between
collisions. Given the close packing of atoms in a metal, this behavior is somewhat surprising.
It’s akin to hitting a golf ball through a dense forest and finding that it travels several
10
Absolute values of EF can be confusing. In the free-electron model, the bottom of the band defines zero energy, and EF
values are positive numbers that get larger as the band filling increases. That is the formalism used here. In tight-binding
theory, the Fermi level is always a negative number determined by orbital overlap and the ionization energies of the
atomic orbitals. In other band-structure calculation methods, EF is often arbitrarily set to zero energy. Note that in Figure
10.4 both plots employ the scale used for tight-binding calculations.
410 Conducting Materials

hundred yards without hitting a tree. As we will soon see, the mean free paths at low
temperatures can be even longer. How can we understand this behavior?
Two facets of quantum mechanics are responsible for the unexpectedly long mean free path of
the electron. Firstly, as the conducting electron approaches an atom, it experiences repulsions
with the atom’s electrons, which tend to cancel out the Coulomb attraction to the positively
charged nucleus. Secondly, wave–particle duality allows the electron to behave as a wave with
a wavelength given by the de Broglie relation, λ = h/(mev). Using vF = 2×106 m/s, we obtain
a wavelength of 3.6 Å for conduction electrons in aluminum. This value is comparable to the
atom spacing in a crystal. By acting as a wave, the electron can pass through the crystal
indefinitely without scattering. When an electron wave travels through the crystal, atoms absorb
energy from the wave and radiate it back, so that the wave continues with the same direction and
intensity, much in the same way a light wave traveling through a crystal would behave.11
In a perfect crystal, the conduction electrons would not be scattered by collisions with
atoms and there would be no resistance to current flow. In real materials, deviations from
perfect periodicity disrupt the transport of conduction electrons and give rise to finite
resistance. We can classify the imperfections into two categories: defects and lattice vibra-
tions. Defects, such as substitutional impurities, interstitial atoms, and vacancies, break the
periodicity of the lattice and lead to scattering of the electrons. Even if you could create
a defect-free crystal, at finite temperatures lattice vibrations break the periodicity of the
lattice and result in scattering of the conduction electrons.
The resistivities of high-purity aluminum and copper are plotted as a function of tempera-
ture in Figure 10.8. Notice that both metals become more conductive as the temperature
decreases. At 1 K, the resistivity of aluminum is 1×10−12 Ω m and that of copper 2×10−11 Ω m.
If we convert these values to conductivities (1×1012 and 5×1010 S/m, respectively) and
compare with the room-temperature values, we see that aluminum and copper are 25000
and 830 times more conductive at 1 K than at room temperature.
What is responsible for the increased conductivity of metals at low temperature? To answer
this question, consider Equation (10.19). The Fermi velocity depends on the Fermi energy
through Equation (10.17), which is essentially independent of temperature. For a metal, the
carrier concentration n does not change dramatically with temperature. If anything, n might
decrease subtly at low temperature as the Fermi–Dirac function sharpens. Therefore, we can
only explain the increased conductivity by concluding that the mean free path of the conduc-
tion electrons has increased significantly at low temperature. By assuming n is constant as
T decreases, we can estimate a lower bound on the increase in mean free path. In high-purity
aluminum, such an estimate gives a mean free path at 1 K of 0.7 mm, which means the electron
is able to pass through more than a million unit cells between collisions!
As seen on the right-hand side of Figure 10.8, a crossover in the temperature dependence
of the resistivity happens at low temperature. At low temperatures, the thermal vibrations of

11
The velocity of the electron wave is slowed just like a light wave is slowed when it moves through a crystal, in the
latter case by an amount equal to the refractive index.
 10.2 Metals 411

Figure 10.8 The resistivities for high-purity single crystals of aluminum and copper as a function of
temperature. The plot on the right-hand side shows an expanded view of the low-temperature region.
Data taken from ref. [1].

the atoms become negligible, and scattering due to impurities and defects dominates. This
contribution to the scattering is independent of temperature. At sufficiently low temperat-
ures, below roughly 20 K for high-purity Al and Cu, the resistivity reaches a nearly constant
value known as the residual resistivity. The value of the residual resistivity of a metal is
determined largely by the concentration of defects. As the defect concentration decreases, the
residual resistivity also goes down. At higher temperatures, scattering due to lattice vibra-
tions dominates, and the resistivity increases linearly as a function of temperature.12

10.2.8 Band Structure and Conductivity of Aluminum

The free-electron model can be applied with reasonable success to the alkali metals, but for most
other metals we need a more sophisticated model. Let’s use the concepts developed in Chapter 6 to
discuss the band structure and conductivity of aluminum. Aluminum has a cubic closest-packed
structure with a face-centered cubic (fcc) unit cell. Instead of using the fcc cell, we choose to
construct the band structure with the smaller primitive unit cell that contains a single atom. Since
there are four valence orbitals per atom (3s, 3px, 3py, 3pz), the band structure contains four bands.
The 3s and 3p orbitals do not mix at Γ, which leads to a large energy separation because the
3s orbitals have a bonding overlap at Γ while the 3p orbitals overlap in an antibonding manner, as
shown on the right-hand side of Figure 10.9. The observation of a minimum for the “3s” band at Γ
parallels the band structures of the 1D (Figure 6.5) chain and 2D (Figure 6.17) sheet of H atoms.
This feature arises because the maximum bonding overlap of a lattice of s orbitals always occurs at
Γ, where all atomic orbitals that make up the crystal orbital have the same phase. The lower half of
the “3s” band has a parabolic shape similar to the free-electron band structure, but agreement

12
More complex behavior is seen in some metallic conductors, as discussed in Section 15.10.
412 Conducting Materials

12

energy (eV)
p–p antibonding at Γ
4

0 EF

-4

-8
s–s bonding at Γ
-12
W L Г X W

Figure 10.9 The band structure of aluminum. The sketches on the right show the orbital contributions of
the 3s band and one of the triply degenerate 3p bands at Γ.

with the free-electron model breaks down as the energy increases. The free-electron model is not
a good approximation near the Fermi level. Therefore, we cannot use the free-electron model to
estimate the carrier concentration or mobility in aluminum.
It is more complicated to explain the band structure as we move away from Γ because the
3s and 3p orbitals mix, but we can make some observations about the properties from the band
structure. There is no gap between bands, which means that, regardless of the electron count,
the Fermi level will cut through one or more bands, resulting in metallic conductivity. The
bands are wide, with widths that vary from approximately 10 eV to 16 eV. Wide bands lead to
high electron mobilities, so it is not surprising that aluminum is a highly conducting metal.

10.2.9 Band Structures and Conductivity of Transition Metals

How do transition metals differ from main-group metals? We must now consider the
d orbitals, and how their inclusion impacts the band structure. The room-temperature
electrical conductivities of metals in the d block of the periodic table are given in Table
10.3. Within a group, the conductivities are similar, whereas along a period, the changes are
quite irregular. One conspicuous feature is the abrupt jump in conductivity between groups
10 and 11, even though elements from both groups adopt fcc structures. What is responsible
for the high conductivity of the group 11 elements?
Because we need to include the ns, np, and (n − 1)d orbitals, we expect a total of nine bands
in the band-structure diagram. Figure 10.10 compares the band-structure diagrams and
DOS plots for Ag and Pd. Four wide bands originate from overlap of the 5s and 5p orbitals.
They are similar in shape to those seen in aluminum (although not identical due to mixing
with the 4d orbitals). Moving toward Γ, the energy of the band with predominant
5s character decreases toward a minimum while the bands with predominant 5p character
increase toward a maximum. In addition to these four bands, five much narrower bands arise
from the 4d orbitals. The contribution of the 5s, 5p, and 4d bands can be seen more clearly in
 10.2 Metals 413

Table 10.3 Room-temperature conductivities (in 106 S/m) and structure types of the transition
metals, arranged as found in the periodic table (group numbers shown at the top).

3 4 5 6 7 8 9 10 11

Sc Ti V Cr Mn Fe Co Ni Cu
1.8 2.5 5.0 7.9 0.6 10.0 17 14 59
hcp* hcp bcc* bcc Other bcc hcp fcc* fcc
Y Zr Nb Mo Tc Ru Rh Pd Ag
1.8 2.4 6.7 20 5.0 14 23 10 62
hcp hcp bcc bcc hcp hcp fcc fcc fcc
Lu Hf Ta W Re Os Ir Pt Au
1.8 5.3 7.7 20 5.6 12 21 9.4 45
hcp hcp bcc bcc hcp hcp fcc fcc fcc
* The abbreviations hcp, bcc, and fcc refer to hexagonal closest-packed, body-centered cubic, and face-
centered cubic (cubic closest-packed) structures.

silver
12

8
energy (eV)

0 EFF
Ag 4d
-4 Ag 5s
Ag 5p
-8
W L  X W
partial DOS

palladium
12

8
energy (eV)

0 EF

-4 Pd 5s

Pd 5p Pd 4d
-8
W L  X W
partial DOS

Figure 10.10 The band structure of Ag (top) and Pd (bottom). Fermi energy EF is set to 0 eV. The partial
DOS illustrates the contributions of the 5s, 5p, and 4d orbitals (right).
414 Conducting Materials

the partial DOS plots on the right-hand side of Figure 10.10. The 4d bands are narrower because
the 4d orbitals are more contracted than the 5s and 5p orbitals (Section 5.2.2), limiting overlap
with orbitals on neighboring atoms. Hence, we get a band structure that contains wide bands
with 5s/5p character together with a set of narrow bands of 4d parentage.
How do we rationalize the high conductivity of Ag in terms of its band structure?
Although palladium and silver have similar band structures, the Fermi level cuts through
the bands at different points. In silver, the 4d bands are fully occupied, and the Fermi level
cuts through a wide sp band, which leads to high-mobility carriers. In palladium, the electron
count per atom decreases by one electron and the Fermi level cuts through the upper part of
the 4d bands. These bands are much narrower, and therefore the electrons that populate
these bands have lower mobilities. Consequently, palladium has a six times smaller conduct-
ivity than silver, even though they are isostructural and neighbors on the periodic table. In
fact, all metals in groups 3–10, where the Fermi level cuts through the d bands, have lower
conductivities than Cu, Ag, and Au.
It is worthwhile noting that when EF cuts through the narrow d bands, the DOS at the
Fermi energy is considerably larger than it is in metals where EF cuts through the much wider
sp bands. This leads to an increase in carrier concentration n. With all other things being
equal, this would increase the conductivity, but, as the observed conductivities of Pd and Ag
show, the decrease in mobility has a larger impact on the conductivity than the increase in
carrier concentration.

10.3 Semiconductors
Whereas metals are characterized by the presence of one or more partially filled bands,
semiconductors have a band gap. At T = 0 K, the bands that lie below the band gap are filled
with electrons and those above the gap are empty (Section 6.3). When an external field is
applied to a metal, it promotes electrons in orbitals just below the Fermi level to nearly
degenerate unoccupied orbitals just above the Fermi level, thereby enabling current to flow.
In a semiconductor, conduction can only occur if electrons are excited across the band gap,
and this makes high-purity semiconductors poor conductors at low temperatures. The
conductivity can, however, increase dramatically by raising the temperature and/or introdu-
cing appropriate substitutional impurities. This sensitivity to chemical substitution and
temperature differentiates the conductivity of semiconductors from that of metals.

10.3.1 Carrier Concentrations in Intrinsic Semiconductors

One big difference between the conductivities of metals and semiconductors concerns the
concentration n of the charge carriers. In an intrinsic semiconductor, the number of electrons
thermally excited into the conduction band at any given temperature must be equal to the
number of empty states (holes) created in the valence band. If the densities of states in the
 10.3 Semiconductors 415

valence conduction
f (E) band holes band
1

EF

0.5

electrons

0
-0.5 -0.25 0 0.25 0.5
E − EF (eV)

Figure 10.11 The Fermi–Dirac distribution function f(E) near EF at T = 300 K in a semiconductor with
a band gap of 0.4 eV. The shading represents occupied states.

valence and conduction bands are equal, this condition requires that the Fermi level be
located halfway between the two bands, as shown in Figure 10.11. In practice, the densities of
states of the two bands are not exactly equal, and EF shifts slightly from the halfway point,
but the shift is normally quite small.
We see from Figure 10.11 that the concentration of charge carriers is governed by the tails
of the Fermi–Dirac distribution. For crystal orbitals that lie within the conduction band, E −
EF is typically much larger than kT, and the distribution in Equation (10.13) simplifies to13:

1 1
f ðEÞ ¼ ffi ¼ exp½ðE  EF Þ=kT (10.20)
1 þ exp½ðE  EF Þ=kT exp½ðE  EF Þ=kT

It can be shown that the carrier concentration n in a semiconductor is proportional to the

value of f(E) at the conduction-band edge, where E − EF = Eg/2, thus:

n ∝ exp½Eg =2kT (10.21)

For silicon whose band gap Eg = 1.1 eV, Equation (10.21) gives n = 1×1016 m−3 at T = 300 K,
which translates to roughly one charge carrier for every 1012 atoms, many orders of magnitude
less than in a metal. This difference in carrier concentration helps to explain why high-purity
semiconductors have low conductivities (Table 10.1). Equation (10.21) also reveals another
important difference between semiconductors and metals. Semiconductors become more

13
Expressed in electronvolts, the Boltzmann constant is k = 8.617×10−5 eV/K and kT = 0.0258 eV at 300 K.
416 Conducting Materials

conducting as the temperature increases because the exponential increase in carrier concentration
is much larger than the increased scattering of the carriers due to lattice vibrations.

10.3.2 Doping
The conductivity of a semiconductor is very sensitive to aliovalent substitution (Section 2.4),
called doping. This sensitivity can be used to tailor the conductivity of a semiconductor.
Dopant atoms that have more valence electrons than the atoms in the host lattice are called
donors, because they donate electrons to the conduction band. Dopants that have fewer
valence electrons than the atoms in the host lattice are called acceptors, because they accept
electrons from the valence band.
Let’s take a closer look at what happens when phosphorus is added to an elemental
semiconductor like silicon: donor doping. It is instructive to first consider the local bonding.
Silicon is a covalently bonded solid that adopts the diamond structure type, where each atom
is tetrahedrally coordinated (Figure 1.37). In order to form two-electron bonds with its
neighbors, each silicon atom must use all four of its valence electrons. Phosphorus has five
valence electrons so that when it substitutes for silicon, there is one extra electron that is not
needed to satisfy local bonding requirements. This extra electron can easily be excited into
the conduction band where it is free to move through the crystal.
Let’s shift from this local picture to the delocalized view represented by a band-structure
diagram. In silicon, the valence bands are made up of bonding crystal orbitals and the
conduction bands of antibonding crystal orbitals (Section 6.6.3). As the bonding crystal
orbitals are fully occupied, the extra electron associated with the phosphorus dopant goes
into the conduction band. Because the conduction band is nearly empty, the electron has
many nearly degenerate crystal orbitals to choose from and is relatively free to move through
the crystal. Given our earlier estimate of one free charge carrier for every 1012 atoms in pure
Si, we see that even very low levels of doping can dramatically alter the carrier concentration.
In theory, even one part in 109 of the phosphorus donor would increase the carrier
concentration by three orders of magnitude over that of pure silicon!14 Consequently, carrier
concentration and conductivity can be controlled by introducing dopants in defined quan-
tities, something that cannot be done with a metal. Such extreme sensitivity to dopants also
explains why semiconductor processing takes place in clean rooms where great care is taken
to avoid unintentional incorporation of impurities.
Acceptor doping can be illustrated by aluminum substituting for silicon (Kröger–Vink
notation AlSi 0 in Section 2.6). In this case, the dopant atom does not have enough valence
electrons to satisfy the local bonding requirements of the matrix. The inability to completely
fill the bonding states results in an unfilled crystal orbital at the top of the valence band,
which, as we learned earlier, is called a hole. If we apply an electric field, electrons from

14
In real samples, it’s quite difficult to reduce the impurity levels to the point where they have no influence on the
conductivity. Germanium crystals have been grown with impurity levels as low as 1 part in 1010.
 10.3 Semiconductors 417

e– host lattice
atom (Si)
positively charged
dopant (PSi•)

Figure 10.12 Orbit of a conduction-band electron around a (positively charged) donor.

elsewhere in the crystal can move into the hole. It is easier to keep track of the movement of
holes (which are few in number) than it is to follow the movement of electrons (which are
much more numerous), provided we remember one key distinction—holes are positively
charged and move in the opposite direction to electrons when an electric field is applied.15
The extra electron that phosphorus donates to the conduction band is not quite as free as
the above description would suggest. Phosphorus has one more proton than silicon, which
makes the phosphorus site positively charged (PSi•). The conduction-band electron is
attracted to this positively charged site and will be bound to the impurity much like the
electron in a hydrogen atom is bound to the nucleus (see Figure 10.12). In fact, with some
modifications, we can estimate the energy that binds the electron to the substitutional
impurity just as we would calculate the energy levels in a hydrogen atom.
The differences between the bound states of a substitutional impurity and the energy levels
in a hydrogen atom are as follows. Firstly, the charge of the “nucleus” is screened by other
electrons and reduced by the relative dielectric permittivity, εr, of the host. Secondly, the
kinetic energy of the electron is modified because it is moving in the conduction band of the
semiconductor rather than in free space. Therefore, we need to use the effective mass m* in
place of the free-electron mass me. Finally, the energy zero corresponds not to a vacuum but
to the bottom of the conduction band, Ecb. Taking these factors into account, it can be
shown that the energy required to excite an electron bound to a donor into the conduction
band, the donor ionization energy, Ed, is equal to:
 2    2   2   
1 m Z hcRH Z m
Ed ¼ Ecb  En ¼ ¼ hcRH ðin JÞ (10.22)
εr me n2 nεr me

15
Consider a mostly filled auditorium with a few empty seats at the front. When the show starts and people are
encouraged to move closer to the stage, the people (analogous to electrons) move toward the front while the empty
seats (analogous to holes) move toward the back of the auditorium.
418 Conducting Materials

where c is the speed of light, h is Planck’s constant, RH is the Rydberg constant

(1.0973×107 m−1, the wavenumber 1/λ of the lowest-energy photon that can ionize a hydro-
gen atom), n is the principal quantum number of the hydrogen-like orbital, and Z is the
Kröger–Vink defect charge. As E = hν = hc/λ, the term hcRH gives the energy required to
ionize a hydrogen atom, 2.1799×10−18 J. Division by the elementary charge, e, gives the
Rydberg energy of 13.6 eV in an alternative form of Equation (10.22):
 2   
Z m
Ed ¼ 13:6 ðin eVÞ (10.23)
nεr me

We can also calculate the size of the hydrogen-like “atom” formed by the donor and its
electron. In the Bohr model of the hydrogen atom, the radius of an electron in a 1s orbital is
52.9 pm (the Bohr radius). In a doped semiconductor, the radius of the donor electron, rd, is
modified by the relative dielectric permittivity and the effective mass of the host:
m 
e
rd ¼ 52:9εr  ðin pmÞ (10.24)
m

Ed is the energy required to excite a hydrogenic electron from the ground state (principal
quantum number n = 1), where it is trapped, to the conduction band (n = ∞), where it can
move freely through the crystal. It is referred to as the donor ionization energy, as shown in
Figure 10.13. Due to the dielectric screening of the semiconductor and the effective-mass
correction (in most semiconductors m*/me < 1), a substitutional impurity is much easier to
ionize than a hydrogen atom. For a group-15 dopant in silicon, the ionization energy is
calculated to be 0.03 eV, in reasonable agreement with 0.045 eV measured for phosphorus in
silicon. The ionization energies of selected dopants in silicon and germanium are listed in
Table 10.4, including acceptors. The ionization energies are in general smaller in germanium,
in part because Ge has a higher relative dielectric permittivity (εr = 16 versus 12 for Si). We
also see that for various dopants in a given host, the ionization energies are quite similar, as

conduction
band e− donor
E
ionization
energy, Ed
donor

acceptor
acceptor
ionization
e− energy, Ea
valence
band

Figure 10.13 Impurity ionization energies with respect to the valence- and conduction-band edges for
donor and acceptor doping.
 10.3 Semiconductors 419

Table 10.4 Ionization energies (in eV) of various donors and acceptors in Si and Ge.

Donors Si Ge Acceptors Si Ge

P 0.045 0.013 B 0.045 0.010

As 0.054 0.014 Al 0.057 0.011
Sb 0.042 0.010 Ga 0.065 0.011

we might expect given the fact that the chemical identity of the dopant does not enter into
Equation (10.22) (except its charge Z).
Figure 10.13 and Table 10.4 suggest that this treatment of a donor-bound electron trapped
beneath the conduction band can be inverted into an entirely analogous case of the acceptor-
bound hole trapped above the valence band. The positively charged hole experiences
a weak attraction to the negatively charged dopant, forming a hydrogen-like bound state.
Figure 10.13 shows how the acceptor states sit just above the valence band so that when an
electron jumps into this state, a hole is created in the valence band. The energy required to
excite the bound hole into the valence band is called the acceptor ionization energy, Ea. We
can see from Table 10.4 that Ea and Ed are of similar magnitudes.

10.3.3 Carrier Concentrations and Fermi Energies in Doped Semiconductors

Real semiconductor crystals invariably contain both donors and acceptors. Electrons in the
conduction band arise from ionized donor states, whereas holes in the valence band arise from
ionized acceptor states. In addition, conducting electrons and holes are created by thermal
excitation across the band gap. To further complicate matters, electrons can fall from the
donor levels into empty acceptor levels. Deriving expressions for the concentration of charge
carriers in a completely general manner is therefore quite complicated. Fortunately, the
situation is much simpler in a few important limiting cases of temperature and doping.
Let the total volume concentration of electrons in the semiconductor conduction band be
n (just as it is for metals) and the concentration of holes in the valence band p (both in units
of m−3). In the absence of doping, we can use Equation (10.21) to estimate n, and because
a hole is created for each electron excited into the conduction band, p = n.
We will then use the symbols Nd and Na to denote the concentrations of the respective
donor and acceptor impurities (in m−3). When the concentrations of the two types of dopants
are not equal, the behavior of the semiconductor is determined largely by the majority
dopant. Semiconductors containing more donors than acceptors (Nd > Na) are said to be
n-type semiconductors. Semiconductors where the acceptor dopants are in the majority (Na >
Nd) are said to be p-type semiconductors.
The carrier concentration and Fermi level in doped semiconductors show temperature
dependences that differ from intrinsic semiconductors. Consider the behavior of an n-type
420 Conducting Materials

semiconductor. At very low temperatures, there is not enough thermal energy to ionize the
donor states and the carrier concentration in the conduction band is very low. As the donor
states are almost fully occupied, the Fermi level must lie above the ionization energy Ed of
these donor states.16 This temperature range is called the freeze-out regime. In this regime, we
can estimate the carrier concentration n by substituting Ed for the band gap in Equation
(10.21):
n ∝ exp½Ed =2kT (10.25)

When the temperature is increased above the freeze-out regime, the donor states begin to
ionize. As a result, n increases and EF moves downward. At some higher temperature, nearly
all donor states are ionized and carrier concentration is effectively equal to the dopant
concentration, n = Nd. This is called the saturation regime because the carrier concentration
remains essentially constant as the temperature continues to climb.
As the temperature continues to increase, the concentration of the intrinsic carriers that
are generated via the valence- to conduction-band excitation increases exponentially. At low
temperatures, their concentration is much smaller than the concentration of carriers that
come from doping, and we can effectively ignore the intrinsic-carrier contribution (at least
for the majority carriers, electrons in this case). However, with increasing temperature, the
concentration of intrinsic carriers will eventually approach and exceed the now saturated
concentration of carriers coming from dopants. At high enough temperatures, the concen-
tration of conduction-band electrons generated by thermal excitation becomes dominant
and the carriers coming from doping can be neglected. This is called the intrinsic regime
because the properties of the semiconductor are essentially the same as an undoped semi-
conductor of the same material. EF is then located at the midpoint of the band gap, and n ≈ p,
both of which are governed by Equation (10.21).
The temperature dependence of the carrier concentration for an n-type semiconductor is
plotted as a function of temperature in Figure 10.14. The carrier concentration in a p-type
semiconductor evolves in a similar manner.
The crossover from the freeze-out regime to the saturation regime depends upon the
donor ionization energy. For typical dopants, this occurs well below room temperature.
The crossover from the saturation regime to the intrinsic regime depends upon the dopant
concentration and the band gap. Apart from very small band-gap semiconductors and/or
very low dopant concentrations, the intrinsic regime is only reached at temperatures exceed-
ing room temperature. For devices, it is preferable to operate in the saturation regime where
the carrier concentration is not strongly temperature dependent. For most doped semicon-
ductors, this is the situation at room temperature, which means the Fermi level is not too far
from the ionization energy of the dominant dopant. Thus, EF is strongly shifted toward the

16
Remember that f(E) = 0.5 when E = EF, so that EF must lie above the donor ionization energy until more than 50%
of the donor states are ionized.
 10.3 Semiconductors 421

e−
EF
EF EF

h+
freeze-out regime intrinsic regime
3

saturation regime
(n ≈ Nd)

n/Nd

1
total carriers

intrinsic
carriers
0
0 100 200 300 400 500
T (K)

Figure 10.14 Temperature variation of the total carrier concentration per donor, n/Nd, for an n-type
semiconductor. In the upper schematic, the filled circles and shaded regions represent electrons and the
open circles, holes.

conduction band in an n-type semiconductor, and toward the valence band in a p-type
semiconductor.

10.3.4 Conductivity
The electrical conductivity of a semiconductor is governed by the carrier concentration and
mobility, just as it is for a metal. Unlike a metal, we now have two types of carriers, electrons
and holes, whose currents are additive. We can therefore modify Equation (10.10) to
a general expression for the conductivity of a semiconductor:

σ ¼ σn þ σp ¼ nμe e þ pμh e (10.26)

where μe and μh are the mobilities of the electron and the hole, respectively. Like a metal, the
scattering from lattice vibrations increases with temperature (Section 10.2.7), which tends to
decrease the mobility and hence the conductivity, but, unlike a metal, the carrier concentra-
tion increases exponentially with temperature. The latter term dominates, and, as a result,
the temperature dependence of the conductivity shows approximately the same temperature
variation as the carrier concentration (Figure 10.14). When new materials are prepared, the
422 Conducting Materials

Table 10.5 Band gaps and carrier mobilities for various

semiconductors at room temperature.

Eg, eV μe, m2/(V s) μh, m2/(V s)

Si 1.11 (indirect) 0.19 0.050

Ge 0.67 (indirect) 0.38 0.182
III–V
GaAs 1.43 (direct) 0.90 0.050
InAs 0.36 (direct) 3.30 0.046
InSb 0.18 (direct) 8.00 0.075
II–VI
ZnS 3.6 (direct) 0.012 0.0005
ZnSe 2.58 (direct) 0.053 0.0016
CdTe 1.50 (direct) 0.030 0.0065

temperature dependence of conductivity is often used to differentiate metallic conductors (σ

decreases as T increases) from semiconductors (σ increases as T increases).
For an intrinsic semiconductor, the carrier concentrations depend on the band gap and
temperature through Equation (10.21). Semiconductors with a small band gap will be more
conductive than those with a large band gap. This explains why germanium has a higher
intrinsic conductivity than silicon (see Table 10.1).17 The carrier mobilities depend on
effective masses (Section 10.2.5) which in turn depend on the curvature of the appropriate
bands in the band-structure diagram, Equation (10.15). Because electrons and holes reside in
different bands, their mobilities are different, Table 10.5. The observation that electron
mobilities are higher than hole mobilities is general to nearly all semiconductors. The largest
electron mobilities are found for III–V semiconductors.

10.3.5 p–n Junctions

The p–n junction is an integral part of many electronic devices, including transistors,
light-emitting diodes (LEDs), and solar cells. A p–n junction is formed when a p-type
semiconductor and an n-type semiconductor are brought into intimate contact. Typically,
the host semiconductor is the same on either side of the junction, and the junction is
formed by an abrupt transition from acceptor to donor doping. In practice, rather than
fusing together two separate crystals, dopants of one type are diffused into a layer of
a semiconductor that was already doped in the opposite sense. For example, acceptors can
be diffused into an n-type semiconductor crystal, or conversely donors can be diffused
into a p-type crystal.

17
We need to remember that doping, which is always present to some extent, also plays an important role in
determining the carrier concentration.
 10.3 Semiconductors 423

Setting aside the practicalities of manufacturing a p–n junction, let’s consider what would
happen if we literally brought crystals of the same host semiconductor, one doped p-type and
the other n-type, into intimate contact and let the system come to equilibrium. Initially, the
Fermi level in the n-type material is higher than it is on the p-type side (Figure 10.15). This
imbalance causes the electrons on the n-type side to flow over to the p-type side and holes to
flow in the opposite direction. The electrons and holes recombine, annihilating each other
over a fairly narrow region on either side of the junction. When the electrons flow out of the
n-type side, they leave behind ionized donors that carry a positive charge (e.g. PSi•). In
a similar manner, the holes leave behind ionized acceptors that carry a negative charge (e.g.
AlSi 0 ). As a result, the junction develops an electric field that opposes further diffusion of
electrons from the n-type side and holes from the p-type side into the junction. This process is
illustrated in Figure 10.15.
When two solids are in equilibrium with each other, their respective Fermi levels must be at
the same energy. For this to happen, while maintaining the p- and n-type character of the two
sides far away from the junction, the valence- and conduction-band edges must bend across
the junction, as shown on the lower left-hand side of Figure 10.15. What is the physical origin
of band bending in a p–n junction? It is a direct consequence of the uncompensated charges
of the ionized impurities. As shown on the lower right-hand side of Figure 10.15, a positive
charge develops on the n-type side of the junction, and a corresponding negative charge

Before equilibrium ionized

ionized
acceptor electron
hole donor

EF − +
− +
− +
−
− − + +
EF − +
− − + +

p-type n-type p-type n-type

At equilibrium
depletion region

− − +
− + +
EF − − − + +
− + + +
− − +
p-type − − + +

depletion p-type n-type

region n-type

Figure 10.15 A schematic representation of the development of a p–n junction before (upper) and after
(lower) equilibrium is established. The energy-level diagram (left) and charge distribution (right) across
the junction are shown.
424 Conducting Materials

develops on the p-type side. Because electrons are attracted to a positive charge and repelled
by a negative charge, the band edges on the n-type side of the junction are lowered and those
on the p-type side are raised. The drop in potential energy that occurs upon crossing a p–n
junction made from silicon is on the order of 0.5–0.7 eV. Dividing this potential-energy
difference in electronvolts (eV) by 1 (electron) we obtain the built-in potential, Vbi, in volts.18
At the center of the junction, n ≈ p, and the carrier concentration is characteristic of an
intrinsic semiconductor. As discussed in Section 10.3.1, carrier concentrations in intrinsic
semiconductors are very low unless either Eg is small or the temperature is large. Therefore,
the transition region is almost completely depleted of mobile carriers, and for this reason is
called the depletion region. The depletion region acts as an insulating barrier between n-type
and p-type sides of the junction.
The most basic property of a p–n junction is rectification, which refers to the phenomenon
whereby current can pass much more easily in one direction than in the other. Energy-level
diagrams for a p–n junction in both reverse and forward bias are shown on the left-hand side
of Figure 10.16. In these diagrams, it is thermodynamically favorable for electrons and holes
to occupy the states where they have the lowest potential energy. Electrons in the mostly
empty conduction band will therefore flow downhill, while holes inside the valence band
move uphill. In the absence of an external electric field, the system is at equilibrium and there

conduction
band 0.2
I
p-type
0.15
valence band
p-type
n-type
0.1
n-type

0.05

voltage voltage
source source 0
− + + − -0.4 -0.2 0 0.2 0.4
V
-0.05
reverse bias forward bias
p n p n

Reverse bias Forward bias Current–voltage

(V < 0) (V > 0) (I–V) curve

Figure 10.16 Current as a function of applied voltage (V) for a p–n junction acting as a rectifier. The
band-bending diagrams shown in the upper left and middle panels depict the potential-energy drop across
the junction.

18
For definition of electronvolt, see Footnote 12 in Chapter 5.
 10.3 Semiconductors 425

is no net flow of current. When the equilibrated junction is connected with its p-side to the
negative pole and n-side to the positive pole of an external voltage source, as shown in the
left-hand panel of Figure 10.16, the band bending at the junction increases. Since the applied
voltage results in practically zero current flow, the junction is said to be in reverse bias. If the
polarity of the voltage source is reversed, as shown in the middle panel of Figure 10.16, the
band bending at the junction is reduced, and current begins to flow across the junction, which
is now said to be in forward bias. If you put the junction into a circuit of alternating current,
only one polarity passes through the junction.19
Let’s consider the effects of applied biases at the microscopic level. In reverse bias, the
positive applied voltage pulls electrons on the n-type side away from the junction, and
similarly the negative pole attracts holes on the p-type side. This leads to more ionized
donors on both sides of the junction, increasing the width of the depletion region, as well as
the magnitude of the potential-energy drop across the junction. As a result, the flow of
current across the junction is very small. A forward bias has the opposite effect; once the
applied voltage exceeds Vbi, the band bending is fully suppressed and the current increases
sharply, leading to the asymmetric I–V curve shown in Figure 10.16. This is a feature that is
characteristic of a rectifying junction.

10.3.6 Light-Emitting Diodes and Photovoltaic Cells

When a forward bias is applied to a p–n junction, electrons from the n-type side and holes
from the p-type side are forced into the depletion region, and this leads to a high rate of
recombination. Energy equal to the band gap is released, either radiatively or non-
radiatively, as the electrons fall into the holes. If the semiconductor is a direct-gap semicon-
ductor, radiative-recombination events are often favored, leading to emission of photons with

hν
Figure 10.17 Light
e−
absorption and car-
rier separation in
a photovoltaic cell.
EF

p-type
h+

depletion n-type
region

19
A rectifier is a device that converts alternating current to direct current.
426 Conducting Materials

energies determined by the band gap. This is the principle behind light-emitting diodes
(LEDs) that were discussed in Section 7.9. Non-radiative-recombination events, where the
recombination energy is dissipated by lattice vibrations, reduce the efficiency of LEDs. This
is particularly problematic for semiconductors with indirect band gaps.
Photovoltaic cells operate much like LEDs, but the timing of events is opposite, as shown
in Figure 10.17. Photons with energies greater than the band gap are absorbed by the
semiconductor creating an electron–hole pair. Normally the electron and hole would recom-
bine, but, if the photon is absorbed in the depletion region, band bending drives the electron
downhill towards the n-type side and the hole uphill toward the p-type side. The separated
charge carriers can then be run through an external circuit to do work.
The voltage that can be obtained from a photovoltaic cell is limited by the extent of band
bending at the p–n junction, the upper limit being the band gap of the semiconductor. The
current that can be obtained is limited by the number of photons absorbed in the depletion
region. This creates a trade-off in terms of the semiconductor band gap: increasing the band
gap increases the output voltage yet lowers the current, because photons with energies
smaller than Eg are not absorbed. Decreasing the band gap has the opposite effect. For
generating energy from sunlight incident upon the Earth’s surface, the optimum band gap is
~1.4–1.5 eV.

10.3.7 Transistors
The transistor is a key component in many electronic devices, and its discovery ranks as one
of the most important inventions of the twentieth century. The first three-contact transistor
was built at Bell Labs in 1947. The invention sparked a huge research effort that eventually
led to integrated circuits and computer chips. There are different types of transistors but the
most common transistor in today’s integrated circuits is the metal-oxide-semiconductor field-
effect transistor (MOSFET). The basic design of a MOSFET is shown in Figure 10.18.
Through appropriate doping, two flank regions of a p-type semiconductor are made n-type.

source gate electrode drain

electrode electrode
gate dielectric d

source channel drain

n-Si n-Si

p-Si

Figure 10.18 A schematic representation of an n-channel MOSFET on a Si wafer.

 10.3 Semiconductors 427

These two n-type regions are called the source and the drain, respectively, while the p-type
region that separates them is called the channel. An insulating-oxide dielectric is deposited
on top of the channel. This oxide is called the gate. Finally, metallic contacts are made to the
source, drain, and gate. The architecture described here corresponds to an n-channel
MOSFET.20 If the n- and p-type regions are reversed, a p-channel MOSFET is formed.
In the n-channel MOSFET, the source is connected to the negative voltage and the drain
to the positive voltage. In the absence of voltage on the gate, the consecutive n–p and p–n
junctions at the source–channel and drain–channel interfaces do not allow any current
through the transistor, even in the presence of a potential difference between the source
and drain. When a positive voltage is applied to the gate, the holes in the p-type channel are
repelled, and minority carrier electrons are attracted to the gate. If the gate voltage exceeds
a threshold value, this effect is large enough to create a thin inversion layer (n-type > p-type)
in the channel. Because the inversion layer acts as an n-type semiconductor, there is now
a path for current to flow between the source and the drain. In this way, a MOSFET can act
as a switch: “on” when the gate voltage exceeds the threshold value and “off” when it does
not. The switching speed of a MOSFET depends in part on the carrier mobility. For this
reason, n-channel MOSFETs, where the current is carried by electrons, are generally
preferred to p-channel MOSFETs, where the current is carried by lower-mobility holes
(see Table 10.5).
Until relatively recently, the materials used in a MOSFET were not very diverse. The
source, drain, and channel were made from silicon; the gate dielectric was SiO2 formed by
oxidizing the silicon; and the metallic contact to the gate was heavily doped polycrystalline
silicon. The reliance on silicon is based on the need to simplify the fabrication of millions of
transistors on a single chip. In recent years, things have become more interesting as the drive
to put more and more transistors on a chip has led to a dramatic reduction of the channel
length, L.
The gate voltage needed to create an inversion layer depends upon the capacitance of the
gate dielectric, which in turn depends upon the dimensions and the dielectric permittivity of
the gate dielectric. Recollect from Section 8.1.1 that the capacitance of a parallel-plate
capacitor C = εrε0A/d. As the size of the MOSFET decreases, the area A of the gate dielectric
decreases and the capacitance goes down. If no other changes are made, a higher voltage is
required to create the inversion layer, but this is undesirable because it increases power
consumption and generates more heat. For decades, engineers have been compensating for
the shrinking gate area by steadily decreasing the thickness d of the SiO2 gate dielectric. By
the time the channel length reached 65 nm in 2006, the SiO2 layer was only 1.2 nm thick!
Further reduction in this thickness was not feasible because leakage, including quantum-
mechanical tunneling, across the gate dielectric becomes problematic.

20
When the gate voltage is sufficiently large to turn on the transistor the channel becomes n-type and current flows
across the channel, hence the name n-channel MOSFET.
428 Conducting Materials

The solution to this engineering dilemma is straightforward in principle: compensate for

the decrease in A by increasing εr. Since the relative permittivity of amorphous SiO2 is only
3.9, finding a higher εr alternative does not sound too daunting. However, to be practical, the
replacement material must meet several criteria. It must be highly insulating, not form
compounds with silicon at the gate–channel interface, and have bands that are sufficiently
offset from silicon to avoid unwanted leakage current. The material that emerged from an
extensive search was amorphous HfO2, whose permittivity is approximately 30. The seven-
fold increase in εr allowed engineers to shrink the channel length to 32 nm by 2010, while
increasing the thickness of the gate dielectric. At the same time, it was necessary to change the
gate-electrode material from heavily doped Si to more conducting materials like TiN. To
further increase the density of transistors, manufacturers have since moved away from the
planar geometry of a MOSFET to a geometry called a FinFET, where the gate wraps around
three sides of the channel. Further efforts to keep pace with Moore’s law21 are likely to
involve incorporation of even more exotic materials and unconventional architectures.

10.4 Transition-Metal Compounds

Compounds containing transition-metal ions with partially filled d orbitals are a fascinating
and complex class of materials. Their electrical and magnetic properties vary widely and can
change dramatically in response to subtle changes in temperature, pressure, and/or chemical
substitution. To a first approximation, we would expect most transition-metal compounds to
be metallic because the bands originating from d orbitals are partially filled. A glance at the
properties of transition-metal compounds quickly reveals behavior that cannot be explained
within the conventional band-structure approach. For example, some metal oxides with
partially filled d orbitals are metallic (e.g. TiO and SrFeO3), while others are semiconducting
(e.g. NiO and LaFeO3). Compounds like VO2 and LaTiO3 show both behaviors, semicon-
ducting at low temperature and metallic at high temperature. In the sections that follow we
explore the reasons behind this behavior.

10.4.1 Electron Repulsion: The Hubbard Model

The fundamental reason transition-metal compounds show complex electronic behavior can
be traced back to strong repulsive interactions between electrons. In the band model we have
used up to this point, the effects of electron–electron repulsion have been assumed to be
uniform throughout the crystal. This approximation tends to work well for metals and
semiconductors where the bands are wide and the electron wavefunctions highly delocalized,
but there are many instances where this simplifying approximation breaks down. Hund’s

21
Moore’s law is an empirical prediction that the density of transistors on an integrated circuit will double approxi-
mately every two years.
 10.4 Transition-Metal Compounds 429

Ti2+ Ti3+ Ti4+

e− e− e− e− e− e− e− e−

e− e− e− e− e− e− e−
e−
e− e− e− e− e− e− e− e−

e− e− e− e− e− e− e− e−

(a) (b)

Figure 10.19 (a) A square array of Ti3+ ions. (b) The same array after an electron moves from one site to
another, initiating conduction.

rules (Section 7.3.2) owe their existence to electron–electron repulsion, and high-spin transi-
tion-metal ions (Section 5.3.9) would not exist were it not for electron–electron repulsions.
Compounds where the physical properties cannot be understood without taking electron–
electron interactions into account are called strongly correlated materials. These materials
play a prominent role in many functional materials, particularly those discussed in chapters
11 and 12.
Accounting for the effects of electron repulsion in solids is a challenging problem. One
approach, the Hubbard model, simplifies the problem by assuming that the only electron–
electron repulsions of importance are those between valence electrons on the same atom. To
understand the Hubbard model, consider the 2D array of Ti3+ ions shown in Figure 10.19a.
As Ti3+ is a d1 ion, the above-defined repulsions are absent. The situation changes once we
allow the electrons to move, as they must for the material to be conducting. Once an electron
moves from one titanium atom to another, Figure 10.19b, we create a Ti site containing two
electrons and must pay an energy penalty, U, equal to the on-site repulsion between the two
electrons.
What is the magnitude of this “Hubbard” U ? When the electron moves from one site
to another, the net change in the system can be expressed as a disproportionation reaction:
2Ti3+ → Ti4+ + Ti2+. For titanium ions in the gas phase, we can precisely determine this
quantity from the ionization energy, IE, and the electron-gain enthalpy, Heg, of a Ti3+ ion22:

Ti3+(g) → Ti4+(g) + e− IE of Ti3+ = 4th IE of Ti = 43.3 eV

−
Ti (g) + e → Ti (g)
3+ 2+
Heg of Ti3+ = −(3rd IE of Ti) = −29.3 eV
2Ti (g) → Ti (g) + Ti (g) U = IE + Heg = 14.0 eV
3+ 4+ 2+

22
Remember from Chapter 5 that electron-gain enthalpy Heg has the same magnitude and opposite sign as electron
affinity, EA.
430 Conducting Materials

This is a very large energy, much larger than the band gap of a semiconductor. If U were
always this large, no transition-metal compounds would be metallic. However, in solids, the
magnitude of U is reduced significantly by two factors. Firstly, the electron clouds of the
surrounding anions will be polarized toward the site that lost an electron (Ti4+) and away
from the site that gained the electron (Ti2+). Secondly, the electron wavefunctions in a solid
extend over a larger region than they do in an isolated atom, which reduces the on-site
electron repulsion. The Hubbard U values of transition-metal ions are difficult to determine
precisely, but there is general agreement that energies of 2–5 eV are appropriate for first-row
(3d) transition-metal ions. The 4d and 5d orbitals extend further from the nucleus than the
3d orbitals, which leads to smaller U values for second- and third-row transition-metal ions.
While the on-site repulsion U favors localized electrons, there is a gain in kinetic energy
that results from electron delocalization and hence favors metallic behavior. In strongly
correlated materials, the competition between these two effects determines whether the
material will be insulating or metallic. For transition-metal compounds, the competing
forces tend to be closely balanced, so that either metallic or insulating behavior can be
realized depending on several factors.23
To understand the behavior of transition-metal compounds, we incorporate the Hubbard
model into the band model that we have been using up to this point. To illustrate this
integration, let’s return to our 2D array of Ti3+ ions. To make things a little more realistic, we
add nitride ions so that each titanium is surrounded by a square of nitride ions and vice versa
(see Figure 10.20). Unlike our earlier array of Ti3+ ions, this structure is charge-balanced.

N
Ti
energy

Μ point
EF

Γ Μ Γ point
k

Figure 10.20 The band structure for the Ti 3dxy band in the 2D array of TiN discussed in the text,
including the crystal orbitals at Γ and M.

23
For lanthanoid compounds where the electrons reside in highly contracted 4f orbitals, the Hubbard U values are
large, 5–12 eV, and localized-electron behavior dominates.
 10.4 Transition-Metal Compounds 431

The resulting array is not a completely hypothetical construct, it is identical to a (001) layer in
the NaCl-type structure of TiN.24
To keep things simple, we will only consider the Ti 3dxy band. Firstly, we derive the band
structure in the absence of electron-repulsion effects. There is one titanium atom per unit cell, so
the band structure consists of a single band. One that runs uphill from Γ (kx = ky = 0), where the
Ti–Ti overlap is bonding, to M (kx = ky = π/a), where the Ti–Ti interactions are antibonding
(Figure 10.20). At both Γ and M, the N 2p orbitals cannot contribute to the crystal orbital for
reasons of symmetry (see Section 6.5.2). Each titanium ion has a single d electron, so the dxy
band is half filled. The resulting band structure, where the Fermi level cuts through the band at
the halfway point, is reminiscent of the band structure of the 1D hydrogen-atom chain encoun-
tered in Chapter 6 (Figure 6.5). From this picture, we would expect metallic conductivity.
What happens when we introduce on-site electron repulsion? The Hubbard U splits the
single two-electron band into two one-electron bands, which we will refer to as the lower
Hubbard band and the upper Hubbard band. The lower band reflects the energy when each
titanium ion holds a single electron. The upper band represents the higher energy that would
be experienced by an electron that moves to a site where there is already an electron present.
If the titanium atoms are sufficiently far apart, their interaction is negligible. In this
scenario, each band will be flat and the energy separation between bands will be equal to
the on-site repulsion U of two electrons (the Hubbard U). If we shrink the lattice parameter,
the Ti 3dxy orbitals begin to overlap as the titanium atoms approach each other and the
bands develop a width, W. In the absence of mixing with other bands, the lower and upper
Hubbard bands will have the same width. The bandwidth will increase as the Ti–Ti distance
decreases, similar to what was demonstrated for the 1D hydrogen chain in Figure 6.5. When
the Hubbard U is larger than the bandwidth W, as shown on the left-hand side of Figure
10.21, the lower Hubbard band is filled and separated from the empty upper Hubbard band
(remember the Hubbard bands can only hold one electron each). In this scenario, on-site
repulsion prevails and leads to localized electrons and semiconducting behavior.
Compounds that are semiconducting (insulating) because the on-site electron repulsion is
larger than the bandwidth (U > W) are called Mott–Hubbard insulators. When the band-
width W exceeds the Hubbard U (W > U), as shown on the right-hand side of Figure 10.21,
the two bands overlap, and the Fermi level cuts through both bands. In this case, there is no
gap between the bands, and metallic behavior is observed.

10.4.2 Transition-Metal Compounds with the NaCl-Type Structure

To see the effects of electron–electron repulsion in real compounds, perhaps the best place to
start is with binary oxides and nitrides that crystallize with the NaCl-type structure. In this
structure, the coordination octahedra share edges in all three dimensions, as illustrated on

24
The Ti–Ti distance in TiN (3.00 Å) is very similar to the Ti–Ti distance in titanium metal (2.95 Å), so this structure is
a reasonable approximation to a hypothetical square array of d 1 ions.
432 Conducting Materials

Weak Ti–Ti interaction (W < U) Strong Ti–Ti interaction (W > U)

Mott–Hubbard insulator metal

energy upper Hubbard band

EF EF
U W
W
U

lower Hubbard band

Γ Μ Γ Μ
k k

Figure 10.21 The Ti 3dxy band in our 2D model of TiN for the respective weak (left) and strong (right) Ti–
Ti interactions.

the left-hand side of Figure 10.22 (a 3D analog of the 2D model of TiN discussed in the
previous section). Both metallic and insulating behavior can be found in this family, which
makes it a good place to study the competing forces at work in transition-metal compounds.
The transition-metal ions are octahedrally coordinated by the anions splitting the
d orbitals into t2g (dxy, dyz, dxz) and eg (dx2−y2, dz2) sets (Section 5.3.8). The t2g and eg orbitals
form relatively narrow bands that lie between the filled anion 2p bands and the empty cation
4s and 4p bands, as shown on the right-hand side of Figure 10.22. Because the cations are in
low oxidation states (+2 or +3), the metal–anion interactions are fairly ionic, which in turn
leads to relatively narrow eg bands.25 The bands arising from t2g orbitals are somewhat wider
because the d orbitals can overlap directly across the shared edges, as discussed in the
previous section and illustrated in Figure 10.20. The widths of the t2g and eg bands in TiO
are estimated to be approximately 6 eV and 4 eV, respectively [2, 3]. The widths of the eg
bands are sufficiently small that W < U, and the electrons that go into these orbitals are
localized. For the t2g bands, no such broad generalization can be made.
The properties of transition-metal monoxides with the NaCl-type structure are summar-
ized in Table 10.6. TiO is a metal and Pauli paramagnet (of delocalized electrons, Section
9.6.2). VO lies close to the crossover from metallic to semiconducting [3]. The conductivity is
sensitive to defects (both vanadium and oxygen vacancies are normally present), and the
magnetic susceptibility has a temperature dependence that is intermediate between a Pauli
paramagnet and a Curie–Weiss paramagnet (of localized unpaired electrons, Section 9.6.1).

25
The maximum width of a d orbital band is limited by the difference in energy between nonbonding and antibonding
crystal orbitals. Therefore, when the metal–anion distances are long and the orbital interactions are weak, as is often
the case for larger ions in low oxidation states, we expect narrow bands.
 10.4 Transition-Metal Compounds 433

Table 10.6 Structural, electrical, and magnetic properties of transition-metal monoxides with
NaCl-type structure.

Compound M–M distance (Å) d-orbital rmax (Å)* Electrical properties Magnetic properties†

TiO (d 2) 2.94 0.53 Metallic Pauli PM

VO (d 3) 2.89 0.48 Semimetallic PM
MnO (d 5) 3.14 0.41 Semiconductor AFM, TN =122 K
FeO (d 6) 3.03 0.38 Semiconductor AFM, TN =198 K
CoO (d 7) 3.01 0.36 Semiconductor AFM, TN =293 K
NiO (d 8) 2.95 0.34 Semiconductor AFM, TN =523 K
*rmax is the distance from the nucleus where the radial distribution function of the 3d orbitals reaches
a maximum value (Section 5.2.2). †PM = paramagnetic, AFM = antiferromagnetic.

energy

M 4s and 4p bands
(empty)

M 3d eg
bands
filling determined
by d-electron count
M 3d t2g
bands

anion 2p bands
(filled)

Figure 10.22 The NaCl-type structure (left), along with a schematic DOS diagram for first-row transi-
tion-metal nitrides and oxides with this structure (right).

The properties of the remaining compounds, MnO through NiO, are unambiguously char-
acteristic of localized electrons, implying that W < U.
The crossover from delocalized to localized electronic behavior is driven by contraction of
the d orbitals as we move from early to late transition metals. The calculated values of rmax
for the 3d orbitals in Table 10.6 show that this contraction is significant. The width of the t2g
bands decreases because the orbital contraction is not paralleled by a decrease in the metal–
metal distance,26 hence the orbital overlap is diminished and the t2g bands become
26
The expansion of the unit cell and hence the M–M distance on going from VO to MnO results from filling the eg
orbitals, which are σ antibonding. These orbitals are empty in TiO and VO and doubly occupied in MnO, CoO, and
NiO.
434 Conducting Materials

increasingly narrow. The increase in the Néel temperature, TN, can be attributed to the
increasing electronegativity of the transition-metal ion, which leads to an increase in cova-
lency of the metal–oxygen bonds and hence stronger superexchange interactions that propa-
gate through the M–O–M σ bonds (Section 9.8).
Many early transition-metal nitrides (TiN, ZrN, HfN, NbN, and TaN) are metallic and
Pauli paramagnetic, strong evidence that W > U. VN is also metallic, but its low-temperature
magnetic susceptibility shows a more pronounced temperature dependence than expected
for a Pauli paramagnet (Section 9.6.2). This is the first sign that electron repulsions are
starting to impact the properties. This becomes more pronounced with CrN, which orders
antiferromagnetically with a TN of 280 K, an indication of significant electron–electron
interactions. It is generally accepted that W ≈ U for CrN [4, 5]. MnN crystallizes with
a tetragonally distorted NaCl-type structure and orders antiferromagnetically at quite high
temperature, TN = 650 K [6]. Its electrical properties are not fully characterized, but
antiferromagnetism is typically associated with Mott–Hubbard insulators. The later transi-
tion-metal elements are not known to form nitrides with the NaCl-type structure.

10.4.3 Transition-Metal Compounds with the Perovskite Structure

The band structures of perovskites were examined in some detail in Chapter 6, but com-
pounds with ions that possess partially filled d orbitals were consciously avoided. The
electrical and magnetic properties of several transition-metal perovskites are listed in
Table 10.7. To understand the properties of these compounds, electron repulsion effects
must be considered.
A schematic DOS plot for a perovskite is shown in Figure 10.23. As was the case with the
NaCl-type compounds, the t2g and eg orbitals form bands that lie between the filled anion
2p bands and the empty cation 4s and 4p bands. Unlike NaCl-type compounds, there is no
direct metal–metal bonding in the perovskite structure, consequently the eg bands are wider
than the t2g bands, for reasons discussed in Section 6.6.5. To assess the competition between
localized- and delocalized-electron behavior, the width of these bands is a critical parameter.
Recall that both sets of bands become wider when the transition-metal oxidation state
increases, or when a 3d transition-metal ion is replaced by a 4d or 5d transition-metal ion
from the same group. These considerations are important for making sense of the trends in
Table 10.7.
The perovskites listed in Table 10.7 can be sorted by their electrical properties as either
semiconducting or metallic. The semiconductors can further be sorted into those that are
diamagnetic (LaScO3, SrTiO3, KNbO3, LaCoO3, and LaRhO3) and those that order anti-
ferromagnetically. Those that are diamagnetic are sometimes called band insulators/semi-
conductors because electron-repulsion effects are not needed to explain the presence of
a band gap: LaScO3, SrTiO3, and KNbO3 are semiconducting because the d-orbital bands
are empty, just like the perovskites discussed in Section 6.6.5. LaCoO3 and LaRhO3 are
semiconducting because the transition-metal ions have a low-spin d 6 configuration, resulting
 10.4 Transition-Metal Compounds 435

Table 10.7 The electrical and magnetic properties of selected perovskites.

Compound Electron configuration Electrical properties Magnetic properties

3+
LaM O3, M = 3d
LaScO3 d0 t2g0eg0 Semiconducting Diamagnetic
LaTiO3 d1 t2g1eg0 Semiconducting AFM, TN = 138 K
LaVO3 d2 t2g2eg0 Semiconducting AFM, TN = 142 K
LaCrO3 d3 t2g3eg0 Semiconducting AFM, TN = 290 K
LaMnO3 d4 t2g3eg1 Semiconducting AFM, TN = 100 K
LaFeO3 d5 t2g3eg2 Semiconducting AFM, TN = 750 K
LaCoO3 d6 t2g6eg0 Semiconducting* Diamagnetic*
LaNiO3 d7 t2g6eg1 Metallic Pauli PM
SrM4+O3, M = 3d
SrTiO3 d0 t2g0eg0 Semiconducting Diamagnetic
SrVO3 d1 t2g1eg0 Metallic Pauli PM
SrCrO3 d2 t2g2eg0 Metallic Pauli PM
SrMnO3 d3 t2g3eg0 Semiconducting AFM, TN = 235 K
SrFeO3 d4 t2g3eg1 Metallic FM, TC = 130 K
SrCoO3 d5 t2g4eg1 Metallic FM, TC = 280 K
AMO3, M = 4d
KNbO3 d0 t2g0eg0 Semiconducting Diamagnetic
BaNbO3 d1 t2g1eg0 Metallic Pauli PM
SrMoO3 d2 t2g2eg0 Metallic Pauli PM
SrRuO3 d4 t2g4eg0 Metallic FM, TC = 165 K
LaRuO3 d5 t2g5eg0 Metallic Pauli PM
SrRhO3 d5 t2g5eg0 Metallic PM†
LaRhO3 d6 t2g6eg0 Semiconducting Diamagnetic
LaPdO3 d7 t2g6eg1 Metallic Pauli PM
*Below 100 K, LaCoO3 is a diamagnetic semiconductor with a low-spin t2g6eg0 configuration. At higher
temperatures, electrons are excited into the eg bands leading to paramagnetism and metallic conductivity.
See text for more details. †The paramagnetism of SrRhO3 is intermediate between Pauli and Curie–Weiss.

in filled t2g bands and empty eg bands. In LaCoO3, the gap separating these two bands is very
small. As a result, LaCoO3 is semiconducting and diamagnetic only at low temperatures. As
the temperature rises, electrons in LaCoO3 are thermally excited into eg bands, leading
paramagetism and eventually to metallic conductivity (T > 650 K) [7, 8].
For each of the remaining (non-diamagnetic) semiconducting perovskites, either the t2g
bands, the eg bands, or both, are partially filled. When the bandwidth is smaller than the on-site
repulsion (W < U), they are Mott–Hubbard insulators, whereas, for the opposite case (W > U),
they are Pauli-paramagnetic metallic conductors with fully delocalized electrons. Those com-
pounds that are metallic, but do not show Pauli paramagnetism, namely the ferromagnetic
SrFeO3, SrCoO3, and SrRuO3 lie in a regime where W ≈ U. In these compounds, the
interactions between electrons are not sufficient to fully localize them, but they do impact
the properties in significant ways, similar to the previously discussed compounds VO and CrN.
436 Conducting Materials

energy
M 4s and 4p bands
(empty)

M 3d eg
bands filling determined
by d-electron count
M 3d t2g
bands

O 2p bands
(filled)

Figure 10.23 The perovskite-type structure (left), along with a schematic DOS diagram for first-row
transition-metal oxides with this structure (right).

It is interesting to consider the periodic trends that dictate whether a perovskite will be
a Mott–Hubbard insulator (W < U) or a Pauli paramagnetic metal (W > U). Among the
LaMO3 compounds where M is a 3d transition-metal ion, the combination of the smaller
3d orbitals and the relatively low oxidation state of M (+3) leads to fairly narrow bands.
Hence, all LaMO3 perovskites with partially filled 3d bands are Mott–Hubbard insulators
with the exception of LaNiO3. In LaNiO3, there are two factors that combine to make
W larger than the other LaMO3 compounds. Firstly, +3 is a high oxidation state for a late-
transition metal like nickel. This results in short, highly covalent Ni–O bonds, which in turn
raise the energy of the antibonding crystal orbitals at the R point, leading to wider t2g and eg
bands (see Section 6.6.5). That effect, coupled with the partially filled eg bands, push this
compound toward the W > U category. It should be noted that if La3+ is replaced by
a smaller lanthanoid ion (e.g. Nd3+), the induced octahedral tilting narrows the eg bands
(Section 6.6.5). This triggers a transition from metallic conductivity at high temperature to
semiconducting behavior at low temperature [9]. From this, we can infer that W in LaNiO3
is just barely large enough to stabilize delocalized metallic conductivity to low temperature.
Among SrMO3 perovskites containing 3d transition-metal ions, delocalized electrons and
metallic conductivity are the norm rather than the exception. At first glance, this may seem
surprising because the same M constitutes an antiferromagnetic semiconductor when the
A-site cation is La. This change in behavior can largely be attributed to the increased
bandwidth that results when the oxidation state of M increases from +3 to +4. The sole
exception in Table 10.7 is SrMnO3, where the d3 configuration leads to a half-filled t2g bands.
The on-site repulsion is particularly high for half-filled bands because for a t2g electron to
move from one Mn4+ to another, it must either go into the higher-energy eg bands or flip its
 10.5 Organic Conductors 437

spin and remain in the t2g-orbital manifold. It should also be noted that W is only slightly
larger than the Hubbard U for some of the metallic SrMO3 compounds in Table 10.7.
Replacing Sr2+ with the smaller Ca2+ leads to octahedral tilting distortions that narrow
the bandwidth (Section 6.6.5), resulting in semiconducting behavior for both CaCrO3 and
CaFeO3. The response of CaFeO3 to electron localization is particularly interesting, and is
discussed further in Chapter 11.
When M in SrMO3 is a second-row transition-metal ion with partially filled t2g or eg
orbitals, the larger spatial extent of the 4d orbitals leads to wider bands. Hence W > U, and
metallic conductivity is observed for all of the 4d perovskites in Table 10.7, except for the
band insulators KNbO3 and LaRhO3.

10.5 Organic Conductors

Organic compounds are not an obvious place to look for conducting materials. Most organic
solids consist of small molecules held together by relatively weak intermolecular forces. The
lack of chemical bonds between molecules does not favor electronic wavefunctions that are
delocalized over the entire crystal, as they must be in highly conducting materials. Until the
1970s, there was little enthusiasm for the idea that organic substances could be useful
conductors of electricity, but scientists have subsequently discovered organic substances
that are semiconducting, metallic, and superconducting. In fact, the conductivity of organic
substances spans a range comparable to that of oxides. Because organic conductors are
lightweight and potentially flexible, they lend themselves to novel applications where inor-
ganic conductors cannot be used, such as flexible electronics. They also offer new approaches
to processing. For example, some organic conductors can be dissolved in solvents and
printed like an ink [10]. In this section, we examine some prototypical organic conductors
and take a closer look at what makes these substances different from most organic materials.
A common feature among organic conductors is a conjugated π system: a network of
alternating single and double bonds where each atom is sp2 hybridized and the remaining
p orbital participates in a delocalized π-bonding network (Section 5.3.7). Conjugated mol-
ecules and polymers possess delocalized molecular orbitals (MOs), which are a necessary
criterion for an organic conductor. In conjugated systems, the highest occupied molecular
orbitals (HOMOs) have π-bonding character and the lowest unoccupied molecular orbitals
(LUMOs) have π-antibonding character. As the size of the conjugated network increases, the
HOMO–LUMO gap decreases, but, even for large conjugated systems, a gap generally
remains, which makes most organic conductors inherently semiconducting.27 To attain
metallic conductivity, it is necessary to either remove electrons from the π orbitals or add
electrons to the π* orbitals. As we will see, there are various strategies for introducing charge

27
We can think of graphene (Section 6.5.1) as the largest possible 2D conjugated system and in graphene the gap
between π and π* crystal orbitals goes exactly to zero.
438 Conducting Materials

carriers. Organic conductors can be broadly subdivided into three categories: conducting
polymers, polycyclic aromatic hydrocarbons, and charge-transfer salts.

10.5.1 Conducting Polymers

Polymers are a natural place to look for organic conductors because, like inorganic semi-
conductors, they contain extended networks of covalently bonded atoms. However, most
familiar polymers (polyethylene, polypropylene, polystyrene, polyester, Teflon, etc.) are
insulating because their backbones are made up of sp3-hybridized carbon atoms connected
by single bonds. Polyacetylene is the simplest polymer where every atom along the polymer
backbone is sp2-hybridized, and therefore part of a conjugated π network (Figure 10.24).
Let’s take a closer look at the electronic structure of trans-polyacetylene. As with
benzene (Section 5.3.7) and graphene (Section 6.5.1), the electrical and optical properties
are determined almost exclusively by the π and π* orbitals. For trans-polyacetylene, there
are only two carbon atoms in the repeat unit (Figure 10.25), which means the basis sets of
interest are the π and π* MOs of the C2H2 repeat unit.28 Once the basis sets, u(x), are
determined, the task is analogous to our earlier derivation of the band structure of a 1D
chain of H2 molecules (Section 6.2). The crystal orbitals corresponding to these two MOs
are shown at k = 0 and π/a in Figure 10.25.29 Immediately we see that at k = 0 the band
formed from the π MO is completely bonding and will have the lowest possible energy,
while the band formed from the π* MO is completely antibonding and will have the

H H H H
H H H
C C C C C C C C
C C C C C C C C
H H H
H H H H

trans-polyacetylene cis-polyacetylene

Figure 10.24 The trans- and cis- forms of polyacetylene.

28
In constructing the π and π* bands, we can neglect the orbitals that form σ bonds because the σ and π orbitals are
orthogonal.
29
In this treatment, we consider the C2H2 repeat unit as a unit cell with 1D translational symmetry. This approxima-
tion is a useful construct for visualizing the electronic structure of polyacetylene, but, in practice, long-range
crystalline order in polyacetylene varies depending on synthetic conditions.
 10.5 Organic Conductors 439

σ* bands (empty)

energy
H
π* band (empty)
C
C
π band (filled)
H
n
σ bands (filled)
repeat unit

k = 0, u(r) = π* k = π/a, u(r) = π*

π* band

energy
k = 0, u(r) = π k = π/a, u(r) = π

π band

crystal orbitals π/

Figure 10.25 The C2H2 repeat unit in trans-polyacetylene (top left). A block diagram of the electronic
structure (top right). The crystal orbitals formed from the carbon 2pz orbitals at k = 0 and π/a looking at
a side-on view of the chain (bottom left). The band structure, showing only the valence (π) and conduction
(π*) bands (bottom right).

highest possible energy. At k = π/a, the interaction between carbon atoms in neighboring
unit cells changes sign so that the band formed from the π MO runs uphill from k = 0 to π/
a and the band formed from the π* MO runs downhill. The individual widths of the π and
π* bands are estimated to be 5–6 eV [10].
If all carbon–carbon distances were identical, the bands would touch at k = π/a and trans-
polyacetylene would be a metal. However, the distances alternate between 1.44 Å (single
bond) and 1.36 Å (double bond).30 This alternation lifts the degeneracy and is responsible for
the presence of a band gap of 1.7–1.8 eV. As mentioned in Section 6.2, this is a classic case of
a Peierls distortion.
Conjugated polymers are “doped” by a different mechanism than inorganic semiconduct-
ors: by oxidation/reduction rather than aliovalent substitution. For example, when poly-
acetylene is exposed to iodine vapor, I2 molecules penetrate the substance, partially oxidizing
(CH)x chains and forming dilute cationic charges. In total, one I2 takes two electrons from
a pair of double bonds, turning them into single bonds of one C+ (where the π electron was
taken) and one C• radical (where the π electron remained), as shown in Figure 10.26, left. The
reduction product I− reacts with excess I2 to form I3− counter ions. When the double bond
becomes a single bond, a local distortion involving an increase in the C–C distance occurs.

30
This difference is not quite as pronounced as the difference between carbon–carbon double and single bond lengths
typically found in organic molecules: 1.32 Å (double) and 1.54 Å (single).
440 Conducting Materials

(1) pristine polyacetylene

(4) conversion to solitons

+
oxidation
+
(2) polaron

+ (5) charged solitons

oxidation
+

+
(3) bipolaron
+

Figure 10.26 Formation of charge carriers upon p-doping trans-polyacetylene.

In a conducting polymer like polyacetylene, the cation–radical defect pair can move along
the polymer chain, distorting the C–C distances in its vicinity as it moves. A charge carrier
that distorts the structure when it moves through a crystal is termed a polaron. In polyacet-
ylene, the positive charge of the polaron is not very mobile at low doping levels because it is
localized by attraction to the negatively charged counter ion I3−.
As the doping level increases, the polarons in a given chain eventually begin to interact,
and the potential exerted by negatively charged counter ions becomes more uniform.
When two polarons approach each other, they form a bipolaron as shown in Figure
10.26, left. While bipolarons are not the important charge carriers in trans-
polyacetylene, they are in other conducting polymers. In polyacetylene, the unpaired
electrons in the bipolaron tend to combine, forming a new double bond and leaving
behind two positively charged carbon atoms. At each of these carbocations, shown in
the lower right-hand side of Figure 10.26, the pattern of bond alternation is reversed (i.e.
the bond patterns on either side of the carbon are mirror images of each other). A carbon
atom that forms two single bonds and reverses the pattern of bond alternation is called
a soliton.31 In highly conducting samples of trans-polyacetylene, positively charged soli-
tons are the dominant charge carriers.
Interactions between chains also play an important role. Polymers contain chains of
different lengths, the packing of which invariably introduces some disorder. The interchain
bandwidths are at least an order of magnitude smaller than intrachain bandwidths. For these
reasons, conductivity in polymers is limited by interchain hopping. The mechanisms of
31
Solitons can also be neutral, in which case it is simply a free radical. When it is charged, it can be either positively
charged (carbocation) or negatively charged (carbanion). In all three cases, the pattern of single and double bonds
on either side of the soliton are mirror images of each other.
 10.5 Organic Conductors 441

interchain hopping are beyond the scope of this text, but suffice to say that the properties of
conducting polymers are sensitive to the distribution of chain lengths and crystallinity, both
of which depend on the details of synthesis and processing.
Although trans-polyacetylene is the simplest, first, and most highly conducting polymer,
its use in practical applications is limited by its instability towards oxidizing agents and
moisture. Additional examples of conducting polymers include doped polyaniline, polypyr-
ole, polythiophene, polyethylene dioxythiophene (PEDOT), and numerous others. Many of
these are quite stable in air, although their conductivities are not as high as polyacetylene.
The interested reader is directed to a review by Heeger [10] for more details about other
conducting polymer systems and their applications.

10.5.2 Polycyclic Aromatic Hydrocarbons

Polycyclic aromatic hydrocarbons (PAHs) are planar molecules made up of fused benzene
rings. We examined a subset of PAHs, the acenes, in Section 7.6 where we learned that the
band gap decreases as the number of the rings increases—from 4.69 eV in benzene to 1.85 eV
in pentacene (see Figure 7.14). The latter value is similar to the band gap of trans-
polyacetylene, which suggests that the larger PAHs might have useful electrical properties.

Box 10.1 Materials Spotlight: The discovery of conducting polymers

In the early 1970s, Hideki Shirakawa found a way to prepare films of polyacetylene via Ziegler–
Natta-catalyzed polymerization of gaseous acetylene (HC≡CH). When the polymerization was
carried out at low temperature (−78 °C), a copper-colored film was deposited on the walls of the
reaction vessel. This film was shown to be predominantly (>95%) cis-polyacetylene. When the
polymerization was carried out at higher temperatures (150 °C), a silver-colored film was
deposited that was subsequently shown to be trans-polyacetylene.
While polyacetylene has a conjugated network, Shirakawa’s films were not particularly good
conductors for the reasons discussed in the text. The trans isomer has a band gap of 1.7–1.8 eV
and a conductivity of 10−2 S/m to 10−3 S/m, while the cis isomer has an even larger band gap
(~2.2 eV) and a lower conductivity (10−7 S/m to 10−8 S/m). When Shirakawa told Alan
MacDiarmid about his silvery films, MacDiarmid suggested that it should be possible to
increase their conductivity through doping. Shirakawa, MacDiarmid. and Alan Heeger
showed that when exposed to halogen gas (I2, Br2, or Cl2), polyacetylene is partially oxidized,
which results in p-type doping and leads to a dramatic increase in conductivity. By 1977, they
were able to raise the conductivity of trans-polyacetylene films to 3×103 S/m [11]. Subsequently,
conductivities as high as ~105 S/m have been obtained in doped polyacetylene films. In 2000, the
three scientists shared the Nobel Prize in Chemistry for this discovery.
442 Conducting Materials

While molecules with a conjugated π system possess delocalized MOs, not all conjugated
molecules form conductors when they crystallize. For a molecular crystal to be conducting, there
must be communication between MOs on different molecules. The optical properties of PAHs
provide a clue to this interaction. For example, the longest-wavelength absorption for an isolated
pentacene molecule (in solution or frozen in an argon matrix) is at ~580 nm, corresponding to
a HOMO–LUMO gap of 2.1 eV. When pentacene crystallizes, this absorption shifts to ~670 nm,
corresponding to a band gap of 1.8 eV [12, 13]. The red shift of the absorption edge implies
interactions between discrete MOs to form bands. However, compared with inorganic semicon-
ductors, or even the intrachain π and π* bands in polyacetylene, the bands in pentacene are very
narrow, only a few tenths of an electronvolt wide. Bandwidths of this magnitude are typical of
small-molecule organic conductors.
The widths of the bands formed from π and π* MOs are sensitive to the way molecules
pack in the solid state. Linear acenes, like tetracene and pentacene, tend to pack in
a “herringbone” arrangement (Figure 10.27) where the long edge of one molecule points
toward the face of another. The valence bands32 in pentacene are formed from the HOMO of
a pentacene molecule that has six nodal planes. As shown in Figure 10.27, the highest-energy

side view top view

HOMO

crystal orbital with antibonding crystal orbital with bonding intermolecular

intermolecular interactions interactions

Figure 10.27 Top: The HOMO of pentacene C22H14 with its six nodal planes seen sideways and from top.
Bottom left: Antibonding pattern of the “sideways” intermolecular crystal-orbital interactions at the top
of the valence band. Bottom right: Analogous pattern of bonding interactions at the bottom of the
valence band. After ref. [14].

32
There are two molecules per unit cell and thus two valence bands formed from the HOMO of pentacene.
 10.5 Organic Conductors 443

tetracene (C18H12) rubrene (C42H28)

Figure 10.28 The molecular structure of tetracene (left) and rubrene (right).

point in the valence band belongs to a crystal orbital with antibonding intermolecular
interactions, while the lowest-energy crystal orbital in the valence band features bonding
interactions between HOMOs on neighboring molecules. The difference in these two ener-
gies determines the width of the valence bands and is a measure of the strength of bonding
interactions between molecules. Despite the relatively narrow bandwidth, conductivities as
high as 1×104 S/m have been reported in iodine-doped pentacene films [15].
By functionalizing the conjugated core of molecules like tetracene and pentacene, it is
possible to alter the energy levels of the HOMO and LUMO, the crystal packing, and the
reactivity. Among such substituted variants, rubrene (5,6,11,12-tetraphenyltetracene, C42H28)
is one of the most extensively studied PAHs. Rubrene (Figure 10.28) is less reactive and
therefore more stable than tetracene or pentacene, which makes it possible to grow high-
purity single crystals. Rubrene holds the record mobility for an organic semiconductor, μ =
0.0040 m2/(V s). This value, while much lower than values observed in crystalline inorganic
semiconductors (Table 10.5), is approaching values seen in amorphous silicon. Unfortunately,
films of rubrene tend to be poorly crystalline. For devices that incorporate thin films of organic
conductors, pentacene is often the material of choice.

10.5.3 Charge-Transfer Salts

In the previous section we learned that conjugated polymers and polycyclic aromatic
hydrocarbons are intrinsically semiconducting. Doping is required to attain high conductiv-
ity, even then the conductivity decreases as the temperature is lowered. To find organic
conductors that behave like metals, we turn to charge-transfer salts. These compounds
contain two types of molecules whose frontier orbitals are energetically aligned so that
electron density is transferred from a donor molecule with a high-lying HOMO to an
acceptor molecule with a low-lying LUMO. The energy levels of the two molecules in the
salt are such that in the solid state the transfer is only a fraction of an electron per donor–
acceptor pair. In this way, partially filled bands can be obtained without doping.
444 Conducting Materials

To illustrate the behavior of charge-transfer salts, consider the prototypical example, TTF–
TCNQ, whose structure is shown in Figure 10.29. The donor, tetrathiofulvalene (TTF), first
synthesized in 1970, has a relatively low first ionization energy of ~6.7 eV in the gas phase [16],
due to the contributions of sulfur 3p orbitals to the delocalized π MOs (see Footnote 31 in
Chapter 12). In the acceptor, tetracyanoquinodimethane (TCNQ), electron-withdrawing cyan-
ide groups stabilize low-lying empty π* orbitals, resulting in an electron affinity that is estimated
to be ~2.8 eV in the gas phase [17].
While the energies of the frontier orbitals are such that electron transfer would not occur
between gas-phase molecules, the situation changes when the two molecules form the crystal
structure shown in the lower left-hand side of Figure 10.29. TTF and TCNQ form alternat-
ing cationic and anionic layers parallel to the bc–plane, each composed of planar molecules
that stack tightly along the b axis of the monoclinic unit cell due to dispersion forces between
their π orbitals. This widens the bands shown on the right-hand side of Figure 10.29. The two
bands that cross EF and run uphill from Γ to Z, are derived from the π-acceptor orbitals on
TCNQ. The nearly degenerate pair of bands that cross EF and run downhill from Γ to Z are
derived predominantly from 3p π orbitals of the sulfur atoms in the TTF molecule. The
overlap of these two pairs of bands is responsible for the partial electron transfer from TTF
to TCNQ. From spectroscopic signatures, such as the C–N stretching frequencies of the
TCNQ molecules, the transfer is estimated to be 0.59 e− per donor–acceptor pair.

N N
S S

S S
N N
tetracyanoquinodimethane (TCNQ) tetrathiofulvalene (TTF)

c* a* b*
0.5

0.0 EF
energy (eV)

TCNQ
−0.5
TTF

−1.0

c
−1.5
Γ B Y Γ Z
a

crystal structure of (TTF)(TCNQ) electronic structure of (TTF)(TCNQ)

Figure 10.29 Top: The molecular structures of the electron donor TTF and electron acceptor TCNQ.
Left: The crystal structure of TTF–TCNQ. Right: The band structure of TTF–TCNQ near the Fermi
level. After ref. [18].
 10.6 Carbon 445

The Fermi level cuts through the frontier orbitals of both TTF and TCNQ molecular
stacks, leading to metallic conductivity. The room-temperature conductivity is 6.5×104 S/m
[19]. Upon cooling, conductivity increases as expected for a metal, reaching a value of
1.5×106 S/m at 58 K before undergoing a metal–insulator transition upon further cooling
[20]. The bandwidth is only substantial along the b* axis of the reciprocal lattice, which
corresponds to the b axis of the real-space lattice, the direction along which the molecules
stack. This anisotropy of the electronic structure is confirmed by experimental measurements
that reveal conductivity parallel to b three orders of magnitude larger than it is in perpen-
dicular directions. Because conduction occurs predominantly along the stacks of similar
molecules, TTF–TCNQ is a quasi-1D metal.
The frontier orbitals on the donor and acceptor molecules must be appropriately aligned to
achieve partial charge transfer. For example, if the hydrogens on the central benzene ring of
TCNQ are replaced with fluorine (F4TCNQ), the acceptor becomes more electronegative and
a full electron transfer occurs, leading to an insulating behavior. Another approach to stabilize
partially filled π bands is to have one singly charged polyatomic anion, such as PF6−, AsF6−,
SbF6−, BF4−, or NO3− gaining its anion charge from two π-donor molecules. On average, two
donor molecules give up a total of one e− leading to p-doping of the donor-based valence band
by one-half hole per donor molecule. These compounds, called Bechgaard salts, are known for
their superconductivity and are discussed in more detail in Section 12.5.

10.6 Carbon
In the previous section, we explored the electronic conductivity of hydrocarbons with
networks of conjugated π bonds. In this section, we focus on two low-dimensional allotropes
of carbon with (in principle) infinitely conjugated networks: graphene and carbon nano-
tubes. We will see that the conductivity of these materials is exceptional in many ways. The
fullerenes, yet another conducting allotrope of carbon, are covered in Chapter 12, due to
their superconductivity.

10.6.1 Graphene
Graphene is an infinite 2D network of fused hexagonal rings. Each carbon atom is sp2
hybridized and forms bonds to three equidistant neighbors. As discussed in Section 6.5.1, the
electronic structure of graphene near the Fermi level is dominated by the π and π* bands that
originate from overlap of the carbon 2pz orbitals. Using a tight-binding approximation, it is
possible to derive an analytical expression for the energies of these two bands as a function of
the 2D wave vector, k [21]. A 3D contour plot of the energy throughout the first Brillouin
zone using this model is shown in Figure 10.30. Notice how the two bands are widely
separated at Γ, where the π interactions between nearest neighbors are bonding in the
lower band and antibonding in the upper band. The situation is very different at the
K and K 0 points (Figures 6.18 and 6.19), where the two bands touch. Those six points in
446 Conducting Materials

Figure 10.30 The energies of the π and π* bands in graphene as plotted in 2D reciprocal space (kx and ky in
Cartesian coordinates, for convenience, rather than the hexagonal reciprocal lattice vectors a* and b* shown
in Figure 6.13). The real-space orbital-overlap diagrams at Γ and K are taken from Figure 6.19. Image of the
2D band structure reprinted with permission from C.J.M. Beenakker, “Colloquium: Andreev reflection and
Klein tunneling in graphene” Rev. Mod. Phys. 80, 1337–1354, 2008. Copyright 2008 by the American Physical
Society.

k space where the filled and empty bands touch are called Dirac points and the surrounding
regions Dirac cones. The Fermi level lies at the precise energy where the bands touch, which
makes graphene a perfect semimetal.
Because the filled π band touches but does not overlap with the empty π* band, anything
that perturbs the position of the Fermi level will have large impact on the carrier concentra-
tion. The 2D structure of graphene makes it is possible to apply an electric field across
a graphene sheet, with the electric-field gradient perpendicular to the graphene layer. The
electric field shifts the position of the Fermi level and the charge-carrier concentration
increases as it shifts away from the Dirac points, leading to an increase in conductivity.
This process is called gating. Depending upon the sign of the applied voltage, the Fermi level
can be shifted upward, leading to conductivity dominated by electrons, or downward
whereby holes become the dominant charge carriers. In both cases, the carrier concentration
increases linearly with the voltage for modest fields. Graphene is highly sensitive to molecules
adsorbed on the surface. Exposure to electron-donating species induces n-type conduction,
whereas exposure to electron-withdrawing species leads to p-type conduction.
The semimetallic nature of graphene is not the only aspect of its electronic structure that
sets it apart from most conducting materials. In the free-electron model, the band energy
 10.6 Carbon 447

varies with square of the wave vector (E ∝ k2), as expressed in Equation (10.12). In real
materials the band structure is more complicated, but, in conventional semiconductors, the
bands are approximately parabolic (E ∝ k2) near the minimum of the conduction band and
the maximum of the valence band. In contrast, the band energies in graphene change linearly
with the wave vector k in the vicinity of the K and K 0 points where the π and π* bands touch,
as given by the relationship:

EðkÞ ¼ vF ℏk (10.27)

where vF ≈ 106 m/s is the Fermi velocity in graphene. The Dirac cones and linear relationship
between E and the wave vector k arise from the half-filling of the π bands and the honeycomb
topology of the crystal structure (see Section 6.5.1). See ref. [22] for more details.
Graphene’s electronic structure near the Fermi level is compared with a conventional semicon-
ductor in Figure 10.31. In conventional semiconductors, vF depends on the position of the Fermi
level, Equation (10.17), whereas in graphene it is independent of both EF and k. Since effective
masses are given by the curvature of the bands, Equation (10.15), the carriers in graphene behave
as though they are massless.33 Since mobility is inversely proportional to the effective mass, one
might expect charge carriers in graphene to have infinitely large mobilities.34 While this is not
observed in practice, the carrier mobilities in graphene are exceptionally high. They approach
20 m2/(V s) at low temperatures [23] and can be as high as 2.5 m2/(V s) at room temperature [24].
While mobilities this high are known for III–V semiconductors like InSb, the exceptional feature
of graphene is that its carrier mobility does not decrease as the carrier concentration increases.
The high mobility leads to mean free paths on the micron length scale in high-quality graphene,
even at room temperature! This attribute makes it possible to construct devices from graphene
capable of ballistic transport, meaning that the carriers pass from one end of the graphene sheet to
the other end without being scattered by a defect or lattice vibration. Another feature that sets the
mobilities in graphene apart is the symmetry between electron and hole mobilities; they are the
same in graphene but typically very different in conventional semiconductors.

10.6.2 Carbon Nanotubes

Carbon nanotubes are hollow cylinders of graphene sheets, typically a few nanometers in
diameter. They come in two varieties: single-walled carbon nanotubes (SWCNTs) and multi-
walled carbon nanotubes (MWCNTs). The latter are made up of cylindrical SWCNTs of
increasing diameter wrapped around each other, held together by van der Waals forces.
Because the electronic and transport properties of MWCNTs tend to be a composite of the

33
The effective mass is only zero when the Fermi level is at the energy where the bands touch. If the Fermi level is
shifted, the carriers develop a mass, but it is still quite small compared to conventional materials. For this reason, the
charge carriers in graphene are sometimes referred to as massless Dirac fermions.
34
The charge carriers in graphene obey the Dirac equation, an equation that combines quantum mechanics and
general relativity. This gives rise to interesting analogies between the charge carriers in graphene and the world of
particle physics.
448 Conducting Materials

E(k) E(k)

kx ky ky
kx

Conventional Graphene
semiconductor

• finite band gap • zero band gap

• E is proportional to k2 • E is proportional to k
• vF depends on k • vF independent of k

Figure 10.31 Differences between the linear dispersion near the Fermi level of graphene and the parabolic
dispersion seen in a conventional semiconductor.

individual SWCNT from which they are made, the treatment that follows is limited to
SWCNTs.
We will treat carbon nanotubes as 1D crystals that possess translational symmetry along
the long axis of the tube. Unlike graphene whose structure is uniquely defined, SWCNTs can
adopt an infinite number of structures, each with its own diameter and chirality. Fortunately,
there is a relatively simple nomenclature that can be used to define the structure of any
carbon nanotube in terms of graphene’s lattice vectors.
As illustrated in Figure 10.32, we begin by specifying the circumferential vector, C, which
defines the circumference of the tube by connecting two equivalent sites on a graphene
sheet.35 The circumferential vector can be expressed as a linear sum of the a and b lattice
vectors of graphene in the general form C = na + mb, where n and m are integers. The next
step is to define the translational vector, T, that is perpendicular to C and links two
equivalent sites. The rectangle formed by C and T defines the unit cell of the carbon
nanotube. To complete the process, we imagine excising a strip, of width defined by C and
infinite length, from the graphene sheet and rolling it up into a tube by connecting opposite
sides of the strip. In this manner, the structure of the carbon nanotube is completely defined
by the integers n and m, and is expressed using the notation (n,m). For example, the tube
depicted in Figure 10.32 is a (4,2) carbon nanotube.
Once the (n,m) indices of the circumferential vector are known, many useful characteristics
of the tube can be calculated. The diameter of the tube d is related to its circumference:
35
The terms circumferential vector and roll-up vector are used interchangeably in the literature.
 10.6 Carbon 449

2b
4a
C

(1) Define the circumferential vector C

(2) Define the translational vector T (T C)

roll up
T

C
(3) Cut a strip out of the graphene (4) Roll the strip up into a cylinder and
sheet that becomes the unit cell apply translational symmetry along the
tube axis (parallel to T)

Figure 10.32 A step-by-step illustration of how the (n,m) notation used to describe the structure of a carbon
nanotube is related to the lattice vectors a and b of a graphene sheet. In this example n = 4 and m = 2.

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
jCj a n2 þ nm þ m2
d¼ ¼ (10.28)
π π

where a is the graphene unit-cell edge, whose length is 2.46 Å. The so-called chiral angle, θ,
which is the angle between the lattice vector a and the circumferential vector C (see the top
image in Figure 10.32), is related to n and m through the relationship:
pffiffiffi
3m
tanθ ¼ (10.29)
2n þ m
450 Conducting Materials

The number of carbon atoms in the 1D unit cell Ncell is given by:

4ðn2 þ nm þ m2 Þ
Ncell ¼ (10.30)
Ncd

where Ncd is the greatest common divisor of the sums 2m + n and 2n + m. For the (4,2)
nanotube depicted in Figure 10.32, the chiral angle is 19.1°, and there are 56 carbon atoms in
the unit cell (Ncd = 2).
The chiral angle can take any value between 0° and 30°, but the two limiting cases are
special geometries that deserve further mention. When m = 0, the circumferential vector is
parallel to the a lattice vector of the graphene sheet and the chiral angle is zero. Such tubes
are called zigzag tubes because they exhibit a zigzag pattern of C–C bonds along the
circumference, as shown in the upper panel of Figure 10.33. When n = m, the chiral angle
attains the other limit (θ = 30°), and the tubes are called armchair tubes because they exhibit
an armchair pattern of C–C bonds along the circumference, as shown in the lower panel of
Figure 10.33. Both zigzag and armchair tubes are achiral, whereas all other SWCNTs are
chiral. As we will soon learn, the electronic and transport properties of SWCNTs depend
crucially on the chirality and diameter of the tube.
The electronic structures of SWCNTs can be derived directly from graphene, provided we
neglect the changes in hybridization that result from the slight curvature of the tube. As

b
C = 5a + 0b

C−C bonds adopt a zigzag

pattern along the tube’s
circumference
(5, 0) zigzag nanotube

T
C−C bonds adopt an armchair
C = 3a + 3b pattern along the tube’s
circumference
(3, 3) armchair nanotube

Figure 10.33 A zigzag carbon nanotube (m = 0) whose chiral angle θ = 0° (top), and an armchair carbon
nanotube (n = m) whose chiral angle θ = 30° (bottom). Note the rolled-up tubes on the right are illustrative
of each type, they do not correspond directly to the tubes shown on the left.
 10.7 Problems 451

graphene has full translational symmetry in two dimensions, the wave vector k can take any
value in the first Brillouin zone, whereas a SWCNT is finite along its circumference, which
quantizes the allowed values of k to a series of parallel lines whose spacing is inversely
proportional to the diameter of the tube [25]. When the condition kC = 2πq is met (q is an
integer), the Dirac points fall on a line of allowed k values. This condition is met in SWCNTs
where (n − m)/3 is an integer, and such nanotubes are metallic. All other SWCNTs are
semiconducting with a band gap that is inversely proportional to the diameter. Thus,
armchair nanotubes are always metallic, because n − m = 0, whereas zigzag and chiral
nanotubes can be either metallic or semiconducting, depending upon the values of n and m.
The analysis above assumes all 2p π orbitals are locally orthogonal to the network of
σ bonds. This assumption is not strictly true due to the curvature of the nanotube, and thus
each carbon atom experiences a subtle trigonal distortion from the planar geometry it adopts
in graphene. For example, the carbon atoms in a (5,5) SWCNT with a diameter of 6.8 nm are
bent ~6° from planar. The degree of such “pyramidalization” is smaller in larger tubes. This
distortion perturbs the electronic structure in a couple of important ways. Firstly, it opens
a very small band gap (Eg < 0.1 eV) in zigzag and chiral nanotubes that would otherwise be
metallic because (n − m)/3 is an integer. Interestingly, in armchair nanotubes, this distortion
does not prevent the allowed wave vectors from running through the Dirac points, and the
tubes remain metallic. The band dispersion in metallic nanotubes is nearly linear close to the
Fermi level, and the electrical transport properties along the tube axis are similar to graphene.
The ability to tune from metallic to semiconducting with a subtle change in chirality or
diameter makes SWCNTs highly unusual materials that are attractive for applications. At
the same time, it presents a significant practical barrier to use in electronic devices. Most
syntheses produce a mixture of metallic (or nearly so) SWCNTs, where (n − m) is a multiple
of three, and semiconducting SWCNTs, in roughly a 1:2 ratio. To make devices reprodu-
cibly, the SWCNTs must be separated according to their chirality, not to mention separated
from MWCNTs that are also present. Approaches to doing this include chemically digesting
the more reactive tubes (i.e. those with larger pyramidalization angles), dielectrophoretic
separation, chromatographic methods, and ultracentrifugation [26]. Finding routes to
obtain nanotubes with homogeneous properties is essential if they are to find applications
that fully exploit their electronic transport properties.

10.7 Problems
10.1 Use the conductivity of copper given in Table 10.1 to calculate the resistance of
a copper wire that is 3 mm in diameter and 20 cm long.
10.2 Show by dimensional analysis that Equation (10.9) gives the proper units (A/m2) for
the current density J. Remember 1 A = 1 C/s.
10.3 Confirm Equation (10.10) by dimensional analysis in SI units. Remember that 1 A = 1 C/s.
452 Conducting Materials

10.4 (a) Estimate the carrier mobility μ in Na metal at 300 K via Equation (10.10), using the
conductivity of Na in Table 10.2, and the Drude-model assumption that all valence
electrons can respond to an applied field. (b) Use it to calculate the relaxation time τ. (c)
What is the rms velocity of the conduction electrons at 300 K in the Drude model? (d)
Use your answers to estimate the mean free path of an electron. How does it compare
with the unit-cell parameter of 4.3 Å for Na metal? How does it compare with the
estimates of mean free path given in Section 10.2.7? How do you account for the
discrepancy between the two estimates?
10.5 The conductivities and crystal structures of the heavier s-block metals are given in the
table below. (a) Calculate the valence-electron concentration n in m−3 and arrange the
elements from lowest to highest n. (b) Use these values and those for lighter s-block
elements Na, K, Mg, and Ca in Table 10.2 to construct a plot of electrical conductivity
(in S/m) versus valence-electron concentration (in m−3). (c) What can you say about
how well these metals follow the predictions of the Drude model?

Rb Cs Sr Ba
6 6 6
Conductivity (S/m) 8.3×10 5.0×10 7.7×10 2.9×106
Crystal structure bcc bcc fcc bcc
Lattice parameter (Å) 5.58 6.14 6.08 5.03

10.6 Show that at any finite temperature the crystal orbital at the Fermi level will have
a 50% probability of being occupied.
10.7 At which energy separation from EF is the probability f(E) of an electronic state in
a metal being occupied equal to 0.01 at (a) 10 K, (b) 100 K, (c) 300 K, and (d) 1000 K.
10.8 (a) Within the assumptions of the Drude model, calculate the rms velocity of a valence
electron in sodium metal at T = 300 K. (b) Use a quantum-mechanical model to
estimate the Fermi velocity of the free carriers (assume m* = me) in Na given that EF =
3.1 eV. (c) Which velocity estimate is larger? (d) Do the two models predict different
temperature dependences of the free-carrier velocity?
10.9 What is the temperature dependence of the conductivity for (a) a metal, (b) an intrinsic
semiconductor?
10.10 Compare electrical properties of copper and nickel. Which has higher carrier concen-
tration? Which has higher carrier mobility? Which has higher conductivity and why?
10.11 Calculate the value of f(E) at the conduction-band edge in intrinsic Ge (Eg = 0.7 eV) at
300 K. Assume the Fermi level is located midway between the valence- and conduc-
tion-band edges.
10.12 Calculate the donor ionization energy and Bohr radius for a donor impurity in InP
(the dielectric permittivity εr = 12.1, effective mass m* = 0.07me).
10.13 Will the temperature range over which a semiconductor that is in the saturation
regime increase or decrease in response to the following changes: (a) the band gap
 10.7 Problems 453

of the semiconductor increases, (b) the effective mass of the carriers increases, (c) the
dielectric constant of the semiconductor increases?
10.14 In compounds containing lanthanoid ions, the 4f orbitals are sufficiently localized
that interactions with orbitals on neighboring atoms can be neglected, hence W ≈ 0. In
this limit, the Hubbard U can be experimentally determined from photoelectron
spectra. (a) Use ionization energies for gas-phase ions to estimate U for Pr3+ and
Gd3+ (the third and fourth ionization energies of Pr are 21.6 eV and 39.0 eV,
respectively, those for Gd are 20.6 eV and 44.0 eV). (b) Measured values of U for
these two ions in solids are estimated as 6 eV for Pr3+ and 12 eV for Gd3+. Why are the
estimated values from part (a) higher than the measured values? (c) Why is U so much
higher for Gd3+ than it is for Pr3+? (d) By analogy, would you expect a higher U for
TiO or MnO?
10.15 The perovskite NaOsO3 undergoes a transition from a paramagnetic metal to an
antiferromagnetic semiconductor upon cooling below 410 K. (a) What is the electron
configuration of osmium in this compound. (b) Most perovskites containing ions with
partially filled 5d orbitals are metals at all temperatures, what feature of NaOsO3
makes it different?
10.16 Each of the following compounds contains an octahedrally coordinated transition-
metal ion and behaves as an insulator/semiconductor. Identify each as either a band
insulator or a Mott–Hubbard insulator: (a) Fe2O3, (b) MgCr2O4, (c) Li2PtO3.
10.17 The images below show a portion of a chain of the conducting polymer polypyrrole,
before (top) and after (bottom) oxidative doping. (a) Is the charge carrier formed in
this process a polaron, bipolaron, or soliton? (b) Is the conducing polymer that results
doped n- or p-type?

S S S
S S S

oxidation

S S S
S S S
+

10.18 Consider the electronic structure of graphene. (a) At how many points in the first
Brillouin zone do the conduction and valence bands of graphene meet? (b) What is the
name given to those points? (c) The bands that meet are derived from which atomic
orbitals on carbon: 2s, 2px, 2py, and/or 2pz?
10.19 Name at least three ways in which the electronic structure and transport properties of
graphene differ from a conventional semiconductor.
454 Conducting Materials

10.20 What percentage of single-walled carbon nanotubes in a typical synthesis batch will be
semiconducting? Explain your answer.
10.21 Consider the circumferential and translational vectors of a potential carbon nanotube
shown below. (a) What are the n and m values of the circumferential vector C? (b) Is
the T vector shown below a valid translational vector?

T
C

10.22 Consider the unit of a carbon nanotube projected onto a graphene sheet shown
below. (a) What are the n and m values of this carbon nanotube? (b) Is this
a metallic or semiconducting nanotube? (c) What is the diameter and chiral angle
of this nanotube? (d) How many atoms are contained in the unit cell?

10.8 Further Reading

P.A. Cox, “The Electronic Structure and Chemistry of Solids” (1987) Oxford University Press.
S. Elliot, “The Physics and Chemistry of Solids” (1998) Wiley.
D.I. Khomskii, “Transition Metal Compounds” (2014) Cambridge University Press.
A.J. Heeger, “Semiconducting polymers: The third generation” Chem. Soc. Rev. 39 (2010), 2354–2371.
M. Ouyang, J.L. Huang, C.M. Lieber, “Fundamental electronic properties and applications of
single-walled carbon nanotubes” Acc. Chem. Res. 35 (2002), 1018–1025.
R. Saito, G. Dresselhaus, M.S. Dresselhaus, “Physical Properties of Carbon Nanotubes” (1998)
Imperial College Press.
A.H. Casto Neto, F. Guinea, N.M.R. Peres, K.S. Novoselov, A.K. Geim, “The electronic properties of
graphene” Rev. Mod. Phys. 81 (2009), 110–155.
 10.9 References 455

10.9 References
[1] “Electrical resistivity of pure metals,” in “CRC Handbook of Chemistry and Physics” 98th edition,
Editor: J.R. Rumble (Internet Version 2018) CRC Press/Taylor & Francis.
[2] A. Yamasaki, T. Fujiwara, “Electronic structure of the MO oxides (M = Mg, Ca, Ti, V) in the
GW approximation” Phys. Rev. B 66 (2002), 245108.
[3] F. Rivadulla, J. Fernández-Rossier, M. García-Hernández, M.A. López-Quintela, J. Rivas,
J.B. Goodenough, “VO: A strongly correlated metal close to a Mott–Hubbard transition”
Phys. Rev. B 76 (2007), 205110.
[4] A. Herwadkar, W.R.L. Lambrecht, “Electronic structure of CrN: A borderline Mott insulator”
Phys. Rev. B 79 (2009), 035125.
[5] R.M. Ibberson, R. Cywinski, “The magnetic and structural transitions in CrN and (Cr, Mo)N”
Physica B 180 (1992), 329–332.
[6] K. Suzuki, Y. Yamaguchi, T. Kaneko, H. Yoshida, Y. Obi, H. Fujimori, H. Morita, “Neutron
diffraction studies of the compounds MnN and FeN” J. Phys. Soc. Japan 70 (2001), 1084–1089.
[7] M.A. Señarís-Rodríguez, J.B. Goodenough, “LaCoO3 revisited” J. Solid State Chem. 116 (1995),
224–231.
[8] M.W. Haverkort, Z. Hu, J.C. Cezar, T. Burnus, H. Hartmann, M. Reuther, C. Zobel, T. Lorenz,
A. Tanaka, N.B. Brookes, H.H. Hsieh, H.J. Lin, C.T. Chen, L.H. Tjeng, “Spin state transition in
LaCoO3 studied using soft X-ray absorption spectroscopy and magnetic circular dichroism”
Phys. Rev. Lett. 97 (2006), 176405.
[9] G. Catalan, “Progress in perovskite nickelate research” Phase Transitions 81 (2008), 729–748.
[10] A.J. Heeger, “Semiconducting polymers: The third generation” Chem. Soc. Rev. 39 (2010),
2354–2371.
[11] H. Shirakawa, E.J. Louis, A.G. MacDiarmid, C.K. Chiang, A.J. Heeger, “Synthesis of electric-
ally conducting polymers: Halogen derivatives of polyacetylene (CH)x” J. Chem. Soc., Chem.
Commun. (1977), 578–580.
[12] M. Bendikov, F. Wudl, D.F. Perepichka, “Tetrathiafulvalenes, oligoacenenes, and their buck-
minsterfullerene derivatives: The brick and mortar of organic electronics” Chem. Rev. 104 (2004),
4891–4945.
[13] J.E. Anthony, “The larger acenes: Versatile organic semiconductors” Angew. Chem. Int. Ed. Engl.
47 (2008), 452–483.
[14] G.A. de Wijs, C.C. Mattheus, R.A. de Groot, T.T.M. Palstra, “Anisotropy of the mobility of
pentacene from frustration” Synth. Met. 139 (2003), 109–114.
[15] T. Minakata, I. Nagoya, M. Ozaki, “Highly ordered and conducting thin film of pentacene doped
with iodine vapor” J. Appl. Phys. 69 (1991), 7354–7356.
[16] D.L. Lichtenberger, R.L. Johnston, K. Hinkelmann, T. Suzuki, F. Wudl, “Relative electron
donor strengths of tetrathiafulvene derivatives: Effects of chemical substitutions and the molecu-
lar environment from a combined photoelectron and electrochemical study” J. Am. Chem. Soc.
112 (1990), 3302–3307.
[17] R.N. Compton, C.D. Cooper, “Negative ion properties of tetracyanoquinodimethane: Electron
affinity and compound states” J. Chem. Phys. 66 (1977), 4325–4329.
[18] M. Sing, U. Schwingenschlögl, R. Claessen, P. Blaha, J.M.P. Carmelo, L.M. Martelo,
P.D. Sacramento, M. Dressel, C.S. Jacobsen, “Electronic structure of the quasi-one-
dimensional organic conductor TTF–TCNQ” Phys. Rev. B 68 (2003), 125111.
456 Conducting Materials

[19] J. Ferraris, D.O. Cowan, V. Walatka, J.H. Perlstein, “Electron transfer in a new highly conduct-
ive donor–acceptor complex” J. Am. Chem. Soc. 95 (1973), 948–949.
[20] M.J. Cohen, L.B. Coleman, A.F. Garito, A.J. Heeger, “Electrical conductivity of tethrathioful-
valinium tetracyanoquinodimethane (TTF)(TCNQ)” Phys. Rev. B 10 (1974), 1298–1307.
[21] A.H. Casto Neto, F. Guinea, N.M.R. Peres, K.S. Novoselov, A.K. Geim, “The electronic
properties of graphene” Rev. Mod. Phys. 81 (2009), 110–155.
[22] A. Jorio, R. Saito, G. Dresselhaus, M.S. Dresselhaus, Chapter 2 in “Raman Spectroscopy in
Graphene Related Systems” (2011) Wiley-VCH.
[23] K.I. Bolotin, K.J. Sikes, Z. Jiang, M. Klima, G. Fudenberg, J. Hone, P. Kim, H.L. Stormer,
“Ultrahigh electron mobility in suspended graphene” Solid State Commun. 146 (2008), 351–355.
[24] M. Chhowalla, D. Jena, H. Zhang, “Two-dimensional semiconductors for transistors” Nat. Rev.
Mater. 1 (2016), 1–15.
[25] M. Ouyang, J.L. Huang, C.M. Lieber, “Fundamental electronic properties and applications of
single-walled carbon nanotubes” Acc. Chem. Res. 35 (2002), 1018–1025.
[26] M.C. Hersam, “Progress towards monodisperse single walled carbon nanotubes” Nat.
Nanotechnol. 3 (2008), 387–394.
11 Magnetotransport Materials

11.1 Magnetotransport and Its Applications

While present-day data processing is based on binary digits of electric charge being on and
off, a future spin-electronics paradigm may utilize not only the charge but also electron spin.
Materials that only carry electrons with spin of one direction (spin-polarized current) would
be essential ingredients of spintronic devices. The quest for materials in which magnetization
and magnetic order is correlated with spin-polarized conductivity has therefore been one of
the research motors in solid state chemistry.
Magnetoresistance, the change in resistivity upon applying an external magnetic field, has
already been widely adopted in applications. Computer hard disks have read/write heads
made of materials that rely on this property. Magnetoresistance is normally a weak effect
occurring due to the distribution of electron velocities. However, if a junction of two
conductors with opposite spins is made, one creates a magnetoresistant device called a spin
valve (Figure 11.1). One of the materials is chosen so that it can easily flip its magnetization in
an external magnetic field. Without a field, the assembly is insulating because the conduction
electrons in the two materials are oppositely polarized. Under a field, their spins align
parallel, allowing the spin-polarized portion of the current through. If the magnetic field
stems from the pattern of magnetized particles on a spinning hard-disk platter, you have
converted the recorded information into a time-resolved pattern of electric current. In some
materials, the spin flip needed can be induced by a sufficiently large pulse of spin-polarized
current. This is termed spin injection and represents a first step towards the holy grail of
spintronics; the spin transistor.
Materials with spin-polarized itinerant electrons are closely associated with ferromagnet-
ism. However, the degree of spin polarization in ferromagnetic metals is not high: Jullière [1]
studied the change in conductance across two ferromagnetic films, first oriented with parallel
and then with antiparallel magnetizations (upon tunneling through an intermediate 100 Å
sheet of semiconducting Ge). The maximum relative change in conductance1 G (=1/R)

1
A conductance (symbol G) in siemens (S) is the inverse of resistance (symbol R) in ohms (Ω).

457
458 Magnetotransport Materials

Figure 11.1 Principle of the spin valve with two conductors of strictly opposite spin orientations and an
external field that can switch the orientation parallel.

between these orientations, (Gpar − Gapar)/Gpar, was a mere 0.14. This gives the spin-polar-
ization fraction P = 0.27 with the formula Jullière used,

Rapar  Rpar Gpar  Gapar 2P2

magnetoresistance ≡ ¼ ¼ (11.1)
Rapar Gpar 1 þ P2

of spin polarization P defined as

n↑ ðEF Þ  n↓ ðEF Þ
P¼ (11.2)
n↑ ðEF Þ þ n↓ ðEF Þ

where n↑ ðEF Þ and n↓(EF) are the densities of states at the Fermi level for the respective spin-
up and spin-down tunneling electrons. In Section 11.4 we return to this in more detail.
Higher degrees of spin polarization may be obtained from itinerant valence-mixing electrons
of ferro- or ferrimagnetics such as magnetite Fe3O4.

11.2 Charge, Orbital, and Spin Ordering in Iron Oxides

11.2.1 The Verwey Transition in Magnetite, Fe3O4
Magnetite, Fe3O4, is the oldest known magnetic material, with a long history of use in
compasses. It is also one of the earliest examples of a mixed-valence [2] oxide. Fe3O4 adopts
the inverse spinel structure (more details in Section 1.5.1), [Fe3+]tet[Fe2+Fe3+]octO4, whereby
a mixture of high-spin Fe2+ and Fe3+ is accommodated on octahedral cation sites.
Spectroscopic measurements at ambient conditions show that all iron atoms on the octahe-
dral sites are identical. This is reminiscent of aromatic conjugation in benzene where
delocalization of six π electrons over six carbons leads to six equivalent carbon-to-carbon
bonds of bond order 1.5. In magnetite, each minority-spin electron of Fe2+ is delocalized
 11.2 Charge, Orbital, and Spin Ordering in Iron Oxides 459

10
[Fe ] tet [Fe
3+ 2+
Fe3+ ] oct O 4

resistivity (Ω m)
1

0.1

0.01

0.001 easy hopping

0.0001

100 150 200 T (K) 250 300

Figure 11.2 Electrical resistivity across the Verwey transition in an Fe3O4 single crystal, plotted from data
in [4], and extrapolation of the low-temperature trend of relatively easy hopping from Fe2+ to Fe3+.

over the octahedral sites. Just as the σ electrons in benzene remain localized, so do the five
majority-spin electrons per Fe in Fe3O4. Here the analogy ends, though. While conjugation
stabilizes benzene by some −150 kJ/mol (released energy), the delocalization in magnetite is
weakly endothermic, it consumes about 1 kJ per mole Fe of heat that feeds the increased
entropy of the delocalization product when the thermal energy becomes high enough upon
warming; just like melting of ice. As a consequence, the delocalized electron localizes upon
cooling, and the high-temperature “valence-mixed” state freezes into a low-temperature
“charge-ordered” state.
The transition into the valence-mixed2 state in magnetite is called the Verwey transition,
after Evert Verwey, who first reported [3] a spectacular drop in electrical resistivity upon
heating through the T = 125 K transition temperature (Figure 11.2). Magnetite is a good
electrical conductor because of its magnetism: The unpaired spins of iron at all octahedral
sites are parallel as they point in the opposite direction to the tetrahedral spins, forming a
ferrimagnet (Section 9.10). The minority-spin electron of Fe2+ can easily hop from one
octahedral iron to the other, with no need to flip its spin. On top of this situation, the
resistivity suddenly drops upon heating through 125 K when the valence-mixing delocaliza-
tion sets in.
The valence-mixing electrons are all of the minority spin (Figure 11.2), and this makes the
magnetite crystal a source of spin-polarized current; an electrical conductor where the
majority of carriers are electrons of one spin only. The fact that magnetite is a

2
For clarity, we shall use “valence mixed” when the “valences” of two oxidation states are thermally mixed via sharing
electrons. The low-temperature situation is called “charge ordered”. Note that the term “mixed valence” introduced
by Robin and Day [2] is general and covers both cases.
460 Magnetotransport Materials

Y Y

308 K
Ba Ba

Fe 2+ Fe 3+ Fe 2.5+ Fe 2.5+
z
y
x

Figure 11.3 Charge-ordered (left) and valence-mixed (right) structures YBaFe2O5.

compositionally simple oxide easily grown as single crystals and epitaxial3 layers, with a very
high magnetic ordering temperature (TC = 858 K), makes it an attractive candidate for
spintronic applications. Since its crystal structure becomes remarkably complex in the low-
temperature charge-ordered state [5], we’ll use a perovskite derivative to illustrate a simple
valence-mixing process.

11.2.2 Double-Cell Perovskite, YBaFe2O5

The crystal structure of YBaFe2O5 [6] is derived from the perovskite aristotype ABO3
by ordering two types of A atoms, Y and Ba, and removing one of every six oxygens
(Figure 11.3). The size difference of Y3+ and Ba2+ drives their ordering into layers, doubling
one of the unit-cell edges. By confining the oxygen vacancies to the yttrium layer, Y3+
becomes eight-coordinate, while the larger Ba2+ retains a twelve-coordinate environment.
More importantly, all iron coordinations become square pyramidal, and the oxidation state
of Fe is +2.5 on average.
The charge-ordered YBaFe2O5 (Figure 11.3, left) undergoes a thermally induced phase
transition where (most of) the Fe2+ and Fe3+ pairs start sharing the minority-spin Fe2+
electron upon formation of valence-mixed Fe2.5+ (Figure 11.3, right). The two phases coexist
while the latent heat is supplied (or removed), like ice and water, marking a first-order
transition. The resistivity drops by two orders of magnitude upon heating through the
transition, as with magnetite.
The high-temperature structure has all Fe-coordination square pyramids identical, even
locally in both space and time. At low temperatures (Figure 11.3, left), chains of Fe3+ and
chains of Fe2+ pyramids pack into the checkerboard arrangement shown. This is surprising
because the coulombic minimum for point charges would entail Fe3+ and Fe2+ alternating in
all three directions, not just in two. Why does YBaFe2O5 adopt a structure that is not
predicted from electrostatic interactions?

3
An MgO substrate with unit-cell parameter a = 4.21 Å is about a half of a = 8.40 Å for magnetite.
 11.2 Charge, Orbital, and Spin Ordering in Iron Oxides 461

Y x2–y2
x2–y2
Fe 2+

Ba z2 z2

Fe 3+
xy xy
yz xz
xz yz
z dxz
y Fe2+ Fe3+
x
1.96Å 1.98Å

a/2 2.05Å 1.89Å

expansion along a 1.96Å 1.98Å
(versus dashed square)

Figure 11.4 Orbital ordering of dxz in YBaFe2O5 at low temperatures (left). Fe2+ and Fe3+ coordinations
with distances and energies of their d orbitals (right).

Such a violation is indicative of ordering of spatial charges, entire orbitals. Expansion

along x and contraction along y of the Fe2+ square pyramid identify dxz as the ordered
orbitals, as illustrated in Figure 11.4, top left. Such a preferential orientation of an orbital
throughout the periodic structure is called orbital ordering. As the Fe2+ dxz orbital is doubly
occupied (Figure 11.4, right), the uniaxial expansion is a Jahn–Teller distortion. Because
these dxz orbitals are ordered in the periodic structure, we speak of a cooperative Jahn–Teller
distortion.
Below TN of about 430 K, the valence-mixed YBaFe2O5 becomes an antiferromagnet. The
antiferromagnetic coupling occurs in all three directions of each double-pyramidal slab of
Figure 11.3, right. Only Fe atoms of two different slabs, which face one another across the Y
layer, couple ferromagnetically in the valence-mixed state. This parallel alignment of spins in
these Fe pairs facilitates the sharing of the minority-spin electron. Upon cooling into the
charge-ordered state, the minority spin localizes, these Fe pairs begin to couple antiferro-
magnetically, and the Fe2+ and Fe3+ spins alternate in all three directions (the G-type order
illustrated in Figure 9.19).
Mössbauer spectroscopy is essential in studies of the local distribution of the iron valence and
spin states. Analysis of YBaFe2O5 spectra reveals the evolution of the minority-spin electron’s
orbital occupancies from the doubly occupied dxz in the charge-ordered state at low temperat-
ures to shared occupancy at high temperatures by two dz2 orbitals of the thus faintly attracted
valence-mixing Fe atoms that face each other across the Y layer [7]. The valence mixing
preserves the t2g degeneracy to a large extent and occurs when the orbital-energy difference
462 Magnetotransport Materials

between the two sites in the charge-ordered state is small enough to be overcome by thermal
energy.4

11.2.3 CaFeO3 and SrFeO3

The high-spin iron d 4 of the CaFeO3 and SrFeO3 perovskites contains one electron per two
eg orbitals in the octahedral ligand field. The system has three ways in which to respond to the
need to lift the orbital degeneracy: One is Jahn–Teller distortion that splits the eg levels by
c-axial expansion of the Fe-coordination octahedron (Section 5.3.10). The second is dispro-
portionation of d 4 into the more stable d 3 and d 5 high-spin half-filled orbital levels of FeV
and FeIII. The third is to make the single eg electron itinerant by mixing with oxygen 2p
bands. The latter two are found in CaFeO3 and SrFeO3, respectively.
CaFeO3 has FeIV that disproportionates into FeV and FeIII on cooling through 290 K
(Figure 11.5). SrFeO3 is metallic down to at least 4 K. The structural difference is that the
relatively large Sr2+ makes SrFeO3 a cubic perovskite, whereas in CaFeO3 the octahedra tilt,
lowering the symmetry. The subtle competition between electron delocalization and dispro-
portionation depends upon the width of the conduction band. As discussed in Section 6.6.5,
linear M–O–M bonds (as in cubic SrFeO3) lead to a relatively wide conduction band. This
favors delocalization of the eg electrons. Bending the M–O–M bonds due to tilts in CaFeO3
narrows the conduction band, destabilizing the metallic state and favoring disproportion-
ation into a charge-ordered semiconductor5 below 290 K.

290 K
Fe3+

Ca Ca

Fe5+ z Fe4+
y
x

Figure 11.5 Charge disproportionation in CaFeO3.

4
In most general terms, such a difference will not be high when the orbitals about to share the mixing electron are
weakly antibonding (practically nonbonding). Localization of the electron will cause only a small expansion of
bonds, a small structural distortion; a shallow potential well that will not trap the electron unless the temperature is
very low. A strong antibonding character would cause a larger lattice distortion, creating a deep potential well that
could trap the electron even at high temperatures.
5
Flat (narrow) bands have a small curvature, small second derivative, and this leads to a large effective mass (Section
10.2.5) that favors localization.
 11.2 Charge, Orbital, and Spin Ordering in Iron Oxides 463

Box 11.1 Characterizations: Mössbauer spectroscopy

While resonant X-ray absorption creates excited states in the electron shell of atoms, resonant γ-
ray absorption excites the nucleus. The complication is that the γ photon has such a high energy
that its momentum recoils the atom it hits in a solution or gas. The energy loss then prevents
resonant fluorescence by that atom; a radiative emission that would carry meaningful informa-
tion. Rudolf Mössbauer realized that atoms in solids might allow absorption of the γ photon
without recoil if the photon momentum brings less energy than that of a vibronic quantum.* The
fraction of absorption events that proceeds in such a manner is called the recoilless fraction.
Only relatively low-energy γ radiation gives a high recoilless fraction. The most popular is the
radiation associated with the transition of the metastable excited state I = 3/2 (of the nuclear-
spin angular momentum I; mean lifetime 0.14 μs) to the stable state I = 1/2, in the 57Fe isotope
(about 2% of natural iron). Radioactive 57Co is used as the γ source (the half-life is 272 days).
About 90% of its decay path goes via the excited I = 3/2 state of 57Fe that then de-excites via two
main paths: In 89% of cases, it ejects an electron. In 11% of cases, it emits a γ photon of energy
14.412 keV. This photon is then resonantly absorbed in 57Fe of the sample. The energy peak of
this photon is extremely sharp; 14412 eV high yet only 5×10−9 eV wide at half its height. This
extreme sharpness allows tuning the incoming γ energy over the narrow span of the γ-absorp-
tion peaks by the Doppler effect. All that is needed is to mechanically oscillate the source back
and forth relative to the flat transmission sample. As an example, a velocity v = 1 mm/s changes
the γ-photon energy by a fraction v/c, hence by 48×10−9 eV; about ten times the peak’s natural
width. With this resolution, the so-called hyperfine structure of the resonance absorption
appears as shifts and splitting of absorption energies, as illustrated in the figure below. A
disadvantage is that only samples containing 57Fe, 119Sn, 151Eu, and 197Au do not have to be
cooled down in order to limit the recoil. The ultimate disadvantage is that some elements have
no Mössbauer-suitable isotopes.

Hyperfine interactions: quadrupole

isomer shift magnetic splitting
splitting
+3/2
I = 3/2 57Fe absorber ±3/2 +1/2
±1/2 −1/2
nuclear −3/2
levels mI
+1/2
I = 1/2 ±1/2 −1/2
DmI = 0, ±1
0

Spectra:

velocity (mm/s)
electric-field
electron density magnetic field
gradient
464 Magnetotransport Materials

Box 11.1 (cont.)

The hyperfine features are isomer shift, quadrupole splitting, and magnetic splitting. The
isomer shift reflects the chemical environment around the 57Fe nucleus—oxidation state,
covalency, coordination number—anything that affects the electron density of the nearby
3s and 4s electrons (only s electrons have non-zero probability at the nucleus). The
quadrupole splitting is caused by the electric-field gradient that the 57Fe nucleus experi-
ences as a result of its own valence electrons and the surrounding ligands. It tells us
something about the symmetry of its coordination environment. The magnetic splitting is
caused by cooperative magnetism as seen at the nucleus. It is very sensitive to both local
and long-range magnetic ordering. The isomer shift, the “electric” quadrupole splitting,
and the magnetic splitting build on each other, and the mutual orientation of the electric
and magnetic fields affects the shape of the sextet of peaks. The 57Fe Mössbauer spectrum
therefore provides an account of all local valence, coordination, and spin states of iron in
the sample.
Two additional techniques take a different approach. One is conversion-electron
Mössbauer spectroscopy, a reflection technique for thin-film samples (~100 nm) or
very thick samples that can’t be thinned. It utilizes the 89% probability of the sample’s
excited 57Fe state not emitting a γ photon but ejecting an electron. The ejected electrons
are detected in backscattering mode, together with secondary electrons emitted by the
X-ray radiation released when higher-lying electrons fill the holes vacated by the ejected
electrons. Simplified, the spectrum is the inverse of the absorption spectrum described
above, but probes only a thin surface layer of the measured sample. The second method
utilizes synchrotron γ radiation (6–30 keV) with high-intensity pulses. After exciting
the 57Fe atoms in the sample by an extremely short γ pulse (0.1 ns), a coherent
(fluorescent) γ re-emission upon de-excitation of the atoms occurs in the forward
direction (while the incoherent portion emits in all directions). It is detected as a
function of time up to some hundreds of nanoseconds, yielding an emission spectrum
that looks unlike any classical Mössbauer result yet still contains information about the
hyperfine parameters. The overall intensity of the forward emission is a simple function
of the actual recoil-free absorption fraction, and thus provides additional information
about the sample.

* The de Broglie relation for photon momentum p = h/λ combines with E = hc/λ to p = E/c for the γ photon of
E = 14.4 keV in 57Fe Mössbauer experiments. Since linear momentum p = mv (mass times velocity), v = p/m, and the
kinetic energy the 57Fe atom may receive is ½mv2 = 3.13×10−22 J, where m is the atom’s mass. The atom’s vibronic
quantum is hν (ν is the vibration frequency of 57Fe in the solid), and a rough estimate of ν = 1013 s−1 (Chapter 3.3.1) yields
hν = 66×10−22 J; ~20 times larger than the 3.13×10−22 J brought to 57Fe by the γ photon. As a consequence, most of the
photons can be absorbed without exciting vibrational modes in the crystal. At room temperature, 91% of the interactions
between 57Fe and γ photons are recoil-free.
 11.3 Charge and Orbital Ordering in Perovskite-Type Manganites 465

11.3 Charge and Orbital Ordering in Perovskite-Type Manganites

The instability of the d 4 configuration encountered in the previous section on CaFeO3 and
SrFeO3 is an important ingredient of magnetotransport in the perovskite oxides containing
trivalent and tetravalent manganese. Before we proceed by way of example to the series
La1−xCaxMnO3, let’s have a look at the pure parent phases.

11.3.1 Spin and Orbital Ordering in CaMnO3 and LaMnO3

CaMnO3 is isotypical but not isoelectronic with CaFeO3. The nearly linear d 3–d 3 super-
exchange interaction in CaMnO3 is mediated by an oxygen-anion p orbital linking the empty
eg orbitals of neighboring Mn4+ cations. The top of Figure 9.20 shows that this dictates
alternating electron spins across the oxygen links. All superexchange interactions are anti-
ferromagnetic, and the spin order is therefore G-type (see Figure 9.19, top left). The
degenerate half-filled d 3 shell of pseudospherical symmetry does not permit any charge or
orbital ordering in stoichiometric CaMnO3.
Stoichiometric LaMnO3 and CaFeO3 are isotypical and their transition-metal ions are
isoelectronic, yet the two compounds find different solutions to the d 4 instability. Mn3+ has a
lower electronegativity than Fe4+, hence reduced covalency that narrows the Mn(eg)–O(2p)
conduction band and prevents itinerancy of the single eg electron. Instead, the coordination
octahedron removes the eg degeneracy by uniaxial expansion (Jahn–Teller distortion) to a
4.36 Å apex-to-apex distance (versus 3.81 Å and 3.92 Å for the other two distances) [8]. This
lowers the dz2 orbital energy relative to dx2–y2 and accommodates the single electron in dz2.
The dz2 orbitals follow the elongated axis of each tilted octahedron, an orbital ordering, see
Figure 11.6.

Along b: dx2−y2 and p are non-orthogonal (overlap) → AFM

La
O dx2−y2
O p orbital

dz2

t2g
c Mn d4 orbitals Mn d4 orbitals

a Within ac: dx2−y2 and p are orthogonal (no overlap) → FM

Figure 11.6 LaMnO3: Ordering of half-filled Mn dz2 orbitals (dashed, shape simplified), via uniaxial
Jahn–Teller distortion of Mn coordination octahedra (Pnma, neutron diffraction at 9 K [8]), yields
ferromagnetism (FM) in the ac plane and antiferromagnetic coupling (AFM) along b. See text for
description of superexchange interactions.
466 Magnetotransport Materials

Box 11.2 Synthetic Methods: Valence control via O2 partial pressures

LaMnO3+δ has a relatively wide oxygen nonstoichiometry range of about −0.1 < δ < 0.3. Upon
partial substitution of La3+ with suitable alkaline-earth cations, the range remains wide but
shifts to lower values. How can one synthesize a material with a specific desired mixed valence
of Mn?
From Chapter 3, we know that equilibrium is established between point defects of the
oxide and O2(g) at high temperatures. Let’s start with LaMnO3 and let’s say we are not
aware of the actual type of intrinsic defects in it, so we simply consider a plain oxygen deficit
and excess. This defines anion-Frenkel defects (Figure 3.1) that act as compensators of the Mn
redox equilibria we wish to control. Following Table 3.2, we write equations for oxidation,
O2(g) = 2Oi 0 0 + 4h•; reduction, 2OO˟ = 2vO•• + 4e 0 + O2(g); the intrinsic anion-Frenkel defect
formation, OO˟ = vO•• + Oi 0 0 ; and the intrinsic ionization, nil = e 0 + h•. The respective mass-action
terms are: Kox = [Oi 0 0 ]2[h• ] 4·pO−1
2
, Kred = [vO•• ]2 [e 0 ]4·pO2, KF = [Oi 0 0 ][vO•• ], and Ki = [e 0 ][h•], where
square brackets mean molar fractions. Let’s now acceptor dope LaMnO3 into La1−xSrxMnO3.
The electroneutrality condition (Section 3.1.5) states 2[Oi 0 0 ] + [e 0 ] + [SrLa 0 ] = [h• ] + 2[vO•• ]. From
this, with the help of the mass-action terms, one evaluates the f([Oi 0 0 ]) = 0 and f ([vO•• ]) = 0
polynomial functions for the oxygen defects, with pO2 as the independent variable and three of
the four equilibrium constants as parameters. Each of the equilibrium constants, K, is associ-
ated with a standard reaction-enthalpy and -entropy change ΔH and ΔS, as K = exp[(ΔS/R) −
(ΔH/RT)]. This determines the temperature dependence. The La1−xSrxMnO3+δ oxygen non-
stoichiometry, δ = [Oi 0 0 ] − [vO••], is least-squares fitted to the δ = f(pO2, T) data obtained by
experiment [9] and plotted over a range of temperatures for a given x:

intrinsic Frenkel disorder

La0.8Sr0.2MnO3+δ

0.0
δ in

-0.1

-0.2

-0.3
-10 log pO2 (bar) 0

We see a plateau of oxygen contents at δ = 0, referring to the integer (non-vacant) structure.

No plateau is observed at the Mn integer-valence point (δ = −0.1, when the substitution is
compensated by oxygen vacancies). This implies that electronic defects strongly outnumber
anion-Frenkel defects (Ki > KF more than Ki > KS in Figure 3.4).
Say, we now realize from the literature [10, 11] that the intrinsic defects in LaMnO3+δ
actually are of the Schottky type. We can approach it as an M2O3+δ oxide (M = La, Mn), in
 11.3 Charge and Orbital Ordering in Perovskite-Type Manganites 467

Box 11.2 (cont.)

which metal and oxygen vacancies are the intrinsic defects. The new equation for oxidation is
O2(g) = (4/3)vM 0 0 0 + 4h• + 2OO˟, the Schottky-defect formation nil = 2vM 0 0 0 + 3vO••, the other two
equations remain the same. The corresponding new mass-action terms are: Kox = [vM 0 0 0 ]4/3
[h•] 4·pO2−1 and KS = [vM 0 0 0 ]2[vO•• ]3. The electroneutrality condition for the acceptor-doped phase
La1−xSrxMnO3 is 3[vM 0 0 0 ] + [e 0 ] + [SrLa 0 ] = [h• ] + 2[vO•• ]. From it, with the help of the mass-
action terms, the f([vM 0 0 0 ]) = 0 and f([vO•• ]) = 0 polynomial functions are evaluated, and the
oxygen nonstoichiometry δ = (3/2)[vM 0 0 0 ] − [vO•• ] is fitted to the experimental δ = f(pO2, T) data;
here again for La0.8Sr0.2MnO3+δ:

intrinsic Schottky disorder

La0.8Sr0.2MnO3+δ

0.0
δ in

-0.1

-0.2

-0.3
-10 log pO2 (bar) 0

Because both Frenkel and Schottky defects involve oxygen vacancies, the reductive non-
stoichiometry is the same in both plots, only the oxidative parts differ slightly. The message for
La0.8Sr0.2MnO3+δ synthesis is the same. Despite Sr doping and high temperatures both
promoting O2 release, some extra oxygen (δ > 0) is still accommodated at all feasible temperat-
ures under syntheses in air where log (pO2/bar) = −0.71. Graphs like these help finding the right
pO2 and temperature to equilibrate the product before it is rapidly quenched to obtain the
desired oxidation state of Mn.

Ordering of the half-filled dz2 orbitals dictates the orientation of the empty dx2−y2 orbitals that
participate in superexchange interactions via oxygen atoms. Along the b axis, we get antiferro-
magnetic superexchange just as we did in all three directions in CaMnO3. In the ac plane, the p
orbital of each O overlaps on one side with the empty spin-up sub-band dx2−y2 of Mn; on the other
side it points against the half-occupied dz2, hence overlaps with its spin-down sub-band, making
the ac-plane spins parallel (the coupling is analogous to the d 3–d 5 case shown in Figure 9.20).
The bulk magnetic order is antiferromagnetic (TN = 140 K), of the type A shown in Figure 9.19,
top right.
468 Magnetotransport Materials

11.3.2 The La1−xCaxMnO3 Phase Diagram

Although the most interesting magnetoresistance properties of La1−xCaxMnO3 occur in a
limited range of compositions around x ≈ ⅓, their understanding is assisted by considering
the wider structural and magnetic phase diagram of this solid solution. Let’s recollect that
the magnetic order of LaMnO3 is antiferromagnetic A type whereas CaMnO3 is G type. At
high temperatures, solid LaMnO3 and CaMnO3 are completely miscible, paramagnetic, with
Mn3+ and Mn4+ charges mixed. The phase diagram in Figure 11.7 shows that at the
La0.5Ca0.5MnO3 composition, a 1:1 ordering of the Mn charges occurs below the critical
charge-ordering temperature, Tco. Simultaneously, the conflict of differing magnetic orders
unravels, as described in following paragraphs.
On the Mn3+-rich side (0.2 < x < 0.5), the antiferromagnetic coupling of the LaMnO3 type
between ferromagnetic layers (Figure 11.6) gives way to complete ferromagnetism. On the
Mn4+-rich side, several charge-ordered “superstructures” appear, such as La1/3Ca2/3MnO3
shown in Figure 11.8, where complex antiferromagnetism (see Problem 11.6) occurs due to
empty eg orbitals of Mn4+ and ordered dz2 of Mn3+ in a manner similar to the next example.
The charge-ordered arrangement of La0.5Ca0.5MnO3 supports antiferromagnetism when the
dz2 Mn3+ orbitals order upon uniaxial expansion of MnO6/2 octahedra (Figure 11.9). This orbital
ordering dictates magnetic interactions by fixing the orientations of the empty dx2−y2 orbitals of
Mn3+. Because Mn4+ has both dz2 and dx2−y2 orbitals empty, it is always able to make
antiferromagnetic superexchange via O p orbitals. The magnetic order therefore depends on
Mn3+. The four lobes of its empty dx2−y2 orbital overlap with p orbitals of four oxygens

300
Tco
TC
T (K)
Too
200
TN

100
FM AFM

0
0.2 0.3 0.4 0.5 0.6 0.7 0.8
x in La1−xCaxMnO3

Figure 11.7 La1−xCaxMnO3 phase diagram. Ordering of spins (ferromagnetic below TC, antiferromag-
netic below TN), of orbitals below Too, and of the Mn3+ and Mn4+ charges below Tco. Compiled from a
variety of sources. Vertical lines mark compositions for which the actual ordered “superstructure” was
determined experimentally. Stability regions of different phases change with magnetic field.
 11.3 Charge and Orbital Ordering in Perovskite-Type Manganites 469

La/Ca Mn4+ Mn3+

Figure 11.8 Charge ordering below Tco in La1/3Ca2/3MnO3 [12].

Mn3+
E C E C
Mn4+
C E C E
dz2
dx2−y2 E C E C

C E C E
AF
AF AF
AF

Figure 11.9 Magnetism in charge- and orbital-ordered La0.5Ca0.5MnO3 (dark Mn4+, white Mn3+). Left:
A plane with ordered half-filled dz2 Mn3+ orbitals (hatched) that fit the structural distortions. The empty
dx2−y2 orbitals each make four antiferromagnetic Mn–O–Mn couplings via p orbitals of oxygen (not
shown) with empty eg orbitals (not drawn) of two Mn4+ (in plane) and two Mn3+ (out of plane) neighbors.
Right: The resulting CE-type order of spins in a slightly tilted view for clarity. After [13,14].

mediating four antiferromagnetic couplings towards Mn neighbors. The other two couplings are
ferromagnetic.6 The total antiferromagnetic order combines the C and E types introduced in
Figure 9.19, and is denoted CE.7
The Mn3+-rich ferromagnetic phase and the CE-type antiferromagnetic phase compete
when the composition parameter x increases towards the x = 0.5 line in Figure 11.7. The

6
Note in Figure 11.9 that the half-filled dz2 orbitals of Mn3+ expand the Mn4+–O–Mn3+–O–Mn4+ distance, along
which the three Mn couple ferromagnetically.
7
To compare the C (or E) cells identified in Figure 11.9 with the types of antiferromagnetic order in Figure 9.19, sketch
any C (or E) cell from Figure 11.9 with a + and − on the corners for the two opposing moment orientations, do the
same with the dotted cell in Figure 9.19, and rotate till they unite.
470 Magnetotransport Materials

100
applied field 0 T
resistivity (Ω m) 10

0.1

0.01
applied field 8 T
0.001

0.0001
50 100 150 200 T (K) 250

Figure 11.10 Colossal (negative) magnetoresistance in La0.5Ca0.5MnO3, as observed by measuring resist-

ivity under an applied magnetic field (of flux density) 0 T and 8 T, with both measurements performed as a
function of temperature. In the 0 T field, the sample is antiferromagnetic and has thermally activated
conductivity. In 8 T, it is a ferromagnetic “bad” metal above 125 K and a frustrated antiferromagnet
below. After data in [15].

proportion of these two magnetic phases changes with temperature and also with applied
magnetic field that tends to align the moments ferromagnetically. In this region where the
phases compete, the increased filling of eg electrons makes metallic delocalization the
preferred method of removing the eg degeneracy, further stabilizing the ferromagnetism by
double-exchange interactions (Section 9.9.1). In contrast to this metallic delocalization, the
competing antiferromagnetic phase has a thermally activated conductivity of localized
electrons. As a consequence, up to a six-orders-of-magnitude decrease in electrical resistivity
(Figure 11.10) may be observed in such samples upon exposure to a strong magnetic field.
This negative8 magnetoresistance observed in manganites is therefore termed colossal
magnetoresistance.

11.3.3 Tuning the Colossal Magnetoresistance

Replacement of the large La3+ in La0.5Ca0.5MnO3 by smaller trivalent lanthanide ions
reduces the width of the Mn(eg)–O(p) conduction band by increasing the octahedral tilting
that bends the Mn–O–Mn angles (Table 6.2). The narrow band favors electron localization
(Section 10.2.5) and stabilizes the charge-ordered phase. This also means that the range over
which the ferromagnetic and antiferromagnetic phases coexist in Figure 11.7 (controllable by
8
It is called “negative” because the resistance decreases under the magnetic field.
 11.3 Charge and Orbital Ordering in Perovskite-Type Manganites 471

the magnetic field) shifts to lower x values. Phases like Nd0.7Ca0.3MnO3 show the typical
two-phase instabilities and colossal magnetoresistance.
Replacement of Ca2+ with larger Sr2+ destabilizes the charge-ordered phase in favor of the
metallic ferromagnetic phase because the reduced octahedral tilting increases the bandwidth,
and this favors delocalized electrons. This general trend can be seen in Figure 11.11. While
La0.55Sr0.45MnO3 (on the right) is a metallic ferromagnet with high TC, its calcium analogue
La0.55Ca0.45MnO3 (middle) has a much lower TC, and replacement of La with a smaller
lanthanide converts it into a charge-ordered antiferromagnet such as Nd0.55Ca0.45MnO3
(left). As discussed below, two parameters are responsible for this complex behavior—the
general size effect and the local variance effect of the solid solution at the A site.
From Chapter 1, we know how the size of the perovskite A-site cation determines the
octahedral tilting through the tolerance factor. As discussed in Sections 10.4.3 and
6.6.5, the tilting in turn controls the width of the Mn(eg)–O(p) conduction band and
thus the stability of the ferromagnetic metallic state. When the cations occupying the A
site have similar size, the properties can be easily tuned by controlling the average
radius. However, in most cases these cations differ in size, often considerably. Then we
have to consider not only the average size, but also the mismatch in size of the A-site
cations, which creates additional local distortions. A suitable parameter to evaluate the
mismatch is the statistical variance σ2 of ionic radii. Termed the variance effect [16], it
P P
is expressed as σ2 ¼ i ðxi r2i  rA2 Þ, where rA2 ¼ i xi ri2 is the weighted average of ionic-radii

0.02
spin glass
variance
radius

Gd
0.01
Sm

Nd TC

Gd Pr
Tco Sm Ca
Nd Sr,R = La
0.00 Pr = R = La

1.25 1.30 1.35 1.40

radius average (Å)

Figure 11.11 Magnetic behavior in a contour map of charge-ordering (Tco) and Curie-temperature (TC)
isotherms (in K), in the A-site ionic-radius [17] variance–average plot [18] for R0.55AE0.45MnO3 (R =
lanthanoid, AE = Ca1−xSrx and Sr1−xBax). Of many such compositions, two paths are marked, AE = Ca
with squares, AE = Sr with circles (ferromagnetic shaded, antiferromagnetic not).
472 Magnetotransport Materials

squares for the in total i atoms of fractional site occupancies xi at A. With fine compositional
scaling in terms of a three-atom solid solution at the A site, the effects of the size parameter
and of the variance parameter can be discerned [17]. The result is shown in the contour plot
of Figure 11.11 that maps the stability of the charge- and spin-ordered variants of
R0.55AE0.45MnO3 (R = lanthanoid, AE = Ca1−xSrx and Sr1−xBax) in terms of their ordering
temperature.
The contour map of isotherms in Figure 11.11 explains all. Low variance (equal size) means
low size-induced disorder at the A site, which favors either the ferromagnetic metallic phase or
the charge-ordered antiferromagnetic phase, which in turn depend on the average radius at the
A site as described above. Increasing the variance, i.e. the local size disorder, quickly disrupts
the charge ordering (as seen on left) that facilitates the antiferromagnetic order, yet much less
so the ferromagnetic spin ordering (seen on right). Ferromagnetism persists to moderate
values of variance, but eventually a magnetically disordered spin-glass phase wins out.
In total, three compositional parameters tailor magnetotransport properties of these
perovskite manganates: The doping level of the alkaline-earth ion(s), which controls the
oxidation state of Mn. The average radius at the perovskite A site, which varies the stability
of the antiferromagnetic charge-ordered versus ferromagnetic states. The A-site cations’ size
variance, which fine tunes the critical competition between these two.

11.4 Half-Metals and Spin-Polarized Transport

In Section 11.2.1 we learned that the itinerant minority-spin electron makes a single crystal of
ferromagnetic magnetite a source of spin-polarized current. Half-metals (Section 9.9.1),
being metallic for one spin direction and insulators for the other spin direction, have spin
polarization P = 1, Equation (11.2). They are by definition pure spin-polarized conductors.
The phenomenon of half-metallicity first gained attention when spin polarization and
magnetic order were included in band-structure calculations. The term itself was coined by
de Groot et al. [19] in 1983 to describe the band structure of the Heusler alloy NiMnSb.
Calculation remains an important tool to identify possible half-metals because specific and
quantitative experimental proof of half-metallicity is difficult to obtain (see also Box 11.3).
An inherent problem is that a complete spin polarization may be thwarted by temperature,
defects, and spin–orbit interactions. Non-zero temperatures tend to disorder the ferromag-
netism and excite carriers of the opposite spin into the conduction band. Defects play a
similar role, in particular the largest of them, the crystallite surface.

11.4.1 Magnetoresistant Properties of Half-Metals

Tunneling magnetoresistance is one of the potentially interesting transport effects that can
arise between two half-metallic electrodes separated by a thin tunneling barrier. Let’s
consider in more detail the previously mentioned Jullière experiment [1] (Section 11.1) with
 11.4 Half-Metals and Spin-Polarized Transport 473

two ferromagnetic epitaxial films as the electrodes. The tunneling from one electrode (emit-
ter, filled states) to the other electrode (collector, empty states) occurs in two independent
channels, one for ↑ and one for ↓ electrons (spin conservation assumed). The two conduct-
ances, G↑ and G↓, are each assumed to be proportional to the product of the densities of
states9 at the Fermi level for the two ferromagnetic electrodes, for example G↓ = n1↓·n2↓ (the
qualifier EF is omitted for simplicity). Let’s now consider varying directions of magnetization
at the two electrodes. Two limiting cases emerge—Gpar for the two ferromagnetic films of
parallel and Gapar of antiparallel magnetization:

Gpar ¼ n↑1 n↑2 þ n↓1 n↓2 (11.3)

Gapar ¼ n↑1 n↓2 þ n↓1 n↑2 (11.4)

In a magnetoresistance experiment, a strong enough applied field turns the moments parallel.
The maximum tunneling magnetoresistance, TMR, is the normalized difference between
Gpar and Gapar. The normalization is performed against either value, giving two alternative
expressions justifying the Jullière formula10 in Equation (11.1),

Rapar  Rpar Gpar  Gapar ðn↑1 n↑2 þ n↓1 n↓2 Þ  ðn↑1 n↓2 þ n↓1 n↑2 Þ
TMR ≡ ¼ ¼
Rapar Gpar ðn↑1 n↑2 þ n↓1 n↓2 Þ
2P1 P2
¼ (11.5)
1 þ P1 P2
Rapar  Rpar Gpar  Gapar ðn↑1 n↑2 þ n↓1 n↓2 Þ  ðn↑1 n↓2 þ n↓1 n↑2 Þ
TMR ≡ ¼ ¼
Rpar Gapar ðn↑1 n↓2 þ n↓1 n↑2 Þ
2P1 P2
¼ ; (11.6)
1  P1 P2

where spin polarizations Pi of the two electrodes 1 and 2 were introduced by substitution for n1↑
and n2↑ from Equation (11.2). The formula in Equation (11.5) has a maximum TMR = 1, whereas
the maximum is ∞ for Equation (11.6). Both are used in the literature, along with Rapar/Rpar.
As noted in Section 11.1, ferromagnetic metals have low P values. Even with half-metals,
only single-crystalline films have P1 and P2 close to 1 that would yield high TMR for the two
spin alignments at the two sides of the tunneling barrier [20]. Since the grain and domain
structure of half-metallic films can be difficult to control, attention has been given to powder
magnetoresistance, the change in conductance/resistance of half-metal powder compacts
upon a change in the external magnetic field. Its advantage is the isotropy of powders that
eliminates directional variations. The insulating surfaces of the individual powder grains are
the tunneling barriers, as electrons have to percolate across many grains of random spin

9
More precisely, to an effective value called tunneling density of states.
10
The version with the + sign appears in Jullière’s article [1] but now both versions are in use.
474 Magnetotransport Materials

orientation. After Coey and Venkatesan [20], we simplify the general result of Guinea [21] to
assume that the probability of magnetoresistant tunneling between two average half-metallic
grains i and j of the mixture varies as cos2(θij/2) = (1 + cosθij)/2 of the average angle 0° < θij <
90° between any two grain spins in the mixture, and so does the conductance G:

1 þ cosθij
G∝ ð0° < θij < 90°Þ (11.7)
2

The maximum observed TMR is the normalized difference in G (or in R = 1/G). Figure 11.12
illustrates the effect of the external magnetic field on the average angle. The θij = 0° case is the
strong-field limit for all spins aligned parallel, and resistance Rpar is measured in Figure 11.12.
The θij = 90° is the average angle of particle pairs, randomly distributed between having spins 0°
parallel and 180° antiparallel, for the demagnetized powder at the coercive field, and Rrandom is
measured in Figure 11.12. The spin-aligned case means cosθij = 1, say Gpar = 1 after Equation
(11.7). The random demagnetized powder has a cosθij average equal to 0, and Grandom = ½. For
the two types of normalization of the resistance difference given in Equations (11.5) and (11.6),
the following maxima for the TMR in powder compacts of half-metals are obtained, respectively:

Rrandom  Rpar Gpar  Grandom 1  ½

TMR ≡ ¼ ¼ ¼½ (11.8)
Rrandom Gpar 1
Rrandom  Rpar Gpar  Grandom 1  ½
TMR ≡ ¼ ¼ ¼1 (11.9)
Rpar Grandom ½
magnetization

Rrandom
θij
resistance

Rpar
0
magnetic field

Figure 11.12 Magnetoresistance of a powder compact and magnetic configurations in a percolative

conduction link at saturation and coercivity, after [22].
 11.4 Half-Metals and Spin-Polarized Transport 475

Given that half-metals at non-zero temperatures never have 100% spin polarization,
similar spin polarizations can also be achieved for materials that are not strictly half-metallic:
EuO (becomes ferromagnetic at low temperatures; Section 9.9.1) has a calculated band gap
[23] of 3.4 eV for one spin direction and 2.5 eV for the other. The La1−xCaxMnO3-type
perovskites have a small minority-spin density at the Fermi level, which makes them nearly
half-metallic spin-polarized conductors and contributes to the exhibited variety of magneto-
transport features. Similarly, magnetite at ambient conditions has a nearly zero band gap for
minority-spin electrons and ~0.7 eV for majority-spin d electrons. The resulting spin polar-
ization is sometimes termed transport half-metallicity or transport spin polarization. In some
connotations, these phases are simply counted among half-metals as well.
Half-metals can be categorized in several ways. One of them considers the origin of the
spin-polarized itinerant electron: (1) half-metals with integer-valence itinerant electrons,
such as CrO2 and NiMnSb; and (2) half-metals with valence-mixing itinerant electrons,
such as Fe3O4 and Sr2FeMoO6. Coey and Venkatesan [20] list a more detailed categorization
of half-metals that also includes the transport half-metals.

Box 11.3 Characterizations: Experimental proofs of half-metallicity

Magnetization of half-metals, in contrast to magnetic order in metals, leads to integer
values of the saturated magnetic moments per magnetic atom because of the existence
of the insulating spin direction in which the number of localized electrons per atom
must be an integer dictating the number of electrons in the itinerant spin direction. The
integer magnetic moment is a necessary but not sufficient condition of half-metallicity.
Positron annihilation is a selective but demanding test of half-metallicity. A half-metal
has a Fermi surface for one spin direction only, not for both as in ordinary metals.
Spin-polarized positrons radiating from a suitable beta source, such as 22Na, would
annihilate with electrons of opposite spin at the Fermi surface of a half-metallic thin
film or single crystal. Each individual annihilation forms two γ photons, the directions
of which conserve the linear momentum and energy that are detected and evaluated.
One of a small number of such studies was performed on half-metals and concerned
NiMnSb [24].
Andreev reflection [25] occurs for conduction across a metal–superconductor interface.
Say, a ↓ electron crosses from the metal into a superconductor well below its Tc. As it
does, it has to grab a ↑ electron from the metal it leaves because in the superconductor
it will only move as a Cooper pair (↓↑). This leaves a backscattered or “reflected” hole
in the ↑ band of the metal, doubling the conductance G = dI/dV (for voltage V → 0)
against what would be achieved if the collector was non-superconducting. In half-
metals, only the ↓ band is occupied, there is no ↑ electron to grab, and the Andreev
reflection is suppressed [26]. One observes that for V → 0, the conductance dI/dV has a
minimum. The spin polarization P is given by the position of the actual conductance at
V → 0 on the scale between this minimum (P = 1) and the doubling maximum of the
476 Magnetotransport Materials

Box 11.3 (cont.)

Andreev reflection (P = 0). Because the Andreev reflection requires very good interface
contacts with little or no tunneling, its application on half-metals suffers from limita-
tions for half-metallicity at the surface of the solid.
Spin-resolved photoemission directly yields the degree of spin polarization on oriented
surfaces. The information is deduced from the spatial energy distribution of electrons
emitted from thin films or single crystals of half-metals by polarized light. Analogously,
spin-resolved inverse photoemission uses a low-energy polarized-electron beam and
measures the distribution of the emitted photons. The disadvantage is that both
methods only see a few atomic layers below the surface, and the obtained spin polar-
ization depends on how well the surface of the given crystal structure reconstructs to
carry the half-metallicity of the bulk. As a result, it is not easy to compare spin
polarization values among different half-metals.

11.4.2 CrO2
Half-metallicity of CrO2 was identified by Schwarz [27] in 1986. Chromium dioxide at
that time was a popular magnetic tape-recording material. The ionic model of CrO2
implies a 3d 2 electron configuration, and the integer spin-only ferromagnetic moment
of 2.0 µB per Cr at 0 K suggests that the CrO2 ferromagnetism mentioned in Section
9.9.1 is not one of itinerant electrons such as in Fe or Ni (Section 9.9.2). Actually, in a
simple picture, superexchange interactions in its rutile-type structure could well make it
an antiferromagnet, and the integer local moment would be as expected for a semicon-
ductor. Yet CrO2 is a ferromagnet and metallic conductor [28]. Two circumstances are
essential for CrO2 ferromagnetism: (1) The CrO6 octahedra are slightly deformed into
two short and four long Cr–O bonds [29]. The latter expand the xy plane of the
octahedron, which lowers the dxy orbital below the energy of dxz and dyz, so that the
dxy orbital accommodates one electron of the two. (2) The high Cr4+ electronegativity
lends high covalency to the dxz and dyz interaction with the oxygen p orbitals. The final
result is that while the electron in the dxy orbital is localized, the other two orbitals
accommodate the second Cr electron in a band with π* character. The π* electrons are
itinerant and the ensuing double-exchange interaction favors ferromagnetism. Hund’s
rule favors parallel alignment of the two electrons, accounting for the saturation
moment of 2.0 µB per Cr as the temperature approaches 0 K [28].
The half-metallicity of CrO2 at 0 K is manifested by spin polarization of nearly 1.0 by
Andreev reflection (Box 11.3) at freshly made point contacts at liquid-helium temperatures.
Owing to the natural surface layer of antiferromagnetic Cr2O3, tunneling magnetoresistance
of CrO2 powder compacts at low temperatures also yields high levels of spin polarization.
However, increasing temperature brings a strong contribution from hopping (not polarized)
 11.4 Half-Metals and Spin-Polarized Transport 477

conductivity, and this effectively eliminates the tunneling contribution. At room tempera-
ture, the TMR is only ~3% [30].

11.4.3 Heusler Alloys

Half-metallicity occurs in some Heusler intermetallics such as NiMnSb or Co2MnSb
(Section 1.4.2). A simple picture of the electronic band structure that elucidates the half-
metallic behavior is obtained by considering these alloys as Zintl phases (Section 1.5.6),
where the more electronegative Sb atom achieves an octet, Sb3−. In the following treatment,
we assume these anions cause a crystal-field splitting of the transition-metal d orbitals that
then form molecular orbitals and eventually bands.
Figure 11.13 shows the [MnNi]3+ molecular orbitals in the half-Heusler alloy NiMnSb,
formed by overlap (mixing, hybridization) of the Ni and Mn d orbitals that split in the
tetrahedral field of Sb3− into a doubly degenerate e set and a higher-lying triply degenerate t2
set (Figures 7.2 and 5.23). Orbitals of the same symmetry overlap to form five bonding and
five antibonding orbitals. These widen into bands, for which this approximate picture is valid
close to the full-symmetry Γ point in the Brillouin zone. Note that the atomic orbitals of the
more electronegative Ni are lower in energy than those of Mn. The properties of the phase
then depend on the filling of these molecular orbitals.
Filling of the scheme in Figure 11.13 with the 22 valence electrons of NiMnSb is shown in
Figure 11.14. The energy shift between the spin-up and spin-down d bands11 is such that the
electron-filling process, like pouring in water, places the Fermi level between the bonding and

Ni bands Mn Sb
t2 *

e*
t2

e
Mn Sb
t2
Ni
e t2 NiMnSb
NiMn in mutually
e tetrahedral coordination
Not to scale:

s,p

Figure 11.13 Molecular-orbital scheme for the half-Heusler phase NiMnSb, after [31].

11
The electron-pairing energy; a sum of the Hund-rule coupling and coulombic repulsion.
478 Magnetotransport Materials

NiMnSb: 22 valence electrons, 13↑, 9↓

μsat = 13 − 9 = 4 μB

Majority spin Minority spin

9 electrons

Sb3−

Ni ban ds Mn Ni bands Mn

Sb3−

Figure 11.14 Filling of molecular orbitals (bands) in NiMnSb by majority-spin and minority-spin
electrons.

antibonding bands of the minority spin while crossing the incompletely filled antibonding t2*
band of the majority spin. The majority spin is metallic; the minority spin has a gap. We see
four uncompensated majority spin electrons of predominantly Mn parentage that give rise to
ferromagnetism with a saturated moment μsat = 4.0 µB per Mn (Figure 11.14). Positron-
annihilation studies [24] see spin polarization in NiMnSb of nearly 1.0. Half-metallicity is
also obtained when the valence-electron total is decreased to 21 by replacing Ni with Co, or
Sb with Si, and μsat is 3 µB. But when the molecular orbitals (bands) in Figure 11.14 are filled
with only 20 electrons, the Fermi level sinks to cross the minority-spin band, the moment
becomes a non-integer and half-metallicity disappears. The condition for half-metallicity in
the half-Heusler alloy XYZ (Z is typically a main-group metalloid) of z valence electrons per
formula is then μsat = (z − 18) equals 4 μB or 3 μB. While half-Heusler compounds with z = 19
(VCoSb) or 17 (NdNiSb) remain ferromagnetic metals, TiCoSb or TiNiSn with z = 18 lose
ferromagnetism as the spin-up and spin-down electrons cancel. The latter is the basis for the
semi-empirical Slater–Pauling rule [32] that μsat = VEC − 6 (VEC being the valence-electron
count per average atom), which is valid for the so-called localized ferromagnetism of individ-
ual atomic magnetic moments that occurs up to VEC ≈ 8.5.12
The half-metallicity of full-Heusler alloys X2YZ can be treated like the XYZ case. Let’s
consider Co2MnSb. The formally [Co2Mn]3+ molecular orbitals are formed by overlap of the

12
At higher VEC, itinerant ferromagnetism due to conduction electrons takes over, as discussed in Section 9.9.2 and
shown in Figure 9.27 for the example of Ni.
 11.4 Half-Metals and Spin-Polarized Transport 479

Co2MnSb: 30 valence electrons, 18↑, 12↓ bands

t2 *
μsat = 18 − 12 = 6 μB
Mn
e*
Co

eu
minority gap
t1u
Co
Co2MnSb

t2
24 el.
e

s,p Sb3−

Figure 11.15 Formation of the minority gap in the [Co2Mn]3+ molecular orbitals of the Co2MnSb full-
Heusler alloy with 30 valence electrons per formula, after [31].

Mn atomic orbitals with the Co2 molecular orbitals.13 Note in Figure 11.15 that the Co sites
alone would form a primitive cubic lattice of Co cubes filled alternately with Mn and Sb. The
Co 3d orbitals split in the octahedral field of their six Co neighbors into the eg and t2g sets that
overlap to form five bonding Co2 molecular orbitals of e and t2 symmetry. These orbitals can
mix with the Mn orbitals. The five antibonding orbitals have a symmetry that does not mix
with the e and t2 orbitals of Mn. They therefore remain nonbonding with respect to Mn. The
resulting orbital-energy scheme is shown in Figure 11.15.
The full-Heusler alloy becomes half-metallic when filling of the molecular-orbital scheme
places the Fermi level between the minority-spin nonbonding eu and t1u orbitals. This
happens for example in Co2MnSb with 30 valence electrons. These are filled into the bands
in Figure 11.15, placing 12 electrons into the minority bands and 18 in the majority bands.
The spin of 6 electrons remains uncompensated. Of these, two occupy the eu nonbonding
orbitals of Co2 parentage; one electron per Co. The remaining four will occupy the anti-
bonding orbitals of Mn parentage. This corresponds to the actual location of the magnetic
moments in the structure. Owing to ferromagnetic coupling, μsat = 6 µB. Co2MnSi (29 valence
electrons) is also half-metallic, with 2 µB at the two Co atoms and 3 µB at Mn. The Co2YZ
Heusler alloys remain half-metallic down to 26 valence electrons per formula, and the integer

13
The orbital approach is used here as a local precursor of bands present in these extended structures.
480 Magnetotransport Materials

μsat = z − 24 μB is observed. Below z = 26, the Fermi level sinks into the minority-spin band,
leaving the phase merely ferromagnetic (except for z = 24 when the electron spins cancel).
A full generalization of simple rules and properties for all X2YZ compositions is not possible
as the situation is too complex. Not all metals form the desired phase. Even if the correct
composition is formed, it may adopt the so-called inverse Heusler structure. An example is
Mn2CoAl where half of the Mn atoms adopt the zinc-blende sites inside the unit cell. The result is
a spin-gapless semiconductor, a material with a band gap for one spin and a zero band gap for
the other [33]. Another structural variant is the quaternary Heusler compound of two different X
elements, an X and an X 0 , such as (CoMn)VAl, a semiconductor of z = 24 electrons, complying
with the Slater–Pauling rule just like the other Heusler intermetallics [34].

11.4.4 Half-Metals with Valence-Mixing Itinerant Electrons

The prototype of such ferromagnetic phases, the transport half-metal magnetite, has been
discussed in Section 11.2.1. Figure 11.2 illustrates that sharing the minority-spin Fe2+
electron with Fe3+ is responsible for the conductivity jump observed upon heating through
the Verwey transition. A similar valence mixing may occur between two different metal ions,
as illustrated in Figure 11.16 for the Sr2FeMoO6 perovskite where ordered Fe and Mo share
the single minority electron of Fe2+ origin [35] and couple antiferromagnetically:
Sr2FeMoO6 gained attention when a relatively high TMR = 0.1, as defined in Equation
(11.9), was reported [36] even at room temperature, compared with TMR = 0.3 at 4.2 K. This
fueled hopes for a material with large magnetoresistance at room temperature.
Unfortunately, Sr2FeMoO6 preparations tend to suffer from intersite disorder between Fe
and Mo, which affects the observed magnetic moment as well as the magnetic ordering
temperature. By extrapolation of neutron powder-diffraction data to the extremes of order
and disorder, the saturated magnetic moment of the highly ordered phase is shown [37] to be
approximately 4 µB at low temperatures and disappears at a TN of 440 K. This is the normal
moment of a high-spin Fe2+ in an octahedral field, but also that of Fe2.5+ (4.5 µB) minus the
opposite spin of 0.5 µB on Mo (both due to the ½ electron present at Fe and Mo), hence
consistent with the picture of a ferrimagnet with antiferromagnetic coupling of spins between
Fe and Mo. In contrast, the disordered samples form Fe–Fe and Mo–Mo antiferromagnetic
clusters of higher TN than the ordered phase, lowering the total measured magnetic moment.

Fe2+ + Mo6+ ↔ Fe3+ + Mo5+

Figure 11.16 Valence mixing Fe2+ + Mo6+ → Fe2.5+ + Mo5.5+ ← Fe3+ + Mo5+ in the Sr2FeMoO6
perovskite. The minority-spin electron of Fe2+ origin is shared with the thus antiferromagnetically
coupled Mo.
 11.5 Problems 481

11.5 Problems
11.1 Tunneling magnetoresistance between two ferromagnetic films oriented parallel and
antiparallel was measured to be 0.4, referring to Equation (11.1). Calculate the spin
polarization P and the corresponding relative percentages n↑ ðEF Þ and n↓(EF) for the
tunneling electrons.
11.2 All atomic magnetic moments in YBaFe2O5 lie along the y axis. Based on Figure
11.3 (in which two chemical unit cells are drawn on the left, four on the right) and
information in the text, sketch with arrows the Fe magnetic moments in the
smallest magnetic-unit-cell parallelepipeds of charge-ordered and valence-mixed
YBaFe2O5.
11.3 YBaMn2O5 has a structure with two crystallographically different Mn sites:

T = 1.5 K O(1)
Mn(1) bonds
1 × 2.08 Å
4 × 1.91 Å

O(2)
O(1)
Mn(2) bonds
1 × 1.96 Å
4 × 2.09 Å

O(2)

(a) Determine the oxidation state and orbital occupancies of Mn(1) and Mn(2) in charge-
ordered YBaMn2O5. (b) The occupation of one of the d orbitals alternates between
Mn(1) and Mn(2). Which orbital is involved in YBaMn2O5 orbital ordering?
11.4 For each of the following perovskites predict whether there will be a cooperative Jahn–
Teller distortion: (a) LaCrO3, (b) LaTiO3, (c) low-spin LaCoO3, (d) high-spin NdMnO3.
11.5 You are asked to synthesize two samples of La0.8Sr0.2MnO3+δ, with δ = 0.1 and 0. At
which temperature would you equilibrate the δ = 0.1 composition in ambient O2
atmosphere prior to quenching? How would you equilibrate the δ = 0 composition in
commercial Ar gas of pO2 ≈ 10−4 bar? Use the Schottky plot in Box 11.2, where
isotherms are spaced evenly by 100 degrees.
11.6 Into the figure below, sketch the ordered dz2 orbitals of Mn3+ such that
their orientation explains the magnetic moments in this La1/3Ca2/3MnO3 layer.
Use the principle that follows from the magnetic order of La0.5Ca0.5MnO3 in
Figure 11.9.
482 Magnetotransport Materials

3+ 4+

11.7 Perovskites with partially filled eg orbitals at the octahedrally coordinated atom
respond to this impending violation of orbital degeneracy in a variety of ways. In
SrFeO3, the eg electrons are delocalized and metallic behavior ensues. In LaMnO3, the
eg electrons are localized and the octahedra distort through cooperative Jahn–Teller
distortions. In CaFeO3, the orbital degeneracy is removed through charge dispropor-
tionation into an ordered pattern of Fe3+ and Fe5+. The RNiO3 (R = trivalent rare-
earth ion) perovskites are another system with partial filling of eg orbitals. The phase
diagram between critical temperatures for insulator-to-metal (TIM) and antiferromag-
netic-to-paramagnetic (TN) transitions in these nickelates is shown in the figure below
(SmNd approximates Pm).

RNiO3
TIM

400 paramagnetic paramagnetic

metal
Tc (K)

insulator
Nd
200 TN Sm
Sm Pr
Gd Eu Nd
Tm Y
Lu
antiferromagnetic insulator La
0
R3+ radius

(a) What is the oxidation state, d-electron count and occupation of the t2g and eg
orbitals for the low-spin nickel in these compounds? (b) Would you expect the
nickel-centered octahedra in LaNiO3 to be symmetric or to be distorted as a result
of a cooperative Jahn–Teller distortion? (c) Whereas LaNiO3 is a paramagnet and a
metallic conductor at all temperatures, the lanthanoid RNiO3 phases undergo a
metal–insulator transition upon cooling. What stabilizes metallic LaNiO3 even at
low temperatures and why? (d) Why does the insulator-to-metal transition tempera-
ture TIM increase along the rare-earth series? (e) Which changes in the crystal
structure of YNiO3 would you expect upon cooling through TIM? How would the
 11.7 References 483

Ni–O distances allow you to see whether the low-temperature paramagnetic insu-
lating phase is stabilized by a cooperative Jahn–Teller distortion or by a charge
disproportionation?
11.8 Sketch the Andreev reflection of an electron crossing from a metal into a supercon-
ductor in an electric circuit.
11.9 Calculate the total valence-electron content z per formula and predict the properties of
the following half-Heusler alloys:

PdMnSb IrMnSb TiNiSn TiCoSb MnCoSb

z
Ferromagnet?
μsat (in μB)
Half-metal?
Semiconductor?

11.10 Sketch the spin-up (majority) and spin-down (minority) orbitals of Co2MnSi around
the Fermi energy so that half-metallicity is achieved.
11.11 Spin-resolved density of states were obtained by M. Kallmayer et al. [38] on Co2MnSi
thin films. The DOS of the minority spin was 24 units, of the majority spin 306 units.
Calculate the degree of polarization and the optimal tunneling magnetoresistance
between two epitaxial films with both alternative formulas.

11.6 Further Reading

S. Bandyopadhyay, M. Cahay, “Introduction to Spintronics” (2015) CRC Press.
C. Felser, A. Hirohata (editors), “Heusler Alloys” Volume 222 in Springer Series in Materials Science,
(2016) Springer.
F. Hagelberg, “Magnetism in Carbon Nanostructures” (2017) Cambridge University Press, Part IV—
Transport Phenomena”.

11.7 References
[1] M. Jullière, “Tunneling between ferromagnetic films” Phys. Lett. A 54 (1975), 225–226.
[2] M.B. Robin, P. Day, “Mixed valence chemistry. A survey and classification” Adv. Inorg. Chem.
Radiochem. 10 (1967), 247–422.
[3] E.J.W. Verwey, “Electronic conduction of magnetite and its transition point at low temperat-
ures” Nature (London) 144 (1939), 327–328.
[4] Z. Kąkol, “Magnetic and transport properties of magnetite in the vicinity of the Verwey transi-
tion” J. Solid State Chem. 88 (1990), 104–114.
484 Magnetotransport Materials

[5] M.S. Senn, J.P. Wright, J.P. Attfield, “Charge order and three-site distortions in the Verwey
structure of magnetite” Nature 481 (2012), 173–176.
[6] P.M. Woodward, P. Karen, “Mixed valence in YBaFe2O5” Inorg. Chem. 42 (2003), 1121–1129.
[7] J. Lindén, F. Lindroos, P. Karen, “Orbital occupancy evolution across spin- and charge-ordering
transitions in YBaFe2O5” J. Solid State Chem. 252 (2017), 119–128.
[8] B.C. Hauback, H. Fjellvåg, N. Sakai, “Effect of nonstoichiometry on properties of La1−tMnO3+δ
III. Magnetic order studied by powder neutron diffraction” J. Solid State Chem. 124 (1996),
43–51.
[9] J.H. Kuo, H.U. Anderson, D.M. Sparlin, “Oxidation–reduction behavior of undoped and
Sr-doped lanthanum manganese oxide LaMnO3 nonstoichiometry and defect structure” J.
Solid State Chem. 83 (1989), 52–60.
[10] B.C. Tofield, W.R. Scott, “Oxidative nonstoichiometry in perovskites, and experimental survey.
Defect structure of an oxidized lanthanum manganite by powder neutron diffraction” J. Solid
State Chem. 10 (1974), 183–194.
[11] J.A. Alonso, M.J. Martínez-Lope, M.T. Casais, J.L. MacManus-Driscoll, P.S.I.P.N. de Silva,
L.F. Cohen, M.T. Fernandez-Diaz, “Non-stoichiometry, structural defects and properties of
LaMnO3+δ with high delta values (0.11 < δ < 0.29)” J. Mater. Chem. 7 (1997), 2139–2144.
[12] P.G. Radaelli, D.E. Cox, L. Capogna, S.-W. Cheong, M. Marezio, “Wigner-crystal and bi-stripe
models for the magnetic and crystallographic superstructures of La0.333Ca0.667MnO3” Phys. Rev.
B 59 (1999), 14440–14450.
[13] E.O. Wollan, W.C. Koehler, “Neutron-diffraction study of the magnetic properties of the series
of perovskite-type compounds La1−xCaxMnO3” Phys. Rev. 100 (1955), 545–563.
[14] J.B. Goodenough, “Theory of the role of covalence in the perovskite-type manganites [La,M(II)]
MnO3” Phys. Rev. 100 (1955), 564–573.
[15] G.Q. Gong, C.L. Canedy, G. Xia, J.Z. Sun, A. Gupta, W.J. Gallagher, “Colossal magnetoresist-
ance in the antiferromagnetic La0.5Ca0.5MnO3 system” J. Appl. Phys. 79 (1996), 4538–4540.
[16] L.M. Rodríguez-Martínez, J.P. Attfield, “Cation disorder and size effects in magnetoresistive
manganese oxide perovskites” Phys. Rev. B 54 (1996), R15622–R15625.
[17] R.D. Shannon, “Revised effective ionic radii and systematic studies of interatomic distances in
halides and chalcogenides” Acta Crystallogr. Sect. A 232 (1976), 751–767.
[18] Y. Tomioka, Y. Tokura, “Global phase diagram of perovskite manganites in the plane of
quenched disorder versus one-electron bandwidth” Phys. Rev. B 70 (2004), 014432/1–5.
[19] R.A. deGroot, F.M. Mueller, P.G. Van Engen, K.H.J. Buschow, “New class of materials: Half-
metallic ferromagnets” Phys. Rev. Lett. 50 (1983), 2024–2027.
[20] J.M.D. Coey, M. Venkatesan, “Half-metallic ferromagnetism. Example of CrO2” J. Appl. Phys.
91 (2002), 8345–8350.
[21] F. Guinea, “Spin-flip scattering in magnetic junctions” Phys. Rev. B 58 (1998), 9212–9216.
[22] J.M.D. Coey, “Powder magnetoresistance” J. Appl. Phys. 85 (1999), 5576–5581.
[23] Z. Szotek, W.M. Temmerman, A. Svane, L. Petit, P. Strange, G.M. Stocks, D. Koedderitzsch, W.
Hergert, H. Winter, “Electronic structure of half-metallic ferromagnets and spinel ferromagnetic
insulators” J. Phys.: Cond. Matter 16 (2004), S5587–S5600.
[24] K.E.H.M. Hanssen, P.E. Mijnarends, L.P.L.M. Rabou, K.H.J. Buschow, “Positron-annihilation
study of the half-metallic ferromagnet nickel manganese antimonide (NiMnSb): Experiment”
Phys. Rev. B 42 (1990), 1533–1540.
[25] A.F. Andreev, “Thermal conductivity of the intermediate state of superconductors” Zh. Eksp.
Teor. Fiz. 46 (1964), 1823–1828. (Sov. Phys. JETP 19 (1964), 1228).
 11.7 References 485

[26] R.J. Soulen, Jr., J.M. Byers, M.S. Osofsky, B. Nadgorny, T. Ambrose, S.F. Cheng, P.R.
Broussard, C.T. Tanaka, J. Nowak, J.S. Moodera, A. Barry, J.M.D. Coey, “Measuring the
spin polarization of a metal with a superconducting point contact” Science 282 (1998), 85–88.
[27] K. Schwarz, “Chromium dioxide predicted as a half-metallic ferromagnet” J. Phys. F: Met. Phys.
16 (1986), L211–L215.
[28] M.A. Korotin, V.I. Anisimov, D.I. Khomskii, G.A. Sawatzky, “CrO2: A self-doped double
exchange ferromagnet” Phys. Rev. Lett. 80 (1998), 4305–4308.
[29] J.K. Burdett, G.J. Miller, J.W. Richardson, Jr., J.V. Smith, “Low-temperature neutron powder
diffraction study of chromium dioxide and the validity of the Jahn–Teller viewpoint” J. Am.
Chem. Soc. 110 (1988), 8064–8071.
[30] S. Sundar Manoharan, D. Elefant, G. Reiss, J.B. Goodenough, “Extrinsic giant magnetoresist-
ance in chromium(IV) oxide, CrO2” Appl. Phys. Lett. 72 (1998), 984–986.
[31] I. Galanakis, P.H. Dederichs, “Half-metallicity and Slater–Pauling behavior in the ferromagnetic
Heusler alloys” Lecture Notes in Physics 676 (2005), 1–39.
[32] L. Pauling, “The nature of the interatomic forces in metals” Phys. Rev. 54 (1938), 899–904.
[33] S. Ouardi, G.H. Fecher, C. Felser, “Realization of spin gapless semiconductors: The Heusler
compound Mn2CoAl” Phys. Rev. Lett. 110 (2013), 100401/1–5.
[34] T. Graf, C. Felser, S.S.P. Parkin, “Simple rules for the understanding of Heusler compounds”
Prog. Solid State Chem. 39 (2011), 1–50.
[35] J. Lindén, T. Yamamoto, M. Karppinen, H. Yamauchi, T. Pietari, “Evidence for valence
fluctuation of Fe in Sr2FeMoO6−w double perovskite” Appl. Phys. Lett. 76 (2000), 2925–2927.
[36] K.-I. Kobayashi, T. Kimura, H. Sawada, K. Terakura, Y. Tokura, “Room-temperature mag-
netoresistance in an oxide material with an ordered double-perovskite structure” Nature 395
(1998), 677–680.
[37] D. Sánchez-Soria, J.A. Alonso, M. García-Hernández, M.J. Martínez-Lope, J.L. Martínez, A.
Mellergård, “Neutron-diffraction magnetic scattering in ordered and disordered Sr2FeMoO6”
Appl. Phys. A 74 (2002), S1752–S1754.
[38] M. Kallmayer, P. Klaer, H. Schneider, E. Arbelo Jorge, C. Herbort, G. Jakob, M. Jourdan, H.J.
Elmers , “Spin-resolved unoccupied density of states in epitaxial Heusler-alloy films” Phys. Rev. B
80 (2009), 020406/1–4.
12 Superconductivity

Superconductivity is the phenomenon whereby a significant number of elements and many

compounds can conduct electricity with zero resistance below a critical temperature, Tc,
field, Hc and current, Jc. There are many technological applications for materials with this
remarkable property and therefore a large global research and development effort in the
area; around 7000 original research articles are published on superconductivity every year.
In this chapter, we will look at the history of superconductivity and its physical origins in
so-called BCS or conventional systems. We’ll then focus on the solid state chemistry of five
distinct families of superconducting materials: A3C60 alkali-metal intercalates, molecular
superconductors, Ba(Pb,Bi)O3 perovskites, the cuprate- or “high-Tc” superconductors,
and the LaOFeAs-related “iron” superconductors. These families will highlight several
recurrent themes and show how chemistry is used to prepare and tune superconducting
materials.

12.1 Overview of Superconductivity

The discovery of superconductivity is a wonderful example of how “blue skies” research can lead
to completely unexpected discoveries. In 1908, Heike Kamerlingh Onnes, a Dutch physicist
working at the University of Leiden in the Netherlands, succeeded in liquefying helium. Access to
liquid He, which boils at 4.22 K, allowed him to perform physical measurements on materials at
much lower temperatures than previously possible. At that time, little was known about what
would happen to the electrical resistance (R) of metals at very low temperatures. There were three
basic possibilities (Figure 12.1a): (1) Would the known decrease in R with T seen at higher
temperatures continue such that a metal had zero resistance at low temperature? (2) Would R
flatten off such that it approached a finite value? (3) Or would R rise to infinity at low T as mobile
conduction electrons “froze” and became bound to individual metal atoms rather than delocal-
ized? Initial measurements on Au and Pt suggested hypothesis (2), but experiments on Hg (Figure
12.1b) showed a very dramatic drop in R over just a few hundredths of a degree to extremely low

486
 12.1 Overview of Superconductivity 487

0.150
(a) (b)
0.125

0.100

resistance (Ω)
resistance

0.075
(3)
0.050

(2) 0.025

(1)
0.000
0 20 40 60 4.00 4.20 4.40
temperature (K) temperature (K)

Figure 12.1 (a) Early postulated models for how a metal’s resistance might vary with temperature as
discussed in the text. (b) Onnes’ original data on the resistance of Hg metal; below a critical temperature
Tc of 4.2 K (the best modern value is 4.153 K), resistance falls to ~0 Ω.

values. Onnes’ report [1] on a sample with a resistance of 172.7 Ω as a liquid at 0 °C stated that “at
4.3 K [resistance] had sunk to 0.084 Ω . . . at 3 K the resistance was found to have fallen below
3×10−6 Ω, that is to one ten-millionth of the value it would have [in the solid state] at 0 °C.
Mercury has passed into . . . the superconductive state.”
Superconductivity below a critical temperature Tc has since been identified as the ground state
of around a quarter of the elements at ambient pressure and more than half at ambient and
higher pressure or as thin films1, as summarized in Figure 12.2. The element with the lowest
discovered Tc at ambient pressure is Rh (0.000325 K); that with the highest Nb (9.25 K). Li
superconducts at 0.004 K at ambient pressure yet has one of the highest elemental Tc values
under pressure (20 K at ~30 GPa). Eu is the most recently discovered superconducting element
(2009), with a Tc of 1.8 K at 80 GPa. For our later discussions, there are two interesting points to
take from Figure 12.2. Firstly, none of the naturally magnetic elements (Fe, Co, Ni, Gd are
ferromagnetic; Mn antiferromagnetic) superconduct at ambient pressure. Secondly, the
elements with the highest normal metallic conductivity (Cu, Ag, Au have conductivities of
around 6×107 S/m at room temperature, Chapter 10) do not superconduct.
In addition to elements, many compounds have been shown to superconduct, including
some that are magnetically ordered. Until 1986, the alloy Nb3Ge had the highest Tc known
(23.2 K). Superconductivity was then discovered in a large family of copper-containing
oxides at temperatures up to 138 K (~164 K at high pressure), which held the record for

1
Thin films of elements and compounds can be significantly strained relative to the bulk. This influences local
coordination geometries and thereby electronic properties.
488 Superconductivity

H He

Li Be B C N O F Ne

Na Mg Al Si P S Cl Ar

K Ca Sc Ti V Cr Mn Fe Co Ni Cu Zn Ga Ge As Se Br Kr

Rb Sr Y Zr Nb Mo Tc Ru Rh Pd Ag Cd In Sn Sb Te I Xe

Cs Ba La Hf Ta W Re Os Ir Pt Au Hg Tl Pb Bi Po At Rn

Fr Ra Ac

Ce Pr Nd Pm Sm Eu Gd Tb Dy Ho Er Tm Yb Lu

Th Pa U Np Pu Am Cm Bk Cf Es Fm Md No Lr

Figure 12.2 Periodic table of superconducting elements. Those in gray boxes will superconduct at
ambient pressure in bulk form. Elements in black boxes have been shown to superconduct under pressure
or as thin films.

the highest well-established value of Tc for many years. The Nb3Ge Tc barrier has also been
surpassed by A3C60 compounds, MgB2, some nitride halide intercalates, and by a family of
iron-containing materials related to LaOFeAs. A prediction in 1968 by Neil W. Ashcroft [2]
that metallic hydrogen, if formed at high pressures, would be a high-temperature supercon-
ductor, is behind another pathway to high Tc—high-pressure accommodation of super-
stoichiometric hydrogen around pinning points of suitable central atoms in a crystal
structure. In 2015, researchers reported Tc > 200 K in samples of H2S held at very high
pressure (forming H3S), and in 2018 a Tc ≈ 270 K was reported in a superstoichiometric
lanthanum hydride.2 Table 12.1 gives a selection of superconducting materials chosen either
because they led to historically significant breakthroughs, or because they are representative
members of a larger family of materials. The Holy Grail in this area is, of course, ambient-
pressure superconductivity at room temperature or above. There are periodic reports of
higher-Tc materials than those in Table 12.1, but they prove hard or impossible to reproduce.
These are often semi-jokingly referred to as USOs—unidentified superconducting objects.

12.2 Properties of Superconductors

We’ve stated above that superconductors have zero resistance below the critical temperature
Tc. It’s important to consider initially whether there is anything intrinsically special about the
2
Tc = 203 K superconductivity has been reported in samples of H2S pressurized to ~150 GPa [A.P. Drozdov et al.,
Nature 525 (2015), 73–76]. The superconducting phase has been suggested to be H3S (created by decomposition of
3H2S to 2H3S + S) or [H3S]+[SH]−. Various high-pressure metal-poor hydrides were predicted to have Tc close to or
above room temperature, such as LaH10 (~280 K at >200 GPa) [H. Liu et al., PNAS 114 (2017), 6990–6995], the first
experimental support appeared in 2018.
 12.2 Properties of Superconductors 489

Table 12.1 Selected superconductors and their critical temperatures, organized partly chronologically
and partly by chemical type. TMTSF and ET are defined and drawn in Figure 12.12. Tc values are
collated from references in [3].

Material Tc (K) Comment

Rh 0.0003 Lowest Tc of any superconducting element

Hg 4.153 First superconductor discovered in 1911
Pb 7.2 Highest-Tc type-I superconducting element
Nb 9.25 Highest-Tc type-II superconducting element
Li 0.004/20 Ambient/~30 GPa
Ca 21–25 Highest-Tc element ~220 GPa
Nb3Sn 18 High-Tc intermetallic
Nb3Ge 23.2 Highest-Tc intermetallic
NbO 1.5 First oxide superconductor, 1933
SrTiO3−δ 0.3 First perovskite superconductor
BaPb1−xBixO3 13 First high-temperature superconductor of 1970s
Ba0.6K0.4BiO3 34 Highest Tc of BaBiO3-related systems
La1.85Ba0.15CuO4 30 First cuprate superconductor, 1986
La1.85Sr0.15CuO4 or La2CuO4.08 38 Highest-Tc “214” cuprate superconductor
YBa2Cu3O7 93 First >77 K (N2 boiling point) superconductor, 1987
HgBa2Ca2Cu3O8+x 134/164 Highest-Tc cuprate at ambient/elevated pressure
Ba1−xSrxCuO2 90 Infinite-layer cuprate
Ca1−xSrxCuO2 110 Highest-Tc ternary infinite-layer cuprate
Nd2−xCexCuO4 ~25 n-type superconducting cuprate
Sr1−xNdxCuO2 40 n-type superconducting infinite-layer compound
RuSr2(Gd,Eu,Sm)Cu2O8 ~58 Ferromagnetic cuprate superconductor
Nd0.8Sr0.2NiO2 9–15 d 9 nickel oxide potential cuprate analogue
Sr2RuO4 1.5 Non-cuprate "214" superconductor
Nax(H2O)yCoO2 ~5 Layered cobalt oxide
LaO1−xFxFeAs (x ≈ 0.12) 26 First superconducting iron oxide-pnictide
SmO0.9F0.1FeAs 55 Highest-Tc oxide pnictide
Fe1+xSe 37 Under pressure
PbMo6S8 15 Chevrel phase
NbN 17 Widely used nitride superconductor
CaC6 11.5 Graphite intercalate
K3C60 18 C60 intercalate
Cs3C60 38 Highest-Tc molecular system under pressure
Lix(THF)yHfNCl 25.5 Intercalate; highest-Tc nitride based superconductor
MgB2 39 Highest-Tc conventional superconductor
YPd2B2 C 23 Highest-Tc borocarbide
MgCNi3 7–8 A non-oxide anti-perovskite
CeCu2Si2 1 First heavy-fermion superconductor
490 Superconductivity

Table 12.1 (cont.)

Material Tc (K) Comment

UPd2Al3 2 Antiferromagnetic heavy-fermion superconductor
PuCoGa5 18 High-Tc heavy-fermion superconductor
(SN)x 0.33 Inorganic polymer
(TMTSF)2PF6 1 First organic superconductor at 12 kbar
ET2Cu[N(CN)2]Br 11.6 Charge-transfer salt
H2S 203 At 150 GPa

superconducting state, or whether superconductors merely represent extreme examples of

metallic behavior. If we compare resistivities for metals and superconductors, we find an
excellent low-temperature metallic conductor like Ag has ρ ≈ 10−11 Ω m at 1 K, whereas the
best estimates for superconductors are around 10−25 Ω m. The resistivity of superconductors
is lower than even the best metal by many orders of magnitude, suggesting a qualitatively
different electronic state.
Superconductors also display remarkable magnetic properties. When a superconductor is
placed in a magnetic field, it displays the Meissner or Meissner–Ochsenfeld effect—it expels
the magnetic field so that the magnetic induction B within the sample is zero (Figure 12.3).3
From our Chapter 9 definitions of magnetic induction as B = μ0(H + M) and magnetic
susceptibility as χ = M/H, it follows that, for a superconductor, μ0(H + χH) = 0 or χ = −1.4
A superconductor is thus a perfect diamagnet. However, the total field exclusion occurs only
for certain sample geometries and below a critical value of the applied field. The exclusion
arises because the supercurrents flowing in a thin layer at the surface of the superconductor
generate a magnetic field that exactly opposes the external field.
This magnetic behavior is fundamentally different to that of a hypothetical “perfect
metallic” conductor with resistivity ρ = 0. If one took either a superconductor below Tc or
a perfect metal with ρ = 0 and applied a magnetic field, the magnetic field within both samples
would be zero. If, however, one started with both samples in a magnetic field at temperatures
where they had a finite resistance and then cooled them such that their resistivity fell to zero,
the two samples would show different behavior. The superconductor would expel the
magnetic field, whereas the perfect conductor would trap the field.
Superconductors only display the Meissner effect below certain applied field strengths. In
a type-I superconductor, such as elemental Pb, perfect diamagnetism with χ = −1 (i.e.
M = −H) is only observed up to a critical field Hc. Above Hc, the sample reverts to a normal
conducting state, and χ falls to the much smaller value of −1.8×10−5, which is typical for a
normal metal (Figure 12.3a). The magnitude of Hc is temperature dependent, being largest

3
This is the origin of the famous magnet levitation demonstration using a superconductor.
4
H is the magnetic-field strength, M is the magnetization, and μ0 is the permeability of free space; see Section 9.2.2 for
full definitions.
 12.2 Properties of Superconductors 491

type I type II
M (a) M (c)

Hc Hc1 Hc2
H H

super
conductor

H (b)
H
(d)
Hc2
Hc
normal mixed normal
metal Hc1 metal

super- super-
M = magnetization
conductor conductor
H = magnetic-field strength
Tc T Tc T

Figure 12.3 Magnetization at T < Tc and schematic phase diagrams of superconductors in an applied
magnetic field.

near T = 0 K and falling to 0 at T = Tc. This leads to the behavior shown in Figure 12.3b. An
internal field created by current flow in a superconductor wire can also destroy supercon-
ductivity, leading to the related concept of a critical current, Jc.5
Other materials exhibit so-called type-II superconductor behavior (Figure 12.3c and d), in
which they show perfect diamagnetism up to a critical field Hc1 followed by a gradual
decrease of the absolute value of the magnetization up to a critical field Hc2. Above Hc2,
normal magnetic behavior is observed. Critical fields Hc2 are typically much higher than Hc
values for type-I superconductors. As an example, while pure Pb is a type-I superconductor
with Tc = 7.2 K and Hc ≈ 550 G at 4.2 K, alloying with In yields a type-II superconductor; at
20% In, perfect diamagnetism is only observed up to Hc1 of ~150 G but Hc2 is ~3700 G.
Values of Hc2 up to 110 T (1.1×106 G) have been reported in nanocrystalline PbMo6S8 [4].
In the range between Hc1 and Hc2, a type-II superconductor exists in a mixed or vortex state
where certain regions of the sample superconduct and expel the magnetic field, while other
regions don’t. The normal region is separated from the rest of the superconductor by a
circulating supercurrent called a vortex. There is a magnetic repulsion between the mag-
netic-field “tubes” that penetrate the sample, which then arrange themselves to maximize their
separation. This often leads to a regular hexagonal array that can be imaged (Figure 12.4) by
placing fine magnetic particles on the specimen (“decoration”) or by surface-sensitive tech-
niques such as scanning tunneling microcopy (STM) and magnetic force microscopy. When a

5
A high critical current is particularly important in superconducting cables.
492 Superconductivity

Pb0.96In0.04 T = 1.1 K NbSe2 T = 1.8 K

1 μm 1 μm

Figure 12.4 The vortex state of a superconductor. Left: The first such image [U. Essmann, H. Trauble,
Physics Letters A 24 (1967), 526] obtained using decoration methods. Right: STM image [H.F. Hess,
Phys. Rev. Lett. 62 (1989), 214]. In both images the lighter region is superconducting.

supercurrent flows in a vortex-state material, it causes the vortices to move in a direction

perpendicular to both the current and applied field (this can again be imaged). The movement
of the normally conducting magnetic-field tubes induces current flow within them, which
dissipates energy and destroys superconductivity. If, however, the vortices can be spatially
trapped or pinned6 (e.g. by lattice defects in the sample), the sample will transport current with
zero resistance via the fixed superconducting regions. Note that there is no structural or
chemical difference between the normal and superconducting regions in the vortex state.

12.3 Origins of Superconductivity and BCS Theory

How does superconductivity arise? In this section, we will discuss theoretical models of the
so-called conventional superconductors. Some texts that discuss the theory of unconventional
superconductors are given in Further Reading.
We’ve already encountered several clues that will be important in understanding the origin of
superconductivity. Firstly, we’ve seen that there is an inverse relationship between normal
metallic conductivity and superconductivity—the best metals don’t superconduct. Secondly,
internal or external magnetic fields normally disrupt superconductivity.7 We can gain more

6
Pinning is also behind the fascinating levitation properties of type-II superconductors. If you cool a small, flat,
superconductor disc with a small, very strong, magnet on its flat surface, a pinned vortex state will trap the magnetic
field in specific places in the superconductor, and, below Tc, the magnet will start “hovering” there without being
held. This is the Meissner effect. When you then gradually raise the magnet with bamboo tweezers, if the supercon-
ductor is light enough and has enough pinned vortices, it will hang below the magnet. This is not the Meissner effect,
but an attraction due to the magnetized pinned vortex state. If the magnet is removed and then placed back on the
superconductor, the “magnetic memory” of its original position stored in the vortex state returns it to its original
position. With the magnet removed, the still-cooled superconductor will attract ferrous objects for the same reason.
7
There are exceptions. For example, in alloys such as UGe2 [Saxena et al., Nature 406 (2000), 587–592], URhGe [Aoki et
al., Nature 413 (2001), 613–616] and UIr [Akazawa et al., J. Phys. Condens. Matter 16 (2004), L29–32], superconductivity
can coexist with itinerant ferromagnetism. These are examples of unconventional superconductors.
 12.3 Origins of Superconductivity and BCS Theory 493

4 2
(a) (b) (c)

heat capacity (mJ mol-1K-1)

0.0 normal
free energy (mJ mol-1)

entropy (mJ mol-1K-1)

3 superconducting

-0.5 normal
2 1
superconducting
-1.0 super
1 normal
conducting
Tc Tc Tc
-1.5 0 0
0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5
temperature (K) temperature (K) temperature (K)

Figure 12.5 (a) Free energy, (b) heat capacity, and (c) entropy of Al in the superconducting and normal
states of aluminum. Parts (a) and (c) after [5].

insight by considering the thermodynamics of the transition to the superconducting state. Figure
12.5 shows the temperature dependence of the free energy, heat capacity, and entropy of
aluminum in the normal and superconducting state. These data were obtained by performing
the measurements initially in a zero magnetic field and then repeating them in a field above Hc to
suppress the superconducting transition. The free energy of the superconducting state is lower
than that of the normal state at all temperatures below Tc (1.2 K), though by just 0.4 mJ/mol at
0 K. At Tc, the heat capacity shows a marked change, which, along with the lack of a
discontinuity in the free energy, suggests a second-order transition (see Section 4.4.4). In addition,
the electronic contribution to the specific heat at low temperature shows an eaTc=T -like depend-
ence, which suggests an energy gap is present with magnitude ~kTc; we’ll discuss this gap
later. Finally, the entropy is lower in the superconducting state, which suggests that the
electrons must adopt a more ordered configuration.
Since there are no changes in a sample’s cell parameter or structure at Tc, the thermo-
dynamic data of Figure 12.5 led to the idea that there must be attractive electron–electron
interactions in the superconducting state, which lower the free energy of the system despite
the loss of entropy. How do these attractions occur? While it was originally thought that the
structure of the sample played little role, in the 1950s it was discovered that the Tc values of
superconducting elements exhibit an isotope effect whereby Tc ∝ M−α, where M is the
isotope mass. Isotopes of Hg, for example, have Tc values ranging from 4.161 K to
4.126 K as the mean atomic weight varies from 199.7 to 203.4 and α = 0.49(2) [6]. This
dependence mirrors the dependence of vibrational frequencies on mass,8 and suggests that
lattice vibrations, or phonons (Section 4.4.6), give rise to the electron–electron attractions.
The simplest model for the electron–phonon (or Fröhlich) interaction is shown in Figure
12.6a. As an electron (e1) moves right to left through a metal, its negative charge will attract
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
8
A simple diatomic molecule, for example, has vibrational frequency (in Hz) of ð1=2πÞ kbond =μ, where μ is the
reduced mass of the atoms involved (1/μ = 1/m1 + 1/m2) in kg and kbond the force constant in N/m.
494 Superconductivity

+ + + + k2
e2
k2 + q = k2 '
+ +
k1 ' e2
+ +

e1
e1 q k2 '
+ +
+ +
k1 = k1 '+ q
k1
+ + + +
k1 + k2 = k1 '+k2 ' = k0
(a) (b)

Figure 12.6 (a) Real-space illustration of how electron–phonon coupling in a simple metal system can
lead to attractive interaction between two electrons.9 (b) Virtual-phonon exchange behind Cooper-pair
formation.

the positive metal ions around it, creating a temporary deformation of the structure and a
local region with excess positive charge. This moving region of positive charge, or polariza-
tion trail, will then attract a second electron, forming a Cooper pair. We therefore have a
situation in which two electrons attract to form a pair not via a direct interaction but via a
first-electron-to-lattice, lattice-to-second-electron process10 of phonon-mediated pairing.
Since electrons move much more rapidly than the heavier metal ions, there will be a time
delay between the motion of the first electron that causes the lattice distortion, and the
subsequent attraction of the second electron. To estimate the distance, d, the first electron
has passed by the time the second is attracted, we need to know vF (velocity of an electron at
the Fermi level) and the Debye angular frequency, ωD (a measure of the frequency at which
the lattice can respond):

2π
d ¼ vF (12.1)
ωD

For aluminum, vF ≈ 2×106 m/s (see Section 10.2.6) and ωD ≈ 5×1013 s−1, which gives d ≈
2500 Å; i.e. the distance between electrons in pairs is large. At this distance, there will
be effectively no Coulomb repulsion between the electrons to counteract the attraction,
particularly given the screening by the positive charges on the metal ions.
A more precise way to describe this interaction is in terms of the exchange of a virtual
phonon between the electrons. In this process, we imagine electron 1 with an initial wave
vector,11 k1, emitting a phonon with wave vector, q, after which its wave vector becomes k10
(Figure 12.6b). A second electron absorbs this phonon and is scattered from wave vector k2
to k20 . Conservation of momentum leads to the relationships:
9
Note that in BCS superconductors the distance between e1 and e2 is sufficiently large that this local sketch is a gross
approximation; the distance is much smaller in the bismuth oxide and cuprate superconductors discussed later.
10
A common analogy is placing heavy bowling balls on a sprung mattress. One ball creates a depression that the
second ball will roll into. The balls are attracted to each other by an indirect first-ball-to-mattress, mattress-to-
second-ball interaction.
11
Remember from Chapter 6 that the wave vector just labels one of the crystal orbitals or an allowed electron state.
 12.3 Origins of Superconductivity and BCS Theory 495

k1 ¼ k01 þ q and k2 þ q ¼ k02 (12.2)

Since momentum is also conserved for the overall process, Equation (12.2) leads to the
relationship:

k1 þ k2 ¼ k01 þ k02 ¼ k0 (12.3)

The phonon is said to be virtual because it is reabsorbed during the process, and no net
vibrational energy is transferred to the lattice to produce electrical resistance.
Is there any special relationship between the two electrons that will maximize the
overall attractions (i.e. is there a special value of k0)? We can best explore this
in k space using the schematic of Figure 12.7. If the attraction is mediated by a
phonon of energy ℏωD , then the Pauli exclusion principle means that only electron states
within ℏωD of EF can be involved in the coupling,12 corresponding to a shell of finite
width δk in k space. The states involved can be represented graphically, as shown in
Figure 12.7b, and their number is maximized, giving the largest number of electron–electron
attractions, when k0 = 0. Equation (12.3) then means that we have the condition on
the electrons that k1 = −k2. If the two-electron wavefunction of the Cooper pair is
spherically symmetric,13 quantum mechanics requires a singlet spin state with S = 0, leading
to the overall condition that the electrons in the Cooper pair have opposite momenta
and spin: k↑ and −k↓. This is called s-wave pairing and occurs in conventional or BCS (see
below) superconductors.14 Note that the situation is a complex and dynamic one: Cooper
pairs in a superconductor shouldn’t be thought of in the same way as electron pairs in a
chemical bond, but as in a dynamic state in which individual electrons are constantly
exchanging phonons.
From the idea of Cooper pairs, John Bardeen, Leon Cooper, and John Schrieffer were
able to build the first quantitative model for superconductivity, BCS theory, for which
they were awarded the 1972 Nobel Prize in Physics. Their theory showed that, at T < Tc,
many electrons are involved in Cooper pairs and that it is the cooperative properties of
the ensemble of pairs that determines overall properties.15 For example, the pair binding
energy (usually called 2Δ), depends on how many other pairs are present. As the Cooper
pairs combine two fermions (the electrons), they behave as a boson and will all have the

12
Electrons must transition to empty states; at T = 0 K all states < EF are full.
13
This is the case for conventional superconductors but not for unconventional superconductors such as cuprates or
heavy-fermion materials.
14
A d-wave pairing, where the two-electron wavefunction has nodes like those in d orbitals, also requires S = 0 and is
thought to occur in cuprates. Spin triplet states (S = 1, see Section 7.3.2) are also possible but associated with p- or f-
wave pairing. These have been proposed in heavy-fermion superconductors such as Sr2RuO4 and UPt3. See, e.g., the
text by Norman in Further Reading for more detail.
15
The cooperative behavior is perhaps unsurprising given that the radius of a Cooper pair is ~10−7–10−6 m and that
~106–107 other electrons pairs will be present in the corresponding volume.
496 Superconductivity

ky k0 reduces ky

δk
k1 k2
k1 k1 k2
kF
kx k0 kx
k2
EF ℏ D
k2 k1
EF
EF  ℏ D

(a) (b) (c)

Figure 12.7 (a) A k-space sketch of a spherical shell of states within δk of Fermi level kF. Electrons with
arbitrary wave vectors k1 and k2 are shown. The change in k1 on phonon exchange is restricted to a small
area determined by δk. (b) The overall condition of k1 + k2 = k0 for virtual phonon exchange can be
represented geometrically by two such shells offset by k0. For a given value of k0 the overlap region
(shaded) gives the states that can be involved in coupling. The maximum coupling will occur when the two
shells overlap such that k0 = 0 and k1 = −k2.

same quantum state at T = 0 K; i.e., they have undergone a transition similar16 to a Bose–
Einstein condensation.17
One of the important features of the BCS ground state is that its energy is lower
than that of the normal free-electron state (recall Figure 12.5a). In particular, an
energy gap, the superconducting gap, opens up between the superconducting state and
the normal free-electron state.18 We can understand this gap with reference to Figure
12.8. The exchange of virtual phonons between electrons close to EF lowers the
electrons’ energy, which can be represented as a peaking in the density-of-states
(DOS) plot in Figure 12.8. Note that this figure represents the single electron states
and does not specifically show Cooper pairs. The typical energy, ℏωD , of a phonon
giving rise to coupling is around 10−3 eV, whereas EF is typically ~5 eV. This means that only
a small fraction, around 10−3, of the conducting electrons have their energy changed and
explains why the energy difference between the normal and superconducting states is small.
At T = 0 K, levels above the gap are empty. As T rises, electrons are excited across the gap to
normal states. Since this destroys some Cooper pairs, the gap decreases but the sample
remains superconducting. At T = Tc, the gap reduces to zero and the sample becomes a
normal metal.

16
“Similar” as Cooper pairs only exist in the overall cooperative state of a superconductor, the superconducting state
wouldn’t arise from a gas of independent electrons.
17
A Bose–Einstein condensate is a state that forms when a very dilute gas of bosons is cooled to sufficiently low
(nanokelvin) temperatures that a large fraction of them adopt the same quantum state and quantum-mechanically
derived behavior becomes observable on a macroscopic scale. The 2011 Nobel Prize in Physics was awarded to
Cornell, Ketterle, and Wiemann, who first formed Bose–Einstein condensates in Rb and Na gases.
18
More exotic gapless superconductors are also possible, in which Cooper pairs exist without a gap.
 12.3 Origins of Superconductivity and BCS Theory 497

T > Tc T=0K 0 < T < Tc

E
E E E

Δ0
N(E) ~5 eV EF ~10−3 eV

N(E) N(E) N(E)

Figure 12.8 DOS plot of a metal close to EF in its normal and superconducting states. In the supercon-
ducting state, peaks appear in N(EF) just below and above the gap. Away from EF, the DOS of the
superconductor is similar to a normal metal. At T > 0 K, there is some occupancy of normal states above
the gap.

For an isotropic system with a 0 K gap defined as 2Δ0,19 BCS theory predicts that:

ℏωD 0
D0 ¼ ≈ 2ℏωD e1=NðEF ÞV (12.4)
1
sinh
NðEF ÞV 0

where N(EF) is the DOS at the Fermi level and V 0 the strength of the electron–lattice-
vibration interaction (electron–phonon coupling).20 Values for typical materials predict
Δ0 ≈ 5×10−4 eV, which matches experimental measurements of the gap obtained by infrared
radiation adsorption. The theory predicts that the gap decreases continuously with increas-
ing temperature, becoming 0 at Tc. The Tc is related to Δ0 by:

2D0 ¼ 3:52kTc (12.5)

and we see why conventional superconductivity only happens at low temperature.

Combining Equations (12.4) and (12.5) gives:

ℏωD 1=NðEF ÞV 0

Tc ¼ 1:14 e (12.6)
k

The form of Equation (12.6) helps us to rationalize some of the experimental

observations on superconductivity. Firstly, as the Debye frequency, ωD , will scale
with M−1/2 (Footnote 8 in this chapter), Tc will show the same dependence. This explains
the experimentally observed isotope effect. We can also see why light-element-containing

19
At T = 0 K, a BCS superconductor has Cooper pairs in the same quantum state. The first single-electron state is
higher in energy by Δ0 such that energy 2Δ0 must be supplied to break a pair.
20
The approximation only holds for weak coupling N(EF)·V′ ≪ 1.
498 Superconductivity

compounds such as MgB2, where high Debye frequencies are expected, can have high
Tc values.21 Secondly, a high value of the electron–lattice-vibration interaction V 0 will
lead to a high value of Tc. This helps explain why “good” normal metallic conductors
(Cu, Ag, Au) make bad superconductors, yet “bad” normal conductors (such as oxides)
make good superconductors—the same electron–phonon interactions that give rise to
resistivity in a normal metal give rise to the Cooper pairs required for superconductiv-
ity. Finally, Δ0 and Tc can be seen to depend on N(EF). When pressure is applied to a
BCS superconductor, it broadens the bands (Chapter 6) and lowers N(EF), and we
would therefore expect Tc to fall. This has been observed for many superconductors
and will be discussed in Section 12.4. Note that this behavior only applies for conven-
tional BCS superconductors, and the effect of pressure on unconventional non-BCS
systems can be very different.
A superconductor’s signature property of zero resistance is a consequence of the
cooperative nature of the ensemble of Cooper pairs, and stems from the fact that they
are all in the same quantum state. In Chapter 10, we described conductivity as arising
when individual electrons have their momenta shifted by the application of an electric
field from k to k+Δk, and resistivity as arising due to defects and phonons scattering
electrons back into lower momentum states. Why doesn’t this occur in superconduct-
ors? When there is zero current in a superconductor, the net momentum of each
Cooper pair is zero (k↑ and −k↓). When they are accelerated by an electric field, the
entire Fermi surface is shifted and all Cooper pairs gain momentum (changing to
k+Δk↑ and −k+Δk↓). Since the pairs have coupled momenta, the only way to change
the overall momentum is to break up a Cooper pair by excitation across the super-
conducting gap, and this requires an energy of 2Δ. An elastic process such as impurity
scattering clearly won’t be able to do this and can’t give rise to resistivity. What
about inelastic processes such as phonon absorption or emission? As we’ve discussed,
these processes happen continuously in a superconductor at 0 < T < Tc and create a
dynamic equilibrium between normal electrons and Cooper pairs. However, the
Cooper pairs that form in these processes must have the same wavefunction as the
pre-existing pairs and hence the same momentum. This means that their creation and
destruction won’t cause any change in the current. The overall current is therefore
only affected by factors that influence all Cooper pairs in the same way, such as
changing the applied electric field. We can also understand the origin of the super-
conducting critical current Jc from the energy gap. When the applied electric field
increases the pair’s kinetic energy to be comparable with its binding energy, Cooper
pairs are gradually destroyed until resistance re-emerges.
As a final point, we should consider which electrons are involved in carrying the
supercurrent. Is it just the small fraction of electrons close to EF involved in Cooper
pairs, such that we can think of a normal current and supercurrent being present

21
Note also the high Tc of Li (20 K) and H2S-derived compositions (203 K) under pressure (Table 12.1).
 12.3 Origins of Superconductivity and BCS Theory 499

simultaneously? It turns out that the situation is more complex. At 0 K, there are no
excitations possible across the superconducting gap, and all the electrons behave as if
they’re carrying the supercurrent. At higher temperatures, where excitations are
possible, the number of electrons involved in the supercurrent falls, reaching 0 at
Tc. The presence of at least some electrons carrying the supercurrent is sufficient to
give zero resistance.

Box 12.1 Materials Spotlight: A serendipitous superconductor MgB2

In 1954, the synthesis and crystal structure of MgB2 were reported in the Journal of the
American Chemical Society [7]. Other researchers cited the paper just 12 times in the
following 10 years, once in the 1970s and once in the 1990s. Almost 50 years later,
Akimitsu set out to dope the ferromagnetic semiconductor CaB6 with Mg. When the
magnetic and electronic properties of the MgB2 starting material were tested, it was
found to superconduct at a remarkable 39 K. The discovery was reported in a wonder-
fully succinct paper in Nature in February 2001 [8], which was cited almost 300 times by
the end of the calendar year.
The structure of MgB2 (see below, left) is remarkably simple, containing Mg sandwiched
between graphene-like hexagonal B layers. It’s also strikingly similar to the graphite intercal-
ation compounds we’ll meet in Chapter 13. This structural relationship is unsurprising since
electronic calculations show the material is essentially ionic, Mg2+B22−, making the B layers
isoelectronic with graphite. Their band structure is also similar to that of graphene (Chapter 6),
with one major difference; both the 2D sp2-derived σ band and the more three dimensional pz-
derived π band cross the Fermi level. This gives rise to holes in the σ band (it would be full in
graphite) and a corresponding number of electrons in the π band, creating a σ ≈ 105 Ω−1 cm−1
metallic conductivity at room temperature. Hall-effect measurements have shown the presence
of both carriers.

0.06

0.05

Mg
0.04
ρ(T)/ρ(300 K)

B 0.03
Mg11B2 Mg10B2
0.02

0.01

0
c
-0.01
37 38 39 40 41 42 43
a b
temperature (K)

Despite the high Tc, MgB2 appears to be a conventional s-wave superconductor. It shows a
strong isotope effect, with Tc shifts of ~1 K on substitution of 10B for 11B (see right-hand side of
the figure [9]) and ~0.1 K on Mg substitution. The α values in Tc ∝ M−α are 0.26–0.30 and 0.02,
500 Superconductivity

Box 12.1 (cont.)

respectively, suggesting that phonons involving B are particularly important. The strong B–B
bonding gives a stiff material with high-frequency phonons (ωD ≈ 3300 cm−1), and there is
strong electron–lattice-vibration coupling (V 0 = 0.82). Equation (12.6) shows these factors will
lead to a high Tc. One peculiar feature of MgB2 is that the phonons couple differently with
electrons in the σ and π bands, which leads to two distinct superconducting gaps (around 7.1
meV and 2.3 meV). These have been observed by a number of experimental techniques, most
strikingly heat-capacity measurements that show a superposition of two distinct features of the
type plotted for Al in Figure 12.5b.
After the discovery of superconductivity in pure MgB2, many attempts were made at
chemical doping to enhance Tc. To date only three substitutions are possible: Al and Mn for
Mg and C for B. In each case, there is a decrease in Tc. Al and C are both electron dopants and
the band structure shows a decrease in N(EF) on adding electrons, while the magnetic moment
of Mn2+ is predicted to destroy Cooper pairs. Despite reducing Tc, carbon doping raises the
T = 0 K critical field from Hc2 ≈ 16 T in MgB2 to Hc2 ≈ 36 T in Mg(B0.95C0.05)2. This high Hc2 in
a light and relatively cheap material has meant that MgB2 is already being used in MRI
magnets.
It’s remarkable that an optimally doped conventional superconductor with the highest Tc of
any binary compound remained overlooked for so long!

12.4 C60-Derived Superconductors

In this section, we will discuss the superconducting alkali- and alkaline-earth AxC60
intercalates. These have been shown to superconduct up to a maximum Tc of 38 K,
and currently hold the Tc record for ambient-pressure molecular systems. While the
full detail of the electronic properties of these materials is complex [10], they provide
an illustrative example of how superconducting properties can be chemically
controlled.
C60 itself is a truncated icosahedron with point group Ih. The 60 carbon atoms form 20
hexagonal and 12 pentagonal faces that share edges to give the molecule’s familiar soccer-
ball shape. Hexagon–hexagon edges are slightly shorter than pentagon–hexagon edges, and
are referred to as double and single bonds, respectively.22 Crystalline C60 adopts a structure
which, to a first approximation, can be described as a cubic closest-packed (ccp) arrangement
of C60 spheres with cell parameter a = 14.17 Å. At room temperature, the C60 molecules are
rotationally disordered, but, on cooling, orientational ordering occurs23 and the

22
Average bond lengths of the two sets are 1.391 Å and 1.455 Å; intermediate between typical carbon–carbon double-
and single-bond lengths of 1.32 Å and 1.54 Å.
23
In fact, the molecules remain dynamic down to ~100 K in the Pa3 form but flip between discrete orientations.
 12.4 C60-Derived Superconductors 501

incompatibility of the m3m site symmetry of the ccp arrangement of spheres and the
icosohedral symmetry of the C60 molecules lowers the space-group symmetry. If we ignore
this subtlety, the ccp arrangement makes two tetrahedral and one octahedral holes per C60
(Section 1.4.2). The tetrahedral and octahedral holes have a radius of 1.12 Å and 2.07 Å,
respectively. C60 has a relatively high electronegativity and is easily reduced down to C606−.
Since the interstitial holes are comparable in size to alkali and alkaline-earth ions (six-
coordinate “CR radius” [11] r = 0.9 Å for Li+ and r = 1.81 Å for Cs+), intercalation24
compounds AxC60 form. Filling all the octahedral and tetrahedral holes leads to an A3C60
composition. In Na2AxC60 (A = K, Rb, Cs), the smaller Na+ ions occupy tetrahedral holes
and the larger A+ octahedral holes, and the symmetry remains Pa3. In A3C60 (A = K, Rb,
Cs), larger cations must occupy the tetrahedral holes and this forces the C60 molecules to
realign. They become disordered over two possible orientations such that the symmetry of
the average structure becomes Fm3m. Upon switching to body-centered cubic (bcc) packing,
A6C60 compositions can be formed, and metal content up to x = 12 is possible in compounds
such as Li12C60.
In 1991 it was reported that A3C60 compounds are metallic and superconduct at low
temperature [12], with Tc = 18 K for K3C60 and 29 K for Rb3C60. The origin of the
metallic behavior can be understood from the simplified molecular orbital (MO) dia-
gram of Figure 12.9. In the C60 molecule, the highest occupied molecular orbital
(HOMO) is fivefold degenerate (hu symmetry25), and separated by ~1.8 eV from a triply
degenerate t1u lowest unoccupied molecular orbital (LUMO) and by ~2.5 eV from the t1g
LUMO+1. In the solid state, these molecular orbitals broaden to form bands. For the
A3C60 composition, the band derived from the t1u level will be half filled, and one might
anticipate metallic properties.
In fact, much of the interesting behavior of A3C60 materials arises because they lie
close to a Mott–Hubbard metal-to-insulator boundary (see Section 10.4.1). In an iso-
lated C60 molecule, the electron–electron Coulomb repulsion is estimated to be around
3 eV and the Hubbard U in the extended solid around 1 eV. The bandwidth, W, is
around 0.5 eV. Since U > W, we would predict the materials to be Mott–Hubbard
insulators. In fact, when the triply degenerate nature of the LUMO is taken into
account, the Mott–Hubbard boundary moves to U > 2.5W for insulating behavior,
and we predict metallic properties for A3C60. We’ll see below that proximity to the
metal–insulator boundary can lead to dramatic changes in conduction properties as the
bandwidth W is changed.
One key question is whether these materials are BCS superconductors. Equation
(12.6) showed that one prediction from BCS theory is that Tc depends on the DOS
0
at the Fermi level N(EF), and Tc ∝ e1=NðEF ÞV implies that Tc will decrease as N(EF) is

24
Intercalation chemistry is discussed more fully in Chapter 13.
25
See Footnote 5.19 for symbolism; h implies fivefold degeneracy.
502 Superconductivity

t1g (C60 LUMO+1)

t1u (C60 LUMO)

octahedral sites
tetrahedral sites
hu (C60 HOMO)

A3C60
E E

pressure
t1u W

N(EF)
N(EF)

N(E) N(E)

Figure 12.9 Top: Structure of A3C60 and C60 MO diagram with electron filling appropriate for C603−.
Bottom: The t1u becomes a half-filled band in solid K3C60; the bandwidth W and N(EF) depend on
pressure.

lowered. Do A3C60 compounds follow this behavior? The easiest way to change N(EF) would
be to change the separation between the C60 molecules. As the molecules are moved closer
together, we would expect the width W of the conduction band to increase as bonding
interactions move to lower energy and antibonding interactions to higher energy. As the
molecules are moved apart, the opposite would happen. In the limit of infinitely separated
C60 molecules, one would have infinitely sharp bands, or the discrete energy levels of
the molecule, Figure 12.9. Since the area of the t1u-derived band must stay constant (it
can always adopt up to six electrons per C60), N(EF) will increase as molecular
separation increases. Equation (12.6) then predicts that Tc will increase with the cell param-
eter a. The easiest chemical way to increase a is to increase the size, or average size, of the
alkali metal. The size of the cation on the smaller tetrahedral site will be of particular
importance.
Figure 12.10 shows how Tc changes as a increases for a large number of A3C60
compounds,26 and reveals that they broadly follow the predictions of BCS theory up to a
volume per C60 of ~775 Å3, or face-centered cubic (fcc) cell parameter of ~14.6 Å. As we
discuss below, a Mott–Hubbard transition occurs beyond this value and the compounds
become antiferromagnetic insulators at low temperature. This behavior is similar to that of
non-BCS highly correlated electronic systems such as the cuprates and pnictides of Sections
12.7 and 12.8.

26
This and related plots actually show the volume per C60 molecule, allowing direct comparison of different structure
types.
 12.4 C60-Derived Superconductors 503

40 40

35 (a) Rb2Cs1 Rb1Cs2 35 (b)

30 Rb3 30

25 25
Rb0.35Cs2.65
Tc (K)

Tc (K)
20 20
K3
15 15
Na2Cs1
10 BCS- 10

5
like 5
Cs3
0 0
660 700 740 780 820 660 680 700 720 740 760
volume per C60 (Å3) volume per C60 (Å3)

Figure 12.10 (a) Dependence of Tc on volume per C60 molecule (adjusted to room temperature) for a
variety of C603− containing materials. Open circles from ref. [13], shaded from ref. [14]. (b) Evolution of Tc
as volume per C60 changes by external pressure for Rb3C60. Solid lines are guides to the eye.

The dependence of Tc on N(EF) is also revealed by the effect of external pressure. As

C60 molecules are pushed together, orbital overlap will increase, leading to broader bands
and lower N(EF). This would be expected to decrease the Tc of a BCS superconductor
(though note that it can have the opposite effect on non-BCS materials), and this is again
observed experimentally. Figure 12.10b shows data for Rb3C60. The reason why these
A3C60 materials show this apparently simple behavior is that the two important param-
eters for influencing Tc, i.e. the electron–phonon coupling constant and the DOS at the
Fermi level, can be influenced separately (at least to a first approximation). In fact, the
phonons responsible for superconductivity are thought to be essentially intramolecular
C60 vibrations and not to involve the alkali metal. This is supported by isotope-effect
values α in Tc ∝ M−α, which are significant for carbon but zero for rubidium (carbon α =
0.30±0.05 in K3C60 [15] and either 0.30±0.05 or 0.21±0.012 [16, 17] in Rb3C60, whereas
rubidium α = −0.028±0.036 [18]).
There have been considerable efforts to increase Tc in C60-derived materials, including
some spectacular high-profile claims that were later retracted.27 A genuine example of how to
increase the Tc of a parent A3C60 compound is via ammoniation of the 10.5 K supercon-
ductor Na2CsC60. This reaction leads to the formation of (NH3)4Na2CsC60, which contains

27
See, for example, “Plastic Fantastic: How the Biggest Fraud in Physics Shook the Scientific World” by E.S. Reich
(MacMillan Science).
504 Superconductivity

[Na(NH3)4]+ units of approximate radius 2.9 Å on the octahedral sites. The room-tempera-
ture lattice parameter increases from 14.13 Å to 14.47 Å and Tc rises to ~30 K. Similar
strategies have been tried with other A3C60 compounds, though Tc values are generally lower
than would be expected based on lattice parameter increases alone. This is thought to be due
to H-bonding interactions between NH3 molecules and σ electrons on C603− disrupting the
simple rigid band-structure picture, and changing transport properties.
For many years, the highest Tc that could be obtained reproducibly was in Rb2CsC60 with
Tc ≈ 29 K. Despite many efforts, the larger Cs3C60 (a = 14.76 Å) proved extremely hard to
prepare as it is thermodynamically unstable against disproportionation to Cs1C60 and
Cs4C60. Eventually, reproducible low-temperature routes to both a bcc and an fcc form of
Cs3C60 were reported. These materials help us understand the dip in Tc values beyond a
volume per C60 of ~760 Å3 shown in Figure 12.10. At ambient pressure, both Cs3C60 phases
are Mott insulators28 and the unpaired spins on C603− (Figure 12.9) order
antiferromagnetically at low temperature. However, on application of pressure, both
Cs3C60 polymorphs become metallic below 280 K and superconduct at low temperature.
The bcc material has a maximum Tc of 38 K and the fcc material, 35 K. The former Tc is the
highest of any C60 compound, and the latter the highest for an fcc A3C60 structure. Tc has a
strong dependence on pressure leading to the “dome-like” superconducting-phase diagram
of Figure 12.11.
The effect of pressure is to reduce the C60 separation, broaden the conduction bands and
cause a transition to the metallic state. It’s not surprising that similar behavior can be
achieved chemically. For example, introduction of Rb to form RbxCs3−xC60 will also move
C60 molecules closer together and broaden bands. Compounds in this family are found to be
metallic for x ≳ 0.35 and Tc varies from 26.9 K (x = 0.35) to 32.9 K (x = 1.0; the highest Tc in
this family) and back down to 31.8 K (x = 2) on increasing the Rb content. There is again a
dome-like dependence of Tc (shown with the shaded circles in Figure 12.10) similar to Cs3C60
under pressure. The three compositions are said to be respectively over-, optimally- and
under-expanded. The under-expanded composition Rb3C60 (Tc = 29.3 K) shows a value of
2Δ0/kTc of 3.6(1), close to the value of 3.52 predicted by BCS theory in Equation (12.5),
whereas the over-expanded samples show values up to 5.6(2) for x = 0.35, indicating non-
BCS behavior.
We can understand the domes of the superconducting-phase diagrams as arising from two
competing effects. If we start in the over-expanded region, contracting the cell destroys the
Mott insulator as the bandwidth increases and the electronic states become less molecule-
like. This initially gives rise to a metal with strong electron–electron correlations,29 where the
superconductivity is non-BCS in origin. In this region, decreasing the volume per C60

28
More formally they are Mott–Jahn–Teller insulators as the C60 molecules undergo a Jahn–Teller distortion, in
which the degeneracy of the three t1u orbitals is lost, giving a low-spin S = ½ state. As discussed in the text, removing
orbital degeneracy reduces the U/W ratio, above which insulating behavior occurs.
29
The initial metallic state has been described as retaining Jahn–Teller distortions, suggesting the presence of both
localized and extended states and has been called a Jahn–Teller metal.
 12.5 Molecular Superconductors 505

pressure (kbar)
19.5 11.6 5.4 0.8

metal

anti-

Tc (K)
ferromagnetic
insulator

antiferromagnetic
insulator +
superconductor

superconductor
bcc Cs3C60
volume per C60 (Å3)

Figure 12.11 bcc Cs3C60 undergoes an antiferromagnetic insulator to superconducting phase transition
on application of pressure; figures reproduced from [19] with permission.

increases Tc. As contraction continues and the bandwidth increases, one reaches more
conventional metallic behavior and BCS superconductivity occurs. In the BCS region, Tc
decreases with increasing bandwidth, as N(EF) decreases. The maximum value of Tc appears
to occur on the boundary of these two behaviors, highlighting the importance of both local
molecule-like features and extended properties in controlling superconductivity.
Superconductivity has also been observed in compounds with a higher level of doping,
such as Ba4C60 that has a Tc of 6.7 K. This material has a body-centered arrangement of
ordered C60 molecules with space group Immm. In this formally C608−compound, the
superconductivity comes from the t1g LUMO+1-derived band (Figure 12.9). The t1g band
appears to simultaneously give rise to better metallic conductivity and lower Tc values. For
example, the isostructural Ba4C60 (electrons in t1g band) and K4C60 (electrons in t1u band),
are a Tc = 6.7 K metal and an insulator, respectively. It seems likely that mixing of Ba and C
orbitals broadens the t1g band.

12.5 Molecular Superconductors

In addition to the C60 superconductors described above, there has been huge interest in the
development of so-called molecular or organic superconductors. Many of these are based on
salts of organic electron-donor (D) and -acceptor (X) molecules, Dδ+mXx−n. What distin-
guishes these from the C60 intercalates is that they are prepared from molecular precursors in
solution. This is often accomplished electrochemically by placing a solution of D and a salt of
Xx− in the anode compartment of an electrochemical cell, and a solution containing the
506 Superconductivity

anion in the cathode compartment; the two compartments are separated by a porous
membrane [20]. On passing a low constant current (typically ~1 μA), the D molecule oxidizes
into a cation, and crystals of DmXn grow slowly on the positive anode over a period of
days to weeks.
Interest in this field was sparked by the discovery of metallic conductivity in the organic
charge-transfer salt [TTF]δ+[TCNQ]δ− in 1972 (see Figures 12.12 and 10.29).30 The struc-
ture of this compound contains segregated stacks of TTF and TCNQ species, and their
electron donor/acceptor properties in the conductor originate in a charge transfer of ~0.59
electrons from TTF to TCNQ to form a “cation” and “anion” of fractional charge.31 Band-
structure calculations (Section 10.5.3) show that electron transfer occurs from a band
derived from the HOMO of TTF, which is a π orbital largely based on sulfur and the
two central carbon atoms, to a TCNQ LUMO that is more uniformly distributed across the
molecule.32 The intermolecular interactions of π orbitals along the stacks are then sufficient
to give metallic conductivity in a band that contains contributions from both the donor
HOMO and acceptor LUMO. Conductivity along the stack increases from ~4×104 S/m at
room temperature to ~106 S/m at 60 K.33 Below this temperature, a Peierls distortion34 (or
charge-density wave35) occurs, mainly on the TCNQ stacks, and the material becomes
semiconducting.
The scientific insights from TTF–TCNQ led to the synthesis of many related salts with
metallic conductivity containing TTF or one of the chemically related donors shown in
Figure 12.12. The first organic superconductors (often called Bechgaard salts after their
discoverer) have the formula (TMTSF)2X, where X can be a variety of inorganic anions
(PF6−, ClO4−, ReO4−, etc.). The anions lie between layers of [TMTSF]0.5+ cations formed
when two octet-obeying TMTSF molecules gave up 1 π electron to achieve partial aromatic
stabilization. Despite its fixed oxidation state, the anion is key to controlling superconduct-
ivity as it determines the distances, hence the interactions, between donor molecules. It also

30
It’s common to use abbreviations for donor and acceptor molecules: TTF = tetrathiafulvalene; TCNQ = tetra-
cyanoquinodimethane; TMTTF = tetramethyltetrathiafulvalene; TMTSF = tetramethyltetraselenafulvalene;
BEDT-TTF (or ET) = bis(ethylenedithio)tetrathiafulvalene; dmit2− = 1,3-dithiole-2-thione-4,5-dithiolate.
31
TTF in Figure 12.12 has two rings, each of them having seven σ electrons (two from each S, one from each C).
According to the Hückel rule, losing one electron per ring (off the C=C central bond) by oxidation to a [TTF]2+
cation would generate full aromatic stabilization. In the salt with organic acceptors, this tendency leads to a partial
oxidation that is more precisely described in band-structure terms.
32
Calculations suggest that the electron density of the TTF HOMO is around 61% S, 22% from the central C atoms
and 17% from the four outer C [Fraxedas et al., Phys. Rev. B. 68 (2003), 195115].
33
Cu has σ ≈ 108 S/m at room temperature. 34
Similar to that described for polyacetylene in Section 10.5.1.
35
The terms charge-density wave (CDW) and spin-density wave (SDW) are frequently encountered when discussing
complex electronic materials. If we have a charge-ordered metal oxide with alternating sites, such as Mn3+ and Mn4+
in La0.5Ca0.5MnO3 (Chapter 11), we could imagine describing the charge distribution using a periodic (charge-density)
wave propagating through the structure. Similarly, we could represent magnetic-moment direction and magnitude on
different sites of an antiferromagnet by a wave propagating through the structure. Such waves could have wavelengths
matching the lattice repeat and give rise to commensurate charge/spin ordering, or could have wavelengths that don’t
have a simple relationship to the lattice periodicity and give rise to incommensurate charge/spin ordering.
 12.5 Molecular Superconductors 507

Donors Acceptors

TTF TMTSF (TMTTF with S) TCNQ

BEDT-TTF or ET [M(dmit)2]x–

(TMTSF)2ClO4 β-(ET)2I3

Figure 12.12 Top: Molecular structures of donor and acceptor molecules discussed in the text. Bottom:
Structures of two molecular superconductors (H atoms omitted in β-(ET)2I3 for clarity).

influences subtle structural distortions of the structure, which can destroy metallic conduct-
ivity. In the first such superconductor discovered, (TMTSF)2PF6, a transition to an insulat-
ing state occurs on cooling below 12 K, in which the residual unpaired spins of the cations’
HOMO couple antiferromagnetically (a spin-density wave). If this transition is suppressed
by applied pressures of 0.6 GPa, superconductivity is observed below Tc ≈ 1.5 K (Figure
12.13a).
At room temperature, the molecular anions in these materials are rotationally disordered,
but this disorder freezes out on cooling. In the case of X = ClO4−, the freezing leads to
doubling of the cell dimensions in the same manner as observed for a Peierls distortion, and a
metal–insulator transition occurs. Under slow cooling, however, anion ordering occurs at
24 K in such a way that the periodicity is not doubled along the stacking axis, the metal–
insulator transition is suppressed, and superconductivity occurs below Tc = 1.4 K at ambient
pressure.
We saw in A3C60 compounds that both external and internal chemical pressure can
influence superconductivity. Similar observations hold for molecular superconductors as
illustrated in the schematic phase diagrams of Figure 12.13. As discussed above, Figure
12.13a shows how pressure tunes the properties of (TMTSF)2PF6 by influencing molecular
overlap in the donor stacks, leading to superconductivity above 0.6 GPa. Figure 12.13b
shows how the same effect can be achieved chemically. On the left-hand side of the diagram,
we have (TMTTF)2SbF6 that has sulfur-containing organic electron donors and becomes
508 Superconductivity

(TMTTF)2PF6 (TMTSF)2PF6
(TMTSF)2PF6 (TMTTF)2SbF6 (TMTTF)2Br (TMTSF)2ClO4

(a) (b)
12
temperature (K)

metal metal

temperature
8 CO

i/c
4
SDW superconductor SP
AF AF/SDW
SC
0.3 0.6 0.9 1.2
external pressure (GPa) chemical pressure

Figure 12.13 Schematic phase diagrams showing how superconductivity depends on either external or
chemical pressure in Bechgaard salts. (a) (TMTSF)2PF6 as a function of pressure. (b) Chemical-pressure
effect on a series of related compounds. i/c = incommensurate, SDW = spin-density wave, CO = charge
ordered, AF = antiferromagnetic, SP = spin Peierls, SC = superconductor. Figure adapted from Brown
(see Further Reading).

insulating on cooling at relatively high temperatures. If we change the anion to the smaller
PF6−, donor–donor interactions increase and charge ordering is suppressed to lower tem-
peratures. When we swap TMTTF for Se-containing TMTSF, even higher donor–donor
interactions further suppress charge ordering, and the application of moderate pressure then
induces superconductivity. The smaller ClO4− salt will superconduct even at ambient pres-
sure. It has also been shown that application of sufficiently high pressures will drive TMTTF
salts into the superconducting state.
The origin of the pairing needed for superconductivity in these materials isn’t known
definitively, but there is strong evidence that intra-stack antiferromagnetic spin fluctu-
ations are important; i.e. that a non-BCS mechanism is responsible. This is consistent
with the observation that superconductivity in (TMTSF)2PF6 emerges just above the
pressure at which the spin-density wave is destroyed, and that Tc has its highest value
here.
Far higher superconducting temperatures are observed in (ET)2X salts (e.g. Figure
12.12, lower right). These compounds again have structures in which layers of electron
donors are separated by anions, with the donor packing leading to more 2D electronic
properties. In the β-(ET)2X (X = [I3]−, [IBr2]−, [AuI2]−, etc.) compounds, the donors
have their molecular planes parallel (Figure 12.12). In the so-called κ-phase salts, e.g. κ-
(ET)2Cu[N(CN)2]Br, the donor layers contain “dimers” of closely separated pairs of
face-to-face donors, which then interact with other dimers that are oriented at approxi-
mately 90°. Conductivity is highly anisotropic; typically ~5000 S/m in plane but 1000
times lower perpendicular to the plane. The compound κ-(ET)2Cu[N(CN)2]Br has one
 12.6 BaBiO3 Perovskite Superconductors 509

of the highest ambient pressure Tc values (11.6 K) known for a molecular system. By
contrast, the Cl analogue undergoes a metal-to-insulator transition at 40–50 K on
cooling. However, a modest pressure (~0.3 kbar) suppresses this and a 12.8 K super-
conductor results.
Salts of metal complexes containing ligands with strong structural similarities to TTF have
also shown superconductivity. TTF[Ni(dmit)2]2 contains TTF-like species as both the elec-
tron donor and acceptor, and superconducts at 1.6 K under pressure. Superconductivity
supported purely by the metal-complex anions has also been observed at 5 K under pressure
in NMe4[Ni(dmit)2]2.
The range of molecular superconductors is large36 and will undoubtedly continue to
grow. One much-heralded advantage of molecular systems is the ability of the chemist to
tailor the structure of the electron-donor and electron-acceptor species at will. Such
changes can help suppress the transitions (Peierls and Mott–Hubbard) that compete with
superconductivity. The disadvantage is that the large number of intermolecular inter-
actions present in the crystal structure means that different structural arrangements can
have similar energies, and polymorphism is common. It’s therefore not surprising that even
making subtle changes to the electron donors and acceptors can frequently lead to major
changes in structure and properties. Such structural changes are extremely difficult to
predict or control a priori.

12.6 BaBiO3 Perovskite Superconductors

Perovskite-derived superconductors have been extensively studied since the discovery of
superconductivity in reduced SrTiO3−δ (Tc ≈ 0.3 K) in 1964 [21]. One early example, which
paved the way for later work on cuprates, was found by Sleight and co-workers in the solid
solution between two perovskite materials, BaPbO3 and BaBiO3 [22].
BaPbO3 has a tilted version of the perovskite structure with orthorhombic symmetry
Ibmm at room temperature and might be expected to be an electronic insulator like most
other main-group perovskites. In fact, the energy of the Pb 6s orbital is such that the empty 6s
conduction band partially overlaps the O 2p valence band, and BaPbO3 has metallic
conductivity with σ ≈ 3000 S/cm at 300 K.
From simple size arguments, one might expect BaBiO3 to be a cubic perovskite and that
the single 6s electron of Bi4+ could also give rise to metallic conductivity. In reality, the
electronic structure is more complex. We know from chemistry that 4+ oxidation states are
unusual for group 15 elements, and in BaBiO3 the nominal Bi4+ should disproportionate to
Bi3+ and Bi5+, giving Ba2BiIIIBiVO6, analogous to compounds like Ba2BiIIITaVO6. As with
the molecular systems of Section 12.5, this disproportionation can be described as a charge-
density wave. Structural studies support this view and reveal two distinct octahedral sites,

36
Over 100 superconducting phases are known with donors chemically related to BEDT-TTF (ET) and TMTSF.
510 Superconductivity

1000 Pm3m Fm3m

I4/mcm R–3

temperature (K)
I2/m
100 charge-
metal ordered
Ibmm semiconductor
Ba
Ibmm P21/n
Tc
10
super-
conductor
“Bi5+” “Bi3+” “Bi5+” I4/mcm
0.0 0.2 0.4 0.6 0.8 1.0
BaBiO3 x in BaPb1–xBixO3

Figure 12.14 Structure of BaBiO3 and schematic structural phase diagram of BaPb1−xBixO3. Space
groups of important phases are given. Issues with sample purity mean that it is hard to be definitive
about the position of the metal-to-semiconductor boundary.

one (Bi3+) with Bi–O bond distances around 2.28–2.30 Å and a second (Bi5+) with shorter
distances of 2.11–2.13 Å. The combined Bi-site ordering and tilting of octahedra gives P21/n
monoclinic symmetry at room temperature and below. Band-structure calculations show
that it is a semiconductor. We can understand the conductivity of BaBiO3 using a polaron
model, in which a pair of electrons moves from Bi3+ to Bi5+, turning it to Bi3+, etc. This
requires a significant change in the size of the BiO6/2 octahedron.37
In BaPb1−xBixO3 solid solutions, one can tune the electronic properties across a metal-to-
semiconductor boundary at x ≈ 0.3. On cooling, superconductivity is observed for samples
with 0.05 < x < 0.30, with the maximum Tc of 13 K at x ≈ 0.27 (Figure 12.14), occurring very
close to the metal-to-semiconductor boundary. The structural phase diagram of
BaPb1−xBixO3 is complex, and superconducting samples are often found to be a metastable
mixture of the orthorhombic phase of space group Ibmm and a tetragonal phase (I4/mcm)
that is believed to be responsible for superconductivity. What electronic role does the
seemingly innocent Pb4+ play in controlling the structure of these materials and therefore
their superconductivity? If we start at the Bi-rich end of the series, increasing the Pb content
disrupts the Bi3+/Bi5+ charge order as the intermediate size and valence of Pb4+ [r(Pb4+) =
0.78 Å; r(Bi3+/5+) = 1.03/0.76 Å] breaks down the large/small ordered pattern of octahedra.
The onset of superconductivity occurs close to the doping level at which charge ordering
disappears. At this level, the material is metallic at high temperatures and the electrons are
delocalized.
Superconductivity at a higher Tc of up to 34 K has also been created via A-site doping of
BaBiO3 to produce Ba1−xKxBiO3, with superconductivity observed for 0.30 < x < 0.45. The
doping by the K+ acceptor is compensated by oxidation of Bi3+, which will again disrupt the

37
As a pair of electrons moves, it swaps the identities of Bi3+ and Bi5+ sites. Oxygens will rearrange such that Bi3+ sites
have longer Bi–O distances and Bi5+ short, as electrons move between different Bi–O bonding orbitals. We can
imagine a “breathing” mode of the octahedra being involved.
 12.7 Cuprate Superconductors 511

pattern of charge ordering. The phase diagram is again complex, but increasing the K
content initially destroys charge ordering, and the material superconducts once the low-
temperature symmetry becomes I4/mcm at x > 0.30 (i.e. the same symmetry as observed for
BaPb1−xBixO3 superconductors).38 A 12 K superconductivity has been observed in
Sr0.4K0.6BiO3 that can be prepared at high pressure. Charge ordering is again evident in
the undoped material and suppressed in the superconducting phase. Superconductivity has
also been observed in the A-site-deficient (K0.45Na0.25)Ba3Bi4O12 below 27 K and in the
related Sb phase BaPb0.75Sb0.25O3 below 3.5 K.
The origin of superconductivity in these compounds is still debated, and detailed
understanding is made difficult by the fact that two structurally similar phases are
usually present in superconducting samples of BaPb1−xBixO3. There is, however, good
evidence that they have BCS-like properties with both BaPb1−xBixO3 and Ba0.6K0.4BiO3
having 16O/18O isotope effects (Tc ∝ M−α), with α of 0.4–0.5. It’s clear from their
structural behavior that the significant electron–lattice-vibration interactions required by
BCS theory exist, and that superconductivity occurs close to the metal–insulator bound-
ary and close to charge-localizing phase transitions. It’s worth concluding by emphasiz-
ing two features we’ll see again in cuprates: electrons are in a band with significant
oxygen and metal character, and there is a tendency for charge disproportionation of Bi
over oxidation states ranging from BiIII to BiV.

12.7 Cuprate Superconductors

One of the most significant breakthroughs in the field of superconductivity came in
November 1986 when Georg Bednorz and Alex Müller at IBM Research Zurich reported
superconductivity at 28 K in the La–Ba–Cu–O system [23]. By the end of that year, the
superconducting phase had been identified as La1.85Ba0.15CuO4, which has a Tc up to
30 K. This released a frenzy of world-wide research activity aimed at understanding and
exploiting these materials. Superconductivity below 38 K was rapidly found in
La1.85Sr0.15CuO4, 93 K superconductivity (above the 77 K boiling point of readily
available liquid nitrogen) in YBa2Cu3O7, and eventually 134 K superconductivity
(164 K under pressure) in HgBa2Ca2Cu3O8. The observation of such high Tc values
overthrew predictions from BCS theory that the maximum Tc would be limited to
~30 K39 and suggested that another pairing mechanism must be at play. We will look
at the structural chemistry of key cuprate superconductors and the electronic features that
lead to their remarkable properties.

38
These samples are cubic at room temperature.
39
More recent work on MgB2 shows a higher BCS limit of ~39 K.
512 Superconductivity

12.7.1 La2CuO4 “214" Materials

Pure La2CuO4 is an antiferromagnetic insulator with a Néel temperature of ~317 K. It
has a slightly distorted version of the K2NiF4 structure, as shown in Figure 12.15. The
K2NiF4 structure can be viewed as an intergrowth of the perovskite and rock-salt
structures. If one separates 2D infinite slabs of corner-sharing MO2O4/2 octahedra from
the perovskite structure (Figure 12.15a) and translates them by a/2 + b/2 alternately back
and forth such that the axial oxygens lie directly above and below positions of
A-site cations from neighboring layers, one forms the K2NiF4 structure. The A-site
cation is nine-coordinate, with one oxygen directly above it (see the lowest A site
in Figure 12.15c), four oxygens in its plane and four oxygens in a plane below. From
the “waist down”, this cation has the coordination environment found for the A site in
perovskite; from the “waist up”, it has the coordination of a cation in the rock-salt
structure. The crucial structural feature as far as superconductivity is concerned is that
there are infinite 2D layers of distorted corner-sharing CuO2O4/2 octahedra that have
axial (often called apical in the cuprate literature) Cu–O bond lengths of ~2.4 Å and
equatorial bond lengths of ~1.9 Å (ICSD 68381).
There are two distinct ways in which La2CuO4 can be doped to make it superconducting.
Both rely on hole doping (which is a partial/fractional oxidation, here of Cu2+, as opposed to
electron doping that would be a partial reduction), and we’ll discuss the electronic changes
that occur in Section 12.7.4. One way is to partially replace La3+ with a divalent ion to form
compounds such as La2−xSrxCuO4, with the highest Tc = 38 K at x = 0.15.40 The average
copper oxidation state is then Cu2.15+. Alternatively, samples can be doped with excess
oxygen to give La2CuO4+δ. The highest-Tc material has δ ≈ 0.08, implying a similar average
oxidation state of Cu2.16+. The oxygen interstitial occupies a site midway between adjacent
La layers in the structure such that it is surrounded by a tetrahedron of La atoms, as shown in
Figure 12.15c. Significant local distortion is required to allow this, and nearby oxygen atoms
move by up to 0.5 Å. These oxygen interstitials are often mobile in the structure, and the
structural and electronic properties of these and related cuprate superconductors can there-
fore vary with time.
It is interesting to note that optimal superconducting properties occur at similar Cu
oxidation states for both doping routes and occur at doping levels just sufficient to suppress
low-temperature antiferromagnetic ordering of the Cu2+ spins. If such a material is under-
doped, superconductivity is not observed; if overdoped, normal metallic behavior results
without superconductivity. As with A3C60, superconductivity occurs near the metal-to-
insulator boundary. A schematic phase diagram of cuprate superconductors illustrating
these and other points is discussed later in Section 12.7.4.

40
Tc up to 52 K can be achieved in related compositions under pressure (ΔTc ≈ 0.009 K per GPa) or in deliberately
strained thin films. Attempts to increase Tc by chemical rather than external pressure by replacing La3+ with the
smaller Y3+ led to the identification of YBa2Cu3O7.
 12.7 Cuprate Superconductors 513

CuO4/2
layers

La

interstitial
site

(a) perovskite (b) expanded perovskite (c) La2CuO4

Figure 12.15 Relationship between the (a) perovskite and (c) La2CuO4 structures. The central layer of a
hypothetical expanded perovskite (b) is displaced by a/2 and b/2 such that its apical oxygens lie above and
below the La atoms of adjacent layers. A potential La4O interstitial site, as found in La2CuO4+δ, is
labelled.

It’s also worth noting that even rather subtle changes in the local structure of a material can
influence superconducting properties. For example, it’s found experimentally that for a variety
of “optimally doped” Ln3+1.85 A2+0.15CuO4 materials, all with the same Cu2.15+ oxidation state
but with different combinations of Ln3+ and A2+ cations, there is a maximum Tc for a mean A-
site nine-coordinated ionic radius of ~1.22 Å. In addition, Attfield et al. [24] have shown that
not only the mean cation radius but the variance41 of the cation radius is important—at a
constant doping level, Tc is lowered if a large variety of small and large cations is used to
produce a given average radius.

12.7.2 YBa2Cu3O7−δ “YBCO” or “123” Materials

Perhaps the best known of the cuprate superconductors is the yttrium barium copper oxide
YBa2Cu3O7−δ first reported by Wu and Chu in 1987 [25]. It is frequently referred to as
YBCO or 123 (after the cation ratios). YBa2Cu3O7 has a Tc of 93 K.
The structure of YBa2Cu3O7 can again be derived from perovskite. As shown in Figure
12.16, the prototype perovskite cell is tripled such that it contains A3M3O9 with three MO6/2
octahedra. The simplest way (Pauling’s parsimony rule, Chapter 1) to remove two O atoms is
to form 1MO4/2 + 2MO5/2. The square-planar CuO4/2 link in chains, the corner-sharing bases
of the CuO5/2 square pyramids link in planes. This leads to the colloquial names for the two
sites as chains and planes, respectively.

XN
41
The population variance, σ2, is defined as σ2 ¼ 1=N ðxi  μÞ2 , where there are N values xi and µ is their mean. A
range of different radii will give a large variance. i¼1
514 Superconductivity

×× ×× × ×× chains
Ba

planes

×× ×× Y

Ba
O

×× × ××
Cu
××

A3M3O9 YBa2Cu3O7 YBa2Cu3O7 YBa2Cu3O6

Figure 12.16 A hypothetical triple perovskite structure A3M3O9. Deleting two oxygens per unit cell
(crosses) leads to YBa2Cu3O7, shown in ball-and-stick and polyhedral views. Deleting a further oxygen
leads to YBa2Cu3O6.

4.00 12.00 100

(a) (b)
3.96 80
11.80

3.92 a 60
cell (Å)

TC (K)

b 11.60
c
3.88 40

11.40
3.84 20

3.80 11.20 0
6.0 6.2 6.4 6.6 6.8 7.0 6.0 6.2 6.4 6.6 6.8 7.0
x x

Figure 12.17 The dependence of (a) unit-cell parameters and (b) Tc on oxygen content for a series of
YBa2Cu3Ox samples. Data from [27].

The structural and superconducting properties of YBa2Cu3O7−δ depend critically on

oxygen content (Figure 12.17 shows data from an early publication). As the oxygen content
is lowered below 7, oxygen is removed from CuO4/2 at sites halfway along the b cell edge,
until at δ = 1 one reaches the YBa2Cu3O6 structure shown in Figure 12.16. Below
YBa2Cu3O6.41, the material no longer superconducts. This loss of superconductivity is
associated with a structural change from orthorhombic (a ≈ b ≠ c) to tetragonal (a = b ≠ c)
 12.7 Cuprate Superconductors 515

Box 12.2 Synthetic Methods: Synthesis and tuning superconductivity in

YBa2Cu3O7−δ
It’s straightforward to synthesize a sample of YBa2Cu3O7−δ that will show superconductivity.
A simple procedure is to grind and mix a stoichiometric ratio of 0.5Y2O3, 2BaCO3, and 3CuO
and then heat it to 950 °C for a few hours in air. The mixture is then reground, pelletized, and
reheated to 950 °C, then slow-cooled at around 1 °C a minute in air (or better in O2) to room
temperature. Such samples will show the Meissner effect by repelling a magnet when cooled by
liquid N2. However, we know that superconductivity in this system depends sensitively on the
oxygen nonstoichiometry, with the highest Tc when δ approaches 0. How do we make an
optimal sample? At what temperatures and in what specific atmosphere should it be prepared?
Experimental data for the dependence of δ in YBa2Cu3O7−δ on temperature and oxygen
partial pressure are shown below:

In Chapter 3, we simplified the defect equilibria in YBa2Cu3O7−δ to a single pseudochemical

oxidation of oxygen vacancies with the equilibrium constant Kvox = pO2−½·4(1−δ)2/δ;
see Equation (3.5). How do we derive the temperature dependence for this equilibrium
constant? The easiest approach is to express the Gibbs energy as ΔG = ΔH − TΔS and to
introduce ΔG = −RTlnK, which gives Kvox = exp[(ΔSvox/R) − (ΔHvox/RT)]. A least-squares
fitting of the experimental data gives ΔSvox = −69.8(4) J/(mol K) and ΔHvox = −79.8(3) kJ/mol
as parameters. The calculated isotherms shown in the graph above agree best with the experi-
ment in the middle range of nonstoichiometry where the concentration of the neglected defects
(see Section 3.3) is lowest. We can see from this graph that to prepare samples with δ ≈ 0, one
should anneal the sample at relatively low temperatures (<500 °C) in pO2 > 1.

symmetry, related to oxygen atoms randomly occupying ½ 0 0 and 0 ½ 0 positions in the

tetragonal unit cell; a short-range order occurs at the local level [26].
In YBa2Cu3O7, the average Cu oxidation state +2.33 is distributed over one square-planar
CuO4/2 and two CuO5/2 square pyramids. Since we know that d 8 ions favor square-planar
516 Superconductivity

coordination, a first approximation is that YBa2Cu3O7 contains one square-planar Cu3+ and
two square-pyramidal Cu2+. YBa2Cu3O6 then has the typical linear Cu+ and two Cu2+. As in
La2CuO4, both compounds have their planes occupied by Cu2+ in this first approximation.
What happens next in superconducting YBa2Cu3O7 is that the mismatch of the two average
Cu–O bond orders 3+/4 versus 2+/5 along the same length of the short unit-cell edge, and the
highly oxidizing nature of Cu3+, both make the chains partially hole dope (oxidize) the Cu2+
of the planes, as can be shown by bond-valence analysis. This intrinsic hole doping again leads
to p-type superconductivity of this cuprate, suggesting the superconducting pairing derives
from diluted holes in the planes. More in Section 12.7.4.

12.7.3 Other Cuprates

The remarkable properties of the 214 and 123 cuprates prompted the search and subsequent
discovery of superconductivity in a wide variety of copper-containing mixed metal oxides.
All of them contain CuO4/2 planes (with or without additional apical oxygens to give five- or
six-coordinate Cu), with bewilderingly complex chemical formulae and structures. Just as
with the 214 and 123 families, many of them can be described and understood in terms of the
intergrowth of slabs from simple structure types (like layers from the rock-salt, fluorite, or
perovskite structures). Some contain increasing numbers of adjacent Cu layers, leading to
homologous series of ideal formulae42 such as HgBa2Can−1CunO2n+2 (n = 1–7),
TlBa2Can−1CunO2n+3 (n = 1–4), Tl2Ba2Can−1CunO2n+4 (n = 1–4), or Bi2Sr2Can−1CunO2n+4 (n =
1–3). It’s common to use abbreviated names for many of these phases. For example, Bi-2201 is
used for the n = 1 member of the last family (Bi2Sr2Ca0Cu1O6). To illustrate the close relation-
ships between these phases, the structures of the n = 1, 2, and 3 members of the
HgBa2Can−1CunO2n+2 family are shown in Figure 12.18. Note that CuO2 layers separated by
small Ca2+ cations don’t contain apical O atoms. The n = 1 material has the highest Tc of any
cuprate with a single CuO2 plane (98 K). In general, it’s found that Tc of such materials increases
as the number of adjacent Cu planes increases from n = 1 to n = 3 but decreases thereafter.
These chemically complex materials are difficult to prepare as single phases, and they
frequently have compositions that differ from ideal stoichiometric formulae. This can be
caused by a variety of defects (e.g. in TlBa2Can−1CunO2n+3 by Tl/Ca substitution, Cu/Tl
substitution, Tl/Ca vacancies, and/or oxygen nonstoichiometry), by loss of volatile species
(e.g. CuO or Bi2O3) during synthesis, or by intergrowths of different structures between the
cuprate planes. This leads to a variety of mechanisms for the hole doping responsible for
their superconductivity. Many materials, in particular the Bi phases, have complex incom-
mensurate structures due to the size mismatch between adjacent structural layers. Many
other cuprate superconductors are known with species such as halide anions, CO32−, BO32−,
or (GaO33−)∞ zig-zag chains of corner-sharing tetrahedra between Cu planes.

42
As in YBCO, the exact oxygen content may differ from these formulae. For example, the Hg series is often expressed
as HgBa2Can−1CunO2n+3−δ.
 12.7 Cuprate Superconductors 517

Hg
Ba
O

3×CuO2
2×CuO2 Ca
1×CuO2

HgBa2CuO4 HgBa2CaCu2O6 HgBa2Ca2Cu3O8

n = 1, Tc = 98 K n = 2, Tc = 128 K n = 3, Tc = 134 K

Figure 12.18 Structures of the n = 1, 2, and 3 members of the HgBa2Can−1CunO2n+2 series. The ½ ½ 0
oxygen positions would be occupied by O in the related TlBa2Can−1CunO2n+3 series. The n = 3 member
has Tc = 164 K under pressure; the reason for the Tc enhancement is not understood.

In addition to hole doping leading to p-type superconductivity, some cuprates can be

electron-doped to n-type superconductivity. These contain CuO2 planes without apical
oxygens. The first such family discovered was based on Nd2CuO4 doped with CeO2, or
with substitution of F for O, to give Nd2−xCexCuO4 or Nd2CuO4−xFx. A Tc up to ~25 K
has been found. The structure of Nd2CuO4 is shown in Figure 12.19a.43 The smaller radius
of Nd3+ compared to La3+ leads to a different structure type containing slabs of fluorite-
like [Nd2O2]2+ separated by [CuO4/2]2−. The size of the Nd2O2 block is such that CuO4/2
planes are under slight stretch tension. Electron doping puts electrons in Cu–O antibond-
ing states (see below), relieving this tension. In these systems, it is again difficult to produce
phase-pure samples, and the exact composition of the superconducting component is
unsure. n-type superconductivity has also been achieved at T = 40 K in Sr1−xNdxCuO2
synthesized at high pressures. This material has the so-called infinite-layer44 superconductor
structure (Figure 12.19b), of alternating CuO4/2 plane and cation layers, and can be
thought of as the “parent structure” of all cuprate superconductors.45

12.7.4 Electronic Properties of Cuprates

What can we say about the origin of superconductivity in the cuprates? Currently there
is still no universally accepted theory to explain their remarkable properties. One

43
Colloquially known as the T 0 phase versus La2CuO4 the T phase.
44
“Infinite” refers to a hypothetical n = ∞ member of the series HgBa2Can−1CunO2n+2 in Figure 12.18, a CaCuO2 with
only the apex-free CuO2 planes.
45
Ca0.84Sr0.16CuO2 appears to be the only phase with the infinite-layer structure accessible at ambient pressure.
518 Superconductivity

Nd2O2 layers
with ONd4/4
tetrahedra Sr/Nd
CuO4/2
CuO4/2 square planes
square planes

(a) Nd2CuO4 (b) Sr1–xNdxCuO2

Figure 12.19 Nd2CuO4 and the “infinite-layer” compound, Sr1−xNdxCuO2.

La2CuO4 Ln3+/A2+

YBa2Cu3O7 chains

HgBa2CuO4 HgBa2O2

CuO4/2 layers charge reservoirs

Figure 12.20 Schematic representation of the key structural features of superconducting cuprates.

common feature of all the cuprates discussed is the presence of one or more CuO4/2
planes. Controlling their electronic properties by doping is key to controlling super-
conductivity. This leads to a generic description of cuprates (Figure 12.20) as electron-
ically active CuO4/2 planes separated by so-called charge reservoirs that dope the
planes. The CuO4/2 planes can contain purely corner-sharing square-planar copper or
additional apical oxygens may be present to form square pyramids (as in YBa2Cu3O7)
or octahedra (as in La2CuO4).
As we’ve seen in Section 6.5.2, d 9 Cu2+ in a square-planar oxygen environment will have a
d-orbital splitting pattern that leads to a half-filled band of energy levels derived from the
antibonding overlap of p orbitals on oxygen and the dx2 y2 orbital on Cu (Figure 12.21). Due
to the similarity in energy of Cu 3d and O 2p orbitals, there is significant covalence, and the
band has both Cu and O character. The width of the band is also such that we expect the
material to be a Mott–Hubbard insulator rather than a metallic conductor. In the majority of
cuprates, the role of the charge reservoir is to introduce holes in this band, which induces an
 12.7 Cuprate Superconductors 519

upper Hubbard
E E band
Cu 3dx2-y2 U
lower Hubbard
band

O 2p and O 2p and
Cu-filled Cu-filled
bands bands

DOS, N(E) DOS, N(E)

Figure 12.21 Left: Schematic density-of-states plots for CuO4/2 planes; Right: Splitting of the dx2−y2 into
an upper and lower Hubbard band due to energy cost of electron pairing.

insulator–metal transition.46 The mechanisms for doping vary. In the La2CuO4 series, the
extrinsic hole doping by either La3+/A2+ substitution or incorporation of excess oxygens
produces a Cu oxidation state of +2.16. In YBa2Cu3O7 the hole doping is intrinsic: Cu3+
of the chains partially oxidize the Cu2+ of planes to Cu(2+x)+ and are themselves reduced to
Cu(3−2x)+. A variety of similar doping mechanisms occur in the chemically more complex
systems.
Many cuprates have a T-versus-doping phase diagram similar to Figure 12.22, in
which undoped materials are antiferromagnetic insulators at low temperature. Long-
range antiferromagnetism is suppressed at ~3% hole doping and superconductivity
emerges at ~5% doping (between these regions local phase segregation occurs). Tc is
typically maximized at ~15–20% doping, and superconductivity destroyed above ~25%
doping, leading to a superconducting dome similar to those we’ve seen in other mater-
ials. The regions prior to and after the maximum Tc are referred to as under- and over-
doped regions, respectively.
The electronic properties of the cuprates above Tc turn out to be just as unusual as
those below, and are still not properly understood. Initial doping of the Mott insulator
causes a gradual transition to what’s known as the pseudogap regime. In this region
there is a gap in the electronic density of states for some values of k (i.e. some
momentum directions of electrons) but not for others. There is also a strong tendency

46
In 2019, superconductivity was discovered in thin films of Nd0.8Sr0.2NiO2 [D. Li et al., Nature 572 (2019), 624–627].
The Ni oxidation state is around one unit lower than in the Cu superconductors, suggesting superconductivity again
occurs by hole doping a d 9-derived half-filled band. The Ni possibility was discussed by Karen et al. [J. Solid State
Chem. 97 (1992) 257–273] when they correlated high-Tc superconductivity with the presence of a well-separated
nearly half-filled Cu-localized HOMO so that a dilute occurrence of a hole pair (if less than half filled) or an electron
pair (if more than half filled) in that orbital was imaginable.
520 Superconductivity

strange metal

temperature

AF Mott insulator
pseudogap
insulator

normal
metal
superconductor
0.0 0.05 0.10 0.15 0.20 0.25 0.30
x in Cu(2+x)+ of CuO2 layers

Figure 12.22 A generic phase diagram for hole-doped cuprate superconductors. Dashed lines separate
regions where there are poorly defined changes in electronic behavior.

towards forming a variety of ordered states such as charge-density waves,47 spin-

density waves, and more exotic forms of electronic order. The pseudogap regime has
several electronic features in common with the superconducting state, though there is
debate about whether it is a precursor to superconductivity or a competing state.
Beyond the pseudogap region, one enters a region where the material is often described
as a “strange” or “bad” metal to reflect its differences to a normal metal. Examples of
these differences include high-temperature resistivities two orders of magnitude higher
than normal metals48 and a linear dependence of resistivity with temperature up to the
highest temperatures measured.49 It is perhaps unsurprising that the electronic proper-
ties of cuprates in this region are unlike those of a normal metal. The reason for the
insulating nature of the undoped materials (which have band gaps of ~2 eV) is the high
Hubbard U that reflects strong electron–electron repulsions on moving an electron
from one site to another that already has an electron (see Section 10.4.1). Some of
these repulsions will still be present in doped cuprates, and we can imagine electrons as
having to “get out of each other’s way” to allow conductivity by moving in concert. As
such, the mobile electrons will be strongly correlated, unlike those in a normal metal.
As the hole-doping levels are increased, correlations reduce and the high-temperature
properties gradually become more like those of a normal metal.

47
One form of separate ordering of charge and spin leads to so-called “stripes” of electron density in some cuprates.
48
The in-plane resistivities (per CuO2 layer) are very similar for different families of optimally doped cuprates and
typically around 5×10−4 Ω cm at 300 K. Out-of-plane resistivities are higher by a factor rising from ~30 for
YBa2Cu3O6.95 to ~1×106 for Bi2Sr2Ca2O8+δ just above Tc [N. Hussey, J. Phys. Condens. Matter. 20 (2008), 123201].
49
In a normal metal, resistivity saturates at high temperatures when the mean free path becomes comparable to the de
Broglie wavelength; see Chapter 15.
 12.8 Iron Pnictides and Related Superconductors 521

How do these arguments influence superconductivity? While it’s clear that the cuprates
obey the general rule of “bad-metal, good-superconductor”, this doesn’t explain Cooper-
pair formation at temperatures far higher than those predicted by BCS theory for conven-
tional superconductors. There’s good evidence that the antiferromagnetic nature of the
parent insulating phases survives into the superconducting regime in the form of short-
range dynamic magnetic fluctuations, and that these are much stronger than in a normal
metal. Most experts think that these are important for pair formation.
Despite over 30 years of intense scientific effort, it’s still not possible to provide a complete
explanation for superconductivity in the cuprates. We shouldn’t feel too bad about this.
Writing in 2015, Chu (co-discoverer of YBa2Cu3O7) and co-workers concluded that it is still
necessary to “let the experts resolve the different views” [28].

12.8 Iron Pnictides and Related Superconductors

In February 2008, Hosono and co-workers reported superconductivity at Tc = 26 K in
LaO1−xFxFeAs with 0.14 < x < 0.4 [29]. The observation of such a high Tc in an Fe-based
superconductor was a considerable surprise and rapidly led on to the discovery of ambient-
pressure Tc = 55 K in SmO1−xFxFeAs, a temperature only surpassed by the cuprates.
Superconductivity has since been discovered in a range of related materials, most notably
those based on BaFe2As2 (Tc up to 30 K), LiFeAs (Tc = 18 K), and even in the binary phase
β-FeSe50 (Tc = 37 K at 8.9 GPa). In contrast to other families, superconductivity in these

Fe
Fe

[FeAs]− Li
Fe Ca
As
La

[LaO]+

LaOFeAs CaFe2As2 LiFeAs

1111 122 111

Figure 12.23 Pnictide superconductors. All contain layers of edge-sharing FeAs4/4 tetrahedra.

50
It is the tetragonal β-FeSe and not the hexagonal α form that superconducts; there is some confusion in the literature
over the use of α/β.
522 Superconductivity

materials can be remarkably tolerant to compositional changes, and chemical changes that
can modify superconducting properties are typically possible on all sites.
As with the cuprates, these materials are often referred to by a shorthand notation:
LaOFeAs as “1111", CaFe2As2 as “122”, LiFeAs as “111” (Figure 12.23). All of them
contain [FeAs]− layers made up of edge-sharing FeAs4/4 tetrahedra. In the 1111 family,
these layers alternate with [LaO]+ layers of OLa4/4 edge-sharing tetrahedra. The La ions
“bridge” the oxide and pnictide portions of the structure and have a 4 + 4 square-
antiprismatic coordination LaO4Se4. In the 122 family, A2+ replaces [La2O2]2+ segments,
giving a ThCr2Si2-type structure (Th → Ca, Cr → Fe, Si → As). LiFeAs has Li+ ions in
five-coordinate square-pyramidal sites between the [FeAs]− tetrahedral layers (Cu2Sb
structure type). Finally, FeSe adopts the α-PbO structure type and has the same FeSe4/4
layers but with (in the ideal structure) no cations separating them.
All these pure/undoped compounds share common features. Each contains Fe2+ arranged
on a square grid of edge 2.63–2.85 Å at ~200 K.51 The Fe–Fe distance is much shorter than
the Cu–Cu distances in the cuprates (~3.9 Å), and band-structure calculations show that the
important electronic states near EF are dominated by iron d–d overlap. Experimentally, these
materials are metallic conductors at room temperature with resistivities of ρ ≈ 10−2 Ω cm for
LaOFeAs and ~10−3 Ω cm for β-FeSe. This is in contrast to the Mott-insulating or bad
metallic behavior of the cuprates.
The most straightforward way of generating superconductivity in the 1111 family is by
electron doping. In LaO1−xFxFeAs, superconductivity is observed for 0.04 < x < 0.14 with a
maximum Tc around x ≈ 0.11. Doping has several effects. The room-temperature resistivity is
approximately halved; the antiferromagnetic ordering, which occurs around 150 K in

150
poor metal
temperature (K)

paramagnetic

100
antiferromagnetic

50

superconductor

doping level

Figure 12.24 Generic phase diagram for LaOFeAs-derived superconductors.

51
Small distortions from tetragonal to orthorhombic symmetry occur on cooling most systems.
 12.9 Problems 523

LaOFeAs,52 is shifted to lower temperature and eventually suppressed; and superconductiv-

ity is observed at low T. The schematic phase diagram of the pnictide superconductors
(Figure 12.24) is therefore strikingly similar to that of the cuprates, with superconductivity
emerging as the antiferromagnetic order is suppressed. Superconducting hole-doped com-
positions such as La0.85Sr0.15OFeAs have also been prepared.
In the 122 family, superconductivity can again be achieved by both hole doping in
materials such as A1−xBxFe2As2 (A = Ca, Sr, Ba; B = Na, K, Cs)53 or electron doping in
materials such as AFe2−xCoxAs2. For FeSe, the superconducting phase is believed to be the
stoichiometric β-Fe1.01±0.02Se,54 suggesting that electron doping is not required in this
compound. The ambient-pressure Tc of 8.5 K can be raised to ~15 K by partial substitution
of Te for Se, even though FeTe itself is non-superconducting. Under pressure (8.9 GPa), a Tc
of 37 K has been reported in FeSe.
At the time of writing, the iron pnictide superconductors are being intensively
researched, and the origin of their superconductivity is unknown. However, their super-
conducting phase diagram suggests that, like the cuprates, antiferromagnetic spin fluctu-
ations may be important in electron pairing. One hope is that understanding their
properties will help shed further light on the cuprates. The pnictides’ metallic parent
state and their relatively high values of critical field and critical current may also make
them more suitable for certain applications.

12.9 Problems
12.1 Consider the main families of superconducting materials discussed in this chapter:
elemental metals, intermetallic compounds, fullerenes, organic conductors, cuprates,
and iron pnictides. (a) State the highest Tc (in K) observed in each of the different
families. (b) Order the six families in ascending order of the year in which supercon-
ductivity was first discovered. (c) In which families does BCS theory provide a reason-
able description for superconductivity?
12.2 State two properties that distinguish a superconductor from a normal metallic
conductor.
12.3 Al has Tc = 1.14 K. Assuming it is in the BCS weak-coupling limit, calculate the
superconducting gap 2Δ0. Give your answer in J, eV, and cm−1.
12.4 Al has a Debye frequency ωD = 5×1013 s−1 (~1670 cm−1) and N(EF)·V 0 = 0.168.
Estimate the superconducting gap 2Δ0.
12.5 Sn and Pb have Tc = 3.7 K and 7.2 K respectively. When illuminated with far-IR
radiation, they show a marked drop in reflectivity at frequencies of ~10 and ~20 cm−1,
respectively. Rationalize these observations.

52
Magnetic ordering is often closely associated with the tetragonal to orthorhombic transition. While doping
suppresses this transition, superconductivity is observed for both symmetries.
53 54
For example Ba0.6K0.4Fe2As2 has Tc = 38 K. Early reports suggested Fe1+xSe was responsible.
524 Superconductivity

12.6 The table below contains Tc values for different isotopically enriched samples of
Sn and Hg. Assuming Tc ∝ M−α, determine α in each case. Give a short
explanation of why Tc depends on the isotopic mass. Sn data are from ref. [30];
Hg data are from ref. [6].

Sn samples, mean atomic mass 113.58 116.67 118.05 118.7 119.78 123.01
Tc (K) 3.8082 3.7708 3.7442 3.7419 3.7238 3.6651
Hg samples, mean atomic mass 199.7 200.7 202.0 203.4
Tc (K) 4.161 4.150 4.143 4.126

12.7 Draw a sketch of the fcc structure of C60. Show the positions of octahedral and
tetrahedral holes and state the number of each. Assuming that C60 has a radius of
5 Å, estimate the cubic cell parameter and size of these sites. Compare the diameters of
the alkali-metal cations with the size of the octahedral holes. You can find ionic radii at
webelements.com.
12.8 The table below contains data for the alkali-metal fullerides A3C60. (a) Give a brief
explanation for the Tc changes observed in the 0 GPa data. (b) Give a brief explanation
for the pressure dependence of Tc in Rb3C60. (c) Comment on differences in Tc between
the two series.

Compound P (GPa) a (Å) Tc (K)

K3C60 0 14.25 18
K2RbC60 0 14.30 22
K1.5Rb1.5C60 0 14.35 24
Rb2KC60 0 14.36 26
Rb3C60 0 14.44 29
Rb3C60 0.113 14.40 28.1
Rb3C60 0.258 14.37 26.8
Rb3C60 0.434 14.32 25.4
Rb3C60 0.661 14.27 23.5

12.9 RbxCs3−xC60 compounds have Tc = 26.9 K, 32.9 K, and 31.8 K and room-temperature cell
parameters a = 14.70 Å, 14.60 Å, and 14.45 Å for x = 0.35, 1.0, and 2.0, respectively. Predict
the dependence of Tc on pressure for each of these compositions.
12.10 Perovskite-related SrBiO3 and Sr0.4K0.6BiO3 can both be prepared at high pressure.
SrBiO3 has two BiO6 octahedral sites, one with bond distances of 2 × 2.281, 2 × 2.317,
2 × 2.321 Å and a second with bond distances of 2 × 2.140, 2 × 2.209, and 2 × 2.123 Å.
Sr0.4K0.6BiO3 has a single site with Bi–O distances of 2 × 2.110 and 4 × 2.114 Å. Discuss
the origins of these structural differences and predict the differences in low-temperature
 12.9 Problems 525

conductivity you might expect for each material. Calculate the Bi bond valence sum for
0
each site assuming RBi–O = 2.094 Å.
12.11 Although the true structure of La2CuO4 is orthorhombic, it can be approximated
as tetragonal with space group #139 I4/mmm, unit-cell parameters of a = 3.814 Å
and c = 13.15 Å, and crystallographically non-equivalent atoms at the
following coordinates (and Wyckoff sites): Cu 0 0 0 (2a); La 0 0 0.362 (4e); O1
0 ½ 0 (4c); O2 0 0 0.187 (4e). The symmetry generates equivalent positions:
Rotational symmetry generates the additional coordinate ½ 0 0 for Wyckoff site
4c and 0 0 −z for Wyckoff site 4e. Translational symmetry generates equivalent
positions at (½,½,½)+ for all coordinates of the four Wyckoff sites. (a) Sketch the
structure. (b) State the coordination geometry around each metal ion. (c) Calculate
the Cu–O bond distances.
12.12 Depending on the level of oxygen vacancies and temperature, YBa2Cu3O7−δ can be
either a (poor) metallic conductor, an antiferromagnetically ordered insulator, or a
superconductor. A simplified phase diagram separating the different regions is
shown below. (a) Give a brief description of the structure of YBa2Cu3O7−δ,
emphasizing the features relevant for its superconducting properties. (b) Describe
key changes in structural or physical properties of YBa2Cu3O7−δ as you
move along each of the arrows 1, 2, and 3 in the figure. (c) State which character-
ization techniques would give information on each of the arrowed transitions. (d)
Describe how a material with optimum superconducting properties could be
prepared.

12.13 A black material A was prepared by heating 0.3978 g of CuO, 0.1480 g of BaCO3,
and 1.5069 g of La2O3 to 700 °C under N2 and rapidly cooling to room tempera-
ture. A contains 15.79% O by mass and exhibited no weight loss on heating under a
flow of He. A second black material B was prepared by heating La2CuO4 to 1070 °C
for 17 hours in air and then slowly cooled under flow of O2. When 20 mg of B was
heated to 380 °C under flow of Ar, it lost 0.059 mg. Calculate the chemical formulae
526 Superconductivity

of A and B, and indicate how their electronic properties might be expected to differ
from those of pure La2CuO4.
12.14 What is unusual about the superconducting phases Nd2−xCexCuO4 and Sr1−xNdxCuO2?

12.10 Further Reading

W. Buckel, “Superconductivity Fundamentals and Applications” (1991) VCH.
S.J. Blundell, “Superconductivity: A Very Short Introduction” (2009) Oxford University Press.
J.E. Hirsch, M.B.Maple, F. Marsiglio (editors), “Superconducting Materials: Conventional,
Unconventional and Undetermined (Physica C, volume 514)” (2015) Elsevier. This special edition
contains reviews of many different superconducting families. The editorial by Hirsch, Maple, and
Marsiglio gives an overview of the area
P. Batail, “Molecular conductors” Chemical Reviews 11 (2004), 4887–4890 (editorial); 4891–5737
(various authors).
S.E. Brown, “Organic superconductors: The Bechgaard salts and relatives” Physica C 514 (2015),
279–289.
A.W. Sleight, “Bismuthates: BaBiO3 and related superconducting phases” Physica C 514 (2015),
152–165.
J.P. Attfield, “Chemistry and high temperature superconductivity” J. Mater. Chem. 21 (2011),
4756–4764.
C.W. Chu, L.Z. Deng, B. Lv, “Hole-doped cuprate high temperature superconductors”, Physica C 514
(2015), 290–313.
B. Keimer, S.A. Kivelson, M.R. Norman, S. Uchida, J. Zaanen, “From quantum matter to high-
temperature superconductivity in copper oxides” Nature 518 (2015), 179–186.
M.R. Norman, “Unconventional superconductors”, Chapter 13 in “Novel Superfluids, Volume 2”,
Editors: K.H. Bennermann, J.B. Ketterson (2014) Oxford University Press.

12.11 References
[1] H. Kamerlingh Onnes, “Through Measurement to Knowledge: The Selected Papers of Heike
Kamerlingh Onnes 1853–1926” (1991) Springer.
[2] N.W. Ashcroft, “Metallic hydrogen: A high-temperature superconductor?” Phys. Rev. Lett. 21
(1968), 1748–1749.
[3] J.E. Hirsch, M.B. Maple, F. Marsiglio, “Superconducting materials classes: Introduction and
overview” Physica C 514 (2015), 1–8.
[4] H.J. Niu, D.P. Hampshire, “Critical parameters of disordered nanocrystalline superconducting
Chevrel-phase PbMo6S8” Phys. Rev B 69 (2004), 175403.
[5] C. Kittel, “Introduction to Solid State Physics”, 8th edition (2005) John Wiley and Sons.
[6] C. Reynolds, B. Serin, W. Wright, L. Nesbitt, “Superconductivity of isotopes of mercury” Phys.
Rev. 78 (1950), 487.
 12.11 References 527

[7] M.E. Jones, R.E. Marsh, “The preparation and structure of magnesium boride, MgB2” J. Am.
Chem. Soc. 76 (1954), 1434–6.
[8] J. Nagamatsu, N. Nakagawa, T. Muranaka, Yuji Zenitani, J. Akimitsu, “Superconductivity at 39
K in magnesium diboride” Nature 410 (2001), 63–64.
[9] S.L. Bud’ko, P.C. Canfield, “Superconductivity of magnesium diboride” Physica C 514 (2015),
142–151.
[10] O. Gunnarsson, “Alkali Doped Fullerides” (2004) World Scientific.
[11] R.D. Shannon, “Revised effective ionic radii and systematic studies of interatomic distances in
halides and chalcogenides” Acta. Crystallogr. Sect. A 32 (1976), 751–767.
[12] A.F. Hebard, M.J. Rosseinsky, R.C. Haddon, D.W. Murphy, S.H. Glarum, T.T.M. Palstra,
A.P. Ramirez, A.R. Kortan, “Superconductivity at 18 K in potassium-doped C60” Nature
350 (1991), 600–601.
[13] S. Margadonna, Y. Iwasa, T. Takenobu, K. Prassides in “Fullerene based materials structure and
property” Structure and Bonding 109 (2004), 127–164.
[14] R.H. Zadik, Y. Takabayashi, G. Klupp, R.H. Colman, A.Y. Ganin, A. Potocnik, P Jeglic, D.
Arcon, P. Matus, K. Kamaras, Y. Kasahara, “Optimized unconventional superconductivity in a
molecular Jahn–Teller metal” Sci. Adv. 1 (2015), e1500059.
[15] C.C. Chen, C.M. Lieber, “Synthesis of pure 13C60 and determination of the isotope effect for
fullerene superconductors” J. Am. Chem. Soc. 114 (1992), 3141–3142.
[16] C.C. Chen, C.M. Lieber, “Isotope effect and superconductivity in metal-doped C60” Science 259
(1993), 655–658.
[17] M.S. Fuhrer, K. Cherrey, A. Zettl, M.L. Cohen, “Carbon isotope effect in single-crystal Rb3C60”
Phys. Rev. Lett. 83 (1999), 404–407.
[18] B. Burk, V.H. Crespi, A. Zettl, M.L. Cohen, “Rubidium isotope effect in superconducting
Rb3C60” Phys. Rev. Lett. 72 (1994), 3706–3709.
[19] Y. Takabayashi, A.Y. Ganin, P. Jeglic, D. Arcon, T. Takano, Y. Iwasa, Y. Ohishi, M.
Takata, N. Takeshita, K. Prassides, M.J. Rosseinsky, “The disorder-free non-BCS supercon-
ductor Cs3C60 emerges from an antiferromagnetic insulator parent state” Science 323 (2009),
1585–1590.
[20] P. Batail, K. Boubekeur, M. Fourmigué, J.-C.P. Gabriel, “Electrocrystallization, an invaluable
tool for the construction of ordered, electroactive molecular solids” Chem. Mater. 10 (1998),
3005–3015.
[21] J.F. Schooley, W.R. Hosler, M. Cohen, “Superconductivity in semiconducting SrTiO3” Phys.
Rev. Lett. 12 (1964), 474–475.
[22] A.W. Sleight, J.L. Gillson, P.E. Bierstedt, “High-temperature superconductivity in the BaPb1−xBixO3
system” Solid State Commun. 17 (1975), 27–28.
[23] J.G. Bednorz, K.A. Müller, “Possible high Tc superconductivity in the Ba–La–Cu–O system” Z.
Phys. B 64 (1986), 189–193.
[24] J.P. Attfield, A.L. Kharlanov, J.A. McAllister, “Cation effects in doped La2CuO4 superconduct-
ors”, Nature 34 (1998), 157–159.
[25] M.K. Wu, J.R. Ashburn, C.J. Torng, P.H. Hor, R.L. Meng, L. Gao, Z.J. Huang, Y.Q. Wang, C.
W. Chu, “Superconductivity at 93 K in a new mixed-phase Y–Ba–Cu–O compound system at
ambient pressure” Phys. Rev. Lett. 58 (1987), 908–910.
528 Superconductivity

[26] M. v. Zimmermann, J.R. Schneider, T. Frello, N.H. Andersen, J. Madsen, M. Käll, H.F.
Poulsen, R. Liang, P. Dosanjh, W.N. Hardy, “Oxygen-ordering in underdoped YBa2Cu3O6+x
studied by hard x-ray diffraction” Phys. Rev B 68 (2003), 104515/1–13.
[27] R.J. Cava, A.W. Hewat, E.A. Hewat, B. Batlogg, M. Marezio, K.M. Rabe, J.J. Krajewski, W.F.
Peck Jr, L.W. Rupp Jr., “Structural anomalies, oxygen ordering and superconductivity in oxygen
deficient Ba2YCu3Ox” Physica C 165 (1990), 419–433.
[28] C.W. Chu, L.Z. Deng, B. Lv, “Hole-doped cuprate high temperature superconductors” Physica
C 514 (2015), 290–313.
[29] Y. Kamihara, T. Watanabe, M. Hirano, H. Hosono. “Iron-based layered superconductor
LaO1−xFxFeAs (x = 0.05–0.12) with Tc = 26 K” J. Am. Chem. Soc. 130 (2008), 3296–3297.
[30] E. Maxwell, “Superconductivity of the isotopes of tin” Phys. Rev. 86 (1952), 235–242.
13 Energy Materials: Ionic Conductors, Mixed
Conductors, and Intercalation Chemistry

In previous chapters we have concentrated on properties of materials that are largely derived
from the electrons they contain. In this chapter and the next, the focus will change to
materials in which the main applications derive from the transport of ions, atoms, or
molecules through them. Long-range atomic, ionic, or molecular migration in a solid may
initially come as a surprise, especially given the conventional view of solids as strongly
bonded arrays, with atoms undergoing only small vibrations around their mean positions.
Some solids do, however, allow high ionic migration in the solid state. We will see that they
can have important applications as electrolytes in batteries, sensors, and fuel cells. We’ll look
at the general requirements for this functionality, then focus on conductors of metal cations,
protons, and oxide anions. We will also consider mixed ionic electronic conductors: materials
that can conduct both ions and electrons and are used as electrodes in many energy-related
devices. Finally, we will discuss intercalation chemistry that involves the reversible insertion/
removal of chemical species into a host structure. This chemistry is exploited in most
lightweight rechargeable batteries.

13.1 Electrochemical Cells and Batteries

Since energy storage and production is the reason why many of the materials in this chapter
are of interest, it’s worth reminding ourselves of the components of an electrochemical cell,
Figure 13.1. Such a cell consists of electrodes separated by an electrolyte. In some cases (such
as the Zn(s)|ZnSO4(aq)||CuSO4(aq)|Cu(s)1 cell shown), the electrode compartments might be
physically separated by a salt bridge.
A galvanic cell is one in which a spontaneous reaction produces electricity, and an
electrolytic cell is one in which a non-spontaneous reaction is driven by an external current
source. The anode is always defined as the electrode at which oxidation occurs; the cathode as

1
In this notation, a solid line | represents a phase boundary between cell components and a double line || a phase
boundary such as an ideal salt bridge that doesn’t contribute to the cell potential.

529
530 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

E°cell = 1.10 V
e−
V
anode cathode
negative salt bridge positive

Zn Cu
SO42−

1M ZnSO4 1M CuSO4

Figure 13.1 Example of a galvanic cell under standard conditions, with standard cell potential E°cell. The
salt bridge (typically KCl dissolved in agar) allows ion flow without excessive mixing of electrolytes.

where reduction occurs. In a galvanic cell such as the one shown in Figure 13.1, oxidation of
Zn metal occurs at the anode, Zn2+ ions pass into solution, and the electrons left behind give
the anode a negative charge. These electrons flow through the external circuit to the cathode.
Here, Cu2+ ions being reduced from solution attract the electrons, giving the cathode
a positive charge. The circuit is completed by the flow of ions through the salt bridge and
the electrolytes.
In an electrolytic cell, the spontaneous direction of the reaction is reversed using an
external current source. Reduction must now occur at the electrode where oxidation
occurred originally. This can only be achieved by supplying electrons, i.e. giving that
electrode a negative charge. Since the electrode at which reduction occurs is always defined
as the cathode, the cathode is negative (and anode positive) in an electrolytic cell.
In any galvanic cell that has not reached its final state of chemical equilibrium, the
movement of electrons through a potential difference can be used to do electrical work: the
higher the potential difference and larger the number of electrons, the more work. The
maximum amount of work occurs when the cell operates reversibly at a fixed chemical
composition. This occurs when the electric current approaches zero (the resistance of the
voltmeter approaches infinity) and the measured voltage is called the cell potential, Ecell.2 The
zero-current cell potential is related to the Gibbs free energy of the cell’s chemical reaction by:

DG ¼ nFEcell (13.1)

where n is the number of electrons in the cell half equations and F is Faraday’s
constant (96485 C/mol). A spontaneous reaction (ΔG negative) means a positive cell
potential. The cell potential depends on the chemical state of the cell. When it is far
from equilibrium, the cell potential will be high; when it is close to equilibrium, the cell
potential will be close to zero.
2
Or the zero-current potential or (obsolete) electromotive force (emf) of the cell.
 13.1 Electrochemical Cells and Batteries 531

For a cell such as Zn(s)|ZnSO4(aq)||CuSO4(aq)|Cu(s) (Figure 13.1), we can use

standard (reduction) electrode potentials3, E°, to determine both the direction of
spontaneous change and the cell potential under standard conditions. We write down
half equations for the two electrodes and realize that the oxidizing agent will be in the
half equation with higher standard (reduction) potential. In our cell, the oxidizing
agent is Cu2+ because E° = +0.34 V for the Cu2+(aq) + 2e− → Cu(s) half equation
and the reducing agent is Zn because E°= −0.76 V for Zn2+(aq) + 2e− → Zn(s). The
total reaction is then Cu2+(aq) + Zn(s) → Cu(s) + Zn2+(aq), and the standard cell
potential is the difference of the two E° values: 1.10 V.
Under non-standard conditions, the electrode potential is given by the Nernst equation:

RT
E ¼ E°  lnðQÞ (13.2)
nF

where Q is the reaction quotient (a mass-action term) for the half equation. Since the mass-
action term depends on activities or concentrations of species present, we can see how
electrochemical cells can be used as chemical sensors—they will give a changing electrical
signal as the chemical concentration of a particular metal or ion changes. Application of the
Nernst equation is explored in the end-of-chapter problems.
These basic ideas have led to a number of different battery designs based on different
chemical processes. A battery is, strictly speaking, an array of cells connected together, but
the term is used loosely. Primary batteries are based on reactions that are irreversible under
normal conditions and can only be used once. Secondary batteries have reversible reactions
and can be recharged.
The standard disposable alkaline cell is a primary battery. It relies on the reaction between
a Zn anode (connected to the flat base) and an MnO2 cathode (connected to the raised button)
with a KOH electrolyte (hence the alkaline cell name). The cathode is a cylinder of MnO2 powder
mixed with carbon black to improve conductivity, into which a central cylinder of a gel
containing zinc anode powder and electrolyte is inserted. The surface shared by the two cylinders
is electrically and mechanically isolated by an ion-conducting porous polymeric membrane.
The most familiar secondary battery is the lead–acid rechargeable car battery. It contains
Pb and PbO2 plates in a sulfuric acid electrolyte when charged, which both form PbSO4 on
discharge. Rechargeable “nickel–cadmium” batteries were used in many portable devices
and rely on the reversible reaction of NiIII-containing NiO(OH) and Cd to produce Ni(OH)2
and Cd(OH)2. In many devices, these were replaced by “nickel–metal hydride” batteries.
These again rely on an NiO(OH) cathode that is reduced to Ni(OH)2 as the cell is discharged.
The hydride reducing agent is stored in the charged cell in a metal hydride anode based on the
LaNi5 alloys discussed in Chapter 2. The chemistry of these traditional batteries is explored
further in the end-of-chapter problems.

3
The potential versus the standard hydrogen electrode under unity activities. Standard reduction potential because it is
a potential for reduction of the oxidized species in the redox couple.
532 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

13.2 Fuel Cells

A second energy-producing device that requires high-functionality materials is the fuel cell.
Fuel cells again convert chemical energy into electrical energy but, in contrast to batteries,
the fuel and oxidant are supplied from an external source. Fuel cells have high efficiency
(60–70% or up to 90% if waste heat is also used), no moving parts, are quiet to operate, and
can have zero or low polluting emissions at the point of use. There are many possible designs,
and Figure 13.2 shows a configuration using H2 as fuel. In this cell, H2 releases electrons at
the anode, which flow through an external circuit. The protons formed migrate through the
electrolyte, and, at the cathode, combine with electrons and oxygen to form H2O. The Gibbs
free energy of H2(g) + ½O2(g) → H2O(l) (ΔG = −236 kJ/mol) can, in theory, be used to do
electrical work; the work is limited in practice by ohmic-heat losses and residual concentra-
tions of the two reactants. At low temperatures, a Pt catalyst is typically employed at both
electrodes to speed up reaction rates.
Fuel cells are used in a variety of large-scale applications, and several companies have
demonstrated small devices that could replace rechargeable batteries in some portable
appliances [1]. Other applications of similar technology include reversing the direction in
which the fuel cell operates to produce an electrochemical H2 pump capable of removing H2
from a system. In a NEMCA (nonfaradaic electrochemical modification of catalytic activity)
reactor, this has been applied to reactions such as the dehydrogenation of ethane or conver-
sion of methane to other hydrocarbons. We will discuss the materials for different fuel-cell
designs later in the chapter.

e− e−
excess H2 H2O out
2H+ + 2e− + ½ O2 = H2O
H2 = 2H+ + 2e−

H+

H2 in air in

proton-
conductor
membrane
porous anode & catalyst porous cathode & catalyst

Figure 13.2 A schematic diagram of a hydrogen fuel cell with a proton-conductor electrolyte. 2–5-nm Pt
nanoparticles are typically used as the catalyst.
 13.3 Conductivity in Ionic Compounds 533

13.3 Conductivity in Ionic Compounds

In Chapter 10, we defined conductivity as σ ¼ L=RA where L, R, and A are the sample length,
resistance, and area. Values of conductivity vary over a huge range from <10−20 S/cm for
insulators to >1025 S/cm for the superconductors discussed in Chapter 12.4 Figure 13.3
reviews the typical ranges for different categories of conductors. Although ionic compounds
such as NaCl are considered insulators at low temperatures, their conductivity rises dramat-
ically on heating and can approach the values of semiconductors. Certain ionic materials
such as AgI and PbF2 have conductivities approaching those of metals even at relatively
modest temperatures. How does this occur?
In Chapter 2, we saw that the structures of ionic materials always contain defects. NaCl, for
example, will always have a certain level of cation vacancies due to minor aliovalent impurities of
the type Na1−2xCax □ xCl (extrinsic defects). In addition, at temperatures above 0 K, it is
thermodynamically favorable to form intrinsic Schottky defects (which predominate in NaCl)
or Frenkel defects (e.g. in AgCl). When NaCl is heated, Na+ can hop to an adjacent vacant
defect site as described in Section 3.3.1 and shown schematically in Figure 13.4. The position of
the vacancy has then moved, and a different Na+ can hop into the newly vacant site. Under an
applied field, the cation vacancy (which has a formal negative charge; see Section 2.2) can
migrate giving rise to a net current. The overall conductivity will depend on the number of
vacant sites available, as we know from the relationship between conductivity (σ), the number of
charge carriers (n), their charge (z), and their mobility (μ), developed in Chapter 10:

semiconductors

insulators metals
conductivity,
10−16 10−12 10−8 10−4 100 104 σ (S/cm)

RbAg4I5
25 °C
Y0.1Zr0.9O1.95 β
Na-β-alumina
750 °C 300 °C

σ (S/cm)
10−5 10−4 10−3 10−2 10−1 100 101

NaCl β-AgI NaCl β-AgI PbF2

500 °C 125 °C 900 °C 146 °C 600 °C

Figure 13.3 The range of conductivities σ, in S/cm, for different conductors.

4
We use siemens per centimeter (S/cm) for conductivity in this chapter rather than siemens per meter (S/m) as these are
the units most commonly used in the literature; 1 S/cm is 100 S/m.
534 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

interstitial migration

interstitialcy mechanism

vacancy migration vacancy

Na+ Cl−

Figure 13.4 Left: Schematic mechanisms of vacancy and interstitial migration. Right: Na+ migration
from an occupied to a vacant site.

σ ¼ nzμ (13.3)

For a compound where interstitial defects dominate, such as AgCl, one can imagine two
mechanisms that allow charge movement. Given the large number of interstitial sites
available, one possibility is that direct hopping occurs from an occupied interstitial to
a vacant interstitial. A second possibility is that an Ag+ cation hops from an interstitial
site to a normal lattice site, provided that in so doing it drives the Ag+ off this site and pushes
it onto a new interstitial position: a “billiard ball” or interstitialcy-type mechanism. In AgCl,
this latter mechanism is believed to predominate.
In Section 3.3, we explored the thermodynamics of vacancy migration quantitatively, and
we can use these ideas to predict how the conductivity of a simple ionic compound like NaCl
will vary with temperature. For NaCl with a normal level of M2+ impurities, the dominant
source of low-temperature charge carriers will be extrinsic cation vacancies. Their number
will not change with temperature, and any temperature dependence of conductivity will be
governed by changes in ionic mobility. For Na+ migration to occur, the Na+ ion must pass
from one octahedral site to another (Figure 13.4). Regardless of the precise pathway, there
will be an activation energy associated with this process since the Na+ must pass through an
unfavorable environment.
In Section 3.3.8, we saw that the diffusivity D in this situation has the form
D ¼ D0 expðEA =kTÞ with D0 ¼ λ2 fc pdir ν, where λ is the jump distance, ν is a vibrational
frequency, pdir is a crystal-structure-dependent probability of a given jump moving an ion
forwards, and fc is a correlation factor related to the probability of an ion returning to its
original site. For a simple 1D case with each jump allowed, this simplifies to D0 ¼ λ2 ν. In
Section 3.3.6, we derived the Nernst–Einstein equation that relates the conductivity of
a species to its diffusivity via σ=D ¼ q2 n=kT, where we use n as the number of mobile ions
per unit volume in place of c in Equation (3.23) and q is the charge. This relationship applies
 13.3 Conductivity in Ionic Compounds 535

when the same random-walk mechanism applies for both conduction and diffusion. We can
then express the conductivity as:
 
λ2 νq2 n Em
σ¼ exp (13.4)
kT kT

where Em is the energy barrier height for vacancy (or interstitial) diffusion. A plot of
ln(σT) versus 1/T will therefore give the activation energy for vacancy migration from
its slope of −Em/k. For samples of NaCl with different dopant levels, provided the
activation energy doesn’t change significantly on doping, the conductivity will be
determined largely by the number of dopants in the pre-exponential factor of
Equation (13.4).5 For different doping levels, plots of ln(σT) against 1/T would there-
fore give a series of parallel lines (see Figure 13.5).
We saw in Chapter 2 that the number of intrinsic defects increases rapidly as temperature
increases, such that we may have to consider the contribution of both extrinsic and intrinsic
defects to conductivity. The intrinsic contribution will be particularly important for high-
purity samples at high temperature. In Section 2.3, we derived the number of Schottky
defects (ignoring vibrational contributions to entropy) as:

temperature (°C)
700 600 500 400 300

-1

-3 intrinsic
log(σT) (KScm−1)

-5
extrinsic
-7

-9
higher doping

-11

-13
1.0 1.2 1.4 1.6 1.8
1000/T (K−1)

Figure 13.5 Plots of ln(σT) against 1000/T for an ionic compound at two levels of extrinsic defects: high
( ■ ) and low ( □ ) doping. Data are calculated with Equation (13.6) for an NaCl crystal with activation
energies of Em = 0.70 eV and ES = 2.37 eV.

5
Note the 1/T term in the pre-exponential factor arises in the presence of an applied field but isn’t present in the
equivalent expressions using diffusivity. If a plot of ln(σ) versus 1/T [as opposed to ln(σT) versus 1/T] is used to
determine activation energies from conductivity measurements, values may be in error by ~10 kJ mol−1.
536 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

n ¼ N0 expðDHS =2kTÞ (13.5)

where N0 is the number of sites per cm3 of sample and ΔHS is the formation enthalpy of
a Schottky defect pair. This can be introduced into Equation (13.4) giving:

λ2 νq2
σ¼ N0 expðDHS =2kTÞexpðEm =kTÞ (13.6)
kT

A plot of ln(σT) against 1/T (Figure 13.5) will then give two broad regions: one where
vacancies from extrinsic doping dominate conductivity with a slope of −Em/k (low T), and
one where intrinsic vacancies dominate with slope ðEm þ DHS =2Þ=k (high T). Between these
two approximately linear regions, a curved region will occur where extrinsic and intrinsic
contributions are comparable.
It’s worth noting that while doping a compound like NaCl, where vacancy migration
dominates, with a 2+ ion will increase conductivity, in AgCl, where interstitial migration
dominates, the conductivity may decrease. This happens because doping will create add-
itional vacancies on normal Ag+ sites (vAg 0 ), which, by simple equilibrium considerations,
will decrease the number of Ag+ interstitials (the product of the number of interstitials and
the number of vacancies remains constant). At higher levels of doping, the number of
vacancies may rise sufficiently that, despite lower mobility compared to interstitial sites,
their increasing number may cause the overall conductivity to rise again.

13.4 Superionic Conductors

In some compounds, ionic conductivity in the solid state approaches or exceeds that of the
molten state (Figure 13.6). Such materials are referred to as superionic or fast ionic conduct-
ors. There is, however, nothing intrinsically “super” about the conduction mechanism
(certainly it shouldn’t be equated to the electronic superconductors of Chapter 12), nor is
the intrinsic migration rate in these materials particularly fast. The key feature of
these materials is that their structure (crystalline or amorphous) gives a large overall value
of n × z × μ in Equation (13.3).

13.4.1 AgI: A Cation Superionic Conductor

One material that illustrates many of the features required for high cationic conductivity is
AgI. At room temperature, it adopts the wurtzite (β-AgI) or sphalerite (γ-AgI) structures
(Figure 1.32) with Ag+ occupying half the tetrahedral holes of a hexagonally or cubic closest
packed (hcp or ccp) array of I−, respectively. Upon warming to 420 K, there is a transition to
α-AgI, in which iodide anions adopt a body-centered cubic (bcc) packing. This is accompan-
ied by a ~104-fold increase in conductivity to values near 2 S/cm. The conductivity actually
 13.4 Superionic Conductors 537

2
Tmelt = 1103 K

Tmelt = 829 K

log(σ) (S/cm)
-2 AgI
Tα→β = 420 K PbF2

-4

NaCl
-6
Tmelt = 1074 K

-8
200 400 600 800 1000 1200
temperature (K)

Figure 13.6 Conductivities of the archetypal fast cation conductor AgI and fast anion conductor PbF2
compared to NaCl. After ref. [2].

drops by 10% when the material melts at 829 K. The electronic contribution to the overall
conductivity is around 107 times lower than the ionic.
The bcc arrangement of two I− ions per α-AgI cell creates a large number of different sites
that could be occupied by two Ag+ ions: Wyckoff site 12d on the cell face, which has
a distorted tetrahedral coordination environment of I−; site 6b on the cell edge (a distorted
octahedron); and 24h (distorted triangle); see Figure 13.7.6 Neutron-scattering experiments
and simulations suggest that Ag+ cations spend around 75% of their time on tetrahedral sites,
with hops between positions via trigonal sites being around six times as likely as those via
octahedral sites. Calculations also suggest that after a given hop, there is a slight preference
for the ion to migrate back to its starting position.7 The diffuse orbitals of the large
polarizable I− anion facilitate low-energy pathways for Ag+ migration, and α-AgI has an
extremely low activation energy for ionic migration of 0.03 eV (~3 kJ/mol). The Ag+ ions in
α-AgI are sufficiently disordered that they have been described as “liquid like” above 420 K,
and the β/γ → α transition is frequently described as a “melting” of the cation sublattice. This
view is supported by thermodynamic measurements, which show an entropy gain of ΔS =
+14.5 J/(mol K) for the β → α transition and +11.3 J/(mol K) for melting of AgI at 829 K [3].
The sum of these two values is similar to the entropy of melting of a “normal” ionic solid such
as NaCl of ~24 J/(mol K). The structural evidence, however, suggests that a picture of

6
Bond-valence sums for Ag+ at the three sites are 1.1, 1.2, and 1.4, respectively. The sites are highly distorted from
ideal geometries: the “tetrahedral” site has bond angles of 101.5° and 126.9°; the triangular site has bond angles of
109.5° and 141.1°; and the “octahedral” site has bond distances of 2 × 2.5 Å and 4 × 3.6 Å.
7
After an initial hop, we would statistically expect a 25% probability of Ag+ hopping to each neighboring tetrahedral
site. There is actually a ~40% chance of a hop back to the original position.
538 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

I− I−
I− I−
12d Ag+ distorted tetrahedron

6b Ag+ distorted octahedron

I−
24h Ag+ distorted triangle of I−

I− I−
I− I−

Figure 13.7 Cubic α-AgI showing Im3m sites that can accommodate Ag+. Distorted tetrahedral sites are
small black spheres with gray lines showing how they link via trigonal sites (small gray spheres). Distorted
octahedral sites are the small white spheres at the cell edges.

frequent discrete hops between specific cation sites is a more appropriate description than
a molten array.
α-AgI exemplifies many of the key requirements for a material to show high conductivity:

1. A large number of mobile ions to maximize n in σ = nzμ.

2. An arrangement of oppositely charged ions that creates energetically favorable sites into
which the mobile ion can hop, with low-energy migration routes between them.8
3. Polarizable immobile ions that can deform their electronic clouds to allow migration of
mobile ions.
4. The absence of conditions that lead to trapping of mobile ions, such as attractions to the
local charges of aliovalent dopants or vacancies; a process that leads to significant aging
(deterioration of conduction over time) in some ionic conductors.
5. A low charge on mobile ions to minimize the activation energy for migration.
6. Other things being equal, small ions will have a lower activation energy for migration than
large ones.

Many attempts have been made to further increase the ionic conductivity of AgI.
Substitution on the cation site has led to materials such as RbAg4I5 (which has one of the
highest room-temperature ionic conductivities known, σ = 0.21 S/cm), Ag1−xCuxI, and many

8
It is important to realize that conduction requires both mobile cations and vacant sites for them to hop to. In several
materials, vacancies are introduced by doping, and one finds a maximum in conductivity when one has 50%
vacancies and 50% ions on the site of interest. A good example occurs in Li4–3xAlxSiO4. As Al is introduced,
vacancies start to appear on one Li site that is completely empty by x = 0.5. When x = 0.25, the mobile Li site is 50%
occupied and 50% vacant and maximum conductivity is observed [A. García, G. Torres-Trevino, A.R. West, Solid
State Ionics 13 (1990), 40−41].
 13.4 Superionic Conductors 539

other materials. Substitutions on the anion sublattice leads to materials such as Ag3SI
and Ag2S. At high temperatures, the structure of β-Ag2S is similar to α-AgI but with
twice as many of the tetrahedral sites occupied. Derivatives with oxoanions, such as
Ag6I4WO4, have also been investigated. Some have been successfully used as electro-
lytes in batteries.
For those fascinated by the ubiquity of the perovskite structure throughout materials
chemistry, it’s worth noting Ag3SI. The α*-Ag3SI form has a high conductivity of ~0.3 S/cm
at room temperature and a structure like that of α-AgI but with anions disordered over the
bcc sites. The thermodynamically more stable β form is ordered and contains I− at the center
of the unit cell, S2− at the corners, and has Ag+ about halfway along each cell edge. As such,
the structure is an anti-perovskite ISAg3.

13.4.2 PbF2: An Anionic Superionic Conductor

In the same way that AgI exemplifies many of the key features of cationic conductors, PbF2
can be considered the archetypal anion conductor. Indeed, its unusual properties were
reported as long ago as 1838 by Faraday. He noted that solid PbF2 acted as an insulator
but that on heating it became conducting long before it melted. The conductivity of PbF2 as
a function of temperature is shown in Figure 13.6. In contrast to AgI, it rises continuously
from ~10−7 S/cm at ambient temperature to ~4 S/cm at 711 K before leveling off. The
conductivity shows no significant change on melting at 1103 K. Despite the fact that there
is no abrupt change in conductivity with temperature, evidence from heat-capacity and
thermal-expansion data suggest that a transition to the superionic state can be defined as
occurring at Tc = 711 K.
At low temperature, anion-Frenkel defects prevail in PbF2. The interstitial F− is thought
to adopt a position at the center of the ccp cell of Figure 13.8 (marked as F4). This site is
equivalent to the anion site in the rock-salt structure. A large number of these sites is
available (one per formula unit), which favors high conductivity. As the material is heated
and the number of Frenkel defects increases above the ~1% level, the proximity of neighbor-
ing defects is thought to destabilize this site such that it is not significantly occupied in the
superionic state. Instead, the superionic state is thought to contain the interstitial ions
predominantly at site F1 in Figure 13.8; midway between two anions but displaced towards
the center of an empty anion site. A second anion site, F2, located approximately 1.5 Å from
a cube corner along the body diagonal towards its center, is also observed. This location is
thought to be due to anions close to F1 interstitials, relaxing away from their ideal positions.
Sites F1 and F2 aren’t occupied simultaneously in any local region, but appear in the average
structure due to local structural clusters distributed throughout the material. One model that
fits neutron-scattering data well has these local clusters containing vacancies, interstitial
anions, and anions relaxed off their ideal position in a 3:1:2 ratio. The Frenkel defects in
these clusters move dynamically through the crystal in the superionic state and have
540 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

symmetry-generated
interstitial sites

F1 unique
F4
interstitial sites
F− F2

Pb2+ Pb2+

Figure 13.8 The fluorite structure of PbF2 is a ccp array of Pb2+ with F− in tetrahedral holes (left). It can
also be described as an array of F− cubes with centers alternately empty or filled by Pb2+. Possible F−
interstitial sites in empty cubes shown on right as small dark gray spheres; labels are discussed in the text.

a lifetime of the order of 10−12 s. Similar conduction mechanisms are believed to occur in the
fluorite-related oxygen-ion conductors of Section 13.7.1.

13.5 Cation Conductors

In the following sections we will discuss some of the important categories of cationic
conductors, with a focus on materials that have found technological applications.

13.5.1 Sodium β-alumina

Sodium β-alumina is one member of a family of materials discovered in the 1960s, which
have sufficiently high Na+ conductivity to be used as electrolytes in high energy-density
rechargeable sodium–sulfur batteries. Traditionally, the formula of these materials has been
expressed as Na2O∙nAl2O3, with n ranging from ~8 to ~11. To understand the structure, it’s
probably easier to express the formula as NaAl11O17 (the n = 11 member), while keeping in
mind that the material can contain extra Na+, charge-balanced by O2−, up to a composition
of Na1.375Al11O17.185 (n = 8). These limiting compositions correspond to Na0.18Al2O3.09 and
Na0.25Al2O3.125, respectively.
The ideal structure of NaAl11O17 is shown in Figure 13.9 and is related to that of spinel,
MgAl2O4, which has a ccp arrangement of O2− ions with Mg2+ filling one-eighth of the
tetrahedral holes and Al3+ half of the octahedral holes. Blocks with cation site occupancies
similar to those in spinel are present in sodium β-alumina, but in every fifth layer, three-
quarters of the ccp O2− ions are missing. The Na+ ions are located in these anion-deficient
layers, which contain a mirror plane such that the layer of oxygen atoms immediately above
the plane is the same as that below. The occupied Na+ sites (often referred to as Beevers–Ross,
 13.5 Cation Conductors 541

spinel
octahedral and tetrahedral Al
block

O mirror
plane = m site
A = abr site
C
br site
B
A O
vacancy Na
layer Na O
4
3
2
1

Figure 13.9 Sodium β-alumina. The side-on view (left) emphasizes the four spinel-like ccp layers of O2−
(large spheres), with Al3+ in tetrahedral and octahedral holes, separated by a single sodium/oxygen
vacancy layer. A view perpendicular to the layers is shown on the right.

or br, sites) have coordination number nine (three O atoms below/above at 2.83 Å and three in
plane at ~3.23 Å). The bond-valence sum for a Na+ cation at the br site is 0.43, significantly
lower than would normally be expected.
There are, in addition, other potential Na+ sites possible. These are referred to as m (for
midway, marked with a triangle in Figure 13.9) and abr (for anti-Beevers–Ross, marked with
a square). These sites would be eight- and five-coordinate and have Na+ bond-valence sums
of 0.43 and 0.48, respectively. We therefore have a 2D network of possible Na+ sites with
similar bond-valence sums. As a result, the material has both a high conductivity (σ = 0.1 S/
cm at 573 K) and a low activation energy (0.16 eV) for migration (see Figure 13.11). The
mechanism for conductivity is thought to be of the billiard-ball type with Na+ migrating
from its original site via a pathway br → m → abr → m → br. When it reaches an m site
adjacent to an occupied br site, that cation is expelled from its position.
The exceptional properties of sodium β-alumina led to the development of the sodium–sulfur
rechargeable cell (Figure 13.10) that operates at around 300 °C. It comprises a liquid-sodium
anode separated from a liquid-sulfur/carbon cathode9 by a thin tube of sodium β-alumina. The
use of liquid electrodes ensures good contact with the solid electrolyte. Upon discharge, Na from
the anode releases electrons that flow through an external circuit, and the Na+ ions formed
migrate through the β-alumina electrolyte to the cathode where they combine with liquid sulfur
and electrons from the external circuit to form Na2Sx. The cell can be operated over
a composition range of Na2S5.2 to Na2S2.7 and gives a voltage that varies from 2.07 V to
1.78 V.10 The reaction 2Na + 3S → Na2S3 could theoretically yield 760 W h/kg, and batteries
9
Carbon is added to increase conductivity.
10
Note that Na2S5 contains two Na+ cations and one polysulfide S52− anion.
542 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

load

t erminal
anode: 2Na(l) → 2Na+ + 2e−
cathode: xS(l) + 2e− → Sx2−
overall: 2Na(l) + xS(l) → Na2Sx(l)
insulator

metal insert

Na Na electrode Na

Na β-alumina

sulfur

cell container

Figure 13.10 A cut-away diagram for one design of a sodium–sulfur rechargeable cell, of about 9 × 50 cm.
Right: A battery of many cells within an electrically heated thermal enclosure with an output power of 50
kW and a storage capacity of 360 kW h. 3D images and photograph provided by NGK Insulators Ltd
(www.ngk.co.jp).

with an energy density around 150 W h/kg have been built.11 They have been shown to have
around 86% efficiency on charge/discharge and can operate for thousands of cycles. Their main
practical applications are as backup power supplies and load levelers. In the latter application,
they help meet energy demands from power stations at peak times or smooth the output of wind-
or solar-power facilities. For example, a 34 MW system containing around 250000 cells has been
installed by NGK Insulators at a Japanese wind farm [4].
There are many materials closely related to sodium β-alumina with similar properties. The
so-called β″ aluminas have a related but subtly different structure and occur at lower Na
content. The β″ structure can also be stabilized with dopants such as Li+ and Mg2+ and can
incorporate a range of other cations such as K+ and Ag+.

13.5.2 Other Ceramic Cation Conductors

Many other materials have been examined for high cationic conductivity; the conductivities of
some are compared with materials already discussed in Figure 13.11. One family that has
undergone extensive investigation are the so-called NASICON phases (for Na Super Ionic
Conductors). These are solid solutions of the type Na1+xZr2(PO4)3−x(SiO4)x, in which a variety
of Na+ sites are occupied within a framework of corner-linked ZrO6/2 octahedra and (P,Si)O4/2
tetrahedra. The two end members of the series are relatively poor ionic conductors, but the x = 2

11
Compared to 20–40 W h kg–1 for lead–acid cells.
 13.5 Cation Conductors 543

temperature (°C)
800 400 200 100 0
1

0 RbAg4I5

Na β-alumina
-1

log(σ) (S/cm)
Li0.34La0.51TiO2.94
-2
Na3Zr2(PO4)(SiO4)2
-3

-4
Na β''-alumina
-5 PEO8LiClO4

-6
0.5 1.5 2.5 3.5 4.5
1000/T (K−1)

Figure 13.11 Electrical conductivities for selected cationic conductors. PEO = polyethylene oxide.

member Na3Zr2(PO4)(SiO4)2, which has a mixture of full and empty Na+ sites, has a high
conductivity of ~0.2 S/cm at 573 K and activation energy of ~0.3 eV.
A number of ceramic phases showing high Li+ conductivity have been investigated, but all
have much lower conductivity than liquid electrolytes formed by dissolving salts such as
LiPF6 in non-aqueous solvents at low temperature. Examples include Li2SO4, Li4SiO4 and
derivatives, and various sulfides. The name LISICON has been given to conductors based on
solid solutions of the type Li2+2xZn1−xGeO4 (−0.36 < x < 0.87) with Li3Zn0.5GeO4 having
σ = 0.125 S/cm at 573 K. Perovskites such as Li0.5−3xLa0.5+xTiO3, also show high conductiv-
ity; these contain A-site layers that alternate between La3+ and a mixture of La3+, Li+, and
vacancies, with Li+ displaced from the ideal A-site position.

13.5.3 Polymeric Cation Conductors

In addition to ceramic ionic conductors, there is considerable interest in polymer
electrolytes for many applications. These are potentially easier to process and more
readily integrated into devices than ceramic materials. They also offer the benefit of
intrinsic flexibility: as other components of devices change size due to temperature and/
or chemical changes, the polymer can adjust without cracking. Compared to liquid
electrolytes, polymers eliminate problems related to leakage of flammable/toxic solvents
from devices and have mechanical stabilities that remove the need for spacers to
separate electrodes.
Polymer electrolytes can be divided into two main categories: polymer–salt complexes
and polyelectrolytes. In the former, an ionic salt is dissolved in a polymer matrix leading
544 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

to the possibility of both cation and anion motion. In the latter, charged groups are
covalently bound to the polymer backbone such that only the counter ion is mobile. The
NAFION proton conductor that will be described in Section 13.6.1 is one example of
a polyelectrolyte.
The early breakthroughs in the area of polymer–salt complexes were made in the mid
1970s when it was realized that materials such as polyethylene oxide [PEO, (CH2CH2O)n]
could dissolve large quantities of salts (in excess of 2 mol/L in some cases) and that the
resulting materials have high ionic conductivity. A wide range of simple ionic salts have
been investigated as well as salts containing large anions such as [(CF3SO2)3C]−,12 in
which the delocalization of the negative charge over several atoms aids solubility.
Conductivities up to around 10−4 S/cm at room temperature can be achieved, which,
although considerably lower than high-temperature ceramic conductors or liquid elec-
trolytes, are high enough for polymer electrolytes to be of technological importance.
PEO is a semicrystalline solid that contains helical (CH2CH2O)n chains, with two
turns of the helix occurring every 19.3 Å. While the majority of the best PEO-derived
conductors are amorphous, it has been possible to gain insight into the structures of
their crystalline counterparts using powder-diffraction methods (single crystals are
extremely difficult to prepare). For example, in (PEO)3·NaClO4, Na+ ions are accom-
modated within helical polymer chains and coordinated by four ether oxygens and
two oxygens from the ClO4− anions. In other compounds, such as (PEO)6·LiSbF6
(Figure 13.12), cations are fully coordinated by the polymer chains and completely
isolated from the charge-balancing anions. The polymer chains provide a large num-
ber of donor O atoms, giving rise to an array of available cation sites along the
chains. These pre-formed cation sites may be one reason why the crystalline form of
this material has a higher conductivity than its amorphous counterpart, even above
the glass-transition temperature of the latter [5].
For the majority of polymer–salt complexes, the conductivity of amorphous materials is
several orders of magnitude higher than crystalline analogues. This is related to segmental
motion contributing to ionic migration. In addition to local bond vibrations, so-called crank-
shaft torsional motions around C–C or C–O bonds occur in an amorphous polymer above its
glass-transition temperature Tg. These motions are believed to promote the migration of ions
by breaking bonds in one local coordination environment of Li+ and simultaneously creating
a more favorable nearby location for the ion to migrate to. This mechanism leads to an
enhanced dependence of conductivity on temperature over the other materials we discuss,
and polymer–salt electrolyte conductivity is often described using the Vogel–Tammann–
 
Fulcher expression σ ¼ σ0 exp  B=ðT  T0 Þ , which is discussed further in Section 15.8.

12
Tris[(trifluoromethyl)sulfonyl]methanide(1−).
 13.6 Proton Conductors 545

SbF6−

Li+
O

C
H

Figure 13.12 The structure of the crystalline polymer–salt complex (PEO)6·LiSbF6. Left: Axial view of
helical polymer chains that surround Li+ to create tubes interspersed with SbF6− octahedra. Right: Side-
on view of a single helical tube.

13.6 Proton Conductors

Materials that display high proton conductivity are required for a number of applications.
One is in the low-temperature fuel cell we discussed in Section 13.2, which relies on proton
migration through a membrane separating fuel and oxidant to turn chemical into electrical
energy. Unfortunately, the highest-conductivity materials only work well at low temperat-
ures, where the only effective electrocatalyst is Pt that is both expensive and prone to
deactivation by CO poisoning.13 The challenge in this area is to identify a material that
remains conducting at a temperature high enough to use cheap electrode materials and to
either tolerate low-purity H2 or to run directly on fuel sources such as methanol. A number of
materials exhibiting high proton mobility, both organic and inorganic, have been discovered,
and conductivity data of some are compared in Figure 13.13.
The mechanisms of proton conductivity are somewhat different to those encountered in
other cationic conductors, due to the unique bonding requirements of the proton. In the
electronically insulating materials required for pure proton conduction, hydrogen typically
forms a short bond to oxygen (~0.95 Å) as well as an additional weaker interaction to a more
distant oxygen; a hydrogen bond. The proton often lies in a double potential well, with a local
minimum close to each oxygen. As O–O separations in the material decrease, the energy
barrier for H migration from one oxygen to the other decreases until, for short O–O

13
Consequently, the technological drive is to increase the operating temperatures of these polymer fuel cells, whereas
it is to decrease those of the high-temperature solid-oxide fuel cells (based on either proton or oxide conducting solid
electrolytes) from ~800–1000 °C to ~500–700 °C.
546 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

temperature (°C)
500 300 100 0 –50
0
H3PO4
NAFION
-2

-4 CsHSO4 H3OUO2AsO4·3H2O

log(σ) (S/cm)
BaCe0.9Y0.1O3-x

-6

-8
H3O+ β -alumina ice

-10

-12
1 2 3 4 5
1000/T (K−1)

Figure 13.13 Conductivities of selected solid proton conductors. Data for liquid H3PO4 and water ice are
included for comparison. Data are extracted from refs. [6, 7].

separations of ~2.4 Å, the potential has a single minimum with hydrogen equally bonded to
both oxygens. Thus, although the strong proton–oxygen interaction means that protons don’t
diffuse freely, there is a low-energy mechanism for migration provided the proton motion is
coupled to the motion (either local or long range) of the oxygen to which it is bound.

13.6.1 Water-Containing Proton Conductors

Many of the earliest proton conductors studied were materials containing structural water. The
most widely used are the protonated forms of NAFION (for Na Fast ION) polymers (Figure
13.14) that were developed as sodium ion-exchange resins, but also show high H+ conductivity
when protonated and hydrated. They are
[—(CF2)m—CF—CF2—]n
| often referred to as proton-exchange mem-
O
| branes or polymer electrolyte membranes
CF2
| (both PEMs).
CF3—CF—O—(CF2)2—SO3−H+
PEMs contain a stable, hydrophobic,
Figure 13.14 The chemical structure of a typical fluorinated polymer backbone with pen-
NAFION-related polymeric proton conductor. dant acidic –SO3H groups separated by
around 6–9 Å. The high proton migra-
tion is believed to occur along low-dimensional water channels between polymer chains
and to have a similar mechanism to that in water itself. The picture in pure water is
complex, but excess protons are thought to exist predominantly in H5O2+ or H9O4+
clusters that occur with approximately equal probabilities, have lifetimes of the order of
10−13 s, and are hydrogen bonded to surrounding H2O molecules. Proton migration
 13.6 Proton Conductors 547

occurs via the largely uncorrelated shifting of protons within the hydrogen-bonded network
and the dynamic destruction and creation of these clusters. As one H bond weakens prior
to migration, others strengthen, meaning that the overall activation energy for proton
mobility in pure water (~0.1 eV) is significantly less than the energy required to break an
individual H bond (~1 eV). This mechanism is supported by the ratio of the proton-
conduction diffusivity to the diffusivity of water, Dproton =DH2 O , of around 4.5, which shows
that proton movement greatly exceeds molecular migration.
In highly hydrated NAFIONs, proton mobility approaches that in water. Unfortunately, if
the material is heated, it dehydrates and conductivity falls due to individual H bonds in
the water-containing region becoming stronger, as well as due to trapping of protons
by the anionic charge on –SO3− groups. Hydrated NAFIONs can therefore only be
used up to ~90 °C, which is their main technological limitation. In addition, mem-
branes can be relatively permeable to fuel molecules, leading to the equivalent of
a chemical short circuit in the fuel cell. Despite these shortcomings, fuel cells using
hydrated NAFIONs and related polymers are widely used.

13.6.2 Acid Salts

A second category of proton conductors are acid salts such as MHXO4 (e.g. M = NH4, Cs,
Rb, K; X = S, Se) and CsH2PO4. As they don’t rely on water molecules for the conduction
pathway, they can operate at higher temperatures. Perhaps the best understood of these
materials is CsHSO4. Its conductivity jumps from around 10−6 S/cm to 10−3 S/cm when it
undergoes a first-order phase transition at ~413 K (Figure 13.13). This behavior is reminis-
cent of that of AgI, but, as with all proton conductors, the conductivity remains several
orders of magnitude lower than the best cationic conductors.
The jump in conductivity at the phase transition is caused by the onset of disorder in
the tetrahedral HSO4− groups, which tumble around their average site in the high-
temperature phase much as they would in solution; such phases are frequently
described as plastic crystals. As neighboring HSO4− groups tumble, adjacent oxygens
will come into closer contact than usual (Figure 13.15). If a proton-carrying oxygen
comes within ~2.4 Å of an oxygen on an adjacent tetrahedron, there will be little or no
energy cost for a proton transfer between groups, leading to a low activation-energy
pathway for migration. From an order–disorder point of view, the proton changes
from being located on one specific S–O bond to being dynamically disordered over all
four. Thermodynamic measurements support the high disorder of the high-temperature
phase with ΔS values for the 413 K phase transition and eventual melting at 484 K of
13.2 J/(mol K) and 27.2 J/(mol K), respectively; i.e. a significant fraction of the entropy
gain expected for melting occurs at the solid-solid phase transition. Phase-transition
temperatures and resistivities in MHXO4 generally follow the order Cs+ < Rb+ < K+,
such that the Cs-containing materials have the most favorable properties. This is
548 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

proton

rotation transfer

sulfur
oxygen

metal distortion transfer

Figure 13.15 Proton migration from one sulfate tetrahedron to another in CsHSO4 (top) and from one
octahedron to another in the octahedral network of BaHxCe1−xYxO3 (bottom).

consistent with weaker cation–anion bonding and lower activation energies for HXO4−
reorientation with a larger, more polarizable counter ion. Working fuel cells based on
CsHSO4 have been demonstrated [8].

13.6.3 Perovskite Proton Conductors

High-temperature proton conductivity in perovskites is associated with OH− groups, which
can be stable up to very high temperature in some compounds. Hydroxide anions are formed
when water molecules react with oxygen vacancies in a process represented in Kröger–Vink
notation as:

H2 OðgÞ þ vO •• þ OO  ¼ 2OHO •• (13.7)

(see Section 2.6 for a definition of terms). Essentially, the oxygen of a water molecule
enters a vacant site in the structure and its protons form two OH groups. As the
reaction involves deprotonation of H2O, it is most favored for basic oxides. Perovskites
that undergo this process include acceptor-doped phases compensated by oxygen
vacancies, such as BaCe1−xYxO3−x/2 (with BaCe0.8Y0.2O2.9 forming the best proton
conductor) and BaZr1−xYxO3−x/2. Equation (13.7) is exothermic for both systems.
The mechanism for conductivity in these phases is related to that in the acid salts and
involves proton jumps from OH− to adjacent O2− (see Figure 13.15). In a compound such as
BaCeO3, the average O–O separation (~3.2 Å) is too long for strong OH–O hydrogen
bonding, unless there is a significant local distortion of the CeO6 octahedron. Calculations
suggest that the energy penalty for distorting the structure to produce O–O separations in the
2.5–3.0 Å range is almost perfectly offset by the gain in hydrogen-bond strength that follows,
giving a range of structural configurations of similar energy. Locally, each OH group is
thought to form transient H bonds to its eight nearest O sites such that there is a rotational
diffusion of the H bond with an activation energy of around 0.1 eV. Proton transfer is
 13.7 Oxide-Ion Conductors 549

believed to occur when local distortions lead to short instantaneous O–O distances of ~2.4 Å
and has an activation energy of ~0.4–0.6 eV.
Note that even though there is an apparent net migration of OH groups through the
structure, this occurs via proton transfer between adjacent oxide ions and not via the motion
of an intact OH group. The concentration of oxygen vacancies is too low to allow significant
OH migration. This can be proved by 18O tracer diffusion experiments; for example, in
BaCeO3, 18O diffusivities are three orders of magnitude too low to explain the proton
diffusivity. The mechanism at work in these materials also implies that there will be an
optimum temperature for proton migration. At too low a temperature, the H+ mobility will
be low; at too high a temperature, the equilibrium of Equation (13.7) will lie too far to the
left, protons will be lost as water gas, and the number of carriers is low.

13.7 Oxide-Ion Conductors

While PbF2 taught us many of the key features that lead to high anion conductivity, most
technological applications require oxide-anion conductors. Two typical devices requiring
oxide conductivity are shown in Figures 13.16 and 13.17. In the solid-oxide fuel cell (SOFC)
shown in Figure 13.16, H2 is oxidized to protons at the fuel–anode interface releasing
electrons that flow through the external circuit to the cathode. At the cathode, these electrons
reduce O2 to O2− ions that are sucked into the oxide-anion conducting electrolyte and

interconnect

e− e− fuel
H2O excess
out air electrode

air electrolyte
H2 + O2− = H2O + 2e−

1/2O2 + 2e− = O2−

air
electrode
O2−
porous support
tube

inter-
fuel connect
electrode
H2 air electrolyte
in in air
oxide-ion electrode fuel
conductor
electrolyte
air
anode cathode

Figure 13.16 Left: Schematic design of a typical solid-oxide fuel cell. Right: Typical tubular design (3D
image modified from ref. [9]).
550 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

e− e−
V

oxygen p2 oxygen p1
(pressure to be O2− (e.g. air reference
measured, p2 < p1) gas)
oxide-ion
conductor

anode cathode

Figure 13.17 Schematic of a sensor to measure the partial pressure of O2.

migrate along the strong concentration gradient to the anode where they react with the
protons to form water. As for the proton-conductor fuel cell of Section 13.2, the free energy
of the reaction H2(g) + ½O2(g) → H2O(g) can be used to do electrical work, though note that
the direction of ion flow is opposite to that in Figure 13.2. The right-hand side of Figure 13.16
shows the design of a commercial tubular device in which air passes down the center of
a hollow tube and fuel down its outside.
Much of the challenge in commercializing these devices lies in the difficulty of producing
thin-enough components to keep the overall cell resistance low and in providing gas-tight
seals between compartments that will withstand the temperatures (and temperature changes)
when the cell operates. After many years of research, companies have developed high-
efficiency large-scale fuel cells with long operating lives based on this technology. Smaller
systems for domestic power installation are also becoming available.
A second important device where an oxide electrolyte is needed is the oxygen sensor. These
are widely used in internal combustion engines where running at an optimal air-to-fuel ratio
greatly improves efficiency and reduces toxic emissions. A simple design is shown in Figure
13.17. At the right-hand electrode, ½O2 at a high partial pressure p1 picks up two electrons to
form O2−. This then migrates under the concentration gradient through the electrolyte to the
left-hand electrode, where it releases electrons to form ½O2 at a lower O2 partial pressure p2.
One has thus transferred ½O2(g) between environments at two different partial pressures
and therefore two different chemical potentials. The overall voltage of the cell is given by the
Nernst equation as Ecell ¼ RT=nFlnðp1 =p2 Þ. If p1 is a reference gas (typically air) at a known
partial pressure, one can determine p2 from the sensor voltage.
For an oxide-anion conductor to be useful, it must fulfill many requirements, often under
extreme operating conditions. In an SOFC, the material must have high ionic conductivity
(typically >0.01 S/cm at operating temperatures), negligible electronic conductivity, thermo-
dynamic stability, low volatility of components, good mechanical properties, low reactivity
with electrode materials, thermal expansion compatible with other components, as well as be
 13.7 Oxide-Ion Conductors 551

cheap, environmentally benign, and easy to fabricate as dense thin layers of a few tens of
microns. This is a demanding set of requirements! Many structural families have been
investigated and shown to support oxide-ion conductivity (fluorites, perovskites, brownmil-
lerites, Aurivillius phases, pyrochlores, melilites, apatites, and others). Conductivities of
selected members of important families are shown in Table 13.1 and Figure 13.18; and the
origins of their conductivities are discussed in the following sections.

Table 13.1 Ionic conductivities in S/cm at 1000 K of selected SOFC

electrolytes in air. Data largely extracted from ref. [10].

Parent Optimally doped electrolyte Colloquial name log(σ)

ZrO2 Sc0.093Zr0.907O1.9535 ScSZ −1.1

ZrO2 Y0.08Zr0.92O1.96 YSZ −1.4
CeO2 Ce0.8Gd0.2O2−x GDC −1.3
Bi2O3 Bi1.6Er0.4O3 ESB −0.4
Bi2O3a Bi0.913V0.087O1.587 BiVO −1.4
LaAlO3b Sr0.1La0.9AlO3−x −3.1
LaGaO3 Sr0.2La0.8Ga0.76Mg0.19Ca0.05O3−x LSGM −1.1
Ba2In2O5c Ba2In2O5 −3.5
Bi4V2O11 Bi2V0.9Cu0.1O5.35 bimevox −0.8
A10(SiO4)6O2±x La9.75Sr0.25(SiO4)6O2.875 apatite −1.6
La2Mo2O9 La2Mo2O9 lamox −1.4
773 K, bunder N2, cat 10−6 atm where conductivity is predominantly ionic.
a

temperature (°C)
900 700 500 300
0
Bi1.6Er0.4O3
-1
Bi2V0.9Cu0.1O5.35

σ = 10−2 S/cm
-2
log(σ) (S/cm)

-3

Y0.08Zr0.92O1.96
-4
Ba2In2O5 Bi0.913V0.087O1.587
-5
La2Mo2O9

-6
0.7 1.0 1.3 1.6 1.9
1000/T (K−1)

Figure 13.18 Conductivities of selected oxide-anion conductors.

552 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

13.7.1 Fluorite-Type Oxide-Ion Conductors

The most widely used oxide-anion conductors are derived from ZrO2. ZrO2 has a monoclinic
structure at room temperature with seven-coordinate Zr. On heating, it undergoes phase
transitions firstly to a tetragonal (~1370 K) and then (~2643 K) to a cubic fluorite-type
form that remains stable up to the melting point of ~2988 K. The cubic structure can be
stabilized at lower temperature by acceptor doping with Ca2+, Sc3+, or Y3+ to form solid
solutions AxZr1−xO2−x (A2+) or RxZr1−xO2−x/2 (R3+). For the A2+ doping, the dopant-oxide
dissolution equation can be expressed in Kröger–Vink (Section 2.6) notation as:

AO ¼ AZr 00 þ vO •• þ OO  (13.8)

and we see that doping produces an oxide vacancy. With Ca2+, the cubic material is stable for
0.15 < x < 0.28 and with Y for 0.13 < x < 0.68 at 1500 °C. These materials are known as
calcia- and yttria-stabilized zirconias (CSZ and YSZ, respectively) and have high oxide-
anion conductivities. YSZ is the most widely used material commercially, and its conductiv-
ity exceeds 10−2 S/cm above ~650 °C.
How do we optimize O2− conductivity in these compounds? In a simple picture, we might
expect conductivity to increase with vacancy concentration, reaching a maximum when 50%
of the sites are vacant (if chemically possible). In practice, the conductivity peaks at much
lower concentrations due to vacancy trapping; in YSZ this occurs at around 4% vacancies
(x = 0.16). One contribution to the trapping comes from the different sizes [11] of Zr4+ (rVIII
= 0.84 Å) and Y3+ (rVIII = 1.02 Å) cations. If we think of the fluorite structure in terms of
edge-shared M4O tetrahedra, the larger size of Y3+ means that the energy barrier for an O2−
to cross a Y–Y or Y–Zr tetrahedral edge is higher than a Zr–Zr edge. This reduces the
mobility of anions close to a Y dopant. In terms of our expression for diffusivity of
D ¼ D0 expðEA =kTÞ, higher Y3+ content increases D0 via the number of vacancies, but
this is offset by a higher EA due to the larger number of Y edges. These competing effects lead
to the maximum in conductivity at a specific Y content. From the vacancy’s perspective, we
can think of local lattice strain trapping it in positions close to the smaller Zr4+ ions, thereby
reducing its mobility. As this discussion suggests, the highest conductivity in any zirconia-
based system is found in Zr1−xScxO2−x/2 where substitution with Sc3+, which has a similar
radius (rVIII = 0.87 Å) to Zr4+, minimizes the vacancy trapping. Even in this material,
a maximum in conductivity is observed, though at higher x. This occurs as it becomes
energetically favorable for the charged vacancies to order as their concentration is increased.
Local vacancy–vacancy pairs form along the ⟨ 111 ⟩ direction and make longer-range
clusters. The clusters again act as traps for the vacancies, reducing overall O2− mobility.
The useful working range of an oxide conductor, its electrolyte domain, is typically defined
as the temperature and oxygen partial-pressure range, over which more than 99% of its
conduction is ionic rather than electronic. Stabilized zirconias can operate at oxygen partial
pressures of 100–200 bar down to 10−25 bar at ~1000 K. By switching to materials such as
Y-doped ThO2, this range can be extended to even lower oxygen partial pressure.
 13.7 Oxide-Ion Conductors 553

A number of other fluorite-type materials have been investigated in the search for higher
conductivities. Ceria-derived (CeO2; Ce4+ rVIII = 0.97 Å) materials have larger cubic cell para-
meters than zirconias, providing a more open metal–oxide framework that should make oxide-
ion migration easier. Under oxidizing conditions, doped compositions such as GdxCe1−xO2−x/2
with 0.1 < x < 0.2 do have higher conductivities, particularly at low temperature. Unfortunately,
under the reducing conditions at the anode side of the fuel cell,14 Ce4+ can be reduced to Ce3+.
This leads to n-type electronic conduction and a partial electronic short circuit in the cell. It also
causes significant lattice expansion (Ce3+ rVIII = 1.14 Å) that can lead to mechanical failure [12].
Despite these problems, working cells have been made with ceria electrolytes.
While ZrO2 and CeO2 must be acceptor doped to create vacancies, high vacancy content and
ionic conductivity can be achieved in pure Bi2O3. At room temperature, α-Bi2O3 has a monoclinic
structure, but at 1002 K it converts to a face-centered cubic δ form with a defect-fluorite structure,
BiO1.5 □ 0.5, which is stable up to the melting point of 1103 K. The 25% of vacant oxygen sites
leads to a conductivity of 1 S/cm at 1000 K, considerably higher than CSZ or YSZ. Doping is
again required to stabilize the cubic structure at low temperature, and vacancy-trapping effects
mean that the minimum level of dopant again gives the highest conductivity; materials such as
Bi2−xErxO3 (x ≈ 0.4) and Bi2−xYxO3 (x ≈ 0.46 to 0.50) have the highest conductivities. The highest
conductivity reported to date in a stable δ-Bi2O3 derivative is in Bi1−xVxO1.5+x with x = 0.087,
which surpasses the σ = 10−2 S/cm threshold at a mere 350 °C [13]. Its properties and stability have
been related to three key structural features: variable coordination of V, tumbling motions of VOn
polyhedra (which help O2− migration in a manner reminiscent of proton transfer in CsHSO4), and
adoption of a stable superstructure with ordered cation sites (to reduce aging). Low stability under
reducing conditions and volatilization of Bi2O3 at relatively low temperatures have traditionally
been seen as barriers to the practical use of Bi-containing electrolytes. New innovations in
electrode and electrolyte designs and the possibility of operating at lower temperature, where Bi
is less reducible, are helping to overcome these problems.

13.7.2 Perovskite, Aurivillius, Brownmillerite, and Other Oxide Conductors

A number of interesting oxide-ion conductors have been found in materials with perovskite-
related structures. The earliest reports of ionic conductivity in pure perovskites centered on
acceptor-doped RAlO3 (R = rare earth), with materials such as Sr0.1La0.9AlO3−x having the
highest conductivities. Later work focused on LaGaO3-based materials that have higher con-
ductivities than stabilized zirconias and operate at lower oxygen partial pressures. The best
properties are achieved by acceptor doping at both La and Ga sites, and there is a trade-off
between vacancy concentration and vacancy trapping. Optimal materials use size-matched
substitutions, giving compositions La1−xSrxGa1−yMgyO3−δ, with 0.1 < x < 0.2 and 0.15 < y <
0.20. These materials don’t absorb water, don’t show significant aging at 800 °C, and have >98%
oxide-anion conduction under the typical operating conditions of a fuel cell.

14
Reducing conditions due to the excess of fuel at the anode side. Typical partial pressures might be 10–19 bar O2.
554 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

At higher oxygen deficiency levels, some perovskite-related materials adopt structures in

which the vacancies are ordered. Compositions ABO3−x with x = 0.5 (i.e. 16.7% oxygen
vacancies) can adopt the structure of the mineral brownmillerite, Ca2FeAlO5. One example
of this is Ba2In2O5 (Figure 13.19) that can be thought of as containing rows of oxygen
vacancies parallel to [110] of the simple perovskite cell, running in every other equatorial
plane of the corner-linked InO6 octahedra. In doing this, one removes two of the six
octahedral corners in every second layer, leading to alternate layers of tetrahedra and
octahedra. Conductivity data for Ba2In2O5 are included in Figure 13.18. At 1000 K and
p(O2) = 10−6 atm, the conductivity is 10−3 S/cm, but jumps to 0.1 S/cm at 1200 K, when the
material undergoes a first-order phase transition where vacancies disorder, ultimately reach-
ing a disordered cubic perovskite structure at around 1300 K. Improvement in conductivity
at lower temperature can be achieved in compositions such as Ba2In1.75Ce0.25O5.125, where
the disorder is stabilized. The high number of vacancies means, as suggested by Equation
(13.7), that Ba2In2O5 is susceptible to water uptake and it is in fact a good proton conductor
from 300 °C to 700 °C. Problems of water uptake at low temperature can be reduced by
substitution to give materials such as Ba1.2La0.8In2O5.4.
Other structure types investigated include apatite-related A10−x(MO4)6O2±y (A = Ln3+/alka-
line earth, M = Si/Ge), Aurivillius-type “bimevox” materials, and La2Mo2O9-derived phases.
The Aurivillius structure of Bi2WO6 is shown in Figure 13.19. It contains layers of corner-
sharing [WO4/2O2]2− octahedra alternating with [Bi2O2]2+ layers. The [Bi2O2]2+ portion has
a central layer of oxide ions tetrahedrally coordinated by Bi (similar to [La2O2]2+ in the
LaOFeAs superconductors of Chapter 12). The Bi3+ ions are coordinated by four of these
oxygens (at 2.18–2.51 Å) and also by two oxygens of the WO6 layers (2.44–2.58 Å) in an
asymmetric coordination environment typical of a lone-pair cation. Replacement of WVI by VV
leads to Bi2VO5.5 or Bi4V2O11, which has a similar structure, but with oxygen vacancies
predominantly in the octahedral layers. Bi4V2O11 has a conductivity of 0.1–1.0 S/cm in its

InO6/2

InO4/2

Bi
O
Bi

BaIn2O5 Bi2WO6

Figure 13.19 Brownmillerite-type Ba2In2O5 and Aurivillius-type Bi2WO6.

 13.8 Intercalation Chemistry and Its Applications 555

high-temperature (> 843 K) γ form. The substituted “bicuvox” phase15 Bi2V1−xCuxO5.5−3x/2

with x ≈ 0.1 retains a conductivity of 10−2 S/cm down to 623 K. La2Mo2O9 has also been shown
to exhibit high conductivity in its high-temperature structural form, and this has been related to
the ability of Mo to adopt variable coordination numbers. This particular material highlights
some of the complexities of functional oxide materials; its room-temperature structure contains
a remarkable 312 crystallographically unique atoms!

13.7.3 SOFC Electrode Materials and Mixed Conductors

There are also significant challenges when considering materials to use as electrodes in SOFCs.
The anode must catalyze the oxidation of the fuel, conduct O2− ions from the electrolyte, and
simultaneously have sufficiently high electronic conductivity (~100 S/cm or higher) to channel the
electrons produced into the external circuit. It must also have thermal expansion compatible with
the electrolyte. Traditionally, porous ceramic–metal composites or cermets have been used, such
as Ni–YSZ composites. Having a fully connected (percolated) metal component ensures elec-
trical conductivity, and fuel oxidation can occur at the three-phase boundary between metal,
electrolyte, and fuel. This is, however, only a 1D intersection.16 There is therefore significant
benefit in using mixed ionic and electronic conductors (MIECs) where fuel oxidation can occur
over the whole of the porous anode’s high surface.17 A variety of perovskite materials have been
investigated for this purpose such as (La0.75Sr0.25)0.5Cr0.5Mn0.5O3−x, Mn/Ga-doped Sr1−xLax
TiO3+δ,18 and PrBaMn2O5+δ [14, 15, 16].
At the cathode, the key process is reduction of molecular O2 and its transport to the
electrolyte. Materials such as La1−xSrxMnO3−x (see Chapter 11) are frequently used.
A variety of perovskite-related materials such as Ba0.5Sr0.5Co0.8Fe0.2O3 and LnBaCo2O5+δ
[17] are also being investigated for this application.

13.8 Intercalation Chemistry and Its Applications

The last category of materials we will discuss where the migration of ionic species leads to
energy-related applications are intercalation compounds. In everyday usage, the term inter-
calation refers to the insertion of extra days into the calendar, such as the introduction of
February 29 in leap years to allow for the fact that the Earth takes 365.26 days to orbit the
Sun. In solid state chemistry, intercalation is defined as a reversible insertion of guest species
(atoms, ions, molecules, or molecular ions) into vacant sites of a host structure. It is a process
that has been practiced inadvertently for millennia in the processing of clays by pottery
industries around the world and is shown schematically in Figure 13.20.

15
One of the wider family of “bimevox” materials.
16
The use of Ni in cermets can also lead to significant coke formation if hydrocarbons are used as fuels.
17
Mixed conductors have other areas of application, including as semi-permeable oxygen membranes.
18
Typical composition La0.33Sr0.67Ti0.92Mn0.04Ga0.04O3.125.
556 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

intercalate
host
+ guest

− guest

Figure 13.20 Schematic intercalation reaction of a layered 2D host.

13.8.1 Graphite Intercalation Chemistry

The first scientific report of intercalation chemistry is probably that of Schäuffatl, who
described the reaction of graphite with concentrated sulfuric acid in the 1840s. Since then,
a large number of graphite intercalation compounds have been prepared with guest species
between the graphite layers that move apart along the c axis to accommodate them. One of the
most controllable synthetic methods for combining the graphite and guest is by vapor-phase
transport. This is done by heating the guest and solid graphite to slightly different temperatures
in different compartments of an evacuated glass tube. Potassium intercalates K0.125C (≡ KC8)
with K between all the graphite sheets (Figure 13.21) can be prepared at ~250 °C.19 A bromine
intercalate (Br2)0.0625C can be obtained by holding the guest and graphite at ~20 °C.
Intercalation can also be achieved using a solution of K in liquid ammonia20 or Br2 in CCl4.
Intercalates of a wide variety of metals and molecular guests can be prepared in similar ways.
Graphite is unusual as an intercalation host in that it will intercalate both electron donors like
K and acceptors like Br2. The key to this behavior lies in its semimetallic band structure discussed
in Chapter 6. On intercalation, K is ionized to K+ and electrons are donated to the empty
conduction band. With an acceptor, on the other hand, electrons from the valence band partially
reduce the guest, leaving holes at the top of the valence band. Both these processes lead to partially
filled bands, and intercalation causes significant changes in conductivity, as shown in Table 13.2.
As discussed in Chapter 6, graphite shows significant anisotropy in its conductivity. The aniso-
tropy is decreased in the alkali-metal intercalates but increased in molecular intercalates due to
poor overlap between the graphite p orbitals and molecular orbitals of the guest. It is interesting to
note that the carrier mobility of graphite layers is sufficiently high that AsF5-acceptor
intercalates21 have a conductivity comparable to Cu despite having only around a quarter of
the density.22
In addition to compositions such as K0.125C, which contains guest molecules between
all host layers as shown on the left of Figure 13.21, it is also possible to prepare
compounds containing fewer guests in which only every nth layer is occupied. This
19
The graphite is held at a slightly higher temperature than the guest to prevent condensation of the guest on the
graphite. The K content is limited to 0.125 due to the relative sizes of K and graphite layers (see Problem 13.13).
20
Here ammonia co-intercalation to form Kx(NH3)yC occurs.
21
The intercalate formed from graphite and AsF5 is believed to contain a mixture of molecular ions and neutral
molecules between the layers formed via reactions such as 3AsF5 + 2e− → 2AsF6− + AsF3.
22
Graphite/AsF5 has mobility of 1.3 m2 per volt per second [m2/(V s)] at room temperature. Cu has σ = 5.8×105 S/cm
and mobility 0.0035 m2/(V s).
 13.8 Intercalation Chemistry and Its Applications 557

E
conduction
band
D + Cn → D+Cn−

D → D+ + e−

EF

A + e− → A−

A + Cn → A−Cn+ valence
band

N(E)

Figure 13.21 Left: A K0.125C intercalation compound of graphite viewed parallel and perpendicular to
the layers. Right: DOS plot for pristine graphite with indication how a donor (D) partly fills the
conduction band or an acceptor (A) partly empties the valence band.

Table 13.2 Room-temperature conductivity of graphite and various

intercalation compounds in plane (σa) and out of plane (σc) [18].

Guest Stage σa (S/cm) σc (S/cm) σa/σc

Graphite (none) – 2.5×104 8.3 3000

K 1 1.1×105 1.9×103 56
Li 1 2.4×105 1.8×104 14
Br2 2 2.2×105 1.6 1.4×105
AsF5 2 6.3×105 0.24 2.7×106

phenomenon is called staging. For the potassium–graphite intercalate, stage-2 com-

pounds with every other layer full can again be prepared by the vapor-transport
method, if the graphite is heated to ~375 °C rather than 250 °C to lower the driving
force for intercalation. The simplest model for staging would be that shown on the left
of Figure 13.22, with host layer gaps being either completely full or completely empty.
However, the experimental observation that it is possible for different stages to evolve
smoothly from n to n−1 as more guest is added is inconsistent with this model. For
example, transforming from a stage-1 to a stage-2 compound would require the
improbable removal of guest species entirely from one layer and their complete reintro-
duction in another. A more plausible model was suggested by Daumas and Herold [19],
and is shown on the right of Figure 13.22. Here, staged compounds have pleated host
layers with islands of guest between them. With this model, smooth transformation
between stages is possible with only lateral movement of guests. A possible driving
force for staging that is consistent with experimental observations is that it is due to
558 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

stage 1

graphite

stage 2

stage-3 stage-2 stage-1

K0.03C K0.04C K0.08C
or or or stage 3
KC36 KC24 KC12

classical Daumas-Herold

Figure 13.22 Staging in graphite intercalation compounds.

strain in host layers. As guests are intercalated, the addition or removal of electrons in
the graphite layers changes the equilibrium C–C bond distance, creating local strain
(C–C bond lengths increase on K intercalation as its electron is accommodated in
graphene antibonding orbitals). The strain throughout the crystal can be minimized by
packing guest species as closely as possible within one layer and maximizing the
distance to the next intercalated layer. Stages can order over large distances, giving
rise to sharp (00l) peaks in diffraction patterns, and n = 8 and higher stages are well
documented in alkali-metal intercalates. Different staged compounds have different
conductivities and different colors. In the case of K, stage-1 compounds are gold,
stage-2 are blue, stage-3 are blue-black, and stage-4 and higher are black.

Box 13.1 Synthetic Methods: Chimie Douce

The reversibility of intercalation chemistry offers unusual opportunities to synthesize materials
at low temperature that would not otherwise be stable. These are examples of “Chimie Douce”,
French for “Soft Chemistry”. One of the earliest applications of this was the preparation of
a metastable layered form of VS2 that couldn’t be prepared by other methods. When Li, V, and
S are heated together in a sealed silica-glass tube, crystals of LiVS2 form, with Li between VS2
layers. If this compound is stirred with a solution of I2 in acetonitrile, lithium is removed,
leaving a metastable layered form of VS2. The reactions are:

Li + V + 2S → LiVS2
LiVS2 + ½I2 → LiI + VS2
 13.8 Intercalation Chemistry and Its Applications 559

Box 13.1 (cont.)

The low reaction temperature traps the metastable layered form of VS2 [20]. Similar tricks have
led to a metastable λ-MnO2 by Li extraction from LiMn2O4 [21].
Chimie Douce has also been used to synthesise layered LiMnO2 [22, 23], analogous to
LiCoO2. Due to size considerations, the layered material can’t be prepared directly at high
temperatures. However, its Na analogue can be made at 923 K and LiMnO2 prepared by
stirring NaMnO2 with LiBr in hexanol at 433 K.

13.8.2 Lithium Intercalation Chemistry and Battery Electrodes

One of the main reasons for interest in intercalation chemistry is its application in the electrodes
of rechargeable batteries. Lithium is the most electropositive metal (E° = −3.04 V) and has the
lowest density (ρ = 0.53 g/cm3), making it particularly suitable for application in high-energy,
low-weight batteries. We will therefore look at intercalation reactions of Li with layered metal
chalcogenides and metal oxides, and how they can be exploited in lithium-ion batteries.
A large number of MX2 metal dichalcogenides adopt layered CdI2 structures (Figure 1.28),
where octahedral holes in every second layer of an hcp array of anions are occupied by cations
(e.g. TiS2, ZrS2, SnS2). Owing to the relatively weak van der Waals forces holding layers
together, these materials are ideal intercalation hosts and take up a variety of guest species.
A simple room-temperature reaction with n-butyllithium, for example, leads to the rapid
formation of LixMS2, with octane as the side product. The Li “atoms” formed from butyl-
lithium reduce the MS2 host, and the resulting Li+ ions enter the empty octahedral sites
between the host lattice layers; the electrons enter the conduction band of the host.

e− e−
TiS2
TiS2

Li TiS2/teflon Li+ Li-metal

cathode (+) anode (−)
electrolyte
TiS2x− e.g. LiPF6
Li+ Li+ Li+ Li+ Li+ xLi+ in solvent

TiS2x−

anode: xLi xLi+ + xe−

cathode: xLi+ + xe − + TiS2 LixTiS2
overall: xLi + TiS2 LixTiS2

Figure 13.23 Left: Schematic intercalation reaction. Right: Schematic design for a rechargeable Li cell.
Typical solvents used are mixtures of dimethyl carbonate, diethyl carbonate, and ethylene carbonate.
560 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

Table 13.3 Requirements for intercalation-based electrode materials for rechargeable Li batteries.

Battery requirement Chemistry to meet requirement

High cell voltage, Ecell Requires a large ΔG for the cell reaction and will be governed by the change in
chemical potential of Li in the charged and discharged states.
Approximately constant Ecell Requires ΔG to change little as a function of x during the xLi + cathode → Lix
during discharge [cathode] intercalation reaction.25
High capacity The more Li ions per formula unit that can be reversibly intercalated/
deintercalated, the higher the charge that can be stored. The capacity is usually
expressed in A h/kg (≡ mA h/g).
Storage energy density A battery needs high energy density either per unit mass or volume, depending on
the application. The more exchangeable Li per electrode, the better. The overall
energy density (ed) is the product of the electrode potential E and the charge per
unit mass or volume Q: ed = EQ in W h/kg or W h/L.
High current/power Rapid intercalation is required to give high current flows and high power. This
requires high mobility of both Li+ and electrons through electrodes and is
favored by minimal structural rearrangement during intercalation. It is helped
by cell designs that minimize diffusion length.
Reversibility/capacity fade Intercalation must be reversible over a wide composition range and allow many
cycles of charge/discharge without damage to either the atomic, micro, or bulk
structure of the electrodes or on-going reaction between electrodes and electrolyte.
Manufacturing and The device must be easy to manufacture, cheap, safe to use, have minimal
commercial environmental impact, use abundant elements, have a long shelf life, be
considerations recyclable at the end of life, and outperform established technology sufficiently
to warrant commercialization.

This reaction can also be performed in an electrochemical cell of the type shown in
Figure 13.23. At the negative Li-metal anode, Li dissolves as Li+ ions, and the electrons released
pass round the external circuit. At the TiS2 positive cathode, Li+ ions from solution are
intercalated between the host layers, with charge balance provided by electrons from the external
circuit entering the TiS2 conduction band. The overall chemical reaction Li + TiS2 → LiTiS2 has
a Gibbs free energy of −206 kJ/mol, and can be used to do electrical work; ΔG = −nFEcell implies
a cell voltage of 2.13 V. Since intercalation is reversible, applying an external voltage can reverse
the chemistry and recharge the cell. Such cells have been operated successfully in the lab over
many cycles, giving cell voltages varying smoothly from ~2.25 V to ~1.75 V (see Figure 13.28) as
the cell is discharged and recharged to the 90% level.23
Table 13.3 summarizes the key desirable features of a rechargeable battery. While the
Li–TiS2 system meets many of these, it was never commercialized.24 The main reason was
23
Note that, in contrast to the lead–acid and Ni–Cd rechargeable batteries discussed earlier, there is no reaction with
the electrolyte as the cell operates.
24
Though TiS2 cathode–LiAl anode cells were sold as watch batteries in the late 1970s.
25
Note that a small change in voltage during discharge can be advantageous as it then gives an indication of the
battery charge. This is how the battery lifetime indicators on portable devices function.
 13.8 Intercalation Chemistry and Its Applications 561

Figure 13.24 Electron energy levels in a typical Li cell in its charged state. As drawn, the formation of an
SEI (see text) passivating layer gives kinetic stability such that electrolyte reduction doesn’t occur despite
e
μ anode
e being higher than its LUMO. All levels are in joules per mole (J/mol) as F is the Faraday constant.

safety. As the cell is discharged, Li+ ions are removed from the anode, then redeposited under
charging, but not necessarily at the same point. Over many cycles, dendritic growth of Li metal
at the anode surface is observed. The growing “strands” of Li have such a high surface area that
reactions with the electrolyte become a problem. If they become too large, they can short-circuit
the cell, leading to sudden temperature rises and the cell igniting.

13.8.3 Lithium-Ion Batteries with Oxide Cathodes

How can we design a better intercalation-based battery; one that meets more of the require-
ments in Table 13.3? We’ll need to consider the chemistry of each component of the cell, the
reactions that can occur at the interfaces between components, and the passage of both ions
and electrons during its operation.
Let’s consider first what determines Ecell. Since we no longer have a simple aqueous galvanic
cell (like in Figure 13.1) operating under standard conditions, we can’t just determine Ecell from
the difference of standard reduction potentials of, say, Ti4+/Ti3+ and Li+/Li. Instead, we have to
consider the energetics of adding or removing electrons at the Fermi levels, EF, of the two
crystalline electrodes. The important electron-energy levels are shown in Figure 13.24. When
these energy levels are in units of eV26 (units of voltage times the elementary charge) the EF
difference divided by 1 (electron) yields directly the cell voltage Ecell.
26
EF values are typically given in electronvolts (1 eV = 1.602176634×10−19 J) and represent the energy released for
each electron added to the solid. Multiplying by NA converts 1 eV per added electron to 96485 J per added mole of
electrons (the charge of 1 mole of electrons is 96485 C or the Faraday constant; and 1 J = 1 C × 1 V).
562 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

Since the chemistry of an operating cell changes as it is charged or discharged, it is useful to

express Ecell via the electrochemical potentials e
μ A ¼ μA þ zA Fϕ per mole of electrons27 at the
two electrodes. In a charged cell, electrons have a high electrochemical potential e
μ anode
e at the
reduced anode, and a low electrochemical potential e μ cathode
e at the oxidized cathode. If we
connect leads to each electrode, the voltage difference between them, Ecell, will be propor-
tional to the difference in electrochemical potentials or in Fermi levels of the two electrodes:
!  
μ anode
e eμ cathode Eanode  Ecathode
Ecell ¼ e e
¼ F F
½V; J=mol1 =ðC=mol1 Þ; J=mol1 =ðC=mol1 Þ
F F
(13.9)

Our first consideration in cell design is the stability of the electrolyte to oxidation or
reduction. If e
μ anode
e in the charged state lies above the electrolyte’s lowest unoccupied
molecular orbital (LUMO), the electrolyte will be reduced28; if the corresponding e μ cathode
e is
below the highest occupied molecular orbital (HOMO), the electrolyte will be oxidized. As
both processes could decompose the electrolyte, there is an electrolyte stability window that
limits the cell potential to Ecell ≤ Eg/F, where Eg is the electrolyte HOMO–LUMO gap in J/
mol.29 With aqueous electrolytes, this limits practical voltages to ≲ 1.5 V.30 Much larger
Ecell values are possible with organic electrolytes, such as for LiPF6 dissolved in a mixture of
organic carbonates.31 With a Li-metal anode, e μ anode
e lies above the LUMO of even these
electrolytes. Fortunately, the electrolyte stability window can be expanded via formation of
a thin amorphous passivating layer on the electrode surface, which is produced by reaction
with the electrolyte during the initial charge. This is called the solid electrolyte interphase
(SEI), and ethylene carbonate is particularly effective in this role. SEI formation reduces the
battery capacity but, since it provides a kinetic barrier to further electrolyte reduction in use,
it improves the lifetime.
Let’s now consider how to choose optimum electrode materials for an intercalation-based Li
cell. When we discharge the cell, we move Li from the anode (where it has a high chemical
potential) to the cathode (where it is low) in the reaction Li(anode) → Li(cathode). Li+ ions move
through the electrolyte and electrons flow through the external circuit where they do work on the
surroundings. If we break the circuit so that no current can flow, we create an “electrochemical
27
The electrochemical potential of the electrons has two contributions. The first is the plain chemical potential μA of the
electron, taken as a component A in a reacting mixture of A, B, C, . . . and describes how the Gibbs energy of this
“system” changes per unit change in the number of moles of the reaction component. At a given temperature and
−1
pressure and fixed amount n of other species μA = (∂G/∂nA)T,p,nB,nC . . ., and is typically expressed in J mol .
The second contribution is zAFϕ, where zA is the charge on A and ϕ the electrostatic potential in V, and
reflects the energy required to bring a charged species to a specific location. Note that e μ A ¼ μA for
a neutral species such as Li.
28
As electrons move to lower energy if they transfer from the anode to the empty electrolyte LUMO.
29
Or Ecell ≤ Eg for the HOMO–LUMO gap given in eV.
30
Sum of standard potentials for H+ + e− → ½H2 and ½O2 + 2H+ + 2e− → H2O plus a small kinetic overpotential.
31
Dimethyl carbonate, diethyl carbonate, and ethylene carbonate are commonly used.
 13.8 Intercalation Chemistry and Its Applications 563

Li metal LixC6 LixTiS2 Li1−xCoO2 LixCoPO4

E E E E E
CB Co 4s Co 4s
Ti 4s
EF

0.2 eV

2.3 eV

4.0 eV
Ti4+/Ti3+

4.7 eV
Co4+/Co3+
S 3p Co3+/Co2+
VB
O 2p
VB
O 2p
VB

DOS, N(E) DOS, N(E) DOS, N(E) DOS, N(E) DOS, N(E)

anodes cathodes

Figure 13.25 DOS plots showing relative positions of EF (in eV = 1.602×10−19 J) in Li metal and LixC6
anodes and approximate energies and characters of charged cathode-material orbitals that will be filled
on Li intercalation. VB = valence band, CB = conduction band. After ref. [24].

equilibrium” state,32 in which local equilibrium for the process Li ⇄ Li+ + e− at each electrode
μ Liþ þ e
requires e μ e ¼ μLi .33 Since Li+ can diffuse freely through the electrolyte, under these
conditions e μ Liþ will be identical at both electrodes and it follows from Equation (13.9) that:
!  
e
μ anode
  e
μ cathode
 μanode  μcathode
Ecell ¼ e e
¼ Li Li
(13.10)
F F

We can see from this equation that we need to manipulate the chemical potential of Li in the
anode and cathode environments to influence Ecell.
In commercial batteries, the most widely used trick at the anode is to use an intercalation
compound such as LixC6 instead of Li metal. Here, μanodeLi is lower than in Li metal (Li is more
stable in LixC6, Figure 13.25). This has the disadvantage of reducing Ecell, as can be seen
from Equation (13.10), but two major advantages: Firstly, reduction of the electrolyte when
the cell is charged becomes less likely. Secondly, the formation of Li metal dendrimers during
cell charging is eliminated,34 greatly improving battery safety.
How do we compensate for the loss of Ecell caused by using a graphite anode? The only way is
to look for a cathode in which μcathode
Li is as low as possible. Since Li is invariably present in the

32
Note that this is not a global chemical equilibrium state.
33
At equilibrium, products and reactants must have equal electrochemical potentials.
34
At least under slow charging, which approximates equilibrium conditions, when all Li will intercalate.
564 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

(oxidized)
phosphates Li1−xMn2O4

cathode (+) materials

4.0 LixMPOnFm Li1-xCoO2

electrolyte -stability window

M = Fe, Co, V
potential vs. Li metal (V)

LixVyOz

LixC6/Li1-xCoO2 E ≈ 3.6 V
3.0 Li1+xMn2O4

LixMS2
2.0 M = Ti, V

(reduced)
Li4Ti5O12

anode (−) materials

Li/TiS2 E ≈ 2 V

1.0

Lix(Si,Sb,Sn)
Li3−xCoxN2
LixC
0.0 Li metal

Figure 13.26 Voltage ranges for various cathode/anode combinations. The cell voltage is the difference
between cathode and anode potentials. The electrolyte-stability window to reduction/oxidation for LiPF6
in a 1:1 mixture of ethylene carbonate and diethyl carbonate is indicated. Batteries must be operated
within this window.

cathode as Li+ ions and electrons, it is convenient to consider the terms e μ Li

cathode
þ μ cathode
and e e
separately. A good cathode will therefore need low-energy sites to adopt Li+ ions35 and to have
a low EF (e μ cathode
e ), meaning low-energy orbitals for the electrons. It is therefore important to
understand the density of states (DOS) of the cathode, using the concepts introduced in
Chapter 6. DOS plots for some of the cathodes discussed in this chapter are included on the
right of Figure 13.25. The first thing we can appreciate from Figure 13.25 is that the energy of the
top of the cathode valence band will set an upper limit on the value of Ecell achievable with a given
anode. The maximum Ecell achievable using the Co4+/Co3+ couple in an oxide cathode such as
Li1−xCoO2 will therefore be much larger than using Ti4+/Ti3+ in a chalcogenide such as LixTiS2.
Even in an oxide, changing the counter ion for a given transition metal has a significant impact on
Ecell. For example, Ecell is greater using the Co3+/Co2+ couple in LixCoPO4 than the Co4+/Co3+
couple in Li1−xCoO2.36 Figure 13.26 summarizes these ideas and shows the range of cell voltages
achievable with different anode and cathode materials.

35
We explore the influence of different Li+ sites in the end-of-chapter problems where we see Ecell changing for a given
cathode when Li+ adopts different coordination sites.
36
Another example is the observation that the position of the Fe3+/Fe2+ couple can be moved by around 1 eV within
a closely related series of LixFe2(XO4)3 materials with X = As, P, Mo, or S (see ref. [20]).
 13.8 Intercalation Chemistry and Its Applications 565

discharge: Li 1−xCoO2 + LixC6 → LiCoO2 + C6

(+) cathode LixCoO2 (−) anode LixC6

e− e−
O
Li+ discharge
Co
O
Li
O Li+ charge

charge: LiCoO2 + C6 → Li1−xCoO2 + LixC6

Figure 13.27 The simplified38 structure of LiCoO2 and a schematic representation of its function in a Li-
ion cell. The direction of electron flow shown is that during discharge.

The first commercially successful rechargeable battery was introduced by Sony in the early
1990s and used all of these ideas. It employed a LixC6 intercalation anode to prevent
dendrimer formation and an oxide Li1−xCoO2 cathode based on the Co4+/Co3+ couple.
The structure of LiCoO2, Figure 13.27, is related to that of NaCl, and can be described as
a slightly distorted ccp of oxide ions with the octahedral holes in alternate layers occupied by
Li+ and Co3+.37
The cell is manufactured in a discharged state with a LiCoO2 cathode, carbon anode, and
electrolyte of LiPF6 dissolved in a mixture of alkyl carbonates,39 with a porous polymer film
separating the electrodes. There are various physical designs that try to maximize electrode
surface areas to maximize Li+ diffusion rates.40 On the first charge, Li ions are partially
deintercalated from LiCoO2, forming Li0.5CoO2, and intercalated by the anode, forming
LiC6. When the cell discharges, the reverse occurs: Li is deintercalated from graphite and Li+
ions from the electrolyte intercalate between cathode layers, with the charge-balancing
electrons flowing through the external circuit. The overall cell reaction is 0.5LiC6 +
Li0.5CoO2 → LiCoO2 + 0.5C6, the free energy of which arises from moving Li ions and
electrons from the carbon to the oxide material where their chemical potential is lower. An
important operating feature of the Li1−xCoO2 cell is that only half the available Li ions are
exploited in the cell reaction. Beyond this point, one approaches the stability limit of
Li1−xCoO2 with respect to O2 loss, and oxidation of the electrolyte becomes more likely.41

37
Remember from Chapter 1 that the close-packed layers are perpendicular to the ⟨ 111 ⟩ (body-diagonal) directions
of the cubic unit cell. Note the similarity to LiTiS2 that has alternate Li+/Ti3+ layers but in an hcp anion array.
38
Li/Co ordering lowers the symmetry from cubic to rhombohedral in LiCoO2.
39
For example, dimethyl carbonate, diethyl carbonate, and ethylene carbonate (see Footnote 31). Ethylene carbonate
is particularly important for anode passivation.
40
The ionic current density in the electrolyte and through the SEI layer is lower than the electronic current density of
the external circuit, requiring a thin electrolyte and high-surface-area electrodes.
41
As shown in Figure 13.25, the Co4+/Co3+ couple is sufficiently close to the O 2p valence band that electrons may be
lost from valence-band orbitals with high oxygen character if more Li is removed on charging; i.e. O2− oxidation
occurs. Near the surface, peroxide formation can lead to the reaction O22− = O2− + ½O2(g) and oxygen loss.
566 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

Batteries can be made with a voltage exceeding 3.6 V, an energy density of ~180 W h/kg (two
to three times higher than achievable for Ni–Cd batteries, see the end-of-chapter problems),
which are rechargeable tens of thousands of times. The design is usually called a LION (lithium
ion) or rocking-chair42 cell to distinguish it from those containing Li metal. The rechargeable Li
battery has revolutionized modern society was described as “the most important advance in
energy storage for 100 years” [25]. It led to the award of the 2019 Nobel Prize in Chemistry to
Goodenough, Whittingham, and Yoshino.43 In addition to LiCoO2, a wide variety of substitu-
tion-related phases have been studied as potential cathodes, and materials such as Li(Ni~0.33
Mn~0.33Co~0.33)O2 (NMC) show good performance, as discussed in Box 13.2.

Box 13.2 Nanoscale Concepts: Cathode structure control from the micro- to
nanoscale
A rechargeable battery places extreme demands on its constituent materials and on the
interfaces between them. Electrode materials must undergo major yet rapid and reversible
changes in their local structure and bonding, and must shuttle between highly reactive oxida-
tion states. Going to the nanoscale offers the potential benefit of rapid reactions due to short
diffusion lengths, but the high surface area of components can lead to significant side reactions.
Properties therefore need to be optimized across the length scales from ångstroms (controlling
atomic-level structure) to nanometers (morphology and composition of primary particles or
domains) to microns (particles made up of several primary particles).
One challenge in improving the properties of LixMO2 cathodes is to find cheaper
materials that simultaneously improve capacity without compromising safety or lifetime.
Compositions like Li(Ni0.8Mn0.1Co0.1)O2 have higher capacity (up to 200 mA h/g versus
140 mA h/g for LiCoO2) but have severe safety implications. On charging to low Li
levels, they can undergo an exothermic transformation to spinel structures at around
200 °C (3MO2 → M3O4 + O2). The heat and oxygen released can lead to the cell catching
fire. In contrast, the higher Mn content in Li(Ni0.5Mn0.5)O2 stabilizes the layered struc-
ture such that decomposition doesn’t occur until 300 °C, but leads to lower capacity since
Mn4+ doesn’t take part in the electrochemical reaction.
How might we get the best of both worlds? One possibility is to prepare core-shell
cathode particles (see Figure B13.2.1) that contain a high-capacity Ni-rich core enclosed
by a high-stability Mn-rich shell. Such particles can be synthesized by reacting aqueous
transition-metal solutions with hydroxide to form particles of (NiaMnbCoc)(OH)2(s) of
controlled and uniform size. By varying the concentration of metal ions in the feedstock
over time it’s possible, for example, to precipitate Ni-rich (Ni0.8Mn0.1Co0.1)(OH)2 cores
surrounded by (Ni0.5Mn0.5)(OH)2. On calcining these hydroxides with lithium carbonate,
hydroxide, or nitrate, the spherical particle morphology and transition-metal distribution

42
The widely used “rocking-chair” term seems to have been introduced to reflect the fact that the Li ions move
reversibly to and fro between the cathode and anode.
43
Stanley Whittingham for the Li/TiS2 battery, John Goodenough for metal oxide cathodes, and Akiro Yoshino for
introducing carbon-based anodes.
 13.8 Intercalation Chemistry and Its Applications 567

Box 13.2 (cont.)

is retained, leading to core-shell particles. The particles show good stability when used as
cathodes, with just a small capacity reduction due to the shell layer.

core shell concentration gradient

(a) (b) (c)

Li(Ni0.46Mn0.31Co0.23)O2 Li(Ni0.68Mn0.22Co0.10)O2
Li(Ni0.64Mn0.18Co0.18)O2 Li(Ni0.90Co0.10)O2
~12 μm

Figure B13.2.1 (a) Scanning electron microscope (SEM) image of a core-shell LiMO2 cathode particle with
compositions indicated in (b). (c) Concentration-gradient particle. SEM image from Sun et al., Nat. Mater.,
8 (2009), 320–324 with permission.

One problem with this design is that the core and shell show very different volume
changes on delithiation (around 9–10% and 2–3%, respectively). The resulting stresses
could ultimately lead to the protective shell cracking away from the core. One solution to
this is to produce particles in which the shell has a composition that gradually becomes
richer in the stabilizing manganese as one moves from the core to the surface. Figure
B13.2.1a shows such a particle with overall composition Li(Ni0.64Mn0.18Co0.18)O2 but with
a Mn-rich and therefore stabilized surface of Li(Ni0.46Mn0.31Co0.23)O2. These particles
showed similar initial capacity to uniform particles, but with 95% capacity retention over
500 cycles in a cell compared to 70%.
An alternative approach is to produce full concentration-gradient particles in which the
composition varies continuously from the core to the outer surface (Figure B13.2.1c). Particles
of overall composition Li[Ni0.75Mn0.15Co0.10)O2 could be prepared, in which Mn content
varied from 0% at the core to 22% at the surface. By adjusting the OH− concentration it
was also possible to produce the primary particles as extended nanorods. Importantly,
microspheres could be prepared with these nanorods aligned radially. This morphology
maintains good diffusion pathways for Li+ into the particle (10 times higher than a core-
shell particle). These particles showed exceptional stability in working cells (e.g. 95% capacity
retention over 1000 cycles at 25 °C). They also showed good thermal stability in their highly
delithiated state, with an onset temperature for exothermic decomposition to spinel phases of
280 °C [26].
568 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

13.8.4 Electrochemical Characteristics of Lithium Batteries

Electrochemical measurements give considerable insight into the performance of battery elec-
trodes and the details of the chemical changes that occur. Figure 13.28 shows voltage as
a function of Li content for cells using two cathode materials, LixTiS2 and LixFePO4, against
metallic lithium anodes. These data were recorded using a slow discharge rate that approximates
equilibrium conditions. The discharge rate of a cell is often expressed as a C rate (in h−1). A 1 C
rate uses the current needed to discharge the given cell in 1 hour, 0.1 C in 10 hours.
We know from Equation (13.10) that Ecell depends on the difference in chemical potential
of Li in the cathode and anode. If we use Li metal as the anode, its chemical potential there is
constant, and the −Ecell directly reflects the changing Li chemical potential at the cathode.
The two plots showing Ecell as a function of Li content for the two different electrodes in
Figure 13.28 have different shapes that reflect the different chemistries that occur.
LixTiS2 shows simple solid-solution-like behavior as the Li content changes. We can
understand the shape of the curve in terms of how the chemical potential of inserting
a single Li+ and electron is influenced by evolving enthalpic and entropic contributions as
a function of x (defect energetics were discussed in Section 2.3).44 In a simple model, the
entropic term reflects the number of ways of placing the cation in the host sites available
(configurational entropy) and is therefore particularly important near x = 0 and x = 1 where
the increase in configurational entropy per Li is highest. The enthalpic term can be related to
repulsive Li+–Li+ interactions that lead to an increase in μcathode
Li with increasing x, thus
causing Ecell to decrease with x. The combination of these effects leads to the S-shaped cell
voltage observed. The difference in voltage on charge and discharge is due to the internal

3.50 4.50

3.00 4.00

charge
))

(V)
vollttaaggee(V(V

2.50 3.50
oltage (V
voltage )

2.00 3.00
V
V

discharge
1.50 2.50

1.00 2.00
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
x in LixTiS2 x in LixFePO4

Figure 13.28 Voltage versus Li+/Li for LixTiS2 and LixFePO4 at current density 10 mA/cm2 and discharge
rate 0.1 C after several charge/discharge cycles [27, 28].

     
∂G ∂H ∂S
44
Since μ is the partial molar Gibbs energy at constant pressure it follows that μ ¼ ¼ T .
∂n ∂n ∂n
 13.8 Intercalation Chemistry and Its Applications 569

resistance R of the cell to the ionic current I (which equals the external current). This reduces/
increases the discharge/charge voltage by an Ohm’s law I·R term.
In contrast, the electrochemical behavior of LixFePO4 (Figure 13.28, right) is affected by
the miscibility gap for 0.1 < x < 0.95 that leads to a different Ecell dependence on x.45 When
the cell is discharged between x = 0.1 and x = 0.95, the chemical change that occurs is a direct
conversion of Li0.1[cathode] to Li0.95[cathode], and there are two phases present. The cell
voltage versus x therefore shows a plateau.
In practice, the intercalation chemistry and therefore electrochemical properties of cathodes
can be much more complex due to changes in the Li-ordering patterns within layers as
a function of x, or changing coordination requirements of the cathode transition metal on
reduction. For example, Mn4+ in an MnO2 cathode would become Jahn–Teller active on
reduction to Mn3+ during discharge, changing local coordination geometry and influencing
Ecell. Voltage steps might also occur when particularly favorable Li+ sites in a material become
fully occupied, or a particular host band becomes filled by electrons. A careful combination of
electrochemical and structural studies (often in situ) is required to unravel this behavior.

13.8.5 Other Lithium Battery Electrode Materials

Despite its excellent properties, LiCoO2 has a drawback of high cost.46 In addition, the cell’s
internal resistance means that full charge and discharge rates cannot be quicker than ~1 hour in
order to avoid overheating. In addition to the NMC materials mentioned above, there has also
been considerable interest in materials with the olivine47 structure such as LiFePO4 [29]. The
structure (Figure 13.29) contains edge- and corner-sharing FeO6 octahedra and PO4 tetrahedra
and has chains of edge-sharing LiO6 octahedra that allow a 1D Li diffusion pathway through

PO4

LiO6

FeO6

Figure 13.29 LiFePO4 viewed down c and b axes.

45
For example a material of nominal composition Li0.525[cathode] would phase-separate to domains of 0.5Li0.1
[cathode] and 0.5Li0.95[cathode].
46
Co is relatively scarce, and projected usage in automotive batteries exceeds current mining production.
47
Olivine is a mineral with composition (Fe,Mg)2SiO4.
570 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

the material. By coating individual crystallites with thin layers of carbon, it has been possible to
overcome the low intrinsic electronic conductivity of LiFePO4 (~10−9 S/cm) and produce cells
with capacities of ~160 A h/kg (close to the theoretical limit of 170 A h/kg), which can be
charged and discharged 30000 times. As discussed in the previous section, a bulk
LixFePO4 cathode acts as a two-phase system with x = 0.1 and x = 0.95 at room temperature
and a resultant flat Ecell versus x curve (Figure 13.28, right). A single-phase behavior, with the
normal sloping Ecell versus x curve, only occurs above ~450 °C or when working with
metastable 40 nm particles. These small particles give good battery performance even without
carbon coating. A range of related materials with Mn, Co, and Ni substitution, and mixed-
anion materials such as LiVPO3F have been studied or already commercialized. These cathode
materials are of particular interest for high-power applications.
In the discussion so far, we’ve focused on the cathode systems that have already been
exploited commercially. There is, however, a huge research effort to improve the properties of
all the components used in rechargeable batteries, how they can be prepared from cheaper
materials, and how they can eventually be recycled. Alternative anodes to LixC6 have been
investigated, including Si and Sn or Sb-based alloys. These have higher capacities (the saturated
stoichiometry is up to Li4Si for pure Si) but undergo large volume changes (300% for Si) on
charging, meaning that they must be prepared as sponges or nanoparticles surrounded by an Li+
conducting medium. Oxides such as Li4Ti5O12 have also been studied as potential anodes. They
lead to lower cell voltages (Figure 13.26) but can be charged more rapidly without Li metal
plating the anode, improving safety. Methods that could allow Li-metal anodes to be re-
introduced are also important, as Li metal has a much higher gravimetric capacity than
LiC6 (3361 mA h/g versus 339 mA h/g). These could include using solid polymeric or ceramic
electrolytes, changes to electrolyte solvents/additives, or more advanced thin-film cell designs to
give more controlled Li deposition. We’ve also seen that the cathode in a cell can be highly
oxidizing (potentials of > 4 V versus Li+/Li), meaning that even non-aqueous electrolytes are
operating outside their thermodynamic stability window of ~3.5 V. Cell reversibility and
ultimate lifetime is frequently associated with electrolyte decomposition at the electrode inter-
face, and much research has been performed to try and control this.
One exciting idea is that it may not be necessary to rely on the relatively few host
structures that will allow intercalation reactions for electrode materials. It is known, for
example, that reaction kinetics of nanostructured materials are sufficiently fast that
conversion reactions such as MaXb + (b·n)Li = aM + bLinX (X = O, S, N, P, F, H) can
be used reversibly at reasonable operating temperatures [30]. The reactions proceed via
formation of nanometer-scale islands of the transition metal embedded in LinX, and
controlling the morphology of the reactants and products is crucial. There is also
considerable interest in lithium–oxygen and lithium–sulfur batteries. The former
operate on the reaction 2Li + O2 = Li2O2 and could offer capacities up to 1200
mA h/g, but there are huge materials challenges that need to be overcome before
commercialization [31].
 13.8 Intercalation Chemistry and Its Applications 571

Box 13.3 Materials Spotlight: Supercapacitors

In addition to the batteries and fuel cells discussed in the main text, there is a third
category of electrochemical devices used for energy storage: electrochemical capacitors or
supercapacitors. The basic set-up of a supercapacitor is shown in Figure B13.3.1 and
again contains two electrodes (usually called the anode and cathode even though redox
reactions don’t necessarily occur) in contact with a liquid electrolyte. At the electrode–
electrolyte interface, ions accumulate, attracted by the charge on the electrode surface,
forming an electrically charged double layer separated by 2–10 Å. This gives rise to
a capacitance we can estimate from the expression C ¼ Q=V ¼ ε0 εr A=d as being of the
order of 10 μF/cm2, assuming εr of 10 for ions in water and a separation d = 9 Å. If an
electrode such as a porous carbon is used, which can have a surface area up to 1000 m2/g,
one can achieve extremely high capacitances of ~100 F/g [32]. Since the device needs two
electrodes, its overall capacitance is given by C ¼ 1=ð1=C1 þ 1=C2 Þ ¼ C1 =2 for symmetric
electrodes. The specific capacitance is then 25 F/g (given double the mass), much higher
than conventional capacitors which are in the pF to μF range. The energy stored by such
a supercapacitor is given by 1/2 CV 2. With organic electrolytes, which are stable to
decomposition up to a few volts, energy densities of ~25 W h/kg are possible.

− + − + −
+
negative electrode surface

− + − + −
negative electrode

positive electrode

−
+
−

−
+

+
− +

− +
+ −

−
+

− +
solvated
− + − +
cation
− + − + −
+

− + − + − solvent

electrolyte
separator

Figure 13.3.1 Schematic of a symmetric supercapacitor. Right-hand figure shows a close-up of the electrical
double layer formed at the negative electrode. The overall capacitance is due to the charge from any
specifically adsorbed ions and solvent and the charge in the outer layer.

One key difference between supercapacitors and batteries or fuel cells is the timescale of the
processes responsible for energy storage. The double layer at the electrode surface of
a supercapacitor forms rapidly (10−8 s) compared to the redox reactions in batteries and fuel
cells (10−2–10−4 s). This means that charging and discharging processes are extremely fast. There is
also no structural change needed in these processes and there are fewer side reactions between
electrodes and electrolytes. Supercapacitors can therefore be charged and discharged millions of
times without capacity loss.
Significantly higher capacitance can be achieved by using electrochemically active electrodes in
what are called faradaic supercapacitors or pseudocapacitors. In these devices, redox reactions take
572 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

place at the electrode surface, with charge passing from the electrolyte to the electrode in a process
similar to that in batteries [33]. Capacitances 10–100 times higher than for electrostatic double-layer
electrodes are possible, though one pays the price of lower cycling stability and lower power as the
(dis)charging reactions are slower. Hybrid supercapacitors use one electrode of each type to overcome
some of these limitations. Electrodes using conducting polymers and metal oxides such as RuO2,
MnO2, and Co3O4 have been extensively investigated. RuO2 works via reactions such as RuO2 + xH+
+ xe− = RuO2−x(OH)x (0 < x < 2) and has many attractive properties: a theoretical capacitance of up
to 2000 F/g, reversible redox reactions, good electronic and proton conductivity, and high stability. It
has the drawback of being very expensive.
The properties of the different energy devices we’ve considered are captured in the Ragone plot
shown in Figure B13.3.2, which plots their specific power (in W/kg) and specific energy (W h/kg).
Fuel cells are high-energy sources but have low power, supercapacitors have low energy but high
power, and batteries lie somewhere in between. While none of these systems can approach the
simultaneous high energy and power of a traditional combustion engine, hybrid systems offer this
important option. For example, in an automotive application one could combine a high-energy
fuel cell (for range) with batteries and supercapacitors to provide high power when needed for
acceleration. The high charging rate of supercapacitors also means that they can be used for
regenerative energy storage on braking. Supercapacitors are also widely used when a short-term
energy supply is needed such as in power-supply backups.

107

106
capacitors

combustion
specific power (W /kg )

105 engines, gas

turbines
104
supercapacitors
103

102
batteries fuel
cells
101

1
0.01 0.1 1 10 100 1000
specific energy (W h/kg)

Figure B13.3.2 A Ragone plot comparing the power and energy density of different electrochemical storage
devices. The sloped dashed lines indicate the relative times to store or extract energy.
 13.9 Problems 573

13.9 Problems
13.1 Write out half equations, the cell reaction, and determine the standard potential of the
following cells given standard electrode potentials E°(Cu2+/Cu) = +0.34 V, E°(Sn2+/Sn) =
−0.14 V, E°(Co3+/Co2+) = +1.81 V, E°(Sn4+/Sn2+) = +0.15 V: (a) Sn(s)|SnSO4(aq)||CuSO4
(aq)|Cu(s), (b) Pt|Co3+(aq),Co2+(aq)||Sn4+(aq),Sn2+(aq)|Pt.
13.2 In its charged state, a lead–acid battery contains PbO2 and Pb electrodes. The half
equations that occur during discharge can be written as below. Give the overall cell
equation and estimate E°cell. State which electrode is the cathode and which the anode.

PbO2(s) + 3H3O+(aq) + HSO4−(aq) + 2e− ↔ PbSO4(s) + 5H2O(l); E° = 1.685 V

PbSO4(s) + H3O+(aq) + 2e− ↔ Pb(s) + HSO4−(aq) +H2O(l); E° = −0.356 V

13.3 The non-rechargeable alkaline battery contains a Zn anode, MnO2 cathode, and
KOH electrolyte, with MnO2 present slightly in excess of a 1:2 molar ratio for
safety reasons. Propose a balanced equation for the cell reaction and give individ-
ual half equations.
13.4 Nickel–cadmium batteries rely on the reaction between NiO(OH) and Cd metal with an
alkaline electrolyte, with both metals forming hydroxides. Propose a balanced equation
for the cell reaction and give individual half equations.
13.5 Use the Nernst equation to estimate the potential of an electrochemical cell Zn(s)|Zn2+
(c1, aq)||Zn2+(c2, aq)|Zn(s) in which Zn concentrations c1 and c2 are 0.3 M and 0.8 M,
respectively.
13.6 Write down the half equations that occur at each electrode in the oxygen sensor of
Figure 13.17 and state the overall cell equation. Oxygen sensors of this type are used
in car engines to monitor the air to fuel ratio via changes in the partial pressure of
O2 in the exhaust gas relative to air. Calculate the voltage expected from the sensor
for the following air to fuel (A:F) ratios: (a) a lean mixture of A:F = 10.5 and p(O2)
= 1×10−14 Pa, (b) a stoichiometric ratio of A:F = 14.7 and p(O2) = 5×10−11 Pa,
(c) a rich mixture of A:F = 18 and p(O2) = 4500 Pa. Assume the partial pressure of
O2 in air is 21.2 kPa and the exhaust gases are at 1000 K.
13.7 The table below contains conductivity data for a crystal of NaCl. Comment on
the shape of a plot of ln(σT) versus 1/T. Estimate the activation energy for
vacancy migration and Schottky defect formation from your graph.

T (K) 550 600 650 700 750

σ (S/cm−1) 1.60×10−8 5.06×10−8 1.37×10−7 3.51×10−7 9.92×10−7

T (K) 800 850 900 950 1000

σ (S/cm−1) 3.36×10−6 1.25×10−5 4.47×10−5 1.47×10−4 4.34×10−4

574 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

13.8 AgI is often described as a fast ion conductor. Given that NaCl has around 10−4
carriers per formula unit immediately below its melting point and that conductiv-
ities are ≈ 1 S/cm (AgI) and ≈ 10−5 S/cm (NaCl), comment on the term fast ion
conductor. Assume cell parameters of 5.84 Å for NaCl and 5.06 Å for AgI.
13.9 The table below contains conductivity data for two materials. The melting point of
PbF2 is 1103 K. (a) Describe the structure of each material and the origin of the
relatively high conductivities. (b) Comment on the temperature dependence of conduct-
ivities. (c) For Zr1−xYxO2−x/2, estimate the activation energy for charge-carrier migra-
tion. (d) State two possible uses of Zr1−xYxO2−x/2.

T (K) PbF2 σ (S/cm) Zr1−xYxO2−x/2 σ (S/cm)

670 0.1 0.0001

840 4.6 0.002
1250 4.6 0.1

13.10 At high temperature, La2Mo2O9 has a structure closely related to cubic β-SnWO4, which
contains WO4 tetrahedra and Sn in a distorted six-coordinate environment. On cooling
through 580 °C, a 2a × 3b × 4c monoclinic superstructure is formed. Correlate these
observations with the conductivity data in Figure 13.18, and suggest a mechanism for
conductivity above 580 °C.
13.11 The table below lists the in-plane conductivity for graphite and K0.125C at two different
temperatures. Comment on these values.

Substance σ(90 K) (S/cm) σ(300 K) (S/cm)

Graphite 1.3×104 2.5×104

K0.125C 1.1×106 1.1×105

13.12 The table below lists d spacings of the first four strong diffraction peaks observed in the
diffraction patterns of graphite and two Rb intercalation compounds. Note that the
graphite interlayer spacing is 3.35 Å and that the preferred orientation of crystallites
means that (00l) reflections are typically strongest. Comment on these data.

Substance dhkl (Å) dhkl (Å) dhkl (Å) dhkl (Å)

Graphite 3.35 1.68 1.12 0.84

Rb0.03C 12.34 6.17 4.11 3.08
Rb0.125C 5.65 2.83 1.41 0.94

13.13 The structure of the stage-1 KxC intercalate has been reported in space group Fddd with
a pseudohexagonal unit cell of a = 4.92 Å, b = 8.51 Å, c = 21.4 Å with K at 0 0 0, C1 at
1 1
/4 /12 1/8, C2 at 1/4 1/12 5/8. With the assistance of a structure-drawing program: (a) Sketch
 13.9 Problems 575

the structure of a single layer and annotate a simple 2D unit cell. (b) Comment on the
stacking sequence of layers relative to graphite. (c) State the value of x in KxC and
comment on the value given that bcc K has a = 5.21 Å. (d) Comment on the fact that Rb
(Rb metal bcc a = 5.63 Å) and Cs (a = 6.06 Å) intercalates have similar structures.
13.14 The Li graphite intercalate has a limiting composition LiC6. Sketch a possible unit cell in
2D and estimate the Li–Li distance assuming a C–C bond length of 1.42 Å.
13.15 Calculate the theoretical capacity in A h/kg for the electrode materials: (a) Li metal
anode, (b) LiC6 anode, (c) LiTiS2 cathode, (d) LiCoO2 cathode.
13.16 For LiC6 calculate the theoretical capacity in A h/kg of carbon.
13.17 Assuming densities for LiFePO4, carbon black, and Teflon® to be 3.6, 1.8, and 2.2 g/cm3
respectively, calculate the reduction in volumetric density of a cathode comprising 10 wt%
carbon and 5 wt% Teflon.
13.18 Calculate the energy density in W h/kg for the Li/TiS2 cell described in Section 13.8.2.
Assume an operating voltage of 2 V and that the electrolyte and packaging make up
50% of the mass.
13.19 Calculate the energy density in W h/kg for a battery based on a LixC6 anode and a
Li1−xCoO2 cathode, operating at an average voltage of 3.6 V. Assume that the electro-
lyte and casing make up 50% of the battery mass and that the cathode can only be
charged to Li0.5CoO2. Repeat the calculation for a pure Li anode assuming the cell
voltage is 0.6 V higher. Comment on these values.
13.20 The figure below shows E against x at 30 °C for a cell with a spinel-based LixMn2O4
cathode (0 < x < 1.8) and a Li anode. (a) Describe the structure you might expect
for a LiMn2O4 spinel. (b) State what can you conclude from Ecell versus x in the regions 0
< x < 0.5, 0.5 < x < 1, and 1 < x < 1.8. (c) State what could cause the marked drop in Ecell
at x = 1.

4.5

4.0
Ecell (V)

3.5

3.0

2.5
0.0 0.5 1.0 1.5 2.0
x in LixMn2O4
576 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

13.21 Li1Mn2O4 (a = 8.2495 Å) is cubic, whereas Li2Mn2O4 is tetragonal (a = 5.653 and

c = 9.329 Å). Suggest a reason for lower symmetry of Li2Mn2O4. Calculate the
volume change upon its formation from cubic Li1Mn2O4 and state how that might
impact the long-term performance of a cathode cycling between these two
compositions.

13.10 Further Reading

M. Winter, R.J. Brodd, “What are batteries, fuel cells and supercapacitors” Chem. Rev. 104 (2004),
4245–4269. The October 2004 issue of Chemical Reviews contains a number of articles on energy-
related materials.
S. Hull, “Superionics: Crystal structures and conduction processes” Rep. Prog. Phys. 67 (2004),
1233–1314.
L. Malavasi, C.A.J. Fisher, M.S. Islam, “Oxide-ion and proton conducting electrolyte materials for
clean energy applications: Structural and mechanistic features” Chem. Soc. Rev. 39 (2010),
4370−4387.
P.G. Bruce (editor), “Solid State Electrochemistry” (1995) Cambridge University Press.
J. Maier, “Thermodynamics of electrochemical lithium storage” Angew. Chem. Int. Ed. Engl. 52
(2013), 4998–5026.
M. Armand, J.-M. Tarascon, “Building better batteries” Nature 451 (2008), 652–657.
M.S. Whittingham, “Lithium batteries and cathode materials” Chem. Rev. 104 (2004), 4271–4301.
J.B. Goodenough, Y. Kim, “Challenges for rechargeable Li batteries” Chem. Mater. 22 (2010),
587–603.
J.B. Goodenough, “Evolution of strategies for modern rechargeable batteries” Acc. Chem. Res. 46
(2013), 1053–1061.
L. Croguennec, M. Rosa Palacin, “Recent achievements on inorganic electrode materials for
lithium-ion batteries” J. Am. Chem. Soc. 137 (2015), 3140–3156.
T. Famprikis, P. Canepa, J.A. Dawson, M.S. Islam, C. Masquelier, “Fundamentals of inorganic
solid-state electrolytes for batteries” Nat. Mater. 18 (2019), 1278–1291.

13.11 References
[1] B.C.H. Steele, “Material science and engineering: The enabling technology for the commercial-
isation of fuel cell systems” J. Mater. Sci. 36 (2001), 1053–1068.
[2] S. Hull, “Superionics: Crystal structures and conduction processes” Rep. Prog. Phys. 67 (2004),
1233–1314.
[3] M. O’Keeffe, B.G. Hyde, “The solid electrolyte transition and melting in salts” Philos. Mag. 33
(1976), 219–224.
[4] A. Poullikkas, “A comparative overview of large-scale battery systems for electricity storage”
Renewable and Sustainable Energy Rev. 27 (2013), 778–788.
 13.11 References 577

[5] Z. Gadjourova, Y.G. Andreev, D.P. Tunstall, P.G. Bruce, “Ionic conductivity in crystalline
polymer electrolytes” Nature 412 (2001), 520–523.
[6] K.D. Kreuer, “Proton conductivity: Materials and applications” Chem. Mater. 8 (1996), 610–641
[7] K.D. Kreuer, “Proton conducting oxides” Annu. Rev. Mater. Res. 33 (2003), 333–359.
[8] S.M. Haile, D.A. Boysen, C.R.I. Chisholm, R.B. Merle, “Solid acids as fuel cell electrolytes”
Nature 410 (2001), 910–913.
[9] http://commons.wikimedia.org/wiki/File%3ATubular_sofc_de.png.
[10] V.V. Kharton, F.M.B. Marques, A. Atkinson, “Transport properties of solid oxide electrolyte
ceramics: A brief review” Solid State Ionics 174 (2004), 135–149.
[11] R.D. Shannon, “Revised effective ionic radii and systematic studies of interatomic distances in
halides and chalcogenides” Acta Crystallogr. Sect. A 232 (1976), 751–767.
[12] M. Mogensen, N.M. Sammes, G.A. Tompsett, “Physical, chemical and electrochemical proper-
ties of pure and doped ceria” Solid State Ionics 129 (2000), 63–94.
[13] X. Kuang, J.L. Payne, M.R. Johnson, I. Radosavljevic Evans, “Remarkably high oxide ion
conductivity at low temperature in an ordered fluorite-type superstructure” Angew. Chem. Int.
Ed. Engl. 51 (2012), 690–694.
[14] S. Tao, J.T.S. Irvine, “A redox-stable efficient anode for solid-oxide fuel cells” Nat. Mater. 2
(2003), 320–323.
[15] J.C. Ruiz-Morales, J. Canales-Vázquez, C. Savaniu, D. Marrero-López, W. Zhou, J.T.S. Irvine,
“Disruption of extended defects in solid oxide fuel cell anodes for methane oxidation” Nature 439
(2006), 568–571.
[16] S. Sengodan, S. Choi, A. Jun, T.H. Shin, Y.-W. Ju, H.Y. Jeong, J. Shin, J.T.S. Irvine, G. Kim,
“Layered oxygen-deficient double perovskite as an efficient and stable anode for direct hydrocar-
bon solid oxide fuel cells” Nat. Mater. 14 (2015), 205–209.
[17] S. Yoo, A. Jun, Y.-W. Ju, D. Odkhuu, J. Hyodo, H.Y. Jeong, N. Park, J. Shin, T. Ishihara,
G. Kim, “Development of double-perovskite compounds as cathode materials for
low-temperature solid oxide fuel cells” Angew. Chem. Int. Ed. Engl. 53 (2014), 13064–13067.
[18] M.S. Dresselhaus, G. Dresselhaus, “Intercalation compounds of graphite” Adv. Phys. 51 (2002),
1–186.
[19] N. Daumas, A. Herold, “Relations between phase concept and reaction mechanics in graphite
insertion compounds” C. R. Seances Acad. Sci., Ser. C 268 (1969), 373.
[20] D.W. Murphy, C. Cros, F.J. DiSalvo, J.V. Waszczak, “Preparation and properties of LixVS2 (0 ≤
x ≤ 1)” Inorg. Chem. 16 (1977), 3027–31
[21] J.C. Hunter, “Preparation of a new crystal form of manganese dioxide: λ-MnO2” J. Solid State
Chem. 39 (1981), 142–147.
[22] A.R. Armstrong, P.G. Bruce, “Synthesis of layered LiMnO2 as an electrode for rechargeable
lithium batteries” Nature 381 (1996), 499.
[23] F.C. Capitaine, P. Gravereau, C. Delmas, “A new variety of LiMnO2 with a layered structure”
Solid State Ionics 89 (1996), 197–202.
[24] J.B. Goodenough, Y. Kim, “Challenges for rechargeable Li batteries” Chem. Mater. 22 (2010),
587–603.
[25] J.M. Tarascon, S. Grugeon, M. Morcette, S. Laruelle, P. Rozier, P. Poizot, “New concepts for the
search of better electrode materials for rechargeable lithium batteries” C. R. Chim. 8 (2005), 9–15.
[26] S-T. Myung, H-J. Noh, S-J. Yoon, E-J. Lee, Y-K. Sun, “Progress in high-capacity core–shell
cathode materials for rechargeable lithium batteries” J. Phys. Chem. Lett. 5 (2014), 671–679.
578 Energy Materials: Ionic Conductors, Mixed Conductors, and Intercalation Chemistry

[27] M.S. Whittingham, “Chemistry of intercalation compounds: Metal guests in chalcogenide hosts”
Prog. Solid State Chem. 12 (1978), 41–99.
[28] J.M. Tarascon, M. Armand, “Issues and challenges facing rechargeable lithium batteries” Nature
414 (2001), 359–367.
[29] A.K. Padhi, K.S. Nanjundaswamy, J.B. Goodenough, “Phospho-olivines as positive electrode
materials for lithium batteries” J. Electrochem. Soc. 144 (1997), 1188–1194.
[30] J. Cabana, L. Monconduit, D. Larcher, M. Rosa Palacín, “Beyond intercalation-based Li-ion
batteries: The state of the art and challenges of electrode materials reacting through conversion
reactions” Adv. Mater. 22 (2010), E170–E192.
[31] P.G. Bruce, S.A. Freunberger, L.J. Hardwick, J.M. Tarascon, “Li–O2 and Li–S batteries with
high energy storage” Nat. Mater. 11 (2012), 19–29.
[32] X. Zhao, B. Mendoza Sánchez, P.J. Dobson, P.S. Grant, “The role of nanomaterials in
redox-based supercapacitors for next generation energy storage devices” Nanoscale 3 (2011),
839–855.
[33] G. Wang, L. Zhang, J. Zhang, “A review of electrode materials for electrochemical
supercapacitors” Chem. Soc. Rev. 41 (2012), 797–828.
14 Zeolites and Other Porous Materials

In this chapter we will discuss the chemistry of porous materials, that is materials
whose 3D structure contains pores or channels accessible by chemical species. IUPAC
has defined three classes of porous materials—microporous (<2 nm), mesoporous
(2–50 nm), and macroporous (>50 nm). We’ll see that these pore sizes allow uptake
of atoms and molecules leading to important technological applications. For example,
crude-oil cracking using microporous zeolites underpins a multibillion US dollar indus-
try. Materials with significantly larger pore sizes, such as the mesoporous MCM family
of aluminosilicates and similar materials based on metal oxides and chalcogenides, have
additional exciting applications. We will also discuss metal–organic frameworks
(MOFs), where one can design materials with extremely high porosity and correspond-
ingly large internal surface areas.

14.1 Zeolites
Naturally occurring zeolites were first identified in 1756 by the Swedish mineralogist Axel
Cronstedt, who noted that certain minerals appeared to bubble and froth when heated.
He named these minerals after the Greek words zeo (boil) and lithos (stone). We will see
that this behavior is directly related to the structures and chemistry of these materials.
Zeolites can be described as framework aluminosilicates constructed from corner-sharing
SiO4/2 and AlO4/2 tetrahedra, which contain cations, anions, water, and/or other guest
molecules in the framework pores. As such, we can represent the general formula of a
zeolite as Gn+x/n[(AlO2)x(SiO2)1−x]⋅mH2O.1 Materials with similar structures but contain-
ing other elements on the tetrahedral sites are called zeotypes.

1
We show cations only in this formula; several zeolites contain cations and anions in their pores.

579
580 Zeolites and Other Porous Materials

We can understand many of the structural features of aluminosilicates by starting with a

simple SiO2 structure such as cubic β-cristobalite shown in Figure 1.39.2 If we replace half the
Si4+ sites in each unit cell with Al3+, we will need some form of charge compensation. One
possibility is to introduce Na+ ions into 12-coordinate sites between the corner-linked
tetrahedra to obtain NaAlSiO4, the mineral carnegieite. A similar derivation can be per-
formed for the different crystalline polymorphs of SiO2, giving the so-called stuffed silica
structures.
Zeolite structures are conceptually similar, and the large number of ways in which
tetrahedra can be corner-linked gives a rich structural diversity. Each zeolite framework is
built from different rings (see Section 1.4.3) forming cages that surround various “pores” and
“channels” containing the charge-balancing ions. The important chemistry of zeolites is
derived from their ability to exchange non-framework ions, absorb molecules into their
pores, and catalyze chemical reactions, and is a direct consequence of their structure. The 3D
porosity of zeolites gives them extremely high effective surface areas of up to 900 m2/g.3
From the discussion above, the distinction between zeolites and other aluminosilicates
may seem slightly arbitrary. In practice, it is based on the relative amounts of “pore” and
“framework” present and quantified by the framework density (FD) or number of tetrahe-
dron-center atoms (T) per nm3. Corner-linked tetrahedral aluminosilicates typically have
framework densities ≳ 21 T/nm3, while zeolites have low framework densities ranging from
~10 to ~20 T/nm3 (see Table 14.1).

Table 14.1 Characteristics of selected zeolites and related phosphates.

Name Datea Frameworkb Ring Channels (Å) T/nm3 vol%c Coded SBUe

Sodalite 1930 Al6Si6O24 6 – 17.2 0 SOD 6–2

Zeolite A 1956 Al12Si12O48 8 4.1 × 4.1 12.9 21.4 LTA 8
Faujasite 1958 Al58Si134O384 12 7.4 × 7.4 12.7 27.4 FAU 6–6
ZSM-5 1978 AlnSi96−nO192 n<27 10 5.5 × 5.1 17.9 9.8 MFI 5–1
Mordenite 1961 Al8Si40O96 12 7.0 × 6.5 17.2 12.3 MOR 5–1
UTD-1 1996 Si64O128 14 8.1 × 8.2 17.2 15.6 DON 5–3
ITQ-33 2006 Si20Ge26O92 18 12.3 × 12.3 12.3 25.1 ITT 6
VPI-5 1988 Al18P18O72 18 12.7 × 12.7 14.2 25.4 VFI 6
Cloverite 1991 Ga96P96O372F24(OH)24 20 4 × 13.2 11.1 33.8 -CLO 4–4
a
Date of first reference in atlas of zeolite types. bFramework composition of species first used to establish the
framework type. cVolume percent accessible by a 2.8 Å diameter spherical probe molecule (see text). dA three-letter
code used to identify the framework (see Section 14.1.1); a hyphen precedes for frameworks where the T sites are
not all 4-connected. eSecondary building units (SBUs, shown in Figure 14.4).

2
Fd3m (227, origin 1) cubic, a = 7.133 Å at 300 °C, Si at 0 0 0 (8a), O at ⅛ ⅛ ⅛ (16c ) in the ideal structure.
3
The “soccer pitch” is a common non-SI unit of area. The most common championship pitch size is 7140 m2.
 14.1 Zeolites 581

14.1.1 Representative Structures of Zeolites

All zeolites have a 3D framework of corner-sharing tetrahedra. One common unit found in
several important zeolites is the sodalite or β cage, an approximately spherical unit made up
of 24 tetrahedra (Figure 1.57). There are various ways in which this unit can be represented,
as shown in Figure 14.1. Figure 14.1a shows a ball-and-stick view, Figure 14.1b a polyhedral
view, and Figure 14.1c a view in which oxygen atoms have been omitted and tetrahedral
centers are linked by a single line. This third view simplifies the sodalite cage into a truncated
cuboctahedron (24 vertices) and is frequently used when drawing structures of zeolites. It
also emphasizes the two types of rings of tetrahedra present: 6T-rings4 (of 6 Al/Si central
atoms) and 4T-rings. Figure 14.1d gives an idea of the size of the ring openings. Taking a
1.35 Å Shannon [1] ionic radius for O2− in ideal cage geometry, the O–O separations across
6T- and 4T-rings are around 2.4 Å and 1 Å respectively.
If we share each of the 4T-rings of a sodalite cage with another cage, we create the
structure shown in Figure 14.2a, which we can think of as a sodalite cage occupying each
corner of a cube. Interestingly, the truncated cuboctahedron of the sodalite cage tessellates in
three dimensions such that the cavity we might expect to form at the center of our cube is
itself a sodalite cage. This is the structure of the mineral sodalite,5 which has a typical
formula Na8Cl2[(AlO2)6(SiO2)6]⋅8H2O6 and a framework density of 17.2 T/nm3.
The space available inside different zeolite frameworks can be estimated by calculating the
volume percent (vol%) of the unit cell that can be occupied by a spherical probe molecule of
water (assumed to be a 2.8 Å diameter sphere). This is called the occupiable volume. However,
there is a snag. For the sodalite framework, 10.3% of the volume can be occupied but cannot

(a) (b) (c) (d)

Figure 14.1 Views of a sodalite cage showing (a) atoms involved, (b) TO4/2 tetrahedra, (c) T centers linked
by a line, and (d) with oxygens shown using a radius of 1.35 Å. The central sphere emphasizes the
approximately spherical shape of the cage. In (a)–(c) it is drawn with a radius of 4.2 Å; in (d) a 3.1 Å radius
sphere just contacts the O atoms.

4
In the Liebau nomenclature of ordinary silicates (Appendix C), the 6-ring of tetrahedra would be termed a “sechser
ring”. This 6-ring is part of the structure’s 6–2 secondary building units (defined in Figure 14.4).
5
We note that sodalite is mineralogically classified as a feldspathoid (similar to feldspar), yet its structural features are
similar to the other zeolites we discuss.
6
The mineral sodalite contains Cl− at the center of each cage tetrahedrally coordinated by Na+.
582 Zeolites and Other Porous Materials

(c)
(b)
(a)

sodalite zeolite A faujasite or zeolite X or Y

Figure 14.2 Schematic representation of how the structures of (a) sodalite, (b) zeolite A, and (c) faujasite
can be derived by sharing the light gray sodalite cage via the dark gray linkers.

be reached once the sodalite has been synthesized. This volume lies inside the sodalite cages,
but the size of the ring opening makes it inaccessible to a molecule. The accessible volume is
therefore 0% and the corresponding accessible surface area 0%.
Instead of directly sharing 4T-rings between sodalite cages, the cages can be linked by a T–
O–T bridge, as shown in Figure 14.2b. Remember that each line in this figure represents a T–
O–T linkage (two tetrahedra sharing a vertex oxygen). The introduction of this extra linker
unit (a 4–4 or double-4-ring, see Figure 14.4) leads to a far more open structure and creates a
large 48-sided cavity (called an α cage) between the sodalite cages. The structure also now
contains large 8T-ring windows, with a pore size of about 4.1 Å × 4.1 Å, which link the α cages,
resulting in a much lower framework density of 12.9 T/nm3. The pore sizes are comparable to
the dimensions of small molecules. The accessible pore volume is 21.4% and the accessible area
281.5 Å2 per cell, which corresponds to a surface area of 1205 m2/g. This is the structure of
zeolite A that has a typical composition Na12[(AlO2)12(SiO2)12]⋅27H2O. The sites occupied by
non-framework Na+ and H2O depend on the precise composition and level of hydration. In
the Na form of zeolite A, eight of twelve Na+ adopt the Na1 site adjacent to the 6T-rings that
open onto the β cage, three occupy Na2 sites that lie near the center of the 8T-rings and one is in
the center of the α cage (Figure 14.3). Of the 27 water molecules, 20 form a pentagonal
dodecahedron within the α cage similar to that found in water clathrates. Approximately
four water molecules lie in the β cage, and the remaining approximately three hydrate the Na+
cation in the center of the α cage.
Instead of linking 4-rings via 4–4 units one can link 6T-rings via 6–6 units in a similar way.
This leads to the faujasite structure (Figure 14.2c), in which the sodalite cages again adopt a
cubic arrangement, but with positions analogous to those of carbon in diamond. The structure
contains 12T-rings leading to larger pore sizes of 7.4 Å × 7.4 Å, the formation of a large
supercage with an internal diameter ~14 Å, an overall framework density of 12.7 T/nm3,
accessible pore volume of 27.4%, and surface area of 1211 m2/g. Faujasite itself is a naturally
 14.1 Zeolites 583

O
Figure 14.3 Approximate
Si
Na sites in zeolite A. Na1
lies at the center of the 6-
ring, Na2 at the center of
Na2 the 8-ring, and Na3 (col-
Na1 ored white) at the center of
the α cage.

occurring mineral (Ca,Mg,Na2)29[(AlO2)58(SiO2)134]⋅240H2O. Synthetic analogues include

zeolite X Na86[(AlO2)86(SiO2)106]⋅220H2O and zeolite Y; X and Y differ only in their Si:Al
ratio, X has a ratio around 1.0–1.4 and Y typically above ~2.5. The formula is written in this
apparently complex way to reflect the fact that, due to the high space-group symmetry (Fd3̄m),
each T atom on a general position (192i Wyckoff site) generates 192 equivalent atoms in the
unit cell. The faujasite structure again contains a number of potential cation sites, whose
occupation depends on the degree of hydration and the precise cations present.
In addition to the sodalite-cage zeolites, many other topologies are possible. The unique
framework types known are documented in the Atlas of Zeolite Framework Types [2]. The
2018 online atlas (www.iza-structure.org/databases/) listed 230 ordered frameworks, with
each given a unique three-letter code identifier. It also contains coordinates, drawings, and
interactive 3D views of all the framework types known.
There are several useful ways of categorizing different zeolite structures. One of the early
systems used the secondary building units (SBUs) they contain [2]. These are units containing
several tetrahedra (basic building units, BBUs) from which the entire framework can be built.
The SBUs important for the zeolites in this chapter are shown in Figure 14.4 and included in
Table 14.1. As more framework types were discovered, the zeolite community switched to
discussing structures in terms of the common composite building units (CBUs) present [3] or the
more mathematical language of natural tilings [4, 5]; information on both is listed for each
framework in the online zeolite atlas [2]. To help summarize a given framework, the atlas lists
two other terms—the coordination sequence and the vertex symbol. The combination of these is
thought to be unique to a given framework and can be used to decide if a newly prepared
material has a novel framework. The coordination sequence is evaluated for each crystallo-
graphically unique7 T (tetrahedron taken as a unit, or as if represented by just its central atom)
in the structure, and it is obtained by counting the number of neighboring T that each such
unique T is bonded to in increasing shells. For example, each T will link to N1 = 4 neighboring
T in the first shell, each of which will link to a maximum of 3 T in the second shell (N2 ≤ 4×3 =

7
Atoms generated from a given site by space-group symmetry are crystallographically equivalent. Only crystallo-
graphically unique sites need to be specified when describing a crystal structure or considered when generating the
coordination sequence.
584 Zeolites and Other Porous Materials

4 5–1 6 8

4–4 6–2 5–3 6–6 8–8

Figure 14.4 Secondary building units found in some of the zeolite structures discussed.

12), etc. Coordination sequences are listed up to N10. The vertex symbol was introduced in
Section 1.4.38 and gives the size of the smallest ring associated with each of the six tetrahedral
angles of a T site, with rings from the three pairs of opposite angles placed adjacent to each
other in the list. If more than one ring of a given size is found, the number of such rings is added
as a subscript (e.g. 62 for two 6-rings at an angle). For zeolite A (of the three-letter code LTA
for “Linde Type A”), the coordination sequence (CS) and vertex symbol (VS) of the single T
are: CS 4 9 17 28 42 60 81 105 132 162 and VS 4·6·4·6·4·8, and we can see that the largest ring
size is 8, consistent with Figure 14.3.
Computer algorithms have been developed to investigate possible zeolite structure types,
and many thousands of hypothetical framework structures have been predicted [6, 7].
Some of these have subsequently been found experimentally. One such example, whose
synthesis adopted many of the lessons learned from years of zeolite research, is the large-
pore ITQ-33 [8].9 The guiding principle used in its preparation was the observation that
large-pore structures generally contain a significant proportion of 3- and 4-rings, which are
favored by including Ge in the synthesis [9].10 ITQ-33 was isolated and identified using
automated high-throughput synthetic methods that allowed exploration of unusual syn-
thetic conditions. It contains 18T cavities (~12.3 Å across) interconnected by perpendicular
10T pores (~6.1 Å × ~4.3 Å). These large pores lead to an unusually low framework density
of 12.3 T atoms per 1000 A3, and a density of only 0.37 g/cm3. The material is more
thermodynamically stable than other large-pore materials, and the presence of Al in the
framework leads to catalytic properties (Section 14.1.3).

8
We also discuss rings and coordination sequences in the context of amorphous materials in Chapter 15.
9
Formula: (hexamethonium)0.07F0.07(H2O)0.37[Si0.66Al0.04Ge0.30O2.02].
10
The same synthetic tactic has also led to ITQ-40 that has the lowest framework density of any zeolite (10.1 T atoms
per nm3) and a framework composition of Ge32Si43.6O150(OH)4. The presence of OH groups means that not all T
sites are 4-connected. Calculations show that Ge is stabilized in structures with small T–O–T angles [9].
 14.1 Zeolites 585

Box 14.1 Synthetic Methods: Zeolites and hydrothermal synthesis

Modern synthetic routes to zeolites stem from the works of Barrer, Milton, and Breck in the
1940s and 1950s. For example, Barrer investigated the synthesis of mordenite by taking a
“batch composition” of Na2O:Al2O3:8.2–12.3SiO2 and heating to ~290 °C in the presence of
water [10]. Milton and Breck, working at Union Carbide, realized that adding hydroxide
anions to reaction mixtures would greatly increase the solubility of Al via formation of [Al
(OH)4]−, and were consequently able to lower reaction temperatures to ~100 °C. They success-
fully prepared many previously known natural zeolites as well as zeolite A, a material with no
natural counterpart. The addition of organic cations as templates (see main text) to these
hydrothermal preparations later led to the production of many other new zeolites.
In a typical modern zeolite hydrothermal synthesis, a silica source and an alumina source are
combined with water, template molecules, and guest cations at high pH to produce an
inhomogeneous amorphous gel (see Section 14.3 for definitions). Typical alumina sources
include NaAlO2, Al(OH)3, and Al2(SO4)3; typical silica sources are soluble silicates such as
Na2SiO3⋅nH2O, aqueous silica sols (typically 30% by weight SiO2), and other forms of amorph-
ous silica. The addition of F− anions as mineralizers allows syntheses to proceed at lower pH,
helps incorporation of heteroatoms, and can promote the growth of large single crystals. The
gel is aged and then heated until the amorphous material transforms to a crystalline phase. This
is often performed above 100 °C, under the autogenous pressure generated in a Teflon-lined
steel autoclave (colloquially known as a “bomb”). Common lab-scale autoclaves can be used
up to 250°C and 1800 psi (~12.5 MPa), and are designed to be “fail safe” in the event of
unexpected pressure build-up.
The transformation from gel to final crystalline zeolite often occurs over an extended time
period, during which different products may evolve. Frequently, Ostwald’s rule of successive
transformations applies, with initial tiny metastable particles evolving into more stable ones.
High-density zeolites such as analcime (FD = 18.6 T/nm3) and sodalite (FD = 17.2 T/nm3) are
therefore typically observed later in reactions.
There has been considerable debate over the precise roles the initial gel, species in solution,
free cations, and template molecules play in the synthesis. Does the amorphous gel transform
directly in the solid state to form the zeolite? Does it merely act as a nutrient source to produce
soluble species from which the zeolite grows? The commercial importance of zeolites means
that huge efforts have been made to understand the process [11]. The best picture is probably as
summarized in Figure B14.1.1. The initial amorphous gel (a) will be in equilibrium with solutes
such as [Al(OH)4]−, various oligomeric silicates, template molecules, and other cations (b). As
species rearrange in solution and reprecipitate by making and breaking T–O–T bonds, guest
cations or templates will energetically favor the formation of certain rings (c), increasing their
likelihood and leading to areas of local order (d). Ordered nuclei of a critical size will then
progress to form a fully ordered crystalline material (e). As apparent from this figure, trad-
itional distinctions between solid- and solution-phase syntheses are probably not applicable. In
fact, even apparently clear synthesis liquids that might imply a solution mechanism can
586 Zeolites and Other Porous Materials

Box 14.1 (cont.)

actually be optically transparent colloidal suspensions. Crystallization within or on such
colloidal particles has been observed experimentally.

(b)
(c)

amorphous
crystalline

(a) (d) (e)

Figure B14.1.1 Schematic of processes involved in the synthesis of zeolites (after [11]).

14.1.2 Roles of Template Molecules in Zeolite Synthesis

One factor that led to an explosion in the number of synthetic zeolites was the inclusion of
organic cations during synthesis (see Box 14.1). These cations play two important roles: they
control the Si:Al ratio and they influence the specific framework topology formed. An
illustration of how the Si:Al ratio can be controlled is given by the synthesis of compounds
with the sodalite structure. In the sodalite structure, each of the 24 tetrahedra making up
a sodalite cage is shared by four cages (when we remember that the void in the center of
Figure 1.57 is itself a sodalite cage). There are therefore six T sites per cage. When sodalite is
prepared in the presence of NaOH, the size of each cage is such (Figure 14.1) that it can adopt
four Na+ cations and an OH− anion. These will charge-balance three Al3+ and three Si4+
giving Na8Al6Si6O24(OH)2·nH2O [12] with an Si:Al ratio of 1:1 and n = 3–4.11 If the synthesis
is performed in the presence of N(CH3)4OH [13], the larger size of the [N(CH3)4]+ cation
means that there’s only space for one cation per cage and the Si:Al ratio is increased to 5:1
([N(CH3)4]2Al2Si10O24). The final zeolite contains [N(CH3)4]+ trapped inside a cage that it
can’t diffuse into or out of. This suggests a templating role where the cage crystallizes around
the [N(CH3)4]+ cation during synthesis.

11
Similarly, Figure 1.57 shows that in the cubic mineral sodalite there are three Na+ cations to charge-balance the
framework along with an additional one Na+ charge-compensated by one Cl−.
 14.1 Zeolites 587

The use of different organic templates can allow the Si:Al ratio to be varied continuously
from ∞:1 down to 1:1 in some zeolites. At the high limit, the zeolite would essentially be a
polymorph of crystalline SiO2. The lower Si:Al limit of 1:1 is rationalized by Loewenstein’s
rule, which states that Al–O–Al linkages are not stable in the tetrahedral framework of
zeolites.12 Since each Al site must therefore be surrounded by four Si (each of which links on
to three more T sites that may contain Al), one cannot have a Si:Al ratio lower than 1:1. This
rule does not hold for less open networks of aluminosilicates produced at higher temperat-
ures, which can have more Al than Si.
The Si:Al ratio has a significant impact on the properties and applications of zeolites.
Firstly, when the Si:Al ratio increases, the internal surface changes from hydrophilic to
hydrophobic as the amount of charge-compensating cations decreases; high-silica zeolites
will therefore sorb nonpolar molecules. Secondly, if an Al-containing zeolite charge-
compensated by NH4+ or [N(CH3)4]+, or another protonated organic cation is calcined
(heated to remove or burn the volatile template) in a final synthetic step, protons will be left
behind to maintain charge balance. By increasing the Si:Al ratio, one reduces the number
of such acid sites.
The exact role of organic cations in controlling the framework topology is more contro-
versial, and the term structure-directing agent is perhaps more realistic than template. For
example, zeolite ZSM-513 was first prepared in the presence of tetrapropylammonium
cations, and the size and shape of this molecular ion was thought to be crucial for
producing that specific framework. However, the same ZSM-5 has since been prepared
in the presence of over 20 different organic species, in the presence of purely inorganic
cations and, in 1997, it was found in nature as the rare mineral mutinaite. In fact, the
organic cation plays multiple roles: It may simply be a “space filler”, with non-specific van
der Waals interactions helping stabilize the growing zeolite. It may work as a structure-
directing agent that increases the likelihood of a certain framework topology under specific
conditions. It can be a true template whereby specific electronic and steric properties
uniquely direct formation of the desired framework. Perhaps the best example of pure
templating is the use of “tri-quat”, [C18H36N3]3+, in the synthesis of ZSM-18. The size- and
shape-match between the guest and framework suggests templating (Figure 14.5), though
subsequent theoretical studies have successfully predicted other molecules that also lead to
ZSM-18 formation [14]. It has also been possible to design template molecules of specific
chirality to produce enantiomerically enriched bulk zeolite samples [15]. In addition to
using organic molecules to favor the formation of certain frameworks, the alternate
approach of blocking formation of an undesired framework is also possible. One uses
molecules that either wouldn’t fit inside common frameworks, or that would bind to
surfaces of unwanted frameworks inhibiting their growth.

12
Al on a Si site carries a formal negative charge (AlSi 0 ) that repels other Al sites.
13
Zeolite Socony-Mobil (Socony = Standard Oil Company of New York) after the research labs where it was
developed.
588 Zeolites and Other Porous Materials

Figure 14.5 The structure of

the MEI framework of
NMe2
ZSM-18 and the tri-quat
molecule first used in its
Me2N
synthesis.

NMe2

14.1.3 Zeolites in Catalysis

One extremely important application of zeolites is in heterogeneous catalysis, where they
underpin multi-billion dollar sectors of the chemical industry. Many of the most important
applications rely on their properties as Brønsted or Lewis solid acids.14 We’ve discussed
above that zeolites contain cations in their pores to balance framework charge. These cations
can be exchanged for protons either by stirring in dilute acid (for high-silica zeolites such as
ZSM-5), by heating an ammonium-exchanged form of the zeolite, or via calcination of
organic-template-containing zeolites. Typical reactions for these procceses are:
exchange
1. Naþ ½zeolite þ H3 Oþ → H3 Oþ ½zeolite →
heat
Hþ ½zeolite þ H2 O
Naþ
exchange
2. Naþ ½zeolite þ NH4 þ → NH4 þ ½zeolite →
heat
Hþ ½zeolite þ NH3
Naþ

3. ½templateþ ½zeolite →
heat
Hþ ½zeolite þ CO2 þ H2 O þ other products

The protons are present in the zeolite as Brønsted-acid sites with the H atom covalently
bonded to a framework O adjacent to an Al site.
The true structures of commercially useful catalysts are often significantly more complex
than a simple crystallographic picture would suggest. For example, a HY zeolite15 with a
Si:Al ratio of ~2.5 can be prepared by reaction 2 above, but is unfortunately unstable under
the conditions needed in real catalytic applications.16 Catalysts with a higher Si:Al ratio
would be more stable, but are more expensive to prepare directly. It is, however, possible to
turn the ammonium form of Y into so-called “ultrastable” HY (USY) by contacting it with
steam at ~750 °C for around 24 hours. In this process, Al is lost from the framework, leaving

14
IUPAC defines a Brønsted acid as a molecular entity capable of donating a proton to a base, whereas a Lewis acid is a
molecular entity that is an electron-pair acceptor and able to react with a Lewis base (an electron-pair donor). The base in
the specific case of zeolites is a sorbed molecule. Extra-framework Al and rare-earth cations (both discussed below) are
believed to act as Lewis-acid sites in zeolites. The term solid acid refers to it remaining undissolved, solid, in the gas or
liquid reaction medium.
15
Here and later, the nomenclature term HY implies the H+ form of zeolite Y, similarly NaA implies the Na+ form of
zeolite A.
16
Particularly under the high-temperature conditions described later to regenerate catalysts.
 14.1 Zeolites 589

defects behind and initially producing so-called extra-framework aluminum in the zeolite
pores and cages.17 Parts of the zeolite framework collapse and become amorphous as more Al is
lost, but local recrystallization allows Si from these regions to “heal” nearby framework defects,
effectively refilling them with Si. A portion of the Al migrates to the surface of the crystallites
during the process, where amorphous alumina has been observed experimentally. Framework
healing and Al loss occur at comparable rates and result in large mesoporous cavities forming
inside the zeolite crystallites, which can make up 20–30% of the crystallite volume. The frame-
work Si:Al ratio increases during the process and the HY formed is more stable.18 The process
has two additional benefits: The large mesopores can speed diffusion through the zeolite, leading
to better catalytic activity, and the extra-framework aluminum species increase activity. Low-
coordinate Al species can act as Lewis acids, and cationic species have been calculated to lower
the activation energies of key steps of the catalyzed reactions. Overall, catalysts prepared in this
way can have activities two orders of magnitude higher than otherwise.
The commercial importance of zeolite acid catalysis means that there has been considerable
effort to try and quantify the acidity of the Brønsted-acid sites19 present and how acidity
influences reactivity. Acidity, referring to a general removal of the proton (no base), can be
determined computationally20 or inferred using infrared spectroscopy to measure framework
O–H stretching frequencies. Alternatively, thermal analysis21 or spectroscopic techniques can
provide insight using different probe-molecule bases [16]. Perhaps the most direct insight
comes from NMR studies using probe molecules [17]. These show, for example, that a
molecule that is basic in an aqueous environment, such as p-fluoroaniline (C6H4FNH2), is
protonated in the pores of HZSM-5 and HY, whereas a less basic one such as p-fluoronitro-
benzene (C6H4FNO2) is not. The picture that emerges in this specific case is of a degree of base
protonation comparable to what would be found in 70% H2SO4.22
In a given framework, acidity usually increases as the Si:Al ratio increases up to about 6–
10. This can be related to the increasing number of Si atoms in the framework sites increasing

17
A variety of different extra-framework Al species can be present in zeolites depending on the specific conditions
(varying as a function of temperature, water content, Al content, etc.). 27Al NMR suggests the presence of four-,
five-, and six-coordinate species and theoretical studies show that Al3+ can coordinate to O2− (framework and non-
framework), OH−, and H2O depending on conditions. Single-Al species are mobile within the framework, and
clusters containing a few Al3+ may form within the zeolite pores. As mesopores develop or Al migrates towards the
surface, larger clusters can evolve towards nanoscale hydrated alumina.
18
The increasing Si content of the framework is supported by 29Si NMR measurements and a decrease in unit-cell
parameter. A portion of the extra-framework Al can be removed from the material by acid washing or ion exchange.
19
Many catalysts contain rare-earth ions as additional Lewis-acid sites.
20
By calculating the energy required to remove a proton to an infinite distance from the resulting framework anion.
21
For example, a material containing sorbed amines can be gently warmed to remove weakly bound molecules and then
heated to higher temperatures to remove more strongly bound molecules. Acidity correlates with the temperature at
which amines that are protonated in the pores desorb. The number of accessible acid sites correlates with the number of
desorbing species.
22
This comparison leads to zeolites often being referred to as strong acids. Acidity is the ability to protonate a Lewis
base in a particular solvent or medium, and comparing strengths between different media is problematic. Note that
early literature often stated zeolite acidity as being in the so-called super-acid range. Super acids would, for example,
protonate benzene to form a benzenium ion, C6H7+. This is not observed in zeolites.
590 Zeolites and Other Porous Materials

+H+ + migration −H+

+

Me Me Me Me
H Me H Me H + H H H
+H+ H migration −H+
+ Me
H H H H H H H Me
H H H H

β-
+H+ scission −H+
+ +
+ +

Figure 14.6 Acid-catalyzed transformations of simple organic molecules. The top two processes exemplify
isomerization reactions and the lower process, cracking.

the framework electronegativity and hence its Brønsted acidity. After a plateau,23 the acidity
starts to fall as the number of protons present (to charge-balance Al) decreases. Whether a
specific H-zeolite can act as an effective acid will also depend on the chemical transformation
under consideration, as it will be influenced by how the protonated form of the molecule in
question is stabilized when coordinated by the zeolite framework. These factors are probably
the most important in determining the overall acidity of a specific zeolite.
Figure 14.6 shows examples of the acid-catalyzed transformations that can occur in zeolites.
These are thought to proceed via carbenium-ion-like species.24 For an alkene, protonation to
produce a carbenium ion can be easily imagined: H2C=CH2 + H+ → H2C+–CH3. For an
alkane, one has to envisage either protonation to produce a five-coordinate carbocation
followed by H2 loss, or a direct hydride abstraction, both leading to the carbenium ion; either
H3C–CH3 + H+ → H4C+–CH3 → H2C+–CH3 + H2 or H3C–CH3 → H2C+–CH3 + H−.25 Once
formed, a carbenium ion can undergo a variety of processes, such as hydride or alkyl migration
leading to isomerization, or β-scission leading to cracking (see Figure 14.6). Such a rearranged
carbenium (cat)ion can then either deprotonate and desorb as an alkene or abstract hydride
from another hydrocarbon and desorb as an alkane.26
23
The major influence on acidity is the number of second-nearest-neighbor sites occupied by Al for the AlOH site in
question (second neighbor, as Loewenstein’s rule dictates that first neighbors are all Si). For high Si:Al ratios there is
a high probability that these sites will all contain Si such that the local electronegativity will only change slightly as
more Si is added.
24
Carbenium ions have a carbon with three covalent bonds and a positive charge. In contrast, carbonium-ions contain C
with five bonds (e.g. CH5+ or C2H7+). Although protonated zeolites aren’t generally sufficiently strong acids to directly
protonate sorbed alkenes to form carbenium ions, the transition states involved in reactions are often thought to be
carbenium-like. Some particularly stable carbenium ions have been observed experimentally in zeolite pores.
25
More exotic non-classical carbonium ions involving two-electron, three-center C–H–C bridges formed by proton-
ation of C–C bonds have also been proposed.
26
The actual chemical processes occurring within zeolites can often be significantly more complex than the overall
chemical transformation might suggest. For example, there is good experimental evidence that the transformation
of methanol to short-chain alkenes is mediated by aromatic methyl benzenium cations and other cyclic species in the
 14.1 Zeolites 591

Table 14.2 Relative cracking rates of various hexanes using HZSM-5 at 340 °C.

n-Hexane 2-Methyl-pentane 3-Methyl-pentane 2,3-Dimethyl-butane 2,2-Dimethyl-butane

Relative rate 0.71 0.38 0.22 0.09 0.09

(a) reactant selectivity

CH3OH

(b) product selectivity

(c) transition-state selectivity

Figure 14.7 Shape-selective catalysis in zeolites.

The huge importance of these transformations to the chemical industry explains the enormous
interest in acid catalysts. Solid zeolites have the advantage that they can be readily separated
from products and recycled. They also have extremely high and easily accessible external and
internal surface areas. Most importantly, the dimensions of their pores and cages can control the
course of the catalyzed reaction, hence the pore size constrains the chemistry leading to a shape-
selective catalysis. In this way, zeolites are analogous to enzymes since the local geometry fine-
tunes the properties of the active center. Different pore and void sizes and shapes control
reactant-, product-, or transition-state selectivity, as illustrated schematically in Figure 14.7.
In reactant-size selectivity, Figure 14.7a, access to the catalytically active sites of the zeolite is
controlled by the size of the framework pores. An example of this is the observation of different
rates of catalytic cracking for different isomers of C6H14 by ZSM-5 as given in Table 14.2.
Straight-chain n-hexane, which diffuses rapidly into the pores, cracks far more readily than
branched isomers. Similarly, both n- and i-butanol are readily dehydrated by large-pore NaX
zeolite at ~250 °C but only n-butanol by smaller-pore CaA zeolite. The size of the pores can be
modified in a number of ways to influence reactant selectivity. In zeolite A, for example, alkali
metals take up sufficient pore space so that only methane can penetrate inside. With divalent
pore cations, only half the sites are occupied, and this allows straight-chain hydrocarbons to
diffuse into the zeolite but not branched hydrocarbons. In general, molecules up to ~0.5 Å larger
than the apparent pore size can diffuse through zeolites. This has been explained in terms of local
vibrations increasing pore diameter, via the rigid-unit mode ideas discussed in Box 14.2.

framework via a so-called carbon-pool mechanism. Zeolite-bound alkoxide groups are long-lived intermediates in
several processes. More details are given in refs. [16] and [17].
592 Zeolites and Other Porous Materials

-6
H2 H2

log(diffusion coefficient) (m2/s)

C C
H3C CH2 CH3
-8 H2 H2
C C CH3 CH3
H3C C C
H2 H2
-10 1,4 para
CH3 H3C CH3
H3C CH3
-12 H3C
C
H2

CH3
-14
CH3
CH3

-16 1,2 meta

H3C CH3
-18
2 4 6 8 10
critical diameter (Å)

Figure 14.8 Diffusion coefficients of simple hydrocarbons in ZSM-5. Data on aromatics at 315 °C; aliphatics
500 °C (after [19]).

Product-size selectivity relies on the fact that different products of a reaction can have
vastly different diffusion rates out of the zeolite (Figure 14.8). A wonderful example of this is
provided by the disproportionation reaction of methylbenzene in ZSM-5 to produce benzene
and dimethylbenzene; the 1,4-dimethylbenzene isomer is a valuable feedstock for the pro-
duction of nylon [18]. Using unmodified or small crystallites of ZSM-5, one obtains an
equilibrium mixture of isomers containing ~24% of the desired 1,4 (para) product. However,
the diffusion coefficient of the para isomer is ~104 times higher than for the other isomers
(Figure 14.8), such that increasing the average time spent in the catalyst should increase the
proportion of the para isomer produced. This can be achieved by either blocking a fraction of
zeolite pores with coke, or by increasing the crystallite size of the catalyst. At low conversion,
0.7 μm crystallites give ~60% selectivity for 1,4-dimethylbenzene while 3.6 μm crystallites
give 80%; 97% selectivity can be achieved in coated or impregnated ZSM-5. In-situ solid state
NMR studies on related reactions have supported this mechanism by showing that the
isomer distribution within the pores of the zeolite is much closer to that expected from
thermodynamic equilibrium than that observed in the extracted product mixture.
The final category of shape selectivity is transition-state selectivity, Figure 14.7c. In the
example shown, the transfer of a methyl group from one dimethylbenzene molecule to
another can produce two isomers of trimethylbenzene. The transition state required to
produce 1,3,5-trimethylbenzene is sterically disfavored in the pores of mordenite, resulting
in preferential production of 1,2,4-trimethylbenzene. Similarly, in the acid-catalyzed iso-
merization of dimethylbenzenes, no trimethylbenzene forms. This is presumably due to pore
size disfavoring the bimolecular transition state needed to form the latter product. Another
area in which transition-state selectivity is thought to be important is in preventing the build-
 14.1 Zeolites 593

up of coke in zeolite catalysts. For large-pore zeolites such as mordenite, coking can block
pores, leading to a decrease in catalytic activity. In zeolites such as ZSM-5, the transition
states leading to coke formation are believed to be disfavored by pore size, such that coke
only forms on the zeolite surface. Transition-state selectivity can be distinguished from
reactant and product selectivity in that there should be little dependence on crystallite size,
aside from the indirect influence of surface sites. All the forms of selectivity we have discussed
can be optimized by treatments (e.g. dealumination) to remove or block surface sites.
The largest catalytic application of zeolites in catalysis is in the cracking of crude oil to produce
more useful fractions [20]. Over 99% of the world’s petrol is produced in this way using >300000
tonnes of zeolites annually. In a typical system, preheated crude oil (~370 °C) is passed onto a
fluidized bed of catalyst (fluid catalytic cracking, FCC) at ~500 °C, where cracking occurs. In a
subsequent reactor, the catalyst is then treated with steam to remove sorbed products before
flowing to a regeneration vessel, where it is heated in air at up to ~730 °C to remove coke before it
returns to the catalyst bed. Most cracking is carried out using FAU zeolites (Figure 14.2c).
Zeolites are also widely used in hydrocracking, in which low-grade oils react with hydrogen over a
fixed catalyst bed at high temperature and pressure to produce higher-grade oils. Dual-function
catalysts with both acidity and the capability to catalyze hydrogenation and dehydrogenation
reactions are needed. This is particularly important to help promote the formation of carbenium-
ion-like species when processing alkanes. These latter properties can be achieved by producing
metal dispersions both on the surface and in the cages of the zeolites.27 In all these applications, it
is often the ease of catalyst regeneration and cost rather than its intrinsic properties that lead to
commercial use. Large-pore materials, such as those in Section 14.1.6 and 14.2, are often unstable
under regeneration conditions, and the exotic templates required to prepare some frameworks
may make them too expensive for large-scale use.
Other catalytic applications involve modifying the chemistry of the framework to intro-
duce active species. One example is the catalytic oxidation of simple organic species by H2O2
on Ti-containing zeolites such as TS-1 of the same structure as ZSM-5. This process works
without toxic reagents and produces water as a byproduct. A fuller coverage of catalytic uses
of zeolites both on industrial and small scales can be found in Further Reading.

14.1.4 Ion-Exchange Properties

Another extremely important aspect of zeolite chemistry is that the charge-balancing cations
inside their frameworks can often be swapped for other cations in aqueous solution:

bAaþ ðaqÞ þ aBbþ ½zeolite ↔ bAaþ ½zeolite þ aBbþ ðaqÞ

The preference for forming a given zeolite–cation combination is determined by a number of

factors: the number and size of sites available to coordinate cations (note that the type of site
27
For example by ion exchange of the compensating cation with Ni2+ and reduction, or with [(Pd,Pt)(NH3)4]2+
followed by decomposition and reduction. Protons charge-balance the final catalyst.
594 Zeolites and Other Porous Materials

may change during the process as some sites may exchange preferentially); the cation charge,
radius, and electronegativity; and the ability of water molecules to hydrate the cation and
form hydrogen-bonding networks. It is also important to consider the entropy changes in
both the zeolite and solution that occur on exchange.28 It is common to report selectivity29
via so-called ion-exchange isotherms that compare the amount of a given cation in the zeolite
with the amount in solution at equilibrium. For zeolite A, for example, selectivities of alkali
metals are Na+ > K+ > Rb+ > Li+ > Cs+; alkaline earths are preferred over Na+ with Ca2+ ≈
Sr2+ ≈ Ba2+ > Mg2+. The higher Si:Al ratio of ZSM-5 increases the affinity for larger weakly
hydrated cations: Cs+ > H+ ≈ NH4+ ≈ Rb+ > K+ > Na+ > Li+. Zeolite ion exchange will
happen even with distilled water via NaX þ H2 O ↔ HX þ NaOH. This means that a zeolite
such as NaX in contact with water gives an alkaline solution. Other zeolites undergo more
complex slow hydrolysis reactions.
The maximum ion-exchange capacity of a given zeolite is determined by the Si:Al ratio
that dictates the number of charge-balancing cations. In many cases, full ion exchange isn’t
possible. For example, Cs+ ions are too large to fully replace all Na+ ions in the Na+ form of
zeolite A (NaA); Rb+ and Cs+ are too large to ion-exchange with smaller cations in the
β cages of zeolite X. Zeolite ion exchangers can therefore show size selectivity similar to the
catalytic reactant-size selectivity discussed above.
The microporosity of zeolites also affects the kinetics of ion exchange. For example, while
Mg2+ and Ca2+ will both undergo full ion exchange with NaA, the rate of exchange at room
temperature is around 10 times slower for Mg2+. This is believed to be due to the stronger
hydration shell of Mg2+ that must be removed before the cation can enter the zeolite pores. At
higher temperature, rates become comparable. The huge importance of ion exchange to industry
has led to a large body of research in this area and more details can be found elsewhere [21, 22].
One major application area is in solid laundry detergents, where NaA (among other effects)
softens water by removing Ca2+ and Mg2+. Zeolites have the added benefit of reducing powder
agglomeration during storage, allowing compact formulations. Up to 15% zeolite A might be
present in a typical detergent.30 The zeolite synthesis is carefully controlled to produce
crystallites of 3–5 μm with beveled edges [23] so that they rinse from clothes more readily.
28
Possible entropic contributions include: changes in the number of cations in the zeolite or solution for a ≠ b,
changing numbers of water in the framework for different cations, changes in the water coordination shell of cations
in solution, etc.
29
One way of depicting selectivity is through an ion-exchange isotherm that plots a quantity ZA representing the
equilibrium amount of cation A in the zeolite against SA representing the amount in solution:
amAZ aMAs
ZA ¼ and SA ¼ with mAZ the molality of Aa+ in the zeolite in mol/kg and MAs the
amAZ þ bmBZ aMAs þ bMBs
molarity of Aa+ in solution in mol/L. A straight-line plot with gradient 1 shows no preference for cation A over B. A
high initial slope shows a selectivity for A over B and a lower slope, the opposite. More complex shapes can reflect
exchange occurring at different cation sites.
30
In the detergent industry, the species used to control Ca2+ and Mg2+ levels are traditionally called builders. Various
species called chelates or co-builders (e.g. citrate) are added to control other metal-ion concentrations. A typical
solid formulation will also include sodium carbonate and silicates to produce the alkaline conditions needed for the
wash, surfactant molecules, dispersing agents (e.g. polycarboxylates) to avoid reprecipitation of soil particles, as
well as bleaches, perfumes, and optical brighteners.
 14.1 Zeolites 595

This usage declined somewhat as the industry moved to liquid formulations and as surfactant
systems were improved, but zeolites are still present in over 30% of laundry products (around
70000 tonnes of zeolite are used annually in a ~$40 billion market).
Zeolites are also used in water purification. Natural clinoptilolite31 is used to remove
NH4+ from waste water at sewage plants, and radioactive ions such as 137Cs and 90Sr after
nuclear spills [21, 22]. Clinoptilolite was even added to the feedstock of sheep that had grazed
on 137Cs-contaminated land following the 1986 explosion at the Chernobyl nuclear power
plant! Zeolites also provide a potential method for the medium-to-long-term storage of
radioactive waste from the nuclear industry.

14.1.5 Drying Agents, Molecular Sieving, and Sorption

All chemical laboratories will have a bottle of 3A, 4A, or 5A zeolite beads that are used as
general-purpose drying agents. The zeolites are heated in a vacuum to drive off water that
coordinates the extra-framework cations, after which they have a high affinity for water
uptake. The number in the symbol corresponds to the nominal size (in Å) of molecules the
zeolite will absorb. Zeolite 4A is the dehydrated Na form of zeolite A, Na12[(AlO2)12(SiO2)12].
Zeolite 3A is the K form, and the larger K+ ions reduce the accessible pore size. Zeolite 5A is
(Ca4Na4)[(AlO2)12(SiO2)12]. In 5A, the cations only adopt site 1 in the structure of Figure 14.3,
effectively leaving the 8-rings “open” and creating a larger accessible pore. Zeolites 4A and 5A
can be used to remove water from most alcohols, except methanol for which 3A is used because
methanol itself would enter the larger pores of 4A and 5A. Zeolite 3A is widely used for drying
natural gas, whereas 4A might be used if there is a need to also remove small amounts of the
larger CO2 molecule. Zeolite 5A is commonly used in the separation of straight-chain and
branched hydrocarbons. For example, there are several industrial processes that isolate linear
C10 to C18 hydrocarbons from gas streams containing branched and cyclic species by their
sorption in the pores of 5A.
The separation of O2 from air is another important application of zeolites. Historically,
oxygen was produced by fractional distillation of liquid air, relying on the different boiling
points of N2 and O2 (77 and 90 K, respectively). However, the need for complex liquefaction
apparatus can be avoided by using zeolites such as Li-chabazite or a (Li+,M2+)X zeolite,
which both operate at room temperature [24]. Since neither N2 or O2 have a permanent
dipole moment, separation relies on the fact that N2 has both a larger quadrupole moment
and higher polarizability32 than O2 [25], leading to stronger electrostatic interaction with

31
Natural clinoptilolite (framework code HEU) has typical composition A6Al6Si30O72·~20H2O, A = (K, Na, Ca0.5,
Sr0.5, Ba0.5, Mg0.5) with higher Na + K content than alkaline earths.
32
The simplest electric dipole is a vector of a + − pair of point charges. The simplest quadrupole moment can be
thought of as alternating point charges at the corners of a square (+ − + − clockwise round the square).
Homonuclear diatomics have no permanent dipole but, since they are not spherically symmetric, have a quadrupole

moment þ þ. Quadrupole moments are −1.115 and −0.225 atomic units (equal to ea02, where e is the elementary

charge and a0 = 0.529 Å is the radius of the first Bohr orbit) and polarizabilities are 11.74 and 10.61 in units of a03 for
N2 and O2, respectively [25].
596 Zeolites and Other Porous Materials

extra-framework cations. Under typical conditions this leads to a ~4–10-fold higher uptake
of N2 than O2. Oxygen can thus be produced by a vacuum-swing-adsorption method
whereby gases are loaded at ~1.2 bar and preferential N2 adsorption leaves behind an O2-
rich gas. When the adsorbent is saturated, a vacuum is applied (~0.35 bar) and the zeolite
sorbent regenerated. Multiple sorbent beds can be used to give a constant flow of O2.
While the processes above all work via an absorb–desorb cycle, there is also considerable
interest in producing extended membranes of zeolites to enable continuous sieve-like separ-
ation; incorporating catalytic activity in the membranes makes them more exciting still. It has,
for example, been possible to produce extremely thin highly oriented (so pores are aligned)
membranes of MFI zeolites that allow high fluxes of molecules and can separate pairs of
species such as 1,4-dimethylbenzene/1,2-dimethylbenzene or benzene/cyclohexane [26].

14.1.6 AlPOs and Related Materials

Given the structural chemistry of zeolites, it is not surprising that similar materials can be
prepared containing other tetrahedral metal cations. For example, starting from SiO2, site
ordering with AlIII and PV (the idea shown in Figure 1.38) leads to dense aluminum
phosphates, isostructural to quartz, tridymite, or cristobalite polymorphs of SiO2. By
heating mixtures of Al2O3:P2O5:amine-template:40–400·H2O under hydrothermal condi-
tions, porous aluminophosphates (AlPOs) similar to zeolites have been prepared. It is
worth noting that the site ordering of Al and P sites in such AlPO frameworks constrains
them to those containing even-numbered rings. The main difference to zeolite synthesis is
that reactions are typically performed at low pH where it is easier to include elements in
addition to Al and P in the framework during synthesis.33 A wide variety of other tetrahedral
MO4/2 groups (commonly M = Ga, Ge, As, B, Be, Li, Co, Zn) can be introduced.
One interesting feature of AlPOs is that materials with large pores (>12 T) can be
synthesized (see Table 14.1 and Figure 14.9). VPI-5, for example, can be synthesized
hydrothermally at 150 °C from orthophosphoric acid and hydrated alumina in the presence
of tetrabutylammonium hydroxide. It contains 12 Å-wide channels and has a 25% accessible
volume. The gallophosphate cloverite (C7H14N)24[F24Ga96P96O372(OH)24] (Figure 14.9,
right) contains 20-membered rings and has four terminal OH groups that protrude into
the pores (reducing their size slightly), giving the characteristic clover-leaf pore shape after
which the material was named. In cloverite, the F is located in 4–4-rings and increases the
coordination of each Ga from four to five (GaO4F).
Large-pore AlPOs differ from the zeolites in four main ways: They may contain non-
tetrahedral metal atoms (e.g. Al in octahedral coordination). They may contain terminal OH
groups on metals. They may contain non-tetrahedral species (e.g. OH, H2O, F) as part of the
framework. Finally, they tend to be less stable than zeolites under thermal or hydrothermal
conditions, which limits their potential applications.

33
Under the high-pH conditions of zeolite synthesis, many transition elements are insoluble.
 14.2 Mesoporous Aluminosilicates 597

VPI-5 cloverite

Figure 14.9 Schematic views of the structures of VPI-5 and cloverite. In cloverite GaO4F polyhedra shaded
in light gray, PO4 in dark gray.

14.2 Mesoporous Aluminosilicates

The crystalline zeolites and related frameworks we’ve discussed so far contain pores that
have regular shape, but are limited in size to ≲ 12 Å. Before the early 1990s, materials
containing mesopores (2–50 nm) were restricted to amorphous silicas (Section 14.3) or
modified layered materials in which a distribution of pore sizes are present. Since then, a
variety of synthetic methods have been developed for producing mesoporous ceramics with
uniform pore sizes up to several hundred ångstrøms.
The most famous of these are the MCM-4134 aluminosilicates first described in detail
by workers at Mobil [27]. They performed hydrothermal syntheses under basic condi-
tions, using various silica and alumina sources in the presence of long-chain alkylam-
monium cations such as hexadecyltrimethylammonium, [C16H33(CH3)3N]+.35 These
molecules are well-known cationic surfactants with charged hydrophilic headgroups
and neutral hydrophobic hydrocarbon tails. Above a critical concentration in solution
(the critical micelle concentration, CMC), they self-assemble into spherical or rod-like
micelles, more complex 3D surfactant networks, or lamellar structures. Under the
concentration, temperature, and pressure conditions used to prepare MCM-41, long
rod-like micelles are formed (Figure 14.10, top), which cluster into ordered arrays with
their headgroups coordinated by silicate anions of the initial mixture. The remaining
anions are protonated by water and slowly condense during heating onto these pre-
arranged anions, forming a solid polyacid aluminosilica network. This allows the base

34
MCM after Mobil Composition of Matter.
35
A typical preparation used 2 g of “Catapal alumina” [AlO(OH) and Al(OH)3 in ~1:1 mixture], 25 g of “HiSil” (a
precipitated form of SiO2), and 100 g of a tetramethylammonium silicate solution containing 10 wt% SiO2 along
with 200 g of a 29 wt% surfactant solution of [C16H33(CH3)3N]+[OH/Cl]− (overall Si:surfactant ratio of ≤1). The
mixture was heated to 150 °C in an autoclave for 48 hours followed by heating to 540 °C to remove the template.
598 Zeolites and Other Porous Materials

Figure 14.10 The formation of MCM-41. Surfactant molecules assemble into rod-like micelles around
which inorganic walls form. Transmission electron microscope (TEM) images (bottom center/right) of
typical samples looking along the pore direction. Dark regions are walls and lighter regions pores. Center
image is a sample prepared with [C12H25(CH3)3N]+. The material on the right has a pore size of 40 Å and
was prepared with [C16H33(CH3)3N]+. Images adapted from ref. [28] with permission.

present to dissolve a new portion of the amorphous silica into silicate. The process
continues until the final templated aluminosilicate particles precipitate. Calcination
then removes the template molecules leading to a porous material. This process is
shown schematically in Figure 14.10. Electron microscopy reveals the formation of a
regular array of ~40 Å pores running in one dimension, showing that the micelle
morphology is retained in the solid. The walls themselves are amorphous, typically
~10–15 Å thick and contain a variety of silica species (Q4, Q3, and Q2).36
The size of the pores can be controlled using surfactant templates with different alkyl-
chain lengths, or by adding other organic molecules that can dissolve in the hydrophobic
core of the micelles and increase their size during synthesis. For example, using a shorter-
chain surfactant like [C12H25(CH3)3N]+ in the synthesis gives ~30 Å pores in the final
product. Materials with pores up to ~100 Å can be prepared by adding 1,3,5-trimethylben-
zene to a [C16H33(CH3)3N]+ synthesis.
Similar materials can also be prepared using non-ionic surfactant molecules. For
example, high-stability large-pore structures can be prepared using block copolymers37
of poly(ethylene oxide) and poly(propylene oxide) of the form (EO)n(PO)m(EO)n. These
again contain more hydrophilic (EO) and more hydrophobic (PO) regions and form

36
The Qn symbol is used to represent a SiO4 tetrahedron linked to n other tetrahedra. In quartz, for example, all Si sites
are Q4 (Niggli formula SiO4/2). An otherwise linked tetrahedron with a single Si–OH group would be Q3. Different
Qn species can be identified and quantified by 29Si NMR.
37
Block copolymers contain regular sequences of different monomers and are known to undergo ordering on the
mesoscale due to the differing physical properties (e.g. hydrophobic or hydrophilic) of different blocks.
 14.2 Mesoporous Aluminosilicates 599

various micellar structures in a solution, with EO surfaces and PO cores. With these
surfactants, the mesoporous silicate synthesis is performed under acidic conditions
where the silicate and surfactant species will be protonated. Anions from the added
acid maintain charge balance in the initial products. Materials with pore sizes up to
~300 Å and thick (~30–65 Å) walls can be prepared using an (EO)20(PO)70(EO)20
surfactant with 1,3,5-trimethylbenzene added to swell the micelle size. One was quoted
as being stable up to 500 °C or in boiling water for 24 hours, had a surface area of
910 m2/g, an average pore size of 260 Å, and a pore volume of 2.2 cm3/g [29]. The
expensive polymer template can be removed by solvent extraction and reused in
subsequent syntheses. These materials often exhibit microporosity in the silica walls
in addition to the microporosity of the large channels, though with an irregular pore
distribution reflecting the amorphous nature of the walls.
By changing the surfactant or reaction conditions, it is also possible to make materials
with 3D pore structures, or with spherical pores arranged in hexagonal- or cubic-closest-
packed-like patterns. A variety of related porous metal oxides have also been produced,
including ZrO2, TiO2, Nb2O5, Ta2O5, and MnxOy examples.
Applications of mesoporous materials generally take advantage of their large-pore
structures. Gas-sorption studies show that they have a high available pore volume and
surface area, though they exhibit so-called type-IV38 isotherms with a capillary conden-
sation step such that the gaseous sorbate must be at relatively high pressure. While
MCM materials don’t show the high acidity of zeolites required for catalytic cracking,39
their uniform large-pore structure leads to exciting opportunities for supporting mol-
ecules usually used for homogeneous catalysis within a controlled and uniform environ-
ment. The resulting heterogeneous catalyst is more easily separable from reaction
mixtures for re-use, and the local pore environment can lead to shape-selective control
of products.
In one early demonstration example, a chiral40 Pd-containing catalytic molecule was
attached to the walls of MCM-41.41 The resulting catalyst was shown to catalyze the
following reaction (Ac is acetyl, CH3CO) of two alternative products:

*
+ PhCH2NH2 x + (1−x) + HOAc

38
See e.g. Barton et al. in Further Reading for a discussion of isotherms in porous materials.
39
Acidity appears to correlate with Al–O–T angles: the larger the angle, the higher the acidity. In the amorphous walls
of MCMs, there are no structural features that force large angles.
40
Chiral molecules have non-superimposable mirror images called enantiomers. They are discussed in Box 14.3. An
enantioselective synthesis is one that produces more of one enantiomer than the other; an enantiomeric excess (ee).
A chiral carbon in an organic molecule has four different groups attached and is marked with *.
41
A ligand derived from 1,1 0 -bis(diphenylphosphino)ferrocene (dppf) was first attached to the walls of MCM-41 by
the reaction of a chiral-amine side arm on one ring of the ferrocene with a 3-bromopropyltrichlorosilane molecule
grafted onto the MCM-41 surface. The catalyst formed on subsequent coordination with Pd2+.
600 Zeolites and Other Porous Materials

and to be more active than an equivalent soluble Pd-containing catalyst. The confined
geometry of the pore yields 50% of the branched isomer (x = 0.5), compared to just 2%
(x = 0.02) for a catalyst supported on the surface of non-porous silica [30]. More
importantly, the branched product had a 95% enantiomeric excess; much higher than
the 43% found with the unconfined catalyst. The confined catalyst therefore shows both
regioselectivity (the place where the double bond is attacked) and enantioselectivity over
other forms.
There are numerous other applications of MCM materials. For example, mesopor-
ous oxidation catalysts similar to the TS-1 zeolites can be prepared either by includ-
ing Ti(OC2H5)4 during synthesis or via a post-synthetic grafting of active
organometallic species such as Ti(η-C5H5)2Cl2 and heat treatment. Supported metal-
nanoparticle catalysts can also be prepared. More details are available in Further
Reading, and we will encounter many similar ideas when we discuss applications of
MOFs in Section 14.4.
It has also been possible to combine mesoporous (>2 nm) and microporous (<2 nm)
cavities in materials [31]. For example, when some high-silica zeolites (e.g. ZSM-5 or Y) are
treated with surfactants such as [C16H33(CH3)3N]+Br− in the presence of a weak base, they
appear to regrow around surfactant micelle species that become encapsulated in the zeolite
crystallite. The surfactants can subsequently be calcined off to give zeolite crystallites with
essentially unchanged shape but containing ~45–50 Å mesopores in addition to the normal
framework micropores. The exact mechanism by which this occurs isn’t fully understood,
but it probably involves local recrystallization of the zeolites via a process similar to USY
formation discussed in Section 14.1.3. These mesostructured zeolites have been prepared
on a multi-tonne scale and show high activity in oil-cracking applications. In particular,
the mesopores allow rapid diffusion of large molecules into the zeolite for cracking, and
rapid diffusion of small molecules out of the zeolite, reducing the “over-cracking” that
produces coke.

14.3 Other Porous Oxide Materials

There are a number of other ways to produce highly porous oxide materials with pore
sizes controllable over a range of length scales. One widely used synthetic approach is
the sol–gel method. A typical example would be the formation of amorphous porous
silicas either by acidifying an aqueous solution of sodium silicate, or by the controlled
hydrolysis of silicon alkoxides in a non-aqueous (typically alcohol) solvent, via the
following summary reactions:

SiO32− + 2H+ → SiO2 + H2O or

Si(OR)4 + 2H2O → SiO2 + 4ROH
 14.3 Other Porous Oxide Materials 601

In this process, soluble molecular species gradually condense to form a colloidal

solution of high-molecular-weight polysilicate particles (the sol). These particles
then link together to form a 3D network (the gel42) with interparticle pores filled by
solvent:

Si OH + HO Si Si O Si + H2O

On heating to remove solvent and any other organic species added during the preparation,
porous xerogels43 are formed. In contrast to the zeolites and mesoporous materials discussed
above, neither the pores nor constituent silica particles have long-range order. Other porous
metal oxides can be prepared using suitable metal-alkoxide precursors in place of the Si
source.
Xerogels can be either microporous or mesoporous depending on the specific syn-
thetic conditions, with the porosity governed by the sizes of particles formed during the
early stages of polymerization. Porosity control can also be achieved by using pre-
formed building blocks such as [Si8O12](OCH3)8 or [Ti16O16](OC2H5)32 molecular pre-
cursors. Sol–gel synthesis can also be used to prepare materials with macroscopic
porosity. In the preparation of so-called inverse opals, the interstices in a highly ordered
close-packed array of polymer spheres (typically ~1 μm diameter polystyrene) are filled
with a desired silica sol. Following gelation, the polystyrene spheres can be removed by
careful calcination or dissolved in solvent to produce silica with ordered interconnected
micron-sized spherical cavities.
Aerogels are closely related materials that typically have >50% free volume and can have
extremely low densities down to 1 mg/cm3. They are prepared by replacing the solvent in a
sol–gel synthesis by a supercritical gas such as CO2. Silica aerogels can have extremely low
thermal conductivities as their open structures inhibit both conduction and convection
modes of heat transport. They are used for thermal insulation, particularly in space applica-
tions due to their low weight.
Another conceptually simple but elegant approach to making porous solids is the use
of selective leaching of a two-phase sample, as shown schematically in Figure 14.11
[32]. Here, one forms an intimate composite of two phases, one of which can be
selectively removed by dissolving it in an appropriate solvent, either directly or after
suitable chemical transformation (e.g. selectively reducing one oxide to a metal fol-
lowed by acid leaching). The resulting porous material can itself be chemically altered
in several ways: an oxide could be reduced to form a porous metal; the surface could
be coated with other species; or an additional chemical reaction could render the walls
microporous, creating a hierarchy of pore sizes.

42 43
In a gel, the network of particles spans the entire volume of the container. Greek xeros means dry.
602 Zeolites and Other Porous Materials

(a) (b) (c) (d)

Figure 14.11 The production of porous materials by selective leaching. (a), (b) Schematics of the
approach. (c), (d) Scanning electron microscope (SEM) images of how leaching of ZnO from a dense NiO/
ZnO composite leads to a porous material. Reproduced from Chem. Commun. (2006), 3159–3165.

Box 14.2 Materials Spotlight: Fascinating flexible frameworks—shrinkage on

heating and expansion on squeezing
The 3D structures of framework materials can give rise to some fascinating counterintui-
tive physical properties, several of which are technologically exploitable. For example, we
expect that materials expand on heating. Linear coefficients of thermal expansion,
αℓ ¼ ðℓT2  ℓT1 Þ=ðℓT1 ½T2  T1 Þ, are typically around +2.5 for Si, +8 for Al2O3, and +17 for
Cu (all ×10−6 K−1) near room temperature. This expansion can ultimately be traced back to the
asymmetric nature of interatomic potentials (the shape is similar to the lattice-formation
energy curve in Figure 5.2). As a bond is heated, higher vibrational levels are populated,
which corresponds to larger mean atomic separations causing thermal expansion. Some
framework materials, however, show the opposite behavior and contract on heating. Perhaps
the most famous of these is cubic ZrW2O8, which shows isotropic contraction (negative thermal
expansion, NTE) from 0.3 K to 1050 K and has αℓ = −9×10−6 K−1 between 2 K and 300 K [33].
The origin of this effect can be related to its structure, which contains O in two-coordinate Zr–
O–W linkages between corner-sharing ZrO6 octahedra and WO4 tetrahedra (see Figure
B14.2.1). The metal–oxygen bonds themselves are relatively strong and show low thermal
expansion. At a local level, however, the O atom undergoes a transverse or sideways vibration
on heating. The larger the magnitude of this dynamic displacement, the closer Zr and W
polyhedral centers are pulled together, and the smaller the volume V.* This is called the tension
effect.
To fully understand thermal expansion in a solid, we have to consider phonons that affect the
whole structure; we cannot just focus on the local picture. Interestingly, the 3D network of
ZrW2O8 has a set of phonon modes that keep local coordination polyhedra (nearly) undis-
torted. These low-energy, low-frequency states are called rigid-unit modes (RUMs), and can
dominate thermal expansion. Many RUMs in ZrW2O8 involve transverse motions of O and
tend to contract the structure. Similar behavior is found in some zeolites, some cyanide
networks, and some MOFs [34].
 14.3 Other Porous Oxide Materials 603

Box 14.2 (cont.)

C
O N
ZrO6 Ag
M M
N
C
WO4
T

Co(CN)6

contract
ZrW 2O8 Ag3[Co(CN)6]

Figure B14.2.1 The structure of ZrW2O8 contains corner-sharing octahedra and tetrahedra. Ag3[Co(CN)6]
contains [Co(CN)6]3− octahedra linked by Ag+ into three interpenetrating pseudo-cubic networks (shown
with different bond shadings). Exemplar short (~3.5 Å) Ag–Ag interactions between these networks are
shown with double-headed arrows.

The framework material Ag3[Co(CN)6] has an exceptionally anisotropic thermal expansion

and contraction [35]. In its trigonal structure (P31m), large Ag+ ions link [Co(CN)6]3− octahedra
into three interpenetrating 3D networks and themselves form densest-packed planes parallel with
ab of the hexagonal cell. The geometry of the material at any temperature is governed by these
weak Ag–Ag interactions. On heating, Ag–Ag distances expand profoundly in the ab plane, and
the stronger network bonds force the structure to contract along c. The situation is a 3D
equivalent of stretching a piece of garden lattice fencing or a folding wine rack. Values of αℓ =
−120×10−6 K−1 along c are a truly colossal uniaxial NTE.
Coupled RUM rotations of rigid tetrahedra are also important in zeolites. Many zeolites have
“windows of flexibility”, which are cell-parameter ranges over which the structure can distort
without individual TO4 tetrahedra distorting [36]. Most zeolites with a pure SiO2 composition
adopt the highest cell volume that is possible without tetrahedral distortion. In some ways, this
goes against traditional ideas of solids trying to maximize their density, but makes sense if one
considers the driving force as maximizing the distance between O atoms of different tetrahedra to
minimize Coulomb repulsion. We can also consider RUM-like distortions in zeolites as a low-
energy mechanism whereby the framework structure can expand locally to allow the passage of
diffusing guest molecules, and for understanding their displacive phase transitions.
Another counterintuitive property of zeolites is that under special circumstances some can appear
to expand at the unit-cell level when put under pressure. The explanation for this apparent paradox is
that when the sample is squeezed in a water-containing pressure medium, the pores take up more
water allowing the overall volume to reduce under pressure despite the zeolite volume increasing in the
process. Natrolite, for example, transforms from Na16Al16Si24O80⋅16H2O to Na16Al16Si24O80⋅32H2O
under pressure and the super-hydrated form has a 4.5% larger volume than the normal phase [37, 38].
604 Zeolites and Other Porous Materials

Box 14.2 (cont.)

Remarkable changes in cell volume on the uptake or loss of solvents and gases are also found in some
MOFs, with materials such as MIL-88 showing a volume change of 100% on CO2 uptake [39, 40, 41].
Some of these frameworks initially contract on gas uptake then expand at higher pressures as more gas
is adsorbed.

pull pull

natrolite transverse expansion

Figure B14.2.2 The structure of natrolite and a simple mechanism by which corner-hinged squares can give a
negative Poisson’s ratio by expanding in a transverse direction when pulled. Similar coupled rotations can occur
on heating, giving rise to RUMs and NTE.

Finally, let’s consider the Poisson’s ratio of a material. This measures how much a material
contracts in a direction perpendicular (transverse) to the direction in which it is pulled by a
mechanical force (longitudinal) and is defined as PR ¼ εtransverse =εlongitudinal , where ε is a strain
(deformation expressed as a fractional length increase). Most materials such as rubber contract in
the transverse direction when stretched and therefore have a positive Poisson’s ratio. Cork is an
interesting material in that its Poisson’s ratio is very close to zero. This has huge practical
importance. It means a cork can be pushed into (and, more importantly, later pulled out of) a
wine bottle’s neck to produce a tight seal; a bulging rubber bung couldn’t be! There are rare
examples of materials, auxetic materials, which have a negative Poisson’s ratio and become fatter
when stretched or thinner when compressed [42]. One simple geometric way of achieving this is
shown in Figure B14.2.2. If we have a 2D network of squares hinged at the corners, pulling the
system in one direction will cause the squares to rotate cooperatively and expand in the other
direction. The zeolite natrolite has been shown to display auxetic behavior [43]. Similar behavior
can be engineered in macroscale objects, has been shown in foams processed to contain re-entrant
pores [44], and leads to a number of exciting applications.

* These types of vibrations have a negative Grüneisen parameter, γ, which relates vibrational frequency ν to volume via
γ ¼ dðlnνÞ=dðlnVÞ. The volume thermal expansion of a material depends on γ via the relationship αV ¼ γCV K=Vmol (see, for
example, ref. [34]; CV is the specific heat per mole at constant volume, K is the isothermal compressibility, γ averaged over all
modes). Since CV and K are both positive, negative Grüneisen parameter modes lead to a negative αV and contraction.
 14.4 Metal–Organic Frameworks (MOFs) 605

14.4 Metal–Organic Frameworks (MOFs)

14.4.1 MOF Structures
While the microporous zeolites, mesoporous silicas, and porous oxides discussed above take
much of their inspiration from mineralogy and oxide chemistry, our final family of porous
materials, metal–organic frameworks (MOFs) (also known as coordination polymers or
hybrid porous materials), makes use of ideas from coordination chemistry. The key design
concept behind these materials is to take metal ions or small metal-containing clusters as
nodes and to link them with multidentate ligands to produce framework structures.
The simplest illustration of this idea is given by Prussian-blue-related materials (Section
7.4.2). If one mixes Cd2+ and [Pd(CN)6]2− ions in solution, a precipitate is formed, in which
the N atoms of six different [Pd(CN)6]2− octahedra coordinate Cd2+ to produce an infinite
network of linked octahedral centers (Figure 14.12).44 The simplest analogy is to the network
of corner-linked octahedra of ReO3, but with the single O linking atom replaced by the CN
group (a network expansion) and Re by Pd/Cd (site ordering); an alternative analogy is to the
net of B atoms in CaB6 (vertex decoration).45
Similarly, when Zn(CN)2 is precipitated from solution, the preference of Zn for tetrahedral
coordination leads to a structure containing a diamond-like network of linked Zn(CN)4/2
tetrahedra.46 The large distance between the tetrahedral Zn centers (Zn–CN–Zn is ~5.11 Å)
would, however, lead to a structure with too high a porosity to be stable. Zn(CN)2 overcomes

Pd
Zn

C
N

Cd

Figure 14.12 The structures of Cd[Pd(CN)6] (left) and Zn(CN)2 (right). C/N sites are disordered in Zn(CN)2.
One network is drawn with dashed bonds.

44
Prussian blue itself can be made by adding K4[FeII(CN)6] solution into an excess of Fe3+(aq) producing a blue
precipitate Fe4[Fe(CN)6]3 . xH2O with 14 ≤ x ≤ 16. Although the structure is based on a network of linked octahedral
Fe centers, there are significant defects on the FeII and CN sites that are occupied by water; FeIII is coordinated by
an average of 4.5 N atoms and 1.5 O atoms.
45
The structure of CaB6 is shown in Figure 1.40. A B62− octahedral anion decorates the Te2− sites of the CsCl-
structured CaTe, with short B–B distances to neighboring octahedra.
46
CN groups show CN/NC disorder.
606 Zeolites and Other Porous Materials

bdc bpdc tpdc btc bipy

square trigonal prism octahedron octahedron

Zn dimer Fe3O Zn4O Re6S8

Figure 14.13 Top: Linkers commonly used in the preparation of MOFs; abbreviations are given for those
discussed in text. Bottom: Nodes of specified geometrical connectivity with an increasing number of metal
atoms (linkers’ functional groups are drawn to indicate the connectivity).

this problem by forming an interpenetrated structure, in which one diamond network is

interleaved within another (analogous to the cuprite network in Figure 1.42). A simple way
to avoid interpenetration is to fill the pore space of the framework. An example of this is
[N(CH3)4]+[CuZn(CN)4]−, in which the tetramethylammonium cation that charge-balances
the framework takes up the space that the second penetrating framework would have adopted
[45]. For many MOFs, preparing them from very dilute solutions reduces the chances of
interpenetration.
One of the attractions of this approach to producing porous frameworks is the essentially
unlimited set of “building blocks” one can choose from. Figure 14.13 shows some common
multidentate linking ligands and a tiny selection of the huge array of metal nodes possible. In
addition, it is possible to complete the coordination environment of metals at nodes with
non-linking ligands. This leads to an enormous range of potential structures and the
possibility of designing materials with specific pore shapes and controllable chemical and
physical properties. Note that with MOFs the structure’s topology is controlled by the
connectivity enforced by the nodes employed, rather than by the size/shape of a template,
as in zeolite syntheses.
The natural language for describing MOF structures is that of nets [46]. In many
cases, it is found that structures adopt one of a relatively small number of default nets
that represent the simple high-symmetry ways of linking together nodes of different
 14.4 Metal–Organic Frameworks (MOFs) 607

connectivity. Many of these are familiar from the structures in Chapter 1 and the
discussion in Section 1.4.3. As shown in Table 1.6, for six-coordinate octahedral nodes
we expect the α-Po or NaCl net; for four-coordinate tetrahedral nodes we expect nets
related to diamond, quartz, cristobalite, or the zeolites; with equal numbers of square
planar and tetrahedral nodes the PtS net; and for octahedral and trigonal centers in
equal proportion the TiO2 rutile net. For three-coordinate nodes a common arrange-
ment is the 105105105 network drawn in Figure 1.36, which is often referred to as the
(10,3)-a network in the literature.47 This network is chiral and contains fourfold helices
of the same handedness parallel to each of the cubic axes [47, 48].
To describe and systematize MOF structures, their nets are often given simple codes of the
form abc-d. For example, a primitive cubic net is called pcu. When a vertex is placed halfway
along the edge, the expanded net is called pcu-e. This net is the O net in ReO3 and sufficiently
common that it is given its own name—reo (i.e. pcu-e ≡ reo). These symbols can also be used
for zeolites. For example, the net of four-coordinated T vertices of the zeolite faujasite has
been given the symbol fau. When considering the O atoms of faujasite, the expanded
structure is called fau-e. There are excellent reviews available [49] and various online
databases and software packages available for analyzing nets.48
The discipline of targeting a MOF with a specific topology has been termed reticular
chemistry.49 The MOF-5 family of materials illustrates this approach: MOF-5 (Figure
14.14) can be crystallized from Zn salts and benzenedicarboxylic acid (bdc) in N,N 0 -
diethylformamide at 85–105 °C. These experimental conditions favor the in situ

(b)

(a)

bdc tpdc Zn4O

Figure 14.14 Structures of two MOF-5-related materials that form part of an isoreticular series. Spheres
of radii 8 Å (a) and 11.8 Å (b) are drawn inside the pores to emphasize porosity.

47
Here the (n,m) nomenclature refers to a network made of nodes of connectivity m, in which the shortest closed loop
has n nodes; the attached “a” says it has the highest symmetry of several arrangements possible.
48
For example, the reticular chemistry structure resource at http://rcsr.anu.edu.au/, the EPINET (Euclidean Patterns
in Non-Euclidean Tilings) resource at http://epinet.anu.edu.au/, and ToposPro at http://topospro.com.
49
From the Latin reticulum, meaning little net.
608 Zeolites and Other Porous Materials

formation of Zn4O(CO2R)6 clusters made up of four ZnO4 tetrahedra that share a

common corner (Figure 14.13). The six carboxylate groups, which complete the Zn
coordination, emanate octahedrally from the cluster and link to other clusters in a
CaB6-type network. The framework of MOF-5 remains intact on heating to remove
solvent,50 leading to a material that is truly porous and stable up to around 400 °C (for
the related IRMOF-6). The multidentate binding of the carboxylate anion linker and
the inherent rigidity of the Zn4O building block are thought to impart this stability. It
is also possible to replace the short bdc linker by a variety of larger linkers, including
bpdc and tpdc of Figure 14.13, to produce more open frameworks. This is the process
of expansion that gives rise to structures that are members of an isoreticular series
(maintain the node connectivity).51 The tpdc material (Figure 14.14b) has a cell
parameter of 21.49 Å, a percentage free volume of 91.1%, and a remarkably low density
of just 0.21 g/cm3.52 A sphere of diameter 28.8 Å will fit in the cavity without
overlapping any of the framework atoms. The available pore space in these and later
generations of MOFs is far higher than in any zeolite and only exceeded in amorphous
xerogels and aerogels. Surface areas up to 10000 m2/g are possible. However, as with
the cyanidometallates discussed above, the likelihood of forming interpenetrated net-
works increases with linker length.

14.4.2 Some Applications of MOFs

While zeolites and other porous oxides have had enormous industrial impact, the take-
up of MOFs by industry has been slower and commercial applications are only just
beginning to emerge. What’s clear is that MOFs will not replace zeolites in their large-
scale catalytic applications, where stability at high temperatures in harsh chemical
environments and relatively low cost are key. Indeed, the framework linkers in
MOFs are exactly the type of species zeolites are designed to chemically transform.
Instead, applications are likely to exploit MOFs’ high porosities for molecular storage
and release, the possibility of building responsive frameworks in which bonds can
break and reform dynamically, the ability to incorporate specific chemical functionality
either during or after MOF synthesis, and the possibility of engineering multifunction-
ality in a single material. We will highlight just some of the potential uses of MOFs in
this section. Starting points to discover other potential applications are given in
Further Reading.
There has been significant interest in exploiting the large internal surface area of MOFs for
the physisorption of gases such as CH4, H2, CO2, CO, and NO. In the case of H2, the

50 51
This is not the case for many MOFs. An isoreticular series of structures have the same topology or net.
52
When discovered, this was the lowest density known for a crystalline material. Still lower densities have since been
found. For example the covalent organic framework COF-108 has a density of 0.17 g/cm3 [El-Kaderi et al., Science
316 (2007), 268–272] and MOF-339 a density of 0.13 g/cm3 and void volume of 94% [Furukawa et al., Inorg. Chem.
50 (2011), 9147–9152].
 14.4 Metal–Organic Frameworks (MOFs) 609

possibility of using MOFs for on-vehicle storage stimulated large amounts of research.
MOF-5, for example, stores up to 10 wt% H2 at 77 K and 100 bar, corresponding to a
volumetric density of 66 g H2 per liter, more than twice that of pure H2 under the same
conditions (31 g/L). Unfortunately, at 298 K the adsorption capacity decreases significantly
to around 8.9 g/L, so it is little better than the pure gas (8.7 g/L). Similar behavior occurs with
most hydrogen-storage MOFs and the best materials have gravimetric capacities of ≲ 0.5–1
wt% at room temperature. The capacity decrease with temperature arises from relatively low
isosteric heats of adsorption53 of 5–12 kJ/mol; values of 15–25 kJ/mol are probably needed
for realistic applications. The best room-temperature storage is achieved in MOFs that
contain metal nodes with open coordination sites, but volumetric capacities remain limited
to ≲ 12 g/L.
A second fuel-storage application that was researched extensively targets methane-
powered vehicles. MOF-loaded tanks capable of storing two to three times more CH4
than regular tanks at comparable pressures and temperatures have been successfully
cycled thousands of times, though have not yet been commercialized. Another fascin-
ating potential application is the use of MOFs to capture water from the air. For
example, Kim et al. [50] have shown that Zr4O4(OH)4(fumarate)6 (MOF-801) will take
up water from air of a mere 20% humidity at 25 °C in the dark. Sunlight can then be
used to heat the MOF and release the water. Devices capable of producing 2.8 L of
water per kg of MOF have been described.
The gas-storage properties of MOFs are particularly dramatic when specific host–
guest interactions can be built into the framework. For example, Cu2(pzdc)2pyr (pzdc =
pyrazine-2,3-dicarboxylate, pyr = pyrazine) contains 1D channels in which selective
adsorption of acetylene over the similarly sized CO2 occurs due to strong hydrogen
bonding between acetylene and a carboxylate oxygen. The amount of acetylene stored is
0.43 g/cm3 and would correspond to a hypothetical acetylene pressure of 41 MPa at
room temperature, 200 times the pressure at which it would normally explode [51].
There is also interest in MOFs that can selectively trap CO2 from mixed-gas streams and
even transform it chemically [52]. Tricks such as decorating the interior of pores with
amine groups to enhance the interaction with CO2 have been particularly successful. At
the time of writing, the first smaller-scale MOF-based products, based on gas capture/
release properties, are starting to be produced commercially [53]. These include: MOF-
loaded gas cylinders to allow the storage of toxic gases at sub-atmospheric pressures,
removing the risk of leakage; MOF sachets that release the fruit-ripening inhibitor 1-
methylcyclopropane; and antimicrobial-release devices.
A second interesting aspect of MOF chemistry is the potential for coordination
“flexibility”. In the zeolites we discussed in the first part of this chapter, strong Si–O
and Al–O bonds mean that their frameworks remain intact during most sorption and

53
Isosteric heat of absorption is a measure of the strength of interaction between the gas and MOF surface. It’s defined
as the difference between the partial molar enthalpies of a species in the bulk fluid and adsorbed phases.
610 Zeolites and Other Porous Materials

catalytic applications. With MOFs, it is possible to use weakly binding groups such
that framework linkages can break and reform reversibly. For example, the structure
obtained on dehydrating Ni2(4,4 0 -bipy)3(NO3)4⋅4H2O contains cavities large enough to
hold a toluene molecule, but has connecting windows that are only ~50% of the size
required for toluene to diffuse through the framework to reach them. Nonetheless
toluene can be absorbed into the structure suggesting that local flexibility imparted
by breaking weak C–H····O hydrogen bonds allows the windows to open, permitting
passage of the guest, then close, to “trap” the guest in the structure. Guest uptake is
therefore controlled by the size of the cavity itself rather than that of the window
leading to it.
A third attractive feature of MOFs is the way in which their structures can be modified and
exploited to facilitate useful chemistry. We’ll exemplify this by touching on their possible use
in heterogeneous catalysis. We can conceptually think of five different loci of MOF-catalysis
centers, each of which can make use of the shape selectivities discussed in Sections 14.1.3 and
14.2 [54]. Metal-centered catalysis can occur either at (1) the nodes (particularly if they have
vacant coordination sites), (2) at linkers containing catalytically active metal centers incorp-
orated during MOF synthesis, or (3) at linkers modified post-synthesis to incorporate metals.
Similarly, an active organic catalyst can be used as (4) a linker during MOF synthesis or (5)
created by modifying the linker post-synthesis. In addition, one can prepare MOFs from
chiral linkers. This leads to the possibility of performing enantioselective catalysis as
explored in Box 14.3.
The advantages of a MOF-based catalyst over a homogeneous catalyst combine the
advantages of zeolites and porous oxides discussed in earlier sections with additional
factors: Their heterogeneous nature means that expensive catalysts can be readily
separated from liquid products and re-used. This also helps reduce the contamination
of products with potentially toxic metals. The ability to anchor and isolate catalytic
species in pores can increase the catalyst lifetime as any deactivating processes involv-
ing more than one catalytic center (e.g. clustering or the attack of the fragile ligand set
on one catalytic molecule by another) are decreased. MOFs also have a particular
advantage in the tunability of their shape selectivity as pore sizes can be systematically
controlled through reticular chemistry. One would expect, and often finds experimen-
tally, that decreasing the pore size will increase the product selectivity, though often at
the expense of the reaction rate. Conversely, increasing the pore size typically increases
reaction rates as reagents and products can diffuse more rapidly to and away from
active centers. Rates and selectivities are expected to approach those of homogeneous
catalysts as pore sizes increase, and this has again been verified experimentally. One can
even imagine dual MOF catalysts where tandem or sequential processes occur at
different active centers, or using photochemistry at a metal center to promote organic
catalysis at a linker. This range of possibilities and post-synthetic control sets MOFs
apart from zeolites and mesoporous materials.
 14.4 Metal–Organic Frameworks (MOFs) 611

Box 14.3 Materials Spotlight: Asymmetric catalysis using MOFs

There is enormous interest in the synthesis of chiral molecules, that is molecules whose object
and mirror image are non-superimposable (enantiomers). Enantiomers often have significantly
different biological activity, such that enantioselective synthesis or catalysis is enormously
important to the pharmaceutical industry. MOFs offer exciting opportunities in this arena
since their design-and-synthesis strategy allows much of the learning from homogeneous
asymmetric catalysis to be reapplied in heterogeneous catalysts, with the added benefits of
easy catalyst capture and recycling and pore-size control. Enantioselective catalysis using
MOFs has been demonstrated using each of the five strategies to produce catalytic sites
discussed in the main text and in Figure B14.3.1.

(1) vacant node

node coordination
site or…

(2)
. either metal- (3) linker that can
containing linker have metals
linker

Ti(OiPr)4
or… attached or…
(4) catalytic organic (5) a chemically
linker or… modified linker BINOL activated
derivative BINOL
(chiral or non-chiral)
removable
node
node ligand

Figure B14.3.1 Types of catalytic center possible in MOFs and schematic example of post-synthesis linker
functionalization.

The first demonstration of MOF-based asymmetric catalysis used an enantiopure Zn MOF

called POST-154 and relied on organocatalysis using the linker [55]. POST-1 was synthesized using
an enantiopure linker containing one carboxylate and one pyridyl binding site, producing large
chiral 1D channels. Half the pyridyl nitrogens coordinate to Zn clusters at the nodes and half
point into the framework channels. Its chemical tunability was demonstrated by showing that
channel nitrogens could undergo chemical reactions post-synthesis, and the chiral properties of
the framework by demonstrating enantioselective uptake of metal complexes from solution. Most
importantly, the framework pyridyl groups were shown to catalyze an organic reaction55 with a

54
POST-1 is [Zn3(μ-O)(LH)6]·2H3O·12H2O, where linker L contains one carboxylate and one pyridyl group and can
be prepared enantiopure from tartaric acid.
55
Specifically a transesterification reaction with an ee of 8%.
612 Zeolites and Other Porous Materials

Box 14.3 (cont.)

small enantiomeric excess (ee). The relatively low ee was ascribed to the active centers being
remote from the full influence of the chiral pore walls.
Similar issues of low ee might be expected if one relied on using chiral framework linkers to
induce enantioselective catalysis at metal nodes. Despite this, early work demonstrated that a
framework made from Zn and a chiral triangular linker56 could achieve a 40% ee for a Lewis-acid-
catalyzed reaction.57
The first example of linker–metal-based asymmetric catalysis in a crystalline MOF was
achieved in a Cd MOF built with a pyridyl-functionalized BINOL58 linker (see Figure B14.3.1)
[56]. After synthesis, the two free OH groups on the linker could be reacted with Ti(OiPr)4 to give a
[MOF]Ti(BINOL)(OiPr)2 material. This mimics many solution-based catalysts and was shown to
catalyze the addition of diethylzinc to aldehydes with ee up to 93%.
Another route to enantioselectivity is to take advantage of the built-in chirality of some
networks, e.g. of the (10,3)-a network. This gives the possibility of producing a homochiral
material using enantiopure templates, solvents, or ancilliary ligands, which can later be removed
and reused. One example of ancillary-ligand control is Ni3(btc)2(pd)3(py)6, where py is pyridine,
pd is propan-1,2-diol, and btc is shown in Figure 14.13 [57]. Here, one enantiomer of pd gives a
structure with anticlockwise helices and the other clockwise. This control occurs because the
unobserved structures would contain unreasonably short nonbonded interactions. Related mater-
ials have been shown to retain porosity on guest loss, allowing enantioselective (though with
modest ee) sorption of suitably sized guest materials [58].

14.5 Problems
14.1 The α-cristobalite polymorph of SiO2 has a tetragonal cell with a = 4.971 Å, b =
6.922 Å, with Si on 2a and O on 4f Wyckoff sites. The Si–O bond length is 1.60 Å
and the average O–O distance within tetrahedra is 2.62 Å. The high-pressure stishovite
polymorph of SiO2 has the rutile structure (Figure 1.45) with a tetragonal cell of
dimensions a = 4.177 Å and c = 2.665 Å, and occupied Wyckoff sites of 2a (Si) and
4f (O). Data for a zeolite A polymorph of SiO2 (LTA) are given in the table accom-
panying Problem 14.4. (a) Calculate the theoretical density of each polymorph. (b)
Assuming that Si and O can be treated as hard spheres and that O atoms are in contact
in cristobalite, estimate hard-sphere radii for Si and O. (c) Estimate the volume percent
filled by atoms in each polymorph. Comment on your answers.
14.2 Calculate the surface area of (a) a 1 g single crystal of cristobalite (density = 2.3 g/cm3,
assume the crystal takes a spherical shape), (b) a uniform powder of spherical crystals
56
The material was Zn3(ChirBTB-1)2 where BTB-1 is [4,4 0 ,4″-benzene-1,3,5-triyl-tribenzoate] with a chiral oxazoline
group at the ortho position of each benzoate.
57
Specifically conversion of 1-napthylaldehyde in the Mukaiyama aldol reaction.
58
BINOL is the colloquial name of [1,1 0 -binapthelene]-2,2 0 -diol. It is widely used as a ligand in homogeneous catalysis
as its two enantiomers are easily separable and stable to racemization.
 14.5 Problems 613

each with a diameter of 10 μm, and (c) uniform spherical particles with a diameter of
10 Å. Compare your values to the surface area of a typical zeolite of 900 m2/g.
14.3 Zeolite A can be described using a 11.9 Å primitive cubic cell with the origin at a β-cage
center in Figure 14.3. From the figure, make a visual estimate of the size of a sphere that
would fit inside the β and α cages. By using the information in the table accompanying
Problem 14.4, estimate the surface area of these spheres in a 1 g sample of zeolite A.
Compare your answer to the accessible area of 1205 m2/g quoted in the text.
14.4 Calculate the number of tetrahedral atoms per nm3 for each zeolite framework type
listed in the table below.

Name Space group Cell (Å) Site x y z Wyckoff

3
SOD Im m 8.9650 T1 0.2500 0.5000 0.0000 12d
O1 0.1467 0.5000 0.8533 24h
LTA Pm3 m 11.919 T1 0.0000 0.1823 0.3684 24k
O1 0.0000 0.2122 0.5000 12h
O2 0.1103 0.1103 0.3384 24m
O3 0.0000 0.2967 0.2967 12i
FAU Fd3 m 24.34 T1 0.9469 0.1251 0.0364 192i
O1 0.8958 0.1042 0.0000 96h
O2 0.9669 0.0762 0.0762 96g
O3 0.9966 0.1429 0.9966 96g
O4 0.9283 0.1770 0.0730 96g
ANA Ia3 d 13.567 T1 0.6616 0.5884 0.1250 48g
O1 0.6417 0.4716 0.1180 96h
RHO Im3 m 14.919 T1 0.2500 0.1037 0.3963 48i
O1 0.2754 0.1193 0.5000 48j
O2 0.3344 0.1298 0.3344 48k

14.5 The table above contains ideal framework coordinates for different zeolites. Use a
package such as vesta (http://jp-minerals.org/vesta/en/) to produce a 3D drawing of
each structure. For each example state: (a) which SBUs are present (see Figure 14.4);
(b) the vertex symbol of the T site; (c) the coordination sequence (out to the level N3).
14.6 Assuming that O atoms are in contact and have a radius of 1.35 Å, estimate the pore
opening formed by a planar six-, eight-, ten-, and twelve-membered SiO4/2 ring. Note
that if the rings are nonplanar, pore sizes can be considerably reduced. Hint: Consider
the internal size of a circle drawn inside a ring of circles.
14.7 In NaA, the total void space accessible is approximately 926 Å3. Sorption studies
suggest H2O fills a volume ~833 Å3, whereas N2 only fills ~755 Å3. Comment on these
observations.
14.8 Under certain conditions, a zeolite with the ZSM-5 framework and formula
NaAlSi23O48 (A) can be synthesized. In the presence of (CH3CH2CH2)4NOH it is
possible to prepare a material (B) with an extremely similar X-ray-diffraction pattern
614 Zeolites and Other Porous Materials

and the elemental composition N 0.86%, C 8.86%, H 1.73%, Al 1.66%, Si 39.70%, O

47.19% by weight. On heating to 500°C B loses 11.39% of its mass to form C. (a)
Calculate the empirical formula of B from the elemental analysis. (b) Given the
similarity in the X-ray-diffraction patterns of A and B, suggest a sensible chemical
formula for B. (c) Calculate the empirical formula of C. (d) State how a Na-containing
zeolite such as A can be converted to an acidic form. (e) State the roles played by
organic cations such as [(CH3CH2CH2)4N]+ in zeolite synthesis.
14.9 A mixture of Si(OCH3)4, (CH3)4NOH(aq), and Al(OCH2CH2CH3)3 was heated to drive
off alcohol, resulting in a clear solution. This solution was heated at 130 °C in a sealed
autoclave for 8 days. The resulting white precipitate was washed with distilled water and
dried at 100 °C, yielding a white powder A whose X-ray-diffraction pattern was very
similar to that of the mineral gismondine (CaAl2Si2O8·4H2O). Chemical analysis of A
gave weight percentages N 4.2%, C 14.5%, H 4.2%, Al 8.1%, Si 25.4%, O 43.5%. (a)
Calculate the empirical formula of A. (b) Given that the X-ray-diffraction pattern is
similar to gismondine, suggest a sensible molecular formula for A. (c) In light of your
answer to (a) and (b), discuss how template molecules may be used to influence the
hydrophilicity of zeolites.
14.10 Haag and co-workers measured the cracking rate of n-hexane by HZSM-5 relative to a
high-surface-area catalyst under identical experimental conditions [59]. The depend-
ence of the rate on the Al:Si ratio and on the intensity of the signal due to tetrahedral
Al sites in the 27Al solid state NMR are shown in the figures below. State what you can
conclude from the linear dependence of activity on both the Al:Si ratio and the NMR
signal from tetrahedral sites.

1000 250
relative cracking activity
r elative cracking activity

200
100

150
10
100

1
50

0.1 0
0.00001 0.001 0.1 0 50 100
t etrahedral 27Al NMR signal
Al:Si ratio

14.11 When dimethylbenzenes (xylenes) are passed over acidic zeolites (HZSM-5 that
contains 10-rings and HY that contains larger 12-rings), two processes can occur:
isomerization to different mixtures of ortho, meta, and para isomers; or dispropor-
tionation to toluene (methylbenzene) and trimethylbenzenes, as shown in the table.
State which of the shape-selective catalysis effects shown in Figure 14.7 is responsible
 14.6 Further Reading 615

for the enhanced para- to ortho-selectivity of HZSM-5 and for favoring the isomer-
ization reaction over disproportionation in HZSM-5.

Zeolite para-/ortho-selectivity Isomerization/disproportionation

HZSM-5 2.9 33
HY 1.0 1.5

14.12 The first peak in the powder diffraction pattern of a mesoporous MCM material with
30 Å pores recorded with a wavelength λ of 1.54 Å is at 2.2° 2θ. Estimate the thickness
of the silica walls. A hexagonal material has a = b ≠ c, α = β = 90, γ = 120; Bragg’s law
states λ = 2dhklsinθ.
14.13 Crystallographers have a rough “rule of thumb” that non-H atoms occupy around 18
Å3 in many crystal structures. Based on this assumption, estimate the percent pore
space in the tpdc MOF on the right of Figure 14.14.
14.14 Consider the PtS net listed in Table 1.6. The dehydrated form of MOF-11 has compos-
ition Cu2(ATC) where ATC is 1,3,5,7-adamantane tetracarboxylate and contains a Cu
paddle-wheel unit analogous to the Zn dimer shown in Figure 14.13. Describe the
structure of PtS. Based on the information given and your chemical knowledge, suggest
possible structures for MOF-11 and ZnPt(CN)4.
14.15 Consider a heterogeneous enantioselective catalyst based on a catalytically active metal
center attached to a MOF linker. (a) How would you expect the reaction rate to compare
to the equivalent catalyst in solution? How might reaction rate vary as the pore size
varied? (b) How might the enantiomeric excess compare to the equivalent catalyst in
solution? (c) How might catalyst stability and turn-over compare to the equivalent
catalyst in solution? (d) How would you prove that catalysis is truly heterogeneous and
not caused by catalyst leaching from the MOF into solution? (e) How would you prove
that catalysis occurs in the pores of the MOF and is not restricted to surface sites?

14.6 Further Reading

P.A. Wright, “Microporous Framework Solids” RSC Materials Monographs (2008) Cambridge.
M.E. Davis, “Ordered porous materials for emerging applications” Nature 417 (2002), 813–821.
T.J. Barton, L.M. Bull, W.G. Klemperer, D.A. Loy, B. McEnaney, M. Misono, P.A. Monson, G. Pez, G.
W. Scherer, J.C. Vartuli, O.M. Yaghi, “Tailored porous materials” Chem. Mat. 11 (1999), 2633–2656.
F. Schüth, “Engineered porous catalytic materials” Ann. Rev. Mater. Res. 35 (2005), 209–238.
M. Eddaoudi, D.B. Moler, H. Li, B. Chen, T.M. Reineke, M. O’Keeffe, O.M. Yaghi, “Modular
chemistry: Secondary building units as a basis for the design of highly porous and robust metal–
organic frameworks” Acc. Chem. Res. 34 (2001), 319–330.
616 Zeolites and Other Porous Materials

S. Kitagawa, R. Kitaura, S. Noro, “Functional porous coordination polymers” Angew. Chem., Int. Ed.
43 (2004), 2334–2375.
S. Kitagawa, K. Uemara, “Dynamic porous properties of coordination polymers inspired by hydrogen
bonds” Chem. Soc. Rev. 34 (2005), 109–119.
There are many other excellent reviews of MOFs available. For example, the February 2012
issue of Chemical Reviews (112, 673–1268) contains 17 review articles on different aspects of
their chemistry and applications. The August 2014 issue (43, 5403–6176) of Chemical Society
Reviews contains 30 further reviews.

14.7 References
[1] R.D. Shannon, “Revised effective ionic radii and systematic studies of interatomic distances in
halides and chalcogenides” Acta Crystallogr. Ser. A 32 (1976), 751–767.
[2] Ch. Baerlocher, L.B. McClusker, D.H. Olson, “Atlas of Zeolite Framework Types” 6th edition
(2007) Elsevier; www.iza-structure.org/databases/.
[3] L.B. McCusker, F. Liebau, G. Engelhardt, “Nomenclature of structural and compositional
characteristics of ordered microporous and mesoporous materials with inorganic hosts” Pure
Appl. Chem. 73 (2001), 381–394.
[4] V.A. Blatov, O. Delgado-Friedrichs, M. O’Keeffe, D.M. Proserpio, “Three-periodic nets and
tilings: Natural tilings for nets” Acta Crystallogr. Ser. A 63 (2007), 418–425.
[5] N.A. Anurova, V.A. Blatov, G.D. Ilyusin, D.M. Prosperio, “Natural tilings for zeolite-type
frameworks” J. Phys. Chem. C 114 (2010), 10160–10170.
[6] M.M.J. Treacey, I. Rivin, E. Balkowsky, K.H. Randall, M.D. Foster, “Enumeration of periodic
tetrahedral frameworks, II. Polynodal graphs” Micropor. Mesopor. Mater. 74 (2004), 121–132.
[7] O. Delgado-Friedrichs, A.W.M. Dress, D.H. Huson, J. Klinowski, A.L. Mackay, “Systematic
enumeration of crystalline networks” Nature 400 (1999), 644–647.
[8] A. Corma, M.J. Díaz-Cabañas, J.L. Jordá, C. Martinez, M. Moliner, “High-throughput synthesis and
catalytic properties of a molecular sieve with 18- and 10-member rings” Nature 443 (2006), 842–845.
[9] G. Sastre, A. Corma, “Predicting structural feasibility of silica and germania zeolites” J. Phys.
Chem. C 114 (2010), 1667–1673.
[10] R.M. Barrer, “Syntheses and reactions of mordenite” J. Chem. Soc. (1948), 2158–2163.
[11] C.S. Cundy, P.A. Cox, “The hydrothermal synthesis of zeolites: Precursors, intermediates and
reaction mechanisms” Microporous and Mesoporous Materials 82 (2005), 1–78.
[12] J. Felsche, S. Lugher, C. Baerlocher, “Crystal structure of the hydro-sodalite Na6[AlSiO4]6·8H2O
and of the anhydrous sodalite Na6[AlSiO4]6” Zeolites 6 (1986), 367–372.
[13] C. Baerlocher, W.M. Meier, “Synthese und Kristallstruktur von Tetramethylammonium-soda-
lith” Helv. Chim. Acta. 52 (1969), 1853–1860.
[14] K.D. Schmitt, G.J. Kennedy, “Toward the rational design of zeolite synthesis: The synthesis of
zeolite ZSM-18” Zeolites 14 (1994), 635–642.
[15] S.K. Brand, J.E. Schmidt, M.W. Deem, F. Daeyaert, Y. Ma, O. Terasaki, M. Orazov, M.E.
Davis, “Enantiomerically enriched, polycrystalline molecular sieves” Proc. Natl. Acad. Sci. USA
(2017), 201704638.
[16] M. Boronat, A. Corma, “Factors controlling the acidity of zeolites” Catal. Lett. 145 (2015), 162–172.
[17] J.F. Haw, “Zeolite acid strength and reaction mechanism in catalysis” Phys. Chem. Chem. Phys. 4
(2002), 5431–5441.
 14.7 References 617

[18] D.H. Olson, W.O. Haag, “Structure-selectivity relationship in xylene isomerization and selective
toluene disproportionation” ACS Symp. Ser. 248 (1984), 275–307.
[19] N.Y Chen, W.O. Haag, “Hydrogen transfer in catalysis on zeolites” in “Hydrogen Effects in
Catalysis” Editors: Z. Paal, P.G. Mena (1988) Marcel Dekker, 695–722.
[20] A. Primo, H. Garcia, “Zeolites as catalysts in oil refining” Chem. Soc. Rev. 43 (2014), 7548–7561.
[21] A. Dyer, “Ion exchange properties of zeolites and related materials” in “Handbook of Porous
Solids” Editors: F. Schüth, K.S.W. Sing, J. Weitkamp (2002) Wiley, 525–554.
[22] W. Schmidt, “Applications of microporous materials as ion-exchangers” in “Handbook of Porous
Solids” Editors: F. Schüth, K.S.W. Sing, J. Weitkamp (2002) Wiley, 1058–1097.
[23] A. Dyer, “An Introduction to Zeolite Molecular Sieves” (1988) John Wiley and Sons.
[24] T.R. Gaffeny, “Porous solids for air separation” Curr. Opin. Mater. Sci. 1 (1996), 69–75.
[25] M. Bartolomei, E. Carmona-Novillo, M.I. Hernandez, J. Campos-Martinez, R. Hernandez-
Lamoneda, “Long-range interaction for dimers of atmospheric interest: Dispersion, induction
and eletrostatic contributions for O2–O2, N2–N2 and O2–N2, J. Comp. Chem. 32 (2010), 279–290.
[26] Z.P. Lai, G. Bonilla, I. Diaz, J.G. Nery, K. Sujaoti, M.A. Amat, E. Kokkoli, O. Terasaki, R.W.
Thompson, M. Tsapatsis, D.G. Vlachos, “Microstructural optimization of a zeolite membrane
for organic vapor separation” Science 300 (2003), 456–460.
[27] C.T. Kresge, M.E. Loeonowicz, W.J. Roth, J.C. Vartuli, J.S. Beck, “Ordered mesoporous
molecular sieves synthesized by a liquid-crystal mechanism” Nature 359 (1992), 710–712.
[28] J.S. Beck, J.C. Vartuli, W.J. Roth, M.E. Leonowicz, C.T. Kresge, K.D. Schmitt, C.T-W. Chu, D.
H. Olson, E.W. Sheppard, S.B. McCullen, “A new family of mesoporous molecular sieves
prepared with liquid crystal templates” J. Am. Chem. Soc. 114 (1992), 10834–10843.
[29] D. Zhao, J. Feng, Q. Huo, N. Melosh, G.H. Fredrickson, B.F. Chmelka, G.D. Stucky, “Triblock
copolymer syntheses of mesoporous silica with periodic 50 to 300 ångstrom pores” Science 279
(1998), 548–552.
[30] J.M. Thomas, R. Raja “Exploiting nanospace for asymmetric catalysis: Confinement of immobil-
ized, single-site chiral catalysts enhances enantioselectivity” Acc. Chem. Res. 41 (2008), 708–720.
[31] T. Prasomsri, W. Jiao, S.Z. Weng, J. Garcia Martinez, “Mesostructured zeolites: Bridging the
gap between zeolites and MCM-41” Chem. Commun. 51 (2015), 8900–8911.
[32] E.S. Toberer, R. Sheshadri, “Template-free routes to porous inorganic materials” Chem.
Commun. (2006), 3159–3165.
[33] T.A. Mary, J.S.O. Evans, T. Vogt, A.W. Sleight, “Negative thermal expansion from 0.3 to 1050
kelvin in ZrW2O8” Science 272 (1996), 90–92.
[34] M. Dove, H. Fang, “Lattice dynamics, negative thermal expansion, and associated anomalous
physical properties of materials” Rep. Progr. Phys. (2016), 066503.
[35] A.L. Goodwin, M. Calleja, M.J. Conterio, M.T. Dove, J.S.O. Evans, D.A. Keen, L. Peters, M.G.
Tucker, “Colossal positive and negative thermal expansion in the framework material Ag3[Co
(CN)6]” Science 319 (2008), 794–797.
[36] A. Sartbaeva, S.A. Wells, M.M.J. Treacy, M.F. Thorpe, “The flexibility window in zeolites”
Nature Mater. 5 (2006), 962–965.
[37] Y, Lee, J.A. Hriljac, T. Vogt, J.B. Parise, G.J. Artioli, “First structural investigation of a super-
hydrated zeolite” J. Am. Chem. Soc. 123 (2001), 12732–12733.
[38] M. Colligan, Y. Lee, T. Vogt, A.J. Celestian, J.B. Parise, W.G. Marshall, J.A. Hriljac, “High-pressure
neutron diffraction study of superhydrated natrolite” J. Phys. Chem. B 109 (2005), 18223–18225.
[39] F. Millange, C. Serre, N. Guillou, G. Féréy, R.I. Walton, “Structural effects of solvents on the
breathing of metal-organic frameworks: An in-situ diffraction study” Angew. Chemie Int. Ed.
Engl. 47 (2008), 4100–4105.
618 Zeolites and Other Porous Materials

[40] C. Mellot-Draznieks, C. Serre, S. Srublé, N. Audebrand, G. Féréy, “Very large swelling in hybrid
frameworks: A combined computational and powder diffraction study” J. Am. Chem. Soc. 127
(2005), 16273–16278.
[41] S. Bourrelly, P.L. Llewellyn, C. Serre, F. Millange, T. Loiseau, G. Féréy, “Different adsorption
behaviors of methane and carbon dioxide in the isotypic nanoporous metal terephthalates MIL-
53 and MIL-47” J. Am. Chem. Soc. 127 (2005), 13519–13521.
[42] K.E. Evans, A. Alderson, “Auxetic materials: Functional materials and structures from lateral
thinking!” Adv. Mat. 12 (2000), 617–628.
[43] J.N. Grima, R. Gatt, V. Zammit, J.J. Williams, K.E. Evans, A. Alderson, R.I. Walton,
“Natrolite: A zeolite with negative Poisson’s ratios” J. Appl. Phys. 101 (2007), 086102.
[44] R. Lakes, “Foam structures with a negative Poisson’s ratio” Science 235 (1987), 1038–1040.
[45] B.F. Hoskins, R. Robson, “Design and construction of a new class of scaffolding-like materials
comprising infinite polymeric frameworks of 3D-linked molecular rods. A reappraisal of the zinc
cyanide and cadmium cyanide structures and the synthesis and structure of the diamond-related
frameworks [N(CH3)4][CuIZnII(CN)4] and CuI[4,4 0 ,4″,4‴-tetracyanotetraphenylmethane]
BF4·xC6H5NO2” J. Am. Chem. Soc. 112 (1990), 1546–1554.
[46] A.F. Wells, “Three Dimensional Nets and Polyhedra” (1977) Wiley-Interscience.
[47] R. Robson, “A net-based approach to coordination polymers” J. Chem. Soc., Dalton Trans.
(2000), 3735–3744.
[48] O. Yaghi, M. O’Keeffe, N.W. Ockwig, H.K. Chae, M. Eddaoudi, J. Kim, “Recticular synthesis
and the design of new materials” Nature 423 (2003), 705–714.
[49] M. O’Keeffe, O.M. Yaghi, “Deconstructing the crystal structures of metal organic frameworks
and related materials into their underlying nets” Chem. Rev. 112 (2012), 675–702.
[50] H. Kim, S. Yang, S.R. Rao, S. Narayanan, E.A. Kapustin, H. Furukawa, A.S. Umans, O.M.
Yaghi, E.N. Wang. “Water harvesting from air with metal–organic frameworks powered by
natural sunlight” Science 356 (2017), 430–434.
[51] R. Matsuda, R. Kitaura, S. Kitagawa, Y. Kubota, R.V. Belosudov, T.C. Kobayashi, H.
Sakomoto, T. Chiba, M. Takata, Y. Kawazoe, Y. Mita, “Highly controlled acetylene accommo-
dation in a metal–organic microporous material” Nature 236 (2005), 238–241.
[52] C.A. Trickett, A. Helal, B. Al-Maythalony, Z.H. Yamani, K.E. Cordova1, O.M. Yaghi, “The
chemistry of metal–organic frameworks for CO2 capture, regeneration and conversion” Nature
Reviews Materials 2 (2017), 17045.
[53] N. Notman, “MOFs find a use” Chem. World (May 2017), 44–47.
[54] M. Yoon, R. Srirambalaji, K. Kim, “Homochiral metal–organic frameworks for asymmetric
heterogeneous catalysis” Chem. Rev. 112 (2012), 1196–1231.
[55] J.S. Seo, D. Whang, H. Lee, S.I. Jun, J. Oh, Y.J. Jeon, K. Kim, “A homochiral metal–organic
porous material for enantioselective separation and catalysis” Nature 404 (2000), 982–986.
[56] C. Wu, A. Hu, L. Zhang, W. Lin, “A homochiral porous metal–organic framework for highly
enantioselective heterogeneous asymmetric catalysis” J. Am. Chem. Soc. 127 (2005), 8940–8941.
[57] C.J. Kepert, T.J. Prior, M.J. Rosseinsky, “A versatile family of interconvertible microporous
chiral molecular frameworks: The first example of ligand control of network chirality” J. Am.
Chem. Soc. 122 (2000), 5158–5168.
[58] D. Bradshaw, T.J. Prior, E.J. Cussen, J.B. Claridge, M.J. Rosseinsky, “Permanent microporosity and
enantioselective sorption in a chiral open framework” J. Am. Chem. Soc. 126 (2004), 6106–6114.
[59] W.O. Haag, R.M. Lago, P.B. Weisz, “The active site of acidic aluminosilicate catalysts” Nature
309 (1984), 589–591.
15 Amorphous and Disordered Materials

While the majority of this book has focused on crystalline materials with long-range
translational order and symmetry, many of the oldest technological materials are
amorphous. Glazed Egyptian artifacts have been dated back to 4000 BCE and glass
beads used as currency to around 3000 BCE. Glasses are used in a wide variety of
scientific instruments such as optical microscopes, telescopes, thermometers, barom-
eters, and the plethora of glassware that chemists use. Many major scientific advances
have relied on the ability to engineer a certain type of glass. Without glass, Hooke, van
Leeuwenhoek, Pasteur, and Koch would not have been able to investigate microorgan-
isms under a microscope. Without glass lenses in his telescope, Galileo would not have
made the celestial observations that changed our view of the universe. Three hundred
years after Galileo, the Hooker telescope at the Mount Wilson Institute was used by
Edwin Hubble to devise his theory of an expanding Universe. Over 2 tonnes of glass had
to be slow-cooled over a year-long period in the first stage of manufacturing its 2.5-m-
diameter mirror.
In this chapter, we cover structural and dynamical aspects of glasses, amorphous, and
disordered materials; terms that are often used interchangeably in this field. We discuss
their formation and the concept of the glass transition, of which P.W. Anderson wrote 25
years ago: “The deepest and most interesting unsolved problem in solid state theory is
probably the theory of the nature of glass and the glass transition” [1]. We highlight some
of the unique physical properties of glasses, in particular their low-temperature dynamics
and conductivity. We will focus mainly on inorganic glasses, but many of the concepts
apply equally to organic glasses, amorphous polymers, and biopolymers (e.g. plastics,
RNA, DNA, proteins), which can form randomly entangled and intermeshed chains and
constitute an important group of amorphous materials called random coil glasses. We
point to some of the most important technological uses of glasses. At various points in the
chapter, we touch on general structural concepts regarding topology and space filling,
which apply to amorphous, quasicrystalline, and crystalline materials.

619
620 Amorphous and Disordered Materials

15.1 The Atomic Structure of Glasses

Glasses differ from crystalline materials in that they don’t have a lattice. Atoms positioned
on lattices have long-range order and form a regular array described by translational
symmetry as discussed in Chapter 1. While long-range order is absent in glasses, the
distribution of atoms is not completely random since the presence of chemical bonds imposes
similar interatomic distances throughout the solid. This local similarity is called short-range
order and is a property of all solids. The absence of long-range order yields diffraction data
(see Box 9.1) with no or very broad Bragg peaks, and it has important implications for the
mechanical, electronic, and vibrational properties of glasses. The inorganic glasses we’ll
focus on can be classified into two large families—covalent glasses such as amorphous silica,
and random-packed glasses such as metallic glasses.
Short-range order in glasses arises due to excessive bonding constraints on the connectiv-
ity of local coordination polyhedra. Too many, too strong, or too incompatible bonding
constraints1 destroy lattice periodicity rather than just decreasing the crystal symmetry. This
may also happen when a system that would otherwise crystallize is not given enough time to
equilibrate.
The nature of the short-range order in glasses is determined by local polyhedra like BO3/2
triangles or SiO4/2 tetrahedra in a covalent glass, or specific metal polyhedra in random-
packed metallic glasses. Glassy SiO2 is built of corner-sharing SiO4 tetrahedra with Si–O
bond lengths ~1.6 Å and O–Si–O bond angles near 109.5°. The range of distances encoun-
tered experimentally is shown in Figure 15.1. As one enlarges the range beyond a single
tetrahedron, certain structural configurations with similar distances and angles tend to
reappear, but not with lattice periodicity. The loss of periodicity is due to differing angles
between tetrahedra. This calls for a new way to represent glass structures.
One of the most common ways to describe the structure of a glass is the pair distribution
function (PDF), G(r), which describes the distribution (a statistics of occurrences) of all
pairs of all atoms as a function of their interatomic distances, r, throughout the structure.
An example of a PDF for silica glass is depicted in Figure 15.1. The function G(r) gives
the probability of finding any two atoms at a distance r, with the area of each peak
corresponding to the number of contributing pairs scaled by the sensitivity of the
experimental probe to the specific atoms forming the pair (see ref. [2] for more details).
We can see that with increasing r, the peaks increase in width and decrease in amplitude.
Above ~5 Å, the large number of different pair distances present in a glass means that
only broad, weak peaks are observed in this representation.2 The first peak near 1.6 Å is
due to the four Si–O distances in the SiO4 tetrahedron.3 The next peak, due to six O–O

1
An example is B2O3 that has to accommodate BO3/2 vertex-sharing triangles in a 3D structure.
2
In contrast, the PDF of a crystalline material shows sharp peaks out to very large r and is only limited by
experimental constraints, not by the long-range order of the sample.
3
The small peaks below 1.6 Å are due to termination errors of the Fourier transformation used to calculate G(r) and
arise from the finite experimental data range.
 15.1 The Atomic Structure of Glasses 621

O O
Si–O O 1.6 Å
O
2.6 Å
Si Si
3.1 Å
O
G(r) O O
O–O

Si–Si

0 1 2 3 4 5 6
r (Å)

Figure 15.1 PDF describing the distribution of interatomic distances r in amorphous SiO2. Adapted from
ref. [3].

distances within each SiO4 tetrahedron, appears near 2.6 Å. The increase in width of the
second peak over the first one shows the variations in the O–Si–O angles around the ideal
tetrahedral value of 109.5°. The average inter-tetrahedral Si–O–Si angle of ~135° can
only be reliably extracted using a larger segment of structural configurations. The peak at
3.1 Å is caused by four Si–Si distances, and, as evidenced by its width, it is impacted by
the distribution of the angles within the relatively large set of structural configurations.
The peaks at 4 Å and 5 Å correspond to the O–O distances between two tetrahedra. These
distances will vary upon rotating either tetrahedron around the bond to their shared
vertex (which can be described with a torsion angle).
Many glasses form when a melt is cooled sufficiently quickly through a certain tempera-
ture, called the glass-forming temperature, Tg (more detail in Section 15.7). This approach,
called quenching, does not allow enough time for an ordered crystalline phase to nucleate,
and instead a solid snapshot of a liquid-like structure is formed. We can get some insight into
the type of structures that form by considering a random close packing of spheres. If one pours
spheres into a container and gently taps them down, they do not form the periodic closest-
packed arrangement described in Chapter 1.4 J.D. Bernal explored this type of random
packed structure by making physical models using ball bearings and pouring wax over them
so he could determine the positions of individual spheres [4, 5]. He observed a predominance
4
Gentle tapping traps the system in a metastable state while more rigorous shaking would result in closer packing.
Random close packing is not an intrinsic property of a given material but depends on the packing procedure. These
ideas are further explored in Section 15.7.
622 Amorphous and Disordered Materials

of pentagonal faces among the various polygons present; we will return to this important
observation later on (Section 15.11) when we discuss icosahedral coordination polyhedra
and their inability to tile 3D space. Bernal found that the packing fraction of these arrange-
ments is near 64%, which is significantly less than is achieved in an ordered crystalline
packing (74.05% for cubic and hexagonal closest packing, ccp and hcp; see Chapter 1).
The difference in packing fraction between random and ordered structures led to the
concept of free volume, defined as the volume in excess of the ideal closest packing. Free
volume implies that there will be mass-density fluctuations in the sample, which can be
probed by tomography, small-angle scattering, or other non-Bragg scattering methods
[2]. Changes in the free volume result in substantial changes of viscosity and flow and
also affect elastic constants, electrical conductivity, and other material properties [6]. The
structural and dynamical characterization of random voids making up the free volume in
glasses and heterogeneities in supercooled liquids is a very active research area (see
Further Reading).
Bernal’s random space filling of ~64% is a local minimum of the random packing density
for spheres and gives an average coordination number ⟨ Z ⟩ ≈ 6. For both stretched or
flattened spheres with gradually increasing aspect ratios (more prolate and more oblate5), the
packing volume fraction initially increases above 70% of ⟨ Z ⟩ ≈ 10 and then falls off again as
the aspect ratio grows further [7].

15.2 Topology and the Structure of Glasses

The topology of glasses describes the connectivity of the constituent objects rather than
distances between them. Both covalent and random-packed glasses have distinctive topolo-
gies that can be described using graphs. A graph in this context is a three-dimensional object
made up of vertices and edges. The graph vertices can be atoms and the edges can be bonds.
The graph can therefore correspond to a real structure.6 Atoms in glasses often have a fixed
chemical coordination number, hence the graph vertices have fixed connectivity (e.g. 4 for Si
and 2 for O in amorphous SiO2 or 4 for amorphous Si7). Such glasses are called continuous
random networks (CRN). Models of CRNs were initially built using simple rules: (1) atoms
are fully coordinated, (2) there is no long-range order hence no translational symmetry, and
(3) there are no coordination defects or voids. Due to their importance in understanding the
properties of amorphous materials, advanced CRN-generating algorithms now also take

5
Rotating an ellipse about its major or minor axis leads to a prolate (like a rugby ball) or oblate (like an M&M’s
chocolate drop) shape, respectively.
6
This would be typical for covalent glasses. For random-packed glasses, the graph is a so-called Voronoi network
where the graph edges are those of polyhedra describing the volume of influence of an individual atom.
7
This is an idealization for amorphous Si that is known to have macroscopic voids into which some uncoordinated
electron density of Si atoms can project. These “dangling bonds” are intrinsic defects occurring at a concentration of
about one unpaired electron per 103 atoms, and they introduce electron states at the Fermi energy within the band
gap. See Section 15.10 for more details.
 15.2 Topology and the Structure of Glasses 623

defects and voids into account. The generated CRN models are used to calculate structural,
vibrational, and electronic properties that can be compared to experiments [8].
In Section 1.4.3, we defined a ring as the smallest loop (cycle) that together with other such
rings, the same or different, completes an “infinite” structural network. Such rings are an
important topological feature found in both glasses and crystalline materials. The search for
the smallest set of the smallest rings in liquids, crystals, and amorphous materials is also done
using sophisticated computer algorithms [9].8
A fixed-connectivity network free of rings would give us an ever-growing tree-like top-
ology called a Bethe lattice, with the number of nodes in the nth shell being Nn = Z(Z − 1)n−1,
where Z is the vertex coordination number.9 In the Bethe lattice of SiO2, an SiO4/2
tetrahedron that is the nth neighbor of an arbitrary original “seed tetrahedron” will share
one vertex with its (n − 1)th shell neighbor and three vertices with its (n + 1)th shell neighbors
giving 4 × 3n−1 tetrahedra in the nth shell. This Bethe lattice with vertex coordination number
4 is shown in Figure 15.2 up to the third shell as a surface. In this network, we can form rings
by pruning branches from the “seed tetrahedron” outward and connecting the resulting
underbonded atoms with other atoms in close proximity.
These ideas are also useful in crystalline materials. 3-rings are the smallest ring we
can form by “pruning two tetrahedral branches” that grow from the first shell. As a
result, the seed tetrahedron is still surrounded by four nearest-neighbor tetrahedra but
only by 10 and not 12 second-neighbor tetrahedra [N2 = (4 × 31) − 2]. The formation
of 4-rings by pruning just one tetrahedral branch in the first shell is observed in the

(a) (b)

(c) (d)

Figure 15.2 The “pruning of tetrahedral branches” to form rings in a Bethe lattice of 4-connected vertices
(each represents one tetrahedron, say, of SiO4/2 Niggli formula) out to the 3rd shell. The formation of 3-,
4-, 5- and 6-rings of closed paths is depicted in (a), (b), (c), and (d), respectively.

8
Ring statistics are, for example, included in the program suite ISAACS (Interactive Structure Analysis of
Amorphous and Crystalline Systems) found at http://isaacs.sourceforge.net/over.html.
9
Dendrimers, first made by Vögtle in 1978, are repetitively branched molecules forming a Bethe lattice.
624 Amorphous and Disordered Materials

crystalline SiO2 framework structure of coesite,10 where the seed tetrahedron is

surrounded by four nearest-neighbor tetrahedra but only by 11 and not 12 second-
neighbor tetrahedra [N2 = (4 × 31) − 1]. 5-rings are the result of pruning two tetrahe-
dral branches in the second shell and are found in keatite [10].10 6-rings are made by
pruning only one tetrahedral branch in the second shell, while higher-membered rings
result from pruning in more distant shells. 6-rings are observed in the tridymite and
cristobalite forms of SiO2. The cristobalite structure can be described as 1 seed
tetrahedron + 4 first-neighbor tetrahedra + 12 second-neighbor tetrahedra + 24 third-
neighbor tetrahedra; in total 41 tetrahedra going out to the third shell. In a Bethe
lattice (without forming rings) there would be 52 tetrahedra, N1,2,3 = (4 × 30) + (4 × 31)
+ (4 × 32), out to the third shell as shown in Figure 15.2. The structure of tridymite
differs. It is built up of 1 seed tetrahedron + 4 first-neighbor tetrahedra + 12 second-
neighbor tetrahedra + 25 third-neighbor tetrahedra; in total 42 tetrahedra out to the
third shell. It is thus only in the third neighbor shell that one can distinguish the
topology of these two SiO2 polymorphs from the Bethe lattice. Depicting this in either
a polyhedral or network diagram results in very complex figures. This topological
analysis provides us an explanation of why tridymite cannot transform into cristobalite
by a low-energy distortion of the network but requires heating to break chemical bonds
in a reconstructive phase transition: cristobalite has 24 third-neighbor tetrahedra and
tridymite has 25. Note that both tridymite and cristobalite can undergo topology-
preserving displacive phase transitions between their α and β forms at 114 °C and
270 °C, respectively, where no bonds are broken.
One can also correlate ring size and density. Ring formation always reduces the
density of a Bethe lattice, and small rings reduce the density more than larger ones
since one prunes tetrahedral branches in lower coordination shells. This can be seen in
Figure 15.2 by comparing the amount of tetrahedral pruning when forming a 3- or a
6-ring. In the case of a smaller ring, the sample’s density will be lower. In crystalline
materials, we find 9-, 10-, 11-, and 12-rings in coesite (3.01 g/cm3), 6- and 8-rings
in quartz (2.65 g/cm3), but only 6-rings in cristobalite (2.33 g/cm3) and tridymite
(2.26 g/cm3). Metamict11 and amorphous SiO2 at high temperatures have densities of
2.26 g/cm3 and 2.21 g/cm3, respectively, suggesting small rings are present here. SiO2
melts increase their density under pressure while their short-range order, as shown by
PDF analysis, remains unchanged. This may be related to changing ring sizes. We will
see later that the presence of rings in glasses might also be used to explain vibrational
dynamics observed at very low temperatures (Section 15.9).

10
Besides 4-rings, coesite also contains, 6-, 8-, 9-, 10-, 11-, and 12-rings. See ref. [8]. Keatite also contains 7- and 8-rings.
11
Metamictization describes the process where radiation damage results in a gradual amorphization of materials.
Natural examples are zircon (ZrSiO4), where U and Th atoms sit on Zr sites and their α-radiation destroys the
mineral’s crystallinity. Metamict materials form also by radiation in nuclear reactors and particle accelerators.
Space travel requires materials resistant to metamictization.
 15.4 Optical Properties and Refractive Index 625

15.3 Oxide Glasses

The best-known example of a technologically important amorphous material is silica-based
window glass, which is made by fusing silica sand (SiO2), soda ash (Na2CO3), and limestone
(CaCO3) in an approximate molar ratio of 0.7:0.15:0.15. After mixing, the raw materials are
heated to 1500 °C for chemical homogenization. The temperature is then lowered to 1200 °C
and the melt is poured onto molten tin, forming a floating ribbon with a smooth surface.
During this step, the temperature is lowered to about 600 °C, and the glass is transferred to
rollers and passed through annealing furnaces while continuing to cool. After exiting the
annealing furnace, the glass is cut into its desired dimensions.
In general, oxide glasses are made from network-forming inorganic oxides SiO2, B2O3, or
P2O5, to which network-modifying oxides such as alkali- and alkaline-earth-metal oxides are
added. Empirical rules (Zachariasen rules) for oxide-glass formation are: (1) the coordin-
ation number of the network-former atom is 3 or 4, (2) no oxygen atom links more than two
polyhedra, (3) coordination polyhedra share only corners and not edges, and (4) for 3D
networks at least three polyhedral corners are shared. Rules (2) and (3) facilitate construc-
tion of a network without long-range order; the corner-sharing allows variable angles of
the M–O–M bonds, and these linkages can be randomly converted into terminal anionic
oxygens that bond to cations originating from the network-modifying oxide.
In the following, we will focus on SiO2-based glasses that, being based on a 3D network of
corner-sharing SiO4 tetrahedra, obey Zachariasen’s rules. While pure SiO2 glass only forms
on rapid cooling, adding alkali-metal or alkaline-earth network modifiers in the form of
carbonates or nitrates12 increases the likelihood of glass formation. Cations, such as Li+,
Na+, K+, Sr2+, Ca2+, decrease the viscosity of the melt by orders of magnitude and reduce the
melting temperature and the glass-forming temperature, Tg. The density, electrical conduct-
ivity, coefficient of thermal expansion, and refractive index of the glass are also influenced.
We will touch on all of these properties in this chapter. Network modifiers also influence
the mechanical strength of glasses, as shown in Box 15.1. The coordination requirements
of alkali-metal and alkaline-earth cations underlie their role as network modifiers. They
alter the tetrahedral SiO4/2 network such that the Si–O–Si bridging oxygen is converted into
R3Si–O− units with non-bridging oxygen via R3Si–O–SiR3 + O2− → 2R3Si–O− (where R
stands for tetrahedral silicate units that make up the rest of the network and O2− comes from,
for example, Na2O). The negative charge is compensated by the cations.

15.4 Optical Properties and Refractive Index

The technologically most important uses of glasses are, of course, derived from their optical
properties. These allow us to build microscope and telescope mirrors, lenses, optical fibers,

12
These decompose to oxides at the high temperature of the melt.
626 Amorphous and Disordered Materials

Box 15.1 Materials Spotlight: Strengthening glass—the serendipitous road to

Corning’s Gorilla® Glass
In 1952, Don Stookey at Corning Glass Works made an error when controlling the tempera-
ture of a furnace, which resulted in the serendipitous discovery of the first synthetic lithium-
silicate glass ceramic [11]. It was lighter than aluminum metal, harder than high-carbon steel,
and many times stronger than regular window glass. It was later named Pyroceram® and used
for chemical glassware and cookware for microwave ovens. Another unexpected property was
that this glass bounced without breaking when dropped on the floor. Corning subsequently
launched “Project Muscle”, exploring ways to strengthen glass. This led to a material called
Chemcor®, which could be bent and twisted considerably before it broke. It could also
withstand pressures of 690 MPa—over 10 times as much as normal glass.
One way of strengthening common glass is by careful thermal treatment known as tempering.
The glass is initially poured into a form of desired shape and thickness. It is then reheated until
it softens and rapidly cooled by forced air so that its surface cools much faster than the interior.
Glass forms first in the cool surface region. Its faster thermal contraction compared to the
interior means it forms under tensile stress due to the counterforce of the warmer, higher-
volume interior still above Tg. At this stage, the interior itself is under compressive stress. When
the interior of the sample eventually also cools below Tg and contracts, it will experience a
tensile stress from the frozen outer layers, which themselves then experience compressive stress.
The final frozen-in-stress profile of a tempered glass leaves the surface in compressive and the
core in tensile stress. Compressive surface stresses make tempered glass much more mechanic-
ally resilient than normal glass. A mechanically induced defect can be contained in the
compressed outer layer by blocking paths along which cracks could otherwise propagate; the
sample only breaks once this layer is penetrated.13 Tempered glass should have a minimum
compressive stress near 70 MPa, while in safety glass it should exceed 100 MPa.
However, it was chemical strengthening that gave Chemcor® its superior performance. The
glass is dipped into a 450 °C bath containing potassium salts, typically KNO3. By diffusion, one
partially replaces Na+ (Shannon [12] ionic radius 99 pm) by larger K+ (137 pm) in the near-
surface region, forming a layer with inner compressive forces close to 700 MPa, significantly
stronger than the ones created by thermal tempering.
Applications for car and airplane windscreens, phone booths, prison windows, and eye-
glasses were envisioned, but these products never came to market due to too high a price.
Corning discontinued the production of Chemcor®, and “Project Muscle” was shelved in 1971.
However, about 35 years later, global demand for smart-phone screens led to chemical
strengthening becoming the foundation for Corning’s “Gorilla Glass®”. While relying on a
similar strategy for chemical strengthening, Gorilla Glass® has a different composition than
Chemcor®, allowing its viscosity to be tailored for the specific manufacturing process needed to
produce very thin glass sheets. In this process, called fusion-draw, molten glass is poured into a
trough and allowed to overflow on both sides, rejoin underneath and be drawn down by rollers

13
This can be observed in the mechanical stability found in Prince Rupert’s drops made by dropping molten glass into
ice water. Videos can be found online.
 15.4 Optical Properties and Refractive Index 627

Box 15.1 (cont.)

to continuous sheets. The faster the rollers draw the melt, the thinner the glass. The specifica-
tions for Apple’s iPhone® released in February 2007 demanded massive amounts of chemically
strengthened glass with a thickness of only 1.33 mm. Corning was able to meet the specifica-
tions by May of the same year! Gorilla Glass® 3 was the result of a compositional modeling
approach and resulted in improved damage resistance [13]. Gorilla Glass® 6, which is only
0.4 mm thick, survives on average 15 drops from a 1 m height. It is almost certain that you have
touched Gorilla Glass®.

eyewear, and many other components of optical devices. One of the key features controlling
the optical properties of a material is its refractive index, n, which relates the velocity of light
inside a material, v, to that in vacuum, c (see Appendix J for the value of c):
n = c/v (15.1)

While n for a vacuum is 1, most materials have n > 1, and typical optical glasses have n ≈ 1.5,
meaning that light travels at about two-thirds of its speed in vacuum. The refractive index
depends on the wavelength of light, but it is conventionally specified at 589 nm.14
Let’s consider what happens when light passes from one medium to another (Figure 15.3).
The angles of incidence and refraction are measured with respect to the normal of the
interface plane of the two media, 1 and 2. The ratio of the angles in medium 1 and medium
2 is given by Snell’s law:

sinθ1 v1 n2
¼ ¼ (15.2)
sinθ2 v2 n1

In the cases depicted in Figure 15.3, n2 > n1 and therefore v2 < v1 and θ2 < θ1. This means
that light rays in medium 2 of higher refractive index n travel closer to the normal of the
interface. An example would be light passing from air to water of higher refractive index, as
shown in Figure 15.3a. This refraction explains why a straight rod appears broken when
dipped in water.15 In the opposite case, shown in Figure 15.3b, light is transmitted
from water to air of lower refractive index, and Snell’s law no longer holds for all angles.
The larger angle of refraction (θ1 in Figure 15.3b) reaches 90° at the critical incidence angle
θcritical = arcsin(n2/n1), beyond which the light is no longer refracted but reflected. This
phenomenon, called total internal reflection, is illustrated in Figure 15.3c. In this case, light

14
The well-known yellow doublet in the sodium emission spectrum.
15
For the same reason the Global Positioning System (GPS) has to take into account the delay of the radio signal due
to the refractive index of the Earth’s atmosphere to achieve high accuracy. The refractive index of air at 0 °C and 1
atm pressure is 1.000293.
628 Amorphous and Disordered Materials

θ1 θ1
n1 n1 n1

n2 n2 n2

θ2 θ2 θcrit

(a) (b) (c)

Figure 15.3 The refraction of light passing between two media. Medium 1 has a lower refractive index
than medium 2. The angles are defined towards the normal of the media interface. Above a critical angle,
passing from medium 2 to 1 results in total internal reflection (indicated by dashed lines).

propagates only in the medium with the higher refractive index. At a water–air interface, this
critical incidence angle is 48.6°. This effect is particularly important for the transport of laser
light through optical fibers (see Section 15.5 and Figure 15.4). A very special case occurs for
X-rays with wavelengths near an atom’s electronic absorption edge, where materials can
have n < 1. This means that total reflection of the X-ray beam can occur outside such a
material, a total external reflection. A negative n can even be created in some hybrid
materials, leading to the fascinating possibility of “cloaking devices” that hide objects [14].
To understand these effects better, we have to know something about how light interacts with
matter. In the expression for refractive index of Equation (15.1), v is actually the phase velocity
of the light in the material, v = c/n,16 which describes how fast the crests of the wave move. As an
electromagnetic wave enters a material, its electrical field will interact with electrons. The
strength of this interaction depends on the electric susceptibility χe (Section 8.1.1), which
describes the coupling to the bulk polarization P (the average dipole moment per unit volume,
Section 8.1.2). The electromagnetic wave causes the dipoles in the material to emit an electro-
magnetic wave with the same frequency as the original, but a different phase φ. The overall wave
traveling in the material becomes a superposition of the original and emitted wave. This is an
important concept in nonlinear optical materials and was discussed in detail in Section 8.7.
The refractive index is properly described as a complex quantity n = nr + iκ, where the real
part of the refractive index, nr, refers to the actual phase velocity and the imaginary part, κ,
describes the attenuation of the light’s intensity. Depending on the phase difference Δφ
between the original and emitted wave, different physical effects are observed. In most
materials, Δφ is between 90° and 180°, and the overall wave will travel slower than the
original, yielding “normal refraction” with n > 1 (e.g. the case in Figure 15.3). If Δφ is 180°, a
destructive interference of the waves leads to an imaginary refractive index that describes
light absorption in opaque materials. There are two less common but interesting cases to
consider: (1) If Δφ is 270°, the wave will travel faster than the original. This “anomalous
refraction” is possible as the phase velocity can be faster than the speed of light in a vacuum
without violating the theory of relativity. An example of this is given by the total external

16
The wavelength in a material is then λ = λ0/n with λ0 being the wavelength in a vacuum. The light frequency in the
material, ν = v/λ, remains therefore as in a vacuum and is not dependent on the refractive index.
 15.4 Optical Properties and Refractive Index 629

reflection discussed above: 0.4 Å wavelength X-rays have a refractive index of 0.99999974
(i.e. n < 1) in water. (2) If Δφ is 0°, we observe constructive wave interferences that will
amplify the light intensity. This occurs in lasers during stimulated emission.
We can link the refractive index of a material back to chemical composition, as discussed
in Chapter 8, where we related the optical dielectric constant εopt to the refractive index
n through Equation (8.14):
n2 ¼ εopt (15.3)

The refractive index n and polarizability α are related through the Lorenz–Lorentz relation-
ship we encountered in Equation (8.13) and can now write as:
 2 
3 n 1
αLL ¼ Va 2 (15.4)
4π n þ2

in the CGSes unit system with αLL in Å3 referring to the volume Va in Å3 per atom (for
elements) or formula unit (for compounds), as determined by dividing the unit-cell volume
by the number of formula units it contains. One can rearrange the Lorenz–Lorentz
equation to calculate n from αLL. In this approach, the overall polarizability of the
material, αLL, is estimated by assuming polarizabilities of the individual ions to be additive
[see Equation (8.19) for the additivity rule]. Cation polarizabilities, which are assumed to
be transferrable from one compound to another, were evaluated by Shannon and Fischer
[15]. In contrast, anion polarizabilities show considerable variability from one compound
to another. They depend on both coordination number and cation–anion distances and
correlate strongly with the volume they occupy. Defining Van as the volume of the formula
unit divided by the number of anions, Shannon and Fischer [15] parameterized anion
polarizabilities as:17
1:2
αan ¼ α0 10N0 =ðVan Þ (15.5)

where α0 (the polarizability of the free, non-coordinated anion) and N018 are empirical
parameters that were obtained by fitting a regression model to αan from ~2600
refractive-index measurements of minerals and synthetic compounds [15]. Selected
pairs for α0 and N0 are 1.79 Å3 and 1.776 Å3 for O2−, 0.82 Å3 and 3.00 Å3 for F−,
3.99 Å3 and 1.800 Å3 for Cl−, and 1.62 Å3 and 0.00 Å3 for bound H2O in hydrated
materials.
The accuracy of refractive-index calculations with Equation (15.4) depends on the degree
of covalent character in the ionic bonds. To parametrize this effect, we can correlate α and n
using:

17
In a 2006 paper (referenced in [15]), Shannon and Fischer used ⅔ as exponent. However, 1.2 yields smaller
deviations between observed and calculated total polarizabilities and is therefore used in [15].
18
These calculations are for polarizabilities at an infinite wavelength. For details, see [15].
630 Amorphous and Disordered Materials

ðn2  1ÞVa
α¼   (15.6)
4π
4π þ  Zeo ðn2  1Þ
3

where Zeo is an adjustable parameter with values between 0 for a hypothetical purely
ionic case described by the Lorenz–Lorentz equation (15.4) and 4π/3 for the limiting
covalent case that yields the Drude equation, αDr = (n2 − 1)Va/4π. With Zeo = 2.26,
Equation (15.6) is called the Anderson–Eggleton equation [16, 17]. It works well for
network oxides and silicates, but less so for compounds containing individual poly-
atomic anions such as carbonates, nitrates, sulfates, and perchlorates. By rearranging
Equation (15.6) with Zeo = 2.26, the Anderson–Eggleton refractive index, nAE, can be
calculated:19
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
u 4παAE
nAE ¼ u  þ1 (15.7)
t 4π
2:26  αAE þ Va
3

Using this approach, predictions with errors of 1–2% can be made for oxides. For
compounds containing edge- and corner-sharing transition-metal coordination poly-
hedra, predicted values are typically up to 3% smaller than experimental ones.
Furthermore, cations with a lone pair of electrons (e.g. Tl+, Sn2+, Pb2+, Bi3+) do
not permit simple addition of their polarizabilities. The refractive index of the mineral
orthoclase (KAlSi3O8) is explored in Problem 15.9. When attempting to use a similar
approach for glasses, we are faced with the challenge of determining the volume per
formula unit, Va, of a material that has no unit cell. Given the density, ρ, and
empirical formula of a glass, one can calculate the number of formula units in a
bulk sample and then divide the volume by that number to obtain Va. An example is
given in Problem 15.11.
The Gladstone–Dale relationship, originally used to relate the density ρ in g/cm3 of a
liquid to its refractive index nGD, is yet another approach to estimate the refractive index of
a glass:
X
nGD ¼ 1 þ ρ ki pi (15.8)
i

where ki (listed in Table 15.1) is the chemical refractivity20 (in cm3/g) and pi is the mass
fraction of each oxide component of the glass [18]. See Problem 15.10 for an example of how
Equation (15.8) can be used to estimate the refractive index.

19
This approach is one of the methods used in the program POLARIO to calculate refractive indices; another is based
on a modern adaptation of the approach of Gladstone and Dale.
20
These are sometimes also called specific refractive energies of the components.
 15.5 Optical Fibers 631

Table 15.1 Selected chemical refractivities ki in cm3/g for common

oxides used in glasses. Values are from Table 3 in ref. [18].

Li2O 0.307 MgO 0.200 B2O3 0.215

Na2O 0.190 CaO 0.210 Al2O3 0.207
K2O 0.196 SrO 0.145 P2O5 0.176
Rb2O 0.128 BaO 0.128 As2O5 0.340
Cs2O 0.119 PbO 0.133 SiO2 0.208

15.5 Optical Fibers

The insulating network-forming compounds SiO2, B2O3, and P2O5 have large electronic band
gaps (Chapter 6), and this makes them transparent to visible and near-ultraviolet (UV) light
(Chapter 7). When network-modifying oxides such as Na2O are added to form a glass,
different local bonding configurations and electronic energy levels are created. For example,
a bridging Si–O–Si unit will have lower-energy bonding orbitals than a non-bridging Si–O
unit, and the energy of these orbitals will be further influenced by their proximity to other non-
bridging or bridging units. As a result of this variability, a range of different electronic energy
levels will be present in glasses. This energy distribution leads to an increase in UV absorption
as more Na2O is added, and is referred to as a UV tail.21 At low energies, an infrared (IR)
absorption band peaks around 1385 nm due to hydroxyl groups bonded to the silicon.22 These
originate from the presence of minute quantities of water during glass manufacturing. The tails
of these two absorption maxima bracket a window of low optical absorption between ~650 nm
and ~1350 nm. It is also important to consider attenuation due to light scattering by inhomo-
geneities (impurities or density fluctuations) in a sample. Another important contribution to
attenuation is Rayleigh scattering. It depends on λ−4 and on Δni, the latter being the refractive-
index change induced by doping silica with a dopant i. For fiber-optics communication, we
therefore use IR light at 850 nm and 1300 nm within the low-loss window or at 1550 nm
beyond the strongest IR band (Figure 15.4).23 These wavelengths allow the transmission of
laser beams in glass fibers over thousands of kilometers with minimal losses by the process of
internal reflection (Figure 15.3). This concept was the technological foundation of the rapid
expansion of telecommunications at the end of the twentieth century. In 2009, Charles Kao
received the Nobel Prize in Physics for his work on fiber optics.
Transmission under internal reflection above the critical angle (see Figure 15.3) is made
possible by cladding a glass with lower refractive index around a core fiber with a higher

21
These localized tails in the energy gap between the valence and conduction band are called Urbach tails and are
discussed in Section 15.10.
22
The “water peak” typically shows five main absorption bands at 1247 nm, 1352 nm, 1381 nm, 1391 nm, and 1412 nm.
23
Multimode graded fibers are optimized for 850 nm and 1300 nm, while single-mode fibers are optimized for 1300 nm
and 1550 nm. Transmitters for these wavelengths are lasers and light-emitting diodes.
632 Amorphous and Disordered Materials

1000000

attenuation (db/km)
10000 cladding with low
refractive index

850 nm 1550 nm
1300 nm
100
total core with high
refractive index

1
UV edge

vibraonal
water peaks
losses
0.01
200 600 1000 1400 1800
wavelength (nm)

Figure 15.4 Contributions to attenuation in a typical glass fiber. Shape of water peaks taken from ref.
[19]. Light with wavelengths of 850 nm, 1300 nm, and 1550 nm is transmitted with low attenuation.

refractive index.24 Ultra-pure amorphous SiO2 (n = 1.458) is used as the cladding material.
The glass of the core needs to have a refractive index higher than SiO2, and this is achieved by
mixing SiO2 with GeO2, P2O5, and sometimes Al2O3 of refractive indices 1.609, 1.490, and
1.768, respectively [20]. Additionally, a gradient of the refractive index is introduced in the
core to increase the amount of data that can be transmitted by overlapping a number of
optical signals in one optical fiber using different wavelengths of laser light (multiplexing).
This also allows communication in both directions of a fiber.
Attenuation α in optical fibers is measured in decibels (dB) per kilometer as
α ¼ ð10=LÞlogðPin =Pout Þ, where Pin (in mW) is the electrical power of the signal needed on
the input so that the light transmits Pout = 1 mW at the output distance L in kilometers. The
initial target for a material useful for telecommunications was 20 dB/km, even though this
means that just 1% of the input optical power has been transmitted. Modern fibers have
much lower optical losses (near 0.1 dB/km), allowing the use of extremely long fiber cables
[21]. On land, uninterrupted fiber lengths are of the order of 10 km. “Repeater” stations are
placed between individual fiber segments that amplify and recondition the light signal.
Initially, this was done by converting the optical signal into an electrical one and back.
Nowadays, erbium-doped fiber amplifiers are used as a gain medium and “pumped” by a
laser with a wavelength of either 980 nm or 1480 nm to produce stimulated emission near
1550 nm.
Low attenuation demands extreme purity. Over a 1 km length, 90% of the transmission can
be lost by a metal ion impurity concentration of just 0.01 ppm. A second source of attenu-
ation is due to small but unavoidable surface roughness and inhomogeneity, which reflects
light in random directions. The production of optical fibers is therefore an extreme challenge.
A hollow silica-glass tube (a “preform”) is made and coated inside with high-purity oxide
24
In practice, refractive index differences are about 1%.
 15.6 Nucleation and Growth 633

glass rod

metal-halide vapor
rotate
sintered glass layer deposit

silica tube fixed heater

moving heater

pulling direction

Figure 15.5 Optical-fiber growth using CVD in a tubular SiO2 glass preform that is continuously rotated
while layers with varying refractive index are deposited. An optical fiber is pulled vertically in a drawing
tower.

glass using chemical vapor deposition (CVD) from a moderate-temperature flowing gas of
SiCl4 mixed with GeCl4 or AlCl3 or POCl3, as well as fluorine sources, SiF4 or SF6, and
oxygen. In single-clad fibers, the core has a higher refractive index than the silica cladding,
and this is achieved by Ge-, P-, and Al-doping in the core. F-incorporation decreases the
refractive index, which is needed in more complex fibers. P-containing glass has a lower
viscosity, and this becomes important during fiber pulling. Subsequent layer-by-layer deposi-
tions with varying gas compositions allow a precise control of the refractive index and
viscosity. Then an external torch heats the tube up to near 1900 K, sintering the oxide
glass on its inner walls and collapsing the tube into a rod. Afterwards, the glass rod is either
jacketed by an additional silica tube or directly heated locally in a drawing tower to near
2200 K where the tip of the preform melts and is rapidly stretched into a cooler area, forming
an optical fiber with a typical diameter of 125 μm (Figure 15.5).

15.6 Nucleation and Growth

Before discussing the topic of glass formation, it is useful to consider some of the concepts in
the alternative process of crystallization. The two key concepts are nucleation and growth.
We can understand nucleation as a first-order phase transition, in which a new phase with a
lower free energy emerges inside a large volume of the initial phase after having overcome an
energy barrier associated with the formation of small crystallites. Growth describes the
evolution of these nuclei to a bulk phase.
Let’s consider crystallization from a melt. While the bulk of the new solid phase that
emerges during crystallization will have a lower free energy than the liquid phase, its surface
region, where atoms are underbonded, will be less stable. This energy difference between the
634 Amorphous and Disordered Materials

bulk and surface, called interfacial free energy, is always positive and destabilizes newly
formed small crystallites where the fraction of surface versus bulk is the largest. Small
crystallites are therefore unstable and tend to re-dissolve. Larger crystallites, which initially
form in small numbers during a nucleation process, will have lower free energy and their
growth becomes favored. This is called the Gibbs–Thomson effect, and a critical nucleus size
can be defined, beyond which crystallization of the new phase occurs. The smaller the
interfacial energy, the smaller the critical nucleus size, and the higher the probability of
crystallization.
There are two limiting types of nucleation—homogeneous and heterogeneous nucleation. In
the former, we imagine that a spherical nucleus with radius r precipitates from the melt. In
the latter, we imagine a hemisphere of radius r forms on a solid substrate that could be an
impurity, a seed crystal, or the walls of the crystallization vessel. We would then have to
consider two additional interfacial energies, one between the substrate and crystallite, and
one between the liquid and substrate.
Normally, a melt will crystallize at its melting point, Tm, either spontaneously through
homogeneous nucleation or via heterogeneous nucleation. Under very clean conditions, the
absence of nucleation sites may allow the melt to be cooled considerably below Tm in a
process called supercooling. At Tm, the free energies of the liquid melt and the solid are equal.
At T > Tm, the melt is stable; at T < Tm, the solid. A supercooled liquid is therefore a
metastable state that is out of chemical equilibrium. Water exemplifies this behavior. Pristine
liquid water of Tm = 273 K can be supercooled to 232 K in the absence of any crystallization
centers. Under the slightest disturbance, or on further cooling, the supercooled water
crystallizes rapidly.
The goal in glass formation is to avoid crystallization, that is to cool the sample in such a
way that nucleation and growth don’t occur. The rate of both these crystallization steps is
strongly temperature dependent. For example, close to Tm, the critical size for the nucleus to
be stable is rather large, and most of the nuclei that form remelt. At lower temperatures, the
critical radius is small and remelting is less likely. However, as the temperature is lowered, the
viscosity of the liquid increases, and this slows down atomic transport and therefore nucle-
ation. Similar considerations apply to growth. Glasses therefore form if a melt is cooled
rapidly to a sufficiently low temperature so that neither nucleation or growth can occur.

15.7 The Glass Transition

In this section, we consider how kinetics and thermodynamics impact the formation of
glasses, and we will see close analogies to the supercooled liquids just discussed. The most
direct way to understand the glass transition is by comparing it with its alternative—
crystallization (Figure 15.6). In the initial high-temperature liquid, the atomic arrangement
at any point is in thermal equilibrium. Crystallization is a first-order phase transition
(Chapter 4) manifested by a discontinuous decrease in the entropy of the crystallizing
 15.7 The Glass Transition 635

(a) cooling the liquid (b) heating the glass

volume
volume

heat slower
than cooled

fast cool
fast cooled
overshoot
glass
med cooled heat faster
than cooled
slow cool slow cooled

Tg(slow) Tg(fast) Tm
temperature temperature
Figure 15.6 (a) Volume changes upon cooling a melt as a function of temperature and time at constant
pressure. Tg is the glass-transition temperature at which a supercooled melt becomes a glass at a given rate
of cooling. Glass formed at Tg(fast) was cooled faster than glass formed at Tg(slow). (b) Volume pathways on
heating and cooling differ.

substance upon cooling through its Tm, which releases heat Hm = TmSm that is absorbed by
the cooler surroundings. This is accompanied by a change in volume (as shown in Figure
15.6) and enthalpy. If crystallization is avoided, the system changes continuously into the
supercooled liquid regime with the viscosity gradually increasing on cooling (Section 15.8).
At some point, the atoms no longer have sufficient time to fully rearrange to the (metastable)
equilibrium liquid configuration for their temperature at the cooling rate used. When even
lower temperatures are reached, the structure becomes frozen on the timescale of the
experiment and a glass is formed. As this process occurs, the gradient of volume (or enthalpy)
versus temperature changes gradually until at low temperature it becomes determined by the
thermal expansion (or heat capacity) of the glass.25 The temperature range over which the
changes occur is called the glass-transformation range. It is often useful to quote a single
representative temperature for this process. One way is to define the glass-forming tempera-
ture, Tg, from the intersection of the low- and high-temperature extrapolated behaviors
measured at a specified cooling rate.26

25
We therefore see corresponding changes in heat capacity and thermal expansion at the glass transition. There are
similarities to the second-order transitions of Chapter 4, but the system is not at equilibrium.
26
A second temperature related to glass formation, the fictive temperature, Tf, is often encountered in the
literature. It is obtained just like Tg but upon heating the glass. If a sample is cooled then heated at identical
rates, Tf and Tg will be essentially identical. If a glass is thermally annealed below Tm, the fictive temperature
will change as discussed in Box 15.2. Note that different communities use slightly different definitions for Tf
and Tg.
636 Amorphous and Disordered Materials

It should be clear from this discussion that Tg for a given liquid will depend on the rate of
cooling. If a supercooled liquid is cooled more slowly, the time available for structural
rearrangement at a given temperature will be longer, and the volume of the sample will
follow the trend of the supercooled liquid to lower temperatures. Tg for this slow-cooled
liquid will therefore be lower than for a faster-cooled liquid. The non-equilibrium nature of
the glass transition is also observed when glasses are reheated, and this behavior is explored
in Box 15.2 on hyper-aged glasses. For example, if we reheat a glass at the same rate at which
it was originally cooled, it will follow the same volume (or enthalpy) curve. If it is heated at a
faster rate, the low atomic mobility in the glassy state means that it will overshoot the
extrapolated liquid curve and remain a glass until higher temperatures than when it was
cooled (Figure 15.6b). In contrast, if it is heated more slowly than it was cooled, the glass can
start to rearrange on the timescale of the experiment as its original Tg is approached, leading
to a volume curve higher than that on cooling.
Glass-forming temperatures of real materials on cooling vary enormously: SiO2 has a Tg ≈
1200 °C, the metallic glass Pd0.4Ni0.4P0.2 has a Tg ≈ 580 °C, atactic polypropylene has a Tg ≈
−20 °C, and low-density polyethylene has a Tg ≈ −125 °C. Good glass formers (e.g. SiO2,
GeO2, Se) can be made by melt quenching and typically obey the Kauzmann rule,
Tg ¼ 0:7Tm . Poor glass formers (Tg < 0:7Tm ), like metal alloys, have to be cooled very
quickly to prevent crystallization.
To complicate matters further, deciding whether an actual sample is a glass, supercooled
liquid, or liquid depends on the experimental probe used. In rheology, these properties are
distinguished by the Deborah number, De = τc/τp, where τc is the timescale of the viscosity
(time needed for a reference amount of viscous flow) and τp is the timescale of the experiment
used to probe it. Whether a material responds like a liquid (low De) or a solid (high De)
depends on the experimental timescale as well as the intrinsic properties of the material. As
an example, one can happily swim slowly through water without bodily damage; the picture
is very different if you hit water at high speed – τp, and therefore De, is different in these two
“experiments”. For similar reasons, a high-frequency (1015 Hz) technique like NMR will
“see” the glass transition at a different temperature than a lower frequency technique such as
calorimetry.
At De ≫ 1, the system has insufficient time to explore all possible structural configurations
(Section 15.1), as the probing timescale τp is too small. Glass is such a system and is called
non-ergodic. In contrast, a melt is an ergodic system (same homogeneous behavior across the
system over time) with De ≪ 1, as it has sufficient time to equilibrate and probe all possible
structural configurations. In an ergodic system, the configurational- and time-averages of
properties such as viscosity are the same. In non-ergodic glasses, frozen-in configurations
might average out over time if one probes at a longer timescale. At the glass-forming
transition, ergodicity is lost, and the system can only experience a subset of configurations.
To help understand the interplay between structural configurations and their stability, it is
useful to introduce the potential-energy surface (PES) or energy landscape. This is a
 15.7 The Glass Transition 637

transition states
melt

potential energy
C
B
basins

A
crystal

configurational coordinates

Figure 15.7 The PES of an N-body system separating the melt from various glass states, as well as the
crystalline state as the lowest-energy configuration. The glass states A, B, C and D are discussed in
Materials Spotlight 15.2.

representation of how the energy of an N-body system depends on its structural configur-
ations. Any atomic structural configuration can be described by a multidimensional vector
that contains the coordinates defining interatomic distances and angles of its atoms. For
illustrative simplicity, we collapse all these configurational coordinates27 onto one dimension
of the horizontal axis in Figure 15.7, with the potential energy separating melt from glass or
crystal on the vertical axis.
The PES captures the potential-energy states a material can access depending on its
thermal history. At high temperatures, the system has sufficient internal energy to explore
the whole energy landscape, and the dynamics resembles free atomic diffusion. Such an
ergodic system with De ≪ 1 has sufficient time to equilibrate. At lower temperatures, the
system no longer has the energy to overcome the highest energy barriers and is then forced to
adopt a minimum close to its current configuration. Its accessible configurational space is
reduced or constrained. A system trapped in local low-energy arrangements (e.g. points C or
D in Figure 15.7) is therefore unable to reach the thermodynamically more favorable global
structural configurations of the glass or crystal because the diffusion is too slow. The kinetics
of structural relaxation changes with temperature, and the probability that a certain activa-
tion barrier can be overcome becomes smaller with decreasing temperature; a regime called
super-Arrhenius (Section 15.8). The PES also provides us with an intuitive visualization of
why cooling rates are so important in glass transitions. Fast cooling rates will not allow the
system to adequately probe many configurations before leaving the liquid state and forming
a glass. Aspects of the liquid’s structure then appear frozen in the glass. Slower cooling rates
27
Whereas in PES diagrams and chemical reactions one uses 1D collective configurational coordinates describing the
complete chemical system, in Section 7.8.2 we used configurational coordinates to describe the differences between
electronic ground and excited states of localized areas near emission activators.
638 Amorphous and Disordered Materials

allow enough time to probe the rarer deeper minima (e.g. A and B) in the potential-energy
landscape. The result is a more stable glass, as more configurations are explored, and a lower
Tg.28 That more stable glass then melts at a higher temperature than the less stable glass
formed at points C or D.
Another temperature is occasionally encountered in the literature on glasses—the
Kauzmann temperature, TK. It is the temperature below which a supercooled metastable
liquid would have an entropy lower than the equilibrium crystal at the same temperature and
pressure, were it not for the intervention of a glass transition. Such a situation, known as the
Kauzmann paradox, is an interesting but problematic concept. Modern theories [22] postu-
late a temperature below which an equilibrium supercooled liquid no longer exists, which
invalidates the relevance of the Kauzmann paradox.

Box 15.2 Materials Spotlight: Amber—thermal history of a hyper-aged organic glass

Plant resins can polymerize over hundreds of millions of years and form organic glasses called
amber, which often have a beautiful yellow-brown color leading to their use in jewelry [23].
Amber is different to other materials discussed in this chapter as the glass forms due to chemical
and not physical (e.g. by cooling or applying pressure) processes. Ancient amber presents us
with a unique opportunity to investigate a glass that has aged significantly beyond what is
possible in a laboratory setting. Ambers are hyper-aged organic glasses.
In Figure 15.7, we showed a PES with distinct local minima for different glassy states denoted A,
B, C, and D. In Section 15.7, glass formation is discussed as a process of the melt exploring
different parts of the PES as we cool. Slow cooling allows more of the PES to be explored, leading
to a more stable glass. With the same PES, we can also understand the behavior of amber upon
heating, even though it formed by aging over time rather than by cooling. A possible location for
the potential energy of hyper-aged (110-million-year-old) amber would be in a very deep basin such
as A. If this amber is heated above its glass-melting temperature heating Tg of 438 K during a heat
capacity (Cp) measurement, an endothermic peak is observed [24]. This represents the heat required
to dislodge structural entities from the glassy environments in which they settled after a hundred
million years of structural relaxation close to ambient conditions. In the PES diagram, a sample
that’s heated above heatingTg and cooled down again on a normal laboratory timescale might find
itself located in basin D. One can quantify the stability difference between A and D by measuring
the endotherm on melting amber D, which will be much smaller. Furthermore, one can investigate
what happens to the size of the melting endotherm when amber A is annealed at different
temperatures below heatingTg, resulting in the formation of glasses located in basins B and C of
the PES. If one gradually increases the annealing temperature towards heating Tg, the amber
becomes progressively less stable and the endothermic peak in the Cp measurement smaller and
smaller. The materials located in basins B and C are referred to as partially rejuvenated ambers.

28
Tg typically changes up to 5 °C as the cooling rate is changed by an order of magnitude.
 15.8 Strong and Fragile Behavior of Liquids and Melts 639

Box 15.2 (cont.)

Heating amber tells us about the evolution of such a hyper-aged glass as it escapes the constraints
of its local PES landscape and transitions towards free diffusion in the melt.
Heating amber thus erases the history of hundreds of millions of years of structural relax-
ation. But not all is erased. Remarkably, the study also showed that the specific heat at
temperatures below 1 K is the same in hyper-aged, partially rejuvenated, and fully annealed
amber; they fall on the same curve and show that Cp is proportional to T 1.27. This supports the
conclusion that the tunneling two-level system we will discuss in Section 15.9 is an intrinsic and
universal low-temperature dynamical property of glasses irrespective of their history.

15.8 Strong and Fragile Behavior of Liquids and Melts

According to Cohen and Turnbull [25], viscosity in glasses can be understood as a material
flow into the voids or “free volume” caused by random packing (Section 15.1). An alternative
purely empirical definition of Tg, based on the flow properties of a cooling melt, is to define Tg
as the point at which the shear viscosity η reaches 1012 Pa s.29 Shear viscosity is a measure of the
resistance of a liquid to a deformation under shear stress.30 At ambient conditions, liquids such
as acetone, methanol, and ethanol have viscosities in the range of 10−3 Pa s, while the value for
pitch is about 108 Pa s.31 One of the remarkable features of the glass transition is that by
cooling over a few hundred degrees, the shear viscosity can increase by 15 orders of magnitude.
Just above Tg, the temperature dependence of viscosity for some melts like SiO2 can be
approximated using an Arrhenius-law expression, η = η0 exp(−EA/RT), with the pre-exponen-
tial factor η0 and activation energy EA being material-specific parameters. Plotting ln(η) versus
1/T allows us to identify −EA/R as the slope and ln(η0) as the y intercept at infinitely high T.
Over a wider temperature range, however, the viscosity of many liquids and most polymers
and organic glass formers shows a significant deviation from this simple Arrhenius-type
behavior and is better described using the Vogel–Fulcher–Tammann32 equation:

η ¼ η0 exp½DT0 =ðT  T0 Þ (15.9)

29
This value is close to the annealing point of glasses where they are already hard enough to break when dropped on
the ground but still soft enough to relax internal strains by microscopic flow when modeled at the glassblowers’ pipe.
At higher viscosities, microscopic flow essentially stops.
30
Shear stress is caused by a force parallel to the surface of a sample.
31
One of the longest ever scientific experiments, the pitch-drop experiment, was started in 1927 at the University of
Queensland by Parnell. He poured heated pitch into a sealed funnel, cut the seal in 1930, and monitored the falling
drops. The ninth drop was observed in April 2014. There is now a webcam on the experiment and the tenth drop is
expected in the 2020s.
32
In polymer science, the equivalent equation is also known as the Williams–Landel–Ferry equation.
640 Amorphous and Disordered Materials

η= 1012 Pa s
gradient at Tg gives fragility index, m
11

7 )
log(η)in Pa s s ,D
uid gth
5 liq tren
ng s
rt o ility
s g
3 fra ica
gh sil iO 2
(h
i 2S
O
1 Na 2

-1 glycerol

-3 o-terphenyl

-5 η= 10−5 Pa s
0 0.2 0.4 0.6 0.8 1
(T T )
−1
cooling the liquid g
Tg

Figure 15.8 Viscosity of strong and fragile liquids versus inverse normalized temperature (an Angell plot).
Arrhenius-like behavior would lead to a straight line. Cooling occurs from left to right on the (T/Tg)−1
scale. Fragility index m is the slope of the curve at Tg.

with η0 being the viscosity at infinitely high temperatures, D a material-dependent fit

parameter for the liquid, called the fragility strength33 (meaning strength against the liquid’s
fragility to easily crystallize as opposed to form glass; there is no relation to the brittleness of
the material), and T0 the Vogel–Fulcher–Tammann temperature where kinetic barriers rise
to stop the flow.34 We can see that Equation (15.9) approaches Arrhenius-type behavior as
the temperature goes to infinity. This simple equation is capable of describing viscosity data
over 15 orders of magnitude.
Experimental viscosities can be evaluated in a so-called Angell plot of log(η) versus
the inverse normalized temperature, (T/Tg)−1 = Tg/T, as shown in Figure 15.8. The glass
transition is then formally defined to occur at 1012 Pa s, and all curves are extrapolated to an
infinite-temperature viscosity of 10−5 Pa s (a low viscosity). 35 Angell defined a fragility index,
m, as the slope of the logarithm of the viscosity curve at Tg. Figure 15.8 shows that materials
with a high fragility strength D in Equation (15.9) have a low fragility index and vice versa.
Typical values are 2 < D < 100. GeO2 and SiO2 are both strong liquids (strong glass formers)
with D of 113 and 63, respectively, whereas toluene is a fragile liquid (easy to crystallize, poor
glass former) with D ≈ 5.

33
The D equals B/T0 of the conventional Vogel–Fulcher–Tammann equation, η = η0 exp[B/(T−T0].
34
T0 is always less than Tg. As T0 approaches zero, Arrhenius behavior is approached.
35
The value 10−5 Pa s corresponds to relaxation times of 10−14 s, significantly shorter than the typical timescale (10−12 s
to 10−13 s) of one atomic vibration. Typical liquids have viscosities near 10−3 Pa s. Extrapolation to 10−5 Pa s at infinite
temperature is an arbitrary but reasonable choice.
 15.8 Strong and Fragile Behavior of Liquids and Melts 641

Silica is an example of a strong liquid (a strong glass former; D ≈ 63), and its shear viscosity is
well approximated by the Arrhenius equation. Its activation energy is almost constant over a
wide temperature range (constant slope in Figure 15.8), which indicates that the breaking and
forming of Si–O bonds controls the temperature dependence of the viscosity. Another way to
understand a strong liquid is via its PES (Figure 15.7), which is dominated by a single large
basin modulated by smaller local energy variations due to the presence of different SiO4/2
tetrahedral rings and clusters. A strong liquid has a close structural relationship to its glass and
consequently has a high tendency for glass formation. Melts leading to bulk metallic glasses
(see Section 15.11) are also typically strong liquids. They have high viscosities η, large fragility
strengths D, show near-Arrhenius behavior, and have good glass-forming abilities.
Many organic molecules such as toluene or chlorobenzene are examples of fragile liquids (poor
glass formers). Their viscosity can increase 1015 times as one cools the already supercooled liquid
(Figure 15.8) towards the glass transition. The PES (Figure 15.7) of these fragile liquids is more
complex than that of strong glasses and shows multiple relatively large basins modulated at the
bottom by local energy variations. The structural rearrangements required to transition between
different basins are complex and characterized by a broad spectrum of different bonding and
nonbonding interactions with different relaxation times. In these molecular systems, it is the
multitude of collective intermolecular interactions rather than chemical bonds that gives rise to
the more complex multi-basin PES. The flow behavior is better accounted for by the Vogel–
Tamman–Fulcher equation, as it isn’t dominated by a single chemical bond strength. This has
consequences for ionic conductivity in polymers (see Section 13.5).
High fragility (easy crystallization), manifested by a strongly non-Arrhenius behavior of
η(T), is an essential property of chalcogenide-based phase-change materials and the cause of
their rapid crystallization that is exploited in phase-change memory. For example, if an alloy
of Ge, Sb, and Te such as Ge2Sb2Te5 is heated above its melting point (Tm ≈ 600 °C) and
cooled rapidly, it forms an amorphous low-conductivity phase (see Section 15.10) that is
kinetically stable at room temperature. If this phase is subsequently heated to temperatures
above Tg but below Tm, it rapidly crystallizes to a high-conductivity crystalline phase. These
two states can be easily read out as 0 and 1 in an electronic device, leading to a non-volatile
phase-change memory [26]. It is even possible to produce intermediate conductivity states so
that each memory element can store more than one bit of information. The accompanying
change in optical properties from a low-reflectivity amorphous to a high-reflectivity crystal-
line state has also been exploited in optical storage devices such as Blu-ray disks. Research
has led to the optimization of an alloy Sc0.2Sb2Te3 with a recrystallization speed below 1 ns
that reduces the rate-limiting step of the read/write operation [27]. It appears that the role of
Sc is to seed heterogeneous nucleation, thereby accelerating crystallization.
To conclude our discussion of viscosity in amorphous materials, we will address the
popular myth that the ambient-temperature viscous flow of glass can be observed in
the windows of medieval cathedrals, since the panes are often thicker at the bottom than
at the top. Work using the glass composition found in Westminster Abbey (dated 1268),
revealed that, while the thermal history of the glass can alter the viscosity between 1024 Pa s
642 Amorphous and Disordered Materials

and 1025 Pa s, the maximum flow under a gravitational force amounts to a thickness increase
of 1 nm in 109 years. The actual observed thickness variation is due to the medieval
manufacturing methods and the way glaziers chose to mount individual panes [28].

15.9 Low-Temperature Dynamics of Amorphous Materials

Glasses have different thermal properties than crystalline materials. Heat-capacity and
thermal-conductivity measurements of glasses at temperatures below 1 K reveal they all
show the same physical behavior (see Box 15.2). This can be related to the existence of
dynamical excitations, which occur regardless of structure and composition and are there-
fore called universal low-temperature states.
In crystalline materials, we use the concept of phonons to understand the collective
thermal motion of atoms. Phonons are wave-like dynamical displacements of atoms that
occur in structures with translational symmetry and are the extended-structure analog of the
vibrational modes of isolated molecules (Section 4.4.6). Atoms collectively and concertedly
vibrate about their equilibrium positions, and the pattern of displacements can be described
by long-range waves that propagate through the entire crystal. In a perfect harmonic crystal,
the phonons are plane waves36 with infinite lifetime, and, since phonons are responsible for
heat conduction, one should observe an infinite thermal conductivity.37 In reality, crystals
are not perfect, and phonons are anharmonic, resulting in finite phonon lifetimes and finite
thermal conductivity. A good approximation for thermal conductivity is:

1
κ ¼ CV ν s ℓ p (15.10)
3

where νs is the average sound velocity in m/s, CV the molar heat capacity at constant volume
in J/(mol K), and ℓp is a parameter called the phonon mean free path in m, the average
distance between two consecutive phonon-scattering events. Impurities and lattice defects
reduce ℓp, suggesting that we might expect lower thermal conductivity in disordered
materials.
Our model for phonons is based on crystalline materials, but, since glasses have no unit
cell, no translational symmetry, and no long-range order, do they have phonons? The answer
is, yes; collective excitations we call phonons exist, but they are no longer plane waves at the
nanometer scale and, due to disorder, their lifetimes are shorter than in crystals.
Debye established [29] that for insulating crystalline materials (thus lacking mobile
conduction electrons), the heat capacity Cp at constant pressure and the thermal conductiv-
ity κ should be proportional to T 3 at low temperatures. Zeller and Pohl [30] found that the
36
A plane wave propagates with constant frequency perpendicular to 2D planes in which the wave fronts comprising
points with the same phase are located (infinite parallel planes of constant amplitude).
37
Superfluids, such as 3He below 2 K, have infinite thermal conductivity, meaning that any volume of a sample will
always have a uniform temperature; there is no resistance to heat transport.
 15.9 Low-Temperature Dynamics of Amorphous Materials 643

(W/cm K)
(W s/g K)

Figure 15.9 Comparison of specific heat and thermal conductivity of silica glass (open squares) and
crystalline quartz (filled squares) where thermal conductivity was measured along the c axis. Data from
ref. [26].

low-temperature Cp and κ of glasses are strikingly different; below 1 K, Cp is approximately

proportional to T (not T 3) and κ is approximately proportional to T 2 (not T 3). We encoun-
tered this behavior in Box 15.2, where we learned that Cp of all ambers is proportional to
T 1.27 below 1 K. The low-temperature thermal properties of crystalline α-quartz and
amorphous silica are compared in Figure 15.9 and reveal that Cp is larger, whereas κ is
lower in the glass.38 Furthermore, a plateau is observed between 3 K and 10 K in κ(T ) for
silica glass, which is due to the presence of an excess of low-frequency modes in the energy
distribution of vibrations over the predictions of the Debye model. This plateau is referred to
as the boson peak and is a universal feature of glasses (phonons are quasi-particles with zero
or integer spin and therefore bosons). These excess vibrational modes are believed to scatter
phonons, causing a precipitous reduction of their mean free path ℓp and a concomitant
temporary saturation of κ(T).39
One explanation for the universal ultra-low temperature dynamical behavior of amorph-
ous materials is given by the two-level tunneling model developed by Anderson, Halperin,
and Varma [31] and independently by Phillips [32]. In this model, localized excitations
distinct from phonons are introduced as the lowest-energy excitations of glasses. The
model assumes the presence of many local groups of atoms with nearly the same energy in
the PES. In Section 15.7, we have seen that at high temperatures, atomic relaxations between

38
The low-temperature peak in κ(T) can be understood from the temperature dependence of the three terms in
Equation (15.10). To a first approximation, ℓp increases as temperature decreases and phonon–phonon scattering is
reduced but saturates at low T due to residual scattering by defects, v is independent of T, and CV falls from a
constant value to zero. The combination of these effects leads to the observed peak.
39
Phonons are only well-defined quasi-particles when their mean free-path length ℓp is larger than the phonon half
wavelength, λ/2. This is the Ioffe–Regel criterion for phonons that explains the saturation of thermal conductivity.
In Section 15.10, it is discussed in more depth to explain saturation of electrical conductivity when the electron free-
path length approaches the interatomic distances.
644 Amorphous and Disordered Materials

structures in different local minima are thermally activated and can overcome the barriers
that separate them. At very low temperatures, the thermal energy is insufficient to overcome
the barriers. They can, however, be overcome through quantum-mechanical tunneling. With
many atoms involved in this process, the displacements of an individual atom can be tiny yet
the total result substantial. Due to quantum-mechanical tunneling, two-level-tunneling
excitations are still active even at temperatures below 1 K and contribute to Cp, causing it
to be larger than in the crystalline counterpart. Low-energy phonons with very large
wavelengths are scattered by two-level-tunneling excitations and the resulting decrease of
their mean free path ℓp reduces the thermal conductivity.
Taking two-level tunneling states into account can be important as they act as sources of
noise in electronic devices at the cryogenic temperatures explored for quantum computing
[33]. High-precision measurements at higher temperatures can also be affected by two-level-
tunneling states in materials. For example, the observation of gravitational waves is based on
laser interferometry using reflective coatings of multiple alternating layers of amorphous
silica glass and TiO2-doped Ta2O5 glass40 as mirrors. To detect metric changes of 10−18 m,
two laser beams, split at 90° to each other, are sent down 4 km vacuum tubes and reflected by
mirrors. Interference occurs when they are recombined in the presence of a gravitational
wave. The main experimental obstacle to measuring miniscule interferometric signals is the
absorption loss caused by the coating’s thermal noise. This results in part from external
friction, as the force resisting deformation in a material as stress (force per unit area) is never
completely converted into strain/deformation without losses. Two-level-tunneling states are
a significant source of internal friction and thermal noise in amorphous oxides. They have
been included in modeling internal friction to help search for better coating materials for
more precise interferometric gravitational-wave detectors [34].
In summary, the low-temperature structural and dynamic ground states that crystalline
and amorphous materials display are quite different. Crystalline materials have structures
described by unit cells and translational symmetry. Dynamically, they display a complex
range of excitations described by phonons. Glasses have complex atomic-network structures
with no translational symmetry and no unique structural ground state. In contrast to
crystals, their dynamics at ultra-low T is described by universal two-level tunneling.

15.10 Electronic Properties: Anderson Localization

Another fundamental difference between amorphous and crystalline materials concerns
electrical conductivity. We discussed in Chapter 10 how the conductivity of crystalline metals
can be understood in terms of the mean free path ℓ travelled by an electron accelerated in an
external electric field. Electrons are scattered by defects such as vacancies, interstitials,

40
Besides reducing loss due to absorption, doping Ta2O5 with TiO2 increases the refractive-index contrast versus SiO2,
enhancing the coating’s reflectivity.
 15.10 Electronic Properties: Anderson Localization 645

dislocations, and impurities, or by instantaneous local irregularities caused by phonons. In

most metals, this scattering leads to a random walk of electrons, reducing ℓ (see Figure 10.2),
and constrains conductivity to a diffusion process. The room-temperature value of ℓ for a
metal such as Al is ~29 nm, about two orders of magnitude larger than its lattice parameter.
In high-purity Al, ℓ increases to ~700 μm at low temperatures when phonon scattering is
greatly reduced (see Chapter 10).
It is worth considering two extremes of conduction-electron behavior. In the case of a perfect
crystal, scattering will be rare, and ℓ might exceed the size of the sample, particularly at low
temperatures. This leads to so-called ballistic conduction, discussed for carbon nanotubes in
Chapter 10, where the only scattering interactions are with the “imperfections” caused by the
surfaces of the sample. At the other extreme, when there is a large amount of electron scattering
due to phonons, disorder, and impurities, the path length ℓ between two collisions can approach
the interatomic distance. From Equation (10.19), we define electrical resistivity as:

1 m vF 1
ρ¼ ¼ (15.11)
σ ne2 ℓ

Since an electron can at most be scattered by every atom41 as ℓ approaches the interatomic
distance, we expect a limiting high value of ρsat , the saturation resistivity. This is called the
Ioffe–Regel condition.42
The relevant question in this chapter is how the inherent high disorder of amorphous and
glassy materials impacts electrical conductivity. Does the increasing disorder in a metallic
conductor lower the mean free path ℓ and hence conductivity smoothly? P.W. Anderson
provided the answer to this question. Surprisingly, he predicted a precipitous drop in electrical
conductivity and localization of all the electrons after a critical amount of disorder is present.
This disorder-induced metal-to-insulator transition is known as Anderson localization.43
Before we can understand Anderson localization, we have to introduce some results from
quantum mechanics. The conductivity of a material is related to the probability of an electron
moving from a point I to a point II. In a real material, there will be many different ways in
which an electron can get from I to II, two of which are shown in Figure 15.10. Electrons have
wave character described by a wavefunction with an amplitude and a phase. In our case, an
electron moving as a wave along either path a or b has the (vector) amplitude A or B and phase
φa or φb, respectively. We learned in Section 6.1.2 that the square of the wavefunction ψ2 is
called the probability density; the probability of finding an electron in a given region is equal to
ψ2 divided by the volume of the region. To calculate the probability of getting from I to II,
quantum mechanics requires us to sum up the amplitudes of the two wavefunctions taking
different paths, including an interference term that takes into account the phase difference, φ =

41
In semi-classical theories like the Drude theory, the uncertainty of the k vector of an electron wavefunction becomes
comparable to the size of the Brillouin zone when ℓ ≈ the interatomic distance. See Section 6.4.
42
Alternatively named Mott–Ioffe–Regel limit.
43
Anderson localization explains early experiments in George Feher’s group at Bell labs where phosphorous-doped Si
at 2 K had long electron-spin relaxation times (>1000 s), consistent with electron localization.
646 Amorphous and Disordered Materials

path
a

path
I b path II
a

path
b

Figure 15.10 Two paths, a and b, for the wavefunction of an electron to propagate from point I to point II
and back to I, which will result in constructive interference at point I.

φa − φb, of the two wavefunctions. If crests meet crests at II, we have constructive interference,
and this occurs when there is zero phase difference (φ = 0), hence cos φ = 1. Our probability is
given by jA þ Bj2 ¼ jAj2 þ jBj2 þ 2jAjjBj cosφ. One might expect that in materials with
random disorder, the phase difference φ (in the interval 0–180°) from many different
pathways would average to 90°, and the interference term could be ignored. However, this
is not the case. Mathematically, we can see that paths with the same phase (φ = 0) have the
largest interference term (cos φ = 1) and give the highest probability as they involve
constructive interference of the two wavefunctions. One situation in which this is guaranteed
to occur is when two waves take the same path but travel in opposite directions, as shown in
Figure 15.10. The interference term in this situation is always 1, which increases the prob-
ability that electrons will “circle back on themselves” and become localized. This is called
weak backscattering.
So how do these considerations impact the electronic structure of amorphous materials? In
Section 6.1.6, we introduced electronic density-of-states (DOS) plots and showed that the
valence and conduction bands in a semiconductor like Si are separated by an energy gap, and
that conductivity occurs when either of them is partially occupied. This is depicted in
Figure 15.11 and compared with the DOS of amorphous Si. The latter has two new features:
defects in the middle of the gap due to “dangling bonds” in the macroscopic voids of the
network (see Footnote 7 in this chapter) and regions of localized electron states at the band
edges that extend into the gap, called Urbach tails.44 The states lying near the middle of the
gap are highly localized, not unlike localized energy levels of doped activator ions that were
discussed in Section 7.8. States nearer the band edges are more extended but not over the
entire sample, as they would in a crystalline material. Electrons and holes occupying these
localized states do not contribute to the conductivity. The interface between extended and
localized wavefunctions within a band is called the mobility edge, a concept introduced by
Mott. The energy separation between the two mobility edges at the top and bottom of a
single band is called the mobility gap.

44
Exponential absorption edges are a universal property of all elemental and compound semiconductors and were
described first in 1953 by Franz Urbach.
 15.11 Metallic Glasses 647

conduction band conduction band

energy
energy
½ Emg

mobility edge
Urbach tail

localized
EF EF electrons

Urbach tail
mobility edge
valence band

valence band

N(E) N(E)

Figure 15.11 DOS of crystalline (left) and amorphous (right) semiconductors. Gray shadowing indicates
localized electronic states including those at EF which are due to “dangling bonds”. As disorder increases,
the mobility edges will grow within the bands and the mobility gap Emg between the mobility edges at the
top and bottom of a band will narrow, eventually leading to complete electron localization.

As disorder increases and backscattering becomes dominant, more electron states become
localized, and the mobility gap decreases. Above a critical disorder, all electron states in a band
will localize in what is called a Fermi glass. Such electron localization has been demonstrated in
single crystals of LixFe7Se8, where Li above a critical concentration of x = 0.53 creates a
sufficiently strong disorder to induce a metal–insulator transition and destroy both the crystallo-
graphic and magnetic order of Fe. Specific heat, electrical conductivity, as well as magnetic and
optical measurements suggest that Anderson localization is taking place [35]. Such strong
Anderson localization into a Fermi glass is a general quantum effect found also for light,
sound, and microwaves that can be stopped in media; a high level of random disorder traps the
quantum particle (or quasi particle) and prevents its random-walk migration. In 1977, Anderson
and Mott received the Nobel Prize in Physics for their work on the conductivity of disordered
materials.

15.11 Metallic Glasses

When metals or alloys cool down from the melt, they will generally solidify into the lowest
energy state, usually crystalline. This typically occurs within microseconds and leads to
polycrystalline samples with grains separated by grain boundaries and with line and planar
defects. These areas with less than optimal atomic packing easily slip past each other under
stress, leading to plastic deformation.
Although metals are generally “bad glass formers”, in 1960 Klement et al. [36] discovered
that the molten binary alloy Au75Si25 solidifies as a metallic glass if cooled at rates of at least
648 Amorphous and Disordered Materials

106 K per second. During rapid quenching, its atoms do not have sufficient time or energy to
reach a configuration suitable for crystalline nucleation (Section 15.6). Metallic glasses are
typically similar in visual appearance to crystalline metals; they are shiny and have smooth
surfaces. However, other physical properties differ significantly from those of their crystal-
line counterparts. For the reasons discussed in Sections 15.9 and 15.10, their electrical and
thermal conductivities are often orders of magnitude lower. Because of their disordered
atomic structure, shear due to dislocations is hindered. Under impact, a metallic glass
therefore absorbs less energy and shows less damping45 than a crystalline metal and is
more likely to rebound elastically to its initial shape. Some metallic glasses have strengths
twice that of stainless steel. They are more fracture-resistant than ceramics and can have very
high yield strengths. Metallic glasses also have fewer extended crystal defects and structural
inhomogeneities, where fracture, corrosion, and wear can be initiated.
In order to understand the structure of metallic glasses, we need to consider the options for
non-crystalline packing of metal atoms, and, in particular, the type of units that might be formed
when a melt is rapidly cooled. Metallic bonding is much less directional than covalent bonding,
and metal atoms can be approximated as spheres. These spheres will try to pack together
efficiently so as to maximize the metallic bonding. The smallest 3D element comprising touching
spheres is a tetrahedron. As more spheres are brought together they will touch in various
directions, forming tetrahedra in different orientations. How can these regular tetrahedra be
packed in three dimensions? The important observation is that tetrahedra can’t be packed to fill
space 100% efficiently, that is to say they don’t tessellate 3D space. This is illustrated in Figure
15.12; five regular tetrahedra each trying to share two faces form a body similar to pentagonal
bipyramid, but a gap is left of 7.4°. No such gap is possible in a real structure, so the tetrahedra
must deform slightly in order to form the pentagonal bipyramid. This inability to tessellate space
with a regular object is an example of geometric frustration. When the top five triangular faces of
the pentagonal pyramid are each “capped” with one atom to complete new tetrahedra, and

Figure 15.12 Regular pentagonal bipyramid (left) compared with a cluster of five regular tetrahedra that
leaves a gap of ~7.4° (center). An icosahedron (right) of 13 atoms is built of 20 tetrahedra having their
bases at its 20 triangular faces (icosa = 20 in Greek, hedros = bottom) and vertex at the center atom
marked with a square dot.

45
The process by which a material subjected to oscillatory deformations converts mechanical into thermal energy that
is dissipated.
 15.11 Metallic Glasses 649

when the pentagon that these five new tetrahedra form is itself capped by one atom, an
icosahedron is obtained. It is composed of 20 slightly deformed tetrahedra, the triangular
bases of which form its 20 faces, and their common vertex is at center of the icosahedron.
Analogous capping of the icosahedron’s faces into subsequent layers is possible, and this creates
series of larger and larger icosahedra, the Mackay icosahedra. Note that the icosahedron
contains fivefold axes that aren’t found in normal crystal structures.
This type of packing of spheres into random geometrically frustrated structures of deformed
tetrahedral units is what Bernal found (Section 15.1) in his wax and ball models. In addition to
being the principle behind packing of metallic glasses, it has another interesting consequence—
the formation of quasicrystals. Quasicrystals of icosahedral or fivefold symmetry were dis-
covered by Dan Shechtman in 1982. Fivefold symmetry is not one of the crystallographic
lattice symmetries (introduced in Chapter 1), and a single equilateral pentagon will not tile 2D
space. However, two tiles can be designed such that they tile 2D space in an aperiodic way
called Penrose tiling, which has a fivefold and tenfold rotational symmetry.
In the context of metallic glasses, an interesting quasicrystal is one found for a TixZryNiz
alloy that normally forms a stable crystalline Laves phase [37]. As a droplet of its TixZryNiz
melt is cooled in a differential scanning calorimeter (see Box 4.1), two phase transitions
occur. The first one corresponds to the formation of the metastable quasicrystal that, like the
melt, has local icosahedral order. The second one is the crystallization of the Laves phase
(with some intersite disorder of atoms), in which all icosahedral order is removed. This
sequence complies with the Ostwald step rule that, in general, metastable products crystallize
first, upon exploring/probing high-energy states (like in Section 15.7; see also Box 1.3). It
suggests icosahedral symmetry of the local order in metallic melts, and this has been
confirmed experimentally by electron diffraction [38].
One challenge in utilizing all properties of metallic glasses stems from the need for rapid
cooling despite their low thermal conductivity. In early examples, the required cooling rates
could only be achieved in thin films formed by casting the melt onto a spinning wheel with a
very cold surface (“splat cooling”). It was not possible to cast samples in a mold, and this
precluded many applications. To overcome this challenge, metallic glasses with slower rates
of crystallization were needed.
Early experiments led to the development of empirical rules to compositionally precondi-
tion alloys to slow crystallization, thereby increasing the probability of glass formation.
These include the following: (1) Use of multi-component alloys with three or more elements
and high structural complexity to reduce the formation enthalpy of the undesirable crystal.46
(2) Use of elements with an atomic radius mismatch greater than 12%, which leads to filling
voids, higher packing density, and smaller free volume in the molten state [39], thus requiring
an energetically unfavorable volume expansion during crystallization. (3) Use of elements
with negative enthalpy of mixing that form weaker bonds to other elements in the glass than
46
This is called the principle of confusion as the more complex chemical composition translates into a greater number
of compounds that could nucleate. This competition frustrates crystal nucleation and growth, which will not occur
during sufficiently rapid cooling.
650 Amorphous and Disordered Materials

to themselves, which increases the energy barrier at the liquid–solid interface, reduces atomic
diffusivity, inhibits crystallization, and thereby extends the temperature range over which the
supercooled liquid is stable. (4) Use of alloys with low-temperature eutectics to give a stable
liquid state closer to ambient conditions.
In 1969, Chen and Turnbull [40] used these ideas to make metallic-glass spheres of ternary
PdxMySiz alloys (M = Ag, Cu, Au) at cooling rates of only 100 K/s. This made it possible to
prepare particles 0.5 mm in diameter. In 1974, the critical casting thickness was increased to about
1 mm in PtxMyPz systems (M = Ni, Co, Fe) [41], and, in the early 1980s, glassy ingots of
Au55Pb22.5Sb22.5 with diameters of 5 mm were made [42]. Inoue et al. [43] and Peker and
Johnson [44] later discovered metallic glasses based on La-, Mg-, Zr-, Pd-, Fe-, and Ti-containing
alloys with even lower critical cooling rates (1–100 K/s), which were amenable to conventional
molding.
Metallic glasses are found in a wide variety of commercial applications. Initially,
they were used in sporting equipment such as golf-club heads, tennis rackets, and
baseball bats. Golf-club heads using Zr41.2Ti13.8Cu12.5Ni10.0Be22.5 (“Vitreloy”) are
twice as hard and four times as elastic as ones based on Ti metal, resulting in the
transformation of 99% of the impact energy to the ball, compared to only 70% for Ti-
based heads. Other applications exploit magnetic properties of metallic glasses, in
particular those made of B, Si, P and Fe, Co, Ni, which are soft magnets (see
Section 9.9 and Figure 9.23). Ferromagnetic metallic glasses can have high magnetic
susceptibilities, low coercivity, and high electrical resistivity. When subjected to alter-
nating magnetic fields, their high resistance limits eddy currents.47 This, together with
their low coercivity, is utilized in low-loss cores in power-distribution transformers.
Bulk metallic glasses have also found use in optical mirrors, precision gears for micro-
motors, diaphragms for pressure sensors, and structural parts for aircrafts [45].
Furthermore, prototypes of efficient electrical motors demonstrated that permanent
magnets made of metallic glasses can replace rare-earth magnets that have a potentially
limited supply.
A more recent and related class of disordered materials are high-entropy alloys. Their
mixing entropies increase by virtue of the high number of elemental components, and
this lowers the Gibbs energy of the solid solution. In contrast to common alloys
containing two or three elements in various proportions, high-entropy alloys are
located near the middle of their multidimensional phase diagrams yet keep the simple
face-centered-cubic, body-centered-cubic, or hexagonal closest packing due to averaged
intersite disorder [46]. High-entropy alloys typically have high hardness, low ductility,
and enhanced wear resistance compared to conventional metallic alloys [47].

47
An alternating magnetic field will induce eddy currents in a conductor (Faraday’s law of induction), which are
inversely proportional to the resistivity of the material.
 15.12 Problems 651

15.12 Problems
15.1 Using Zacharisen’s rules, explain why SiO2 forms a glass and MgO does not.
15.2 Explain why Zachariasen’s rules suggest that Al2O3 will not form a glass but alumino-
silicates with typical compositions of 11–16 mole percent (mol%) Al2O3, 52–60 mol%
SiO2, and 9–11 mol% K2O will.
15.3 Consider a glass of composition (M2O)x(SiO2)1−x, where M2O is an alkali-metal
oxide. State the fraction of non-bridging oxygen and the ratio of non-bridging oxygen
to silicon.
15.4 A sodium-silicate glass contains 25 mol% Na2O. Calculate its O:Si ratio and derive a
formula for the composition of a glass with the same O:Si ratio yet with Na2O and
CaO as the network modifiers (CaO results in a more durable glass).
15.5 The glass core of an optical fiber has an index of refraction 1.64. The index of
refraction of the cladding is 1.50. (a) What is the maximum angle a light ray can
make with the wall of the core if it is to remain inside the fiber? (b) Would this optical
fiber work without cladding?
15.6 Calculate the maximum angle a light ray can make in internal reflection with the wall
of the core in an optical fiber made with SiO2 cladding (n2 = 1.609) and a core with a
1% higher refractive index.
15.7 Yellow light of the Na doublet with a wavelength of 589.3±0.3 nm and frequency of
5.09×1014 Hz in vacuum enters Fe2O3 that has a refractive index of 3.00. Calculate the
speed, wavelength, and frequency of the light in Fe2O3 and indicate which of the three
change when the light passes from the vacuum to the Fe2O3.
15.8 Prove that Equation (15.6) reduces to the Lorenz–Lorentz equation (15.4) in the case
of purely ionic bonding.
15.9 Use the Anderson–Eggleton relationship to calculate the refractive index of the
mineral orthoclase (KAlSi3O8) with a unit-cell volume 720.4 Å3 containing four
formula units. The cation polarizabilities are 1.35 Å3, 0.533 Å3, and 0.284 Å3, for
K+, Al3+, and Si4+, respectively. The free-ion polarizability α0 for oxygen is 1.79 Å3
and N0 is 1.776 Å3 (all values from Tables 4 and 5 in ref. [15]).
15.10 Calculate the refractive index using the Gladstone–Dale approach for orthoclase of
Problem 15.9 using the chemical refractivities from Table 15.1. Compare the result
with that of Problem 15.9.
15.11 Given the density of 2.655 g/cm3 of an SiO2 glass, estimate Va in Å3 per formula unit.
Use the refractive index of 1.547 to calculate the polarization with the Lorentz–Lorenz
equation (15.4).
15.12 Calculate the percentage transmission for a 0.1 dB/km and 0.001 dB/km optic fiber
after 1 km.
652 Amorphous and Disordered Materials

15.13 Further Reading

J.L. Finney, L.V. Woodstock, “Renaissance of Bernal’s random close packing and hypercritical line in
the theory of liquids” J. Phys. Condens. Matter 26 (2014), 463102.
E. Le Bourhuis “Glass: Mechanics and Technology” (2008) Wiley.
J.C. Mauro, R.J. Loucks, A.K. Varshneya, P.B Gupta, “Enthalpy landscapes and the glass transi-
tions” Sci. Model. Simul. 15 (2008), 241–281.
C.A. Angell, “The old problems of glass and the glass transition, and the many new twists” Proc. Natl.
Acad. Sci. USA 92 (1995), 6675–6682.
M.D. Ediger, C.A. Angell, S.R. Nagel, “Supercooled liquids and glasses” J. Phys. Chem. 100 (1996),
13200–13212.
S.F. Swallen, K.L. Kearns, M.K. Mapes, Y.S. Kim, R.J. McMahon, M.D. Ediger, T. Wu, L. Yu, S.
Satija, “Organic glasses with exceptional thermodynamic and kinetic stability” Science 315 (2007),
353–356.
J.F. Sadoc, R. Mosseri, “Geometrical Frustration” (2007) Cambridge University Press.
V. Lubchenko, P.G. Wolynes “The microscopic quantum theory of low temperature amorphous
solids” Adv. Chem. Phys. 136 (2007), 95–296.
T. Egami “Magnetic amorphous alloys: Physics and technological applications” Rep. Prog. Phys. 47
(1984), 1601–1725.

15.14 References
[1] P.W. Anderson, “Through the glass lightly” Science 267 (1995), 1615–1616.
[2] T. Egami, S. Billinge, “Underneath the Bragg Peaks: Structural Analysis of Complex Materials”
2nd edition (2012) Pergamon Press.
[3] S. Kohara, K. Suzuya, “Intermediate-range order in vitreous SiO2 and GeO2” J. Phys.: Condens.
Matter 17 (2005), S77–S86.
[4] J.D. Bernal, “A geometrical approach to the structure of liquids” Nature 183 (1959), 141–147.
[5] J.D. Bernal, J. Mason, “Packing of spheres: Coordination of randomly packed spheres” Nature
188 (1960), 910–911.
[6] D.B. Miracle, T. Egami, K.M. Flores, K.F. Kelton, “Structural aspects of metallic glasses” MRS
Bull. 32 (2007), 629–634.
[7] A. Donev, I. Cisse, D. Sachs, E.A. Variano, F.H. Stillinger, R. Connelly, S. Torquato, P.M.
Chaikin, “Improving the density of jammed disordered packings using ellipsoids” Science 303
(2004), 990–993.
[8] N. Mousseau, G.T. Barkema, S.M. Nakhmanson, “Recent developments in the study of random
continuous random networks” Phil. Mag. B 82 (2002), 171–183.
[9] L.W. Hobbs, C.E. Jesurum, V. Pulim, B. Berger, “Local topology of silica networks” Phil. Mag.
A 78 (1998), 679–711.
[10] J. Shropshire, P.P. Keat, P.A. Vaughan, “The crystal structure of keatite, a new form of silica” Z.
Kristallogr. Cryst. Mater. 112 (1959), 409–412.
[11] G.H. Beall, “Dr. S. Donald (Don) Stookey (1915–2004): Pioneering researcher and adventurer”
Front. Mater. 3 (2016), article 37.
 15.14 References 653

[12] R.D. Shannon, “Revised effective ionic radii and systematic studies of interatomic distances in
halides and chalcogenides” Acta. Crystallogr. Sect. A 32 (1976), 751–767.
[13] J.C. Mauro, A. Tandia, K.D. Vargheese, Y.Z. Mauro, M.M. Smedskjær, “Accelerating the
design of functional glasses through modeling” Chem. Mater. 28 (2016), 4267–4277.
[14] J.B. Pendry, D. Schurig, D.R. Smith, “Controlling electromagnetic fields” Science 312 (2016),
1780–1782.
[15] R.D. Shannon, R.X. Fischer, “Empirical electronic polarizabilities of ions for the prediction and
interpretation of refractive indices: Oxides and oxysalts” Am. Mineral. 101 (2016), 2288–2300.
[16] O.L. Anderson, “Optical properties of rock-forming minerals from atomic properties” Fortschr.
Mineral. 52 (1975), 611–629.
[17] R.A. Eggleton, “Gladstone–Dale constants for the major elements in silicates: Coordination
number, polarizability and the Lorentz–Lorenz relationship” Can. Mineral. 29 (1991), 525–532.
[18] J. A. Mandarino, “The Gladstone–Dale compatibility of minerals and its use in selected mineral
species for further study” Can. Mineral. 45 (2007), 1307–1324.
[19] D.B. Keck, R.D. Maurer, P.C. Schultz, “On the ultimate lower limit of attenuation in glass
optical waveguides” Appl. Phys. Lett. 22 (1972), 307–308.
[20] C.R. Hammond, S.R. Norman, “Silica based binary glass systems: Refractive index behavior and
composition in optical fibers” Opt. Quantum Electron. 9 (1977), 399–409.
[21] K. Nagayama, M. Kakui, M. Matsui, T. Saitoh, Y. Chigusa, “Ultra low loss (0.1484 dB/km) pure
silica core fiber and extension of transmission distance” Electron. Lett. 38 (2002), 1168–1169.
[22] H. Tanaka, “Possible resolution of the Kauzmann paradox in supercooled liquids” Phys. Rev. E
68 (2003), 011505.
[23] J.B. Lambert, G.O. Poinar, Jr., “Amber: The organic gemstone” Acc. Chem. Res. 35 (2002),
628–636.
[24] T. Pérez-Castañeda, R.J. Jiménez-Riobóo, M.A. Ramos, “Two-level systems and boson peak
remain stable in 110-million-year-old amber glass” Phys. Rev. Lett. 112 (2014), 165901/1–5.
[25] M.H. Cohen, P. Turnbull, “Molecular transport in liquids and glasses” J. Chem. Phys. 31 (1959),
1164–1169.
[26] H.-S. Philip Wong, S. Raoux, S. Kim, J. Liang, J.P. Reifenberg, B. Rajendran, M. Asheghi, K.E.
Goodson, “Phase change memory” Proc. IEEE 98 (2010), 2201–2227.
[27] F. Rao, K. Ding, Y. Zhou, Y. Zheng, M. Xia, S. Lv, Z. Song, S. Feng, I. Ronneberger, R.
Mazzarello, W. Zhang, “Reducing the stochasticity of crystal nucleation to enable subnanose-
cond writing” Science 358 (2017), 1423–1427.
[28] O. Gulbiten, J.C. Mauro, X. Guo, O.N. Boratav, “Viscous flow of medieval cathedral glass” J.
Am. Ceram. Soc. 101 (2018), 5–11.
[29] P. Debye, “Zur Theorie der spezifischen Wärme” Ann. Phys. 39 (1912), 789–839.
[30] R.C. Zeller, R.O. Pohl, “Thermal conductivity and specific heat of noncrystalline solids” Phys.
Rev. B 4 (1971), 2029–2041.
[31] P.W. Anderson, B.L. Halperin, C.M. Varma, “Anomalous low-temperature thermal properties
of glasses and spin-glasses” Philos. Mag. 25 (1972), 1–9.
[32] W.A. Phillips, “Tunneling states in amorphous solids” J. Low Temp. Phys. 7 (1972), 351–360.
[33] C. Müller, J.H. Cole, L. Lisenfeld, “Towards understanding two-level systems in amorphous
solids: Insights from quantum circuits” Rep. Prog. Phys. 82 (2019), 12450.
[34] C.R. Billman, J.P. Trinastic, D.J. Davis, R. Hamdan, H.-P. Chen, “Origin of the second peak in
the mechanical loss function of amorphous silica” Phys. Rev. B 95 (2017), 014109.
654 Amorphous and Disordered Materials

[35] T. Ying, Y. Gu, X. Chen, X. Wang, S. Jin, L. Zhao, W. Zhang, X. Chen, “Anderson localization
of electrons in single crystals: LixFe7Se8” Sci. Adv. 2 (2016), e1501283.
[36] W. Klement, Jr., R.H. Willens, P. Duwez, “Non-crystalline structure in solidified gold–silicon
alloys” Nature 187 (1960), 869–870.
[37] K.F. Kelton, G.W. Lee, A.K. Gangopadhyay, R.W. Hyers, T.J. Rathz, J.R. Rogers, M.B.
Robinson, D.S. Robinson, “First X-ray scattering studies on electrostatically levitated metallic
liquids: Demonstrated influence of local icosahedral order on the nucleation barrier” Phys. Rev.
Lett. 90 (2003), 195504/1–4.
[38] A. Hirata, L.J. Kang, T. Fujita, B. Klumov, K. Matsue, M. Kotani, A.R. Yavari, M.W. Chen,
“Geometric frustration of icosahedron in metallic glasses” Science 341 (2013), 376–379.
[39] A. Inoue “Stabilization of metallic supercooled liquid and bulk amorphous alloys” Acta Mater.
48 (2000), 279–306.
[40] H.S. Chen, D. Turnbull, “Formation, stability and structure of palladium–silicon alloy based
glasses” Acta Metall. 17 (1969), 1021–1031.
[41] H.S. Chen, D. Turnbull, “Thermodynamic considerations on the formation and stability of
metallic glasses” Acta Metall. 22 (1974), 1505–1511.
[42] M.C. Lee, J. M. Kendall, W. L. Johnson, “Spheres of the metallic glass Au55Pb22.5Sb22.5 and their
surface characteristics” Appl. Phys. Lett. 40 (1982), 382–384.
[43] A. Inoue, A. Kato, T.G.K.S. Zhang, T. Masumoto, “Mg–Cu–Y amorphous alloys with high
mechanical strengths produced by a metallic mold casting” Mater. Trans. JIM 32 (1991),
609–616.
[44] A. Peker, W.L. Johnson, “A highly processable metallic glass: Zr41.2Ti13.8Cu12.5Ni10.0Be22.5”
Appl. Phys. Lett. 63 (1993), 2342–2344.
[45] A. Inoue, N. Nishiyama, “New bulk metallic glasses for applications as magnetic-sensing,
chemical, and structural materials” MRS Bull. 32 (2007), 651–658.
[46] B. Cantor, “Multicomponent and high entropy alloys” Entropy 16 (2014), 4749–4768.
[47] M.-H. Tsai, J.-W. Yeh, “High-entropy alloys: A critical review” Mater. Res. Lett. 2 (2014),
107–123.
 Appendices

APPENDIX A
Crystallographic Point Groups in Schönflies
Symbolism

Throughout the book, the so-called Hermann–Maugin symbolism has been used for
notation of the symmetry elements of crystallographic point groups because it has
traditionally been associated with the analysis of long-range periodic structures. When
the experimental method focuses on the local coordinations, as spectroscopic methods
do, another symbolism is often used, named after Arthur Moritz Schönflies. The
Schönflies symbolism differs from the Hermann–Maugin symbolism not only in symbols
but also in the choice of some symmetry elements; instead of the Hermann–Maugin
rotoinversion axes, rotoreflection axes are used. Instead of listing the generating sym-
metry elements in significant directions, the Schönflies symbolism uses a shorthand for
the visual form or symmetry of the object. In Table A.1, we add the Schönflies notation
to the Hermann–Maugin-style Table 1.1.

Table A.1 Classification of 32 crystallographic point groups into seven crystal systems.

Minimum symmetry Higher-symmetry point groups

Crystal system H.–M. Schön. Hermann–Maugin Schönflies

Triclinic 1 C1 1 Ci
Monoclinic 2, m C2, Cs 2/m C2h
Orthorhombic 222 D2 mm2, mmm C2v, D2h
Tetragonal 4, 4 C4, S4 4/m, 422, 4mm, 42m, 4/mmm C4h, D4, C4v, D2d, D4h
Hexagonal 6, 6 C6, C3h 6/m, 622, 6mm, 6m2, 6/mmm C6h, D6, C6v, D3h, D6h
Trigonal 3, 3 C3, C3i 32, 3m, 3m D3, C3v, D3d
Cubic 23 T m3, 43m, 432, m3m Th, Td, O, Oh
656 Appendices

APPENDIX B
International Tables for Crystallography

The International Tables for Crystallography, Volume A: Space-group symmetry is the first in a
series of authoritative volumes published by the International Union of Crystallography (the first
online edition, 2006, http://dx.doi.org/10.1107/97809553602060000100 or http://it
.iucr.org/Ab/, and the second online edition, 2016, http://dx.doi.org/10.1107/9780955360206000
0114 or http://it.iucr.org/Ac/). Volume A lists symmetry properties of the 230 space groups used
for analysis and description of crystal structures. The symmetry information is presented sym-
bolically, graphically, and mathematically. As an example, the entry for space group Pnma is
explained in this appendix.
International Tables contain a minimum of two pages for each space group. The first of these
two pages for space group Pnma is shown in Figure B.1. The upper line of text contains the space-
group symbol Pnma, the symbols of the point group in Schönflies (D16 2h ) and Hermann–Maugin
(mmm) notations and the crystal system (Orthorhombic). The next line gives the number (62) of
the space group (from 1 to 230; in order of ascending symmetry), the full Hermann–Maugin
space-group symbol (P 21/n 21/m 21/a), and the symbol of Patterson symmetry (Pmmm) of an
electron-density map used to determine crystal structures from diffraction that adds 1. The
symmetry elements in the full space-group symbol are oriented along the main symmetry direc-
tions listed in Table 1.3.
The space-group diagram is given below these lines and shows the location of the symmetry
elements in the unit cell. For an orthorhombic space group, the figure in the top left is the standard
projection/view of the unit cell against its c axis, with the a axis down the projection plane (the
page), b horizontal, c coming out of the plane towards the reader, and the origin in the upper left
corner. The second orthographic projection of the unit cell is on the right (a down, c horizontal),
and the third (c down, b horizontal) is below the standard projection. The standard-projection
symbol is identical with the full space-group symbol. Its corresponding point diagram at the
bottom right of Figure B.1 shows how the initial point x,y,z represented by “○+” is repeated by
the symmetry elements. The z coordinate given as the “+” or “½+” means +z or ½ + z, as we are
looking down against c. A comma in the point’s circle signifies handedness relative to the initial
point. Handedness of two points behind each other would be symbolized by the circle divided into
two halves, one of them with a comma.
The point diagram can be understood if one considers the graphical symbols for symmetry
elements in the standard projection on top left: Here the n, m, and a planes of the space-group
symbol are represented by, respectively, the dashed dotted line for the diagonal glide, by
the full line for the mirror plane and by for the glide of translation in the direction a of the
arrow. This glide is at a height of z = ¼. In the other two projections, however, these same
symmetry elements are drawn with symbols that correspond to their new orientation against the
 Appendices 657

International Tables for Crystallography (2006). Vol. A, Space group 62, pp. 298–299.

mmm Orthorhombic
No. 62 P 21/n 21/m 21/a Patterson symmetry P m m m

Origin at 1̄ on 1 21 1
Asymmetric unit 0 ≤ x ≤ 12 ; 0 ≤ y ≤ 14 ; 0≤z≤1

Symmetry operations
(1) 1 (2) 2(0, 0, 12 ) 1
4 , 0, z (3) 2(0, 12 , 0) 0, y, 0 (4) 2( 21 , 0, 0) x, 14 , 14
(5) 1̄ 0, 0, 0 (6) a x, y, 14 (7) m x, 14 , z (8) n(0, 12 , 12 ) 14 , y, z

Figure B.1 Page 298 of International Tables for Crystallography, Volume A (2006).

plane of the page. In the diagram on the top right (projection onto the ac plane), the mirror plane is
now ¼ above this plane and has the symbol . In the lower left diagram (projection onto bc plane),
the n glide is shown as , and the symbol is the original a glide viewed side on (implying that
the direction of translation is now out of page, against the viewer). The inversions 1 are marked
with small circles ◦, the twofold screw axes 21 have symbol . If this symbol has a white circle in its
658 Appendices

center, it means a 21 with a perpendicular m (21/m; the combination generates a center of inversion
at the crossing). If the 21 axis is parallel to the projection plane, its symbol becomes a half-arrow
.
Origin (in this case listed as 1 on 1 21 1) defines the placement of the origin with respect to
the symmetry elements. By convention, the origin is placed at an inversion where possible. In
some space groups, different origins can be chosen or some of the origin coordinates may be
undefined (a “free” or “floating” origin).
Asymmetric unit, here 0 ≤ x ≤ ½; 0 ≤ y ≤ ¼; 0 ≤ z ≤ 1, defines the smallest unit from which
application of all symmetry operations of the space group fills the space, thus creating the
structure.
Symmetry operations contain a numbered list of those symmetry operations that create
each of the general points (of highest multiplicity, see below) from the initial point x,y,z. In
this case, the numbers (in parentheses) are: (1) the identity; (2), (3), (4) the 21 axes; (5) the
inversion located at the origin (0,0,0); (6) the a glide in the xy plane at z = ¼; (7) the mirror m
in the xz plane at y = ¼; and (8) the n glide in the yz plane at x = ¼. For operations (2), (3),
(4), (8), fractional lattice translations are specified in round brackets. This tells us that (2) to
(4) are actually 21 screw axes rather than twofold rotation axes. The numbering (1) to (8)
relates to points (1) to (8) in the general position listed on the second page for Pnma.
The second page for space group Pnma (Figure B.2) begins with Generators selected. This lists
the symmetry operations required to generate all the symmetry elements and always contains the
identity operation numbered (1) and the three translations corresponding to the three basis
vectors a, b, and c; t(1,0,0), t(0,1,0), and t(0,0,1). For non-primitive space groups, centering
translations would come next. The remaining generators of the space group are identified by the
numbering used in the Symmetry operations; in this particular case, those numbered (2), (3), and
(5), which are the 21 axis along c, 21 axis along b, and the inversion at the origin.
The Positions heading has several subheadings. Multiplicity expresses how many times a
given position is repeated in the cell upon action of the symmetry operations. The highest
multiplicity is characteristic of a general position (a point that does not lie on any symmetry
element, except identity). For Pnma the general position has multiplicity 8. The general
position is followed by special positions, here all have multiplicity 4. These have coordinates
that lie on one or more of the symmetry elements, the symmetry operations of which project
the atom on itself, decreasing the multiplicity.
The Wyckoff letter is an alphabetical label given to the Wyckoff position. It starts with an
“a” for one of the lowest-multiplicity sites at the bottom. The Site symmetry gives the point
symmetry for the site. The Coordinates show how the original point x,y,z representing
coordinates of an atom is transformed after the symmetry operations have been performed
(eight such operations in Figure B.1), and the number of resulting coordinates corresponds
to the multiplicity of the position. Reflection conditions are conditions for the appearance of
hkl Bragg reflections when an atom is placed in that given Wyckoff position. Bragg reflec-
tions not fulfilling the listed condition are absent or extinct. When several special Wyckoff
positions are occupied, their reflection conditions combine towards those listed for the
general position that generates the maximum number of reflections.
 Appendices 659

Figure B.2 International Tables for Crystallography, Volume A (2006) p. 299.

Subgroups and supergroups1 are related by their order. The order of a group is the number
of the symmetry operations it contains (as listed earlier), and it also corresponds to the
multiplicity of the general point x,y,z. A group S is called a (proper) subgroup to the group G

1
We skip the symmetry of special projections entry as it involves plane groups not used in this text.
660 Appendices

if all symmetry elements of S are also present in G and the order of G divided by the order of
S is an integer. Knowledge of group–subgroup relationships gives a clue on how a gradual
removal of symmetry via deformations or via ordering of atoms or charges or magnetic
moments can change the given parent structure into the less symmetric structure of its
“child” subgroup. As we see in Chapters 2 and 4, these considerations may be used to
investigate structural defects such as twinning and antiphase boundaries, and also to
describe structural phase transitions.
The Maximal non-isomorphic subgroups are those closest to the parent group.2 There
is no other group in the hierarchy between the group and its maximal subgroup. The
accompanied dilution of the symmetry operations is expressed by the so-called index,
an integer number placed in square brackets that precedes the symbol of the subgroup.
According to whether the group-to-subgroup transition involves the loss of some
rotational or translational symmetry, two types of maximal non-isomorphic subgroups
are distinguished: Translationengleiche3 subgroups (under heading I) are those that keep
the translations (all lattice points are kept), and the index tells us how many times the
number of (rotational) Symmetry operations per lattice point was decreased upon the
transition into the subgroup. Klassengleiche4 subgroups are those that keep the crystal-
lographic point-group symmetry, the “crystal class”, while some translations (lattice
points) are lost via dilutions such as loss of centering or a cell-edge multiplying. Those
klassengleiche subgroups that retain the same conventional cell are listed under IIa and
arise by removal of centering. Those klassengleiche subgroups that acquire a larger
conventional cell than the parent group are listed under IIb. In both cases, the index in
the square bracket is the dilution factor of lattice points upon the transition into the
subgroup. For our Pnma in Figure B.2, only type I subgroups exist. One of them forms
when Symmetry operations (5) to (8) are lost, lowering the space-group symmetry to
P212121. Another when Symmetry operations (2), (4), (5), (7) are lost—the translatio-
nengleiche subgroup Pn21a and so on.
The category IIc lists the Maximal isomorphic subgroups of the lowest index, where
the term isomorphic indicates that the space group remains the same (only the unit cell
multiplies). Since the number of such enlargements is infinite, the condition of lowest
index (here the dilution factor of lattice points) is applied to ensure that only “nearest”
supercells are listed. Minimal non-isomorphic supergroups are the last category listed on
the page. These relate to the situation in which the present group is a subgroup of a
supergroup and let one investigate the higher possible symmetries. The subdivisions
into I translationengleiche and II klassengleiche subgroups have the same meaning as
explained above.

2
Non-isomorphic; any two groups except for 11 enantiomorphic pairs of space groups: P41–P43 and 10 others (an
inverted structure cannot be rotated to align with the original structure).
3
German for “equal in terms of translations”. Shorthand t-subgroups is also in use.
4
German for “equal in terms of classes”. Shorthand k-subgroups is also in use.
 Appendices 661

APPENDIX C
Nomenclature of Silicates

A systematic nomenclature of silicate anions has been developed by Friedrich Liebau.

The newest version [1] places the formula of the silicate anion in square brackets and
all geometry information in the preceding curly brackets, as {B,P,MD}[SimOna–]. In this
scheme, B stands for branchedness of the silicate anion, P for the periodicity of the
tetrahedral chain, and M for multiplicity with D for dimensionality come as one
combined symbol. The branchedness parameter B describes branching along a chain
or ring; uB stands for unbranched, oB open branched, lB loop branched. The multipli-
city parameter M describes how many parallel strings of tetrahedra the chain has. As
an example, the tremolite chain in Figure 1.56 is a double chain. The dimensionality
parameter D of the silicate anion is a “t” for a terminated anion, r for ring, 1∞ for
infinite chain, 2∞ for a layer, 3∞ for a framework. The periodicity of the chain P is the
number of tetrahedra in the repeat unit along the chain direction (along the ring for
ring chains), excluding any branches. When a name is formed for the silicate anion
from the formula, this number is pronounced as a German enumerative. A chain with a
repeat unit of 1 tetrahedron is called einer, 2 zweier, 3 dreier, 4 vierer, 5 fünfer, 6
sechser, 7 siebener, 8 achter, etc. As an example, the stokesite chain in Figure 1.56 is an
unbranched sechser, the tremolite chain is an unbranched zweier double chain. More
complex silicate chains are shown in Figure C.1 using examples of minerals that also
contain other anions.

open-branched vierer single chain Si6O1812− unbranched vierer double chain Si8O208−
in aenigmatite Fe5Na2Ti{oB,4,1∞1 }[Si6O18]O2 in narsarsukite Na4Ti2{uB,4,2∞1 }[Si8O20]O2

loop-branched vierer single chain Si6O1710− in

deerite Fe62+Fe33+{lB,4,1∞1 }[Si6O1710−]O3(OH)5

Figure C.1 Examples of infinite silicate chains, their formulae, and nomenclature.
662 Appendices

APPENDIX D
Bond-Valence Parameters in Solids

Equation (5.16) contains empirical parameters R0ij and B for the bonded pair of ions ij. Each
P
R0ij ¼ Blnðvi = edij =B Þ is calculated from from i–j bond distances dij in the coordination
polyhedron and from the value vi of the cation’s oxidation state. The values for a select set of
crystal structures with the given i solely coordinated by j are then averaged. Brown and
Altermatt [2] used this procedure to compile a table of bond-valence parameters for use
primarily in oxides. A similar approach was used by Brese and O’Keeffe [3]. Selected values
of R0ij for a variety of atoms bonded to oxygen are compiled in Table D.1. A more extensive
set is at www.iucr.org/resources/data/datasets/bond-valence-parameters. For ions with s2
lone pairs and those that may adopt one of several spin states, the values need to be applied
with caution.
In other approaches, O’Keeffe and Brese [4] derived a formula of two tabulated param-
eters based on size and electronegativity, from which single-bond lengths R0ij can be calcu-
lated for all combinations of 75 elements. More recently, Gagné and Hawthorn [5] evaluated
both R0ij and B independently for bonds to oxygen, with some improvement in standard
deviations.

Table D.1 Selected bond-valence parameters R 0M–O (in Å) for oxides, sorted by oxidation
states of M as column headers. The data are based on B = 0.37 Å and are compiled from refs.
[2] and [3].

+1 +2 +3 +4 +5 +6
Li 1.466 Be 1.381 B 1.371 C 1.390 N 1.432 S 1.624
Cu 1.593 Ni 1.654 Al 1.651 Si 1.624 P 1.604 Se 1.788
Na 1.803 Cu 1.679 Co 1.70 S 1.644 As 1.767 Cr 1.794
Ag 1.842 Co 1.692 Cr 1.724 Ge 1.748 V 1.803 Mo 1.907
Hg 1.90 Mg 1.693 Ga 1.730 Mn 1.753 Br 1.840 Te 1.917
K 2.132 Zn 1.704 V 1.743 V 1.784 Nb 1.911 W 1.921
Tl 2.172 Fe 1.734 Fe 1.759 Os 1.811 Ir 1.916 Os 2.03
Rb 2.263 Pt 1.768 Mn 1.76 Se 1.811 Ta 1.92 U 2.075
Cs 2.417 Mn 1.79 As 1.789 Ti 1.815 Sb 1.942
Pd 1.792 Ti 1.791 Ru 1.834 I 2.003
Cd 1.904 Rh 1.791 Pt 1.879 Bi 2.06
 Appendices 663

Table D.1 (cont.)

Ca 1.967 Au 1.833 Sn 1.905 U 2.075

Hg 1.972 Sc 1.849 Hf 1.923
Sn 1.984 In 1.902 Zr 1.937
Pb 2.112 Lu 1.971 Te 1.977
Sr 2.118 Sb 1.973 Ce 2.028
Eu 2.147 Yb 1.985 Pb 2.042
Ba 2.285 Tm 2.000 U 2.112
Tl 2.003 Th 2.167
Er 2.010
Y 2.014
Ho 2.025
Dy 2.036
Tb 2.049
Gd 2.065
Eu 2.074
Sm 2.088
Bi 2.094
Nd 2.117
Pr 2.138
Ce 2.151
La 2.172
Ac 2.24

APPENDIX E
The Effect of a Magnetic Field on a Moving Charge

A homogeneous magnetic field exerts a force on a wire carrying electric current, whether it is
straight or in a loop. This short recapitulation from physics may help you to grasp some
terms and units used in magnetism.
The left-hand side of Figure E.1 shows a straight segment of wire carrying current I. The
force F exerted along a length l of the wire perpendicular to magnetic field of magnetic
induction B is perpendicular to both the field and the wire, and its magnitude is F = IBl. This
can be used to derive the force and moment exerted on a current loop of area A at an angle φ
to a homogeneous field B. In order to do so, we draw a rectangular current loop of sides a and
b. Forces F1, F2, F3, and F4 act on the four edges of the loop. Of these, F3 and F4 cancel. The
moment of force acting on the loop is then M = F1p = F1bsinφ = Iabsinφ. As ab is the area A
664 Appendices

Figure E.1 Left: Force F exerted on a straight length l of a wire carrying current I in a homogeneous
magnetic field of magnetic induction B. Middle and right: Force exerted on a current loop in magnetic
field (two projections).

of the loop, the vector M of the moment is the cross product M = I(A × B) where A is the
loop-area vector and B is the vector of the magnetic-field induction.

APPENDIX F
Coupling j–j

When spin–orbit coupling is relatively weak, it can be approximated by summing the

spin momenta and orbital momenta separately and combining them afterwards—the
Russell–Saunders coupling. For elements such as the actinoids, where spin–orbit coup-
ling is strong, this scheme breaks down and one must use j–j coupling. For each
electron, its orbital angular momentum is summed with its spin angular momentum
to give the total angular momenta j1 = ℓ − ½ and j2 = ℓ + ½. The new quantum
numbers for these states are j and mj. The j values are then combined to form the total
angular momentum J.
Let’s use an example. Consider a Pb atom which has two unpaired electrons in its p shell,
[Xe]6s26p2. For a configuration with x unpaired electrons in an orbital of an angular-
momentum quantum number ℓ, the number of microstates is given by Nℓ!/[x!(Nℓ − x)!]
where Nℓ =2ℓ(ℓ + 1). Thus for our p2 configuration we expect 6!/[2!4!] = 15 microstates.
Since ℓ = 1 for p orbitals, and |s| = ½ for an unpaired electron, addition of the electron’s
orbital momentum and spin momentum gives j = ³⁄₂ or ½ with three possible combinations of
j1 and j2 for the two electrons; (³⁄₂ ,³⁄₂ ), (³⁄₂ ,½) and (½,½). When j = ³⁄₂ , the mj values possible
are −³⁄₂ , −½, ½, ³⁄₂ , whereas for j = ½, mj may be −½ or ½ (see Figure 9.8). There are only six
allowed pairs of mj1 and mj2 for the j1,j2 combination ³⁄₂ ,³⁄₂ because the Pauli exclusion
 Appendices 665

principle states that if j1 equals j2, mj1 and mj2 must not be equal. The six mj1,mj2 pairs
are: ³⁄₂ ,½ [mj1 + mj2 = MJ = +2]; ³⁄₂ ,−½ [MJ = +1]; ³⁄₂ ,−³⁄₂ [MJ = 0]; ½,−½ [ MJ = 0];
½,−³⁄₂ [MJ = −1]; −½,−³⁄₂ [MJ = −2], corresponding to J = 2, 1, 0. For the second j1,j2
combination ³⁄₂ ,½, there are eight pairs because mj1 and mj2 may be equal; the two
unpaired electrons of Pb in this state are not equivalent, and this also means that ½,−½
and −½,½ combinations of mj1 and mj2 belong to two different sub-states. Again,
J = 2, 1, 0 are the results of these combinations. For the third j1,j2 combination
½,½, the Pauli exclusion principle allows one possible state: mj1 = ½ and mj2 = −½
[MJ = 0] hence J = 0. Only the third Hund rule applies for the choice of the ground
state. Because the shell is less than half full, the minimum J state, J = 0, is the ground
state of the isolated Pb atom and lead is diamagnetic.

APPENDIX G
The Langevin Function

The Langevin function is an important expression that helps us understand the properties of
paramagnets. It can be derived in a semi-classical treatment by considering the competition
between thermal randomization and magnetic alignment. Let’s assume that we have a
magnetic moment free to point in any direction of space (i.e. we are ignoring the effects of
quantization as if J = ∞). We can therefore imagine the moment lying anywhere on the
surface of a unit sphere as in Figure G.1.
The probability of the moment randomly pointing into the interval between angle α
and α+dα towards the field direction z is the area of this interval, 2πsinα·dα, divided by
the entire unit-sphere surface area 4π, hence (sinα·dα)/2. The moments at angle α have a
relative potential energy Um = −Bμcosα. The probability of pointing into the interval α
to α+dα under an applied field at a temperature T is then equal to the product of the
random probability (sinα·dα)/2 and the Boltzmann probability exp(−Um/kT) = exp
(Bμcosα/kT) for the angle α. If we substitute cosα = y and Bμ/kT = x, the average

dα

sinα
α z, H
cosα

Figure G.1 The unit sphere for derivation of the Langevin function.
666 Appendices

atomic moment in the direction of the field can be calculated by integrating over the
top half of the unit sphere:

ð0   ð1
Bμ cos α 1
μ cos α  exp  sin α  dα y expðxyÞdy
kT 2  
π 1 1
μz ¼ ¼μ ¼ μ cothðxÞ 
ð0   ð1 x
Bμ cos α 1
exp  sin α  dα expðxyÞdy
kT 2
π 1

The result is called the Langevin function. In terms of magnetization, it is written as M/Msat =
coth(x) − 1/x, where Msat is the saturation magnetization. At very low T and at high field, the
function is markedly nonlinear. Under weak-field conditions or at higher temperatures, M
versus B is essentially linear (Figure 9.16).
To obtain the Curie law of the paramagnet [Equation (9.17)], two approximations are
made: B ≈ μ0H [because M in Equation (9.7) is small compared to H]; and coth x ≈ 1/x + x/3
(because x ≪ 1), yielding M/Msat = x/3 where x = μ0Hμ/kT. Since Msat in the Langevin
approximation is equal to Nμ, where N is the number of magnetic atoms, expressing M/H
[Equation (9.8) defining χ] gives the Curie law in the form of Equation (9.19).

APPENDIX H
The Brillouin Function

The Brillouin function describes the relative magnetization, M/Msat versus temperature as a
function of the applied field of induction B. One way to derive it is to replace the integrals in the
derivation of the Langevin function with summations over the discrete energy states allowed
by J, but this is a task hard to get right. Another way is to obtain the partition function, Z, for
the magnetic energy levels defined by the quantum numbers S or J, convert it to the Helmholtz
free energy F = −kTlnZ, express the magnetization thermodynamically as M = −(∂F/∂B)T and
relate it to the maximum value of such magnetization. Here, as in most textbooks, we shall
follow a similar procedure, but avoid the explicit thermodynamics.
In the magnetic field B, the ground-state paramagnetic J level splits into an MJ = −J,
−J+1, −J+2, . . . +J ladder of equidistant energy step gJ∙μB∙B (Figure 9.10).5 The distribution
(partition) of the populations found at the ladder rungs, for example among a huge collection
of equivalent atoms in a crystal, follows the partition function Z,

5
Notice that increasing B stretches the ladder.
 Appendices 667

gJ μB MJ  B
X
þJ 
Z¼ e kT (H.1)
MJ ¼J

which represents the sum of the individual populations that are controlled by the ratio
of the magnetic and thermal energy at each rung. The probability that an atom has one
specific MJ rung of the ladder populated (or the fraction of such atoms in the large
collection) is:

gJ μB MJ  B

e kT
pMJ ¼ (H.2)
Z

The mean projection μ z of the collection’s moments onto the axis of the applied field is the
sum, over all rungs, of the ladder-rung probability times that rung’s magnetic moment:

gJ μB MJ B
X
þJ
1 X þJ 
μz ¼ μz pMJ ¼ gJ μB MJ  e kT (H.3)
MJ ¼J
Z MJ ¼J

Let’s differentiate the Z in Equation (H.1) with respect to B, as it will prove useful,

  gJ μB MJ B
∂Z 1 X þJ 
¼ gJ μB MJ  e kT (H.4)
∂B T kT MJ ¼J

because, by comparison of this result with Equation (H.3), we realize that:

   
kT ∂Z ∂lnZ
μz ¼ ¼ kT (H.5)
Z ∂B T ∂B T

Let’s now simplify Equation (H.1) for Z by introducing the negatively taken ratio of the
ladder’s energy step and the thermal energy,

ξ ¼  gJ μΒ Β=kΤ (H.6)

to obtain
X
þJ
Z¼ eξMJ (H.7)
MJ ¼J

in which further substitution x = eξ provides us with an array (in parentheses)

X
þJ
Z¼ xMJ ¼ xJ þ xJþ1 þ . . . þ xJ1 þ xJ ¼ xJ ð1 þ x þ . . . x2J1 þ x2J Þ (H.8)
MJ ¼J
668 Appendices

that has a defined sum, (1 − x2J+1)/(1 − x). Then we can write Z as:
 
1  eð2Jþ1Þξ eJξ  eðJþ1Þξ
Z ¼ eJξ ¼ (H.9)
1  eξ 1  eξ

The fraction, when expanded by e−ξ/2, provides a result in which we recognize the hyperbol-
ical sinus functions:

eðJþ1=2Þξ  eðJþ1=2Þξ sinh½ðJ þ 1=2Þξ

Z¼ ¼ (H.10)
eξ=2  eξ=2 sinhðξ=2Þ

Now we can do the differentiation suggested in Equation (H.5), but, because B appears
in the functional parameter ξ inside a function, the differentiation looks like this:
     
∂lnZ ∂lnZ ∂ξ
μ z ¼ kT ¼ kT (H.11)
∂B T ∂ξ T ∂B

Plugging in the derivative of Equation (H.6) with respect to B gives

 
∂lnZ
μ z ¼ gJ μB (H.12)
∂ξ T

where plugging in Z from Equation (H.10) and differentiating yields:

 
ðJ þ 1=2Þcosh½ðJ þ 1=2Þξ 1 coshðξ=2Þ
μ z ¼ gJ μB   (H.13)
sinh½ðJ þ 1=2Þξ 2 sinhðξ=2Þ

Because gJ∙μB∙B is the saturated paramagnetic moment μsat, we multiply the large
parenthesis with J/J in order to express the ratio μ z =μsat of the average and saturated
moments:
     
μz 1 1 1 1 1
¼ Mr ¼ J þ coth J þ ξ  coth ξ ¼ BJ ðξÞ (H.14)
μsat J 2 2 2 2

This is the Brillouin function BJ(ξ) for the relative magnetization Mr = M/Msat versus ξ of
Equation (H.6) and J.6 In the spin-only case, J is replaced by S (also in ξ).
In a weak field, we expect to obtain the Curie law. Because ξ is small in a weak field, coth ξ ≈
1/ξ + ξ/3 and Equation (H.14) becomes:

6
In contrast to the effective moment, saturated or z-projection moments are summed directly into total
magnetization.
 Appendices 669

μz J þ1
¼ ξ (H.15)
μsat 3

The saturation moment was a strong-field extrapolation, in which only the lowest level MJ = −J is
occupied, and the magnitude μsat = gJ∙μB∙J in analogy with Equation (9.13):

J þ 1 gJ2 μ2B B
μz ¼ J (H.16)
3 kT
Next, we approximate B as μ0H because the paramagnetic magnetization M in Equation
(9.7) is much smaller than the field that caused it. We sum the moments for N atoms and
divide by H to get the magnetic susceptibility χN per that amount of atoms. Finally, we realize
from Equation (9.14) that (J + 1)J·gJ2μB2 is the square of the effective magnetic moment,
μeff2, consistent with the weak-field premise of the Curie law. The result is

μz μ2 μ
χN ¼ N ¼ N  eff 0 (H.17)
H 3kT
the Curie law of Equation (9.19) for the paramagnet of N isolated magnetic atoms.
When combined with the Weiss concept of internal field, the Brillouin function can also be
used to describe the relative magnetization Mr of ferromagnets. One approach uses TC to
express the strength of the ferromagnetic interactions instead of the experimentally less
accessible Weiss-field coefficient, λ. The idea is that at TC the magnetization becomes low,
so the approximation of Equation (H.15) for Mr applies and can be equated with a Weiss
internal-field-based expression of Mr for a ferromagnet at T = TC. The derivation of such a
Weiss-field expression for Mr starts with μ/μsat, where μsat = gJ·μB·J (per atom), as used
above, and μ = HW/λ (per atom) by analogy with Equation (9.22):

HW
Mr ¼ (H.18)
gJ μB Jλ

Now we need to express the Weiss internal field intensity HW as a function of temperature.
That is dictated by the ratio of the magnetic and thermal energy [Equation (H.6)] in which the
field B now includes the Weiss field, μ0(HW +H). We neglect H as it is much weaker than HW:

kT
HW ¼ ξ (H.19)
gJ μΒ μ0

Combination of Equation (H.18) and Equation (H.19) gives the internal-field-based expres-
sion for the relative magnetization Mr of a ferromagnet, in which we replace T with TC
according to the premise:

kTC
Mr ¼ ξ (H.20)
gJ μ2B μ0 Jλ
2
670 Appendices

1.0
µ (µB)
J=½
0.8
J=∞
0.6

0.4

0.2

0.0
0 200 400 600 800 1000 T (K)

Figure H.1 Thermal decay of an ordered moment of 1 μB and Tc = 1000 K, for J = ½, ³⁄₂ , 7⁄₂ , 9 and ∞.

At this temperature, it can be equated with the approximate Mr of Equation (H.15), which
allows us to express λ (per atom) and plug it back into Equation (H.20). This yields the
desired term for Mr, in which we expressed the Weiss field via the magnitude of TC:

J þ1 T
Mr ¼  ξ (H.21)
3 TC

In the ferromagnetic range, Equation (H.21) must be satisfied simultaneously with the
Brillouin function for Mr of Equation (H.14). Unfortunately, plugging ξ from Equation
(H.21) into Equation (H.14) yields no analytical solution for Mr, and these two equations
must be solved numerically. The result, illustrated in Figure H.1, is valid for thermal
disordering of any individual ordered atomic moment with a critical temperature Tc.

APPENDIX I
Measuring and Analyzing Magnetic Properties
Historically, most magnetic measurements were made using the Gouy or Faraday methods
whose basic operating principle is shown in Figure 9.4. Nowadays most measurements are made
using either a vibrating-sample or a SQUID (superconducting quantum interference device)
magnetometer. In a SQUID magnetometer, the sample is moved through a detection coil where
it induces a current. The coil is connected to a SQUID sensor which is a superconducting ring
containing a weak link or Josephson junction and acts as an extremely sensitive current-to-
voltage converter. The output voltage of the SQUID is proportional to the magnetic moment of
the sample, and the instrument can be calibrated using a known material. Modern commercial
 Appendices 671

systems are remarkable instruments with automated field and temperature control and can
routinely operate from 0 T to 7 T over temperatures from 2 K to at least 400 K with sensitivities
down to 10−8 emu. Specialized attachments allow work under non-ambient pressure, as the
sample is laser irradiated, or as other physical properties are monitored.
Samples are mounted in holders of low or known diamagnetic contribution. One common
method is to place the sample in a gelatin drug capsule, then glue (with varnish) the capsule in
a suitably sized drinking straw which is in turn mounted on the magnetometer drive assem-
bly. Alternatively, high-purity silica-glass holders can be used.
In an experiment to test low-temperature magnetic ordering, one might initially measure
moment as a function of field at room temperature. A paramagnet should give a straight line
passing through the origin. A small, non-zero, intercept might indicate the presence of a
minor ferromagnetic impurity (such as Fe or Ni from a spatula). The magnetization of such
an impurity would saturate at low field, leading to an approximately constant offset. A
suitable field for measurements can be assessed from this plot. If there is a non-zero offset, a
field large enough for this to be irrelevant should be chosen, or one could measure at two
fields on the linear portion of M versus H and subtract data.
Common commercial instruments produce data files with columns labeled as
Temperature, Field (Oe), and Moment (emu). The first stage of the analysis is to correct
the measured moment for effects due the sample holder, if any are seen in a blank experi-
ment, in order to obtain the moment due to the sample, μsample. Molar susceptibility is
χmeasured
mol ¼ μsample =½Field  ðmass=molar massÞ and should be corrected at this stage for dia-
magnetic contributions due to core electrons in the sample to give χpara mol . Since
para para
χmol
measured
¼ χmol þ χmol , one must subtract χmol . Because χmol is negative, χmol will be larger
dia dia dia

than χmeasured
mol . For simple compounds, the temperature-independent contribution of χdia mol can
be estimated from tables of Pascal’s constants (see e.g. ref. [6] for a compilation). Let’s take
CuSO4·5H2O as an example. The tabulated χdia(Cu2+), χdia(SO42−), χdia(H2O) are −11, −40.1,
and −13, in units of 10−6 emu/mol, giving χdia(CuSO4·5H2O) as −116.1×10−6 emu/mol to be
used as the diamagnetic correction. Note that its absolute value is significantly smaller than
−6
the χpara
mol of about +1300×10 emu/mol at room temperature. For a Curie–Weiss
para
paramagnet, a plot of 1=χmol versus T should give a straight line as in Figure 9.15 (or
Figure 9.18). The Curie constant C is given by the inverse gradient of this line, and the
Weiss constant θ is the temperature where this line crosses the T axis. The effective moment
μeff in Bohr magnetons per formula unit mol/NA can be determined from this C obtained
from molar susceptibility:7
pffiffiffiffi pffiffiffiffi
μeff ¼ 2:828 C ðCGSemÞ μeff ¼ 797:7 C ðSIÞ (I.1)

A second check of the paramagnetic moment is to plot μeff as a function of T after setting C =
χmolT into Equation (I.1), which should give a clearly recognizable straight horizontal line (if
7
The numerical factor is [3k/(NAμB2μ0)]0.5 in SI, derived from Equation (9.17) and Equation (9.19) after replacing N
with NA and expressing the moment in Bohr magnetons.
672 Appendices

it does not, think). All these procedures can be readily automated by least-squares fitting in a
spreadsheet package, and, with this method, it may be appropriate to fit all diamagnetic
contributions and/or temperature-independent paramagnetic contributions as a single, tem-
perature-independent contribution to χmeasured
mol . For the most careful work, one should
remember to weight fits according to experimental uncertainties in measured moments and
propagate errors correctly.
If the sample is ferro- or antiferromagnetic, plots of reciprocal susceptibility versus
temperature will become linear a little above the magnetic ordering temperature. Here the
Weiss constant (temperature) and effective moment can be obtained, as for the plain para-
magnet. TN of antiferromagnetic powder can be determined from the peak in the χmeasured mol
versus T curve. A ferromagnet’s TC can be found from extrapolation of the sudden drop in
magnetization upon heating towards this temperature.

APPENDIX J
Fundamental Constants of Exact Value

After major revision8 in 2019, all seven SI base units are defined by exact numerical values for
the fundamental physical constants9 to which they relate. As some other constants used in
chemistry and physics are composed solely of these fundamental physical constants, their
values are also exact. A list relevant for this book follows:

Avogadro constant, NA = 6.02214076×1023 mol−1

Boltzmann constant, k = 1.380649×10−23 J/K
elementary charge, e = 1.602176634×10−19 C
Planck constant, h = 6.62607015×10−34 J s
speed of light in vacuum, c = 299792458 m/s
standard atmosphere: 101325 Pa (760 torr or mm Hg)
standard-state pressure: 100000 Pa (1 bar)
molar gas constant (R = kNA): R = 8.314462618. . . J/(mol K)
molar volume (Vm = RT/P) of ideal gas at P = 101325 Pa and T = 273.15 K: Vm =
22.41396954. . . L/mol (1 L = 10−3 m3)

8
See for example https://en.wikipedia.org/wiki/2019_redefinition_of_the_SI_base_units.
9
See for example https://physics.nist.gov/cgi-bin/cuu/Category?view=html&Adopted+values.x=103&Adopted+val
ues.y=10.
 Appendices 673

References for Appendices

[1] H.-J. Klein, F. Liebau, “Computerized crystal-chemical classification of silicates and related
materials with CRYSTANA and formula notation for classified structures” J. Solid State Chem.
181 (2008), 2412–2417, and references therein.
[2] I.D. Brown, D. Altermatt, “Bond valence parameters obtained from a systematic analysis of the
Inorganic Crystal Structure Database” Acta Crystallogr. Sect. B 41 (1985), 244–247.
[3] N.E. Brese, M. O’Keeffe, “Bond valence parameters for solids” Acta Crystallogr. Sect. B 47 (1991),
192–197.
[4] M. O’Keeffe, N.E. Brese, “Atom sizes and bond lengths in molecules and crystals” J. Am. Chem.
Soc. 113 (1991), 3226–3229.
[5] O.C. Gagné, F.C. Hawthorn, “Comprehensive derivation of bond-valence parameters for ion
pairs involving oxygen” Acta Crystallogr. Sect. B 51 (2015), 562–568.
[6] G.A. Bain, J.F. Berry, “Diamagnetic corrections and Pascal’s constants” J. Chem. Educ. 85 (2008),
532–536.
Index

111 superconductors 522 Angell plot 640

1111 superconductors 522 angular momentum quantum numbers in one-electron
122 superconductors 522 atom 358
123 superconductors, see YBa2Cu3O7−δ anion conductors 539
214 superconductors 512 anion polarizabilities 629
8−N rule 17 anode 529
antibonding molecular orbital 170
A3C60 superconductors 501 antiferroelectric 319
(NH3)4Na2CsC60 503 antiferromagnet 356
Cs3C60 504 antiferromagnetic ordering 373
K3C60 501 antiferromagnetism 472
Rb3C60 501 anti-Frenkel intrinsic defect pair, see Figure 3.1
α-Bi2O3 553 anti-perovskite 539
acceptor doping 62, 93, 416 antiphase boundary 72
acceptor ionization energy 419 anti-Schottky intrinsic defect pair, see Figure 3.1
acceptors 416, 419 Arrhenius equation 111
accessible surface area 582 asymmetric unit 12
accessible volume 582 attenuation (in glass fibers) 631
acid salts (MHXO4) 547 Aurivillius structure 554
activator 268, 274, 279 autoclave 585
additivity rule 310 auxetic materials 604
aerogel 601 Avogadro constant in SI 672
Ag3[Co(CN)6] 603 axial glide plane, see glide plane, axial
Ag3SI 539
AgI 536 β-BaB2O4 340, 342
air separation 595 β-(ET)2X 508
aliovalent doping 416 Ba1−xKxBiO3 510
aliovalent substitution 62 Ba1−xVxO1.5+x 553
alizarin (C14H8O4) 265 Ba2In2O5 554
allochromatic 245 BaBiO3 509
alloy superconductors 487 BaCe1−xYxO3−x/2 548
alloy 55 BaCeO3 548
alpha cage 582 bad metal 520
ALPO materials 596 BaFe2As2 521
alternative battery anodes 570 ballistic conduction 645
aluminosilicates 579 ballistic transport 447
aluminum 398, 409, 410, 411–412 band (of crystal orbitals) 207
ambipolar diffusion 108 band emitters 274, 277
ammonia borane (BH3NH3) 57 band gap 213, 214, 262–264
Anderson localization 645, 647 band insulator 434
Anderson–Eggleton equation 630 band width 501
Anderson–Eggleton refractive index 630 band-structure diagram 207, 208
 Index 675

bandwidth 207, 221, 431 buckminsterfullerene, see C60

BaPb1−xBixO3 510 buffer pO2 97
BaPbO3 (barium lead oxide) 509 bulk diffusion 101
bar magnet 349 Burns temperature 320
barium bismuth oxide, see BaBiO3
barium lead oxide, see BaPbO3 C60 superconductors 500
barium titanate, see BaTiO3 C60 500
basic building units (BBUs) 583 Ca2FeAlO5, see brownmillerite
basis set 202, 210, 226 CaFeO3 462
BaTiO3 139, 314–318, 321, 326 calcination 587
BaZr1−xYxO3−x/2 548 CaMnO3 465
Bardeen–Cooper–Schrieffer theory 495, 505 carbenium ions 590
BCS theory, see Bardeen–Cooper–Schrieffer theory carbon nanotubes
Bechgaard salt 445, 506 chiral angle 449
[TMTSF]2X (Bechgaard salt) 507 circumferential vector 448
Bednorz and Müller 511 multi-walled carbon nanotubes (MWCNT) 447,
Beer’s law 255 451
Beevers–Ross sites 540 single-walled carbon nanotubes (SWCNT) 447–451
beta cage 581 translational vector 448
Bethe lattice 623–624 CaSi Zintl phase 48
BH3NH3, see ammonia borane cathode materials 563
Bi2Sr2Can−1CunO2n+4 516 cathode 529
Bi2WO6 554 cation conductors 536, 538
Bi4V2O11 554 cation polarizabilities 629
bicuvox 555 ccp, see cubic closest packing
bimevox 554 cell potential 530
binary phase diagram 123 centered Bravais lattices 6
binodal networks (N,M-connected nets) 30 centering 6
binomial distribution 104 centroid shift 277–279, 294
bipolaron 440 cermet 555
birefringent crystals 332, 334 charge compensation 62
bismuth oxide, see Bi2O3 charge-density wave 506
bismuth 228, 397 charge disproportionation 462
Bloch function 202, 203, 208, 220, 226 charge order 459, 510
Bloch walls 378 charge-transfer excitation 234, 256, 257, 258
block copolymers 598 ligand-to-metal charge transfer (LMCT) 258, 259, 273,
Bohr magneton 352 275, 281
Boltzmann constant in SI 672 metal-to-ligand charge transfer (MLCT) 258
bond graph 15 metal-to-metal charge transfer (MMCT) 259, 261
bond order 173 charge-transfer salts 443–445
bonding molecular orbital 170 chemical refractivity 630
bond-valence method 192 chemical sensors 531
bond valence 662 chemical strengthening of glasses 626
parameter, single-bond length, Rij° 662 chemical-diffusion coefficient 100
Born–Haber cycle 158, 159 chemisorption 56
Born–Mayer equation 157, 158 child structure 149
Bose–Einstein condensate 496 chimie douce 558
boson peak 643 chiral molecules 611
boundary surfaces 164, 168 chiral zeolites 587
Bravais lattices 5, 7 CIF (crystallographic information file) 13
non-standard settings 6 Clausius–Mossotti equation 305, 309, 327
standard settings 7 Clebsch–Gordon series 250
Brillouin function 370, 668 closest packing, hexagonal-, cubic- 19
Brouwer diagrams 92 cloverite 596
brownmillerite (Ca2FeAlO5) 554 coercive field (electrical), Ecoercive 316
676 Index

coercive field (magnetic), Hci 379 critical temperature 122, 135, 487
coercivity, see coercive field cross-relaxation 284
color 244–245 crystal classes 5
color center 58 crystal momentum 207, 407
F center 58 crystal orbitals 202, 205–207, 219, 226
M center 58 crystal systems 5, 655
R center 58 crystal-chemical formula 14
color rendering index (CRI) 292, 293 balance of bond valence 15
color temperature 291 balance of connectivity 15
colossal contraction 603 balance of electroneutrality 15
colossal magnetoresistance 470 crystal-field theory 247, 279
compositional control 470 crystallographic databases 13
variance effect, of ionic radii 471 crystallographic point groups 5, 655
commensurately modulated structure 79 Hermann–Maugin and Schönflies symbols 655
component 120, 121 crystallographic shear 74
composite building units or CBUs 583 crystal structure figures
compound semiconductor 261–264 a perovskite, cubic (+data) (Figure 1.49) 40
concentration gradient electrodes 567 AgI (Figure 13.7) 538
concentration quenching 284 Al2O3 corundum (+data) (Figure 1.30) 26
conductance 457 Ba2In2O5 (Figure 13.19) 554
conducting polymers 289, 438–441, 572 BaMnO3 (+data) (Figure 1.54) 44
conduction bands 213 BaZnF4 cis-layers (Figure 1.44) 36
conductivity 213, 397, 398, 401, 409, 421 Bi2WO6 (Figure 13.19) 554
configuration, structural 620 BiI3 (+data) (Figure 1.29) 26
configurational coordinate model 270 CaB6 (+data) (Figure 1.40) 32
configurational coordinates 637 CaC2 (Figure 1.18) 18
configurational entropy 58 CaF2 (fluorite) (+data) (Figure 1.31) 27
congruent melting 126 CaSi2 (Figure 1.36) 29
conjugated π bonding 181, 339, 437, 438, 441, 445 CaTe (+data) (Figure 1.40) 32
conjugated π network, see conjugated π bonding CdCl2 (+data) (Figure 1.29) 26
conjugated π system, see conjugated π bonding CdI2 (+data) (Figure 1.29) 26
continuous phase transition 137, 141 CdSb (Figure 1.18) 18
continuous random networks 622 Cu (+data) (Figure 1.21) 20
conventional superconductors 492 Cu2O (+data) (Figure 1.42) 33
converse piezoelectric effect 321 diamond (+data) (Figure 1.37) 30
conversion reactions 570 diamond (Figure 1.16) 17
Cooper pair 494 GaSe (Figure 1.17) 18
cooperative Jahn–Teller distortion 461, 465 Heusler alloys, full-, half- (+data) (Figure 1.34) 28
coordination polyhedra 32 HgBa2Can−1CunO2n+2 superconductors (Figure 12.18)
coordination polymers 605 517
coordination sequence 583 iodine, I2 (Figure 1.16) 17
coordinations 34 IrF4 network (Figure 1.44) 36
copper 397, 398, 409 iron, alpha (+data) (Figure 1.23) 21
core-shell electrodes 566 iron-pnictide superconductors (Figure 12.23) 521
correlation diagram 252, 253–255 La2CuO4 (Figure 12.15) 513
coulomb integral 171 LaFeO3 (+data) (Figure 1.52) 43
coulombic potential energy 154–155, 156, 157 LaNiO3 (+data) (Figure 1.52) 43
cracking 590, 593 lanthanum, alpha (+data) (Figure 1.22) 21
critical current 491, 498 LiFePO4 (Figure 13.29) 569
critical exponent 143 LiNbO3 (Figure 8.22) 338
critical field 490 lonsdaleite (+data) (Figure 1.37) 30
critical micelle concentration (CMC) 597 metallic elements (Figure 1.24) 22
critical phase matching 333 Mg (+data) (Figure 1.21) 20
critical point 122 Mg2Al2Si3O12 garnet (+data) (Figure 1.48) 40
critical pressure 122 MgAl2O4 spinel (+data) (Figure 1.47) 39
 Index 677

NaCl (+data) (Figure 1.26) 24 defect equilibria, behind oxygen nonstoichiometry

Nd2CuO4 (Figure 12.19) 518 (redox) 89
NiAs (+data) (Figure 1.13) 13 defect ordering, see defect clustering
NiAs (+data) (Figure 1.26) 24 defect thermodynamics 58
phosphorus, white (Figure 1.16) 17 degrees of freedom 120, 122
polonium, alpha (+data) (Figure 1.23) 21 densest packing of spheres, cubic, see cubic closest
(SnS)1.17NbS2 (Figure 2.14) 79 packing
ReO3 (Figure 1.39) 32 densest packing of spheres, hexagonal, see hexagonal
SiO2 cristobalite, beta (Figure 1.39) 32 closest packing
SiO2 tridymite, beta (Figure 1.39) 32 density of states 209, 223, 405, 497, 503, 564, 646,
SmCo5 (+data) (Figure 1.46) 38 647
SnCl2 (Figure 1.17) 18 Dexter electron transfer 283, 284
SnF4 trans-layers (Figure 1.44) 36 diagonal glide plane, see glide plane, diagonal
sodalite (+data) (Figure 1.57) 47 diamagnet 353
sodium β-alumina (Figure 13.9) 541 diamagnetic correction 671
SrSi2 (Figure 1.36) 29 Pascal’s constants 671
tellurium (Figure 1.16) 17 diamagnetic materials 356
TiF4 triple chains (Figure 1.44) 36 diamagnetism 367
TiO2 anatase (+data) (Figure 1.45) 36 Landau diamagnetism 367
TiO2 brookite (+data) (Figure 1.45) 36 Lenz’s law of induction 367
TiO2 rutile (+data) (Figure 1.45) 36 perfect diamagnet 367, 490
Y5O5F7 (Figure 2.16) 81 diamond 228–230, 303, 306
YBa2Cu3O7−δ (Figure 3.6 and 12.16) 98, 514 diamond glide plane, see glide plane, diamond
YBaFe2O5 (Figure 11.3) 460 Dieke diagrams 275
YCl3 (+data) (Figure 1.29) 26 dielectric loss 308, 312
zeolites A and X (Figure 14.2) 582 differential scanning calorimetry (DSC) 138
Zn4O(O2CC6H4CO2)3 (a MOF structure) (Figure 1.41) diffusion 99
33 diffusion coefficient 100
ZnS sphalerite (+data) (Figure 1.32) 27 diffusionless transition 71
ZnS wurtzite (+data) (Figure 1.32) 27 diffusivity 100, 106
CsCl-type structure 31 diffusivity and redox equilibria of defects 111
CsHSO4 547 diffusivity due to point defects, temperature dependence
cubic closest packing (ccp) 19 111
cubic stabilized zirconia CSZ 552 dimensionless susceptibility, see magnetic susceptibility,
cuprate charge reservoir 518 dimensionless
cuprate superconductors 511 Dirac cones, see Dirac points
Curie constant, C 368 Dirac points 446, 451
Curie law 368 direct-gap semiconductor 214, 263, 287
Curie temperature (electrical) 314, 319 directions in the lattice 3
Curie temperature (magnetic) TC 356, 377 discharge rate 568
Curie–Weiss law 369 discontinuous phase transition 137, 144
current density 396 disorder-induced metal-to-insulator transition 645
Czochralski method 342 displacive phase transition 135
distortion theorem 191, 327
databases in crystallography, see crystallographic domain boundary, see twin boundary
databases domain formation 378
Daumas Herold staging 557 domain structure 377
δ-Bi2O3 553 domains (ferroelectric) 315, 320, 338
de Broglie relation 410 donor doping 62, 94, 416
Deborah number 636 donor ionization energy 417, 420
Debye frequency 494, 497 donors 416, 419
Debye model 308 doping 416, 419, 439
decay time 268 Doppler effect, in Mössbauer spectroscopy 463
defect concentration 59 double exchange 382
defect equilibria 88, 515 double glide plane, see glide plane, double
678 Index

double perovskites 73, 273 ergodic system 636

down-conversion photoluminescence 268, 291 Euler strut 137
drift velocity 400 eutectic point 124
Drude equation 630 excess free energy, entropy, enthalpy 143, 146
Drude model 398–401, 403, 407, 409 exciton 290
drying agents 595 extended defect 66
zeolite 3A 595 extraordinary beam 332
zeolite 4A 595 extrinsic defects 61
zeolite 5A 595 extrinsic ionic conductivity 534
dye 245, 265
faradaic supercapacitors 571
easy axes 378 fast ion conductors 536
edge dislocation 66 faujasite 582
effective mass 403, 406, 407 Fe1−xO 77
effective moment, μeff 361 FePt 73, 139
effective nuclear charge 166, 175, 260 Fermi energy 210, 497, 503
Ehrenfest classification 136 Fermi level, see Fermi energy
electric displacement 304 Fermi velocity 409, 410, 447
electric solenoid 350 Fermi–Dirac distribution function 404, 415
electric susceptibility 303, 332, 628 ferrimagnetic materials 356
electrical conductivity (σi) due to charged defect 107 ferrimagnetism 385
electric-field intensity 396 ferroelectric ferromagnets 388
electrochemical capacitors 571 ferroelectric 313, 314, 314–319, 327, 330
electrochemical cells 529 ferroelectric perovskite 139
electrochemical characterization 568 ferromagnetic materials 356
electrochemical potential 562 half-metals 381
electrode 529 insulators 381
electrolyte 529 metals 381
electrolyte domain 552 ferromagnetism 377
electrolyte stability window 562 ferromagnets of spin-polarized bands 381
electromagnetic radiation 243 FeSe, see iron selenide
electron mobility 401, 406 FeTiO3, see ilmenite
electron–electron interactions 172, 186, 249, 429, 434 Fick’s first law 100
electroneutrality condition 90 Fick’s second law 100
electron–hole pair 290, 426 filling holes 22
electronic conductors 396 first Brillouin zone 204, 216, 217, 219
electron–phonon coupling 270, 271, 272, 286 first-order Jahn–Teller distortion 189, 280
electron–phonon interaction, see electron–phonon first-order phase transition 136, 144
coupling fluid catalytic cracking FCC 593
electron-spin g factor 358 fluorescence 268
electron-transport layer 289 flux 100, 125
elemental semiconductors 230–231, 263 Förster resonant-energy transfer (FRET) 282
elemental superconductors 487 fractional coordinates 2
elementary charge in SI 672 fragility index 640
enantiomeric excess 612 fragility strength 640
enantiomers 611 framework density 580
enantioselective sorption 612 Franck–Condon principle 271
enantioselective synthesis 610, 611 free energy, interfacial, see interfacial free energy
enantioselectivity 600 free volume 622, 639
energy landscape, see potential energy surface free-electron model 402–403, 405, 407, 412, 446
energy product, of ferromagnets, see maximum energy freeze-out regime 420
product Frenkel defects 58, 60, 533, 539
energy transfer 281–283, 284, 292 Frenkel intrinsic defect pair, see Figure 3.1
energy, potential of the magnetic moment 355 Fröhlich interaction 493
enhanced Pauli paramagnetism 383 frontier orbitals 180, 328
 Index 679

frustration, in cooperative magnetism 387 high-entropy alloys 650

fuel cells 532 highest-energy occupied molecular orbital 180, 222
fullerene superconductors, see C60 superconductors high-spin (HS) configuration 187, 254, 256
fundamental SI constants 672 high-spin state, see high-spin (HS) configuration
high-Tc cuprate superconductors 223
gallium nitride (GaN) 288, 314 hole doping 512, 516, 518
galvanic cells 529 hole filling, see filling holes
garnet 39 holes 407
gate dielectric 427 hole-transport layer 289
Gaussian error function (erf) 113 homeotypism 30, 45, 46
gel 585, 601 HOMO, see highest-energy occupied molecular orbital
gemstone 245, 256 hopping in electric field 106
generalized 8−N rule 17 hopping under concentration gradient 105
geometric frustration 648 hopping, random 103
germanium 230, 397, 418, 422 hopping/migration, in chemical potential gradient
Gibbs phase rule, see phase rule 105
Gibbs-energy barrier for migration 104, 111 hopping/migration, in electric potential gradient 106
Gibbs–Thomson effect 634 hopping/migration, under driving force 104
Gladstone–Dale relationship 630 host 268
glass forming range 635 Huang–Rhys parameter 272
glass transition 634 Hubbard model 429–431
glass transition temperature 544 Hubbard U 429, 437
glasses Hund’s rules 360, 367
covalent 620 Hund’s first rule 171, 251
random-packed 620 Hund’s second rule 251
glass-forming temperature 621, 635 Hund’s third rule 251
Glazer tilt (classification) 41 hybrid porous materials 605
glide plane 9 hydride superconductors 488
axial 9 hydrogen bond 545
diagonal 10 hydrogen fuel cell 532
diamond 10 hydrogen storage 56
double glide 10 hydrothermal synthesis 585
Goldschmidt tolerance factor 40, 44 hyper-aged organic glasses 638
graph 622 hysteresis 136, 356
graphene 221–223, 266, 445–447, 448, 450 hysteresis loop (ferroelectric) 316, 319
graphite intercalation chemistry 556 hysteresis loop (magnetic) 378, 379
Gruneisen parameter 604
identity 4
H2S superconductor, see hydride superconductor idiochromatic 245
half-metallic CrO2 476 ilmenite (FeTiO3) 192, 261
half-metallicity immiscibility dome 130
by Andreev reflection 475 incipient ferroelectric 312, 327
by positron annihilation 475 incommensurately modulated structure 78, 79, 149
by spin-resolved photoemission 476 incongruent melting 126, 342
with spin-polarized transport 472 indigo (C15H9N2O) 265
half-metals of itinerant electrons 480 indirect-gap semiconductor 214, 263, 287
hamiltonian operator 161 infinite-layer structure 517
hard (ferro)magnets 379 infinitely adaptive structure 80
hcp, see hexagonal closest packing infrared absorption band (in glass fibers) 631
heat capacity 138, 144, 146, 642 intensity of the magnetic field, H 352
Heusler alloys 477 interaction integral 171
hexagonal closest packing (hcp) 19 intercalation compounds 555
hexagonal perovskites 44 guest species 555
HfO2 428 host structure 555
HgBa2Can−1CunO2n+2 511, 516 interdiffusivity 106
680 Index

interfacial free energy 634 KDP, see KH2PO4

interference term 645 KGe Zintl phase 48
internal conversion 268 KH2PO4 (KDP) 336
internal field, HW 379 klassengleiche subgroup 73, 660
International Tables for Crystallography 656 Koch cluster 77
internal friction 644 Kondo effect 387
interpenetrated network 606 Kröger–Vink notation 65, 87
interstitial 54, 534 KTiOPO4 (KTP) 336
interstitialcy mechanism 534 KTP, see KTiOPO4
intersystem crossing 268, 279, 291 KxC 556
intrinsic defect pairs, Schottky, anti-Schottky, Frenkel,
anti-Frenkel, see Figure 3.1 La1/2Ca1/2MnO3 charge and orbital ordering 469
intrinsic defect 55 La1−xCaxMnO3 phase diagram 468
intrinsic ionic conductivity 535 La1−xSrxGa1−yMgyO3−δ 553
intrinsic regime 420 La2CuO4 512, 513
invariant point 122, 125 La2Mo2O9 555
inverse opal 601 La2CuO4+δ 512
inverse spinel 38 La2−xSrxCuO4 511, 512, 513
inversion center 4 LaGaO3 553
Ioffe–Regel condition 645 LaMnO3 465
ion exchange 593 LaMnO3+δ nonstoichiometry control 466
ion-exchange isotherms 594 Landau theory 140
ionic conductivity 533 Landé factor 363
ionic conductors 396 Langevin function 370, 666
ionic radii 193, 194 LaNi5 56
ionic-radii variance effect, see variance effect LaO1−xFxFeAs 521
iron monoxide (wüstite), see Fe1−xO Laporte selection rule 256
iron platinum, see FePt Larmor precession 363
iron selenide 521 lattice 2
iron pnictide superconductors 521 lattice diffusion 101
isomerization 590 lattice energy 156, 158
isoreticular 608 lattice metrics, see metrics of lattices
isotope effect 493, 503, 511 lattice modes 147
isovalent substitution 61 lattice parameters 2, 8
itinerant ferromagnetism 382 lattice planes, equidistant set 3
ITQ-33 584 lattice-formation enthalpy 158
LCAO, see linear combination of atomic orbitals
Jagodzinski–Wyckoff notation 19 lead acid battery 531
Jahn–Teller distortion, see first-order Jahn–Teller lead fluoride, see PbF2
distortion lever rule 124
Jahn–Teller theorem 188 Li(Ni~0.33Mn~0.33Co~0.33)O2, see NMC
j–j coupling 249, 275, 664 Li4Ti5O12 570
jump and thermal energy kT 101 LiB3O5 340, 341, 342
jump balance 103 LiBH4, see lithium borohydride
jump directional probability, pdir 102 LiCoO2 (lithium cobalt oxide) 565
jump frequency, pBν 101 LiFeAs 521
jump Gibbs-energy barrier, Δ‡Gm 101 LiFePO4 569
jump probability, pB 101 ligand-field splitting 184, 186, 234, 247, 248, 277
jump progression rate, rprogression 102 ligand-field theory 247
jump site availability, pavail 102 light-emitting diodes (LEDs) 287–288, 291, 422, 426
LiNbO3 324, 338
K2NiF4 structure 512 line defect 66
Kamerlingh Onnes, Heike 486 line emitters 274, 284
Kauzmann rule 636 linear combination of atomic orbitals 170, 200, 202
Kauzmann temperature 638 liquidus 124
 Index 681

LISICON 543 of 4f ions 366

lithium air battery 570 magnetic ordering, temperature 356
lithium borohydride (LiBH4) 57 magnetic permeability, relative, μr 354
lithium cobalt oxide, see LiCoO2 magnetic quantum number 162
lithium intercalation 559 magnetic structure 373, 377
lithium-ion batteries 559, 566 magnetic susceptibility
lithium iron phosphate, see LiFePO4 dimensionless 354, 355
lithium–sulfur battery 570 mass, χm 354
lithium–TiS2 battery 560, 568 molar, χmol 354
LiVPO3F 570 volume, χv 353
Loewenstein’s rule 587 magnetic unit cell 374
longitudinal mode 148 magnetic-flux density, see magnetic induction
Lorenz–Lorentz relationship 629, 630 magnetic-moment vector 351
lowest-energy unoccupied occupied molecular orbital 180, magnetism, electron movements and their quantization 357
222 magnetism, unit systems in use 355
low-spin (LS) configuration 187 magnetite 458
low-spin state, see low-spin (LS) configuration magnetization of half-metals 475
luminescence 168, 267, 268 magnetization, M 353
bioluminescence 267 magnetizing field, see intensity of the magnetic field
cathodoluminescence 267 magnetocrystalline energy 378
chemiluminescence 267 magnetoelectric multiferroics 388
electroluminescence 267, 287 magnetoresistant device 457
photoluminescence 267, 268 magnetostatic energy 378
piezoluminescence 267 magnetostriction 378
sonoluminescence 267 magnetostrictive energy 378
thermoluminescence 267, 268 magnetotransport 457
triboluminescence 267 manganite perovskites, contour map of ordering
luminous efficacy 291 isotherms 471
luminous flux 291 martensitic transformations 71
LUMO, see lowest-energy unoccupied occupied mass susceptibility, see magnetic susceptibility, mass
molecular orbital maximum energy product (BH)max 379
MCM-41 597
Mackay icosahedra 649 mean free path 400, 409
macroporous 579, 601 Meissner or Meissner–Ochsenfeld effect 490, 492
Madelung constant 155, 156 mercury superconductivity 487
Magnéli phases 74 mesoporous materials 46, 579, 597
magnesium diboride, (MgB2) 499 mesoporous zeolites 589
magnesium hydride (MgH2) 56 mesostructured zeolites 600
magnetic (dipole) moment, µ 350 metal–organic framework (MOF) 31, 605
magnetic constant 353 metallic glasses 647
magnetic field strength, see intensity of the metallic glasses, ferromagnetic 650
magnetic field metal-oxide-semiconductor field-effect transistor
magnetic induction, B 352 (MOSFET) 426
magnetic measurements 670 metrics of lattices 8
magnetic moment, absolute, μabs 358 MgB2, see magnesium diboride
magnetic moment, ground state 359 MgH2, see magnesium hydride
for actinoids, j–j coupling 360 micelles 597
in free d-metal atom, under Russell–Saunders coupling microporous materials 46, 579
scheme 360 microstates 250, 275, 365
in free f-metal atom, under Russell–Saunders coupling microwave resonator 311
scheme 367 migration/hopping, see hopping/migration
magnetic moment mirror plane 4
spin-only 361 miscibility gap 569
z-projection, μz 358 mixed anion 80
of 3d ions 363 mixed ionic and electronic conductor (MIEC) 555
682 Index

MO diagram, see molecular-orbital diagram neutron diffraction on ordered magnetic moments

MO, see molecular orbital 376
mobility 108 nickel–cadmium battery 531
mobility edge 646 Niggli formula 34
mobility gap 646 nitinol 71
MOF applications NMC 566
asymmetric catalysis 611 nodal planes 164, 176, 182, 265, 442
CO2 capture and storage 609 node 164, 201
dynamic frameworks 610 non-centrosymmetric crystals 313, 324, 332
heterogeneous catalysis 610 noncritical phase matching 334, 341
hydrogen storage 609 non-ergodic system 636
physisorption 608 non-radiative recombination 426
supported catalysis 610 non-steady-state diffusion 100
water capture 609 non-steady-state diffusion, of point defects 112
MOF, see metal–organic framework nonstoichiometry
molar gas constant in SI 672 narrow 87
molar susceptibility, see magnetic susceptibility, molar oxidative 88
molar volume of ideal gas in SI 672 reductive 88
molecular magnets 389 normal spinel 38
molecular-orbital diagram 169–186, 211, 325, 328 n-type cuprate superconductors 517
molecular orbital 169 n-type semiconductor 419, 422
molecular sieves 596 nucleation
molecular superconductors 505 nucleation, heterogeneous 634
monotectic point 130 nucleation, homogeneous 634
morphotropic phase boundary (MPB) 323
Mössbauer spectroscopy 463 occupiable volume 581
motif 2, 222, 225 octahedral holes 22
Mott–Hubbard insulator 431, 434, 435, 436, 518 fractional filling 23
Mott–Hubbard transition 501, 502 full filling 23
multi-phonon emission 273 octahedral ligand-field splitting (Δ), see ligand-field
multiplexing (in glass fibers) 632 splitting
multiplicity 250 octahedral tilting 41, 194, 236
olivine 569
NaAl11O17 540 operation of symmetry, see symmetry operations
NaAlH4, see sodium aluminum hydride optic axis 333
NaCl conductivity 535 optical dielectric constant 307
NAFION 544, 546 optical fibers 631–633
NASICON 542 orbital angular-momentum quantum number 162, 250
NaTl Zintl phase 47, 48 orbital ordering 461
natrolite 603 orbital-degeneracy removal 462
Nb3Ge superconductor 487 by charge disproportionation 462
Nd2CuO4 517 by electron itinerancy 462
Néel temperature, TN 356, 372 by Jahn–Teller distortion 462
negative thermal expansion 602 order parameter 139, 140
NEMCA reactor 532 order, long-range 620, 642
Nernst equation 531, 550 order, short-range 620
Nernst–Einstein equation 107, 534 order–disorder phase transition 135
nets 606 ordinary beam 332
network augmenting 31 organic light-emitting diodes (OLEDs) 287, 289–291
network expansion 31, 608 organic magnets 390
network similarity (site ordering), see similarity (net- organic superconductors 505
work-based) orientational disorder 136
network-forming oxides 625 Ostwald ripening 37
network-modifying oxides 625 Ostwald step rule 37, 649
networks binodal (N,M-connected nets), see binodal overlap integral 171, 173, 174–175, 177
networks (N,M-connected nets) oxidation catalyst 593
 Index 683

oxide fluoride 80 phase-change materials 641

oxide-ion conductors 549 phonons 147, 149, 642
oxyfluoride, see oxide fluoride phosphor-converted light emitting diode 293–294
oxygen control 515 phosphorescence 268, 275, 280, 281, 290
oxygen nonstoichiometry (redox defect equilibria behind phosphors 267, 268, 273, 276, 284, 292–293
it) 89 photoionization 268
oxygen sensor 550 photon 243
oxygen, bridging 625 photovoltaic cell 426
oxygen, non-bridging 625 physisorption 56
oxygen-nonstoichiometry control in oxides 97 piezoelectric effect 321
piezoelectricity 313, 321
packing of spheres 19 pigment 245, 257–258, 259, 265
pair distribution function 318, 620, 621, 624 pinning 66
pairing energy 495 planar defects 67
pairing-s/p/d/f-wave, see s/p/d/f-wave pairing Planck constant in SI 672
paraelectric 314, 318, 320, 323 plastic crystals 547
parallel plate capacitor 302 p–n junction 287, 422–425, 427
paramagnet 353 built-in potential 424
paramagnetic materials 356 depletion region 424, 425
paramagnetism 367 forward bias 425
Curie paramagnets 368 reverse bias 425
Curie–Weiss paramagnets 368 point defect 54
Pauli paramagnets 368 point defects in a pure oxide 87
parent structure 149 point group 3
partial pressure of oxygen, pO2 91, 97 point of integer structure 92, 96, 98
Pauli exclusion principle 162, 171 point of integer structure, see Figures 3.2, 3.4, 3.5
Pauli paramagnetism, enhanced 371 point of integer valence 92, 96, 98
Pauli paramagnetism, temperature independent point of integer valence, see Figures 3.2, 3.4, 3.5
371 point-defect compensations upon oxidation and
Pb(Mg1/3Nb2/3)O3 320 reduction, see Figure 3.1
PbF2 (lead fluoride) 539 point-defect movements 101
PbZr1−xTixO3 (PZT) 323 point-group order 4
Peierls’ distortion 212, 228, 439, 507 point-symmetry elements 4
Peierls’ theorem 212 Poisson’s ratio 604
PEM 546 polar axis 313, 337
penetration depth 113 polar materials 313, 314, 321
pentacene 266, 441–443 polar nanoregions (PNR) 320
PEO, see polyethylene oxide polarizability 278, 304, 305, 309–311, 629
perfect conductor 490 dipolar polarizability 305
peritectic point 126 electronic polarizability 305
peritectic reaction 126 ionic polarizability 305
permeability of free space, see magnetic constant polarization 160, 303, 304, 305, 314, 331, 628
perovskites 40, 233–236, 320, 434–437 polaron 440, 510
microwave dielectrics 312 polonium 227–228
proton conductors 548 polyacetylene 212, 397, 438–441, 442
superconductors 509 polycyclic aromatic hydrocarbons 441
persistent phosphors 268 polyelectrolytes 543
phase diagram 120 polyethylene oxide 544
phase matching 332–334 polyhedral connectivities 34
phase rule 97, 120, 123 edge sharing 35
phase separation 569 face sharing 35
phase transition thermodynamics 142 vertex sharing 34
phase transition 135 polymer-electolyte membrane, see PEM
phase velocity 148 polymeric cation conductors 543
phase 120 polymer-salt complexes 543
684 Index

polymorphism of nanocrystals 37 ReO3 231–233, 234, 235, 397, 605

polymorphism 68 repulsive potential energy 156, 157
polytypes 68 residual resistivity 411
porous oxides 599 resistivity 396, 645
porphyrin 266 reticular chemistry 607
position, see special positions or general positions RGB model 244
potential energy surface (PES) 636, 637, 638, 641 rhenium trioxide, see ReO3
primary batteries 531 rigid unit modes (RUMs) 602
primitive Bravais lattices 6 ring 623
principal quantum number 162, 164, 418 rocking chair battery 566
probability density 164, 202, 645 rotation axes 4
proton conductors 545 rotational axes, see rotation axes
proton-exchange membranes, see PEM rotational symmetry 3
Prussian blue (Fe4(Fe(CN)6)3·xH2O) 257, 261, 605 rotational-symmetry elements 5
pseudocapacitors 571 rotoinversion axis 4
pseudoelasticity 72 rubrene (5,6,11,12-tetraphenyltetracene, C42H28) 443
pseudogap regime 519 rule of parsimony 34
p-type cuprate superconductors 516 Russell–Saunders coupling 249, 275
p-type semiconductor 419, 422
pyroelectricity 313, 314, 330 s/p/d/f-wave pairing 495
PZT, see PbZr1−xTixO3 SALC, see symmetry-adapted linear combinations
sapphire (Al2O3:Fe2+,Ti4+) 260, 288
quality factor 312 saturation moment, μsat 361
quantum efficiency 268 saturation polarization 316
quantum harmonic oscillator 271 saturation regime 420
quasicrystals 649 saturation resistivity 645
quasi-phase matching 334, 338 Schottky defect 55, 60, 533
quenching 621, 648 Schottky intrinsic defect pair, see Figure 3.1
screw axis 9
radial distribution function 164, 169 screw dislocation 66
radial node 164 secondary batteries 531
radiative recombination 287, 425 secondary building units (SBUs) 583
Ragone plot 572 second-harmonic generation 313, 330, 332, 333, 334
Ramsdell symbol 19 second-order Jahn–Teller (SOJT) distortion 189, 280,
random walk (electrons) 645 325–330
random walk 103 second-order phase transition 136
random-close packing of spheres 621 segmental motion 544
Rayleigh scattering 631 Seitz operator 9
RbAg4I5 538 Seitz symbol 9
reciprocal-space lattice 204, 216, 217, 218, 219 selectivity 594
reconstructive phase transition 135 self diffusion 104, 106
rectification 424 self-activating phosphors 269, 284
redox compensation 62 semimetal 223, 446
refraction 627 sensitizer 268, 282, 283, 285, 292
anomalous 628 shape-memory alloys 71
normal 628 shape-selective catalysis 591
refractive index 307, 311, 330, 332, 333, 627, 628 product-selective catalysis 592
regioselectivity 600 reactant-selective catalysis 591
relative coordinates, see fractional coordinates transition-state-selective catalysis 592
relative dielectric permittivity 303, 417, 418 shear plane 74, 75
relaxation time 400, 409 shell 162
relaxor ferroelectric 320 Si:Al ratio 587
remanence, see remanent magnetization silicates, oligo-, cyclo-, catena-, phyllo-, tecto- 45
remanent magnetization, Mremanent 379 silicate nomenclature 661
remanent polarization 316 branchedness 661
 Index 685

multiplicity 661 subgroups and supergroups 659

dimensionality 661 subgroups, klassengleiche 660
periodicity 661 subgroups, maximal isomorphic, of the lowest index 660
silicon 230, 306, 397, 415, 416, 418, 422, 424, 427 subgroups, maximal non-isomorphic 660
silver cobalt cyanide see Ag3[Co(CN)6] subgroups, proper 659
silver iodide, see AgI subgroups, translationengleiche 660
silver 397, 398, 414 subgroups’ dilution index 660
similarity (network-based) 30 supergroups, minimal non-isomorphic 660
single-molecule magnets 391 symmetry elements, graphical symbols 656
SiO2 cristobalite 135 symmetry operations 658
SiO2 quartz 135 symmorphic 8
SiO2 tryidymite 135 Wyckoff letter 658
SiO2 121 space-group symbols 11
site ordering 30, 72 standard 11, 12
diamond–sphalerite, see Figure 1.38 spectroscopic term, see term symbol
lonsdaleite–wurtzite, see Figure 1.38 speed of light in vacuum in SI 672
polonium–sodium chloride, see Figure 1.38 sphere packing, see packing of spheres
site, see Wyckoff site spin-density wave 506
Slater–Pauling rule of localized ferromagnetism 478 spin flip, in antiferromagnet 373
slip plane 66 spin glass 357, 387
SmCo5 structure type, see Figure 1.46 spin glass, reentrant 388
SmO1−xFxFeAs 521 spin polarization, P 458
Sn1.17NbS3.17 78 spin quantum number 162, 250
Snell’s law 627 spin selection rule 256
sodalite cage 47, 581 spin valve 457
sodalite 257, 581 spinel 38
sodium aluminum hydride (NaAlH4) 56 spin-only moment, strong-field limit 362
sodium β-alumina 540 spin-only moment, weak-field limit 362
sodium sulfur battery 541 spin–orbit coupling 249, 291, 359
soft magnets 379 strong-field limit 363
soft mode 149 weak-field limit 362
sol 601 spin-pairing energy 186, 375
sol-gel synthesis 600 spintronics 457
solid acids 588, 589 spontaneous magnetization 377
solid-oxide fuel cell (SOFC) 549 SQUID magnetometer 670
solid solubility 94 Sr0.1La0.9AlO3 553
solid solution 55, 63, 121, 128, 568 Sr0.4K0.6BiO3 511
solid-solution limit 65 Sr0.73CuO2 80
solid electrolyte interphase (SEI) 562 SrFeO3 462
solidus 124 SrGa2 Zintl phase 48
soliton 440 SrTiO3 147, 149, 194, 233, 236, 303, 312, 325, 326, 327,
solvus 129 434
space groups 8, 656 stacking faults 67, 68
absent or extinct Bragg reflections 658 stacking sequence 19
asymmetric unit 658 staging 557
coordinates 658 stage-1 and stage-2 557
diagram 656 standard atmosphere in SI 672
generators selected 658 standard-state pressure in SI 672
multiplicity 658 static dielectric constant 307
non-symmorphic 8 steady-state diffusion 100
order 659 Stokes shift 270, 271, 272, 277, 280
origin placement 658 strain 140
positions 658 strongly correlated materials 429, 520
reflection conditions 658 structure directing agent 587
site symmetry 658 structures of metals 22
686 Index

stuffed silica 580 time-independent Schrödinger equation 161, 402

subshell 162, 167, 168, 274, 360 TiOx 75, 76
substitutional disorder 55, 61, 136 TiS2 559
supercapacitors 571 titanium disulfide, see TiS2
supercell 31 TlBa2Can−1CunO2n+3 516
superconducting dome 504 TlBa2Can−1CunO2n+4 516
superconducting gap 493, 496, 497 TMTTF 508
superconducting levitation 492 top seeded solution growth 342
superconducting transition 498 topology of glasses 622
superconductivity 486 total angular-momentum quantum number 251
supercooled state 122 total external reflection 628
supercooling 634 total internal reflection 627, 628, 631
supercritical fluid 122 transference number (ti) of a point defect 107
superelasticity 72 transition temperature 135
superexchange 374 translational symmetry 2
superferromagnets 384 translationengleiche subgroup 70, 660
superionic conductors see fast ion conductors transport half-metallicity 475
superstructure 31 transverse mode 147
supported catalysts 599 triangle rule 132
surface structure (versus bulk structure) 22, 37 tricritical transition 147
switching field 316 triple point 122
symmetry breaking, see symmetry lowering TS-1 593, 600
symmetry elements rotational, see rotational-symmetry TTF[Ni(dmit)2]2 509
elements TTF–TCNQ 444, 506
symmetry elements translational, see Bravais lattices tungsten trioxide, see WO3
symmetry forbidden 175, 189, 226 tunneling magnetoresistance 472
symmetry lowering 137, 138 between two separated epitaxial films 473
symmetry operations 3 in a powder compact 473
symmetry-adapted linear combinations 178–179, 183, turbostratic disorder 68
184, 202 twinning 68
system 120 deformation twins 69
growth twins 69
Tanabe–Sugano diagram 281 transformation twins 70
temperature coefficient of resonant frequency 312 twin boundary 69
tempering of glasses 626 twin categories 69
template molecules 586, 587 twin component 68
tension effect 602 twin domain 68
term symbol 250, 252 twin law 68
ternary phase diagram 131 twin variants 70
tessellation (in 3D space) 648 two-level tunneling model 643
tetracyanoquinodimethane (TCNQ) 444 type-I superconductor 490
tetrahedral holes 22 type-II superconductor 491
fractional filling, see Figure 1.5
full filling, see Figure 1.5 ultrastable zeolite Y 588
tetrathiofulvalene (TTF) 444 unconventional superconductors 492
thermal conductivity 642 under- and over-doping 519
thermal expansion 602 uninodal network 28
thermal quenching temperature 273 unit cell 2
thermal quenching 272–274, 277 unit conversions in magnetism 355, 356
thermal stimulation of luminescence, see unit-cell parameters 2
thermoluminescence universal low temperature states 639, 642
thermal vibration 101 up-conversion photoluminescence 268, 285
ThrCr2Si2 structure 522 upper consolute temperature 130
tight-binding methods 202 Urbach tails 646
tiling 583 UV tail (in glass fibers) 631
 Index 687

vacancy clustering 58, 66, 77 WO3−x 74

vacancy trapping 552 Wyckoff site 12
vacancy 54 general position 12
vacuum permeability, see magnetic constant special positions 12
valence bands 213
valence mixing, preservation of degeneracy 461 xerogel 601
valence-mixed state 459 X-ray absorption fine structure (EXAFS) 318
valence-sum rule 191, 192
van der Waals forces 159, 447, 559 Y3Al5O12, see yttrium aluminum garnet
vanadium disulfide, see VS2 YBa2Cu3O7 synthesis 515
variance, see degrees of freedom YBa2Cu3O7−δ 98, 108, 112, 134, 511, 513
VECA (valence-electron count per atom A) 17 YBaFe2O5, charge ordered 460
Vegard’s law 63 YBaFe2O5, valence mixed 460
positive deviation 64 YBCO superconductors, see YBa2Cu3O7−δ
negative deviation 64 YO1−mF1+2m 80
vertex augmenting, see vertex decoration yttria-stabilized zirconia (YSZ) 552
vertex decoration 31 yttrium aluminum garnet (Y3Al5O12) 293, 310
vertex symbol 28, 583 yttrium barium copper oxide, see YBa2Cu3O7−δ
Verwey transition 459
vibration (thermal), see thermal vibration Zachariasen rules 625
vibronic coupling, see electron–phonon coupling Zeeman effect 358
virtual phonon 494, 495 Zeeman splitting, see Zeeman effect
Vogel–Fulcher–Tammann equation 544, 639, 641 zeolite catalysts 588
Vogel–Fulcher–Tammann temperature 640 zeolite A 582
volume susceptibility, see magnetic susceptibility, volume zeolites 46, 579
vortex pinning 492 zeolite X 583
vortex state 491, 492 zeolite Y 583
VPI-5 596 zeotypes 579
VS2 558 zero resistance 486, 498
zero-field splitting 365
Wadsley defect 74 zero-point energy 160
water softening 594 Zintl phases 47
wave vector 148, 217, 219, 408, 494 Zintl–Klemm concept 47
wavefunction 161–166, 169, 170, 171, 202, 203, 204, 217 zirconia (ZrO2) 552
weak backscattering 646 zirconium tungstate (ZrW2O8) 130, 602
Weiss constant, θ 369 ZrO2, see zirconia
wide nonstoichiometry 98 ZrW2O8, see zirconium tungstate
WO3 74, 234, 235, 326 ZSM-18 587