Volume 1

D. A. Lavis and G. M.
Bell Department of Mathematics King's College, London
Statistical Mechanics of Lattice Systems

Volume 1: Closed-Form and Exact Solutions
March 16, 1999
Springer-Verlag
Berlin Heidelberg New York London Paris Tokyo Hong Kong Barcelona Budapest
Preface
Most of the interesting and di cult problems in statistical mechanics arise when the constituent particles of the system interact with each other with pair or multiparticle energies. The types of behaviour which occur in systems because of these interactions are referred to as cooperative phenomena giving rise in many cases to phase transitions. This book and its companion volume Lavis and Bell 1999, referred to in the text simply as Volume 2 are principally concerned with phase transitions in lattice systems. Due mainly to the insights gained from scaling theory and renormalization group methods, this subject has developed very rapidly over the last thirty years.1 In our choice of topics we have tried to present a good range of fundamental theory and of applications, some of which re ect our own interests. A broad division of material can be made between exact results and approximation methods. We have found it appropriate to include some of our discussion of exact results in this volume and some in Volume 2. The other main area of discussion in this volume is meaneld theory leading to closedform approximations. Although this is known not to give reliable results close to a critical region, it often provides a good qualitative picture for phase diagrams as a whole. For complicated systems some kind of meaneld method is often the only tractable method available. In Volume 2 we complement the work in this text by discussions of scaling theory, algebraic methods and the renormalization group. Although these two volumes individually contain a less than comprehensive range of topics, we have attempted to ensure that each stands alone as treatment of its material. References to Volume 2 in this volume are normally given alongside other alternative sources. The work in these two volumes was developed in part from courses given by George Bell and myself in the Mathematics Departments of the University of Manchester Institute of Science and Technology and Chelsea and King's Colleges, London. Questions and comments by students in these courses have helped to reduce errors and obscurities. We have also bene ted from the exchange of ideas with friends and co-workers. We are grateful to Professor G. S. Rushbrooke for his careful reading of the original typescript and for his many helpful and instructive comments.
1
An account of the historical development of the subject from its beginnings in the nineteenth century is given by Domb 1985.
VI
Preface
This volume is a revised version of the text rst published by Ellis Horwood Bell and Lavis 1989. In April 1993, before the second volume was complete, my friend and co-author George Bell suddenly died. The consequent delay in the completion of Volume 2 made the production of a revised version of Volume 1 desirable. I am very grateful to Professor Elliott Lieb, who encouraged and facilitated this new edition by Springer. I should also a like to thank Rachel George, who produced a working L TEX version from the published text of the rst edition and created postscript diagrams based on the original drawings by Debbie Bell. Geo rey Bellringer of King's College Library provided me with valuable help in compiling the bibliography. Finally I should like to express my gratitude to Professor Wolf Beiglbock and the sta at Springer for their patience and advice. London, December 1998.
David Lavis
Table of Contents
1. Introduction to Thermodynamics and Phase Transitions : : 1

Thermodynamic Variables: Simple Fluids : : : : : : : : : : : : : : : : : : Change of Variable and Thermodynamic Potentials : : : : : : : : : Response Functions and Thermodynamic Relations : : : : : : : : : Magnetic Systems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : Stationary Properties of Thermodynamic Functions : : : : : : : : : Phase Equilibrium in the Van der Waals Gas : : : : : : : : : : : : : : The Field-Extensive Variable Representation of Thermodynamics : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1.8 The Field-Density Representation of Thermodynamics : : : : : : 1.9 General Theory of Phase Equilibrium : : : : : : : : : : : : : : : : : : : : : 1.9.1 A One-Component Fluid : : : : : : : : : : : : : : : : : : : : : : : : : : 1.9.2 Azeotropy : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1.10 Classical Theory and Metastability : : : : : : : : : : : : : : : : : : : : : : : 1.10.1 Metastability in a One-Component Fluid : : : : : : : : : : : : 1.10.2 The Experimental Situation : : : : : : : : : : : : : : : : : : : : : : : Examples : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1 3 4 6 7 9
12 16 18 19 22 23 25 27 28 31 33 33 35 39 39 39 40 40 42 44 47 49 53
2. Statistical Mechanics and the One-Dimensional Ising Model : : : : : : : : : : : : : : : : : : : : 31

2.1 The Canonical Distribution : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 2.1.1 The Thermodynamic Limit : : : : : : : : : : : : : : : : : : : : : : : : 2.1.2 Kinetic and Con guration Variables : : : : : : : : : : : : : : : : 2.2 Distributions in General : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 2.3 Particular Distributions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 2.3.1 The Constant Magnetic Field Distribution : : : : : : : : : : : 2.3.2 The Constant-Pressure Isobaric Distribution : : : : : : : 2.3.3 The Grand Distribution : : : : : : : : : : : : : : : : : : : : : : : : : : : 2.3.4 Restricted Distributions for Lattice Models : : : : : : : : : : 2.4 Magnetism and the Ising Model : : : : : : : : : : : : : : : : : : : : : : : : : : 2.4.1 The One-Dimensional Ferromagnet in Zero Field : : : : : 2.4.2 The One-Dimensional Ferromagnet in a Field : : : : : : : : 2.5 Fluctuations and Entropy : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 2.6 The Maximum-Term Method : : : : : : : : : : : : : : : : : : : : : : : : : : : :
VIII
Table of Contents
2.6.1 The One-Dimensional Ising Ferromagnet : : : : : : : : : : : : 2.6.2 The General Distribution : : : : : : : : : : : : : : : : : : : : : : : : : : 2.7 A One-Dimensional Model for DNA Denaturation : : : : : : : : : : Examples : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3.1 The Ising Model Ferromagnet : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3.1.1 Free Energy and Magnetization : : : : : : : : : : : : : : : : : : : : 3.1.2 Fluctuations in Zero Field : : : : : : : : : : : : : : : : : : : : : : : : : 3.2 Interpretations of the Mean-Field Method : : : : : : : : : : : : : : : : : 3.2.1 Many-Neighbour Interactions and the Lebowitz Penrose Theorem : : : : : : : : : : : : : : : : 3.2.2 A Distance-Independent Interaction : : : : : : : : : : : : : : : : 3.3 The Mean-Field Method for a More General Model : : : : : : : : : 3.4 Critical Points and Critical Exponents : : : : : : : : : : : : : : : : : : : : 3.5 Scaling and Exponent Relations : : : : : : : : : : : : : : : : : : : : : : : : : : 3.6 Classical Critical Exponents : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3.6.1 The Ising Model Ferromagnet: Mean-Field Approximation : : : : : : : : : : : : : : : : : : : : : : : : 3.6.2 The Van der Waals Gas : : : : : : : : : : : : : : : : : : : : : : : : : : : Examples : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :
54 56 57 63 67 68 71 76 76 77 78 79 84 86 87 88 90
3. The Mean-Field Approximation, Scaling and Critical Exponents : : : : : : : : : : : : : : : : : : : : : : : : : : : 67
4. Antiferromagnets and Other Magnetic Systems : : : : : : : : : : : 93
4.1 The One-Dimensional Antiferromagnet : : : : : : : : : : : : : : : : : : : : 93 4.2 Antiferromagnetic Ising Models : : : : : : : : : : : : : : : : : : : : : : : : : : 94 4.3 Mean-Field Theory : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 99 4.3.1 The Paramagnetic State : : : : : : : : : : : : : : : : : : : : : : : : : : : 100 4.3.2 The Antiferromagnetic State : : : : : : : : : : : : : : : : : : : : : : : 101 4.3.3 The Simple Antiferromagnet : : : : : : : : : : : : : : : : : : : : : : : 103 4.4 Metamagnetism: Tricritical Points and Critical End-Points : : : 105 4.5 Ferrimagnetism: Compensation Points : : : : : : : : : : : : : : : : : : : : : 110 4.5.1 Zero Field : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 113 4.5.2 Non-Zero Field : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 114 Examples : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 117 5.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 119 5.2 The One-Dimensional Lattice Gas and the Continuous Limit : 121 5.2.1 The Lattice Gas : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 121 5.2.2 The Continuum Limit : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 124 5.3 The Simple Lattice Gas and the Ising Model : : : : : : : : : : : : : : : 125 5.4 Phase Separation in the Simple Lattice Gas : : : : : : : : : : : : : : : : 127 5.5 A One-Dimensional Water-Like Model : : : : : : : : : : : : : : : : : : : : 128 Examples : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 132
5. Lattice Gases : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 119
Table of Contents
IX
6. Solid Mixtures and the Dilute Ising Model : : : : : : : : : : : : : : : 135

6.1 The Restricted Grand Partition Function : : : : : : : : : : : : : : : : : : 135 6.2 Binary Mixtures : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 136 6.2.1 The Equivalence to the Ising Model : : : : : : : : : : : : : : : : : 137 6.2.2 The Equivalence to a Lattice Gas : : : : : : : : : : : : : : : : : : 138 6.3 Order Disorder on Loose-Packed Lattices : : : : : : : : : : : : : : : : : : 138 6.4 The Order Parameter and Landau Expansion : : : : : : : : : : : : : : 141 6.5 First-Order Sublattice Transitions : : : : : : : : : : : : : : : : : : : : : : : : 144 6.6 The Equilibrium Dilute Ising Model and Equivalent Models : : 147 6.6.1 Model I: The Equilibrium Dilute Ising Model : : : : : : : : 147 6.6.2 Model II: The Ising Model Lattice Gas : : : : : : : : : : : : : : 149 6.6.3 Model III: The Symmetrical Ternary Solid Mixture : : : 149 6.6.4 Model IV: The Symmetrical Lattice Gas Mixture : : : : : 150 6.6.5 Other Models : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 151 6.6.6 Applications : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 151 6.7 Mean-Field Theory and the Dilute Ising Model : : : : : : : : : : : : : 151 6.8 Multicritical Points in the Dilute Ising Model : : : : : : : : : : : : : : 154 6.9 Multicritical Phenomena with Additional Thermodynamic Dimension : : : : : : : : : : : : : : : 158 6.10 The Unsymmetrical and Completely Symmetrical Models : : : : 163 6.11 Alternative Forms for the Dilute Ising Model : : : : : : : : : : : : : : : 166 6.11.1 Model A: Equilibrium Bond Dilution : : : : : : : : : : : : : : : 166 6.11.2 Model B: Random Site Dilution : : : : : : : : : : : : : : : : : : : : 166 6.11.3 Model C: Random Bond Dilution : : : : : : : : : : : : : : : : : : : 167 6.11.4 Model D: Equilibrium Site Dilution " 0 : : : : : : : : : : : 168 Examples : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 169 7.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 173 7.2 A First-Order Method Using a General Site Group : : : : : : : : : 174 7.2.1 Equivalent Sites : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 174 7.2.2 Sublattice Ordering : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 177 7.3 The Pair Approximation and the Ising Model : : : : : : : : : : : : : : 178 7.3.1 Zero Field : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 179 7.3.2 The Critical Region : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 180 7.3.3 The Linear Lattice : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 181 7.4 Phase Transitions in Amphipathic Monolayers : : : : : : : : : : : : : 181 7.5 A Lattice Gas Model for Fluid Water : : : : : : : : : : : : : : : : : : : : : 187 7.6 1:1 Ordering on the Face-Centred Cubic Lattice : : : : : : : : : : : : 193 7.7 Homogeneous Cacti : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 197 Examples : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 200
7. Cluster Variation Methods : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 173
Table of Contents
8. Exact Results for Two-Dimensional Ising Models : : : : : : : : : 205

Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 205 The Low-Temperature Form and the Dual Lattice : : : : : : : : : : 206 The High-Temperature Form and the Dual Transformation : : 208 The Star-Triangle Transformation : : : : : : : : : : : : : : : : : : : : : : : : 211 The Star-Triangle Transformation with Unequal Interactions : 214 A Linear Relation for Correlations : : : : : : : : : : : : : : : : : : : : : : : : 216 Baxter and Enting's Transformation and the Functional Equation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 218 8.8 The Solution of the Functional Equation : : : : : : : : : : : : : : : : : : 220 8.8.1 A Preliminary Result for f K jk : : : : : : : : : : : : : : : : : : : 221 8.8.2 Expressions for A1jk and B 1jk : : : : : : : : : : : : : : : : 223 8.8.3 An Expression for bk : : : : : : : : : : : : : : : : : : : : : : : : : : : : 224 8.9 Critical Behaviour : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 226 8.10 Thermodynamic Functions for the Square Lattice : : : : : : : : : : : 229 8.11 Thermodynamic Functions for the Triangular and Honeycomb Lattices : : : : : : : : : : : : : : : : 232 8.12 The Antiferromagnet : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 235 Examples : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 238 8.1 8.2 8.3 8.4 8.5 8.6 8.7 9.1 The Decoration Transformation : : : : : : : : : : : : : : : : : : : : : : : : : : 241 9.2 Dilute Decorated Models : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 244 9.2.1 A Superexchange Model : : : : : : : : : : : : : : : : : : : : : : : : : : : 244 9.2.2 The Equilibrium Bond Dilute Ising Model : : : : : : : : : : : 245 9.3 Heat Capacity and Exponent Renormalization : : : : : : : : : : : : : 246 9.3.1 Three-Dimensional Lattices : : : : : : : : : : : : : : : : : : : : : : : : 247 9.3.2 Two-Dimensional Lattices : : : : : : : : : : : : : : : : : : : : : : : : : 250 9.4 Fisher's Decorated Antiferromagnetic Model : : : : : : : : : : : : : : : 251 9.5 The Decorated Lattice Ferrimagnet : : : : : : : : : : : : : : : : : : : : : : : 256 9.6 The Kagom Lattice Ising Model : : : : : : : : : : : : : : : : : : : : : : : : : 262 e 9.7 A Modi ed Star-Triangle Transformation and Three-Spin Correlations : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 264 9.8 The Star-Triangle Ferrimagnet : : : : : : : : : : : : : : : : : : : : : : : : : : : 266 9.9 The Unsymmetrical Ising Model : : : : : : : : : : : : : : : : : : : : : : : : : : 268 9.10 A Competing Interaction Magnetic Model : : : : : : : : : : : : : : : : : 271 9.11 Decorated Lattice Mixtures of Orientable Molecules : : : : : : : : : 273 9.12 The Decorated Lattice Gas : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 278 9.12.1 General Properties : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 278 9.12.2 A Water-Like Model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 281 9.12.3 Maxithermal, Critical Double and Cuspoidal Points : : : 283 9.13 Decorated Lattice Gas Mixtures : : : : : : : : : : : : : : : : : : : : : : : : : : 289 Examples : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 290
9. Applications of Transform Methods : : : : : : : : : : : : : : : : : : : : : : : 241
Table of Contents
XI
10. The Six-Vertex Model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 293

Two-Dimensional Ice-Rule Models : : : : : : : : : : : : : : : : : : : : : : : : 293 Parameter Space : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 295 Graphical Representation and Ground States : : : : : : : : : : : : : : 297 Free Energy and Transfer Matrices : : : : : : : : : : : : : : : : : : : : : : : 299 Transfer Matrix Eigenvalues : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 301 10.5.1 The Case n = 0 : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 302 10.5.2 The Case n = 1 : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 302 10.5.3 The General n Case : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 304 10.6 The Low-Temperature Frozen Ferroelectric State : : : : : : : : : : : 306 10.7 Wave-Number Density : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 307 10.8 The Solution of the Integral Equation for ,1 1 : : : : : : : 309 10.9 The Free Energy of the Disordered State : : : : : : : : : : : : : : : : : : 311 10.9.1 The KDP Model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 312 10.9.2 Square Ice : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 312 10.9.3 Non-Zero Polarization : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 313 10.10 The Ferroelectric Transition in Zero Field : : : : : : : : : : : : : : : : 313 10.11 The Antiferroelectric State : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 316 10.12 Field-Induced Transitions to the Completely Polarized State : : : : : : : : : : : : : : : : : : : : : : : 319 10.13 An Antiferroelectric in an Electric Field : : : : : : : : : : : : : : : : : : 322 10.14 The Potts Model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 325 10.14.1 The Staggered Six-Vertex Model : : : : : : : : : : : : : : : : : : 326 10.14.2 The Solvable Case h v = 1 : : : : : : : : : : : : : : : : : : : : : : : 329 10.14.3 The Solvable Case h v = ,1 : : : : : : : : : : : : : : : : : : : : : 330 10.14.4 The Polarization and Internal Energy : : : : : : : : : : : : : : 331 10.14.5 Critical Exponents : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 331 Examples : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 332 10.1 10.2 10.3 10.4 10.5 A.1 Regular Lattices : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 335 A.2 Elliptic Integrals and Functions : : : : : : : : : : : : : : : : : : : : : : : : : : 337 A.2.1 Elliptic Integrals : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 338 A.2.2 Elliptic Functions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 339 A.2.3 Results Required for Chapter 8 : : : : : : : : : : : : : : : : : : : : 340 A.3 The Water Molecule and Hydrogen Bonding : : : : : : : : : : : : : : : 342 A.4 Results for the Six-Vertex Model : : : : : : : : : : : : : : : : : : : : : : : : : 344 A.4.1 The Proof of I : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 345 A.4.2 The Proof of II : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 345 A.4.3 The Proof of III : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 345 A.4.4 The Proof of IV : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 348 A.5 Fourier Transforms and Series : : : : : : : : : : : : : : : : : : : : : : : : : : : : 349 A.5.1 Fourier Transforms : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 349 A.5.2 Fourier Series : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 349
A. Appendices : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 335
XII
Table of Contents
References and Author Index : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 351 Subject Index : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 365
1. Introduction to Thermodynamics and Phase Transitions
We assume that the reader has some previous knowledge of thermodynamics.1 Our intention here is to revise brie y the essential part of the theory, to introduce critical phenomena and, at the end of this chapter, to put the theory into a suitably generalized form for later use. In the rst instance, thermodynamics deals with systems in equilibrium. This means systems in which, under certain given conditions, all natural processes of change have come to an end. The observable state of such a system is therefore xed and constant in time. In a continuous mass of uid, for example, any initial di erences in temperature or density between di erent regions have been smeared out by convection or di usion when the equilibrium state is reached. Because of this uniformity the state of a system in equilibrium can be speci ed by a small number of variables. For a uid or isotropic solid composed of one type of molecule the state can be speci ed by three independent variables, such as the number of molecules M , the volume e e V and the temperature T . Other possible sets of independent variables are e and the pressure P ; M , T and the volume per molecule v; V , v and P .2 e e e M, T ~ e ~ Once the independent variables are chosen, the other variables are regarded as functions of the independent set. For example, when the state of the uid e e is speci ed by M , V and T the volume per molecule is given by v = V =M . ~ e Thermodynamic variables fall into two classes, intensive and extensive. e e ~ Intensive variables, for instance T , P and v, express the internal constitution of the system. For xed values of the intensive variables, the extensive e variables such as M and V are proportional to the size of the system. To completely specify a system, at least one extensive variable must be used, but the ee others can be intensive, as when the set M; P; T of independent variables is chosen for the uid. Any intensive variable depends only on other intensive e~e variables. A uid or isotropic solid has an equation of state f P; v; T = 0 e, v and T . e which is a relation between the intensive variables P ~
Such as, for instance, is contained in Chaps. 1 5 of Pippard 1967. See also Adkins 1983, Landsberg 1978 and Buchdahl 1966. 2 At this stage it is convenient to attach the tilde symbol to certain thermodynamic variables. This allows a more compact notation to be introduced in Sect. 1.7 and used extensively in this volume and Volume 2.
1
1.1 Thermodynamic Variables: Simple Fluids
The internal energy U of a system arises from the kinetic energies of the molecules and the energy of interaction between them. It is an extensive variable, and from the law of energy conservation its di erential may be written as e e dU = d0 Q , P dV + dM ; 1.1 where is an intensive variable, called the chemical potential, and d0 Q is the heat absorbed. Now suppose that the system is brought from equilibrium state A to equilibrium state B by a `reversible' process; that is one which is slow enough for the intermediate states through which the system passes to be themselves e ectively equilibrium states. Since U is a function of the state, the change in U is independent of the path taken from A to B and can be written UB , UA . Now Q is not a function of the state, and the heat absorbed during the process depends on the path taken from A to B. This is the reason that the symbol `d0 ' rather than `d' has been used for its ee di erential. However, by the second law of thermodynamics d0 Q = T dS for e an in nitesimal change in equilibrium state, where S is an extensive state function, called the entropy. Hence, for such a change e e e e dU = T dS , P dV + dM : 1.2 Since the state of the uid or isotropic solid can be completely speci ed by ee the values of S , V and M , 1.2 gives dU as a total di erential. The pairs of e; T , V ; ,P and M; , one extensive and the other intensive e e e variables S in each case, occurring in the di erential are called conjugate. From 1.2 it follows that3 Now suppose that, with its internal constitution xed, the size of the system is increased by a factor . Since all extensive variables will increase by the factor , e e ee U S ; V ; M = U S ; V ; M : 1.4 Di erentiation with respect to , and substitution from 1.3 gives ee ee U = T S , P V + M ; 1.5 when is put equal to 1.
e = @U T ; e @ S M;V
e = , @U ; P e @ V S;M
= @M : 1.3 S;V
@U
Because of the various choices possible for the independent variables in any thermodynamic system, it is often necessary to specify the variable or variables kept constant during a partial di erentiation by using subscripts.
1.2 Change of Variable and Thermodynamic Potentials
1.2 Change of Variable and Thermodynamic Potentials
An extensive variable, termed the enthalpy, is de ned by ee H = U + PV : 1.6 Substitution from 1.2 yields e e e e ee e e dH = dU + P dV + V dP = T dS + V dP + dM : 1.7 It can be seen that 1.6 represents a Legendre transformation associated with e e replacement of V as an independent variable by its conjugate ,P . From an e or V as an indee experimental point of view it is feasible to treat either P e pendent variable. However, although an experimenter can control T directly, e. Hence to replace S as an independent variable e this is hardly possible for S e by its conjugate T we employ the Legendre transformation ee A = U , TS ; 1.8 which de nes the Helmholtz free energy A. The latter is an example of a thermodynamic potential; the reason for the use of this term will appear in Sect. 1.5. In a similar way to 1.7, ee e e dA = ,S dT , P dV + dM : 1.9 The Gibbs free energy G is obtained by the Legendre transformation ee ee ee G = A + PV = U + PV , TS 1.10 and, either from 1.2 or from 1.9, ee e e dG = ,S dT + V dP + dM : 1.11 It can be seen that, just as A is appropriate to the set of independent variables ee ee T; V ; M , so G is the thermodynamic potential for the set T ; P; M . As indicated above, any intensive variable is dependent only on other intensive variables. For a system speci ed by n + 1 independent variables the maximum number of independent intensive variables is n. We now derive an important relation among such variables. By comparing 1.5 and 1.10 it can be seen that G = M ; 1.12 and hence dG = dM + M d : 1.13 Then substitution in 1.11 yields ee e e ,S dT + V dP , M d = 0 : 1.14 De ning the intensive variables s = S =M ; ~ e v = V =M ; ~ e 1.15
which are respectively the entropy and volume per molecule, we have the Gibbs Duhem equation ,sdT + vdP , d = 0 : ~ e ~ e 1.16 The grand potential D is appropriate to the set of independent variables e e V ; T ; and is important in connection with statistical mechanics, although not often mentioned in thermodynamics texts. Using the Legendre transformation ee D = A , M = U , T S , M ; 1.17 it can be shown, either from 1.2 or from 1.9, that ee e e dD = ,S dT , P dV , M d 1.18 and 1.17 with 1.5 gives ee D = ,P V : 1.19
1.3 Response Functions and Thermodynamic Relations

Response functions specify the reaction of the system to externally imposed changes in the state. The common ones de ned here implicitly assume a xed value for M . They can be expressed in terms of second derivatives of thermodynamic potentials. For instance the isobaric constant-pressure coe cient of thermal expansion of a uid or isotropic solid satis es the relations e! 1 @V = 1 @2G ; = 1.20
P
where the rst member de nes P , and the second follows from 1.11. In a similar way the isothermal constant-temperature compressibility is given by e! 1 @V = , 1 @2G : 1.21 K =,
T
e e V @T P
e e e V @T @P
Relations between derivatives of thermodynamic functions arise from the fact that a state can be speci ed by a small number of variables and from the rst and second laws of thermodynamics, expressed by such relations as 1.2. For instance from 1.20 and 1.21, we may write for a uid of constant composition e e e e e dV = V P dT , V KT dP 1.22 and ! hence
e e V @P T
e e V @P 2
e @P = P : e @ T V KT
1.23
1.3 Response Functions and Thermodynamic Relations
There is thus no need to de ne a separate response function to express the variation of pressure with temperature at constant volume. From their de nitions, P and KT are intensive variables and, using 1.12, 1 @2 1 2 KT = , v @ e : 1.24 P=v e e; ~ @T @P ~ @P 2 A heat capacity gives the amount of heat needed to raise the temperature of the system by 1 K, under stated conditions. For molar heat capacities the system is taken to contain one mole of matter. M = Ma = 6:022 1023 ee molecules, Ma being Avogadro's number. Since d0 Q = T dS , the molar heat capacity molar speci c heat at constant pressure is given by
From the last expression it can be seen that the molar heat capacity is an intensive variable. Similarly the molar heat capacity at constant volume is where a is the Helmholtz free energy per molecule. These four response functions are not independent. To show this, note rst that, at constant composition,
e! e e e 2G e 22 CP = T @ S = @H = ,T @ e2 = ,MaT @ e : e @T P @T P @T @T e! e e e 22 e 22 CV = T @ S = @U = ,T @ eA = ,Ma T @ ea ; e @T V @T V @T @T
1.25
1.26
and hence
e! e! @ S dT + @ S dV e e dS = e e e @T V @V T
1.27 1.28
The importance, pointed out above, of specifying which other variables are held constant when di erentiating with respect to a given variable is clearly displayed by this relation. Now, from 1.9,
e! e! e ! e ! @V @S , @S = @S e P @T V @V T @T P : e e e @T
From 1.25, 1.26, 1.20 and 1.23
e! e! @S = , @2A = @P : e e e @T V e @V T @V @T ee CP , CV = V T KP ;
T
2
1.29 1.30
e where V is the molar volume Ma v. ~
1.4 Magnetic Systems
Until now we have considered energy transfer by heat or matter ow and mechanical work, but energy can also be transferred by the action of a magnetic eld. It will be supposed that the latter has a xed direction and can e be represented by a scalar H of dimension ampre metre A m. The mage netization or magnetic moment is taken as parallel to the eld and can be e represented by a scalar M of dimension weber metre Wb m. Obviously e is an intensive variable and M an extensive variable. If the con guration e H i.e. size and shape and composition are xed then e e ee e e dU = d0 Q + HdM = T dS + HdM : 1.31 e Experimentally it is hard to specify M initially, and it is usually better to e as an independent variable. A magnetic enthalpy H1 may be de ned use H by the Legendre transformation ee H1 = U , HM : 1.32 The thermodynamic potential A may again be de ned by 1.8 and a potential e e F1 , corresponding to independent variables T and H, by ee e e F1 = U , T S , HM : 1.33 From 1.31 1.33 and 1.8 we have ee e e dH1 = T dS , MdH ; ee e e 1.34 dA = ,S dT + HdM ; ee e e dF1 = ,S dT , MdH : Again standardizing to one mole M = Ma , we can de ne heat capacities CH and CM at constant eld and magnetization respectively, the latter being di cult to realize experimentally. We have
Here 0 is the permeability of free space, 410,7 Wb A m, and its inclusion makes T dimensionless.
Another important response function for magnetic systems is the isothermal magnetic susceptibility T which is de ned for unit volume of matter. Noting e ~ that the magnetization for unit volume is M=M v where M is the number of atoms and v the volume per atom, ~ e! @ M = , 1 @ 2 F1 : 1 1.36 T = M v e ~ 0 @H T M v0 @ H2 ~ e
e! e e e 2F CH = T @ S = ,T @ e21 : @T H @T
1.35
1.5 Stationary Properties of Thermodynamic Functions
1.5 Stationary Properties of Thermodynamic Functions
The thermodynamics of irreversible processes is based on the principle that, when a completely isolated system is not in thermodynamic equilibrium, all processes of change increase its entropy. It follows that when an equilibrium state is reached, the entropy is at a maximum and, for small deviations from e e e equilibrium, dS = 0. Note that, while dS = 0, if we put dU = dV = dM = 0 in 1.2, this simply results from the functional relationship which, for equilibe rium states of the uid, exists between S and the independent thermodynamic e and M . The stationary property dS = 0 is more general since e variables U , V it is true for small deviations from equilibrium. We now derive an application of the stationary property. Consider a composite system consisting of two subsystems, each in a rigid container, which are in thermal contact with each other but thermally isolated from the outside world. The subsystems are in internal equilibrium and their temperatures are e e T1 and T2 respectively. Assuming that boundary e ects are negligible the ene e e e e tropy of the composite system S = S1 + S2 where S1 and S2 are the entropies 0 Q is transferred from of the separate subsystems. Now suppose that heat d subsystem 1 to subsystem 2; then e e T1 dS1 = d0 Q1 = ,d0 Q ; e e T2 dS2 = d0 Q2 = d0 Q ; 1.37 1 1 e e e dS = dS + dS = , d0 Q :
1 2
If there is thermodynamic equilibrium in the whole system before the heat e transfer takes place, then, by the stationary property, dS = 0. Hence a cone1 = T2. If T1 T2, so that the subsystems are e e e dition for such equilibrium is T e not in thermal equilibrium, then dS 0 if d0 Q 0. Hence the increasing entropy principle implies that heat is transferred from the higher temperature to the lower temperature. Now suppose that the system is no longer thermally isolated but remains at xed con guration and composition. An example is a uid in a rigid container with heat-conducting walls immersed in a heat bath at given temperae e ture T , the independent thermodynamic variables now being T , the container e and the number M of molecules enclosed. For a system of this type volume V it can be shown that any irreversible process decreases the Helmholtz free energy A. Hence A is at a minimum when the equilibrium state for temperature e T is reached and dA = 0 for small deviations from equilibrium. Thus A is the thermodynamic potential for a system which is mechanically isolated but can exchange heat with the outside world. For a uid the condition of mechanical isolation can be removed by supposing that the rigid container is penetrated by a tube with a close- tting movable piston of xed area on which a constant force is exerted from the
e T2
e T1
e e outside. The independent variables are now T , the pressure P i.e. the piston force divided by area and M . The Gibbs free energy G can be shown to be the appropriate thermodynamic potential, G being at a minimum in the e e equilibrium state for the given values of T and P , and the condition dG = 0 applying for small deviations from equilibrium. Alternatively the condition of molecular isolation can be removed by supposing that the rigid container walls are permeable to the type of molecule composing the uid. The environe ment must now be regarded as both a heat bath at given temperature T and a `molecule bath' at given chemical potential , the independent thermodyee namic variables being T , V and . The appropriate thermodynamic potential is now D, as de ned by 1.17, and dD = 0 for small deviations from the e equilibrium state for the given values of T and . If a volume of uid at an appropriate density is placed in a rigid transparent container and the temperature is lowered, then at a certain point the appearance of critical opalescence indicates that light is being scattered by large-density inhomogeneities. If the temperature is lowered still further, there is a separation into distinct regions or phases of liquid and vapour respectively with a well de ned boundary between them. We shall now derive conditions for such phases to exist in thermodynamic equilibrium at a given e e temperature T . Let one phase contain M1 molecules at volume V1 and the e2, where V1 + V2 = V , the volume of the e e e other M2 molecules at volume V rigid container, and M1 + M2 = M , the total number of molecules in the container. Regarding boundary contributions as negligible, the thermodynamic potential A of the system is ee ee A = A1 T; V1 ; M1 + A2 T ; V2 ; M2 : 1.38 e Now suppose that there is a small deviation from equilibrium in which V1 e , and V2 decreases by dV , while M1 and e e increases by a positive amount dV M2 remain xed. Then, from 1.38 and 1.9, e e e e e e e dA = dA1 + dA2 = ,P1 dV1 , P2 dV2 = P2 , P1 dV : 1.39 Since the phases are assumed to be in thermodynamic equilibrium before the e e small volume changes occur, dA = 0 and hence P1 = P2 . Similarly, if M1 e1 and V2 xed then e increases by dM and M2 decreases by dM with V dA = dA1 + dA2 = 1 dM1 + 2 dM2 = 1 , 2 dM 1.40 but, again, dA = 0 implying 1 = 2 . Hence, for phase equilibrium at tempere ature T , we have the conditions that the pressures and the chemical potentials respectively in the two phases must be equal: e e P1 = P2 ; 1 = 2 : 1.41
1.6 Phase Equilibrium in the Van der Waals Gas
A perfect gas in which the molecules are mass points with no potential energy of interaction has the equation of state e~ e P v = kb T ; v = V =M ; ~ e 1.42 ,23 J K. This is obeyed where kb is Boltzmann's constant, equal to 1:381 10 to a good approximation by dilute real gases, but at higher densities there are large deviations, and the phenomenon of phase separation, discussed in Sect. 1.5, occurs. In a real uid there are strong repulsive forces between molecules when very close together, giving a nite size or hard core e ect, and attractive forces at larger distances apart. Displaying profound physical intuition, van der Waals 18734 modi ed 1.42 in a heuristic fashion by e e ~ replacing v by v , b to allow for nite molecular size, and P by P + cv,2 . ~ ~ The latter change allows for the attractive forces since it implies a smaller external force to hold the M molecules in volume M v than would otherwise ~ be needed. We thus have
e c e P + v2 ~ , b = kb T : 1.43 ~ v If curves of constant temperature isotherms are plotted in the P v plane e for 1.43 it is found that below a temperature Tc , to be deduced below, they have the `s-shape' shown in Fig. 1.1. Now let points A and B on a horizontal line intersecting the isotherm correspond to states labelled 1 and e e 2 respectively. It is obvious that P1 = P2 and it will now be shown that the second phase equilibrium condition of 1.41 is satis ed if the areas of the two loops between the line segment AB and the isotherm are equal Maxwell's e equal-areas rule. To prove this, put dT = 0 in 1.16, giving e~ e v d = vdP = dP v , P d~ ~ e 1.44 and, integrating over the part of the isotherm between A and B,
e v ~ 2 , 1 = P1 ~2 , v1 ,
Z v1 ~
v2 ~
ev P d~ :
1.45
The rst term on the right-hand side of 1.45 is the area between the line segment AB and the v-axis, whereas the integral is the area between the ~ corresponding part of the isotherm and the v-axis. These two quantities are ~ equal when the equal-areas rule applies and hence the latter implies 1 = 2 . Now consider a xed number M of molecules in a rigid container of volume e e V = M vm at the temperature T of the isotherm shown in Fig. 1.1. If v1 ~ ~ vm v2 the uid can separate into the equilibrium phases 1 and 2, the ~ ~ number of molecules in the two phases being respectively M1 and M2 , where
4
For accounts of the impact of van der Waals' work on the development of the theory of phase transitions see Rigby 1970, Rowlinson 1973, de Boer 1974 and Klein 1974.
10
e P
C A B
v1 ~ vm ~ e Fig. 1.1. Van der Waals gas isotherm for T phases A and B.
v2 v ~ ~ e Tc ; equal-areas rule for conjugate
1.46 Condition 1.41 ensures that the two-phase system is locally stable, but for complete thermodynamic stability it must shown that the thermodynamic potential A of the two-phase system is lower than that of the homogeneous system of molecular volume vm , whose state is represented by point C on the ~ isotherm in Fig. 1.1. Since M is xed we can use instead of A the thermodynamic potential a per molecule, which satis es the relations derived from 1.9, 1.10 and 1.12. Now consider the a; v isotherm cor~ e~ responding to the P; v isotherm of Fig. 1.1. From 1.47,
M vm = M1 v1 + M2 v2 ; ~ ~ ~
M = M1 + M2 :
e P = , @a ; @v ~
e~ = a + P v = a , v @a ; ~ @v ~
1.47
e @P = , @2a : 1.48 @v ~ @ v2 ~ e ~ Hence where @ P=@ v 0, that is between the minimum and the maximum e v isotherm, @ 2a=@ v2 0 and the a; v isotherm is concave to on the P; ~ ~ ~ e ~ the axes. At all other points, @ P=@ v 0 and the a; v isotherm is convex to ~
the axes, giving it the form shown in Fig. 1.2. From 1.47 the slope of the e tangent at any value of v gives the corresponding value of ,P whereas the intercept of the tangent with the a-axis gives . The equilibrium conditions 1.41 are thus geometrically equivalent to a double tangent rule. The points on the a; v curve corresponding to the two equilibrium phases lie on a double ~ tangent, as shown in Fig. 1.2. Assuming that the contribution of the phase
11
a a1 am
e Pmin e Pmax
a2 v1 ~
gent.
vm ~
v2 ~
e e Fig. 1.2. Helmholtz free energy volume isotherm for T Tc showing double tanboundaries is negligible, the free energy density value as for the two-phase system is given by Mas = M1a1 + M2 a2 : 1.49 Comparison with the rst relation of 1.46 shows that as is the ordinate of the point on the double tangent lying vertically above the point vm on the ~ v -axis see Fig. 1.2. Since the a; v isotherm lies above the double tangent ~ ~ for the whole range from v1 to v2 , it follows that as is less than the value of a ~ ~ for a homogeneous uid with the same molecular volume. Thus replacing the loop in the homogeneous uid a; v curve between v1 and v2 by the double ~ ~ ~ tangent segment gives the equilibrium a; v curve, indicated by the chain line ~ in Fig. 1.2. This is mathematically the convex envelope of the homogeneous uid a; v curve. ~ Phase separation thus occurs whenever there are maxima and minima on the isotherms given by 1.43 and these appear at the critical point where 1.50 @ v2 T = 0 : ~ e~ e e ~ e The critical values of T , v and P are denoted by Tc, vc and Pc respectively, e e and the isotherm for which T = Tc is called the critical isotherm. From 1.43, e! 2 e! e e @ P = , kb T + 2c ; @ P = 2kb T , 6c ; 1.51 2 v3 2 @v T ~ ~ , b ~ v @ v T ~ , b3 v4 ~ v ~ and, by substituting 1.51 in 1.50, it is easy to show that e 8 e kb Tc = 27cb ; v c = 3b ; ~ Pc = 27cb2 : 1.52
v ~
e! @P
@v T = 0 ; ~
2 e! @P
e P
12
e Pc
e e T Tc e e T = Tc
Phase separation
e e T Tc
v ~
Fig. 1.3. Equilibrium pressure volume isotherms; C is the critical point.
vc ~
e e Typical equilibrium isotherms are shown in Fig. 1.3. For T Tc, an equilibrium isotherm consists of a horizontal segment or `tie line', like AB in Fig. 1.1, and monotonically decreasing portions linked with each end. The equilibrium phases at the end-points of a tie line are known as conjugate phases. The broken curve in Fig. 1.3 is called the coexistence curve, and the e e liquid and vapour regions lie to its left and right respectively. For T Tc the e=Tc or v=vc their shape is e ~~ isotherms are monotonic decreasing and for large T close to the perfect gas form given by 1.42. Though the critical values given by 1.52 are dependent on the particular equation of state 1.43 it should be noted that the thermodynamic arguments above require only that s-shaped isotherms should appear below a certain temperature. A number of other `classical' equations of state, satisfying this condition, have been postulated see Example 1.6 and Chap. 5.
1.7 The Field-Extensive Variable Representation of Thermodynamics

The basic thermodynamic theory will now be expressed in a form applicable to any system and useful for the development of statistical mechanics. The e e expression ,P dV for mechanical work used until now is adequate for uids or isotropic solids, but for non-isotropic solids crystals it must be replaced by a sum over stress and strain components. When a magnetic eld acts
e e on the system a magnetic work term HdM see Sect. 1.4 is added to the mechanical work, and when an electrical eld acts, there is a similar electrical work term. In general the composition of the system must be speci ed by the numbers Mj of molecules or atoms of several components5 labelled with the index j , whereas up to now we have assumed a one-component system with composition speci ed by a single number M . Thus 1.2 or 1.31 must be replaced by X e e X dU = T dS + Yi dXi + j dMj ; 1.53
where the intensive variables Yi are generalized forces, the Xi are the conjugate extensive mechanical variables, and each j is the chemical potential e e of a component j . The di erence in sign between ,P dV and Yi dXi arises from the fact that the work done by the pressure reduces the volume of the system. Here the internal energy U and the chemical potentials j have the dimensions of energy, joule J in SI units, whereas the numbers Mj conjugate to the j are dimensionless. To treat all terms on the right-hand side of 1.53 on the same basis it is convenient to make all the extensive variables whose di erentials appear there dimensionless. Thus we de ne e e e S = S =kb ; V = V =v0 ; M = M=m0 ; 1.54 where kb is Boltzmann's constant, v0 is a standard volume and m0 is the e dipole moment of a single molecule. The transformation from V to V may appear arti cial, but in lattice models, where the position of the molecules is restricted to the lattice sites, v0 can be taken to be the volume per site, and V then becomes the number of sites in the lattice. The transformation of the extensive variables to dimensionless form implies a corresponding transformation in their conjugates. These now all take on the dimensions of energy and will be known as elds. From 1.54, the conjugates of S , V and M respectively are e e e T = kb T ; P = Pv0 ; H = Hm0 : 1.55 In place of 1.53 we can now write, for a system with n + 1 independent variables,
f ig fjg
13
i d Qi ; 1.56 i=1 where T and the i are all elds with the dimensions of energy, while their conjugates S and Qi are dimensionless. The Qi here are not heat terms. The
internal energy U is regarded as a function of the n + 1 extensive variables S , Qi , i = 1; : : : ; n and, in a similar way to 1.5,
5
dU = T dS +
n X
A component is a set of atoms or molecules which are not convertible to atoms or molecules of another component by any process occurring in the system, e.g. oxygen, nitrogen and carbon dioxide in the atmosphere.
14
U = TS + S dT +
n X i=1
n X i=1
i Qi :
1.57 1.58
By taking the di erential of 1.57 and comparing it with 1.56
Qi di = 0 :
A subset Q1 ; : : : ; Q of the extensive variables can now be replaced as independent variables by their intensive conjugates 1 ; : : : ; using the Legendre transformation
H = U ,
X i=1
i Qi = TS +
X i=1
n X
where H is an enthalpy and the last expression of 1.59 follows from 1.57. Taking the di erential of 1.59 and comparing with 1.56 gives dH = T dS ,
i=+1 n X
i Qi ;
1.59
Q i d i +
The enthalpy H is not uniquely de ned since it depends on the order in which we attach the indices 1; : : : ; n to the n variables Qi . For instance, H as de ned for a uid in 1.6 and H1 as de ned for a magnetic system in 1.32 are both examples of the general H1 , with Q1 taken as V and M respectively. The internal energy U could be termed H0 and in a sense the concept of enthalpy is more fundamental than that of internal energy since it is to some extent arbitrary whether energy terms involving independent elds are included in U or treated separately, as in 1.59. The Helmholtz free energy A is de ned as in 1.8 so that
i=+1
i dQi :
1.60
A is the thermodynamic potential which attains a minimum value when the state of thermodynamic equilibrium is reached in a system speci ed by T and the extensive variables Q1 ; : : : ; Qn ; that is, a system where the only
i Qi ; i=1 n X dA = ,S dT + i dQi : i=1
A = U , TS =
n X
1.61 1.62
interchange of energy with the outside world is by heat transfer. The replacement of Q1 ; : : : ; Q as independent variables by their conjugate elds 1 ; : : : ; means that the system is opened to energy interchange by way of these elds. The appropriate thermodynamic potential is now F where
F = H , TS = U , TS ,
X i=1
i Qi =
n X
i=+1
i Qi ;
1.63
dF = ,S dT , giving
X i=1
Qi di +
n X
15
i=+1
i dQi ; i = 1; : : : ; ;
1.64
i = + 1; : : : ; n : From the rst relation of 1.63, S = H , F =T and hence the rst relation
S = , @F ; @T @F ; i = @Qi
Qi = , @Fi ; @
1.65
of 1.65 can be expressed as F =T H = ,T 2 @ F =T = @@1=T ; 1.66 @T which is the generalized Gibbs Helmholtz relation. The remarks made above about the speci cation of H also apply to F . In Sect. 1.2, G and D are both examples of F1 with Q1 taken as V and M , respectively, whereas in Sect. 1.4 the quantity denoted by F1 provides another example of the general F1 with Q1 = M. The Helmholtz free energy A could be denoted as F0 and the relations given above for A and dA can be obtained from 1.63 and 1.64 by putting = 0. For a system with components 1; : : : ; , a generalized grand potential may be obtained by assigning = , Qj = Mj ; j = j ; j = 1; : : : ; 1.67 and then identifying D with F . From 1.63 and 1.64,
D = U , TS ,
dD = ,S dT ,
X
j =1 X j =1
j Mj =
n X
Mj dj +
i=+1 n X i=+1
i Qi ;
1.68 1.69
i dQi :
For the one-component simple uid where the only mechanical eld 1 is ,P , 1.68 and 1.69 reduce to 1.17 1.19. From 1.56 a degree of freedom cannot be a means of transmitting energy to the system if Qi is constant or, on the other hand, if the corresponding eld i is zero. In the rst case the system is closed with respect to energy transmission by way of eld i , and in the second this particular eld is not acting in the system's environment. Clearly if Qi is constant then it must be regarded as one of the independent variables and, in the formalism above, + 1 i n whereas if i is zero then it is an independent variable and 1 i . In neither case does the corresponding term appear on the righthand side of 1.60 or 1.64. Sometimes it is feasible experimentally to put Qi = constant; for instance by enclosing a uid in a rigid vessel we can have, e ectively, V = constant, but it is impossible to put the conjugate eld
16
,P = 0. On the other hand, a system can be isolated magnetically by putting
the eld H = 0. If we try to use only the elds as independent variables by putting = n, then from the last expression of 1.63 the corresponding potential Fn = 0 and 1.64 reduces to 1.58. This `catastrophe' clearly results from an attempt to de ne an extensive thermodynamic potential in terms of intensive quantities only. However, by adopting a di erent approach it is possible to have a representation of thermodynamics in which only intensive variables occur, and this will be done in the next section.
1.8 The Field-Density Representation of Thermodynamics

In this representation, the systematic theory of which is due to Gri ths and Wheeler 1970, extensive variables are replaced by a new set of intensive variables called densities which are ratios of extensive variables. This formulation is useful in relating thermodynamics to statistical mechanics where it is necessary to consider systems in which every extensive variable tends to in nity. The advantage of using properly de ned densities is that they remain nite in the thermodynamic limit. Choosing a particular extensive variable as Qn we de ne the n,1 densities i by i = 1; : : : ; n , 1 : 1.70 i = Qi =Qn ; The entropy, internal energy and Helmholtz free energy densities are de ned in a corresponding way: s = S=Qn ; u = U=Qn ; a = A=Qn : 1.71 Most frequently either Qn = V , so that the densities are `per unit of reduced volume', or Qn = M , where M is the total number of molecules in the system, so that the densities are `per molecule'. We have in fact already used `per molecule' densities for a simple uid in deriving 1.16 and in the treatment of the van der Waals gas in Sect. 1.6. Equations 1.57 and 1.58 can be converted to eld density form by dividing through by Qn to yield
u = sT +
n,1 X
0 = sdT +
i=1 n,1 X i=1
i i + n ; i di + dn :
1.72 1.73
Equation 1.73 is a generalization of the simple uid relation 1.16 and shows that there cannot be more than n independent intensive variables for
1.8 The Field-Density Representation of Thermodynamics

n,1 X i=1 n,1 X i=1
17
a system where there can be as many as n +1 independent extensive variables. Taking the di erential of 1.72 yields du = sdT + T ds +
n,1 X i=1 i di +
xii d i + dn ;
1.74
which can be combined with 1.73 to give du = T ds +
i d i ;
1.75
which is the eld-density analogue of 1.56. Free energy or thermodynamic potential and enthalpy densities may be de ned by f = F =Qn ; h = H =Qn ; = 0; 1; 2; : : :; n , 1 ; 1.76 where f0 = a, the Helmholtz free energy density. Dividing 1.63 by Qn yields immediately
f = h , Ts = u , Ts ,
df = ,sdT ,
X i=1
X i=1
i i =
n,1 X
From 1.77 and 1.75 it is easy to show that

i di + n,1 X i=+1
i=+1
i i + n :
1.77
i d i :
1.78
Since F has a minimum value for thermodynamic equilibrium with xed values of T , 1 ; : : : ; , Q+1 ; : : : ; Qn it follows that f has a minimum value for thermodynamic equilibrium with xed values of T , 1 ; : : : ; , +1 ; : : : ; n,1 . We have already used the minimum property of a for a one-component uid with the ratio V=M xed in Sect. 1.6. A free energy density, which is a function of T and the n , 1 independent elds 1 ; : : : ; n,1 can be obtained by putting = n , 1. Equations 1.77 and 1.78 then become
fn,1 = u , Ts ,
dfn,1 = ,sdT ,
n,1 X i=1
n,1 X i=1
i i = n ; i di :
1.79
The second formula is equivalent to 1.73. For many-component systems it is often convenient to express the composition in terms of the mole fractions xj , given by
xj = Mj =M ;
j = 1; : : : ; ;
X j =1
xj = 1 ;
1.80
18
where
M=
X j =1
X j =1
Mj :
,1 X j =1
1.81
To use the total number of molecules as an extensive variable we must rst note that the energy of the molecular transfer term can be written
j dMj =
j , dMj + dM :
1.82
Let us now assign Qn,+j = Mj ; n,+j = j , ; j = 1; : : : ; , 1 ; 1.83 Qn = M ; n = : With these variables a `per molecule' free energy density can be de ned for any value n , 1, but we shall consider only fn,1 explicitly. From 1.79 and 1.83
fn,1 = = u , Ts ,
dfn,1 = d = ,sdT ,
n, X
i=1 n, X i=1
i i ,
,1 X
i di ,
j =1 ,1 X j =1
xj j , ; xj dj , :
1.84 1.85
The last relation can be re-expressed in the form
sdT +
n, X i=1
i di +
X j =1
xj dj = 0 ;
1.86
which is a generalized form of the Gibbs Duhem equation 1.16.
1.9 General Theory of Phase Equilibrium

Liquid vapour phase separation has been discussed in the context of the van der Waals equation of state although any `classical' theory giving pressure volume isotherms of similar form will yield a critical point and coexistence curve in the same way. We now consider phase equilibrium in the system de ned in Sect. 1.7 with n + 1 extensive variables S; Q1 ; : : : ; Qn . It will be supposed that the state of the system is speci ed by the elds T; 1 ; : : : ; and the extensive variables Q+1 ; : : : ; Qn . The thermodynamic potential with a stationary value at equilibrium, when the above variables have xed values, is F , de ned by 1.63. We now regard the system as split into two homogeneous phases in equilibrium, distinguished by the indices 1 and 2. Assuming as usual that boundary contributions to the extensive variables are negligible,
19
The values of the dependent elds in the conjugate phases will be denoted by i1 and i2 respectively i = + 1; : : : ; n. Now suppose that there is a small deviation from equilibrium in which Qi1 ! Qi1 + dQi ; i = + 1;:::;n: 1.88 Qi2 ! Qi2 , dQi ; Then, by 1.64 and 1.87, dF = dF1 + dF2 = =
n X i=+1 n X i=+1
F = F1 + F2 ; Qi = Qi1 + Qi2 ;
i = + 1; : : : ; n :
1.87
i1 dQi +
n X
i=+1
i2 ,dQi
1.89
i1 , i2 dQi :
Since F has a stationary value at equilibrium, dF = 0 for small deviations from equilibrium and hence the coe cients of the dQi in the last expression of 1.89 are zero. The independent elds i ; : : : ; and T are the same throughout the system by hypothesis. Hence each of the n elds must have the same value in the two phases in equilibrium. This result is seen to be independent of the original choice of independent variables. It can easily be extended to a situation where there are p phases in equilibrium, and we can then write i1 = i2 = = ip,1 = ip ; i = 1; : : : ; n ; 1.90 T1 = T2 = = Tp,1 = Tp ; 1.91 where Tk is the temperature of phase k. It should be noted that phase equilibrium conditions involve intensive variables only; the sizes of phases in equilibrium are arbitrary.
1.9.1 A One-Component Fluid

For a simple one-component uid system, n = 2 since, apart from T , the only elds are ,P and . Hence, for liquid vapour equilibrium, conditions 1.90 reduce to the relations 1.41, derived independently in Sect. 1.5. Fig. 1.3, obtained for a van der Waals gas, qualitatively resembles the liquid vapour part of the phase diagram for a real substance. It can be seen that on any isotherm the pressure at which liquid and vapour phases coexist is determined. The same must therefore be true of the chemical potential, since is a function of P and T . Hence if the temperature of liquid vapour equilibrium is altered, the corresponding changes in pressure and chemical potential are determined. The one-component, two-phase system is said to have one
20
degree of freedom. We now ask: how many degrees of freedom Fn; p are possessed by the p-phase system whose equilibrium conditions are 1.90 and 1.91? If the phases were isolated from each other there would be n independent intensive variables in each phase, making pn in all. However 1.90 imposes np , 1 conditions on the intensive variables and 1.91 a further p , 1 conditions. Thus Fn; p = pn , p , 1n + 1 = n , p + 1 : 1.92 As an alternative to the term `degrees of freedom' we can say that Fn; p is the dimension of the region of coexistence of the p phases. The relation 1.92 is better known in the form of the Gibbs phase rule which applies to an isotropic system with components. Apart from T , the elds in this case are 1 ; 2 ; : : : ; and ,P . Thus from 1.92 the dimension of the region of coexistence is D; p = F + 1; p = , p + 2 : 1.93 For two coexisting one-component phases D1; 2 = 1 in agreement with the result obtained above for liquid vapour equilibrium. Since D; p 0 we conclude that the largest number of phases of a component system which can coexist is + 2. There are exceptions to this, which arise when the phase diagram has some special symmetry. One of these is the triangular ferrimagnetic Ising model discussed in Sect. 4.5 and another is the metamagnet in a staggered eld see Sects. 4.3 and 4.4 and Volume 2, Sects. 2.2 and 2.12. The most well-known example of a single point of three phase coexistence of a one component system is the solid liquid vapour triple point in the T; P plane. Liquid vapour, solid vapour and solid liquid equilibrium respectively exist along curves in the T; P plane, and these curves meet at the triple point see Fig. 1.4. If the pressure on a uid is steadily reduced at a xed temperature T Tc there is a discontinuity in the molar volume when the liquid vapour transition point is reached see Fig. 1.3. This is an example of a rst-order transition, the term rst-order implying a discontinuity in some of the densities i de ned in Sect. 1.8. The terms ` rst-order transition' and `conjugate phases' refer to the same phenomenon. Whether a transition from one phase to another or separation of a homogeneous medium into conjugate phases is observed depends on which thermodynamic quantities are xed and which are allowed to vary. We now consider the behaviour of a free energy thermodynamic potential density function at a rst-order transition. The free energy density fn,1 T; 1 ; : : : ; n,1 , de ned by 1.79, is equal to the eld n . Hence fn,1 is continuous by 1.90. However since, by 1.79,
i=,
and since some or all of the densities i are unequal in the conjugate phases, there will be discontinuities in the corresponding rst derivatives of fn,1 . Geometrically, the function fn,1, i.e. n can be regarded as a surface in
@fn,1 ; @i
i = 1; : : : ; n , 1 ;
1.94
21
Fig. 1.4. Phase equilibP

rium curves in pressure temperature plane; the critical point is circled.
Solid
Liquid
Vapour
an n + 1-dimensional space with coordinates T; 1 ; : : : ; n . The rst-order transition will appear as a `crease' in this surface. Taking a plane through the surface parallel to the fn,1 -axis and the axis corresponding to a eld i we obtain a diagram like that shown in Fig. 1.5. For the simple one-component system, with n = 2; fn,1 = T; P if `per molecule' densities are used. The phase equilibrium lines in Fig. 1.4 are the projections into the T; P plane of the rst-order transition creases in the T; P surface. We now consider the behaviour of fn,2 , which is a function of the elds T; 1; : : : ; n,2 and the density n,1 , at a rst-order transition where n,1 has a discontinuity. Since fn,2 depends on n,1 , fn,2 will normally be discontinuous at the transition. A plot of fn,2 against n,1 for homogeneous phases with the elds T; 1 ; : : : ; n,2 held constant thus consists of two disconnected parts, corresponding to the phases on the two sides of the transition Fig. 1.6. However, putting = n , 2 in 1.77 and 1.78 gives
Transition point
fn,1
i
Fig. 1.5. Free energy eld isotherm showing a rst-order transition point.
22
fn,2 fB fA
A B
Fig. 1.6. Free energy density

isotherm with double tangent AB.
n,1
@f @f n,1 = @ n,2 ; n = fn,2 , n,1 n,1 = fn,2 , n,1 @ n,2 : 1.95 n,1 n,1 The end-points A and B of the two parts of the fn,2 curve correspond to
conjugate phases. Since elds are continuous at a rst-order transition the right-hand sides of the two relations of 1.95 are equal at A and B. Geometrically 1.95 thus implies a double tangent at the points A and B in Fig. 1.6. Let us denote the values of fn,2 at the points A and B by fA and fB respectively and the corresponding values of n,1 by A and B . If the prescribed value of n,1 lies between A and B then the system will split into conjugate phases A and B. If boundary e ects are disregarded, the overall value of the free energy density fn,2 for the system will be a linear combination of fA and fB . Hence the plot of fn,2 for the whole range of n,1 consists of the two curved portions terminating at A and B respectively and the segment AB of the common tangent. For the simple one-component system, with n = 2; fn,2 = aT; v if `per molecule' densities are used. Hence the relations 1.47, derived above, are a special case of 1.95. In Fig. 1.2, derived from the approximate van der Waals theory, the convex envelope indicated by the chain line corresponds to Fig. 1.6. Returning to the general system, the free energy density f , de ned by 1.77, is a function of the independent densities +1 ; : : : ; n,1 and the independent elds T; 1 ; : : : ; . If the latter are held constant then the dependence of f on the independent densities can be represented geometrically by a surface in the n , -dimensional space with coordinates +1 ; : : : ; n,1 ; f . By generalizing 1.95 it can be shown that points corresponding to conjugate phases lie on the same tangent hyperplane.
1.9.2 Azeotropy
We now discuss brie y the phenomenon of azeotropy of which a more detailed account is given by Rowlinson and Swinton 1982. Consider a two-component uid system with mole fractions x1 = x, x2 = 1 , x. Then n = 3 and for `per molecule' densities we use 1.83, putting
1.10 Classical Theory and Metastability
23
Then, from 1.85, df2 = d3 = ,sdT , vd1 , xd2 : 1.97 With = 2 and p = 2 two phase equilibrium of a two-component mixture 1.93 yields D2; 2 = 2, giving a surface in 1 ; 2 ; T space. Consider the curve of intersection of this surface with a plane 1 = ,P = constant. For a small displacement along such a curve, a relation like 1.97, with d1 = 0, is satis ed by each of the two equilibrium phases. Subtracting one of these relations from the other and using the fact that, by 1.90, d3 is the same for both phases we have dT 4x 1.98 d2 = , 4s ; where 4s and 4x respectively are the di erences between the values of entropy per molecule and mole fraction between the conjugate phases. An azeotropic point occurs when the conjugate phases have the same composition 4x = 0 but di erent molecular entropies 4s 6= 0. Then, by 1.98, dT=d2 = 0, so that the phase equilibrium curve in the 2 ; T plane has a horizontal tangent at the azeotropic point. If the coordinates are changed from 2 and T to x and T then the phase equilibrium curve is replaced by two curves each representing one of the conjugate phases. For each of these phases we can write d2 @x dx = @x 1.99 dT @2 T;1 dT + @T 1 ;2 ; where @x=@2 T;1 0 by the stability condition 1.107 and d2 =dT is given by 1.98. The conjugate phase curves meet at the azeotropic point where 4x = 0 and, since d2 =dT = 1, it follows from 1.99 that dT=dx = 0. Hence, in the x; T plane, the conjugate phase curves touch, with a common horizontal tangent at the azeotropic point. Fig. 1.7 shows positive azeotropy; for negative azeotropy the touching curves have minima at the azeotropic point. Azeotropic points in the plane 1 = ,P = constant form a curve in the 1 ; 2 ; T or 1 ; x; T space, and hence, azeotropic behaviour, similar to that discussed above, occurs in T = constant planes. The di erence between an azeotropic point and a critical point for phase separation should be noted. At a critical point, 4s and 4v, as well as 4x, are zero.
1 = ,P ; 2 = 1 , 2 ; 3 = 2 :
1 2
=v; = x;
1.96

Exact solutions for the thermodynamic functions of lattice models exist only for equilibrium states see Chap. 8 and assume the forms shown in Figs. 1.5
24

P
Fig. 1.7. Positive azeotropy; P is azeotropic point and horizontals are tie lines.
and 1.6, when a rst-order transition occurs. We now consider how such a transition appears in classical approximations, taking as our guide the van der Waals gas. In the latter we have seen that intensive thermodynamic variables are continuous over the entire range of the volume per molecule v, including the interval where the homogeneous uid is unstable and separates into conjugate phases of di erent density. From 1.79 dfn,1 = dn = , i di ; 1.100 given that T and j for j 6= i, j 6= n are held constant, which will be assumed for the rest of this paragraph. For a one-component uid with `per molecule' densities, n = and since n = 2 we have to put i = n,1 = ,P , i = v. In a classical theory, where n and i are continuous over the whole range of i , the conditions for phase separation can be satis ed only if the curve of i against i is s-shaped, as in Fig. 1.1. Thus the derivative @i =@ i 0 for large or small values of i , but there is an interval where @i =@ i 0. From 1.100, @fn,1=@ i changes sign when @i =@ i does, assuming i 6= 0 at this point. Hence a plot of fn,1 against i yields a curve like ABCDE in Fig. 1.8. The part ABE represents stable states, since it corresponds to the least value of fn,1 for each i and is equivalent to the equilibrium curve of Fig. 1.5. At the point B, where the lines AC and DE intersect, the slope of the equilibrium curve ABE is discontinuous and hence B corresponds to the rst-order transition. The loop BCD represents unstable states, and the cusps at C and D correspond to the unstable values of i , where @fn,1=@ i and @i =@ i change sign. There are thus three equivalent ways of locating a rst-order transition and determining the corresponding phases in a classical theory: i Plot fn,1 against i with T and the elds j for j 6= i, j 6= n held constant obtaining enough points on the curves AC and ED see Fig. 1.8 to determine the point of intersection B.

C E
25
Fig. 1.8. Classical free energy eld
isotherm; AB and BE represent stable states.
fn,1
i
ii Plot i against i with T and the elds j for j 6= i; j 6= n held constant and draw the equal-areas tie line, as in Fig. 1.1. Numerically this is not as convenient as i. However, the existence of a range of i where is a su cient condition for an instability loop leading to phase separation. Also, the relations for the disappearance of such a range of i determine the position of a critical point end of a rst-order transition line. iii Plot fn,2 = fn,1 + n,1 n,1 against n,1 with T and the elds i , for j = 1; : : : ; n , 2, held constant. If phase separation occurs a part of the curve will be convex upwards and, from 1.95, the conjugate phases can be determined by drawing a double tangent, as in Fig. 1.2. The convex envelope is the equilibrium curve corresponding to Fig. 1.6. This method is as general as i and ii since the elds can be relabelled to make any i for i n into n,1 .
@i @i
1.101
@i @ i = 0;
@ 2 i = 0 @2 i
1.102
1.10.1 Metastability in a One-Component Fluid

So far we have not di erentiated between unstable states, which we have de ned as states where the appropriate thermodynamic potential is greater than its least attainable value. Taking the one-component uid as an example, suppose that M molecules occupy a xed volume V at a temperature T . Now assume that half the mass of uid undergoes a small contraction, taking the
26
molecular volume to v , v. The other half must then undergo a small dilation, taking its molecular volume to v + v. Using Taylor's theorem the change in the Helmholtz free energy of the whole uid is 2 1 A = 1 M @a v + 2 @ a v2 2 2 @v @a @v 1 @ 2a + 1 M @v , v + 2 @v2 , v2 + O v3 2 2 = 1 M @ a v2 + O v3 2 @v2 1.103 = , 1 M @P v2 + O v3 ; 2 @v using 1.48 for the last expression. Hence, if @P=@vT 0 then A 0. Thus, the free energy of the uid is not a minimum with respect to small amplitude, long-wavelength uctuations in the density. Such a situation will be termed internal or intrinsic instability. It follows that for internal stability the condition
@P
must apply. Many other thermodynamic inequalities can be derived. For instance by considering a temperature uctuation at constant volume it can be deduced that
@v T
1.104
CV = Tk @s 0 ; 1.105 b Ma @T V where CV is the molar heat capacity at constant volume, Ma is Avogadro's number and s is according to the conventions we adopted in Sects. 1.7 and
1.8 the dimensionless entropy per molecule. Since 1.104 implies that the isothermal compressibility KT 0 it follows from 1.30 and 1.105 that CP CV 0 ; 1.106 where CP is the molar heat capacity at constant pressure. By a similar proof to that of 1.104 it can be shown for the general system that, for internal stability,
other independent elds or densities and the temperature T being xed. By considering simultaneous uctuations of several variables, many more complicated thermodynamic inequalities can be derived Pippard 1967. We now return to the van der Waals gas and consider the parts of the range v1 ; v2 in Fig. 1.1 where the condition 1.104 is obeyed. For such a value of v the free energy of a homogeneous uid cannot be diminished by small density uctuations, but it can be reduced by separation into conjugate phases with
@i @i
0;
1.107
P
C A
27
B D Metastability
v Fig. 1.9. Liquid-vapour system: metastable region is hatched and C is the critical point.
molecular volumes v1 and v2 respectively. However, for such a separation, large changes in density must occur. A homogeneous uid in such a state is said to be metastable. We have thus distinguished two kinds of unstable state which are termed intrinsically unstable and metastable respectively. Intrinsically unstable states are in nitesimally close to states of lower free energy whereas metastable states are separated from them by nite intervals of certain variables. On Fig. 1.3 the coexistence broken curve divides stable from unstable regions. The curve de ned by @P=@vT = 0, which touches the coexistence curve at the critical point, subdivides the unstable region into domains of metastability and intrinsic instability respectively see Fig. 1.9. Such a curve is known as a spinodal. On the diagram, for a general classical system shown in Fig. 1.8, the segments BC and DB of the instability loop BCD correspond to metastable states whereas the segment CD corresponds to intrinsically unstable states. This latter are said to be subject to spinodal decomposition by long-wavelength uctuations.
1.10.2 The Experimental Situation

Metastable states are not simply artifacts of classical theory, but are frequently encountered experimentally. Suppose that on Fig. 1.9 a vapour is compressed from point D to a point B on the coexistence curve. If the system remains in thermodynamic equilibrium, any further increase in pressure will cause a transition to the liquid part of the isotherm starting at point A, AB being a tie line. However, when Fig. 1.9 represents the behaviour of a real uid, a dust-free vapour may progress along the broken curve continuous
28
with DB into the metastable region in Fig. 1.9. It is then termed supersaturated. Metastable states can also be reached by reduction of temperature, and the substance is then said to be supercooled. Supercooled liquids can sometimes exist at temperatures considerably below the equilibrium temperature of transition to the crystalline form. We now discuss brie y the physical reasons for the existence of metastable states in real substances, using the supersaturated vapour as an example. For the latter to go over to the liquid state, localized low-wavelength density uctuations of large amplitude must occur, resulting in the formation of nuclei, each containing a large number of molecules. Such a nucleus has a lower free energy than an equal mass of vapour and will grow by accretion. However, a nucleus created by a spontaneous local density uctuation is likely to be small and because of its relatively large surface area will have a higher free energy than an equal mass of vapour. This is another aspect of the local stability of metastable states. Nevertheless thermodynamically metastable states are di erent in character from static mechanical metastable states, like that of a ball on a surface at the lowest point of a depression separated by a ridge from a deeper depression. The degree of permanence of a thermodynamically metastable state depends on the relaxation time, the mean waiting time for a su ciently large uctuation to initiate the transformation to the stable state. For the mechanical analogy to be appropriate, in fact, the ball must be regarded as subject to impulses of random strength at random intervals. However, relaxation times can be very large. For instance, glasses are supercooled liquids with higher free energies than the crystalline forms whereas diamond is a metastable crystalline form of carbon. Thus, contrary to the common belief, diamonds are not `for ever', just for a very long time. In a supersaturated vapour dust particles drastically reduce the relaxation time by facilitating the formulation of large nuclei. Charged particles ions have the same e ect, which is used in the cloud chamber apparatus. Here the tracks of charged particles through supersaturated water vapour become visible because of the formation of liquid water droplets. Work on rst-order transition mechanisms is reviewed by Gunton et al. 1983 and Binder 1987.
Examples
1.1 From relations given in Sects. 1.1 and 1.2, show that for a uid of xed composition:
e! e! @T = , @P ; a e e @V @S e !S e ! V c @ P = @ S ; e e @T V @V T
e! e ! b @ T = @ V ; e e @P @S e !S eP! d @ V = , @ S : e e @T P @P T
Examples
29
c is proved in Sect. 1.3 and the other relations can be obtained in a similar way. These are Maxwell's thermodynamic relations. It can be seen that they have the same form if the variables T , S , P and V e e e e are used in place of T , S , P and V . 1.2 From relations given in Sect. 1.2, show that, for a uid of xed composition, e! e! 2 @ A=T ; 2 @ G=T ; e e U = ,T H = ,T
The rst two relations here are known as Gibbs Helmholtz relations and are special cases of 1.66. They can be seen to have the same e e form if T and V replace T and V . 1.3 In terms of the variables T , S , P and V the formulae 1.25 and 1.26 for the heat capacities CP and CV are modi ed to
e @ T e !V e G = ,V 2 @ A=V : e @V T
e @T
and
CP = kb T @T = kb @T P P @ 2 G = ,k M T @ 2 : = ,kb T @T 2 b a @T 2 CV = kb T @T = kb @T V V @ 2 A = ,k M T @ 2a : = ,kb T @T 2 b a @T 2
@S
@H @U
@S
e e e 1.4 De ning CH as in 1.35 and CM = T @ S =@ T M show that @ M 2 " @ M ! ,1 e : CH , CM = T e e @T H @H T
Determine the changes which are needed in the formulae 1.20 and 1.21 for P and KT .
Reference to the proof of 1.30 may be helpful. Determine what changes are needed in this formula if M and H are used in place of e e M and H. 1.5 For a one component uid, where n = 2, show that, putting Qn = M , equations 1.77 and 1.78 can be written
30
and that the equation of state can be expressed in the form " ! e e 1 vc Tc : ~e P =T exp 2 1 , e e e v~ Pc Tc 2~=vc , 1 vT ~ 1.7 Show that the spinodal curves for van der Waals' equation and Dieterici's equation are given respectively by: e~ i kb T v3 , 2c~ , b2 = 0, v e~ v ii kb T v2 , c~ , b = 0. Show that for Dieterici's equation the critical temperature occurs when ii starts to give real values for v. ~
if b is taken as the standard volume v0 of equations 1.54 and 1.55 and " = c=b. Using this equation, show that the constant temperature relation da = ,P dv can be integrated to give " a = , v , T ln v , 1 + T ; where is an arbitrary function of T . Obtain by integrating the constant temperature relation d = vdP and verify that the expressions for and a are compatible with the relation = a + Pv. 1.6 An alternative classical state equation to van der Waals' is that of Dieterici: e , e~ e P = kb T exp, c=kb T v : v b ~ e Show that the critical values of v, T and P are given by ~ e e e kb Tc = 4cb ; v c = 2b ; ~ Pc = 4c2 exp,2 b
Show that van der Waals equation of state 1.43 can be expressed in the form
" P + v2 v , 1 = T
f1 = u , Ts + Pv = a + Pv = ; df1 = ,sdT + vdP ; da = ,sdT , P dv :
2. Statistical Mechanics and the One-Dimensional Ising Model
2.1 The Canonical Distribution

In this chapter we present and discuss necessary results in equilibrium statistical mechanics. After introducing Ising lattice models we illustrate the theory by treating a one-dimensional Ising ferromagnet and nish with an application of the latter to a biophysical system. There are many texts on statistical mechanics which develop the basic theory; a brief selection is Huang 1963, Ma 1985, Chandler 1987, and Thompson 1988. We consider a collection or assembly of a number of atoms or molecules, which are not regarded as directly observable. The atoms or molecules are termed microsystems whereas the whole collection is a macrosystem. The macrosystem corresponds to an observable thermodynamic system, as studied in the previous chapter. It is supposed that the temperature T and the extensive variables Q1 ; : : : ; Qn are speci ed, the numbers of molecules of the various components being included in the Qi . Such a system is mechanically and chemically isolated, but is in thermal equilibrium with its environment, which can be thought of as a heat bath at temperature T . The given values of the extensive variables are compatible with a number of di erent states of the microsystems or microstates. For instance, in an assembly of molecules in the gaseous state, many di erent sets of values of the position coordinates of the molecules are possible with a given volume V . The internal variables specifying the microstates will be denoted by and the space of microstates for prescribed values of Q1 ; : : : ; Qn will be denoted by . Each set of values of determines a value E of the energy of the assembly. The microstate is free to vary with time over all the states in the space and, as it does so, the energy E uctuates in value. However, the microstates successively occupied by the assembly are not uniformly distributed over because a fundamental probability distribution rule is obeyed. For discrete-valued , the probability of a given set of values of is given by the Boltzmann distribution exp ,E =T ; 2.1 p = where
32
Z=
X
exp ,E =T :
2.2
This de nes the canonical distribution with Z , de ned by 2.2, being the canonical partition function. Its form ensures that the necessary consistency condition X p = 1 2.3 is satis ed. For a quantum system, each set of speci es an eigenstate; the E are energy eigenvalues and Z can be regarded as the trace of b b exp,H=T , where H is the Hamiltonian operator. For continuous , p d , with p given by 2.1, is the probability of the microstate lying in the element d of . Correspondingly, the summations in 2.2 and 2.3 are replaced by integrations. We shall use the discrete rather than the continuous formulation since our interest is in discrete lattice systems. In justifying equation 2.1 it is customary to think of a collection or ensemble of a large number N of replicas of the given assembly, and N p then gives the number of members of the ensemble in microstate . The ergodic hypothesis identies averages taken over the ensemble using 2.1 with time averages over the uctuating microstate of a single assembly. We shall simply assume the truth of 2.1 and of the corresponding expressions for di erent sets of independent thermodynamic variables given in Sect. 2.2. For further discussion, see Jancel 1969. The mean or expectation value of the energy E is, by 2.1, P E exp ,E =T X hE i = p E = ; 2.4

Z which is termed the ensemble average. If hE i is identi ed with the equilibrium internal energy U then, from 2.2 and 2.4, 1 ln U = hE i = , Z @ @Z 1 = , @@T ,Z : 2.5 1 T, The energy E depends on the prescribed values of Q1 ; : : : ; Qn and the response of the system in state to variation of Qi is given by @E =@Qi . We identify the eld i with the mean value of this derivative and hence write E X D ln j = @E = p @E = ,T @@QZ ; i = 1; : : : ; n : 2.6 @Qi @Qi i
It can be seen that 2.5 becomes identical with the Gibbs Helmholtz relation for the Helmholtz free energy A 1.66 with = 0 if A = ,T ln Z : 2.7 Equation 2.6 then becomes
2.1 The Canonical Distribution
33
which gives the same relation between elds, thermodynamic potential and extensive variables as 1.65. Hence the basic relations of thermodynamics are consistent with statistical mechanics if the identi cations 2.5, 2.6 and 2.7 are made.
@A i = @Q ;
i
i = 1; : : : ; n ;
2.8
2.1.1 The Thermodynamic Limit

The formulae 2.5 and 2.6 are examples of the general principle that dependent thermodynamic quantities are identi ed in statistical mechanics with the mean or expectation values of uctuating variables. As with any application of probability theory expectation values increase in signi cance as the spread, or variance, of the uctuating variable decreases. As we shall see in Sect. 2.5 this convergence towards `thermodynamic behaviour' occurs, in general, as the size of the system increases. As in Sect. 1.8, we now divide the extensive variables Qi , i = 1; : : : ; n , 1 by Qn to obtain densities i and for de niteness identify Qn with M , the total number of molecules. Then A is a function of T , 1 ; : : : ; n,1 and M . Since A is an extensive variable we can put A = MaT; 1; : : : ; n,1 where a is a thermodynamic potential density as de ned in Sect. 1.8. Hence for statistical mechanics to yield thermodynamic behaviour, ln Z should, like A, be proportional to M but, generally speaking, this will not be true for small M . However, thermodynamic systems consist of a very large number of molecules, roughly of the order of Avogadro's number Ma . In principle, we can go to what is known as the thermodynamic limit and replace 2.7 by a = ,T lim ln Z : 2.9 The model used is a valid one only if the mathematical limit on the right-hand side exists. In practice, we frequently nd that ln Z = Mf T; 1; : : : ; n,1 + f 0 ; 2.10 0 =M ! 0 as M ! 1. Discarding f 0 before substitution in 2.7 is then where f equivalent to using the thermodynamic limit. We can in fact discard factors in Z which are as large as M k , where k is a constant, because k ln M =M ! 0 as M ! 1.1 Size e ects will be discussed again in Sects. 2.5 and 2.6.
M !1
2.1.2 Kinetic and Con guration Variables

It may happen that the internal variables can be divided into two sets 1 and 2 , corresponding to a separation of into subspaces 1 and 2 , such that
1
For further discussion of the thermodynamic limit see Volume 2, Chap. 4 and for rigorous results see Ruelle 1969 and Gri ths 1972.
34
Then, from 2.2, Z = Z1 Z2 ; where X Zi = exp ,Ei i =T ;

i
E = E1 1 + E2 2 :
2.11 2.12
i = 1; 2 :
2.13
2 , irrespective of 1 , is given
Thus the probability of a given set of values of by X p 2 = p = exp ,E2 2 =T

1
= exp ,E2 2 =T
1 exp ,E1 1 =T
and there is a similar expression for p 1 . It can be seen that, provided 2.11 applies, 1 and 2 can be treated as if they corresponded to di erent systems. We shall assume a separation between momentum and con gurational variables such that 2.11 applies with E1 = Ekin , the kinetic energy of the assembly, and E2 = Ec , the con gurational energy, which usually derives from intermolecular interactions. Then, from 2.12 and 2.13, Z = ZkinZc ; 2.15 where X Zc = exp ,Ec c =T ; 2.16 is the canonical con gurational partition function. Now suppose, as in Sect. 1.7, that the system is composed of di erent types of microsystems with Mj microsystems of type j for j = 1; : : : ; . Then, if Ekin is the sum of the kinetic energies of the individual microsystems,
c
Z2
2.14
Zkin =
Y
Thus if we omit Zkin from consideration, as we usually shall, we are omitting terms dependent on temperature alone from U and A. We next de ne the con gurational elds. For the chemical potentials corresponding to components j = 1; : : : ; let jc = jc = j + T ln j T 2.18 and, for elds which are not chemical potentials, we have simply
j =1
j T Mj :
2.17
2.2 Distributions in General
35
Then, from 2.6, 2.15 and 2.17, for all con gurational elds = ,T @ ln Zc = @Ac ; i = 1; : : : ; n ;
ic
ic = i :
2.19 2.20
where the con gurational Helmholtz free energy is Ac = ,T ln Zc : 2.21 By 2.18 and 2.19 the con gurational elds ic are either the same as the original elds i or di erent by a quantity dependent on temperature alone. Hence the i in the phase equilibrium conditions 1.90 can be replaced by the ic for i = 1; : : : ; n. The probability distribution in the con gurational subspace c is independent of kinetic terms by 2.14 and hence we can derive =T @ Z 2.22 Uc = hEc ci = , @ ln,1c = @@Ac,1 T T by arguments similar to those used to obtain 2.5.
@Qi
@Qi

Now consider probability distributions corresponding to the thermodynamic systems discussed in Sect. 1.7, where the independent variables are the temperature T , the elds 1 ; : : : ; and the n , dimensionless extensive variables Q+1 ; : : : ; Qn . The space of microstates compatible with the prescribed values of Q+1 ; : : : ; Qn will be denoted by . It includes as a subspace. Each set of values of determines a value E of the internal enb ergy and values Qi of the dependent extensive variables 1 i . As varies over the space of permitted microstates, there will be uctuations b b not only of the energy but also of Q1 ; : : : ; Q . Note that the character of the distribution depends not only on the value of but on the selection of the pairs of conjugate variables labelled 1; : : : ; and +1; : : :; n respectively. The case = 0, however, is unambiguous. As for the canonical distribution, the general distribution can be regarded as giving microstate probabilities for the members of a collection or ènsemble' of similar assemblies. The Hamiltonian b H , which is de ned by
b H
X b = E , i Qi ; i=1
2.23
is a crucial quantity in the statistical mechanics of the system with independent thermodynamic variables T , 1 ; : : : ; , Q+1 ; : : : ; Qn . It is the microscopic equivalent of the thermodynamic enthalpy de ned by 1.59. With b H de ned, the fundamental distribution law is that the probability of the microstate is given by
36
p
with the generalized partition function X b Z T; 1 ; : : : ; ; Q+1 ; : : : ; Qn = exp ,H =T :

b = exp ,H =T Z

2.24 2.25
From 2.24 and 2.25 it can be seen that the probabilities satisfy a consistency condition similar to 2.3. Now suppose that the thermodynamic enthalpy H is identi ed with the b mean value of H . From 2.23, 2.24 and 2.25, X X @ Z b b Hr = hH i = p H = U , i Qi = , @ ln,1 : 2.26 T i=1 If the equilibrium value Qj of a dependent extensive variable is identi ed b with the mean value of Qj then, from 2.24, X b b Q = hQ i = p Q = T @ ln Z ; j = 1; : : : ; : 2.27
j j
In the microstate , the response of the system to variation of the independent extensive variable Qj , j = + 1; : : : ; n, is the partial derivative of b H with respect to Qj and the mean value of this is identi ed with the thermodynamic value of the conjugate eld j to give D b E X b = @ H = p @ H = ,T @ ln Z ;
j

@j
2.28 The àveraging' relations 2.26, 2.27 and 2.28 become identical with the thermodynamic relations in 1.66 and 1.65, if the identi cation F = ,T ln Z T; 1 ; : : : ; ; Q+1 ; : : : ; Qn 2.29 is made. System size has the same signi cance in the general distribution as in the canonical distribution. Scaling by Qn , as in 1.76, a rigorously correct de nition of the free energy density is f = ,T lim ln Z : 2.30

@Qj

@Qj
@Qj j = + 1; : : : ; n :
However, an equivalent procedure is to retain the part of F , in 2.29, which is proportional to Qn discarding terms of lower order. There is a close connection between the general partition function Z and the canonical partition function Z , which corresponds to the case = 0. This can be demonstrated by performing the summation of 2.25 in two stages, b b the rst over the values of corresponding to xed values of Q1; : : : ; Q , the
Qn !1
Qn
37
second over all values of the latter. However the rst summation, for which all extensive variables are xed, is precisely that required to obtain a canonical partition. Hence we have
X X b ! Z T; 1 ; : : : ; ; Qn,+1; : : : ; Qn = exp i Qi =T i=1 b b fQ1 ;:::;Q g b1; : : : ; Qn : 2.31 b Z T; Q
Now suppose that E = Ekin + Ec, as assumed in Sect. 2.1.2. A con gurational Hamiltonian and partition function are de ned by
b Hc c = Ec c ,
Z c =
X b ic Qi c ; i=1
b where it is assumed that Qi are dependent on the con gurational microstate. The ic i = 1; : : : ; n are de ned by 2.18 and 2.19. By substituting Z = Zkin Zc into 2.31 and using 2.17 for Zkin,
Z =
0 Y fjg
c
b exp ,Hc c=T ;
2.32
j T Mj Zrc ;
2.33
where the prime indicates that only factors corresponding to Mj which are independent variables are included in the product. From 2.18, 2.19, 2.23, and 2.32
b H = Ec + Ekin ,
00 X b b Xc i Qi = Hrc + Ekin + T Mj ln j T ; i=1
c where the double prime denotes a sum of factors corresponding to the Mj which are dependent variables. Hence Y b b b exp,H =T = exp,Hc =T exp,Ekin=T j T ,Mj :
00
fjg
2.34
Substitution of 2.33 and 2.35 into 2.24 followed by summation over the momentum microstates yields b exp ,Hc c =T ; p = 2.36
c c
fj g
2.35
where pc c is the probability of the con gurational microstate con gurational free energy is de ned by
Zc
c.
The
38
2.37 and then, from arguments similar to those used in deriving 2.27, 2.28 and 2.26, Q = T @ ln Zc = , @Fc ; j = 1; : : : ; ; 2.38
j
Fc = ,T ln Zc T; 1c ; : : : ; c ; Q+1 ; : : : ; Qn
@jc
jc
ln Z jc = ,T @ @Q c = @Fc ; @Q
j j
j = + 1; : : : ; n ;
2.39 2.40
i=1 where Sc = ,@Fc=@T is the con gurational entropy and Hc is the con g-
Hc = Uc ,
c ic Qi = , @ ln Z1 = Fc + TSc ; @ T, X i=1 n X
urational enthalpy. From 2.38 and 2.39 dFc = ,Sc dT ,
Qi dic +
The con gurational free energy density is de ned by f = ,T lim ln Zc :

c Qn !1
i=+1
ic dQi :
2.41 2.42
For large enough Qn, we can put Fc = fc Qn and, by 1.70, Qi = i Qn , i = 1; : : : ; n , 1. Allowing the extensive variables to change with all intensive variables xed, 2.41 yields
Qn
0 n,1 X fc dQn = @ ic

i=+1
1 i + nc A dQn ;
2.43
which can be satis ed only if
f c =
n,1 X
A relation similar to the rst member of 1.77 can be deduced by dividing 2.40 by Qn . Again, by allowing the intensive variables to change at a xed value of Qn , 2.41 yields dfc = ,sc dT ,
X i=1 i dic + n,1 X i=+1
i=+1
ic i + nc :
2.44
ic d i :
2.45
From these results and those of Sect. 2.1 it can be deduced that all the relations of Sects. 1.7 and 1.8 are satis ed by the con gurational thermodynamic variables. It follows that we can treat the statistical mechanics and thermodynamics of the con gurational degrees of freedom as if the latter constituted a
2.3 Particular Distributions
39
separate system. Henceforward we shall deal exclusively with con gurational space and, accordingly, omit the index `c'. Again, although the index `' in symbols like Z , F , Fc is useful in general theory, it will be omitted in particular cases where the value of is well understood.

As remarked above, the canonical distribution is the case = 0 of the general theory in Sect. 2.2. We now consider several other particular cases which are useful in practice. The distribution treated in Sect. 2.3.1 will be used in the next section.
2.3.1 The Constant Magnetic Field Distribution

This corresponds to the thermodynamic theory of Sect. 1.4 where T and the magnetic eld H, de ned in 1.55, are the independent intensive variables. It can be obtained from the general theory of Sect. 2.2 by putting = 1 and Q1 = M, the dimensionless magnetization de ned in 1.54, so that 1 = H. We use a con gurational partition function as de ned by 2.32 and, omitting the indices c and = 1 as indicated above, write X b Z = exp ,H =T ; 2.46 b H = E , HM : From 2.38, the equilibrium magnetization is given by M = T @ ln Z = , @F : 2.47
@H
@H
2.3.2 The Constant-Pressure Isobaric Distribution

Consider a liquid or isotropic solid where the pressure is the only mechanical eld. We take T and P as the independent intensive variables so that, in the notation of Sect. 2.2, = 1 and Q1 = V . The independent extensive variables Qi , i 1 are the molecule numbers M1; : : : ; M, where there are components. There is a close analogy with the constant magnetic eld distribution and relations can be translated from one distribution to the other by interchanging ,P and V with H and M respectively. Making the same assumption about separability as in Sect. 2.3.1, a con gurational partition function can be de ned by X b Z = exp ,H =T ; 2.48 b = E + P V : b H
40
The con gurational Gibbs potential is then given by G = ,T ln Z ; 2.49 and, from 2.38 and 2.39, the equilibrium values of the volume V and the con gurational chemical potentials i are given by b V = hV i = ,T @ ln Z = @G ;
@P
@P
ln @G i = ,T @@MZ = @M ; i i
2.50
i = 1; : : : ; :
2.3.3 The Grand Distribution

This corresponds to a thermodynamic system with = , T and the chemical potentials i being independent intensive variables. A con gurational grand partition function is de ned by X b Z = exp ,H =T ; 2.51 b = E , X i Mi c H with the con gurational grand potential given by D = ,T ln Z : From 2.38, the equilibrium values of the molecule numbers are M = T @ ln Z = , @D ; i = 1; : : : ; ;
i i=1
2.52 2.53 2.54
and the elds which are not chemical potentials are = ,T @ ln Z = @D ; j = + 1; : : : ; n :

j
@i
@i
When n + 2, a eld which is not a chemical potential can be added to the set of independent intensive variables, so that = + 1, giving a constanteld grand distribution.
@Qj
@Qj
2.3.4 Restricted Distributions for Lattice Models

Suppose that the centre of each molecule or atom occupies a lattice site and that the N sites are equivalent and form a regular lattice occupying a volume V . Consider the following conditions e i The volume per lattice site v0 = V =N is constant.
41
ii Every lattice site is occupied by the centre of a molecule of one of the components. Condition i will be assumed for all the lattice models discussed in this book, and v0 can be used as the standard volume in 1.54 to de ne the reduced volume e V = V =v0 = N : 2.55 The volume v0 must be replaced by a length per site and an area per site for one- and two-dimensional systems respectively. Condition ii is less general and we shall consider a number of models with vacant sites. However, when ii is assumed, it follows that
V =N =
X j =1
Mj = M ;
2.56
so that the variables M1 ; : : : ; Mk ; V are no longer independent, and n is reduced by one. Any partition function obtained under the condition M = N is called `restricted', and the same description can be applied to the model itself. Using a transformation similar to that of 1.82, but incorporating a pressure term, it follows from 2.56 that
,P dV +
X j =1
j dMj =
,1 X j =1
j , dMj + , P dN ;
2.57
where the j are now con gurational chemical potentials. The set of + 1 elds j , j = 1; : : : ; , ,P are thus replaced by the set of elds j , , j = 1; : : : ; , 1, , P . As we saw in Sect. 1.9 the conditions for phase equilibrium are that T , j , , j = 1; : : : ; , 1, , P and any other relevant elds, such as H, must be equal in the two phases. The new phase equilibrium relations can be explained by the fact that, under conditions i and ii, a change in volume must be accompanied by a transfer of molecules, whereas an increase in the number of molecules of a given component at constant volume must be compensated for by a decrease in the number of molecules of another component. Now consider the distribution in which n = , P and 1 ; : : : ; n,1 , which include the elds j , , j = 1; : : : ; , 1 are taken as independent variables. The only independent extensive variable is Qn = N and the restricted partition function is given by X b Z = exp ,H =T ;
b H
= E ,
n,1 X i=1
b i Q i :
2.58
42
The associated free energy per site f is of type fn,1 in the nomenclature of Sect. 1.8. Using 2.42 and 2.44 we have, for large N , n,1 X f = ,T ln Z = u , Ts , = = ,P : 2.59
From 2.45 df = ,sdT ,
i=1
i i
n,1 X i=1
i di :
2.60
2.4 Magnetism and the Ising Model

Ferromagnetism and other forms of permanent magnetism are due to exchange interaction between unpaired spins occurring either directly or through conduction electrons or non-magnetic component atoms, Martin 1967, Sinha and Kumar 1980, Mattis 1985, Izyumov and Skryabin 1988. In the anisotropic Heisenberg model the energy eigenvalues for an assembly of spins on a regular lattice are derived from a Hamiltonian operator
b H = ,2
n:n: X
where six , siy and siz are the component operators associated with the spin ^ ^ ^ on site i. The symbol n.n. stands for `nearest neighbour', and the summation is over all nearest-neighbour pairs of sites i; j , each pair occurring once only. The isotropic Heisenberg model is obtained when Ix = Iy = Iz Volume 2, Sect. 7.6. In most magnetic materials with localized spins there is a direction of easy magnetization so that jIz j say is greater than jIx j and jIy j Sinha and Kumar 1980. In the Ising model Ix = Iy = 0 and the Hamiltonian is now diagonal in the representation in which the commuting z -component spin operators siz are diagonal. It can thus be treated as the scalar con gurational ^ energy. The siz take the values s; s , 1; : : : ; ,s + 1; ,s, where s is the spin ^ 3 1 magnitude with possible values 1 ; 1; 2 ; 2; : : :. For s = 2 the microstates are 2 more conveniently represented by i = 2îz and each i can take the values s i = 1 and i = ,1, referred to as ùp' and `down' spins respectively. For a lattice of N sites the set of con gurational internal variables is thus given by the vector = 1 ; 2 ; : : : ; N ,1 ; N : 2.62 This has 2N possible values and the con gurational subspace has 2N discrete points. For the ferromagnet with Iz =2 = J 0 the con gurational energy is
fi;jg
Ix six sjx + Iy siy sjy + Iz siz sjz ; ^ ^ ^ ^ ^ ^
2.61
43
E = ,J
n:n: X
Since the contribution of the nearest-neighbour site pair i; j is ,J if i = j and +J if i 6= j , this favours parallel alignment. For general spin magnitude s let i = siz , J = 2Iz for integer s and i = 2îz , J = Iz =2 for half-integer ^ s s. This ensures that all i are whole numbers and that 2.63 applies for all s. The s = 1 case will be referred to simply as the Ising model and for any 2 higher value of s we use the term spin-s Ising model, as in Example 2.4 for the spin-1 model. Since each site is occupied by a spin, Ising model partition functions are restricted in the sense of Sect. 2.3.4. It is assumed that the magnetization of the Ising model is in the z direction, so that the dimensionless magnetization associated with a given spin con guration on the lattice is
fi;jg
i j:
2.63
b X M= i:
N
Now E , as given by 2.63, is invariant under the operation i ! , i all i, that is under reversal of all spins. This important symmetry property is true for all s and for the antiferromagnetic case, where ,J in 2.63 is replaced by +J J 0. In the canonical distribution the magnetization M is a xed independent variable and the summation over i in the partition function Z P is restricted to those lattice spin con gurations for which i i = M. From the symmetry property of E and 2.64 it follows that Z is unchanged when M ! ,M. Hence the Helmholtz free energy A satis es the symmetry relation AN; M; T = AN; ,M; T : 2.65 In the constant magnetic eld distribution it follows from 2.63, 2.64 and 2.46 that the con gurational Hamiltonian for the Ising model ferromagnet is
n:n: b = ,J X H
i=1
2.64
which is invariant under the operation i ! , i all i, con gurational constanteld partition function is thus
fi;jg
i j ,H
X
fig
i;
2.66
Z=
X
0 n:n: X exp @ J
This is a restricted partition function of the type de ned by 2.58. Since the summation is over all spin con gurations, Z is unchanged when H ! ,H. It follows that the corresponding free energy satis es the symmetry relation F N; T; H = F N; T; ,H 2.68
T fi;jg i
1 0 1 X A H i : j A exp @ T
fig
H ! ,H. The
2.67
44
and then, from 2.47, the equilibrium magnetization obeys MN; T; ,H = ,MN; T; H : 2.69 A relative magnetization or magnetization density m is de ned by m = M=N 2.70 and, since m is an intensive variable, 2.69 implies mT; ,H = ,mT; H : 2.71 Although certain cobalt salts are thought to have an Ising-type interaction de Jongh and Miedema 1974, Wood 1975 the extreme degree of anisotropy in the Ising model makes it a rather arti cial model for most real ferromagnets. Indeed any theory which, like the Ising or Heisenberg models, depends on localized spins is unrealistic for iron-group metals Martin 1967, Mattis 1985. Nevertheless the Ising model ferromagnet in two or three dimensions does reproduce the most important property of a real ferromagnet in that it displays spontaneous magnetization below a critical Curie temperature, where there is a singularity in the heat capacity. Also the Ising model ferromagnet is mathematically equivalent to models representing a number of di erent physical phenomena, such as zeroeld antiferromagnetism Chap. 4, liquid vapour phase separation Chap. 5 and phase separation or sublattice ordering in mixtures Chap. 6. When an accurate or approximate mathematical solution has been found for one model, a quantity of physical interest in another model can be obtained by identi cation with the appropriate quantity in the rst model.
2.4.1 The One-Dimensional Ferromagnet in Zero Field

Suppose that N identical microsystems are situated at N equally spaced lattice points along a line and that each microsystem is in one of distinguishable states. Suppose also that in the distribution used the con gurational b Hamiltonian Hc , is the sum of contributions from nearest-neighbour pairs and that when the two microsystems of such a pair are in states and respectively, this contribution is h . The con gurational partition function is X Z= exp ,h + h + + h! =T ; 2.72 where the sum is over all the states ; ; : : : ; ! on the N sites of the lattice. The variables V = exp,h =T 2.73 2 V and 2.72 can be expressed form the elements of the transfer matrix in the form
2
f ; ; ;:::;;!g
A general discussion of the use of transfer matrices is given in Volume 2, Chap. 4.
2.4 Magnetism and the Ising Model = +1 +1
45
,1
+1
,1
,1
,1
+1
+1
Fig. 2.1. One-dimensional Ising model ferromagnet; pair energies are shown underneath.
,J
+J
+J
+J
,J
,J
+J
,J
Z=
X
f ; ; ;:::; ;;!g
V V V V! :
2.74
We now consider the special case of the one-dimensional Ising model ferromagnet shown in Fig. 2.1 for which = 2 see also Thompson 1972a. It may be thought that one-dimensional or linear lattice models are unrealistic. However, they provide examples of systems with interacting molecules `many-body' systems in the terminology of theoretical physics which are soluble and instructive. Moreover, we shall see in Sect. 2.7 that there is at least one case where the one-dimensional Ising model can be used as a simple model for certain aspects of a real physical system. For the Ising model it is useful to de ne the Boltzmann factors X = expJ=T ; Z = expH=T : 2.75 For the one-dimensional Ising model with H = 0 the Hamiltonian de ned by 2.66 consists of a contribution ,J k k+1 from each successive site pair k; k + 1. The matrix V , de ned by 2.73, is given by X X,1 V = X,1 X 2.76 and has the property that X V = X + X,1 ; 2.77 is independent of . Hence, summing over ! in 2.74, the constant magnetic eld con gurational partition function for H = 0 is given by X Z = X + X,1 V V V : 2.78 Repeating this operation over the N , 2 remaining factors gives Z = 2X + X,1 N ,1 = 2N coshJ=T N,1 2.79 and ln Z = N fln 2 + ln coshJ=T g , ln coshJ=T : 2.80 Since the last term divided by N goes to zero as N approaches in nity
f ; ; ;:::; ;g fg
46 0:5
c0 kb
0:25
T=J
Fig. 2.2. Heat capacity plotted against temperature for the model of Fig. 2.1.
in the thermodynamic limit. With H = 0 the enthalpy H = U , the con gurational energy, and, by 2.40 U = @ F=T = ,NJ tanhJ=T : 2.82 @ T ,1 Hence U ! ,NJ as T=J ! 0 and U ! 0 as T=J ! 1. From 1.35 and 2.82 the constanteld con gurational heat capacity per spin with H = 0 is
F = ,NT fln2 + ln coshJ=T g
2.81
the majority of the dipoles are lined up in one of the two directions. However, this cannot be the case since, as will be shown in Sect. 2.4.2, the equilibrium magnetization M is zero for H = 0 at all temperatures T 0. To understand the behaviour of the model it is useful to derive the two-site correlation function h i i+k i, which is positive if the k-th neighbour pair of dipoles tends to point in the same direction. First note that, since i2 = 1 for any site i, 2 2 2 i i+k = i i+1 i+2 i+k,1 i+k 2.84 = i i+1 i+1 i+2 i+k,1 i+k :
2 J=T 2 c0 = ,N ,1kb T @ F = N ,1 kb @U = kb coshJ=T : 2.83 @T 2 @T From 2.83, c0 ! 0 as T=J ! 0 and as T=J ! 1. A plot against temperature is shown in Fig. 2.2. From the limiting value ,NJ of the con gurational energy as T=J ! 0 it might be supposed that at low temperatures T=J 1
47
Now temporarily modify the model by supposing that the interaction parameter J has an individual value for each nearest-neighbour pair so that, in zero eld,
b H =,
N ,1
In exactly the same way as 2.79 was obtained, it can now be shown Stanley 1987 that
k=1
Jk k k+1 :
2.85
Z = 2N
N ,1
b Using 2.84 and the expression 2.85 for H , X b h i i+k i = Z ,1 i i+1 i+1 i+2 i+k,1 i+k exp,H=T
@ = T Z @J @J@ @J@Z : 2.87 i i+1 i+k,1 Substituting from 2.86 and, nally, restoring the original model by putting Ji = J all i, h i i+k i = tanhJ=T k : 2.88 The nearest-neighbour correlation function tanhJ=T is obtained by putting k = 1 in 2.88: this provides an alternative way of deriving the expression 2.82 for U . From 2.88 the correlation function for an k-th neighbour pair is independent of its position on the lattice and is always positive but decreases as k increases. For any xed value of k, h i i+k i ! 1 as T=J ! 0 and to zero as T=J ! 1. The picture which emerges is that, although the total numbers of up and down dipoles remain equal, blocks of similarly oriented dipoles form and steadily increase in length as T=J decreases. The approach of U to the limiting value ,NJ is explained by the number of unlike nearest-neighbour pairs, which represent block boundaries, becoming very small relative to the number of like pairs for T=J 1. Although there is no long-range order in the sense of a majority of the dipoles becoming oriented in one direction, the properties of the assembly at low temperatures are dominated by the short-range order expressed by 2.88.
k,1
k=1
coshJk =T :
2.86
2.4.2 The One-Dimensional Ferromagnet in a Field

We now consider the behaviour of the model in non-zero external eld H 6= 0. With the Hamiltonian given by 2.66 a term ,J k k+1 , 1 H k + k+1 2.89 2
48
can be attributed to each successive site pair and the transfer matrix 2.76 must be replaced by ,1 V = XZ1 X ,1 ; 2.90 X, XZ the Boltzmann factors X and Z being de ned by 2.75. The property, expressed by 2.77, that the two row sums of V are equal, is no longer valid and the method of successive summation employed in Sect. 2.4.1 is inapplicable. The way out is to replace the open linear lattice by a ring, supposing that site N is a nearest neighbour of site 1 and adding a term ,J N 1 to the energy. Since every site now has two nearest neighbours the expression 2.89 summed over the N nearest-neighbour pairs gives the Hamiltonian exactly and there are no ènd-e ect' terms. With the ring assumption an extra factor V! must be multiplied into each term in 2.74 to allow for the interaction between the microsystems on sites 1 and N and 2.74 can now be written Z = TracefV N g : 2.91 For the one-dimensional Ising model 2.91 yields Z = N + N = N 1 + 2 =1 N ; 2.92 1 2 1 where 1 and 2 with 1 2 are the eigenvalues of the symmetric matrix V given by 2.90. These eigenvalues can be obtained from jV , I j = 2 , X Z + Z,1 + X2 , X,2 = 0 2.93 giving 1 1 = 2 XfZ + Z,1 + Z , Z,1 2 + 4X,4 1=2 g ; 2.94 1 XfZ + Z,1 , Z , Z,1 2 + 4X,4 1=2 g : 2 = 2 Since 1 and 2 are both positive and 2 =1 1, 0 ln 1 + 2 =1 N 2 =1 N ! 0 2.95 as N ! 1. Hence, in the thermodynamic limit, F = ,T ln Z = ,NT ln 1 : 2.96 It is of interest to return brie y to the zeroeld case H = 0, where Z = 1. Then 2.94 reduces to 1 = X + X,1 = 2 coshJ=T ; 2.97 2 = X , X,1 = 2 sinhJ=T ; and 2.92 can be written Z = 2N coshJ=T N f1 + tanhJ=T N g: 2.98
2.5 Fluctuations and Entropy
49
The di erence between this and 2.79 is due to the use of the ring lattice in one case and the open lattice in the other. However, substitution of 1 from 2.97 in 2.96 gives a thermodynamic limit expression identical to 2.81, which was derived for the open lattice. We have an example of a general rule that di erences due to boundary conditions disappear in the thermodynamic limit. From 2.98 and 2.79 it can be seen that convergence is faster in the closed ring case, as we might intuitively expect. The transformation H ! ,H gives Z ! 1=Z and it can be seen from 2.94 that the free energy given by 2.96 obeys the symmetry relation 2.68. From 2.47, 2.96, 2.94 and 2.75 the relative magnetization is given by sinhH=T m = Z @ lnZ1 = 2.99 2 H=T + exp,4J=T 1=2 : @ sinh The symmetry condition 2.71 is satis ed and m is a continuous function of H with m = 0 at H = 0 for any T=J 0. For any xed T=J 0, m ! 1 depending on the sign of H as jHj=J ! 1 while, for any xed H=J 6= 0, m ! 1 as T=J ! 0. For T = 0, m thus jumps at H = 0 from the value ,1, for H 0, to the value of +1, for H 0. Equation 2.71 is true for any lattice dimension. However, for two- or three-dimensional lattices, m is discontinuous at H = 0 not only at T = 0 but for a range 0 T Tc, where Tc is a critical temperature. We now consider the isothermal susceptibility T which was de ned by 1.36. However, in the one-dimensional system we must replace the volume v per atom by the lattice spacing, which is denoted by b. Then 2.99 yields ~ exp,4J=T coshH=T 0 b T = @m = 2.100 2 m0 @ H T sinh2 H=T + exp,4J=T 3=2 ; m0 being the moment of an Ising dipole and 0 the permeability of free space. On the H = 0 axis 2.100 reduces to 2.101 For higher-dimensional lattices, T has a singularity on the H = 0 axis for a non-zero critical value of T , which is associated with the onset of long-range order. For the one-dimensional case, where long-range order is absent, 2.101 yields an essential singularity at T=J = 0 and @m=@ H ! 1 as T=J ! 0. For any H=J 6= 0, it can be seen from 2.100 that @m=@ H ! 0 as T=J ! 0 and thus has no singularity, a situation which is paralleled in higher dimensions.
@m ,1 @ H = T exp2J=T :

Thermodynamically a system with prescribed values of the temperature T and the extensive variables Q; : : : ; Qn has a determinate internal energy U ,
50
which we have identi ed with the mean value hE i. However, the canonical distribution displays a range of energy values, with the probability of a microstate of energy E being given by 2.1. For complete consistency with thermodynamics we have to show that for a large enough system the probability of a signi cant deviation of E from U is small. We de ne the energy uctuation by 4E = E , hE i = E , U ; 2.102 and, to estimate the likely deviation of E from U , we obtain the mean of 4E 2 , which is the variance of E denoted by Var E . Thus Var E = hE , U 2 i = hE 2 i , U 2 : 2.103 From 2.4 and 2.5, @U = @ hE i = 1 hE 2 i , U 2 ; 2.104 giving
@T
@T
T2
Var E = T 2 @U : 2.105 @T This is an example of a uctuation response function relation. In this case the uctuating quantity is the energy and the response function is a heat capacity. The spread of energy values is best expressed by the relative uctuation given by Var E 1=2 = T @U 1=2 : 2.106 Since U is proportional to M , the total number of molecules in the assembly, it follows from 2.106 that the relative uctuation will normally be OM ,1=2 and hence very small if M is greater than 1020 , say. Avogadro's number Ma = 6:022 1023. Hence observable deviations of the energy from the equilibrium value U are extremely unlikely. For instance, U = 3 MT for a 2 perfect monatomic gas and 2.106 gives the value 2=3M ,1=2. The mean or equilibrium value of the energy E is determined by a balance between the probability of a single microstate, which by 2.1 decreases monotonically as E increases, and the number of microstates with energy values near to E which, in general, increases with E . From 2.106 there is an overwhelming probability that the system will be in one of a set of microstates, corresponding approximately to an energy range U , Var E E U + Var E . Since all these microstates have an energy near U , a reasonable estimate of can be obtained by supposing that the probability of each microstate is exp,U=T : 2.107 p =
U
jU j
jU j @T
Hence substituting for Z from 2.7 and using 1.61
2.108 The entropy is often described as a measure of disorder, that is of our lack of knowledge about which particular microstate is occupied. For the general distribution discussed in Sect. 2.2, we can obtain a relation b for the variance of the Hamiltonian H :
e = pU ,1 = expU , A=T = expS = expS =kb :
51
b and the relative uctuation of H is then given by @H 1=2 b Var H 1=2 = T : jH j jH j @T

b b Var H = h H , H 2 i = T 2 @H @T
2.109 2.110
b j = 1; : : : ; : Var Qj = T @Qj ; @j Hence the relative uctuation of Qj is given by b Var Qj 1=2 = 1 T @Qj 1=2; j = 1; : : : ; :
jQj j jQj j
@j
The dependent extensive variables also uctuate, and it can be shown from 2.24 and 2.27 that 2.111 2.112
By scaling the extensive quantities with respect to M , the total number of molecules, it can be seen that the relative uctuations given by 2.110 or 2.112 are of OM ,1=2 , and observable deviations from equilibrium values are thus highly unlikely. We showed in the context of the canonical distribution that in any factor space of microstates there is an independent probability distribution 2.14 of the same form as the general probability distribution. If separation between momentum and con gurational terms occurs, giving a con gurational partition function as de ned by 2.32, then there is an independent probability distribution in con guration microspace. Thus in the zeroeld Ising model, we can use 2.105 for the relative uctuation of the con gurational energy, giving, after substituting for U from 2.82, Var E 1=2 = T @U 1=2 = p 1 : 2.113 jU j jU j @T N sinhJ=T 2 2 This is of ON , 1 unless T is so large that T=J itself is of ON 1 . For the Ising model in non-zero eld H, 2.112 gives the relative uctuation of the magnetization as @m 1=2 b Var M 1=2 = p 1 T
jMj
@H exp,2J=T H 1=2 : = cosh2 =T N sinh H=T sinh2 H=T + exp,4J=T 1=4
N jmj
2.114
52
2. Statistical Mechanics and the One-Dimensional Ising Model 2 This is of ON , 1 unless H=T is very small. The relative uctuation com-
pares the likely deviation of a variable from its equilibrium value with the equilibrium value itself and thus ceases to be useful when the equilibrium value approaches zero, as M does in the one-dimensional Ising model when H ! 0. However, uctuation response function relations like 2.109 or 2.111 can still be used. In the last paragraph, variances were calculated from known response functions. However in certain circumstances, response functions can be calculated from variances, using uctuation response function relations. For the zeroeld Ising model, 2.111 yields the uctuation susceptibility relation
@M
@H
since M = 0 at H = 0. From 2.64
H=0
b b = T ,1Var M H=0 = T ,1hM i :

2
2.115
b hM2 i = X
*X !2+ N
i=1 i
=N +2
X
i;k;k 0
i i+k i :
2.116
For the open linear lattice, the correlation function h i i+k i is given by 2.88, and if N , i is large we can write without appreciable error = tanhJ=T 1 , tanhJ=T ,1: 2.117 In the thermodynamic limit it can be assumed that N , i is large for all except a relatively negligible number of sites, and hence b 2i = N 1 + 2 tanhJ=T = N exp2J=T : hM 2.118 1 , tanhJ=T Substitution in 2.115 then reproduces equation 2.101 for @m=@ H which we have thus succeeded in calculating from a uctuation susceptibility relation. We now look at what happens when T is reduced to zero, with independent elds and n , independent extensive variables kept constant. b b Suppose that there is a microstate 0 such that H 0 H for all 6= 0 . Then, by 2.24, the probability ratio p =p 0 ! 0 as T ! 0 for all 6= 0 . Hence, at T = 0, p 0 = 1 and the microstate 0 is called the ground state. Thus the equilibrium state at T = 0 can be identi ed by b nding the least value of H . This remains true if 0 microstates, with b 0 1, have the same least Hamiltonian value H 0 . The ground state is now degenerate and each of the 0 microstates composing the ground state has probability ,1 at T = 0. The one-dimensional ring Ising model 0 ferromagnet has 0 = 2 at zero eld, since the two states i = 1 for all i and j = ,1 for all k both have the least Hamiltonian value ,NJ . When
fk 0 g
i+k i =
1 X
k=1
tanhk J=T
2.6 The Maximum-Term Method
53
0 ! 1 in the thermodynamic limit Qn ! 1, the ground state is said to be in nitely degenerate. b By the rst relation of 2.26, H ! H 0 as T ! 0, so that the ground state can be regarded thermodynamically as that of least enthalpy or of least enthalpy density h = Q,1H . Since the occupation probability of n microstates other than those composing the ground state tends to zero with T , we have ! 0 as T ! 0, so that, by 2.108, S = ln 0 at T = 0. Normally Q,1 ln 0 ! 0 as Qn ! 1 so that the entropy density s = 0 n at T = 0. However, as will be seen in Sect. 8.12 and Chap. 10, there are exceptional cases where s 6= 0 at T = 0, and we then say that a zero-point entropy exists.

In the canonical distribution, all the microstates having the same energy value E` can be regarded as grouped into an energy level. The number of microstates having energy E` is denoted by `, call the weight or degeneracy of the level. The partition function, de ned by 2.2, can be written as a sum over levels: X X Z = ` = 2.119 ` exp,E` =T ; where E`+1 E` . From 2.1 the probability of level ` is ` =Z . For a system with a su cient number of molecules say, more than 1020 there is, according to the discussion in Sect. 2.5, an overwhelming probability that the system will be in a level with E` close to the equilibrium internal energy value U . Thus a plot of ` against E` will be very highly peaked near E` = U and we can identify U with Emax , where max is the largest value of ` . If the gaps between successive levels are small in the vicinity of Emax we can regard E` as a continuous variable and write the maximization conditions as d d2 ` ` 2.120 dE` E` =Emax = 0 ; dE 2 E` =Emax 0 : In view of the work on uctuations we can readily accept the equality of Emax and U , but what is surprising is that we can put S = ln max ; 2.121 A = Emax , T ln max = ,T ln max : Thus, in the thermodynamic limit, we are identifying ln max with ln , being the measure of `disorder'. However, it must be remembered that even if max, the degeneracy of the maximum probability level, is less than by a factor M ,k , for some k 0, this will have no e ect in the thermodynamic limit, since M ,1 lnM k ! 0 as M ! 1.
f`g f`g
54
2.6.1 The One-Dimensional Ising Ferromagnet

The maximum-term method will now be applied to the Ising model ferromagnet on a one-dimensional ring lattice. Since the canonical distribution is used, the magnetization M is xed and hence so are the numbers, denoted by N1 and N2 respectively, of +1 and ,1 spins. We can write
N 1 X1 + ; N1 = 2 i i=1 N 1X N2 = 2 1 , i ; i=1
M = N1 , N2 ;
2.122
where the last relation follows from 2.64. Hence, on a lattice of N sites, 1 1 N1 = 2 N 1 + m ; N2 = 2 N 1 , m ; 2.123 where m is the relative magnetization, de ned by 2.70. For any distribution on a ring lattice, the number of times a +1 spin on site i is followed by a ,1 on site i + 1 is equal to the number of times a ,1 spin is followed by a +1. Denoting this number by X , there are
N 1 X1 , 2X = 2 i i+1 ; i=1
N +1 = 1 ;
2.124
unlike nearest-neighbour pairs and hence the con gurational energy is
E = ,J
N X i=1
i i+1 = ,JN + 4JX :
2.125
The energy of a spin con guration on the ring lattice is determined by X . We need an expression for N1 ; N2 ; X , the number of distinct con gurations for given N1 , N2 , and X , assuming that spins of the same sign are indistinguishable but lattice sites are distinguishable. It is useful to de ne a number gN1 ; N2 ; X by N1 2 gN1 ; N2 ; X = N , X !N!N,! X !X !2 = gN2 ; N1 ; X : 2.126 1 2 Now arrange N1 up spins in a row and distinguish X of them by a `star'. Arrange N2 down spins in another row, put a star on the last spin and distribute X , 1 stars on the remaining ,1 spins. The number of ways of assigning stars in this manner is N1 ! N2 , 1! X 2.127 N1 , X !X ! N2 , X !X , 1! = gN1 ; N2 ; X N2 : Put the rst up spin on lattice site 1, the next up on lattice site 2 and continue until the rst starred up has been placed. Since the star indicates an unlike pair, put the rst down spin on the next lattice site and continue
55
until the rst starred down has been placed. Then resume putting the up spins on successive lattice sites and proceed, changing from spins of one sign to another after a starred spin has been placed. The proviso that the last of the row of down spins must carry a star is necessary because, whether placed on site N or a previous site, this spin is followed by an up spin. Each assignment of stars determines a distinct spin con guration on the lattice, and hence the number of con gurations with site 1 occupied by an up spin is given by 2.127. Interchanging N1 with N2 in 2.127 gives the number of con gurations with site 1 occupied by a down spin and hence, using the symmetry property of gN1; N2 ; X ,
X X N1 ; N2 ; X = gN1 ; N2 ; X N + N 1 2 NX = gN1 ; N2 ; X N N :

1 2
2.128
Since the logarithm of NX=N1N2 remains nite as N ! 1, we can drop this factor from . Hence a relation of form 2.119 for the con gurational partition function can be written X expJN=T Z N; M; T = X =
X
X
gN1 ; N2 ; X exp,4JX=T ;
2.129
where the factor due to the constant energy term ,NJ has been taken over to the left-hand side. Using Stirling's formula for lnn!, with n 1, ln X = ,4J=T X + N1 lnN1 + N2 lnN2 , 2X lnX ,N1 , X lnN1 , X , N2 , X lnN2 , X : 2.130 Hence the stationary point condition for the value Xmax maximizing X is N , X N , X d ln X 1 max 2 max , 4J=T 2 dX X =Xmax= ln Xmax =0: 2.131 It is easy to verify that d2 ln X =dX 2 is negative for all X . From 2.131 2 Xmax = N1 , Xm N2 , XmX,4 ; 2.132 where X is de ned by 2.75 . From 2.121 we have AN; T; M = ,NJ , T ln Xmax : 2.133 From 2.8, the magnetic eld is given by @A = 1 @A = 1 ln N1 , Xmax N2 H = @ M N @m 2 N , X N 2.134 2 max 1
56
using 2.130, 2.131 and 2.123. Hence N1 , XmaxN2 = N2 , XmaxN1 Z2 ; 2.135 Z being de ned by 2.75. The reader will be asked in Example 2.5 to show that these results are equivalent to those of Sect. 2.4.2.
2.6.2 The General Distribution

The maximum-term method can be extended to the general distribution and to a set X1 ; X2 ; : : : ; X of internal variables, of which one may be equal to or dependent on the energy. The Xk are often called order variables. We then have X X Z = Xk = Xk exp ,H Xk =T 2.136 and the conditions for the maximizing values Xkmax of Xk for k = 1; : : : ; are
fX1 ;:::;X g fX1 ;:::;X g
@Xi Xk =Xkmax= 0 ;
i = 1; : : : ; ;
0;
2.137 2.138
for any Xi 6= Ximax . This means that the left-hand side of 2.138 is negative de nite. If, for instance, the constant magnetic eld distribution is used for the Ising model, X and m can be chosen as the two order variables while the Hamiltonian in an equation of type 2.136 for Z is b H = ,NJ + 4JX , N Hm : 2.139 Since = 2, 2.137 gives two relations, which are in fact equivalent to 2.131 and 2.134. Analogously to 2.121, lnZ can be identi ed with the logarithm of the maximum term in the summation of 2.135 and put F = ,T ln Ximax : 2.140 This relation can be given a thermodynamic interpretation. Suppose a nonequilibrium thermodynamic potential is de ned for general Xi values by F Xi = ,T ln Xi : 2.141 The conditions 2.137 and 2.138 which give a maximum value of clearly give a minimum value of F Xi , as T is one of the independent variables which are all kept at xed values. Hence the equilibrium value of F , as given by 2.140, is identi ed as a minimum. This is in accordance with the
Xi , Ximax Xj , Xjmax fi;jg @Xi @Xj Xk =Xkmax
X @2
2.7 A One-Dimensional Model for DNA Denaturation
57
thermodynamic principle that deviations from the equilibrium state at xed values of T; 1 ; : : : ; ; Q+1 ; : : : ; Qn cause F to increase from its equilibrium value. The conditions for a minimum value of F are
@F
@Xi Xk =Xkmax= 0 ;
i = 1; : : : ; ;
2.142 2.143
X @ 2F
i;j
which are equivalent to 2.137 and 2.138. It is possible for two sets, Ximax and Xi0max respectively, of the order variables to satisfy conditions 2.142 and 2.143. Then the states de ned by Ximax and Xi0max are both internally stable, that is stable with respect to small variations of Xi . Now suppose that F Xi0max F Ximax . The state Xi0max is thus internally stable but not stable with respect to the state Xi0max . It is metastable. The true equilibrium or stable state is Ximax and is that given by averaging over the entire microstate space . However, with constraints applying which exclude the true equilibrium range of the Xi , the metastable state can sometimes be obtained by averaging over a subspace 0 . This can be described as `broken ergodicity' Palmer 1982, since, during a normal time span, the microstate can vary only over the subspace 0 rather than the entire space . Real substances can exist in states which are metastable with respect to ordering; for instance quenched alloys where the temperature has been lowered too rapidly for equilibrium rearrangements of the atoms to occur.3 The persistence of these metastable states is determined by considerations similar to those discussed in Sect. 1.10.
@Xi @Xj Xk =XkmaxXi , Ximax Xj , Xjmax 0 ;

In solution, the DNA molecule consists of two very long strands with chemical groups called bases at regular intervals along each strand. At low temperatures each base in one strand is connected to a base in the other strand by a link known as a hydrogen bond H-bond. Successive bonds are slightly angled to each other, leading to the famous double-helix form of DNA. However, this is not important for the present work, and we can regard the molecule as a ladder with the H-bonds as rungs. As the temperature increases, the H-bonds tend to break, a process known as denaturation, unzipping or the helix-coil transition. Experimentally the fraction of bonds broken can be obtained from ultraviolet absorbance measurements and plotted against temperature.
3
When, on the contrary, the temperature is lowered slowly enough for complete equilibrium to be maintained the alloy is said to be annealed.
58
Fig. 2.3. The ladder

= ,1 +1 +1
,1
+1
,1
model for DNA. The vertical segments represent intact and broken H-bonds.
Such a plot is known as a melting curve4 , and the melting temperature Tm is 1 de ned as that at which = 2 . Observed melting curves are quite sharp, with a considerable change in occurring in an interval of a few degrees round Tm . To develop a statistical mechanical theory we introduce a one-dimensional lattice, with each site corresponding to a rung of the ladder see the reviews by Wartell and Montroll 1972, Gotoh 1983, Wartell and Benight 1985. A variable i is associated with each site i, where i = ,1 when the corresponding H-bond is intact and i = +1 when it is broken see Fig. 2.3. A partition function is constructed for the one-dimensional lattice, taking temperature, pressure and the concentrations in the ambient solution as independent variables, though only the rst need be introduced explicitly. An Ising-type theory can then be constructed which, as Wartell and Benight 1985 remark, gives a semi-quantitative understanding of DNA melting. It will be assumed that the number of sites is large enough to regard the last site as adjacent to the rst and to use the thermodynamic limit without serious error. Suppose that the term in the partition function corresponding to all bonds intact is 0 and that each broken bond introduces a factor exp 2LT . Suppose also that it is easier for a bond to break if adjacent bonds are broken, and to allow for this cooperative e ect a factor exp ,2KT , where K 0, is introduced for each +, nearest-neighbour pair. The partition function becomes
Z = 0
X
exp
" X N
L
where is the space of the variables 1 ; : : : ; N and N +1 = 1 . Disregarding constants it can be seen that for each nearest-neighbour pair there is a factor 1 exp K i i+1 + 2 L i + i+1 2.145 in the general term in Z . Comparison with 2.89 shows that, apart from constant factors which have no e ect on the bond con guration, the partition function is identical to that for a one-dimensional Ising model ferromagnet, with5
The term `melting', though sanctioned here by usage, is employed loosely. Properly speaking, melting is a rst-order transition, not a continuous one. 5 The variables K and L are then examples of couplings, which are introduced in the context of the coupling-density formulation of thermodynamics in Volume 2, Sect. 1.4.
4
i=1
i + 1 + K
N X k=1
i i+1 , 1 ;
2.144

K
59
= J=T ; L = H=T : 2.146 The distribution of intact and broken bonds is thus the same as that of up and down spins in the Ising model. In particular, the equilibrium fraction of broken bonds 2.147 = 1 h1 + i i = 1 + 1 h i i = 1 1 + m 2 2 2 2 and then by 2.99, using the identi cations 2.146, sinh L 1 = 2 1+ 2.148 2 L + exp,4K 1=2 : sinh Since exp LT represents the e ect of a complicated chemical interaction between the polymer and the solution, it is not a simple Boltzmann factor and we treat it phenomenologically. First, note that, as Tm is de ned as the temperature where = 1 , then by 2.148, LTm = 0. We may thus expand 2 LT in a Taylor series, LT = aT , Tm + bT , Tm 2 + ; 2.149 but for simplicity, only the rst term is retained. Although the magnitude of K is important, its variation in the temperature range round Tm is not signi cant and we take K as constant. If K were zero, 2.148 would become 1 2.150 = 1 f1 + tanhLg = 2 f1 + tanh aT , Tm g ; 2 a result which could be obtained by deriving h i for a single isolated bond. For K 6= 0, 2.148 can be expressed in the form sinhfaT , Tmg = 1 1+ 2.151 2 faT , T g + exp,4K 1=2 : 2 sinh m The maximum slope of the plot of against T occurs at T = Tm where d 1 2.152 dT T =Tm = 2 a exp2K : The cooperative e ect is embodied in the parameter K , and the slope at Tm increases with K . In the limiting case K = 1, becomes a Heaviside unit function equal to zero for T Tm and to one for T Tm. However, the large experimental slope could, by 2.152, be attributed to either a large a or a large K . To obtain unequivocal evidence of the importance of the cooperative e ect and to consider certain other e ects, we have to examine the structure of the polymer in more detail. The bases which form part of the DNA molecule are not identical and there are in fact four types, labelled A, T, G and C. However the only possible types of H-bond are AT or GC. In a natural DNA molecule the bases occur in an irregular but xed order, and by the bonding rules just given the sequence of bases along one strand determines the sequence along the other strand. Hence if the strands separate, each can attract molecules from the
60
medium to reproduce its complementary strand; this is the important replication property of DNA. Studies of the dependence of melting temperature on the relative proportions of AT and GC indicate that Tm1 Tm2 , where Tm1 and Tm2 respectively denote the melting temperatures of pure AT and pure GC DNA molecules. The di erence between these temperatures is between 40 and 50 degrees Celsius, depending to some extent on the medium Wartell and Montroll 1972. Most early experiments on DNA gave smooth sigmoidal curves of against T , but this is now attributed Wartell and Benight 1985 to the heterogeneity in base-pair sequence and molecular length of the samples. With homogeneous samples, the curves are less smooth and show several peaks of d=dT instead of a single peak in the centre of the melting range Wada et al. 1980. Consider a molecule in which a fraction x of the H-bonds is AT and a fraction 1 , x is GC. For a preliminary treatment, assume that there is no cooperative e ect K = 0 so that each bond is an isolated system and their order is immaterial. Supposing, for simplicity, that the parameter a is the same for both types of bond we de ne L1 = aT , Tm1 ; L2 = aT , Tm2 : 2.153 Equation2.150 can then be applied, with L = L1 , to the fraction x of AT bonds and, with L = L2 , to the fraction 1 , x of GC bonds. This gives, for the overall fraction of broken bonds, 1 , 2 = 1 x tanhL1 + 1 , x tanhL2 : 2.154 2 Consider the case Tm1=Tm = 0:928 and Tm2=Tm = 1:092, Tm being the mean melting temperature de ned in 2.162 below. This corresponds to melting temperatures of 48C and 98C for AT and GC respectively. For a reasonably narrow melting range in the cases x = 1 and x = 0 a large value of a is 1 required so let aTm = 80. With these values and x = 2 , 2.154 gives curve i in Fig. 2.4, which consists of sigmoidal portions near T = Tm1 and T = Tm2 separated by a long nearly horizontal stretch. Since no observed melting curve is like this, the assumption of non-interacting bonds is unsatisfactory. The theory of an irregular sequence at AT and GC bonds with cooperative e ects is di cult see references above. However, the in uence of the cooperative e ect is clearly seen for a regular alternating sequence of AT and GC, and this is the case we shall discuss. For simplicity K is supposed to be the same for all nearest-neighbour pairs and we again assume 2.153. Consider a linear lattice with sites 1; 3; 5; : : : corresponding to AT bonds, sites 2; 4; 6; : : : corresponding to GC bonds. The number of sites N is taken to be even with sites N and 1 nearest neighbours. For the nearest-neighbour site pair i; i +1 there is a factor exp K i i+1 + 1 L1 i + L2 i+1 ; for i odd ; 2.155 2 exp
K
1 i i+1 + 2 L2 i + L1 i+1
;
for i even :
2.156
2.7 A One-Dimensional Model for DNA Denaturation 1:0
61
Fig. 2.4. Theoretical meltii i

0:5
ing curves: i X = 1 for non-interacting bonds; ii X = 0:0125.
0:9
1:0
T=Tm
1:1
The con gurational partition function can now be written as X Z= V V 0 V 0 V! V!0 ;
f ; ; ; ;:::; ;;!g
2.157
1 Z = TracefV V 0 2 N g = 12 N ; 2.159 0 . The last expression depends on where 1 is the largest eigenvalue of V V
where ; ; ; ; : : : ; ; ; ! represent the states of the bonds on sites 1; 2; 3; 4; : : :; N , 2; N , 1; N respectively. Here exp K + 1 1 1 L exp , 2 2.158 V = exp ,K + 1L1L+,2L1 exp K K + 2L1L+,2L2 ; ,1 L 2 2 2 the matrix of the quantities given by 2.155, with V 0 corresponding similarly to 2.156. Thus V 0 is the transpose of V . From 2.157,
1
assuming the thermodynamic limit. The fraction of broken bonds is given by 1 = 4 h1 + i ii even + h1 + i ii odd = 1 + 1 @ ln1 + @ ln1 : 2 4 @ L1 @ L2 After obtaining an expression for 1 see Example 2.6 this yields X,1 sinh 2aT , Tm , 1 = 2 2 fX,1 cosh 2aT , Tm + cm g2 , ,1 , 2 1 ; 2 where
2.160 2.161
62
X = exp,2K ; 1 2.162 Tm = 2 Tm1 + Tm2 ; cm = cosh aTm2 , Tm1 : 1 Thus = 2 for T = Tm , and , 1 is an odd function of T , Tm. At T = Tm 2 the even derivatives of are thus zero and the rst two odd derivatives are given by d dT m = aQcm; X ; 2.163 d3 3 Qc ; X 1 , 3 Qc ; X 2 1 + c X2 ; m m dT 3 m = 4a m where 1 2 Qcm; X = X,1 cm + 1, 1 2 + cm , 1X2 , 2 : 2.164 For the independent bond case when K = 0 X = 1, 2.161 reduces to 2.154 with x = 1 . At the opposite extreme, when K ! 1 X ! 0, it follows 2 1 from 2.161 that , 1 ! 2 signT , Tm, so that the limiting form of is 2 a Heaviside unit function. From 2.163, d=dT m increases monotonically with K and d=dT m ! 1 as K ! 1. When d3 =dT 3m 0, d=dT , which is an even function of T , Tm, has a minimum at T = Tm and so must have a pair of maxima symmetrically placed about T = Tm. For the values of Tm1 =Tm, Tm2=Tm and aTm assumed above, d3 =dT 3m 0 for X 0:006762 K 2:498, implying that the double maxima still exist at values of X where there is no remaining vestige of the horizontal stretch around T = Tm and where the melting range has narrowed considerably. This is illustrated by curve ii in Fig. 2.4, which is drawn for X = 0:0125 K = 2:1910. The two peaks of d=dT are at T=Tm = 1 0:019, approximately. This simple d=dT spectrum re ects the elementary periodic AT GC bond distribution which we have assumed. Although some important factors in denaturation have been identi ed, models of the type discussed above have the disadvantage that intact and broken bonds are treated in an equivalent way as states of linear lattice sites. However, where there are long sequences of broken bonds the two strands will not remain in a xed relative position and will adopt an irregular `loop' conformation. The additional entropy can be estimated by regarding the strands as two free polymers subject to the constraint that they possess common end-points. Various expressions are compared by Gotoh 1983. Under certain assumptions, the theory gives a rst-order transition see the review by Wiegel 1983 which is surprising in a system where the interacting elements are ordered one-dimensionally, but presumably results from loop entropy terms acting as long-range interactions. However, the connection with experimental results is not clear.
Examples
63
For other physical realizations of one-dimensional lattice systems, see Ziman 1979, Chap. 2.
Examples
2.1 Show that in the one-dimensional zeroeld Ising model the con gurational entropy is given by N 2J=T S = 1 + exp2J=T + N ln 1 + exp,2J=T : Verify that S=N ! ln2 as T=J ! 1 and S=N ! 0 as T=J ! 0. 2.2 In the -state Potts Model each microsystem occupies one lattice site and can exist in distinct states. A `like' nearest-neighbour pair, when both members of the pair are in same state, has interaction energy ,R, whereas an ùnlike' nearest-neighbour pair has zero interaction energy. The ferromagnetic Potts model is when R 0, which encourages nearest-neighbour pairs to be in the same state and the antiferromagnetic Potts model when R 0, which encourages nearest-neighbour pairs to be in di erent states. Assuming no interaction with any external eld, construct the transfer matrix V for a one-dimensional Potts Model. Using the method of Sect. 2.4.1, obtain the partition function for an open linear lattice of N sites and show that in the thermodynamic limit, when R 0, the con gurational internal energy is NR expR=T U = , expR=T + , 1 : Show that, apart from an additive constant term in the energy, the two-state Potts Model with = 2 is equivalent to the Ising model 1 with J = 2 R. Derive the heat capacity for general and R 0 and show that the temperature where the heat capacity has a maximum decreases with . 2.3 In the Ising model ferromagnet with H 6= 0 the internal energy U = H + MH where H is the enthalpy derived in the constant magnetic eld distribution. Given that 1 is the largest eigenvalue of the transfer matrix show that H = ,N J X @ ln1 + HZ @ lnZ 1 ; U = ,NJ X @ ln1 ; @X @ @X X and Z being de ned by 2.75. Hence prove that 4X,3 ,1 U = NJ 2 1 X , Z,1 2 + 4X,4 1
64
and show that this expression reduces to ,NJ tanhJ=T when H = 0. 2.4 In the spin-1 Ising model each dipole has three states, in which the spin variable = ,1; 0; 1 respectively. The Hamiltonian for the constant magnetic eld distribution has the same form as for the original Ising model being given by 2.66. Given H = 0, construct the transfer matrix for the one-dimensional model and determine why the successive summation method of Sect. 2.4.1 is inapplicable. Again with H = 0 show that the free energy for a ring lattice of N sites is " p 2c + 1 + 2c , 12 + 8 F = ,NT ln ; c = coshJ=T : 2 Obtain an expression for the con gurational energy U and show that U ! ,NJ as T=J ! 0 and U ! 0 as T ! 1. 2.5 In the maximum-term theory of the linear Ising model show, from 2.135, that 1 2 2 , Xmax = NNNZ2Z N 1 : 1 , 2 By substitution into 2.132 and using 2.123 deduce that ,1 m = p Z ,1Z2 : Z , Z, + 4X,4 Show that this expression for m is equivalent to 2.99. 2.6 i Let V 0 be de ned as the transpose of the matrix V given by 2.158. Show that the largest eigenvalue of the product matrix V V 0 is q 1 = + 2 , 4 sinh2K ; where = coshL1 + L2 exp2K + coshL2 , L1 exp,2K : Deduce the expression 2.161 for the fraction of broken bonds, noting that L1 and L2 are de ned by 2.153. ii Show that, when K = 0, the expression 2.161 reduces to , 1 = 1 ftanh aT , Tm1 + tanh aT , Tm2 g ; 2 2 noting that this is 2.154 for the case x = 1 . 2 iii Show that, when Tm1 = Tm2, 2.161 reduces to 2.151.
Examples
65
iv From 2.161 deduce that d 2 2 ,3 2 dT = a exp2K , 4 sinh 2K f 2 , 4 sinh2 2K coshL1 + L2 , sinh2 L1 + L2 exp2K g : 1 Derive 2.163 for T = Tm = 2 Tm1 + Tm2 .
3. The Mean-Field Approximation, Scaling and Critical Exponents
3.1 The Ising Model Ferromagnet

We consider the Ising model on a regular lattice see Appendix A.1 where each interior site has the same number of nearest-neighbour sites. This is called the coordination number of the lattice and will be denoted by z. We shall denote the dimension of the lattice by the symbol d. It is assumed that, in the thermodynamic limit, boundary sites can be disregarded and that, with N sites, the number of nearest-neighbour site pairs is 1 zN . Since 2 the Ising model is restricted in the sense of Sect. 2.3.4, the Helmholtz free energy A depends on the two extensive variables M and N , the latter being equal to the number of spins and also to the reduced volume V if the volume v0 per site is taken as the standard volume see discussion after 2.55. The con gurational Helmholtz free energy per site a depends on T and the relative magnetization per site m, de ned by 2.70, so that n = 2. The value of m is prescribed for the canonical distribution, and we now consider spin con gurations for given m. Since all sites are regarded as equivalent, 2.64 implies m = h ii ; 3.1 where the right-hand side is the mean of the spin variable on site i in the canonical distribution. Except at the critical point itself, the spins on widely separated sites i and j are independent and thus h i j i = h i ih j i = m2 : 3.2 However, with i and j close together, the situation is di erent. For the Ising ferromagnet the spins tend to lie parallel and thus h i j i m2 : 3.3 This represents short-range order, and the di erence h i j i , m2 , which is a function of the separation of sites i and j , is a measure of the correlation between the spins see Ziman 1979, Chap. 1. This quantity reduces to h i j i when m = 0, a result which holds for d = 1, H = 0 at all temperatures. The correlation h i i+k i for k-th neighbour sites in this case was evaluated in Sect. 2.4. From 2.88, h i i+k i 0 for all k, but tends to zero as k ! 1.
68
Now consider the mean ,J h i j i of the interaction energy of a nearestneighbour i; j pair. The contribution from a particular value of i is ,J i h j i i where h j i i denotes the mean of j for given i and m. Thus spin i can be regarded as acted on by a eld ,J h j i i due to its neighbouring spin j . The meaneld approximation consists in replacing h j i i by m, which by 3.1 is the mean value of j over all spin con gurations compatible with the prescribed m, irrespective of the value of i . Summing over i and using 3.1 again gives ,Jm2 for ,J h i j i. This assumes that 3.2 applies even for nearest-neighbour pairs, and is equivalent to putting the correlation equal to zero, thereby disregarding short-range order. The energy E is the same for all terms in the canonical partition function and thus 3.4 E = , 1 zNJm2 = U : 2 Equations 2.122 and 2.123 apply for a lattice of any dimension and hence the number of spin con gurations compatible with a given value of m is ! N! m = N NN ! = 1 : 3.5 1 1! 2 2 N 1 + m ! 2 N 1 , m ! The canonical partition function in the meaneld approximation is thus Z = m exp,E=T : 3.6 Since 3.4 and 3.5 correspond to a random mixture of N1 positive and N2 negative spins on the lattice, the meaneld method can be termed the randomized approximation. Another term in frequent use is the zeroth-order approximation.
3.1.1 Free Energy and Magnetization

Using 3.4 3.6 and Stirling's formula the Helmholtz free energy density is given by a = N ,1T ln Z 1 1 = 2 T 1 + m ln1 + m + 1 , m ln1 , m , 2 ln 2 , 2 zJm2 : 3.7 It can be seen that a is an even function of m, thus satisfying the symmetry relation 2.65. The second derivative of a with respect to m is and we de ne the critical temperature Tc by Tc = zJ : 3.9 From 3.8 the curve of a against m is convex downwards for all m in the relevant range ,1; 1 when T Tc, as shown by the broken line in Fig. 3.1. At T = Tc the curvature at m = 0 changes sign and for T Tc the form of a
@ 2 a = T , zJ @m2 1 , m2
3.8
69
a + T ln2
A D
,1
,ms
B
ms
C
+1
T Tc .
Fig. 3.1. Free energy magnetization isotherms: broken curve, T Tc ; full curve,
is shown by the full line ABOCD in Fig. 3.1. Since n = 2, a is a free energy density of type fn,2 as discussed in Sect. 1.10 iii, and the points B and C where the curve of a is touched by the double tangent BC represent conjugate phases. The loop BOC, lying above BC, is typical of classical theories and represents unstable states, the stable equilibrium a being given by the convex envelope ABCD. The situation is similar to that shown on Fig. 1.2 for the van der Waals gas. However, because of the symmetry of the a curves about m = 0 the double tangent BC is horizontal and the magnetizations corresponding to the points B and C are equal and opposite. They are denoted by ms ms 0, where ms is a function of T Tc . The eld H is conjugate to the density m, and either from the second relation of 1.34, or from the more general relation 1.78, with = 0 it follows that @a H = @m = 1 T ln 11+ m , zJm : 3.10 2 ,m The last expression derives from 3.7 and, since it is an odd function of m, the symmetry relation 2.71 is satis ed. From the broken curve in Fig. 3.1 the slope @a=@m is a continuous function of m for T Tc. Hence the corresponding plot of m against H is continuous, as shown by the broken line in Fig. 3.2. The situation is quite di erent for T Tc. The slope @a=@m and hence H are zero at the end-points B and C of the common tangent Fig. 3.1. Hence the magnetizations ms correspond to H = 0; ms is thus termed the spontaneous magnetization. The slopes of a against m on the convex envelope approach zero continuously as the ends B and C of the common tangent are approached from the left to right respectively. The form of the plot of m against H for T Tc is thus given by the full line in Fig. 3.2 and
70
3. The Mean-Field Approximation, Scaling and Critical Exponents +1
m ms
H
,ms ,1
Fig. 3.2. Magnetization eld isotherms, as in Fig. 3.1.

There is a discontinuity in the density m as H passes through the value zero at constant T Tc and this constitutes a rst-order transition. Since the internal energy U is an even function of m it remains continuous as H passes through zero. It has been shown that m is a smoothly varying function of H and T except for the discontinuity 3.11. If states of the system are represented by points in the T; H plane then there is a segment of discontinuity rstorder transition line occupying the interval 0; Tc on the T -axis, as shown in Fig. 3.3, and all thermodynamic functions are smooth except on this interval. Although it has been derived by the rather crude meaneld approximation,
H!0
lim m = ms :
3.11
H
0 0
Tc
Fig. 3.3. Phase diagram in the

temperature eld plane.
71
this simple form of phase diagram is believed to be qualitatively correct for the Ising model ferromagnet with d = 2 or 3. The symmetry about the T axis is of course a consequence of the H-reversal symmetry expressed by the relations in Sect. 2.4. For both theoretical and experimental ferromagnets the critical temperature Tc is called the Curie temperature.
3.1.2 Fluctuations in Zero Field

The zeroeld magnetization is obtained by putting H = 0 and m = ms in 3.10 to give 1 ln 1 + ms , zJ m = m 1 , Tc + 1 m2 + 1 m4 + = 0 : 3.12 s s 2 1 , ms T T 3 s 5 s For T Tc the only solution is ms = 0. For T Tc, ms = 0 is still a solution of 3.12 but now corresponds to a non-equilibrium state. The equilibrium value of ms is given by 1 m2 + 1 m4 + = Tc , T : 3.13 3 s 5 s T It follows that ms ! 0 as T ! Tc , 0 so that the magnetization is continuous at T = Tc. As T ! 0 the series on the left-hand side must diverge and so ms ! 1 see Fig. 3.4. From 3.4 the zeroeld con gurational internal energy per spin is given by 80 ; T Tc , 3.14 u=: 1 2 , 2 zJms ; T Tc . The result for T Tc occurs because short-range ordering is neglected in the meaneld approximation. The con gurational heat capacity per spin, when H = 0 and T Tc, is given by
m
0
m = ms Tc m = ,ms T
Fig. 3.4. The spontaneous mag-
netization plotted against temperature.
72
c0 = kb @T
@u
H=0
2 = , 1 zkb J @ ms 2 @T
where the last form follows from 3.13. From 3.15 and 3.14 lim c = 3 k ; T !Tc ,0 0 2 b c0 = 0 ; T Tc :
= 1 kb Tc 2 T
2 h 1
2 2 3 + 5 ms +
i,1
3.15
3.16
Hence there is a nite discontinuity in c0 at T = Tc, as shown in Fig. 3.5. As described in Sect. 1.9 and illustrated by 3.11, a rst-order transition occurs where there is discontinuous change in one or more of the thermodynamic densities. We have just seen that in meaneld theory the Ising model ferromagnet undergoes a transition from a disordered i.e. unmagnetized state to a long-range ordered i.e. spontaneously magnetized state when T decreases through the value Tc with H = 0. Although a change of symmetry has thus occurred, the energy, entropy and magnetization densities are continuous at T = Tc so that the transition cannot be rst-order. In Ehrenfest's classi cation Ehrenfest 1933, Pippard 1967 a pth-order transition, for p 1, is one in which there are nite discontinuities in the p , 1th-order derivatives of the densities with respect to elds, but where all lower order derivatives and the densities themselves are continuous. In this scheme the Ising model transition at T = Tc is second-order since, by 3.15 and 3.16, there is a discontinuity in @u=@T . However, in an exact theory of the d = 2 Ising model see Sect. 8.9, the nite discontinuity is replaced by a singularity where @u=@T ! 1 as the critical temperature is approached from either above or below. Even in meaneld theory there is a symmetrical in nity at
1:5
c0 kb
T=Tc
Fig. 3.5. The zeroeld heat capacity from meaneld theory.
73
and Example 3.3. As an alternative to the Ehrenfest scheme, Fisher 1967 proposed a classi cation of transitions into two categories, ` rst-order' and `continuous' i.e. with continuous densities. The term `higher-order' may be used instead of `continuous'. However, in our view there is no ambiguity in describing transitions where the densities are continuous, but where there are discontinuities or in nities in some of their rst derivatives with respect to elds, as `second-order', and we shall use this term where appropriate. We now consider the zeroeld susceptibility for T Tc . Using 3.9 and a power series expansion of the logarithmic term, 3.10 can be expressed in the form T 1 c + m2 + 1 m4 + , H = 0 ; 3.17 m 1, T 3 5 T from which it follows that m ! 0 as H ! 0 for T Tc . Hence @m = lim m = 1 ; 3.18 which diverges as T ! Tc + 0. To discuss uctuations in m it is necessary to use the constant magnetic eld distribution. From 2.46, 2.65 and 3.4 b 1 ^ H = , 2 zNJ m2 , N Hm ; ^ 3.19 where the `hat' notation for m indicates that the magnetization is no longer ^ an independent variable and must now be summed over in the construction of the partition function. Thus X Z N; T; H = expN Hm=T m ; ^ ^ fmg
^ 3.20 1 2 =T = exp ,Nam=T ; m = m exp 2 zNJ m ^ ^ ^ ^ where m is given by 3.5 and a by 3.7. Since is identical in form to ^ the canonical partition function, 3.20 is an example of 2.31. If the method of the maximum term is used to obtain the equilibrium value m then, from 3.20, the equation 3.10 is recovered. Since m = 0 for T Tc we have X 2 m m ^ ^ fmg ^ Var m = X ^ : 3.21 m ^ Only values of m close to the equilibrium value zero will contribute signi ^ cantly to the sums in 3.21, and hence, using 3.7,
fmg ^
T = Tc in another rst derivative of density, as is shown by 3.18 below,
@H
H=0
H!0
H T , Tc
, ^ ln m = ,NT ,1am = N ln2 , N T 2TTc m2 : ^ ^
3.22
74
Since am is an even function of m, only even powers appear in its Maclaurin ^ ^ expansion which we have truncated after the squared term. The term N ln2 can be omitted since it cancels in 3.21. The sums in this equation can be replaced by integrals since, by 2.123, the interval between successive values of m is only 2=N . Again, as only the neighbourhood of m = 0 is signi cant, ^ ^ the limits 1 for m can be replaced by 1 in the integrals. Finally, after ^ making the substitution r , x = m N T2T Tc ^ 3.23 we have = N T T T : Var m = N T , T Z,1 ^ 1 , c c , exp,x2 dx 2T
,1
Z1
x2 exp,x2 dx
3.24
Now the uctuation susceptibility equation 2.115 applies for any d, provided the equilibrium value of M is zero. Using 2.70, it can be expressed in the form
@m
@H
and, combining with 3.24, we recover 3.18. It is important to consider the microscopic behaviour of the Ising model just above a critical point where the susceptibility diverges. By 2.101 the exact zeroeld susceptibility of the d = 1 Ising model ferromagnet tends to in nity as T ! 0. From Sect. 2.5, this results from the k-th neighbour correlation tanhJ=T k tending to unity as T ! 0. The susceptibility divergence is thus related to the divergence in the size of the region of local order, that is the length of the blocks of parallel spins. An exact treatment would reveal similar behaviour for d = 2 or 3 just above the critical temperature Tc . Although the overall numbers of up and down spins remain equal there are large regions of local order or `patches' of parallel spins. As T ! Tc + 0 the patches become in nitely large which means that a quantity called the correlation length is approaching in nity. This is analogous to critical opalescence in a uid, referred to in Chap. 1, where the local uctuations i.e. regions where the density di ers appreciably from the mean become large just above the critical point. In the meaneld method, only overall uctuations, which change the relative magnetization m from its equilibrium value, ^ are recognized, although 3.24 shows that these are su cient to produce a susceptibility divergence as T ! Tc + 0. In an accurate treatment, which must include the e ect of local uctuations, the susceptibility divergence is sharper, especially for d = 2 see the discussion of critical indices in Sect. 3.5 and the exact solution for d = 2 in Chap. 8.
H=0
= NT ,1Var m ^
3.25
75
For T Tc there is what at rst sight seems to be a paradox in relation to the properties of the zeroeld Ising model ferromagnet for d 1. b With H = 0, H as given by 2.66 is equal to E , which is invariant under a change of sign of the spin variables. Hence h i i = ,h i i = 0 for all sites i, which by 2.64 implies that the mean hmi = 0 in the constant mag^ netic eld distribution with H = 0. The same result follows from meaneld theory since m, as given by 3.20, is an even function of m and hence ^ P m m = 0.^ Since this is true for all T it appears to be inconsistent with ^ ^ m ^ long-range order i.e. a spontaneous magnetization ms 6= 0 for T Tc and H = 0. However, although hmi = 0, there is no concentration of probability ^ density near m = 0 in either accurate or meaneld theory. There is instead a ^ symmetrically bifurcated probability distribution with density concentrated near m = ms and m = ,ms. In meaneld theory, m is given by the last ^ ^ ^ relation of 3.20 and from the curve ABOCD in Fig. 3.1 there are equal maxima of m at m = ms with a minimum at m = 0, for T Tc. For N of ^ ^ ^ the order of Avogadro's number Ma the maxima are very sharp. However, for any non-zero H, however small, one of the two probability maxima becomes dominant. Suppose that, with T Tc, the state m = ms 0 is reached by letting H ! +0. Then, with any realistic assumptions about the dynamics of the system, the waiting time for a ` ip' to the state m = ,ms 0 will be very large compared to normal observation times. The same will apply when the initial state is m = ,ms 0. This situation for T Tc is similar to that in the metastable states discussed near the end of Sect. 2.6, but the microstate now varies over one of the two equal subspaces into which the space of microstates is divided, corresponding respectively to positive and negative magnetization. For either possibility the spatial symmetry is reduced as compared with the T Tc state m = 0, since the system is no longer invariant with respect to rotations or re ections which reverse the magnetization axis. As Palmer 1982 remarks: `Broken symmetry is a particular case of ... broken ergodicity'. The large-scale properties of the Ising model ferromagnet are expressed quantitatively by critical parameters such as Tc=zJ , which are discussed in Sect. 8.9. Behaviour very near the critical point is characterized by critical exponents; see Sect. 3.4. The meaneld method ignores the correlation between spins on nearest-neighbour sites. It might be expected that this correlation would weaken as the coordination number z becomes larger, and it is found that accurate critical parameters do become closer to meaneld values as z increases. However, the dimension d of the lattice is more important. For d = 1 the meaneld method gives Tc = zJ = 2J , but this is spurious since, from the exact results of Chap. 2 there is no non-zero critical temperature. For d = 2 and d = 3, accurate methods do predict a non-zero critical temperature, but the accurate parameters for d = 3 are considerably closer to meaneld values than those for d = 2 see Chap. 8, Table 8.1, and Domb 1974, Table 13.
76
3.2 Interpretations of the Mean-Field Method
The meaneld method was developed independently by a number of authors. As will be seen below, van der Waals' treatment of attractive forces was meaneld in character. The Weiss approach Fowler and Guggenheim 1949, initiated before the development of the quantum theory of magnetism, consisted in adding a term m to the eld H acting on the dipoles. With = zJ this is equivalent to the meaneld method, Example 3.1. There are two limiting cases in which the meaneld results can be considered accurate, both involving long-distance interactions. The rest of this section will be devoted to these.
3.2.1 Many-Neighbour Interactions and the Lebowitz Penrose Theorem

The Ising interaction can be extended to n-th neighbour distance by generalizing 2.63 to
k n X X
E=,
where the summation labelled k is over all k-th neighbour pairs of sites, each pair being counted only once. In the meaneld approximation all pairs i j are replaced by m2 so that 1 E = ,2N
k=1
Jk
fi;jg
i j;
3.26
X n
k=1
zk Jk m2 ;
3.27
Sect. 3.1 are thus still applicable, the only change being the replacement of zJ by Pn=1 zk Jk . Suppose that now Jk is set equal to J for all k = 1; : : : ; n. k Since the correlation between the two spins of a pair will diminish as the distance between them increases it might be expected that accurate critical parameters will become closer to meaneld values as n increases, and this is found to be the case Wood 1975, Table 14. However, the weight of numerical evidence is that the critical exponents for small nite n are the same as those for the nearest neighbour model n = 1. On the other hand, it is possible that meaneld theory may apply for interactions of very long range. Consider, for instance, a uid where the range is so large that any density uctuations may be neglected. The energy of interaction of the given molecule with other molecules is then proportional to the overall density and can be written as ,2KM=V where V is the reduced volume de ned in 1.54. Adding over all molecules and dividing by two to avoid double counting, the long-range interactions contribute ,K M 2 =V to the internal energy U and the Helmholtz free energy A. By 1.9 or 1.65,
zk denoting the number of kth neighbours of any site. All results derived in
3.2 Interpretations of the Mean-Field Method
77
this gives a term ,K 2 in the pressure eld P and ,2K in the chemical potential , where = M=V . The pressure contribution is identical in form to the attractive pressure term in van der Waals theory, described in Sect. 1.6. The contribution to the free energy density a, taken with respect to V , is also ,K 2 . Similar reasoning for a lattice magnetic model gives a term proportional to ,m2 in the energy per site, as in meaneld theory. This intuitive idea about a long-range interaction was made precise by Lebowitz and Penrose 1966 and extended to the quantum mechanical case by Lieb 1966 see also Hemmer and Lebowitz 1976. Take a uid model of dimension d, called the `reference model', with free energy a0 T; per unit volume and introduce an additional interaction energy wr between any two molecules at distance r where wr = du r ; 3.28 ,d at large x. ux being a monotonic function of x, decreasing faster than x For xed r, wr can thus be made as small as we like by increasing , while on the other hand the space integral of wr, which is proportional to
Z1
0
d u rrd,1 dr ;
3.29
is independent of . Lebowitz and Penrose showed that, as ! 1, the limiting form of the free energy per unit volume is the convex envelope of aT; = a0 T; , K 2 ; 3.30 where K is proportional to the space integral. Thompson 1972b has discussed the application of this result to lattice models. The meaneld Ising model theory can thus be regarded as a reference model of independent lattice spins with a Lebowitz Penrose Ising-type limiting interaction added.
3.2.2 A Distance-Independent Interaction

1 Consider a spin- 2 model on N lattice sites with an equal Ising-type interaction between each pair of spins, irrespective of distance. This distanceindependent model, which was rst proposed by Husimi 1953 and Temperley 1954 and further discussed by Kac 1968 and Brankov and Zagrebnov 1983, can easily be shown to give a meaneld form of the congurational energy and the free energy. Extending the summation in 2.63 to every pair of spins and using i2 = 1
E = ,"
allX pairs
X 2 1 = ,1" i + 2 "N: 2
fig
fi6=jg
1 i j = ,2"
X X
fig
i
fj6=ig
3.31
78
For given relative magnetization m this gives 1 3.32 E = , 2 "N 2m2 + 1 "N : 2 Given that, with constant ", E=N would diverge in the thermodynamic limit N ! 1, we choose " = zJ=N to obtain 1 1 E = , 2 zNJm2 + 2 zJ : 3.33 Apart from the last term, which can be disregarded, 3.33 is identical to 3.4.
3.3 The Mean-Field Method for a More General Model

Suppose that the lattice of N sites is occupied by members of species 1; : : : ; , with one member to each site. One species may be `holes' or vacant sites. The number of members of species j on the lattice is denoted by Nj , j = 1; : : : ; and the fraction of sites occupied by species j is xj = Nj =N where
X j =1
Nj = N ;
X j =1
xj = 1 :
3.34
Each j; ` nearest neighbour pair makes a contribution "j` to the con gurational energy, which may therefore be written as X E = "j` Nj` ; 3.35 where Nj` is the number of j; ` nearest neighbour pairs. Now a pair of widely separated sites can be regarded as uncorrelated, and hence the probability of both sites being occupied by species j is x2 , whereas the probability of one j being occupied by a j and the other by an `,for j 6= `, is 2xj x` . The meaneld approximation results from the assumption that these probabilities apply for all site pairs, including nearest neighbours. Hence X 1 X E = 2 zN "jj x2 + zN "j` xj x` 3.36 j since, disregarding boundary e ects, the total number of nearest neighbour site pairs is 1 zN . The Ising model ferromagnet is an example of a model 2 with = 2, species 1 and 2 being up and down spins respectively. Then "11 = "22 = ,J , "12 = J and, from 3.36, 1 3.37 E = , 2 zNJ x2 + x2 + zNJx1 x2 = , 1 zNJ x1 , x2 2 ; 1 2 2 which reproduces 3.4, since, from 2.122, m = x1 , x2 . The number of con gurations for given values of the Nj is
fjg fj `g fj;`g
3.4 Critical Points and Critical Exponents
79
N Nj = Y !
j =1
Nj !
3.38
and hence the corresponding contribution to the canonical partition function is Nj = Nj exp,E=T ; 3.39 with E given by 3.36. If all the Nj are thermodynamic extensive variables or determined by them, then itself is the canonical partition function, and the Helmholtz free energy per site is X X X a = ,N ,1T ln = T xj ln xj + 1 z "jj x2 + z "j` xj x` ; 3.40 j 2 j=1 j =1 where Stirling's formula has been applied to ln Nj . This is applicable to the Ising model where N1 and N2 are determined by M and N so that 3.7 is a particular case of 3.40. If not all the Nj are determined by thermodynamic extensive variables then the canonical partition function is a sum over the Nj . However, the expression 3.40 for a may be regarded as a non-equilibrium free energy as de ned in 2.141, while some of the xj are then order variables with respect to which a must be minimized so as to obtain equilibrium conditions. As discussed in Sect. 2.6, this is equivalent to applying a maximum-term procedure to the partition function. It is easy to construct a non-equilibrium free energy for an alternative distribution. For instance, if some or all of the species are magnetic spins in various orientational states then a constant magnetic eld free energy per site is f = a , Hm ; 3.41 where m may be expressed in terms of the xj see Example 3.4 and Chap. 4.
fj `g

In this section we return to the subject of critical phenomena in general. This is a continuation of the Sect. 1.9, but we now have the advantage of having studied in detail a magnetic as well as a uid system. It is supposed that n = 2 in the sense of Sect. 1.7 so that there are three extensive variables S , Q1 and Q2 which are conjugate to T and the elds 1 and 2 respectively. The eld-density formulation of Sect. 1.8 will be used with densities s = S=Q2 and = Q1=Q2 . For convenience we put 1 = ; 2 = . Then, from 1.77 and 1.78 with = 0, the Helmholtz free energy density a ; T is given by a= + ; 3.42 da = ,sdT + d : 3.43
80
Fig. 3.6. Transition curve in the

c ; Tc temperature eld plane.
First-order transition
In the most common type of critical behaviour, a line of rst-order phase transitions in the ; T plane terminates at a critical point c ; Tc. The situation is shown in Fig. 3.6. The corresponding picture in the ; T plane is shown in Fig. 3.7. The liquid vapour equilibrium line in the T; P plane of Fig. 1.4 and the interval of discontinuity 0; Tc on the T -axis of the T; H plane of Fig. 3.3 are examples of rst-order transition lines terminating at a critical point. The coexistence curve in Fig. 1.3 and the spontaneous magnetization curves in Fig. 3.4 are of the type shown in Fig. 3.7.
T
c ; Tc
I = 1 T
II
Phase boundary
= 2 T
Fig. 3.7. Phase boundary dividing the coexistence region from the single-phase
region in the temperature density plane.
81
The ; T plane in Fig. 3.7 is divided into two regions, I and II, separated by a phase boundary which can be expressed either in terms of the single equation T = T 3.44 or, in terms of the two equations, = 1 T and = 2 T ; 3.45 with 1 T 2 T , for T Tc, and 1 Tc = 2 Tc = c. Region II is a region of coexistence between two phases. We can de ne the diameter d T of the coexistence region by 1 3.46 d T = 2 2 T + 1 T and the width w T of the coexistence region by 3.47 w T = 2 T , 1 T : In order to bring some precision to this rather informal discussion of critical behaviour we need to make a number of assumptions. These are based on Gri ths 1967a. Although they are not universally valid, they apply to a wide class of systems. i a ; T , s ; T and the eld ; T are continuous functions of and T in a neighbourhood around the critical point c ; Tc. They are regular in that neighbourhood except possibly on the phase boundary. ii The phase boundary T = T is a convex upward function of which is regular except possibly at the critical point where it attains its unique maximum. iii The diameter and width of the coexistence region are regular functions of T except possibly at the critical point where w T is more singular than d T . iv For xed T , a ; T is a convex downward function of , linear in region II. v For xed , a ; T is a convex upward function of T . vi The curve = T = 1 T ; T = 2 T ; T ; T Tc 3.48 in the ; T plane is a regular function of T , even at the critical point, which can be analytically continued into the region T Tc . In the simple lattice gas transcription of the Ising model described in Sects. 5.3 and 5.4 the coexistence diameter coincides with the critical isochore = c . This is a special case of the rectilinear diameter law which
82
states that d T is an asymptotically linear function of T in a neighbourhood of the critical point. It is now known that this law is not generally true. There are a number of models Hemmer and Stell 1970, Widom and Rowlinson 1970, Mermin 1971a, 1971b, Wheeler 1977 in which the coexistence diameter is singular at the critical point. One type of such model is treated in Sect. 9.12. In fact, it may be shown Mermin and Rehr 1971, on the basis of some reasonable assumptions, that, unless the model has some special symmetry, as is the case for the Ising model, d d =dT is at least as singular as the constant-volume heat capacity c . Assumption iii would appear to be valid for a wide class of systems, and since 1 1 T = d T , 2 w T ; 3.49 1 T ; 2 T = d T + 2 w it follows from iii that each branch of the phase boundary and functions calculated from them have the same singular behaviour as T ! Tc , 0. From assumption iv it follows that ; T is a monotonic non-decreasing function of along an isotherm. In region II, has a constant value along an isotherm given, from vi, by = T . The response function @ =@ T is thus nonnegative in region I and in nite in region II but not on the phase boundary, where it is de ned by its limiting value when approaching from region I. From assumption v, s ; T is a monotonically non-decreasing function of T along an isochore, and the heat capacity c = T @s=@T is non-negative. We now choose the scaled variables
and expand the free energy density in the form a ; T = a c ; Tc + t + Tcg; t : From this it follows that g0; 0 = 0 and
; t = @g = 1 ; t , t
t= T ,1 ; c = , c
3.50 3.51 3.52
is zero throughout region II and on the phase boundary. Since we have assumed in vi that is a regular function of T , and hence of t, it follows that any non-regular behaviour of a ; T is contained in g; t. In particular the non-regular behaviour of the entropy s is contained in the function ; t = , @g = s + T ddt : 3.53 @t
c
Tc
83
The modern theory of critical phenomena focuses to a substantial extent on a set of indices called critical exponents which describe the behaviour near the critical point see, for example, Stanley 1987 or Volume 2, Chap. 2. There is now a set of agreed symbols for most of these exponents, and for the system under discussion they are de ned in the following way:1 a On the phase boundary, as t ! ,0, 3.54 w ' B1 ,t : In view of assumption iii and equations 3.49, an alternative expression, valid on both branches of the phase boundary, is jj ' B jtj : 3.55 b Along the critical isotherm t = 0, as jj ! 0, , c = ; 0 , 0 = Tc
; 0 ' Cjj ,1 : 3.56 c The limiting properties of the compressibility of a uid or the susceptibility of a magnet are those of
T @ t On the critical isochore = 0 = c , as t ! +0, 'T ' Djtj, and on the phase boundary, as t ! ,0, 8D jtj, 0 along = T , 1 1 'T ' : D2 jtj, 0 along = 2 T .
T
'T = T @ @
Tc @
,1 = :
3.57 3.58 3.59
d For the heat capacity at constant density On the critical isochore = 0, as t ! +0, and on the phase boundary, as t ! ,0, 8K jtj, 0 along = T , 1 1
@s c = kb T @T
' kb T @ : @t

3.60 3.61 3.62
c ' K jtj, c ':
K2jtj,
along = 2 T .
The asymptotic relation symbols ' and are de ned as follows: f x ' gx as x ! x0 means that f x=gx ! 1 as x ! x0 , f x gx as x ! x0 means that f x=gx ! some constant 6= 0 as x ! x0 .
84
From equation 3.61 lim jtj c = K ; jtj!0
3.63
where K is nite and non-zero. Thus if c approaches a nite limit as T ! Tc + 0 the critical exponent = 0. A nite discontinuity in c as the path in the ; T plane passes through the point c ; Tc onto either the = 1 , or the = 2 branch thus implies = 0 = 0. Now suppose that c behaves as ln jtj, as in the exact theory of the d = 2 Ising model see Chap. 8. Then, since jtjn ln jtj ! 0 as jtj ! 0 for any n 0, it follows again that = 0 = 0. It thus appears that zero values for or any other critical exponent can correspond to quite di erent types of singularity. A more detailed theory of critical exponents and scaling is needed to distinguish these cases and this is presented in Volume 2, Chap. 2.
3.5 Scaling and Exponent Relations

Suppose that in a neighbourhood of the critical point in region I we have the asymptotic form
'
; t 3.64 for the dependence of
on and t. It can be seen that the asymptotic forms 3.55 and 3.56 are preserved under the transformation 3.65
3.66 So we now suppose that 3.64 can also be transformed in the same way. This means that
; t '
; 1= t : 3.67 ,1 we now have Setting = jj where +1 and ,1 correspond respectively to those parts of the neighbourhood in which 0 and 0. Given that is a regular monotonic non-decreasing function of along an isotherm in region I it follows that
! ; t ! yt t ;
! y

; for all 0, if yt = 1= ;
y
= :
t
' jj
1; jj1= t
1; jj1=
3.68
t =
1
jj1=
3.69
3.5 Scaling and Exponent Relations
85
and we have
' jj ,1
1
Comparing 3.56 and 3.70 we see that C =

1 0. Since
= 0 on the phase boundary, if
1 x has the root x0 , then the amplitude B of 3.55 is given by B = ,x0 . We now di erentiate 3.67 with respect to to give 1, 'T 'T ; 1= t : 3.71 , we have Setting = jtj where +1 and ,1 now correspond respectively to t 0 and t 0. On the critical isochore for t 0 'T ' jtj 1, 'T 0; 1 : 3.73 On the phase boundary, for 0, 'T ' jtj 1, 'T ,B; ,1 3.74 and on the phase boundary, for 0, 'T ' jtj 1, 'T B; ,1 : 3.75 Comparing equations 3.73 3.75 with 3.58 3.59 we see that = 0 3.76 and D = 'T 0; 1, D1 = 'T ,B; ,1 and D2 = 'T B; ,1. We further see that = , 1 ; 3.77 which is the Widom scaling law, Widom 1964. To obtain the heat capacity exponents, de ned in 3.61 3.62 we need the asymptotic scaling relationship for the free energy. From 3.52
t jj1= :
3.70
'T jtj 1, 'T jt ; 1 ; j
3.72
g; t =
; td + wt :
3.78
We shall assume that wt is a regular function of t in a neighbourhood of the critical point. Then, from 3.67, for the singular part gsing of g, obtained by subtracting o w, we have +1 gsing ' gsing ; 1= t : 3.79 Di erentiating twice with respect to t we have +1,2= c c ; 1= t: 3.80 , we have Setting = jtj
86
c ' jtj
+1,2 c
where +1 and ,1 again correspond to t 0 and t 0 respectively. On the critical isochore, for t 0, c ' jtj +1,2 c 0; 1 ; 3.82 on the phase boundary, for 0, c ' jtj +1,2 c ,B; ,1 3.83 and on the phase boundary, for 0, c ' jtj +1,2 c B; ,1 : 3.84 Comparing equations 3.82 3.84 with 3.61 3.62 we see that = 0 3.85 and E = c 0; 1, E1 = c ,B; ,1 and E2 = c B; ,1. We further see that = 2 , + 1 : 3.86 From 3.77 and 3.86 we have +2 + = 2; 3.87 which is the Essam Fisher scaling law Essam and Fisher 1963. There is one di culty arising in connection with the Ising model ferromagnet which can now be resolved. Here = m so the constant-density heat capacity is cm . However, the heat capacity usually obtained is c0 i.e. c for = H = 0. For T Tc the `critical isochore' m = 0 corresponds to H = 0 and hence c0 = cm . For T Tc, it follows from the result in Example 1.3 that
jtj ; 1 ;
3.81
@m 2 @m ,1 c0 , cm = kb T ;
@T
H=0
where c0 and cm are `per molecule' heat capacities. However H = 0 corresponds to the phase boundary where @m=@T varies as jtj ,1 by 3.55 and @m=@ H as jtj, by 3.59 and 3.76. Hence c0 , cm varies as jtj2 + ,1 = jtj, by the Essam Fisher relation 3.87. It follows that c0 , like cm , varies as jtj, . It should be emphasized that, in general, the exponent of c is not equal to . In fact see Volume 2, Chap. 2, it is equal to , de ned by 3.58 unless certain symmetry relations are satis ed. If they are, then the exponents for c and c are equal, as has just been shown for the Ising model.
@H
3.88
3.6 Classical Critical Exponents

We now apply our analysis to the two classical models of a phase transition considered in Sects. 3.1 and 1.6 respectively. We shall see that these produce the same classical or meaneld values for the critical exponents.
3.6 Classical Critical Exponents
87
3.6.1 The Ising Model Ferromagnet: Mean-Field Approximation Here the treatment is simpli ed by the m ! ,m, H ! ,H symmetry expressed in the relations of Sect. 2.4 and illustrated by Figs. 3.1 3.4. For the Ising model Q1 = M and Q2 = N so that, in the notation introduced at the beginning of Sect. 3.4, = m and its conjugate eld = H. From 3.43, = aT; m , mH = f ; 3.89 where, provided that m has its equilibrium value for the given H, f is the
constant magnetic eld free energy density. In Sect. 3.1 the phase transition conditions were deduced geometrically and it was found that the transition line is the interval 0; Tc of the H = 0 axis so that the critical point eld Hc = 0. The coexistence densities are 1 = ms and 2 = ,ms, and the critical density mc = 0. Since aT; ,ms = aT; ms and H = 0 on the transition line it follows from 3.89 that f is continuous across the transition line. Hence the analytic conditions 1.90 for phase equilibrium are satis ed. Since Hc = 0 and mc = 0, the scaled variables and
, de ned by 3.50 and 3.52, are given by = m and
= H=Tc. The critical exponents introduced in 3.55 3.62 can be obtained from relations given in Sect. 3.1. From 3.17 1
' tm + 3 m3 : 3.90 As this is the asymptotic form corresponding to 3.64 we apply the transformation
m ! m ; t ! 1= t ;
!
3.91
to give
1 3, m3 : 3 Since this must be equivalent to 3.90 for all 0 it follows that 1 =2;
' 1,
+ ,1 tm +
3.92 3.93
=3: 3.94 Alternatively, these values for and follow from 3.13 and 3.17 respectively with T = Tc. The relation 3.18 gives =1 3.95 and hence the Widom scaling law 3.77 is obeyed. Since ms ! 0 as T ! Tc , 0, it follows from the expression given in Example 3.3 for @m=@ HH=0 where T Tc that 0 = 1. Hence the exponent relation 3.76 is satis ed.
88
From the arguments given at the end of Sect. 3.5, the zeroeld heat capacity c0 may be used instead of cm on both sides of the transition point. Since c0 remains nite in fact zero as T ! Tc + 0, = 0; 3.96 which, with 3.93 and 3.95, satis es the Essam Fisher scaling law 3.87. Since, by 3.16, c0 also remains nite as T ! Tc , 0, 0 = 0, satisfying the exponent relation 3.85. In contrast to these meaneld results the accurate critical exponents for the Ising model, change according to dimension d of the system.2 For d = 2, 1 the exact solution Sect. 8.9 gives = 0, = 8 , = 7 , with = 15 4 following from the Widom scaling law 3.77 and the value for in this case corresponding to a logarithmic singularity. For d = 3 Le Guillou and Zinn-Justin 1980 used eld-theory to predict that = 0 = 0:1105, = 0:3252 and = 0 = 1:2412, with = 4:802 given by the Widom scaling law 3.77. High and low-temperature series expansions give good agreement with these values see, for example, Volume 2, Chap. 7. It was suggested by Domb and Sykes 1961 that = 5 and by Essam and Fisher 1963 4 5 that = 16 . This set of attractively simple values is completed, using the Essam Fisher and Widom scaling laws, with = 1 , = 5. 8
3.6.2 The Van der Waals Gas

Here the deduction of the critical exponents for T Tc is harder than for the Ising model because of the lack of symmetry in the T; or T; plane. We now use the parameter b as the standard volume v0 of 1.54 and the critical parameters of 1.52 have the form 1 8 vc = 3 ; Pc = 27 " ; 3.97 Tc = 27 " ; where " = c=b is of the dimensions of energy. Using `per molecule' densities as in Sect. 1.6, Q1 = V , Q2 = M , = v and the conjugate eld = ,P and from 3.43 and 1.47, = a + Pv = : 3.98 In terms of the scaled variables t and de ned in 3.50 the van der Waals equation 1.43 takes the form P 8P 81 + t 27 3.99 Pc = Tc = + 2 , + 32 and along the critical isochore 1 3.100 , @P = , @P @v v=3 @ =0 = 4 T , Tc :
2
The factors which a ect the values of critical exponents for any model and the related idea of universality classes is discussed in Volume 2, Chap. 2.
3.6 Classical Critical Exponents 0:3 T " 0:2 3:0; 0:2963
89
Fig. 3.8. The phase
boundary and the coexistence diameter for the van der Waals uid.
0:1
5:0
10:0 v
15:0
With 3.57 and 3.58 this gives = 1. Using 3.99 we de ne 4P = P t; , P t; 0 = , t + 2 Tc Tc 22 + 4 + 32 and, on the critical isotherm T = Tc t = 0, 1
= P 0; , Pc = , 144 3 + O4 ; Tc giving with 3.56, = 3. Now, from 1.44 and 3.99, v = T v , 1 , 2v" , T lnv , 1 + T ; where the last term depends on T alone. This then gives 4 = t; , t; 0
3.101 3.102 3.103
Tc
Tc h1 i = ,1 + t 2 2 + ln 2 + 2 + 49+ 3 + 3 ht
1 , 1 + 5 2 , 1 3 + + 1 2 , 11 3 i : = ,4 3 36 16 36 432
To satisfy the phase equilibrium conditions 1.41 for T of 1 and 2 must be found such that 4P t; 1 = 4P t; 2 ; 4t; 1 = 4t; 2 :
3.104 Tc t 0, values 3.105
90
2 To order jtj 1 this can be achieved by putting 1 = ,2 = 6jtj1=2 3.106 1 . Thus the Widom relation 3.77 is obeyed and hence, from 3.55, = 2 and 3.106 shows that both branches of the coexistence curve exhibit the same singular behaviour. This is in accordance with assumption iii above concerning the coexistence diameter. It can, in fact, be proved directly that this assumption is correct for the van der Waals gas. Fig. 3.8 shows the phase boundary and the coexistence diameter in the v; T plane.
Examples
3.1 Show that the meaneld relation between m and H for the Ising model ferromagnet can be expressed in the form m = tanh h0 ; h0 = H + zJm=T: This is a Weiss eld equation. 3.2 For the Ising model ferromagnet in the meaneld approximation, show that the constant magnetic eld free energy per site f and the reduced entropy per site s can be expressed as 1 f = ,T ln2 cosh h0 + 2 zJ tanh2 h0 ; s = ln2 cosh h0 , h0 tanh h0 ; where h0 is de ned as in question 1. 3.3 For the Ising model ferromagnet in the meaneld approximation, show that for H = 0, T Tc, 1 , m2 @m = s @ H T , Tc1 , m2 : s Show that for T Tc the expression on the right-hand side is approximately 4=T exp,2Tc=T and thus tends to zero with T . 3.4 For the spin-1 Ising model ferromagnet in the meaneld approximation, show that a non-equilibrium constant magnetic eld free energy per site can be written in the form f = T x1 lnx1 + x2 lnx2 + 1 , x1 , x2 ln1 , x1 , x2 1 , 2 zJ x1 , x2 2 , x1 , x2 H : Here x1 and x2 are fractions of spins with = +1 and = ,1 respectively and have the character of order variables. From the conditions for f to have a stationary value with respect to x1 and x2 , show that the spontaneous relative magnetization is given by
Examples
91
where y = expzJms =T . Show that the critical temperature Tc = 2zJ=3.
y y,1 ms = x1 , x2 equilibrium = 1 + , + y,1 ; y
4. Antiferromagnets and Other Magnetic Systems
4.1 The One-Dimensional Antiferromagnet

The Ising model ferromagnet can be transformed to an antiferromagnet by changing J to ,J J 0 in the con gurational energy expression 2.63. The analysis of the one-dimensional case in Chap. 2 remains valid, and replacing J by ,J the zeroeld correlation relation 2.88 becomes h i i+k i = ,1k tanhJ=T k ; 4.1 which is the same as in the ferromagnet for even k, but is reversed in sign for odd k. At low temperatures the regions of local order are blocks of alternate spins, and nearest-neighbour pairs of parallel spins are the local uctuations which form the boundaries between these regions. As will be shown in Sect. 4.2, the equilibrium state at H = 0 for the antiferromagnet can be obtained from that of the ferromagnet by reversing every other spin, see Fig. 4.1. The values of U , F , and hence S , at H = 0 are unaltered by the change J ! ,J , as can be seen from 2.81 and 2.82. The zeroeld heat capacity given by 2.83 is also una ected. For H 6= 0, there is no simple connection between the states of the ferromagnet and the antiferromagnet although the basic symmetry relations of Sect. 2.4 remain valid. In the ferromagnet the eld reinforces the interaction in promoting parallel spins, but in the antiferromagnet the eld tends to upset the antiparallel nearest-neighbour spin pairs favoured by the interaction. The latter e ect can be clearly seen at T = 0, and to investigate this we replace J by ,J in 2.99 to obtain the relation
AFM FM
Fig. 4.1. One-dimensional antifer-
romagnet top row and ferromagnet lower row.
94
4. Antiferromagnets and Other Magnetic Systems +1
Fig. 4.2. Magnetization
,Hc
as a function of eld at zero temperature: full lines the antiferromagnet, broken lines the ferromagnet.
Hc H
,1
sinhH=T 4.2 sinh2 H=T + exp4J=T 1=2 for the relative magnetization of the d = 1 antiferromagnet. When T jHj, 4.3 sinh2 H=T ' 1 exp2jHj=T 4 and thus, from 4.2, as T=J ! 0, 8,1; H ,H , c m ! : 0; ,Hc H Hc , 4.4 1; H Hc , where Hc = 2J is the critical eld. The behaviour of the antiferromagnet on the T = 0 axis is thus quite di erent from that of the ferromagnet see Fig. 4.2. Since all thermodynamic functions are analytic for T 0 in the d = 1 case, the transition points Hc on the T = 0 axis are isolated singularities. The zeroeld susceptibility of the one-dimensional antiferromagnet also contrasts with that of the ferromagnet. Changing J to ,J J 0 in 2.101 yields @m 1 4.5 @ H = T exp2J=T for the antiferromagnet. Thus @m=@ H ! 0 as T=J ! 0 or T=J ! 1 and has a maximum at T = 2J . The low susceptibility values thus found for T J are another consequence of the `con ict' between the antiferromagnetic spin interaction and the magnetic eld.
m=
4.2 Antiferromagnetic Ising Models

So far it has been assumed that, if boundary e ects are disregarded, all lattice sites in the Ising model are èquivalent'. We must now look more carefully at
95
what this term implies. Firstly, it means that, as for all regular lattices listed in Appendix A.1, the unoccupied lattice looks exactly the same when viewed from any site. That is each site has the same number zp of p-th neighbour sites for any p. It also means that the interaction of a member of any species j with an external eld is the same for all sites and the interaction energy of a member of species j and a member of species ` depends only on the relative position of the lattice sites they occupy.1 This condition could easily be extended from two-body pairwise interactions to n-body interactions. The equivalence of the lattice sites means that it may be possible to divide the lattice into two equal sublattices, labelled a and b. The sublattices are interpenetrating if all nearest-neighbour sites of any a-site are b-sites and vice versa. A lattice which can be divided into two equal interpenetrating sublattices is called loose-packed whereas a lattice for which division of this kind is not possible is close-packed. The d = 1 lattice is loose-packed since it can be divided into interpenetrating sublattices simply by labelling alternate sites a and b. The a and b sublattices for the loose-packed square, honeycomb, simple cubic, body-centred cubic and diamond lattices are shown in Appendix A.1. We shall use the constant magnetic eld distribution exclusively in this chapter. The antiferromagnet with nearest-neighbour interactions only on a loose-packed lattice will be called `simple', and by adapting 2.66 its Hamiltonian may be written as
fi;jg fig fjg a b i and j respectively denoting the spin variables for a site i of sublattice
b H =J
n:n: X
a b , H
a i +
b
4.6
a and a site j of sublattice b. The parameter J of 2.66has been changed to ,J J O and the fact that all nearest-neighbour pairs consist of an a-site and a b-site has been used. For H = 0 4.6 can be written
n:n: b = E = ,J X H
This is the Hamiltonian of a ferromagnet with spin variables ia and , jb . There is a one-to-one correspondence between microstates of equal energy in the zeroeld ferromagnet and antiferromagnet. These microstates are connected by the operation of reversing the spins on one sublattice, leaving those on the other sublattice unchanged, while changing the sign of the interaction parameter. It follows that in zero eld the partition function for the antiferromagnet is equal to that for the ferromagnet. Hence, as we have seen for the one-dimensional case, the free energy, entropy, internal energy and heat
1
fi;jg
a b i , j :
4.7
`Relative position' involves the direction as well as the length of the lattice vector connecting the two sites.
96
capacity are the same. The critical temperature Tc, where there is a heat capacity singularity at H = 0, is the Curie point for the ferromagnet, and is termed the N el point for the antiferromagnet. Experimentally, the phee nomenon of antiferromagnetism is commoner than that of ferromagnetism Martin 1967. The type of long-range order in the antiferromagnet below the critical temperature is quite di erent from that in the ferromagnet. We de ne a relative magnetization for each sublattice by the relations X a X b ma = 2=N mb = 2=N 4.8 i ; j ; where the sums are taken over all sites of sublattice a and sublattice b respectively. The overall relative magnetization m is then given by
X a X b 1 4.9 m = N ,1 i + j = 2 ma + mb : For the zeroeld ferromagnet above the critical temperature Tc, ma = mb = m = 0 whereas below Tc, ma = mb = ms . Now, from 4.7, the zeroeld equilibrium state of the antiferromagnet can be derived from that of the ferromagnet by reversing the magnetization of one of the sublattices. Hence above the N el point T Tc the magnetizations of both sublattices are e zero, but below it T Tc we have either 9 ma = ms ; mb = ,ms ; = or m = 0: 4.10 ma = ,ms ; mb = ms ; ; The magnetizations of the two sublattices are equal and opposite, giving zero overall magnetization. This is the rst example we meet of the important cooperative phenomenon of sublattice ordering. The degeneracy displayed in 4.10 for H = 0, T Tc, parallels the degeneracy m = ms of the ferromagnet. As T=J ! 0, the Ising model antiferromagnet approaches the state ma = 1, mb = 1 where all nearest-neighbour spin pairs are antiparallel see Fig. 4.3. Since, from 4.10, m is a continuous function of H as the interval 0; Tc of the H = 0 axis is crossed, this interval is no longer a rst-order transition line, as it is in the ferromagnet. For H 6= 0 there is no correspondence between the states of the antiferromagnet and those of the ferromagnet. From the equivalence of sites in the antiferromagnetic model it may be expected that, if either the magnetic eld or thermal disordering is strong enough to o set the antiferromagnetic spin interaction, the equilibrium state will be one in which ma = mb = m. Such a state is called paramagnetic as opposed to the antiferromagnetic state in which the spins on sublattices a and b are oppositely oriented. The transition
fig fjg fig fj g
97
curve in the T; H plane on which the change of symmetry from the paramagnetic state to the antiferromagnetic state takes place for d 1 has to be investigated by approximation or series methods even for d = 2, since no exact results are available for either the ferromagnet or antiferromagnet with H 6= 0 and T 0. However, from the zeroeld results above, the transition curve meets the H = 0 axis at T = Tc and from the symmetry relations of Sect. 2.4 it must be symmetrical about H = 0. Also, the points where it meets the T = 0 axis can be determined quite easily. At T = 0, the equilibrium state is one of minimum enthalpy, as shown at the end of Sect. 2.5. In the T = 0 paramagnetic state, all ia = jb = 1, according as H ? 0 and the 1 con gurational energy is 2 zNJ . In the T = 0 antiferromagnetic state, on the other hand, all ia = 1 and all jb = ,1 or vice versa; the con gurational 1 zNJ and the magnetization is zero. Thus, if Hp and Ha denote energy is , 2 the enthalpies in the paramagnetic and antiferromagnetic states respectively, by 1.32, Hp = 1 zNJ , N jHj ; 2 4.11 1 zNJ ; Ha = , 2 and Ha Hp , making the antiferromagnetic state stable if ,Hc H Hc = zJ: 4.12 The last relation de nes the critical eld Hc and the zero-temperature values for the magnetization are the same as those for the one-dimensional case given in 4.4 and Fig. 4.2 with the coordination number z replacing the d = 1 value z = 2. However, for d 1, the transition points H = Hc at T = 0 are no longer isolated singularities. To sum up our deductions from the basic symmetries of the antiferromagnetic model on a loose-packed lattice: i At H = 0 and T Tc, by simultaneous reversal of the signs of all spins on one sublattice and of the sign of the a b interaction, a correspondence is established between the states of the antiferromagnet and those of a ferromagnet in which h i i has the same value for all lattice sites i.
Fig. 4.3. The antiferromagnetic ground state for the

square lattice in zero eld.
98
ii For su ciently large jHj or T the antiferromagnet may go into a paramagnetic state in which h i i has the same value for all lattice sites i. It is useful to generalize the simple antiferromagnetic model by introducing a ferromagnetic interaction i.e. tending to spin parallelism between spins on the same sublattice. In a loose-packed lattice see Appendix A.1 both sites of any second-neighbour pair lie on the same sublattice, and we introduce an energy term ,J2 i j for each such pair. The Hamiltonian 4.6 is modi ed to
n:n: s:n: b = J1 X ia jb , J2 X H fi;jg i;j g
X a X fb
a b , J
s:n: X
,H
fi;jg
b b
`s.n.' denoting the sum over second-neighbour pairs and both J1 and J2 are taken to be positive. At H = 0 we thus have
fig
i +
fjg
4.13
X b H = J1
= , J1
n:n:
fi;jg
a b i j , J2
s:n: X
n:n: X
fi;jg
a a , J
s:n: X
4.14 Property i above is thus still applicable, but the corresponding ferromagnet has second as well as nearest-neighbour interactions, its con gurational energy being given by 3.26 with n = 2. Since all sites remain equivalent, property ii is still applicable. The enthalpies for the paramagnetic and antiferromagnetic states at T = 0 are 1 Hp = 1 z1 NJ1 , 2 z2 NJ2 , N jHj ; 2 4.15 Ha = , 1 z1 NJ1 , 1 z2 NJ2 2 2 and the critical eld is Hc = z1J1 : 4.16 The simple antiferromagnet is recovered by putting J2 = 0, J1 = J and z1 = z. It can be seen that the critical eld is not altered by the introduction of the second-neighbour interaction. This is because the second-neighbour energy contribution is the same for the paramagnetic and antiferromagnetic ground states. We now consider brie y, the thermodynamics of the system, which is restricted in the sense of Sect. 2.3.4, since every lattice site is occupied by a
fi;jg
a b i , j , J2
X s:n:
fi;jg
fi;jg
2
b b
a a , J
s:n: X
fi;jg
, ib , jb :
4.3 Mean-Field Theory
99
spin. As there is only one component we can put 1 = H, 2 = n = ,P . The use of the constanteld Hamiltonian implies that 1 = H is an independent variable so that the partition function is of type 2.58, and by 2.59 the free energy per site f = , P . It follows that magnetized phases are in equilibrium if T , H and f have the same values in each phase. For the Ising model ferromagnet these conditions apply by symmetry to the oppositely magnetized phases on the segment 0 T Tc of the H = 0 axis in the T; H plane. For the models considered here with J2 6= 0, phase transition lines can occur away from the H = 0 axis, and to investigate them the relation df = ,mdH 4.17 at constant T , derived from 1.34 or 2.60, is essential.

A meaneld theory of the model with Hamiltonian 4.13 will be developed by extending the procedure of Sect. 3.1. The nearest-neighbour h ia jb i is replaced by ma mb and the second-neighbour h ia ja i and h ib jb i are replaced by m2 and m2 respectively. The number of con gurations for given a b values of ma and mb is the product ma mb with de ned by 3.5 except that N is replaced by 1 N , the number of sites in each sublattice. 2 Using Stirling's formula, 4.9 and 3.41 a constanteld con gurational free energy per site is f = 1 T 1 + ma ln1 + ma + 1 , ma ln1 , ma 4 + 1 + mb ln1 + mb + 1 , mb ln1 , mb , 4 ln2 1 + 1 z1 J1 mamb , 4 z2 J2 m2 + m2 , 1 Hma + mb : 4.18 a b 2 2 Since, in the constant magnetic eld distribution, the sublattice magnetizations ma and mb are functions of the independent variables T and H, this is an example of a non-equilibrium thermodynamic potential which, as in the general equation 2.141, depends on several order variables. The equilibrium state corresponds to the minimum of f with respect to the order variables ma and mb . Before deriving the conditions for this minimum it is useful to de ne a function 1 ln 1 + m , ,1 1 + m = ,g,m : 4.19 gm = 2 1 , m The two new parameters occurring here and one for later use are de ned by
zJ = z 2 J2 ;
1 1
= zT ; J
1 1
h = zH : J
1 1
4.20
100
Here the parameter expresses the magnitude of the second-neighbour interaction relative to the rst, and and h are respectively a reduced temperature and eld. The rst derivative of gm is given by g0 m = 1 ,1m2 , ,1 1 + : 4.21 The conditions for a stationary value of f can now be expressed in terms of the function gm as 2 @f = gm + ,1 m + m , ,1 h = 0 ; The stationary value is a minimum if " @2f 4 @2f @2f , =
a a b T @ma 2 @f ,1 ,1 T @mb = gmb + ma + mb , h = 0 :
4.22
2
4.23 By substituting conditions 4.22 in 4.18, it can be shown that, at equilibrium, 4f = ln 1 , m2 1 , m2 , ,1 2m m , m2 + m2 : a b 4.24 a b a b T 16
T 2 @m2 @m2 @ma@mb a b = g0 ma g0 mb + ,1 fg0ma + g0 mb g 0 :
4.3.1 The Paramagnetic State

The paramagnetic state, where ma = mb , is stable for large values of or h. From 4.9 and 4.22 ma = mb = m ; 4.25 gm + 2m ,1 = h ,1 : From 4.23 the paramagnetic state is internally stable when g0 m 2 + 2g0m = g0 m g0 m + 2 ,1 0 : 4.26 As h or decreases, g0 m will approach zero before g0m + 2 ,1 , the latter expression being in fact always positive for 1 by 4.21. Hence the paramagnetic state will become unstable with respect to sublattice ordering when g0 m = 0 : 4.27 By eliminating m between 4.27 and 4.25 we obtain the equation h = h of the critical curve in the ; h plane see Example 4.2, outside which the paramagnetic state is stable. Any second-order transitions, that is ones in which m is continuous, from the paramagnetic to the antiferromagnetic state
101
occur on this curve where we shall show that there is a discontinuity in @m=@h . From 4.27, on the critical curve 12 m = m = 1 + + , : 4.28 1 From 4.25 @m 1 4.29 @h = g0 m + 2 ; 1 so that @m=@h ! 2 as m ! m + 0 i.e. from the paramagnetic side. The critical curve cuts the h = 0 axis orthogonally where g0 0 = 0, giving a critical N el temperature e Tc = z1 J1 + z2 J2 : 4.30 c = 1+; This is the point where the paramagnetic state becomes unstable when temperature is reduced at zero eld.
4.3.2 The Antiferromagnetic State

Here there is sublattice order and ma 6= mb . At h = 0, the equilibrium equations 4.22 are satis ed by ma = ms , mb = ms provided that gms = 0 : 4.31 Consider a ferromagnet with nearest and second-neighbour interactions J1 and J2 respectively. From 3.27 all relations of Sect. 3.1 are satis ed, with zJ replaced by z1 J1 + z2 J2 . Hence 4.31 is the relation giving the spontaneous magnetization ms for such a ferromagnetic model. In particular, this implies that the Curie temperature for the ferromagnet is equal to the N el e temperature given by 4.30 for the antiferromagnet. Mean- eld theory thus yields the ferromagnet antiferromagnet equivalence at H = 0 deduced above from the Hamiltonian 4.13. In the antiferromagnetic state the degree of sublattice order can be expressed by the parameter 1 = 2 ma , mb ; 4.32 which, from the previous paragraph, is equal to ms in zero eld. However, for h 6= 0 the simple relationship between antiferromagnet and ferromagnet no longer exists. Some of the most interesting properties of the model depend on behaviour near the critical curve. Accordingly, we de ne = m , m ; 4.33 " = h , h ;
102
Substitution of m = m and h = h in 4.25 and combination with 4.22 yields 1 ,1 4.39 2 gm + + + gm + , , gm = " , 2 : The left-hand side is an even function of . Hence, using Taylor expansions of gm + as above, and rearranging, only even powers of are left. We can then substitute for 2 from 4.36 and, after a good deal of algebra, obtain the relation " = B1 + B2 2 + O 3 ; 4.40 for 0, where B1 = 2 + , ; 4.41 B2 = 8m =5 2 1 + 2 3 , 8 : It is now easy to derive an expression for @m=@h on the antiferromagnetic side of the critical curve. When a higher-order transition occurs on this curve, the magnetizations ma and mb are continuous and both approach m as h ! h , 0. Thus the equilibrium value of ! 0 as " ! ,0. Substituting from 4.41 and 4.38,
which are the deviations of the reduced magnetization and eld respectively from their values on the critical curve. From 4.9, 4.32 and 4.33 the sublattice magnetizations are ma = m + + ; 4.34 mb = m + , : A relation between and is obtained from the di erence of the two equations 4.22. After substitution from 4.34, gm + + , gm + , = 0 : 4.35 Note that the eld terms have been eliminated and that the left-hand side is an odd function of , which can be written in the form f 2 ; . Hence, after Taylor expansions of gm + in powers of and some manipulation, we obtain 2 = A1 + A2 2 + O 3 ; 4.36 where A1 = 2m 1 , 4 ; 4.37 1 A2 = 5 5 , 20 + 64 2 , 64 3 and 1 = 1 + 3m2 = 4 +14+ , 3 : 4.38
@m
1 lim @h h!h ,0 = "!,0 " = B1
103
4 +4 = 2 3 , + , 3 , 1 : 4.42 13 When the critical curve meets the -axis, at = c = 1 + , the last expression reduces to 1 , and it increases from 1 as decreases from c . Since 2 2 1 @m=@h ! 2 , for all , as h ! h + 0 i.e. when the critical curve is approached from the paramagnetic side there is a discontinuity in @m=@h on the critical curve, con rming that the transition there is second-order in the meaneld approximation. This assumes that B1 0; we consider the situation B1 0 in Sect. 4.4.
4.3.3 The Simple Antiferromagnet

We can now apply this analysis to the simple antiferromagnet, for which = 0 since there are no second-neighbour interactions. Now, by 4.16 and 4.20, the critical eld values for the antiferromagnetic paramagnetic transition on the = 0 axis correspond to h = 1. It can be shown see Example 4.2 that the critical curve h = h meets the = 0 axis at the points h = 1 , , which coincide with the transition points when = 0. Thus, when = 0, the antiferromagnetic paramagnetic transitions can be second-order and therefore occur on the critical curve in the ; h plane for the entire range 0 c . However, there is a stability condition which must be satis ed. Since dH = z1 J1 dh by 4.20, it follows from 4.17 and 1.107 that a homogeneous phase is stable only if
@h
@m
4.43
at all points. From 4.29 the condition 4.43 is satis ed everywhere in the paramagnetic region. Just on the antiferromagnetic side of the critical curve i.e. as m ! m , 0 we have, from 4.40, 0 everywhere on the critical curve. We shall show below that 4.43 is satis ed on the h = 0 axis, and it can be con rmed numerically that this is true in the rest of the antiferromagnetic region. Hence, for the simple antiferromagnet, the transition line is identical to the critical curve which is the = 0 line in Fig. 4.4. Although the nite discontinuity in @m=@h at the transition may be an artifact of meaneld theory, the general picture of the phase diagram for the simple antiferromagnet, with the antiferromagnetic and paramagnetic regions separated by a higher-order transition line, is con rmed by accurate series work Domb 1974.
@m = B1 and, for = 0, B1
@h
4.44
104
=0
1:0
Fig. 4.4. Second-order full

and rst-order bro transition curves in temperature eld plane for the antiferromagnet = 0 and metamagnet = 1; the tricritical point is circled.
lines ken line
h
0:5
=1
0:5
T=Tc
1:0
m = 0 and, from 4.21 and 4.29 with = 0, @m 1 1 @h = + 1 = + c :

Hence, from 1.36 and 4.20,
We now consider the zeroeld susceptibility. For T
Tc , ma = mb =
4.45 4.46
It can be shown from 4.9, 4.22 and 4.23 that, for general values of ma , mb , and , @m g0ma + g0mb : 4.47 @h = 2 For H = 0 and T Tc , equation 4.10 applies and then 4.47, 4.21 and 4.23 give, for = 0,
T
C T= zJ
1 1
@m
@h
C = T +T ;
c
m2 C = v0 :
0
From 4.48, 3.12 and 3.13, T and d T =dT ! 0, as T ! 0, whereas d T =dT 0, for T Tc. From 4.46, d T =dT 0, for T Tc. Hence, while T is continuous at T = Tc , d T =dT has a nite discontinuity, giving a cusp in the T ; T curve. This contrasts with accurate series results Fisher and Sykes 1962, Domb 1974 which indicate that d T =dT is continuous at all T and that T has a maximum at a temperature which is about 1:5Tc for d = 2 and 1:08Tc for d = 3.
= T 1 , mC,1 + T : 2
s c
4.48
4.4 Metamagnetism: Tricritical Points and Critical End-Points
105
When an experimental plot of ,1 against T is asymptotic at high temT peratures to a line AT , Tp, for some constant A, a Curie Weiss law is said to be obeyed, and Tp is called a paramagnetic Curie temperature. From 3.18, Tp = Tc in the meaneld approximation for the Ising model ferromagnet. In a real ferromagnet, the paramagnetic Curie temperature is 10 or 20K above the ferromagnetic Curie temperature Martin 1967. From 4.46, Tp = ,Tc for the meaneld antiferromagnet, and a negative paramagnetic Curie temperature is a well-known characteristic of real antiferromagnetic substances for which the susceptibility below the N el point depends on the direction e in which the external magnetic eld is applied. The curve 4.48 corresponds qualitatively to the response to a eld along the zeroeld magnetization axis of the sublattices. For responses to elds perpendicular to the axis of zeroeld magnetization see Fisher 1963, Martin 1967 and Mattis 1985.

In a metamagnetic material the array of sites occupied by the magnetic dipoles is again divided into equivalent sublattices a and b, whose zeroeld magnetization is equal and opposite at low temperatures. However, the a and b sites now lie in alternate planes; there is a ferromagnetic interaction between nearest-neighbour a a or b b pairs of spins and an antiferromagnetic interaction between the closest a b pairs of spins Martin 1967. The reason for this di erence in the interactions lies in the structure of the crystal and particularly in non-magnetic atoms situated between the magnetic sublattice planes. A two-dimensional version of the metamagnetic ground state is shown in Fig. 4.5. For an Ising model metamagnet the Hamiltonian is identical to 4.13, except that the `second neighbour' designation should be replaced by `nearest neighbour in the same sublattice'. The symmetry properties i and ii are una ected. The zeroeld metamagnet is equivalent to a non-
Fig. 4.5. Ground state for square

lattice metamagnet in zero eld.
106
isotropic ferromagnet with interaction parameters J1 and J2 , and a stable paramagnetic state exists at large jHj and T . The meaneld approximation for the free energy per site is again 4.18, z1 and z2 now respectively denoting the numbers of nearest neighbours of a given site in its own and in the other sublattice. A positive is now an essential characteristic of the model. Hence, on the = 0 axis, the end points h = 1 , of the critical curves are not coincident with the transition points h = 1. This indicates that, at low temperatures, there cannot be a second-order transition from the antiferromagnetic state, where ma and mb have opposite signs, to the paramagnetic state where ma = mb = m. If ma = ,mb = ms 0 at h = 0, and h is then increased continuously at a constant value of , there comes a point where the equilibrium equations 4.22 yield mb = 0, and mb changes sign from negative to positive. For 0 0 when c it can be shown see Example 4.3 that @h=@m mb = 0. There is thus an instability loop in the plot of h against m at constant resulting see Sect. 1.10 ii in phase separation, with the point where mb = 0 lying in the region of intrinsic instability. Hence a line of rstorder transitions exists in the upper half of the ; h plane, terminating at = 0, where we would expect it to meet the h-axis at the transition point h = 1. This is con rmed see Example 4.3 by the fact that, if the values of h and where mb = 0 are plotted in the ; h plane, this curve meets the h-axis at h = 1. The remainder of this section will he concerned with how this rst-order transition line terminates at its upper end. There is also a mirror-image transition line in the h 0 half-plane, but we need consider only the h 0 line explicitly. Since @h=@m = B1 at m = m , 0, the stability condition 4.43 is not satis ed on the antiferromagnetic side of the critical curve if B1 0. From 4.42, B1 = 2 at = c = 1 + and decreases steadily with , passing through the value zero at = t where 3 , 1 + 1 = 3 , 1 c : 4.49 t= 3 3 For c t there is a second-order transition on the critical curve, as discussed at the end of Sect. 4.3. For t 0 and su ciently large in fact, 3 , as shown below, the plot of h against m is similar to that in 5 Fig. 4.6 where the loop implies a rst-order transition. The magnetizations m1 and m2 in the conjugate phases correspond to the ends of the equal-areas tie line shown, and the slope discontinuity at m = m falls in the instability interval m1 ; m2 . Hence, for the range t 0, there is a line of rst-order transitions, as shown for the = 1 case in Fig. 4.4. The method of Sect. 1.10 i, with fn,1 = f and i = z1 J1 h, was used for calculation. We thus have a new type of critical phenomenon. The line of rst-order transitions in the ; h plane ends at the point = t , h = h t but is continued by a secondorder transition line. Such a point is called tricritical, and t , given by 4.49, is a tricritical temperature.
4.4 Metamagnetism: Tricritical Points and Critical End-Points 1:0 h
107
h
0:5
0:2 m1 0:4 m 0:6 m m2 Fig. 4.6. Field-magnetization plot for = 1, = 1 c, showing an instability loop 2 and the equal-areas tie line. 0
For small values of the rst-order transition line starting at h = 1 on the = 0 axis cannot end at a tricritical point; in fact from 4.49, t 0 for 1=3. Hence there is an ordinary critical point where, by 1.102,
@h
@m
= 0;
@2h
@m2
= 0:
4.50
Since the condition 4.43 is satis ed everywhere in the paramagnetic region the critical point must lie in the antiferromagnetic region, that is inside the critical curve in the ; h plane. Fig. 4.7 shows part of the ; h plane for = 0:3, the rst-order transition line again being obtained by the method of Sect. 1.10 i. For any between the values corresponding to points Q and C, the plot of h against m will show a loop, but the point m = m , where a slope discontinuity occurs, lies outside the instability interval associated with the loop. Hence, when h is increased at constant , there is a rst-order transition on the segment QC, and then an antiferromagnetic paramagnetic second-order transition at h = h on the critical line. At the temperature corresponding to Q the point m = m enters the instability interval of the rst-order transition, and the line of second-order transitions on the critical curve terminates. We have, in fact, another new critical phenomenon and a point like Q where a line of second-order transitions terminates on a rstorder transition line is known as a critical end-point. It is now easy to nd the value = 1 where the low- critical endpoint phase pattern goes over to the high- tricritical point phase pattern. Consider Fig. 4.7 and suppose that steadily increases. The segment QC
108
will shrink until eventually the critical end-point Q coalesces with the critical point C at = 1 . When coalescence occurs, the conditions 4.50 de ning Q are satis ed on the antiferromagnetic side of the critical curve so that, by 4.40, B1 = B2 = 0. Comparison of the expressions in 4.41 then yields 1 = 3 . This result was derived by Bidaux et al. 1967, who give a 5 comprehensive set of diagrams. We now consider the tricritical regime again and in particular the situation just below the tricritical temperature where = t , , 1. The conjugate antiferromagnetic and paramagnetic phases existing for t , are labelled by the indices 1 and 2. Noting that, from 4.17 and 4.20, z1 J1 ,1 df = ,dhm + hdm 4.51 the phase equilibrium conditions can be written in the form h2 , h1 = 0 ; Z m2 4.52 z1 J1 ,1 f2 , f1 = ,h2 m2 + h1 m1 + hdm = 0 : Using the quantities " and , de ned by 4.33, the equilibrium conditions 4.52 can be expressed as "1 = "2 ; Z 1 Z 2 4.53 "1 1 , "d = "2 2 , "d ;
0 0
m1
h 1:03
1:02 1:01 1:0
Q
Fig. 4.7. As in Fig. 4.4
0:99
0:18
0:2
0:22 T=Tc 0:24
for = 0:3, with critical end-point Q and critical point C.
109
with 1 0 and 2 0. On the antiferromagnetic side of the critical curve where 0 we use 4.40. Since 1 and B1 t = 0 we can, if only leading terms in are required, write B1 = ,A ; 4.54 B2 = B2 t = B ; where 2 62 ; 82 3 1 3 , 5 : B = 51 + 2 3 , 1 4.55 A = 1 + 2 For 0, using 4.29 with g0m = 0,
" = @m =2 : 4.56 m=m +0 For the tricritical range 3 , B 0 so that 4.40 and 4.48 4.56 yield 5
@h
a plot of the same general form as Fig. 4.6. Substitution in the two relations of 4.53 yields 2 ,A 1 + B 1 = 2 2 ; 4.57 1 2 3 2 , 2 A 1 + 2 B 1 = 2 : 3 After a little algebra we then have, for the magnetizations in the conjugate phases and the rst-order transition value of the eld, A m1 = m + 1 = m , 43Bj + O2 ; j 4.58 2 + 2 = m , 3A 2 + O3 ; m2 = m 32jBj 3 2 h = h + 16A j 2 + O3 : jB 4.59
From 4.59 the rst-order transition line and the critical curve have a common tangent at the tricritical point = t , = 0 as shown for = 1 in Fig. 4.4. From 4.59 the paramagnetic branch m = m2 of the coexistence curve in the ; m plane has a common tangent with the critical magnetization curve m = m at = t , but the antiferromagnetic branch m = m1 meets it at a non-zero angle. Hence the coexistence curve has a cusp at the tricritical point, in contrast to the continuous tangent at a critical point. For an example in another system, see Fig. 6.10. At the tricritical point h , h ' Bm , m2 for m m i.e. on the antiferromagnetic side of the critical curve. Now, from 4.55, B 0 for 3 , giving h , h 0. This con rms that, for 3 , the tricritical point 5 5 falls in the instability range of the rst-order transition. As we have seen,
110
t becomes negative for 1 . The subject of tricritical points and critical 3 end-points is taken up again in Sects. 6.8 6.10 and general references are given.
4.5 Ferrimagnetism: Compensation Points

Like an antiferromagnet, a ferrimagnet in zero eld has oppositely magnetized sublattices below a critical temperature. However, the magnetizations of the sublattices do not completely cancel and so there is still a spontaneous magnetization. The crystal structure of a real ferrimagnet can be extremely complicated Martin 1967, but we shall introduce ferrimagnetism through a generalization of the Ising model antiferromagnet considered in Sects. 4.2 and 4.3. Suppose that a lattice of N sites is divided into sublattices a and b of Na and Nb sites respectively, where Na and Nb may be unequal. Each a site has zaa nearest-neighbour a sites and zab nearest-neighbour b sites; zbb and zba are similarly de ned. Then zaa + zab = z ; Nazab = Nb zba ; 4.60 zbb + zba = z ; where z is the coordination number of the lattice, and the nal relation follows from counting the number of a b nearest-neighbour pairs in two di erent ways. A two-dimensional example from Bell 1974a and based on the triangular lattice is shown in Fig. 4.8. Here the b-sites form a Kagom sublattice, e Nb = 3Na, zaa = 0, zab = 6, zbb = 4, zba = 2. We assume ferromag netic interaction energies ,Jaa ia ja and ,Jbb ib jb for a a and b b nearest-neighbour pairs respectively, and an antiferromagnetic interaction en ergy Jab ia jb for a b nearest-neighbour pairs, Jaa , Jbb and Jab being
a b b b a b a b a b b b b b a b a b a b b b a
Fig. 4.8. Plane triangular
lattice ferrimagnet. Reprinted from Bell 1974a, by permission of the publisher IOP Publishing Ltd.
111
all positive by convention. The spins on b-sites have magnetic moment m0 and those on a-sites magnetic moment rm0 . We de ne a parameter s by s = Nb =Na = zab =zba 4.61 and no generality is lost by taking s 1. Sublattice relative magnetizations, ma and mb , and the overall relative magnetization m are de ned by generalizing 4.8 and 4.9: , X b ; , X a ; ma = Na 1 mb = Nb 1 i j
fig
m = rNa + Nb
,1 X
r
fig
a + X b
fjg
With all spins in sublattice a completely aligned, its magnetic moment is of magnitude rNa m0 , whereas the corresponding magnitude for sublattice b is Nb m0 = sNam0 . The ratio of the `magnetic weights' of sublattices a and b is thus r=s. Because of the antiferromagnetic interaction between a and b, ma and mb will have opposite signs in small magnetic elds at T = 0. Hence for small positive H H still de ned by 1.55, ma = ,mb = 1 for r s and mb = ,ma = 1 for r s. The change in interaction energy involved in the transition to the state where ma = mb = 1 at T = 0 is 2Nb za Jab = 2Nazab Jab . For r s the corresponding change in eld energy is ,2NbH, and for r s we have ,2NarH. The critical elds at T = 0 are zbaJab ; r s, Hc = 4.63 s=rzba Jab ; s r. Equation 4.18 for free energy in the meaneld approximation is easily generalized to 1 Nf ma; mb = 2 T fNa 1 + ma ln1 + ma + 1 , ma ln1 , ma + Nb 1 + mb ln1 + mb + 1 , mb ln1 , mb g , NT ln2 + Nb zba Jab ma mb 1 , 1 Na zaa Jaa m2 , 2 Nb zbb Jbb m2 a b 2 , Narma + Nb mb H: 4.64 To express the equilibrium conditions as succinctly as possible we de ne the parameters
fjg
rma + smb = :
r+s
4.62
a = zaa Jaa ; z J h= z H ; baJab

ba ab
z b = zbb Jbb ; J
= z TJ ; ba ab
ba ab
4.65
and then
112
N @f = 1 ln 1 + ma + smb , a ma , rh = 0 ; NaT @ma 2 1 , ma 4.66 N @f = 1 ln 1 + mb + ma , b mb , h = 0 : Nb T @mb 2 1 , mb For any solution ma; mb at given positive h, there is a solution ,ma; ,mb at ,h. A stationary point given by 4.66 is a minimum if " @ 2f 2 N 2T 2 @2f @2f , = N N @m2 @m2 @ma@mb a b a b
= 1 ,1m2 , a 1 ,1m2 , b , s2 0 : 4.67 a b The high-temperature zeroeld state ma = mb = 0 becomes unstable when = 0, giving the critical temperature 1 1 2 4.68 c = 2 fa + b + a , b + 4s 2 g : The model treated in Sect. 4.3 is recovered by putting a = b = , s = 1, r = 1. Since, for c , there is both spontaneous magnetization and oppositely magnetized sublattices, the terms `Curie' and `N el' are equally e appropriate for c . However the former is usually chosen. Since the lattice sites are no longer equivalent, the symmetry properties i and ii, discussed in Sect. 4.2, do not apply for the Ising model ferrimagnet. At H = 0 it is still, of course, possible to nd an equivalent model by simultaneously changing all ib to , ib and Jab to ,Jab . However, though all interactions are then ferromagnetic, h ia i 6= h ib i and the useful relation 4.10 is not satis ed. Property i and equation 4.10 do still apply, however, in the particular case Jaa = Jbb = Jab = J ; 4.69 which will be termed the `simple ferrimagnet'. Here the zeroeld `reversal' operation produces an Ising model ferromagnet with the same interaction parameter for all nearest-neighbour pairs; in fact, it makes all sites equivalent. For the ferrimagnet there is no paramagnetic state with all h i i equal except at H = 0, c when all h i i = 0 and = 0, H ,Hc or H Hc . Secondorder transitions, which correspond to a change of the symmetry of the state, thus no longer occur for H 6= 0. Since there is a spontaneous magnetization which has opposite signs for H = +0 and H = ,0 the interval 0; c of the -axis is a rst-order transition line. There may also be rst-order transition lines starting from H = Hc on the H-axis. In some real ferrimagnets there is a compensation temperature between T = 0 and T = Tc where the spontaneous magnetization is zero. The curve of spontaneous magnetization against T has a downward cusp at this temperature.
113
4.5.1 Zero Field

We now investigate the conditions for the occurrence of a compensation point in the meaneld theory of the Ising model ferrimagnet. For de niteness we put h = +0, that is we consider limiting values as h ! 0 from above, so that the spontaneous magnetization is non-negative and hence, by 4.62, rma + smb 0. A compensation point occurs when rma + smb = 0 which implies rma = ,smb and jrma j=jsmb j = 1 : 4.70 The signs of ma and mb both reverse at the compensation point so that one zeroeld solution ma ; mb of the equilibrium equations 4.66 is exchanged for the alternative ,ma ; ,mb. Although ma and mb are discontinuous, their magnitudes jma j and jmb j and the ratio jrma j=jsmb j are continuous. Just below the N el temperature c , ma and mb are small so that 4.66 with e h = 0 can be linearized to give ma = s = , : 4.71 b m , Equating the last two expressions gives 4.68 for c , again, as would be expected, and hence
T !Tc ,0 jsmb j
1 r0 = c , a = 1 fb , a + b , a 2 + 4s 2 g : 4.73 2 It can be veri ed that r0 s if and only if s b , a + 1. At = 0, jrma j=jsmb j = r=s. Hence, if r lies between r0 and s, jrma j=jsmb j is greater than 1 at one end of the range 0; c and less than 1 at the other. So 4.70
where
lim jrma j = rr ;
0
4.72
will be satis ed at some temperature between 0 and c , giving a compensation point. Given that we take s 1, the conditions for the existence of such a point are therefore s b , a + 1; and r0 r s; or 4.74
s b , a + 1; and s r r0 : With r and s xed, it can be seen that, when r s, a compensation point can be made to occur by increasing b relative to a , and vice versa. Compen-
sation points thus exist because stronger ferromagnetic interactions in one sublattice counteract its lower magnetic weight in an interval below c , but cannot do so for lower values of where jma j and jmb j are nearer to unity. In the simple Ising model ferrimagnet, de ned by 4.69, the spontaneous
114
magnetization is jr , sjmi:m: =r + s, where mi:m: is the spontaneous mags s netization of the equivalent ferromagnet. The sublattice magnetizations are too closely locked together to allow a compensation point. It is easily shown that in this case s = b , a + 1 and r0 = s. For the general case, given that the necessary inequality is satis ed, the compensation temperature can be found by substituting h = 0 and rma + smb = 0 into 4.66 see Example 4.5. Spontaneous magnetization curves are shown in Fig. 4.9, for the model p 1 of Fig. 4.8 with a = 0, b = 1. Here r0 = 2 1 + 13 = 2:3028 s = 3.
0:2
ms
0:1
4:0
2.0
r0
2:6 3:0
0:2
0:4
0:6
0:8 T=Tc
Fig. 4.9. Spontaneous magnetization temperature curves for the model of Fig. 4.8.
Each curve is labelled with the value of r.
4.5.2 Non-Zero Field

Suppose rst that s r. Then, at h = +0, mb 0 and ma 0 for small . However, for large positive values of h, ma and mb , though still unequal, must both be positive. Thus, as h is increased at constant , there must be a point where ma = 0. Substitution in 4.66 yields an ma = 0 curve de ned by
mb = rh ; s
1 1 + mb 0 2 ln 1 , mb , mb = 0 ;
4.75
115
where
= b + s=r : 4.76 This curve meets the = 0 axis at the critical h values of s=r. The second equation of 4.75 is similar in form to 3.12 and has solutions mb 6= 0 when 0 . At such a solution the gradient of the expression on the left-hand side is positive so that 1 , 0 0: 4.77 1 , m2 b If ma = 0 is substituted in 4.67 we obtain 1 b 1 , a , s : 4.78 = 1 , m2 , 2 b From 4.77 the rst bracket is positive and so 0 for a temperature range which includes all a and an interval above a . If h is plotted against m for any in this range there will be an interval where @h=@m 0, giving an instability loop. Thus, in the ; h plane, there are two rst-order transition lines, symmetrical about the h = 0 axis, which start at the critical eld values h = s=r on the = 0 axis. There are discontinuities in ma and mb , involving a change in the sign of ma, at the transition lines. We next investigate how these transition lines end. The curve de ned by 4.75 cuts the h = 0 axis orthogonally at = 0 and from 4.68, 4.73 and 4.76, 0 r0 . Since r s there is therefore c if and only if r no compensation point if the curve 4.75 intersects the h = 0 axis at a point outside the interval 0; c . Noting that, by de nition, jma j 1 and jmb j 1, the stability condition 0 is satis ed for all ma and mb when c . Hence when 0 c the instability regions on the m = m ; h surface associated with the rst-order transition lines starting at = 0, h = s=r do not reach the h = 0 plane. These transition lines are therefore terminated in the ; h plane by critical points at which c , h 6= 0, a situation illustrated by the r = 1 curve in Fig. 9.5 of Chap. 9. Now suppose that 0 c which implies r0 r s and hence the existence of a compensation point. With 0 c , 0 at every point on the curve 4.75 so that the regions of instability on the m = m ; h surface do reach the h = 0 plane. The transition lines starting at = 0, h = s=r in the ; h plane meet the h = 0 axis orthogonally. For h = +0 or h = ,0 there are discontinuities in ma and mb at the compensation temperature and hence the crossing point must be identi ed with the compensation point. The situation is illustrated by the r = 2:7 curve in Fig. 9.5 of Chap. 9. The discontinuities in m are sketched in Fig. 4.10. The case a = 0 needs separate consideration. When ma = 0 = 1 ,1m2 , b , s2 : 4.79 b
0
116
s s+r
h s r
m
0
Compensation point
T=Tc
Fig. 4.10. Discontinuities in the magnetization surface for a Ising model ferrimagnet with a compensation point. The hatched lines are parallel to the m-axis.
Since mb is a solution of the second of equations 4.75 it can be shown that, as ! 0, 1 , m2 ,1 approaches in nity faster than ,n for any positive b integer n. It follows that 0 for small and that critical elds h = s=r on the = 0 axis represent isolated singularities. However if 0 c , implying the existence of a compensation point, then 0 on the curve 4.75 in a temperature interval below 0 . At the compensation point the interval 0; c on the h = 0 axis is crossed orthogonally by a rst-order transition line which terminates at critical points in the ; h plane for which 0. This situation is illustrated by Fig. 9.6 of Chap. 9. So far we have assumed s r in considering rst-order transitions o the h = 0 axis, but the theory for r s runs in parallel, with an mb = 0 curve replacing the ma = 0 curve de ned by 4.75. Finally it may be noted that when the Ising model ferrimagnet was introduced, all interactions were, for simplicity, taken as nearest neighbour. However, interactions of any range can be incorporated into the meaneld theory by replacing such quantities as zaa Jaa by the appropriate sums. Although these results for the Ising model ferrimagnet have been obtained from the meaneld approximation, they are borne out qualitatively by results obtained by exact transformation theory for particular cases in Chap. 9 and by the real-space renormalization group methods described in Volume 2, Chap. 6. Of course, values for such parameters as r0 di er considerably from meaneld ones. The most interesting new cooperative phenomenon is the
Examples
117
compensation point formed by the orthogonal intersection of rst-order transition lines and illustrated in Fig. 4.10.
Examples
4.1 For the one-dimensional antiferromagnet considered in Sect. 4.1 show that @ 2 m=@ H2 = 0 where exp4J=T , 2 = cosh2H=T : Show that the curve de ned by this relation meets the T = 0 axis at H = Hc and the H = 0 axis at T = 4J= ln3. 4.2 For the antiferromagnet or metamagnet in the meaneld approximation, show from 4.25 and 4.27 that the critical curve i.e. the curve where the paramagnetic state becomes internally unstable in the ; h plane is given by 21 1 h = 2 ln 1 + m + 1 , m ; m = 1 + + , : 1,m 1 Hence show that, on the h 0 branch of this curve, as ! 0, h ! 1 , and as ! 1 + , h ! 0. 4.3 For the antiferromagnet or metamagnet in the meaneld approximation, show that, when mb = 0, ma = h = ms where ms 0 is the zeroeld sublattice magnetization at the same temperature. Show that the curve thus de ned in the ; h plane meets the h-axis at h = 1 and the -axis at = c . From 4.47 and 4.23 show that @h 0 0 = g ma + 1 g mb + 1 , 1 : 1 g0 m + g0 m @m a b 2 Hence show that, on the mb = 0 curve, @h 2 ,1 = 1 1 , ms , , , 1 : @m 2 ,1 2 1 , ms + , 21 + Using 4.31 and a relation similar to 3.13, show that, on the mb = 0 curve, @h=@m 0 for . 4.4 For the Ising model ferrimagnet in the meaneld approximation, show that the zeroeld susceptibility above the critical temperature c is given by , a , b , s ,1 = K T r2 , b + s , a , 2rs ;
118
! c + 0. Show that this expression can be transformed to 2 s ,1 = K + 2rs ,rb+ , a r T 2 +s 2 s r 2 2 , , r2 s s r2 +rs a, b s + r ,2s+ 2rs : + a + b r Hence show that the paramagnetic Curie temperature 4.5 Curie temperature!paramagnetic is lower than c and sketch the form of ,1 above the critical temperature. T 4.6 For the Ising model ferrimagnet in the meaneld approximation show, by substituting h = 0 and smb = ,rma into the equilibrium equations 4.66, that at a compensation point s + rma s + r 1 + m b ln s , rm , sr + ln 1 , ma = 0 ; a 1a+ m a 1 a ,1 = ma r + a ln 1 , ma : If r s, show that there is a non-zero solution for ma if r0 r, where r0 is de ned by 4.73. If r s, show that the corresponding condition is r r0 .
4. Antiferromagnets and Other Magnetic Systems where K is a constant. Hence show that ,1 ! 0 as T
5. Lattice Gases
5.1 Introduction
In a classical uid composed of M similar spherical molecules contained in a e volume V , the molecules are regarded as centres of force situated at points e r1; r2; : : : ; rM in V . The con gurational energy is the sum of 1 M M , 1 2 terms, each representing the interaction energy of a pair of centres and dependent on the distance between them. Thus, the con gurational energy is X E r1 ; r2 ; : : : ; rM = ujri , rj j: 5.1 The pair interaction energy ur is positive for small r and ur ! 1 as r ! 0. For larger r values, ur becomes negative with a minimum at r = r0 and, nally, ur ! 0 as r ! 1 see Fig. 5.1. The interaction is, therefore, strongly repulsive at small r, but becomes attractive for r r0 . In Chap. 1 we saw how van der Waals incorporated the essential elements of a hard core repulsion and an attractive interaction into a simple theory.1 Now suppose that the approximation is made of replacing the continuous volume by a regular lattice of N sites and con ning the centres of force to lattice sites with not more than one centre on each site. The M sites coinciding with a centre of force are said to be òccupied' by molecules and the remaining N , M sites are `vacancies' or `holes'. A model of this type is called a lattice gas.2 The volume per site is e v0 = V =N ; 5.2 which is regarded as constant. For d = 1 and d = 2, `volume' must be replaced by `length' and àrea' respectively. An interaction energy "k is associated with each k-th neighbour pair of occupied sites, where "k = urk , rk being the k-th neighbour pair distance on the lattice. An alternative is to put "k = urk , where u is an average of ur in a suitable neighbourhood of the k-th neighbour site.
1
fi jg
For an introduction to continuous uid models, see Rowlinson and Swinton 1982, Chap. 7. 2 This is accepted terminology although the term `gas' is not entirely appropriate since phase separation occurs for d = 2; 3. It would be better to use the name `lattice uid'.
120
5. Lattice Gases
ur r0 r
Fig. 5.1. The interaction energy for
two molecules with centres at a distance r.
If the mesh is so coarse that r1 = r0 see Fig. 5.1 then "1 = ur0 0. The simple lattice gas is obtained if we then put "k = 0 for k 1, so that the interaction is con ned to nearest-neighbour molecules. The hard core and attractive elements are still present, the hard core because two molecules cannot approach nearer than r1 and the attraction because "1 0. The simple lattice gas is mathematically equivalent to the Ising model ferromagnet as is shown in Sect. 5.3. The hard core can be extended by putting "1 = 1 so that no two nearestneighbour sites can both be occupied; this is a nearest-neighbour exclusion model. With "1 = "2 = = "m = 1 we have an m-th neighbour exclusion model. This type of model can undergo a solid uid melting transition Runnels 1972. Generally, the chief advantage of lattice gas models is their versatility as they can be modi ed to allow for non-isotropic and many-body interactions see Chaps. 7 and 9. If the standard volume in 1.54 is taken to be the constant volume per site v0 , then the dimensionless volume is V = N , the number of lattice sites, and e e the conjugate eld is ,P = ,Pv0 . For a d = 3 lattice P has the familiar dimension of energy volume=force area, but for d = 2 and d = 1 it has respectively the dimensions of energy area=force length and energy length=force. A two-dimensional or surface pressure can be measured experimentally for monolayers and the one-dimensional pressure is just a compression applied to the lattice regarded as a rod. The number density is de ned so that = 1 at closest packing. Where two molecules can occupy nearest-neighbour sites, = M=N : 5.3 If we de ne the Helmholtz free energy density as `per site', that is per unit of reduced volume, then 1.77 and 1.78 with n = 2, = 0 give aT; = , P ; 5.4 da = d ; where the second relation applies at constant T . Hence
P=
@a
@
,a:
5.5
5.2 The One-Dimensional Lattice Gas and the Continuous Limit
121
An alternative approach is to think of the volume as divided into N equal `cells'. It is assumed that: i there is not more than one molecule in each cell; ii the interaction energy between two molecules depends only on the relative position of their cells and is independent of their location inside the cells. If a lattice site is associated with each cell, the corresponding `lattice gas' and `cell' models are completely equivalent.

Critical phenomena will not occur at T 0 in a d = 1 model with interactions of nite range because any long-range order can be destroyed by a small number of local uctuations. We have seen in the d = 1 Ising model that with a few `wrong' nearest-neighbour pairs a con gurational energy near the ground-state energy can be attained without a majority of spins pointing in one direction. Hence a d = 1 lattice gas will not display phase separation. However it can, as we shall see in Sect. 5.5, illustrate cooperative phenomena based on short-range order. Also the relation between lattice and continuous models is particularly clear in one dimension. We consider lattice approximants to a Takahashi model, which is a onedimensional continuous model in which M hard-rod molecules of length a lie on a line of total length L. The hard-rod property is introduced by putting ur = 1 for r a, where r is the distance between the centres of a pair of molecules. For r a, ur is piecewise continuous and ur = 0 for r 2a. The last condition ensures that interaction occurs only between successive molecules on the line. It is convenient to introduce a function vR = ur where R = r , a is the length of unoccupied line between the two rods see Fig. 5.2a. Clearly vR = 0 for R a.
5.2.1 The Lattice Gas

Now take a d = 1 lattice of N sites with nearest-neighbour site distance b where b = a=n, n being a positive integer and a measure of the ` neness' of the mesh. Place M molecules on the lattice with k-th neighbour interactions "k = ukb +0 where ur is the interaction energy function for the continuous model.3 Thus "1 = = "n,1 = 1, which ensures that the hard core of each molecule is a rod occupying n sites. Clearly we have an n , 1-th neighbour 3 The +0 symbol indicates that if r = kb is a point where ur has a discontinuity we take the limiting value of ur as r approaches kb from above.
122
5. Lattice Gases
a R r
a
2 b
Fig. 5.2. a The continuous one-dimensional model. b The one-dimensional
lattice model with each molecule occupying three sites. The left-hand boundary is shown with q0 = q1 = 2, q2 = 0, q3 = 4.
exclusion model in the sense of Sect. 5.1. Also, "k = 0 for k 2n. It is convenient to introduce new interaction parameters "0q where "0q = "k for q = k , n. Here q is the number of vacant sites between the two hard rod molecules and corresponds to the continuous variable R introduced above. In fact, "0q = vqb + 0 and "0q = 0 for q n. The microstate is completely speci ed by the set of integers q0 ; q1 ; : : : ; qM ,1 ; qM where qj is the number of vacant sites between the j th molecule from the left and the j + 1th molecule. The integers q0 and qM are the numbers of vacant sites between the rst and last molecules respectively and the boundaries see Fig. 5.2b. It is assumed that there is no interaction between the boundaries and the adjacent molecules. Then the number of sites N and the con gurational energy E are given in terms of the qj by
N = Mn + E=
M ,1
M X
X
j =1
j =0
qj ;
"0qj :
5.6
The constant-pressure distribution of Sect. 2.3.2, will be used, in which the e one-dimensional pressure P = Pb, with M and T independent variables. The number of lattice sites is now a uctuating quantity and according to the b convention prescribed in Sect. 2.2 will be speci ed as N . Then from 2.48, b b H = E + PN : 5.7 Since there is only one component, 2.48, 2.49 and 1.12 yield
Z = expM=T =
1 X NX ^ exp,E=T ; qj b N =M n
5.8
123
where the inner summation is over all values of qj compatible with a given b N and e = exp,Pb=T = exp,P=T : 5.9 The two stage summation of 5.8 is equivalent to summing all qj from 0 to 1. For j in the range 1 to M , 1 each qj yields an identical factor. After substituting the expressions for N and E given by 5.6 into 5.8 and using the condition "0q = 0 for q n this has the form
1 X
q=0
q exp,"0q =T =
n,1 X q=0
n q exp,"0q =T + 1 : ,
5.10
The partition function contains a product of M , 1 of these factors together with factors for q0 and qM , the numbers of vacant sites between the end rods and the boundaries. These each give the geometrical progression 1 Xi 1 = 1, : 5.11 i=0 Finally, the Mn term in the rst of equations 5.6 gives the factor M n and thus n Z = 1 , 2 M ,1 ;
n = n q exp,"0q =T + 1 : , q=0
"n,1 X
5.12
where the Boltzmann factors exp,"0q =T have been replaced by Mayer facb tors exp,"0q =T , 1. Using 5.14 we can average N over the constantpressure distribution to obtain `n , the mean length per molecule for a parb ticular value of n. In this case the equilibrium value of N can be obtained from 2.50. With 5.13 and 5.9 this gives b ` = Nb = ,Tb @ ln = b @ ln : 5.15
n
Since M ,1 ln n 1 , ,2 ,1 ! 0, as M ! 1, at constant P and T , G = M = ,T ln Z = ,MT ln ; 5.13 in the thermodynamic limit. An alternative form for is n,1 n X q exp,"0 =T , 1 + 1 5.14 = q 1, ; q=0
From 5.14 and 5.15, and noting that the constant a = bn,
@P
@
124
5. Lattice Gases
`n = a + 1 b + q=0 ,
n,1 X
q q , q + 1 exp,"0q =T , 1
n,1 X q=0
1 + 1 ,
q exp,"0q =T , 1
5.16
As P ! 1 at constant T the parameter ! 0 and, from 5.16, `n ! a, which means that the hard-core rods are tightly packed on the lattice. For an assembly with hard-core interactions only, all "0q = 0 and the last term of 5.16 disappears. For the general case the rst and second terms on the right-hand side of 5.16 represent hard-core e ects, whereas the last term results from interactions outside the hard core.
5.2.2 The Continuum Limit

The continuum limit is attained by taking n ! 1 and b = a=n ! 0 with a e being kept constant. For this b must be shown explicitly, by using P rather than P and, from 5.9, for small b e 1 , = Pb=T + Ob2 : 5.17 The sums in 5.16 become Riemann integrals as b ! 0 and 5.18 nlim `n = `c ; !1 where Za e e T + 0 PR=T , 1 exp,PR=T exp,vR=T , 1 dR : Z `c = a + e P 1 + P =T a exp,PR=T exp,vR=T , 1 dR e e 0 5.19 This is a form of the equation of state for the Takahashi model described at the beginning of this section Lieb and Mattis 1966, Mattis 1993. Thus it has been shown that, for rods of xed length, the equation of state for the lattice approximation goes over to that of the continuous model when the lattice mesh becomes in nitely ne. It can be shown that the same limiting form is attained if, instead of using "0q = vqb + 0, the averaged value Z q+1b vRdR 5.20 "0 = 1
q
is used. In fact, the convergence is then frequently faster Bell 1980. For a Tonks gas, which is the case where there is the hard-core interaction only, vR = 0, and the nal term in 5.19 disappears giving
b qb
e P=` T a: c,
5.21
5.3 The Simple Lattice Gas and the Ising Model
125
If the attractive term c=v2 is omitted from the van der Waals equation 1.43 ~ then it becomes identical in form to 5.21, with the volumes v and b taking ~ the place of the lengths `c and a respectively. It follows that the van der Waals treatment of hard-core terms is accurate in one dimension. A density can be de ned by
as n ! 1 ; 5.22 = nM = à ! à N n c with = 1, its maximum value, at closest packing. From Sect. 3.2 we can treat the Tonks gas as a reference model and incorporate a Lebowitz Penrosetype long-range interaction by adding a term ,c=`2 to the right-hand side c of 5.21. The resulting equation is then identical in form to that of van der Waals.
5.3 The Simple Lattice Gas and the Ising Model

Consider the simple lattice gas, described in Sect. 5.1, putting "1 = ,", " 0. A microstate can be speci ed by the N quantities i , where i = 1 or ,1 according to whether site i is occupied or vacant. Using the grand distribution of Sect. 2.3.3, the con gurational Hamiltonian for this one-component assembly is, from 2.51,
b 1 H = ,4"
n:n: X
It will be seen that there is a contribution to the rst summation only from terms where i = j = 1. To establish the equivalence with the ferromagnetic Ising model, use the fact that there are in all 1 zN nearest-neighbour site pairs 2 and rewrite 5.23 as 1 b 1 H = , 8 N 4 + z" , 4 "
n:n: X
fi;jg
1 X i + 1 j + 1 , 2 i + 1 :
fig
5.23
Apart from the constant term this is the Hamiltonian 2.66 for a ferromagnetic Ising model, the positive and negative spins respectively corresponding to the molecules and holes in the lattice gas. Formally we can write, for the equivalent Ising model, Je = 1 " ; He = 1 2 + z" ; 5.25 4 4 5.26 where the last term in 5.26 denotes the Ising model Hamiltonian with J = Je and H = He, as given by 5.25. It follows from 2.51 that the grand
fi;jg
i j , 4 2 + z"
X
fig
i:
5.24
b b 8 H = , 1 N 4 + z" + H i:m: ;
126
5. Lattice Gases
partition function for the lattice gas is related to the constant magnetic eld partition function for the ferromagnetic Ising model by + Z N; T; = exp N 4T z" Z i:m: N; T; H: 5.27 8 Unlike the ferromagnet antiferromagnet equivalence, which is con ned to H = 0, this applies for all H. Thermodynamic functions for the lattice gas can now be obtained from those for the Ising model. From 1.19, 2.52, 5.27 and 5.2 1 P = 8 4 + z" + N ,1 T ln Z i:m: N; T; He 1 5.28 = 8 4 + z" , f i:m: T; He ; where f i:m: is the constant magnetic eld free energy per site for the ferromagnetic Ising model. The uid number density satis es the relations 1 = M=N = 1 1 + h i i = 2 1 + m ; 5.29 2 where M is the equilibrium number of lattice gas molecules at the given values of T and , and m is the relative magnetization in the equivalent Ising model. As an illustration, we show that 5.28 is satis ed by the explicit expressions previously obtained for the one-dimensional Ising model and lattice gas respectively. Using the notation of Sect. 5.2, the simple lattice gas case corresponds to putting n = 1 b = a and "00 = "1 = ,". Equations 5.9, 5.12 and 5.13 then yield a relation between , P and T in the form
= exp,=T = exp"=T + 1 , ;
5.30
= exp,P=T : 5.31 To obtain P in terms of and T , 5.30 is rst expressed as a quadratic

equation
,2 , + exp"=T ,1 + exp"=T , 1 = 0 for ,1 . Solving and choosing the larger root to ensure that P
With z = 2, 5.25 gives = X4 =Z2 ; exp"=T = X4 ; 5.34 where, as in 2.75, X = expJe =T , Z = expHe=T . Substitution into 5.33 then gives
1 expP=T = 21 exp"=T + + f exp"=T , 2 + 4 g 2 :
5.32 0 gives 5.33
5.4 Phase Separation in the Simple Lattice Gas
5.35 4T 1 ; where 1 is de ned by 2.94. By 2.96 this is equivalent to the relation 5.28 in the d = 1 case. In Example 5.3 the reader is asked to verify 5.29 for d = 1.
expP=T = ZX,1 1 = exp
2 + "
127
5.4 Phase Separation in the Simple Lattice Gas

We now consider the simple lattice gas for d = 2; 3 and in particular the signi cance of the rst-order transition line on the H = 0 axis in the T; H plane of the equivalent Ising model see Fig. 3.3. De ning 1 4 = + 2 z" = 2Hc ; 5.36 the internal state of the uid depends on the values of T and 4, which may be regarded as de ning a point in a T; 4 plane. By 5.36 the 4 = 0 axis is equivalent to the H = 0 axis in the T; H plane of the Ising model. From 5.29 5.37 lim = 1 1 ms ; 2 4!0 for T Tc, where ms is the Ising model spontaneous magnetization de ned by 3.11. There is thus a discontinuous density increase, equal to ms , as the 4 = 0 axis is crossed from below to above. From the Ising model symmetry
1:0 2T 3" 0:5
1 2 3
Fig. 5.3. Plane triangu0 0:5 1:0
lar lattice gas vapour liquid coexistence curves: 1 meaneld 2 rstorder pair 3 exact.
128
5. Lattice Gases
so that the phase equilibrium conditions 1.41 are satis ed along the interval 0; Tc of the 4 = 0 axis, where is discontinuous, Tc being the Ising model 1 1 critical temperature with J = 4 ". For T Tc, = 2 on the 4 = 0 axis by 5.29 and all thermodynamic quantities are continuous. It follows that Tc is the critical temperature for separation into a high-density liquid and a low-density vapour phase and the interval 0; Tc on the 4 = 0 axis is a rst-order transition line. The condition 4 = 0 de nes the critical isochore and the critical density is c = 1 . In Fig. 5.3 the conjugate phase densities 2 for the triangular lattice z = 6 are shown as functions of T using the meaneld approximation, the rst-order pair approximation see Chap. 7 and the exact solution see Chap. 8 respectively. From 5.37 the coexistence curves 1 in the ; T plane are symmetrical about the line = 2 . Since 4 = 0 at phase equilibrium the corresponding pressure, by 5.28, is 5.40 P = , 1 z" , f I:M:T; 0 : 8 As T ! 0 the Ising model ferromagnet tends to complete spin alignment and 1 1 hence f I:M: ! , 2 zJ . Since Jc = 4 ", this implies P ! 0, as should be the case for a physically reasonable model.
relation 2.68 and from 5.28 a pressure symmetry relation for the uid can be deduced in the form P T; 4 = P T; ,4 : 5.38 From this relation lim P T; 4 = 4!,0 P T; 4 ; lim 5.39 4!+0
5.5 A One-Dimensional Water-Like Model

The existence of a density maximum at 4 C in liquid water is due to the persistence above the melting point of regions of open structure resembling that of ice I see Appendix A.3. As this is a short-range order e ect it is possible to simulate it in one dimension. Although the directional character of hydrogen bonding cannot be built into a one-dimensional model there is an open structure stable at low pressures if the con guration with minimum enthalpy has a density de ned by 5.22 which is less than unity. This is su cient to give density maxima on certain isobars, and it is of interest to compare results for lattice and continuous models. Several òpen structure' models were considered by Bell 1969, 1980. The one discussed here is the displaced parabolic well model with the continuous potential function shown in Fig. 5.4 and de ned by 8 0; 0 R 1 a, a R , 2 5.41 vR = 1 :, 8V20 2R , aa , R ; 2 a R a : a
129
vR a=2
3a=4
Fig. 5.4. Displaced parabolic well potential.
,V0
3 The minimum energy is ,V0 at R = 4 a. Now consider the ground state, which is the con guration of least enthalpy, towards which the system tends as T ! 0. If the separation between successive centres in the ground state is a + R then, since the constant-pressure distribution is used, the enthalpy per molecule is e hR = a + RP + vR : 5.42 e The least value of hR for given P may occur either at R = 0, which gives a close-packed ground state, or at some point R = Rmin, where there is a minimum de ned by dh = 0; d2 h 0 : 5.43 dR dR2 From 5.41 a minimum can occur only in the range 1 a R a where, by 2 5.42, dh = P + 8V0 4R , 3a ; e 2 dR a 5.44 d2 h = 32V0 : dR2 a2 Substitution of 5.44 into 5.43 yields e Pa2 Rmin = 3 a , 32V ; 4 0 5.45 8V0 a2 , 2R2 + Pa : e hRmin = a2 min
e Since, for consistency, Rmin 1 a, a minimum exists only for P in the range 2 0; 8V0 =a. However, the pressure range where hRmin is the least value of hR is much more restricted. From the second relation of 5.45, hRmin p e h0 = Pa, when Rmin a= 2. There will be an open ground state with R =
130
From the rst relation of 5.45, p e 5.46 P0 = 8V0 =a3 , 2 2 = 1:3726V0 =a : The length per molecule ` at T = 0 is a + Rmin in thepopenp ground state e and thus decreases from ` = 1:75a at P = 0 to ` = 1 + 2a= 2 = 1:7071a e e at P = P0 . Then there is a discontinuous decrease to the close-packed value e e ` = a, which applies for all P P0 . For T 0, values of `, and hence of the density , can be obtained by substituting 5.41 into 5.18 and 5.19. e e It can be shown that, for any P in the range 0; P0 , ` decreases initially as T increases from zero Bell 1969. Since ` must eventually increase with T , there is a maximum on any density isobar in the open structure pressure e range 0; P0 . In view of the behaviour of vR in the two halves of the range 0 R a, an even value of n is necessary for lattice approximations to the displaced well model. The simplest case is n = 2 and, using the averaging procedure of equation 5.20, 2 "00 = 0 ; "01 = , 3 V0 : 5.47 There are only two possibilities for the ground state. One is close packing with q = 0 for all successive pairs of molecules, and the other is the open minimum energy con guration with q = 1. The molecular enthalpies for the two states are respectively e h0 = Pa ; 5.48 3e 2 e h1 = P a + b + "01 = 2 Pa , 3 V0 : 3 e 4 For Pa 3 V0 , h0 h1 and the open structure with ` = 2 a is stable at 4 V0 , h1 h0 and the close-packed e T = 0. On the other hand, for Pa 3 state with ` = a is stable at T = 0. Hence there is a discontinuous decrease e 3 e from `2 = 3 a to `2 = a at P = 4 V0 =a, which corresponds to P0 in the 2 continuous model. For T 0, putting n = 2 and, correspondingly, b = 1 a, 5.16 yields, 2 after substitution from 5.47, a , 2 exp2V0 1 `2 = a + 21a + 2f1 1 1 exp2V=3TT ,, 1 g ; 5.49 , + , 0 =3 where e = exp,Pa=2T : 5.50 2 V0 =T ! 0 or ! 1 e Now ! 0 as T ! 0 at xed P , whereas exp 3 3 e 4 V0 or Pa 4 V0. Hence `2 ! 2 a as T ! 0 e 3 according to whether Pa 3 4 V0 and `2 ! a as T ! 0 for Pa 4 V0 , agreeing with the e e 3 for Pa 3 ground-state analysis. After a little manipulation, 5.49 becomes
e e Rmin for the pressure range 0; P0 where P0 corresponds to Rmin = a= 2.
5. Lattice Gases
1:0 0:8 0:6 0:4
131
n=2 n=8 n=1

0:0 0:2 0:4 0:6 T=V0 0:8 1:0
e Fig. 5.5. Temperature density isobars for P = V0 =a with the potential of Fig. 5.4.
1:0 0:8 0:6 0:4
n denotes number of sites occupied by hard core of one molecule. Reprinted from Bell 1980, by permission of the publisher IOP Publishing Ltd.
n=8 n=1
n=2
0:0
0:2
0:4
0:6 T=V0 0:8
1:2
of the publisher IOP Publishing Ltd.
e 3 Fig. 5.6. As Fig. 5.5 with P = 2 V0=a. Reprinted from Bell 1980, by permission
1 2 5.51 `2 = 3 a , 2 1 , f1 + 1a, , exp2V =3T , 1 g : 2 0 Hence, when `2 = 3 a at T = 0, there is a temperature range starting at T = 0 2 where `2 3 a. Thus the density initially increases as T is increased from 0 2 e 3 for constant P 4 V0 =a. From 5.51, `2 regains its T = 0 value of 3 a when e e 2 = 1, which corresponds to T = Pa= ln4. For a pressure P in the open 4 structure range 0; 3 V0 , the density has a maximum at some temperature e T Pa= ln4.
132
5. Lattice Gases
Lattice approximations can also be obtained for higher n values. In Figs. 5.5 and 5.6 density isobars calculated for n = 8, using the averaging relation 5.20 for "0q , are compared with density isobars calculated for the extreme cases n = 2 and n = 1 continuous uid. The curves in Fig. 5.5 are e for P = V0 =a which is in the open structure range in all three cases whereas e 2 those in Fig. 5.6 are for P = 3 V0 =a which is outside the open structure range. It can be seen that a density maximum on an isobar is associated with an open ground state at the same pressure. The curves for lattice approximations are qualitatively similar to those for the continuous uid, and for n = 8 are quite close quantitatively. Lattice gas water models for d = 3 are considered in Sects. 7.5 and 9.12.
Examples
5.1 Using the methods of Sect. 3.3, show that the meaneld Helmholtz free energy per site for the simple lattice gas is aT; = T ln + 1 , ln1 , , 1 z" 2 : 2 By 5.5 or otherwise, nd the critical point by solving the equations Show that 5.25 and 5.28 are satis ed in the meaneld approxi1 mation and that = , 2 z" for conjugate phases. 5.2 A continuous one-dimensional square well attractive potential model is de ned, using the terminology of Sect. 5.2 and with " 0, by the relations vR = ,;" ; 0 R . a , 0 a R Show that the equation of state for the nth lattice approximant is a `n = a + e nfexp Pa=nT , 1g
@P = 0 ; @
@2P = 0 : @2
e a exp"=T , 1 exp,Pa=T : e 1 + exp"=T , 1 1 , exp,Pa=T Given that `c denotes the length per molecule in the original contin,
uous model, show that
a `c , `n = T , : e e P nfexp Pa=nT , 1g Hence prove that `c `n for all n.
Examples
133
5.3 Using the Gibbs-Duhem relation 1.16, show that, for a d = 1 lattice gas at constant temperature, = a @ ; T where the density is de ned by 5.22. Using 5.33 or otherwise, show that for the simple d = 1 lattice gas 1 exp"=T , exp,=T =2+ : 2f exp"=T , exp,=T 2 + 4 exp,=T g1=2 1 Hence show that = 2 1+ m, where m is the relative magnetization of the equivalent ferromagnetic Ising model. 5.4 A continuous hard shoulder model is like the model of question 2, except that ," is changed to " " 0 so that the interaction energy of rods with a gap R a between them is repulsive. Write down an e expression for `n and show that, as T=" ! 0 at xed P , `n ! a for e e Pa " and `n ! 2a for Pa ". Explain this behaviour in terms of the enthalpy of the ground state.
e! @P
6. Solid Mixtures and the Dilute Ising Model
6.1 The Restricted Grand Partition Function

We consider an isotropic solid mixture of Ma atoms of component A and Mb e of component B in volume V . This is modelled by a regular lattice of N sites, the volume v0 per lattice site being as usual taken as constant and used as the standard volume in 1.54. Each site is occupied by an atom so that e M = Ma + Mb = V =v0 = V = N : 6.1 This is a restricted system in the sense of Sect. 2.3.4 with n = 2 and Q1 = Ma , 1 = a , b , Q2 = N and 2 = b , P . A partition function of type 2.58 is obtained by taking N , T and b , b as independent variables with X b Z= exp,H=T ; fMa +Mb =M g 6.2 b = E , a , bMa ; H where E is the con gurational energy. This is known as the restricted grand partition function. The free energy f per site is equal to ,T ln Z=N for large N and, from 2.59, f = a , a , b xa = b , P ; 6.3 where the mole fraction xa = Ma =M = Ma=N is the density conjugate to a , b and a = u , Ts is the Helmholtz free energy per site. From 2.60, df = ,sdT , xa da , b : 6.4 The condition for phase equilibrium is that T , a , b and f have the same values in the conjugate phases. It follows from 6.4 that, on a phase equilibrium line in the A , B ; T plane, 4sdT + 4xa da , b = 0 ; 6.5 where 4s and 4xa denote di erences between the two phases. Hence, at the point where the composition of the two phases is the same 4xa = 0 dT=da ,b = 0, so that the phase equilibrium curve has a stationary value. By a similar argument to that in Sect. 1.9.2 the conjugate phase curves in the xa ; T plane touch at the azeotropic point where 4xa = 0 and their common
136
tangent there is horizontal, as in Fig. 1.7. Such behaviour is known both experimentally and theoretically when rst-order transitions to a sublattice ordered state occur, although the term àzeotrope' is not usually used for solid mixtures.
6.2 Binary Mixtures

In the lattice mixture we associate a variable i with each site i where i = 1 or ,1 respectively according to whether the site is occupied by an A or a B atom. For a nearest-neighbour site pair i; j the expression 1 1 + i 1 + j 4 is equal to 1 if both sites are occupied by A atoms and is zero otherwise. With similar expressions for B B or A B occupation, 1X 1X Mb = 2 1 , i ; Ma = 2 1 + i ;
fig
1 Naa = 4 1 Nbb = 4
n:n: X
fig
n:n: X
fi;jg
1 + i 1 + j ; 6.6
Nab = 1 4
n:n: X
fi;jg
1 , i 1 , j ; 1 + i 1 , j + 1 , i 1 + j

n:n: X
fi;jg
= 1 zN , 1 4 2
where Naa , Nbb and Nab respectively denote the numbers of AA, BB and AB nearest-neighbour pairs. Taking the con gurational energy as a sum of nearest-neighbour pair interaction terms gives E = Naa "aa + Nbb "bb + Nab "ab
fi;jg
i j;
X 1 1 = 8 zN"0 + 1 z"aa , "bb i , 4" 4

where
fig
n:n: X
fi;jg
i j;
6.7
"0 = 2"ab + "aa + "bb ; " = 2"ab , "aa , "bb :
6.8
From 6.6, 6.7 can be expressed in the alternative forms
6.2 Binary Mixtures
137
1 E = 2 z"aa Ma + "bb Mb + 1 "Nab ; 2 1 E = 2 z"bb N + z"ab , "bb Ma , "Naa :
6.9 6.10
6.2.1 The Equivalence to the Ising Model

Substituting 6.7 into the expression 6.2 for the restricted grand Hamiltob nian H and comparison with 2.66 yields b b 1 6.11 H = 8 zN"0 , 1 N a , b + H i:m: ; 2 b where H i:m: is the Hamiltonian of the equivalent Ising model which depends on parameters Je and He de ned by 1 1 Je = 4 " ; He = 2 4 ; 6.12 6.13 4 = a , b , 1 z"aa , "bb : 2 If " is positive then the + sign applies and the equivalent Ising model is ferromagnetic, whereas if " is negative the , sign applies and the equivalent Ising model is antiferromagnetic. From 6.11 the restricted grand partition function satis es the relation Z N; T; 4 = exp fN z"0 , 4a , b =8T g Z i:m: N; T; He 6.14 and the free energy per site is given in terms of the constant magnetic eld free energy per site of the Ising model by 1 f T; 4 = 1 z"0 , 2 a , b + f i:m:T; He : 6.15 8 Since A and B atoms correspond respectively to i = 1 and i = ,1 the mole fractions in the solid mixture are given in terms of the relative magnetization m of the equivalent Ising model by xa = 1 + m ; xb = 1 , m : 6.16 2 2 Now assume that 1 6.17 "ab 2 "aa + "bb or equivalently " 0 implying that AA or BB nearest-neighbour pairs are increasingly preferred to AB pairs as the temperature is reduced. From 6.12 the solid mixture is equivalent to an Ising model ferromagnet at the same temperature with 1 J = Je = 1 " and H = He = 2 4. From 6.13 and 6.15 two solid mixture 4
138
phases have the same values of a , b and f when the corresponding Ising model phases have the same values of f i:m: and He. Thus, if Tc is the Ising ferromagnet critical temperature when J = 1 ", the interval 0 T Tc on 4 the T -axis in the T; 4 plane for the solid mixture is a rst-order transition line. It follows from 6.16 that at each temperature in this interval an A rich 1 1 phase with xa = 2 1+ms is conjugate to a B rich phase with xa = 2 1,ms , where ms is the Ising ferromagnet spontaneous magnetization.
6.2.2 The Equivalence to a Lattice Gas

Comparing 6.12 and 6.13 with 5.25, a solid mixture and a lattice gas at the same temperature are equivalent when " has the same value for both models and = a , b , z"ab , "bb ; 6.18 provided 6.17 applies. Equating the expressions for f i:m: T; H given by 6.15 and 5.28 respectively and, using 6.18, ,Pg = f T; 4 , 1 z"bb ; 6.19 2 where f is the free energy per site in the solid mixture and Pg the pressure eld in the lattice gas. The subscript g distinguishes Pg from the pressure eld in the solid mixture. Again, from 6.16 and 5.29 the number density in the lattice gas and the mole fractions in the solid mixture are connected by xa = ; xb = 1 , : 6.20 Hence the conjugate phase curves shown in Fig. 5.3 apply for the solid mixture if is replaced by xa .
6.3 Order Disorder on Loose-Packed Lattices

Now consider a solid mixture on a loose-packed lattice see Sect. 4.2 and Appendix A.1 with the inequality 6.17 reversed so that 1 6.21 "ab 2 "aa + "bb or equivalently " 0: There is now a tendency to the formation of AB pairs which, at low temperatures, results in sublattice ordering with A atoms preferentially occupying one sublattice and the B atoms the other. This is observed in certain real systems such as body-centred cubic equimolar xa = xb = 1 FeAl where, 2 below a critical temperature, the Fe tends to concentrate on one simple cubic sublattice and the Al on the other see Appendix A.1. Similar behaviour is found in body-centred cubic CuZn brass. For an order-disorder transition to occur, the energy of cohesion of the lattice must be signi cantly
6.3 Order Disorder on Loose-Packed Lattices
139
where ma and mb are the sublattice relative magnetizations. For the antiferromagnet we know from Sect. 4.2 that the overall relative magnetization m = 0, for all T , when H = 0. Hence by 6.16 the corresponding condition 1 4 = 0 gives xa = xb = 2 for all T . For T Tc , ma = mb = 0 in the antiferromagnet, giving the disordered state xaa = xab = xba = xbb = 1 . 2 For T Tc, 6.22 and for de niteness, the rst set of relations of 4.10 yield the ordered state xaa = xbb = 1 +2ms ; 6.23 1 , ms : xab = xba = 2 The change of sign ms ! ,ms corresponds to an interchange of the equal sublattices a and b. As T ! 0 the ferromagnet spontaneous magnetization ms ! 1 and the solid solution approaches a perfectly ordered ground state with all a sites occupied by A atoms and all b sites by B atoms. One of the best-known properties of real alloys is the heat capacity anomaly at the order-disorder transition temperature. From the ferromagnet antiferromagnet equivalence at zero eld and the antiferromagnet solid solu1 tion equivalence, the heat capacity for this model with xa = xb = 2 is equal to that of the Ising ferromagnet at H = 0. Exact results for the d = 2 case of the latter are shown in Fig. 8.6 and a meaneld plot is given in Fig. 3.5. In the context of binary alloys the parameter s = Nb =Na introduced in 4.61 is termed the stoichiometric ratio and a mixture for which xb =xa = s is called stoichiometric. For xb =xa = s, a state of perfect order with all A atoms on sublattice a and all B atoms on sublattice b is possible. For equal sublattices, s = 1 and xb =xa = 1 in a stoichiometric mixture, which thus corresponds to 4 = 0 or He = 0 in the equivalent Ising model. Nonstoichiometric mixtures xb =xa 6= 1 correspond to 1 4 = He 6= 0, and no 2
larger than the energy of sublattice ordering. Otherwise the latter will persist up to the melting point of the solid, as in intermetallic compounds. In terms of our parameters this means that j"0 j must be large compared to j"j. The rst quantitative theory of sublattice ordering was due to Bragg and Williams 1934 and is equivalent to the meaneld treatment of Sect. 6.4 for the particular case of a loose-packed lattice with xa = xb . When 6.21 applies, the equivalent Ising model is the simple antiferro1 magnet, with Je = 4 j"j 0. Denote the fractions of A and B on sublattice a in the solid mixture by xaa and xba respectively and similarly de ne xab and xbb . Then since A and B correspond respectively to positive and negative spins in the antiferromagnet, xba = 1 , xaa = 1 ,2ma ; xaa = 1 +2ma ; 6.22 1 + mb ; 1 , mb ; xa b = 2 xbb = 1 , xab = 2
140 1.0
T Tc
0.5
Sublattice order
Fig. 6.1. The critical curve for sublattice ordering on a loose-packed lattice in the meaneld approximation.
0.5
xa
1.0
exact results are available even for d = 2, so that approximation or series methods must be used. From 6.22 the antiferromagnetic state ma 6= mb corresponds to sublattice order in the solid solution and the paramagnetic state ma = mb = m to disorder. Since the equivalent Ising model is the simple antiferromagnet with = 0 the order-disorder transition is second-order over its whole temperature range and corresponds to the critical curve for the antiferromagnet. A meaneld approximation to the order-disorder transition curve in the xa ; T plane can be obtained from 6.16 and the second relation of Example 4.2, with = 0, p xa = 1 1 1 , ; 2 6.24 T = T = 4T : = This curve, which is clearly symmetrical about xa = 1 , is shown in Fig. 6.1. 2 Every point in the xa ; plane corresponds to an equilibrium state, and the 1 map of the antiferromagnet H = 0 axis is the entire xa = 2 line from T = 1 to T = 0. This contrasts with the super cially similar Fig. 5.3 where the map of the ferromagnetic H = 0 curve bifurcates at the critical point into the right- and left-hand branches of the coexistence curve, and the entire area inside the latter represents non-equilibrium states. We can ask if other forms of sublattice order are possible when xa =xb is far from 1. The body-centred cubic lattice can, for instance, be split into four equal sublattices by dividing the simple cubic sublattice a into two facecentred cubic arrays a1 and a2 and similarly dividing b into b1 and b2 see Fig. 6.2. There may thus be a type of order with stoichiometric ratio s = 3 in which the a1 sites say are preferentially occupied by A and the rest of the body-centred cubic lattice by B. However, this will not occur if there 1 are nearest-neighbour interactions only. Suppose that xa 2 and that all A
Tc
zJe
zj"j
b2 b1 a1 b1
b2
6.4 The Order Parameter and Landau Expansion 141 b1 Fig. 6.2. Division of the body-centred cubic lattice into four face-centred cubic sublattices. a2
b2
b2
b1
atoms are on a sites. Then all the nearest-neighbour sites to any a site are occupied by B atoms and Nab = zMa = 8xa N . By 6.9 the con gurational energy is thus the same for all distributions of the A atoms on the a sites and there is no advantage in the A atoms concentrating on the a1 face-centred cubic sublattice. The degeneracy is removed if there are second-neighbour interactions obeying an inequality similar to 6.21. Experimentally, an ordered state with the Al atoms occupying one face-centred cubic sublattice is found near the composition Fe3 Al for the body-centred cubic Fe Al alloy. However, this is not so for other body-centred cubic alloys such as Fe Co and Fe Cr, the di erence presumably depending on the strength of the e ective secondneighbour interaction. A meaneld theory incorporating second-neighbour interactions on the body-centred cubic lattice and leading to 1:3 as well as 1:1 ordering has been developed Elcock 1958. Other cases when xa =xb 6= 1 are considered in subsequent sections.
6.4 The Order Parameter and Landau Expansion

We now consider the possibility of unequal sublattices and intra-sublattice nearest-neighbour pairs using the nomenclature of Sect. 4.5, so that 4.60 and 4.61 apply. As in Sect. 4.5 we assume s 1. An order parameter is de ned by xaa = xa 1 + s ; xba = 1 , xaa = xb , xa s ; 6.25 xab = xa 1 , ; xbb = 1 , xab = xb + xa : The necessary conditions xb = xba1+ sxbb 6.26 xa = xaa1+ sxab ; +s +s are clearly satis ed. In a stoichiometric mixture, xb =xa = s, giving 1 s xa = 1 + s ; xb = 1 + s : 6.27 For xa 1 + s,1 the range of possible values is ,s,1; 1 and for s1 + s,1 xa 1 + s,1 it is ,s,1 ; s,1 xb =xa . The disordered state xaa = xab = xa , xba = xbb = xb corresponds to = 0. For values of xb =xa in the
142
neighbourhood of the stoichiometric ratio the equilibrium value of can be expected to be positive at low temperatures, giving a concentration of A on a sites and B on b sites. A meaneld theory will now be developed. From Sect. 3.3 the probability of a pair of nearest-neighbour a sites being occupied by an A and a B is 2xaa xba and similarly for a nearest-neighbour pair of b sites. Slightly extending the theory, the probability of this occupation for a nearest-neighbour site pair consisting of one a site and one b site is xaa xbb + xab xba . Hence the expected total number of AB nearest-neighbour pairs is, using the last relation of 4.60, Nab = zaa Naxaa xba + zbb Nb xab xbb 1 + 2 zab Na + zba Nb xaa xbb + xab xba : 6.28 The number of con gurations is the product of expressions like 3.38 for the a and b sublattices respectively, with N replaced by Na in one expression and Nb in the other. In a similar way, from 3.40, the Helmholtz free energy per site is given by a = N ,1T fNa xaa ln xaa + xba ln xba + Nb xab ln xab + xbb ln xbb g 6.29 + 1 z"aa xa + "bb xb , 1 j"jN ,1 Nab ; 2 2 where the expression 6.9 for the con gurational energy is used. Since is an order variable, a is a non-equilibrium free energy which must be minimized with respect to . Using 4.60 and 4.61 a can be expressed in the form a = a0 + 4a ; 6.30 where a0 = T xa ln xa + xb ln xb + 1 z"aa xa + "bb xb 2
, 1 zj"jxa xb ; 2
T 4a = 1 + s fxa 1 + s ln1 + s + xb 1 , ys ln1 , ys
+ s xa 1 , ln1 , + xb 1 + y ln1 + y g with 4a0 = 0 and the mole ratio parameter y = xa =xb :
6.31
, 1 j"jsx2 z , zaa , zbb 2 a 2
6.32 6.33
6.4 The Order Parameter and Landau Expansion
143
For a stoichiometric mixture, y = 1=s. The equilibrium value of satis es the relation @a = @ 4a
1 + s1 + y = Tsxa ln 1 , 1 , ys , j"jsx2 z , zaa , zbb a 1+s =0: 6.34 One solution is always = 0 and other solutions with 6= 0 appear at low temperatures if z zaa + zbb . Landau expansions are described in detail in Volume 2, Chap. 3. In the case considered here the Landau expansion for the Helmholtz free energy density a is obtained by expanding 4a in powers of the order parameter . From 6.32, sx 4a = 2xa T , j"jz , zaa , zbb xa xb 2
b
1 + y,yn,2 1 + s,sn,2 n : 6.35 nn , 1 n=3 The essence of the Landau method is the deduction of critical properties from the behaviour of the rst few coe cients, and this can be illustrated by the loose-packed lattice case. With equal sublattices, s = 1 and the odd powers of disappear in 6.35. This is to be expected since, when s = 1, the expression on the right of 6.32 becomes an even function of . For loose-packed lattices we have, in addition, zaa = zbb = 0 and 6.35 becomes 4a = B2 T 2 + B4 T 4 + B6 T 6 + ; 6.36 where 8 xa 2xb T , zj"jxa xb ; n = 1 , B2n T = Tx 1 + y2n,1 6.37 : 2an2n , 1 ; n 1 . When, as here, the coe cients B2n T for n 1 are intrinsically positive there is a second-order transition at the temperature where B2 T = 0. This is because 4a has a unique minimum at = 0 for B2 T 0, but for B2 T 0 there is a maximum at = 0 and symmetrically placed minima on either side. For the latter case the plot of 4a against is similar to the curve ABOCD in Fig. 3.1. From 6.37, the condition B2 T = 0 yields the critical curve relation T = zj"jxa xb = zj"jxa 1 , xa ; 6.38 + Tsxa 1+s
1 X
144
which is equivalent to 6.24. Comparison of 6.25 with s = y = 1 and 6.23 shows that for an equimolar mixture xa = xb = 1 the order parameter 2 = ms . It follows that where 4, as de ned by 6.13, is zero then = 4x ; 6.39 where 4x = jxa , xb j for either of the conjugate phases in the solid mixture model of Sect. 6.2 with the same values of j"j and T . A case where the sublattices are unsymmetrical so that s 6= 1 and the odd powers of in 6.35 do not disappear is discussed in the next section. z y
a b b
Fig. 6.3. Site of face-centred cubic sublattice a and four

of its twelve nearest neighbours in sublattice b.
6.5 First-Order Sublattice Transitions

We now consider the face-centred cubic lattice see Appendix A.1. This divides into four equal simple cubic sublattices, any site in one of the latter having four nearest-neighbour sites in each of the other three which are thus completely equivalent with respect to the rst sublattice.1 Now attach the label a to the sites in one simple cubic sublattice and the label b to all other sites in the face-centred cubic. Hence Nb = 3Na ; s = Nb =Na = 3 ; zaa = 0 ; zab = 12 ; 6.40 zbb = 8 ; zba = 4 : Figure 6.3 shows an a site at the origin O and four nearest-neighbour b sites in the plane Oxz . There are similar sets of four nearest-neighbour b sites in the planes Oxy and Oyz . We shall study the transition from the disordered state = 0 to the ordered states with 0. Substituting 6.40 into 6.35, 4a = B2 T 2 + B3 T 3 + B4 T 4 + ; 6.41 where
1
This equivalence does not apply when a loose-packed lattice like the body-centred cubic is divided into four equal sublattices. From Fig. 6.2 it can be seen that an a1 site has four nearest neighbours in b1 and four in b2 but none in a2 .
6.5 First-Order Sublattice Transitions

5 4 3 2
145
Fig. 6.4. Free energy
4a
0
0
isotherms plotted against the order parameter for the unsymmetrical case.
1
B3 T = ,xa T 1 , y2 ;
3 B2 T = 2 y T , 4j"jxa xb ; 6.42
B4 T = 14 xa T 1 + y3 : 8 At the temperature T = 4j"jxa xb , where B2 T = 0, the leading term is B3 T 3 , where B3 T 0 for y 1, and 4a has the form of curve 1 on Fig. 6.4. The equilibrium value of corresponds to the minimum of 4a,
and the transition occurs at a higher temperature. Curves 2, 3, 4 and 5 are for progressively increasing temperatures above 4j"jxa xb . On curve 2 the minimum at = 0 represents a metastable state whereas on curve 4 the minimum at 0 represents a metastable state. The intermediate curve 3 corresponds to the transition temperature T0 . Here the minima at = 0 and = 0 both lie on the -axis, and the transition is a rst-order one from 2 = 0 to = 0 , with a discontinuous change ,6j"jx2 0 in the con gurational a energy per site, by 6.32 and 6.40. For a given xa the values of 0 and T0 are determined by the relations @ 4a0 = 0 : 6.43 4a = 0 ;
0 1 For the stoichiometric ratio y = 3 xa = 1 , xb = 3 , B2 T = 0 at 3 j"j. Equation 6.43 yields T0 = 0:8224j"j and 04 = 0:463, Fowler T = 4 and Guggenheim 1949, Sect. 1325. We have discussed the behaviour of the system as T varies at constant xa . Now, the equilibrium Helmholtz free energy per site aT; xa depends on the single density variable xa since xb = 1 , xa and hence is a free energy density fn,2 in the sense of Chap. 1. By considering the variation of aT; xa
@0
146
Fig. 6.5. Free energy-mole fraction

isotherm showing convex envelope.
D C E
xa D
xa E
xa
with xa at constant T we shall show, using the method of Sect. 1.10 iii, that the rst-order transition occurs in a range of xa which is unstable with respect to phase separation. We can express the equilibrium 4aT; xa as 4aT; xa; e , where = eT; xa is the equilibrium value of , that is the value of at the lowest minimum of the plot of 4aT; xa; against . Hence, at constant T ,
@ 4a = @ 4a + @ 4a @ = @ 4a ; 6.44 @xa @xa e @xa @xa @xa e since @ 4a=@ = 0 at equilibrium. It follows that @a = @a0 + @ 4a ; 6.45 @xa @xa @xa e where a0 is given by 6.31. Suppose that xa increases from zero at the xed temperature T = T0 given by 6.43. For small values of xa , e = 0 and hence 4a = 0 so that @a=@xa = @a0=@xa . At the particular value of xa implied in 6.43, e jumps from zero to 0 and @a=@xa undergoes a discontinuous change equal to @ 4a=@xa=0 . This quantity is negative, since the e ect of ordering is to reduce a, and hence the plot of a against xa is as shown in
Fig. 6.5, where the rst-order transition occurs at point C see also Example 6.3. To complete the convex envelope, a double tangent DE must be drawn, and for values of xa between xa D and xa E there is separation into a disordered phase where xa = xa D and an ordered phase where xa = xa E. Conjugate phases in the xa ; T plane are shown in Fig. 6.6. Suppose that the temperature of a mixture with xa = x is reduced slowly enough for a equilibrium to be maintained.2 Then at T = T1 an ordered phase starts to separate out from the disordered mixture, and for T1 T T2 the system consists of ordered and disordered phases in equilibrium. For T T2 the equilibrium system is a homogeneous ordered phase. The temperature T0 at
2
In metallurgical terminology this is called annealed.
6.6 The Equilibrium Dilute Ising Model and Equivalent Models
147
T T1 T0 T2 x a xa
Order
Fig. 6.6. Conjugate phases in the mole

fraction temperature plane.
which the metastable homogeneous mixture would undergo a rst-order transition lies between T1 and T2 . Behaviour of this type is well-known experimentally. In fact, Irani 1972 remarks that òrder disorder transformations in almost every system, except perhaps the L20 structures3 are rst-order transitions' and `most systems exhibit a two-phase eld away from the stoichiometric composition'. There is a meaneld theory Shockley 1938 for the whole composition range on the face-centred cubic lattice, and our treatment of 1:3 ordering is part of this. However, there are discrepancies between experimental observations for the well-known Au Cu alloys and meaneld results. In particular, at the stoichiometric composition AuCu3 xa = 1 the observed phase separation 4 curves have a common maximum or azeotropic point, resembling point P in Fig. 1.7. In meaneld theory, on the other hand, the phase separation curves 1 have a common maximum only at xa = 2 . These discrepancies probably arise both from the simplicity of the model and from the crudity of the meaneld approximation. Ordering on the face-centred cubic lattice is discussed in terms of the rst-order approximation in Sect. 7.6.

We consider an important new set of mathematically equivalent models.
6.6.1 Model I: The Equilibrium Dilute Ising Model

Suppose that in a lattice A B mixture like that of Sect. 6.1 the A atoms carry Ising spins with ferromagnetic interaction. We then have an Ising model
3
Body-centred cubic lattice divided into equal interpenetrating sublattices.
148
ferromagnet diluted by a non-magnetic component B. Associating a spin variable i = 1 with each A atom4 the reduced magnetic moment
b X M= i;
a
b where the summation is over all sites occupied by A atoms and M = hMi is conjugate to the applied magnetic eld H. The system is a restricted one, with n = 3, and we put 1 = a , b ; 2 = H ; 3 = b , P : 6.47 A partition function of type 2.58 is obtained by taking N , T , a , b and H as independent variables with b H = ,"Naa + z"ab , "bb Ma , a , b Ma
,J
aX : ;n:n
fig
6.46
Here the non-magnetic part of the con gurational energy is expressed in the form 6.10, with the constant term 1 z"bb N omitted, and the sums are re2 spectively over nearest-neighbour pairs and single sites occupied by A atoms. It is assumed that J 0 and that 6.17 applies. The partition function can be called constant magnetic eld restricted grand and reduces to 6.2 when J = H = 0. A simpler alternative form for the Hamiltonian is obtained by associating the spin component value i = 0 with each site occupied by a B atom and by de ning parameter by 6.18. Then 6.48 becomes
fi;jg
j ,H
a X
fig
i:
6.48
b H = ,"
n:n: X
We de ne a relative magnetization m with range ,1; 1 by The density per site conjugate to H is then M=N = mxa . From 2.59 the free energy per site is f = a , a , b xa , xa mH = b , P ; 6.51 and, from 2.60, since da , b = d, df = ,sdT , xa d , xa mdH : 6.52 The condition for phase equilibrium is that T , or a , b , H and f take the same values in the conjugate phases.
4
fi;jg
i j
2 2 ,X 2 ,J
fig
n:n: X
fi;jg
i j ,H
X
fig
i:
6.49 6.50
M m = h i ia = M = xM : a aN
Using i avoids confusion with the i of Sect. 6.2, which has a di erent meaning.
b Like the original Ising model Hamiltonian 2.66, H is invariant under a simultaneous change of sign of H and all spin variables. Hence the symmetry relations 2.68, 2.69 and 2.71 are still applicable provided that the values of are the same on both sides. It also follows that xa T; ; H = xa T; ; ,H : 6.53 If " = = 0, then 6.49 gives the Hamiltonian of the spin-1 Ising model with i = 1; 0 see Sect. 2.4 or Example 2.4.
6.6.2 Model II: The Ising Model Lattice Gas
This can be obtained from the results of Sect. 6.6.1 by replacing the B atoms by vacant sites or from the simple lattice gas of Chap. 5 by letting the molecules carry Ising spins. Since M = Ma 6= N = V the system is not a restricted one. We have n = 3 with 1 = ; 2 = H ; 3 = ,Pg ; 6.54 where is the chemical potential and Pg the lattice gas pressure eld. Takb ing N , T , and H as independent variables, the Hamiltonian H has the form 6.49 where ," is the non-magnetic part of the interaction energy of a nearest-neighbour pair of molecules and i = 0 now represents an empty site. This Hamiltonian gives a constant magnetic eld grand partition function. Let M =M ; m= M = M : 6.55 N N Then, from 1.79, ,P g = a , , m
149
H;
6.56
dPg = sdT + d + mdH : 6.57 From the similarity in form of the Hamiltonians the dependent intensive variables for models I and II are related by f = ,Pg ; xa = ; m = m ; 6.58 with the same values of , H and T on both sides.
6.6.3 Model III: The Symmetrical Ternary Solid Mixture

Suppose that each lattice site is occupied by an atom of component B, C or D. The restricted grand Hamilton of 6.2 can be generalized to b H = E , c , b Mc , d , b Md ; 6.59 the independent thermodynamic variables being N , T , c , b and d , b . There are six nearest-neighbour pair interaction energies "bb , "cc , "dd , "bc ,
150
"bd and "cd but this model is equivalent to model I only if the C-D symmetry
conditions "cc = "dd ; "bc = "bd ; 6.60 are satis ed. If C, D and B are identi ed with i = 1; ,1 and 0 respectively and 6.60 is assumed "cc = "aa , J ; "cd = "aa + J ; "bc = "ab ; 6.61 c = a + H ; d = a , H : Then, using 6.18, the Hamiltonian 6.59 can be written in the form 6.49 with 1 1 " = 2"bc , 2 "cc + "cd , "bb ; J = 2 "cd , "cc ; 6.62 1 + , , z" , " ; H = 1 , : = 2 c d b bc bb 2 c d The dependent intensive variables for models I and III are related by 1 1 x 1 , m = x : f = f; 6.63 d 2 xa 1 + m = xc ; 2 a
6.6.4 Model IV: The Symmetrical Lattice Gas Mixture

This model can be derived from model III by replacing each B atom by a vacant site or from that of Sect. 6.6.2 by identifying i = 1 and i = ,1 respectively with occupation by a C or D atom. Taking N , T , c and b as independent variables, the grand not restricted grand Hamiltonian is b H = E , Mcc , Md d : 6.64 There are three nearest-neighbour interaction energies "cc , "dd and "cd , but the model is equivalent to the previous ones only if the symmetry condition "cc = "dd 6.65 applies. Then with "cc = ," , J ; "cd = ," + J ; 6.66 c = + H ; d = , H ; 6.64 can then be written in the form 6.49 with 1 " = , 2 "cc + "cd ; J = 1 "cd , "cc ; 2 6.67 1 1 H = 2 c , d ; = 2 c + d : The dependent intensive variables for models II and IV are related by 1 Pg = Pg ; 2 1 + m = Mc = c ; 1 1 , m = Md = d : 6.68 2 N N
6.7 Mean-Field Theory and the Dilute Ising Model
151
6.6.5 Other Models

Replacement of J by ,J J 0 yields another four equivalent models. Model I is changed to a dilute Ising antiferromagnet and model II to an antiferromagnetic lattice gas. On a loose-packed lattice with H = 0, the Hamiltonian 6.49 is invariant to a change from J to ,J J 0 coupled with a reversal of sign for all i on one of the equal sublattices. Hence the zeroeld ferromagnet antiferromagnet equivalence of Sect. 4.2 is maintained for models I and II. For models III and IV, separation into C-rich and D-rich phases is replaced by sublattice segregation of C and D.
6.6.6 Applications
We now give a brief account of the development and application of the dilute Ising models. Meijering 1950, 1951 developed a theory of ternary mixtures in which the symmetrical case was equivalent to model III in the meaneld approximation. Bell 1953 treated model III using the meaneld approximation and the rst-order pair method see Chap. 7. He pointed out the equivalence of models III to I and also considered the antiferromagnetic case, deriving a relation between the sublattice order parameter and conjugate phase composition which corresponds to the binary mixture relation 6.39. Wheeler and Widom 1970 also used a rst-order method for model III. Blume 1966 and Capel 1966 developed meaneld theories for model I in the restricted case " = 0. This work was extended by Blume, Emery and Gri ths 1971 to model I with " 0, which is hence sometimes called the Blume Emery Gri ths model. These authors pointed out that the phase diagram for the model can contain tricritical points, in the sense de ned by Gri ths 1970 see Sect. 6.9 below. They were concerned with modelling low-temperature liquid mixtures of the helium isotopes 4 He and 3 He, identifying them with A and B respectively. 4 He undergoes a second-order transition to a super uid state characterized by a two-dimensional vector order variable. The representation by model I, with H = 0, involves replacing the real ordering process by a simpler one in the hope, which is largely justi ed, that the general phase characteristics will not be a ected see Sect. 6.8 below. In a rather similar way, monolayers at the air water or oil water interface, which display quite complicated cooperative phenomena, can be represented by model II with H = 0 see Sect. 7.4 below.

We develop a meaneld theory for model I of Sect. 6.6.1, which can easily be transcribed to the models II, III or IV of Sects. 6.6.2 6.6.4. To simplify the notation, let xa = x, xb = 1 , x. By the generalized meaneld theory of Sect. 3.3 the number Naa of AA nearest-neighbour pairs can be replaced by
152
1 zNx2 . From 6.50 the fractions of sites occupied by A atoms with i = 1 2 1 and i = ,1 respectively are 1 x1 + m and 2 x1 , m so that, by an argu2 ment similar to that following 3.36, the magnetic energy can be replaced by , 1 J zx2 m2 . Then using 3.38, 6.10, 6.18 and 6.51, the free energy 2 density f = T x ln x + 1 , x ln1 , x , 1 z"x2 , x 2
1 + 2 x fT 1 + m ln1 + m + 1 , m ln1 , m where = + T ln2 : 6.70 The equilibrium free energy density f is a function of T , and H so that the densities x and m must be regarded as order variables with respect to which the expression 6.69 has to be minimized. Equating to zero the partial derivatives of f with respect to m and x yield respectively 1 T ln 1 + m , Jxm ; H= 2 6.71 1,m
, zJxm2 , 2mH ;
6.69
x = T ln 1 , x , z" + Jm2 x , mH
1 + 2 T f1 + m ln1 + m + 1 , m ln1 , mg : 6.72 We shall need the rst and second derivatives of with respect to x at constant T and H, treating m as a function of x given implicitly by 6.71. From 6.72 and 6.71, @ m2 = x1T x , 1 zJx @ @x , z" + Jm2 ; 6.73 2 @x T;H ,
2 m2 m2 1 1 1 3 = T 1 , x2 , x2 , 2 zJ @ @x , 2 zJx @ @x2 : 6.74 2 @x T;H With H = 0 6.71 is identical to 3.10 except that zJ is replaced by xzJ and it follows that, at equilibrium, m = 0 for T xzJ and m = ms ms 0 for T xzJ . The critical curve on which second-order transitions from the disordered m = 0 to be ordered m 6= 0 state take place is therefore represented in the x; T plane by the line T = xzJ ; 6.75 T in the ; T plane which is shown in Fig. 6.7. The critical curve =
@2
153
T z" 4 T1
C Pt1 I
Fig. 6.7. Critical line and

Pt2
conjugate disordered phase curve in the mole fraction temperature plane for dilute Ising model.
0.0
0.5
1.0
can be obtained by putting m = 0 and x = T=zJ in 6.72. Now consider the disordered state. With m = 0, 6.69 and 6.72 are similar to the corresponding meaneld relations for the binary solid mixture of Sect. 6.2 or, putting x = , for the simple lattice gas. The only di erence is that replaces . Hence, from 5.25 and Sect. 5.4 see also Example 5.1, a rst-order transition line in the H = 0 plane is speci ed by = , 1 z" ; T Tc : 6.76 2 The critical value of x is 1 and, in the meaneld approximation, the critical 2 1 temperature Tc = 4 z". The rst-order transition is from a B-rich disordered phase 1 to an A-rich disordered phase 2 . For the equivalent lattice gas model II of Sect. 6.6.2, 1 and 2 are respectively vapour and liquid. These results can easily be con rmed from the formalism of this section. With T and H constant, 6.52 gives df = ,xd = ,xd : 6.77 From Sect. 1.10ii, phase separation is associated with an instability loop including a range of x where @ 0 6.78 and the critical point is given by @ = 0 ; @2 = 0 :
@x @x
These last relations are satis ed by 6.73 and 6.74, with m2 and its deriva1 tives put equal to zero, if 2 is substituted for x and 1 z" for T . Substituting 4 1 z" into 6.72, with m = 0, gives = ,2 x 6.80 T ln 1 , x , 1 z"2x , 1 = 0 : 2 1 For T 4 z", 6.80 has pairs of solutions x; 1 , x. If we put m = 0, = 1 z" in 6.69 then we obtain an expression such that f T; x = f T; 1 , x. ,2
@x2
6.79
154
Hence 6.80 gives the 1 2 coexistence curve, symmetrical about x = 1 , 2 in the x; T plane, as shown in Fig. 6.7. Here the critical point is labelled by C and the intersection of the curve 6.80 with the critical line 6.75 by Q. From 6.80 and 6.75 the coordinates xq ; Tq of Q are given by For T Tq, the conjugate phase 2 is replaced by an ordered phase and 6.80 no longer represents the coexistence curve. If the e ect of the shortrange magnetic ordering in component A were fully taken into account, as it would be in an exact theory, the disordered phase coexistence curve would no longer be symmetric about x = 1 . The only exception is the case J = 0, 2 discussed in Sect. 9.9.
x " xq ln 1 ,qx , 2J 2xq , 1 = 0 ; q
Tq = xq zJ :
6.81
6.8 Multicritical Points in the Dilute Ising Model

It has been found that there is a critical curve in the ; T plane for the dilute Ising model just as there is in the H; T plane for the metamagnet see Sect. 4.4.5 To look for phase separation instability on the critical curve we need an expression for m at given T valid for values of x just greater than x , where x is the critical value of x for magnetic ordering, de ned by 6.75. From 6.71, with H = 0, and an expansion similar to that of 3.13 we have, for constant T and x x , x, 2 , 6.82 m2 = 3x x x , 275x x + 2 and hence @ m2 @ 2 m2 3 54 = x ; @x x=x+0 @x2 x=x+0 = , 5x 2 : 6.83 Now, using 6.73 and 6.83, we de ne zJ 3 + T = @ B1 @x x=x+0 = 1 , x , z" , 2 zJ ; zJ , T = @ B1 6.84 @x x=x,0 = 1 , x , z" ; 1 @2 : B2 T = 2 @x2 x=x +0
5
Note that x here corresponds to m in the theory of the metamagnet whereas m here corresponds to , given by 4.32.
155
z" 4 1 T1 T
E
2
I
z" 4 Tt
1
D
2
Pt
T
a
b
T Tt
Pt
F0
F c Fig. 6.8. Phase diagrams in the H = 0 plane as "=J decreases. Full lines are second-order transition curves critical curves and broken lines rst-order transition curves. a "=J 3:8019, b 2:7747 "=J 3:8019, c "=J 2:7747.
E
+ Thus B1, T B1 T , and instability will appear on the ordered side + of the critical curve when B1 T = 0. From 6.84 the coordinates of the instability point Pt in the x; T plane are + Tt = zJxt : 6.85 xt = 22""+ 3J ; J A little numerical work shows that Pt and Q coincide i.e. xt = xq when " = 3:8019J . For " 3:8019J , xt xq and, in the x; T plane, the point Pt , like Pt1 in Fig. 6.7, lies inside the 1 2 coexistence curve. The phase diagram in the ; T plane is sketched in Fig. 6.8a, where CQ is the line of 1 2 rst-order transitions and QE is the line of 1 rst-order transitions, the
156
ordered phase being labelled by . The point E represents the equilibrium at T = 0 of a pure B phase with a completely ordered m = 1 pure A phase, and it is easily veri ed that at this point = , 1 z" + J . The line 2 of equilibrium second-order transitions on the critical curve terminates at Q which is therefore a critical end-point. For " 3:8019J , xq xt and, in the x; T plane, the point Pt , like Pt2 in Fig. 6.7, lies outside the 1 2 coexistence curve. Thus, as T passes below Tt , 2 phase separation begins at the point Pt on the critical curve. Pt is therefore a tricritical point. The phase diagram in the ; T plane is sketched in Fig. 6.8b. The 2 rst-order transition line which starts at Pt terminates at a triple point D where it meets the 1 2 and 1 rst-order transition lines CD and DE respectively. When the ratio "=J is further reduced, the 1 2 critical point, which occurs at x = 1 and T = 1 z", falls into the instability range initiated at the 2 4 tricritical point. There are no longer two distinguishable disordered phases, and the phase diagram in the ; T plane assumes the simple form shown in Fig. 6.8c. Here, the rst-order transition line EPt meets the second-order transition line at the tricritical point Pt . The transition between the types of phase diagram shown in Figs. 6.8b and c occurs at the value of "=J where there is equilibrium between the phase at C and an ordered phase. This can be found by setting T = 1 z", and equating f and respectively for 4 1 x = 2 and m = 0 to the appropriate expressions for f and with m 0. The two resulting relations together with the equilibrium equation 6.71 with H = 0 determine the values of x and m for the ordered phase and the ratio "=J . The solution gives " = 2:7747J . Now consider the tricritical phase separation for T = Tt , where 0 Tt. For the ordered and disordered phases respectively + , = B1 Tt , x , x + B2 Tt , x , x 2 + ; 6.86 , = B1, Tt , x , x + ; where B1 T and B2 T are de ned by 6.84 and x by 6.75. Recalling + T = 0 we can, if we require only leading terms in , write, using that B1 q 6.84 and 6.85, 3J + 2" 2 + T , = ,A ; : 6.87 B1 t A= 2J To the same order of accuracy, B1, T and B2 T can be replaced by their B1, Tt and B2 Tt forms. From 6.74, 6.83 and 6.84, 2 "6 2 B1, Tt = 3 zJ ; B2 Tt = zJ 3J + 280J 2JJ + 40J" + 40" : 6.88 2 + 2" The plot of against x has the form shown in Fig. 6.9. Applying analysis similar to that at the end of Sect. 4.4, the compositions of the ordered and
157
A2 2 + O3 8zJB2 Tt and the value of at phase equilibrium by 3 2 = , 16BAT 2 + O3 : 2 t x = x ,
disordered conjugate phases are given respectively by 3 x = x + 4 B AT + O2 ; 2 t
6.89
6.90
From 6.90, the rst-order and second-order transition lines in the ; T plane which meet at the tricritical point have a common tangent there. From 6.89 the tricritical point is the apex of a cusp in a coexistence curve in the x; T plane, as shown in Fig. 6.10 which corresponds to the ; T plane diagram of Fig. 6.8c. A diagram like that of Fig. 6.10 qualitatively resembles that observed for 4 He 3 He mixtures Blume et al. 1971. The meaneld critical curve relation 6.75 can be expressed as x = T=Tc, where Tc is the critical temperature for x = 1 pure A, but the observed values at the tricritical point for the 4 He 3 He mixture are Tt=Tc = 0:4, xt = 0:331. When " 0 and short-range ordering is taken into account, T=Tc x on critical curves Bell 1953. However, in view of the schematic nature of I as a model for the 4 He 3 He mixture, close quantitative agreement between even exact results for model I and experiment is unlikely. For instance, the mole fraction x of A in the A-rich conjugate phase ! 1 as T ! 0, but in the 4 He 3 He systems the 4 He mole fraction ! 0:94, owing to quantum statistical e ects. Finally, note that, in spite of similar behaviour near the tricritical point, there are global di erences between the dilute Ising model and the meta3 magnet of Sect. 4.4. For the latter, there is a critical end-point for 5 ,

Fig. 6.9. Chemical potential mole

x x
fraction isotherm for T Tt showing instability on the critical curve.
158
zJ
T Tt
Fig. 6.10. Phase diagram in mole

fraction temperature plane corresponding to Fig. 6.8c. The horizontal lines are phase separation tie lines.
1.0
and when increases through this value, a tricritical point is formed by the coalescence of the critical end-point and the òrdinary' critical point i.e. Q and C in Fig. 4.7. For the dilute Ising model there is a critical end-point when "=J 3:8019, but when "=J decreases through this value, the critical end-point bifurcates into a tricritical point and a triple point. This leads to the intermediate phase pattern of Fig. 6.8b, which is absent in the metamagnetic case. It is signi cant that the length of the rst-order transition line above the critical end-point lies in the ordered region for the metamagnet but in the disordered region for the dilute Ising model. Hence, for the latter, coalescence of the critical point C with a point on the critical curve would require B1, T = 0, where B1, T is given by 6.84. Since B1, T = 0 + implies B1 T 0, such a coalescence could not occur in an equilibrium state in meaneld theory.
6.9 Multicritical Phenomena with Additional Thermodynamic Dimension

By the results derived at the end of the account of Sect. 6.6.1, H = 0 de nes a plane of symmetry in the three-dimensional ; H; T space of elds. So far, phase diagrams have been obtained only for this plane, but we now consider the extension into three dimensions of the `simple' phase pattern of Fig. 6.8c. A spontaneous magnetization ms ; T 0 is associated with any point ; T in the ordered region of the plane H 0. From 6.71, m ! ms if this point is approached from the H 0 half-space but m ! ,ms if it is approached from the H 0 half-space cf.3.11. Hence there is a discontinuity in m of magnitude 2ms at H = 0, although by 6.53 the mole fraction x is continuous. It is necessary to distinguish two ordered phases
6.9 Multicritical Phenomena with Additional Dimension
159
labelled + H; m 0 and , H; m 0 respectively and in the ; T; H space the ordered region of the H = 0 plane becomes a surface of rstorder transitions from + to , . One boundary of the ordered region in the H = 0 plane is the rst-order transition line Pt E. At any point on PtE we now recognize two phases + and , , with equal and opposite magnetizations, in equilibrium with a m = 0 phase, where the value of x is lower. Hence Pt E is a line of triple points in ; H; T space. These results are in accordance with the generalized phase rule 1.92 which yields F3; p = 4 , p. Here p is the number of phases in equilibrium and F3; p the number of degrees of freedom. Thus, F3; 2 = 2, giving a surface in eld space whereas F3; 3 = 1, giving a line. Now consider a vertical = constant segment in the ordered region of the H = 0 plane, like FG in Fig. 6.8c. The segment FG is a line of rst-order + , transitions in the = constant plane, and hence the point G where this line terminates is a critical point. Since G is an arbitrary point on the critical curve, the latter becomes a line of critical points in ; H; T space.6 As remarked above, three distinct phases, designated + , , and , are in equilibrium at any point on the line Pt E, and Pt E is a boundary of the + , transition surface. It may therefore be supposed that there are two other rst-order transition surfaces which meet on Pt E. These are respectively a + transition surface in the H 0 half-space, and a , surface in the H 0 half-space. By eld-reversal symmetry these surfaces will be mirror images in the H = 0 plane and they can be regarded as a pair of `wings', as shown in Fig. 6.11. This picture is con rmed by the proof given below of the existence of the critical curves Pt K and Pt L, which form the upper boundaries of the `wings'. Although m 6= 0 whenever H 6= 0, the phase region in ; T; H space can be de ned as that which is continuous with the m = 0 area on the H = 0 plane. At a transition surface, jmj and x increase discontinuously. However, for very large positive H, i = 1 for nearly all sites occupied by an A. Thus component A e ectively consists of a single species and the system is a simple binary mixture with the parameter " replaced by " + J . The situation is similar for large negative H. Hence on 1 1 the critical curves Pt K or Pt L, T ! 4 z" + J and x ! 2 as jHj ! 1. At constant and T , from 6.52, df = ,xmdH. From 1.102, critical curves in the H 6= 0 regions are de ned by the relations
@H
Substitution of 6.71 into 6.72 gives x = T ln 1 , x , z"x + 1 T ln1 , m2 : 2

6
@ xm ;T
= 0;
@2H
@ 2 xm ;T = 0 :
6.91 6.92
We now restrict the term `critical curve' to the equilibrium part only.
160

G
Pt
T
L
,,
+,
+, ,
H
Fig. 6.11. Phase diagram in chemical potential magnetic eld temperature space
corresponding to Fig. 6.8c. The broken curve Pt E is a line of triple points and bounds the vertically hatched + , transition surface in the symmetry plane H = 0. The `wings' bounded by PtE, PtK and PtL respectively lie outside the H = 0 plane.
Now 6.91 implies that H has been replaced as independent variable by the density xm. However, since, by 6.92, x and hence xm depend only on m at xed and T , it follows that 6.91 is equivalent to 6.93 @m2 ;T = 0 ; or alternatively, regarding m as a function of x, to similar relations with x replacing m. Thus to demonstrate the presence of critical curves, solutions of 6.93 must be shown to exist. They can be obtained explicitly for " = 0 6.94 6.95
@H
@m ;T = 0 ;
@2H
Blume et al. 1971 when 6.92 can be expressed as pZ 2 ; x= Z = exp=T : Z+ 1,m Substitution into 6.71 yields H as a function of m given by J 1 T ln 1 + m , zpZm : H= 2 1,m Z + 1 , m2
6.9 Multicritical Phenomena with Additional Dimension
161
From 6.93 and 6.95 1 = 2 T ln 4T 4, zJ ; T
" 6.96 p 1 T ln zJ , 2T + pzJ zJ , 3T , pzJ zJ , 3T H= 2 zJ , 2T , zJ zJ , 3T for the critical curve in the H 0 half-space. Changing H to ,H gives the 1 mirror image critical curve in the H 0 half-space. At T = 2 zJ , 6.96 yields H = 0 and = ,T ln2 while 6.95 and 6.94 then give m = 0 and 1 x = 3 . This means that 6.96 and its mirror image terminate on the H = 0 plane at the tricritical point Pt , as may be seen from putting " = 0 in 6.85 and 6.92. Also, from 6.96, H ! 1 as T ! 1 zJ + 0 and it can be shown 4 p from 6.96 and 6.95 that 1 , m2 ' Z so that by 6.94, x ! 1 . For the 2 " = 0 case this agrees with the argument above about the large jHj limit. For " 6= 0, 6.92 and 6.71 can be used to express m and H as functions of x. Then equations for the critical curves can be derived from relations similar to 6.93, but with x replacing m. However, these do not yield an explicit solution like 6.96 and must be solved numerically. It can be seen that in a three-dimensional space of thermodynamic elds, like the ; H; T space considered here, a tricritical point terminates three distinct critical curves as well as a line of triple points. This in fact explains the description `tricritical', introduction by Gri ths 1970. If thermodynamic variables are adjusted to maintain three-phase equilibrium as the temperature increases, that is the system is kept on the line PE in Fig. 6.11, then the tricritical point occurs when the three coexisting phases coalesce. The extension into ; H; T space of the phase pattern of Fig. 6.8a, with a critical end-point Q, is sketched in Fig. 6.12. The ordered region on the plane of symmetry, H = 0, again becomes an + , rst-order transition surface bounded above by a critical curve. The segment QE becomes a line of + , 1 triple points while QC is the intersection of a 1 2 rst-order transition surface with the plane H = 0. A critical end-point is thus the termination of a line of triple points and a single critical curve. If thermodynamic variables are adjusted to keep three phases in equilibrium as the temperature is raised, then a critical end-point occurs when two phases become identical but remain distinct from the third phase. In the present case, + and , coalesce into 2 but remain distinct from 1 . The three-dimensional topology corresponding to the intermediate H = 0 phase pattern of Fig. 6.8b is rather complicated. D becomes a quadruple point where the four phases 1 , 2 , + and , coexist. It is the meeting point of the 1 + , and 2 =+ , triple point lines DE and DPt respectively in the H = 0 plane with 1 2 + and 1 2 , triple point lines which are mirror images of each other in this plane. For 3J " 3:8019J the 1 2 lines terminate at critical end-points where the
162

I
1 , 2
C
T
+, ,
, , 1
+ , 1
H
Fig. 6.12. As for Fig. 6.11, corresponding to Fig. 6.8a; Q is the critical end-point. The broken curve QE is a line of triple points in the plane of symmetry H = 0.
2 and phases coalesce. When " = 3J Mukamel and Blume 1974 there are three tricritical points, 2 + ,, 1 2 + and 1 2 , . For 2:7747 " 3J there are again two critical end-points where the 1 and 2 phases now coalesce. The value " = 3J corresponds to the completely
symmetrical 3-state Potts model see Sect. 6.10. The eld space for the metamagnet can be extended into three dimensions by postulating a staggered eld Hs , conjugate to the order parameter , which acts in opposite senses at the sites of the two sublattices. For 3=5 the `tricritical' phase diagram in H; T; Hs space is similar to Fig. 6.11, the + and , phases in equilibrium at the symmetry plane Hs = 0 corresponding to the 0 0 states speci ed in 4.10. For 3 , four ordered phases + , , , + and , 5 must be envisaged, and when the diagram of Fig. 4.7 is extended into three dimensions, QC becomes a line of quadruple points where the four ordered 0 0 phases are in equilibrium. The + , and , , critical lines terminate at C which is called a bicritical end-point Kincaid and Cohn 1975. For general accounts of tricritical and other multicritical points see Aharony 1983, Lawrie and Sarbach 1984 and Pynn and Skjeltorp 1984.
6.10 The Unsymmetrical and Completely Symmetrical Models
163
In this section we discuss two contrasting cases. In the rst the symmetry about the plane H = 0 is removed whereas in the second a higher symmetry is introduced. Consider the binary lattice gas model IV of Sect. 6.6.4, where the eld variable H is replaced by c , d . This is equivalent to model I of Sect. 6.6.1, with the relation c , d = 0 de ning the symmetry plane, provided that "cc = "dd . However, for real mixtures, symmetry conditions are satis ed only in special cases. When "cc 6= "dd a term see Example 6.4 1 4 "cc , "dd
n:n: X
must be added to the Hamiltonian given by 6.49 and 6.67. Thus the Hamiltonian is no longer invariant under a simultaneous change of sign of c , d and all i ; that is, an interchange of the values of c and d and replacement of each C by a D and each D by a C. We look at the e ect on the three-dimensional phase pattern of Fig. 6.11 where, for model IV of Sect. 6.6.4, labels a vapour phase and + and , correspond to liquid C D mixtures. When "cc 6= "dd the + , rst-order transition surface no longer lies in the plane c = d H = 0, which ceases to be a plane of symmetry. Again, the + and , critical curves cease to be mirror images in the c = d plane and no longer meet the + , transition surface at the same point Pt . Suppose that the + critical curve has slipped downwards relative to the T -axis and meets the + , transition surface at A, where T = Ta . For T Ta, the + and phases are identical and there remains only a , rst-order transition surface, as shown in Fig. 6.13. The broken triple point line and the + critical curve terminate at A, which is thus a critical end-point. In the absence of intrinsic symmetry, such as exists in magnetic systems, or an accidental symmetry, for instance the relation "cc = "dd , a line of triple points will normally terminate at a critical end-point. Suppose that an additional eld is introduced by, say, adding another component to the system. By appropriately adjusting the new eld we may be able to make A move up to point A0 on the , critical curve, and A0 will then become a tricritical point. A tricritical point thus appears at a special position on a line of critical end-points in a fourdimensional eld space. It is helpful to recall the Gibbs phase rule 1.93 for uid mixtures. For three coexisting phases, D; 3 = , 1, where is the number of components, implying a ,1-dimensional domain of triple points in eld space. We then expect , 2- and , 3-dimensional domains of critical end-points and tricritical points respectively. Thus, for lines of critical end-points and isolated tricritical points, = 3, which is a ternary solution. Experimentally, it is hard to nd an isolated tricritical point for a uid system
fi;jg
i j2 + j i2
6.97
164
T
A
A0
c , d
Fig. 6.13. Three-dimensional phase diagram for unsymmetrical binary lattice gas.
The broken line of triple points does not lie in the plane c , d = 0.
so at least a quaternary = 4 mixture is required, giving a line of tricritical points. One of these can then be located at, say, an arbitrary pressure. For multicritical points in uids, see Knobler and Scott 1984, or Rowlinson and Swinton 1982. The Potts model Potts 1952, Wu 1982 was described in Example 2.2 see also Sect. 10.14 and Volume 2, Sect. 3.5. The ferromagnetic 3-state Potts model can be related to the present theory by considering a completely symmetrical form of model III of Sect. 6.6.3 in which 6.60 is replaced by "bb = "cc = "dd = u0 , J ; J 0: 6.98 "bc = "bd = "cd = u0 + J ; The parameter u0 gives a constant additive term Nu0 in the energy and has no e ect on the lattice distribution. Substitution of 6.98 into 6.62 gives 1 " = 3J ; H = 2 c , d ; = 1 c + d , b , 2zJ : 6.99 2 for the completely symmetrical ternary model. In the Potts model, B, C and D are interconvertible species of the same component so that b = c = d and 6.99 becomes " = 3J; H = 0; = ,2zJ: 6.100
165
From 6.98 the di erence between the energies of like and unlike pairs is ,2J , so that J = 1 R, where R is the parameter de ned in Example 2.2. 2 The completely symmetrical ternary model is sometimes regarded as a Potts model with elds. To exhibit the properties of the completely symmetrical model use Cartesian coordinates
p3 c , d ; 1 c + d , 2b ; T 2 2
in eld space. Hence, by 6.99, c , d = 0, = ,2zJ on the T -axis which thus represents the Potts model with b = c = d . The planes b , d = 0 and b , c = 0 meet the plane c , d = 0 on the T -axis at angles 2=3. The three-dimensional phase diagram has three-fold symmetry with respect to rotation about the T -axis,7 a result which applies exactly as well as in the meaneld approximation. Since there is a rst-order transition surface on the c , d = 0 plane there will be similar rst-order transition surfaces on the b , d = 0 and b , c = 0 planes. The three transition surfaces meet in a line of triple points occupying the lower part of the T -axis. Since "=J = 3 lies in the intermediate range for the meaneld approximation, the phase diagram in the c , d = 0 plane is topologically similar to Fig. 6.8b, the three-dimensional extension of which is discussed in Sect. 6.9. The line of triple points DE now lies on the T -axis and is joined at the quadruple point D by another line of triple points in the c , d = 0 plane which terminates at a tricritical point Pt . By the three-fold symmetry condition the phase patterns in the b , d = 0 and the b , c = 0 planes are identical with that in the c , d = 0 plane. Hence the two remaining lines of triple points which emanate from the quadruple point D lie respectively in the b , d = 0 and b , c = 0 planes, and each line terminates at a tricritical point. Hence when "=J = 3 there are three tricritical points in the dilute Ising and equivalent models. At any value of T the Potts model is represented by a point on the T -axis. At high temperatures the state is disordered with xb = xc = xd = 1 , but when T is decreased, a transition will 3 occur at the quadruple point D on the T -axis. The transition is rst-order since the representative points must cross a rst-order transition surface to one of the three equivalent states whose domains meet at the segment DE on the T -axis. The meaneld approximation phase diagram is probably qualitatively correct for three-dimensional lattices Wu 1982. However, it has been shown by Baxter 1973b and Baxter et al. 1978 that the transition in the ferromagnetic -state Potts model on two-dimensional lattices is of higher order for 4. This implies that, in an exact solution of the dilute Ising model, the value "=J = 3 is in a "=J range corresponding to phase diagrams topolog7
There is also symmetry with respect to re ection in the c , d = 0 and b , c = 0 planes. Hence the overall symmetry of the system is that of the permutation group S3 .
166
ically similar to Figs. 6.8c and 6.11. From symmetry considerations the line of triple points Pt E occupies the lower part of the T -axis, and the three critical curves meeting at the tricritical point Pt lie respectively in the c , d = 0, b , d = 0 and b , c = 0 planes and are of identical form. For the Potts model, a higher-order transition occurs on the T -axis at Pt . There is other evidence supporting this general picture Straley and Fisher 1973, Wu 1982.
6.11 Alternative Forms for the Dilute Ising Model

6.11.1 Model A: Equilibrium Bond Dilution
In the site-dilute Ising model considered above, the fraction 1 , x of lattice sites not occupied by spins are ìnert' magnetically. In a bond-dilute model, on the other hand, all N sites are occupied by spins, but a fraction 1 , y of the nearest neighbour pairs are ìnert', with no magnetic interaction between the two spins. We can say that a bond exists between the spins of each interacting nearest neighbour pair and that for the non-interacting pairs the bond is broken. It is assumed that the distribution of unbroken and broken bonds 1 among the 2 zN lattice edges re-adjusts at each temperature to minimize the free energy. As will be shown in Sect. 9.2 the properties of the model can then be deduced from those of the pure Ising model on the same lattice. It turns out that a critical concentration yc exists such that T y ! 0 as y ! yc + 0, where T y is the critical temperature for a fraction y of unbroken bonds. For y yc, no magnetic transition occurs.
6.11.2 Model B: Random Site Dilution

Suppose that a lattice mixture of xN A atoms with magnetic moment and 1 , xN magnetically inert B atoms is rapidly cooled quenched from a temperature T1. It is assumed that the orientations of the spins on the A atoms can re-adjust rapidly enough to maintain thermal equilibrium, but that, due to the relative slowness of A B di usion, the positions of the A and B atoms on the lattice remain xed. The A B lattice con guration of any particular assembly is thus frozen. However, in the theory, it is necessary to consider a distribution of A B con gurations which is independent of the actual value of T but identical to the equilibrium distribution at T1 . It is usually assumed that T1 is large enough for the equilibrium distribution at T1 to be random, giving equal weight to all A B con gurations. Bell 1958 distinguished between this model and the equilibrium site-dilute model introduced in Sect. 6.6. The meaneld method, which implicitly assumes that the environment of each A atom is the same, is unsuitable for dealing with the random model. The rst-order pair approximation described in Chap. 7 makes some allowance for short-range order e ects, and with the pair numbers
6.11 Alternative Forms for the Dilute Ising Model
167
Essam 1972, 1980. A cluster of A atoms is de ned as a set such that there is at least one connected path of AA edges between any two atoms of the set. Long-range ordering of the spin orientations depends on the existence of in nite clusters. Thus, denoting the probability in the limit N ! 1 of an A atom belonging to an in nite cluster by px, the critical concentration xc is the value for which px = 0 when x xc and px 0 for x xc . Hence xc depends on the properties of random distributions on the lattice and not on the details of the AA interaction and is, for instance, the same for the Heisenberg model as for the Ising model. Accurate values for regular lattices di er from the approximate value xc = 1=z , 1; for example, on a 1 1 square lattice, the accurate value is xc = 2 whereas 1=z , 1 = 3 . However, percolation theory yields xc = 1=z , 1 for the loosely connected Bethe lattice described in Sect. 7.7, Fisher and Essam 1961. For more recent references on random systems see Thompson 1985.
Naa , Nbb and Nab xed at their random values gives a critical concentration xc = 1=z , 1 Bell 1958, Sato et al. 1959. The accurate value of xc can be derived from site percolation theory
6.11.3 Model C: Random Bond Dilution

The random bond-dilute model bears the same relation to model A as model B does to the equilibrium site-dilute model. It is assumed that the 1 zNy 2 1 unbroken bonds and the 2 zN 1 , y broken bonds are distributed randomly over the 1 zN edges connecting nearest neighbour pairs of sites. In bond 2 percolation theory Essam 1972, 1980 a cluster of sites is a set such that there is at least one connected path of unbroken bonds between any two atoms of the set. Denoting the probability in the limit N ! 1 of a site belonging to an in nite cluster by pb y the critical concentration yc is the value for which pb y = 0 when y yc and pb y 0 when y yc . For the random cases it can be shown that xc yc for any lattice. For some d = 2 lattices, xc and or yc can be deduced exactly by transformation methods, and for the remaining regular lattices including d = 3 lattices, yc can be obtained with reasonable accuracy by series methods. A striped randomness bond dilution model is reviewed by McCoy 1972. Let the Ising model interaction energy parameters for horizontal edges of the square lattice be J1 and for vertical edges be J2 . The parameter J1 has the same value over the whole lattice, whereas, with the rows of vertical edges labelled by the index `, J2 ` is a random variable with a probability distribution pJ2 . Critical properties di er considerably from those of the pure Ising model. For a review of models A, B and C as well as the equilibrium site-dilute model of Sect. 6.6 see Stinchcombe 1983.
168

i ii iv iii
x 0:5 1.0 Fig. 6.14. Phase diagram, in mole fraction-temperature plane for " ,2J . Reprinted from Bell and Lavis 1965, by permission of the publisher Taylor and Francis Ltd.
0
of AB pairs prevents a transition to a magnetized state and there is a critical concentration xc . Using a rst-order pair method, critical concentrations are found for the Heisenberg model Bell and Fairbairn 1961a and for the Ising model with " ,J Bell 1975 even when A B superlattice formation is not considered see Example 7.5. When A B sublattice ordering can occur there are states of four types: i disordered; ii sublattices ordered but unmagnetized; iii magnetized without sublattice order; iv magnetized with sublattice order. Houska 1963 used the meaneld method with certain restrictive conditions to investigate the properties of Fe Al alloys. Mean- eld theories for A atoms carrying spins of any magnitude s and spin-spin and A B interactions of arbitrary range were developed by Bell and Lavis 1965 and Lavis and Bell 1967, the former being concerned with 1:1 A B sublattice structures and the latter with more elaborate ones. A rst-order approximation was used by Lavis and Fairbairn 1966 with interactions con ned to nearest-neighbour pairs. Superexchange e ects were considered in the meaneld approximation by Bell and Lavis 1968. Curves of i ii, i iii, ii iv and iii iv transitions were derived giving a variety of phase diagrams in the x; T plane whose form depends on the relative magnitude of spin-spin and A B interactions inside and between the sublattices. The simplest case is a loose-packed lattice with interactions con ned to nearest-neighbour pairs. For " ,J and 1 x 2 there are no AA nearest-neighbour pairs at T = 0 and the system is unmagnetized, giving a critical concentration xc = 1 . A phase diagram for 2 " ,2J is shown in Fig. 6.14, where the broken lines are undisturbed i ii and i iii transition curves.
6.11.4 Model D: Equilibrium Site Dilution " 0 For negative " and large enough magnitudes of j"j=J and 1 , x, the formation
Examples
Examples
169
6.1 For the one-dimensional binary mixture with all N sites occupied, show that, apart from a constant factor, the con gurational canonical partition function is given by X Z = gMa; Mb ; X X,X ;
1 where X = exp"=T , " is given by 6.8, X = 2 Nab and g is de ned by 2.126. By the maximum-term method see Sect. 2.6, show that the equilibrium value of X is given by Xmax = 2xa xb ; N 1+
fX g
Show that the con gurational heat capacity per site at constant composition is given by 2X 2 2 c = 4kb lnX + xa xb : 2 1 1 For xa = xb = 2 , show that this reduces to 2 1 c = 4 kb X1=4ln X,1=4 2 : +X Show that this expression has the same value whether " is positive or negative and, putting 4J = j"j, show that it is equivalent to the expression 2.83 for the zeroeld heat capacity of the one-dimensional Ising model. 6.2 From equation 6.32 where satis es the equilibrium relation 6.34, show that @ 4a T ln f s 1 = 1+s ; f = 1 ,+ s1 , s : @xa ys1 + y 1 , show that f 1, where 0 1. For s = 3, y = 3 6.3 If the Ising model con gurational energy for the plain square lattice is modi ed to
= xa , xb 2 + 4Xxa xb : Show that Xmax=N ! xa xb , as T ! 1. Given that xa xb , show that Xmax=N ! 0, as T ! 0, if the inequality 6.17 applies and that Xmax=N ! xb , as T ! 0, if the inequality 6.21 applies.
,J
n:n: X
fi;jg
i j , J4
squares X
fi;j;k;`g
i j k `;
170
where J 0, J4 0 and the second summation is over all squares of four sites such that i; j , j; k, k; ` and `; i are nearest-neighbour pairs, then the Helmholtz free energy per site in the meaneld approximation is a = T 1 + m ln 1 + m + 1 , m ln 1 , m 2 2 2 2 2 , J m4 : , 2Jm 4 Show that a Landau expansion of a gives
a + T ln2 =
where
1 X
n=1
B2n T m2n ; n=1 , n=2 ,
2 T , 2J ; 1 B2n T = 12 T , J4 ;
81
1 Hence show that, if J4 3 J , there is no second-order transition at the temperature T = 4J where B2 T = 0, but instead a rst-order transition at a temperature T0 from m = 0 to m = m0 . De ning 1 x0 = 4 T0 =J and = 3J4 =J , 1, show by truncating the series after the n = 3 term that, approximately, 1 5 x0 x0 , 1 = 48 + 1 , x0 2 ; m2 = 5 + x , x0 : 0 4 0 6.4 For the models described in Sect. 6.6, label sites for which i = 1, i = ,1 and i = 0 by the indices 1, 2 and 3 respectively. Show that 1X N1 = 2 i + 1 i ; fig X N2 = 1 i , 1 i ; 2 f Xig 2 N3 = 1 , i ;
: 2n2T , 1 ; n 3 . n
where the summations are over all sites. Similarly show that the pair numbers Nij are given by
fig
Examples
171
1 N11 = 4
n:n: X n:n: X n:n: X
N22 = 1 4 N12 = 1 2 Na =
fi;jg fi;jg fi;jg
i + 1 j + 1 i j ; i , 1 j , 1 i j ;
i j i j , 1 :
n:n: X
Show that, if component A comprises species 1 and 2 i = 1, then
X
fig
2;
Naa =
The Hamiltonian of model II, Sect. 6.6.2 can be written in the form b H = ,Naa " , N11 + N22 , N12 J , Na , N1 , N2 H : Show that this is equivalent to 6.49 . The Hamiltonian for the unsymmetrical form of model IV, Sect. 6.6.4 is b H = "ccNcc + "dd Ndd + "cd Ncd , c Nc , d Nd : Identifying C and D respectively with i = 1 and i = ,1 and the vacant sites with i = 0, show that ^ 1 H = 4 "cc + "dd + 2"cd
n:n: X
fi;jg
i j
2 2:
fi;jg
i j
2 2
1 + 4 "cc + "dd , 2"cd

n:n:
n:n: X
X 2 i j + i j2 + 1 "cc , "dd 4 fi;jg 1 + X 2 , 1 , X : ,2 c d d i i 2 c

Show that if the symmetry condition "cc = "dd is satis ed then this becomes equivalent to 6.49, using the substitutions given in 6.66. 6.5 By substituting from 6.72 into 6.69 show that 1 f = T ln1 , x + 2 z"x2 + 1 zJx2 m2 : 2 1 Suppose that, at the disordered phase critical temperature T = 4 z" 1 , m = 0 is in equilibrium with a and H = 0, the phase in which x = 2
fig fig
fi;jg
i j
172
phase in which x = x1 1 , m = m1 0. Using the above expression 2 for f and equation 6.92 for show that the equilibrium conditions are 2 ln 21 , x1 + 4x2 , 1 + 4J="x2 m2 = 0; 1 1 1
x 2 ln 1 ,1x , 42x1 , 1 + ln1 , m2 = 0 ; 1 1 to be solved in conjunction with 1 + m ln 1 , m1 , 8J="x1 m1 = 0 : 1 Show that these relations yield x1 = 0:84059, m1 = 0:67192, "=J = 2:7747. 6.6 Denoting the fractions of sites on which i = 1, i = ,1 and i = 0 by x1 , x2 and x3 respectively, show from 6.98 that for the ferromagnetic Potts model b = c = d the free energy is given in the meaneld approximation by 1 f = 2 zu0 , 1 zJ x2 + x2 + x2 , 2x1 x2 , 2x2 x3 , 2x3 x1 1 2 3 2 + T x1 ln x1 + x2 ln x2 + x3 ln x3 : Introducing an order variable s by putting x1 = x2 = 1 1 , s, 3 1 x3 = 3 1 + 2s, show that 2 f = 1 z3u0 + J , T ln3 , 3 zJs2 6 + 1 T 21 , s ln1 , s + 1 + 2s ln1 + 2s : 3 By expanding in a Landau series in powers of s, show that there is a rst-order transition at some temperature greater than 2 zJ . 3
7. Cluster Variation Methods
7.1 Introduction
Cluster variation methods are closed-form approximations which, unlike the meaneld method, make some allowance for short-range order e ects.1 In rst-order approximations, energies and weights for the microsystems on a cluster or group of sites are treated exactly, but the correlations due to sites shared between groups are handled by plausible assumptions. Equivalent versions were invented independently by various authors, notably Guggenheim 1935, the quasi-chemical method, Bethe 1935, the Bethe method and Kasteleyn and van Kranendonk 1956a, 1956b, 1956c, the constant-coupling method. We base our approach on a rst-order method due to Guggenheim and McGlashan 1951 which allows any suitable group of sites to be used and enables quite complicated forms of ordering and intermolecular attraction to be incorporated in a natural way. For the particular problem of sublattice ordering, a similar method had been used by Yang 1945 and Li 1949. Kikuchi 1951 carried the process of approximation to a higher level by minimizing the free energy with respect to the occupation probabilities of subgroups of sites, including nearest-neighbour pairs, as well as those of the main group and this procedure was placed on a systematic basis by Hijmans and de Boer 1955, 1956 see also Domb 1960, Burley 1972, Ziman 1979, Young 1981, Schlijper 1983 and Morita 1984. The cluster counting procedures used by Hijmans and de Boer 1955 have now been used in the nite-lattice series expansion method of de Neef and Enting 1977 see Volume 2, Sect. 7.7. Because they incorporate short-range ordering, cluster variation methods improve on the meaneld zeroth-order method as regards the accuracy of phase diagrams and critical parameters. Sometimes they can treat e ects which the meaneld method cannot and which are not yet accessible to exact analysis. However, the behaviour they predict near critical points is still `classical', giving the same critical exponents as in meaneld theory. In fact, this is bound to be true for any method in which the free energy densities
1
Some authors apply the term `mean eld' to all closed-form approximations. We reserve it for the randomized entropy method introduced in Chap. 3.
174
depend analytically on thermodynamic variables and order parameters. We illustrate this point for the Ising model at the end of Sect. 7.3. An interesting twist to the story is that, though rst-order theory is approximate for d = 2 or d = 3 regular lattices, like those shown in Appendix A.1, it is exact for the interior sites of more loosely connected arrays. This is well known for the rst-order method based on a nearest-neighbour pair of sites Baxter 1982a, Chap. 4, and in Sect. 7.7 we make an extension to a general group of sites like that used in Sect. 7.2.
7.2 A First-Order Method Using a General Site Group

The lattice is divided into Ng similar basic clusters of p sites in such a way that each nearest-neighbour pair belongs to only one group. Since the total number of nearest-neighbour pairs is 1 zN , where N is the number of sites, 2
Ng = z N ; 2q where q is the number of nearest-neighbour pairs in a group.
7.1
7.2.1 Equivalent Sites

All the p sites in a basic group are taken to be equivalent so that each site belongs to pNg =N = zp=2q di erent groups. Where the group is a single nearest-neighbour pair then q = 1 and p = 2 so that, as would be expected, Ng = 1 zN and each site belongs to z pairs. Figure 7.1 shows a division of the 2 square lattice into the groups, shown with faces shaded, each being a square 1 so that q = p = 4. Here, with z = 4, Ng = 2 N and each site belongs to two groups.
Fig. 7.1. Basic groups of four sites on

a square lattice.
7.2 A First-Order Method Using a General Site Group
175
We suppose that there are species present, an atom or molecule of a species occupying just one site. The species may be independent components or dipoles in a given orientation or, as in the case of the lattice gas of Chap. 5, holes. Since there are choices for each site, the total number of possible con gurations on the basic group is p . The con gurations are divided into types labelled by the index `, each type being characterized by the numbers pj` of each species j on sites of the group, where
X j =1
pj` = p ;
7.2
and by a con gurational energy "` . The number of con gurations of type ` is denoted by !` and it is assumed that they are equally likely. Suppose that the fraction of the Ng groups occupied by any con guration of type ` is ` . Then the fraction of sites occupied by species j is X xj = p,1 pj` !` ` : 7.3 The normalization condition
X j =1
f`g
xj =
X
f`g
!` ` = 1
7.4
follows from 7.2 and 7.3. The central feature of the method is an approximation for the number ` of con gurations on the lattice of N sites. If the Ng groups had no sites in common then the number of con gurations for given ` would be Ng ! : 7.5 0 ` = Y Ng ` ! !` To allow approximately for the fact that each site is shared by zp=2q groups, we put, in the absence of sublattice ordering, Y zp ,1 2q ` = 0 ` Nxj ! N ! ; 7.6
f`g
Nxj being the number of sites occupied by species j . From Stirling's formula and 7.1 it follows that
For the random distribution, which will occur in the limit T = 1,
j =1
X X zp N ,1 ln ` = , 2zq !` ` ln ` + 2q , 1 xj ln xj : j =1 f`g

7.7
` =
Y
j =1
xpj` : j
7.8
176
From 7.3 and 7.7,
N ,1 ln ` = ,
X j =1
xj ln xj ;
7.9
which is correct for a random distribution. It is possible to proceed in the reverse direction and deduce the correcting factor in 7.6 by requiring that 7.9 be satis ed Guggenheim 1952. We now construct a non-equilibrium free energy for a given set of ` , using a distribution in which the con gurational Hamiltonian is
b H = Ng
X
f`g
!` "` ` , N
X j =1
j xj :
7.10
Here the rst term is the con gurational energy while in the second term the i are either elds, linearly dependent on elds or zero. From 7.1, the free energy per site is
X X f = ,N ,1 T ln ` + 2zq !f`g "`` , j xj : ` j =1
7.11
The free energy must be minimized at constant temperature; so using 7.7 and the normalization conditions 7.4, 2q f = X ! T ln + " , p X T 1 , 2q ln x + 2q x :
7.12 The procedure now depends on the distribution, that is, the choice of independent thermodynamic variables and order parameters. The nal results will, of course, independent of this choice. For the distribution in which the only independent extensive variable is the number of sites N , which is identi ed with the reduced volume V 7.3 is used to express xj in terms of a linear sum of the ` . From 7.12, 2q f = X ! T ln + " , X p T 1 , 2q ln x + 2q :
f`g
j =1
pz
pz
7.13 The ` are connected by the normalization relation 7.4 and, selecting a reference con guration denoted by index r, the fraction r can be regarded as a dependent variable such that X !r r = , !` ` : 7.14 Substituting this relation into 7.13 and equating the coe cients of ` to zero for ` 6= r gives the equilibrium conditions
f`6=rg
f`g
j =1
j`
pz
pz
The composition of the assembly can be expressed in terms of the , 1 ratios xj =x j 6= . Since, from 7.2,
r
7.2 A First-Order Method Using a General Site Group 177 ` = exp "r , "` Y x1,2q=zp exp 2qi pj` ,pjr : 7.15 j
j =1
zpT
p` , pr = ,
,1 X j =1
pj` , pjr
7.16
the equilibrium conditions 7.15 can be rewritten in the form pj` ,pjr 2q Y ` = exp "r , "` ,1 xj 1, zp exp 2q , : T x zpT j
r j =1
7.17
Since the choice of r is arbitrary, 7.18 is true for any r and 7.15 7.17 for any pair `; r.
By substituting 7.15 back into 7.11 it follows that, at equilibrium, z T ln + " , X p 1 , 2q T ln x + 2qj : 7.18 f = 2q j r r jr zp zp j =1
7.2.2 Sublattice Ordering

The rst-order method can easily be adapted to cases where there is sublattice ordering. It will be supposed that there are equal sublattices labelled a and b and that the basic group has 1 p sites in each sublattice. The sublattice 2 mole fractions are given by X xa = 2p,1 pa !` ` ; j j`
f`g where j` and pb are respectively the number of members of species j on j`
xb = 2p j
,1 X
f`g
pb !` ` ; j`
7.19
the a and b sites of a group in con guration `. For the con guration number, 7.6 is replaced by 2
2q
1 3, zp ,1 1 N! 6 2N! 7 2 : 7.20 ` = 0 ` 6 Y
4 1 a Y
1 b 7 5 Nxj ! Nxj ! 2 2 After writing the corresponding expression for f we have
j =1 j =1
pa
178
2q f = X ! T ln + " ` ` ` ` z
f`g

X j =1
,p
which
T 1 , 2q lnxa xa + lnxb xb + 2q j xj ; j j j j pz pz

7.21 j = 1; : : : ; .
j j
reduces to 7.12 for xa = xb
7.3 The Pair Approximation and the Ising Model

In the simplest type of rst-order approximation also called the Bethe pair method the group is an nearest-neighbour pair, giving p = 2, q = 1. This is now applied to the Ising model ferromagnet taking up and down spins as species 1 and 2 respectively so that in the constanteld Hamiltonian 1 = H ; 2 = ,H ; 7.22 using the notation of 7.10. The four possible con gurations on the pair are 11, 12, 21, and 22, but, in the absence of sublattice ordering, 12 and 21 are equivalent. With an obvious notation the con guration fractions ` are 11 , 22 and 12 with !11 = !22 = 1, !12 = 2, "11 = "22 = ,J and "12 = J . Equations 7.3 reduce to x1 = 1 + m = 11 + 12 ; 2 7.23 1,m = + ; x2 = 2 22 12
where m is the relative magnetization. A parameter is de ned by
" 2 11 = 2 = x1 z,1=z exp 2H 22 x2 zT

12 = X ; 22
7.24
where the second relation follows from 7.22 and the equilibrium condition 7.17, with 11 and 22 substituted for ` and r respectively. Using 7.17, 7.25 7.26 7.27 where X = exp,2J=T : From 7.23 7.26
, 22 2 , 1 ,2 m = 11 + = 2 + 2X + 1 = 1 + 21 , ,1 + ,2 : X 11 + 2 12 22
7.3 The Pair Approximation and the Ising Model
179
For 1, the number of up spins is greater than the number of down spins, so that m 0 and vice versa for 1. From 7.27, m is a monotonically increasing function of at given T and m,1 = ,m ; m1 = 0 : 7.28 The con gurational energy u per site is 2 , 2X + 1 1 7.29 u = , 2 zJ 11 , 212 + 22 = , 1 zJ 2 + 2X + 1 : 2 11 + 212 + 22 Substituting rst 11 and then 22 for r in 7.18 and halving the sum of the expressions obtained, the free energy density f = 1 zT ln11 22 , 2z , 1T lnx1 x2 , 2zJ 4 1 T z , 2 ln + 2X + ,1 =2 , 1 T z , 1 ln 1 + X2 + X + ,1 , 1 zJ : 7.30 2 2 Note that 7.30 is valid only for a value of satisfying the equilibrium relations 7.24 and 7.25. It cannot be treated as a non-equilibrium free energy for other values of . To determine the equilibrium at given values of T and H, the second equality of 7.24 is rewritten as where, from 7.23 and 7.25, 1 + X z,1 = + X = ,1 ,1 : 7.32 From 7.32, reversing the sign of H in 7.31 changes any solution to ,1 . Hence, by 7.28, the symmetry relation 2.71 is satis ed. The symmetry relation 2.68 is also satis ed because the last expression for f in 7.30 is invariant under replacement of by ,1 , and the same is true for u, as given by 7.29.
x z,1 = ; exp 2H=T = z x2

1
7.31
7.3.1 Zero Field When H = 0, 7.31 reduces to
= 1 : 7.33 For any value of T , 7.33 is satis ed by = 1, giving m = 0. By 7.32 any other solutions exist in reciprocal pairs. By considering d =d it can be shown Example 7.2 that, for X z , 2=z, no such solutions exist, but, for X z , 2=z, there is a reciprocal pair 0 ; ,1 , corresponding, by 7.28, 0
180
to m = ms ms by X = z,2 ;

c
0. This indicates that the critical temperature is given
However, to con rm that a transition occurs, it must be shown that f 0 , f 1 0 ; 7.35 where it follows from 7.30 that f ,1 = f 0 . Consider the case z = 3 0 where 7.33 is a cubic equation and it can be deduced that 0 + ,1 = X,2 1 , 2X , X2 : 7.36 0 Substitution in 7.30 gives 21 , X3 1 T ln 7.37 f 0 , f 1 = , 2 1 , 2X1 + X2 : By manipulating the inequality 1 , 3X2 0 it can be shown that, for X X0 = 1 the argument of the logarithm is greater than unity so that 3 7.35 is satis ed. A similar proof is possible for z = 4 Example 7.4, and for higher values of z the inequality 7.35 can be demonstrated numerically. An alternative approach is to consider the variation with respect to m of a Helmholtz free energy density aT; m.
J Tc = ln z=2z , 2 :
7.34
7.3.2 The Critical Region

Pair approximation results depend entirely on the coordination number z and are thus, for instance, the same for the d = 2 triangular lattice as for the d = 3 simple cubic lattice. Critical parameters are compared with accurate values in Chap. 8, Table 8.1, and it can be seen that the pair approximation still under estimates the e ects of short-range correlations for T Tc, although results are better for d = 3 than for d = 2. For the critical region it is convenient to use the variable L = H=T de ned in 2.146. From 7.28, 7.31 and 7.32 m and L are odd functions of = ln ; 7.38 in terms of which they can be expanded in odd powers. From 7.27 and 7.31 H = L = zX , z , 2 + z , 1X1 , X 3 + O5 ; 7.39 T 21 + X 61 + X3
m = 1 + X + O3 : For T Tc, these relations yield
7.40
@m
@H
H=0
7.4 Phase Transitions in Amphipathic Monolayers
181
It can now be shown that the present theory yields the same classical values obtained in Sect. 3.6 for the critical exponents , , and , de ned in Sect. 3.4. In the present case the eld = H and the conjugate density is m. Since dX X , Xc ' dT T , Tc as T , Tc ! 0 , 7.42 c 7.41 yields = 1. At T = Tc, where X = Xc ' z , 2=z, 7.39 and 7.40 give H = z , 1z , 2 m3 + Om5 ; 7.43 T 3z2 so that = 3. For H = 0 the non-zero solutions of 7.39 give ' 3z3 Xc , X 1=2 ; m' 1+X 7.44 2z , 1z , 2 c so that = 1 . It can be shown Example 7.4 that the zeroeld heat capacity 2 has a nite discontinuity at T = Tc, giving = 0.
= !0 H = zT X2, X : lim m

c
7.41
7.3.3 The Linear Lattice

The pair method is exact for the linear d = 1 lattice. For the Ising model this can be demonstrated by putting z = 2, p = 2, q = 1 and = 2 in 7.1 and 7.6 to yield the con guration number 7.45 ` = N Nx1 !Nx2 ! ! 2 : 11 !N 22 ! N 12 With a di erent notation this is identical to gN1 ; N2 ; X , de ned by 2.126, 1 where X = 2 zN 12 = N 12 . However, as shown in Sect. 2.6, gN1 ; N2 ; X can be treated as an exact con guration number in the thermodynamic limit. It is easy to extend the equivalence to a model with an arbitrary number of species.

Amphipathic molecules have polar head groups with an a nity for water and hydrocarbon chains which resist immersion in water. Such molecules form stable or permanent metastable monolayers at the air water surface. One important way of investigating their properties is to plot isotherms of surface pressure against surface area per molecule. Two-dimensional vapour liquid rst-order transitions occur at very low surface pressures, and at higher
182
surface pressure in the appropriate temperature range a further transition is observed. The states on the lower pressure higher area and higher pressure lower area sides of this transition are usually termed liquid-expanded and liquid-condensed respectively. As the temperature is lowered, the separate vapour liquid-expanded and liquid-expanded liquid-condensed transitions coalesce into a single rst-order transition. Increase in carbon chain length has a similar e ect. These phase phenomena occur both for single chain compounds such as fatty acids with simple head groups and for the phospholipids, which have complex head groups attached to double chains. On a large number of experimental pressure area isotherms the liquid-expanded liquidcondensed transition is shown as second-order with the area per molecule and hence surface density continuous but with a discontinuity in the slope. The rate of change of pressure against area is smallest in magnitude on the liquidexpanded side of the transition. Similar liquid-expanded liquid-condensed transitions are observed in phospho-lipid monolayers at the oil water interface, though in this case the low surface pressure vapour liquid-expanded transitions are absent. As the temperature decreases, the slope on the liquidexpanded side of the transition decreases for both air water and oil water monolayers. Pallas and Pethica 1985 assert that for air water monolayers the observed slope should be zero, giving a horizontal part of the isotherm, which represents a rst-order liquid-expanded liquid-condensed transition. The fact that this has been missed in so many experiments is attributed to lack of purity in the chemicals used and poor humidity control. However, these authors accept the slope discontinuity second-order transition as genuine for oil water monolayers. The transition from the liquid-condensed to the liquid-expanded state is thought to be due to disordering of the chain conformations, a process which can take two forms. One of these is isomerism or chain `bending' which is the basis of a theory of the transition due to Nagle 1975, 1986, discussed in Volume 2, Chap. 9. The other is cooperative rotation of the carbon chains, which are planar in the lowest energy isomer of each molecule, about axes perpendicular to the interface. Kirkwood 1943 was the rst author to base a theory on this `rotational' or òrientational' e ect. Here we present the simplest type of orientational model Bell et al. 1978 which is a lattice gas with vacant sites and molecules with two orientational states. This is equivalent to the zeroeld case of the Ising model lattice gas, that is model II of Sect. 6.6.2. The detailed treatment will di er from that of Chap. 6 in that the emphasis is on pressure as a function of density, and the rst-order pair approximation is used rather than the meaneld method. As in Sect. 7.3 the two orientational species will be labelled `1' and `2', and the new hole species accordingly has the label `3'. Hence, if denotes the number density, x1 + x2 = ; x3 = 1 , : 7.46 We shall use the grand distribution so that, since H = 0,
183
7.47 dropping the index g used in Sect. 6.6, which is no longer necessary. From Sect. 6.6.2 the energies of the various types of pair are "11 = "22 = ," , J ;
1 = 2 = ;
3 = 0 ;
f = ,P ;
"12 = ," + J ;
We now introduce parameters w and y by and, from 7.17, 7.46 7.52, x1 z x2 = 1 + wz ; x = w ;
2
7.48
"13 = "23 = "33 = 0 :
11 2z,1 ; 22 = w
33 = y2 22
7.49 7.50 7.51
where
x z,1=z
y = Z x3 exp , zT ; 2
+ Z = exp " 2TJ : 7.52 It can be seen, from 7.50, that w = 1 in the disordered state with x1 = x2 . With w 1 or w 1 we have ordered states with x1 x2 or x1 x2 respectively, corresponding to m 0 or m 0 in Sect. 7.3. The use of wz,1 in place of the parameter of Sect. 7.3 avoids fractional powers. From 7.17 and 7.49
where X is de ned in 7.26. We now express the intensive variables , and P in terms of w and y. From 7.3, x1 = 11 + 12 + 13 ;
12 = X wz,1 ; 22
13 = Z wz,1 y ; 22
23 = Z y ; 22
7.53
x2 = 22 + 12 + 23 ;

and, from 7.46, 7.49 7.54, 1, = yy + Z wz,1 + Z 1 + wz 1 + Z wz,1 + Z y : From 7.46 and7.51
7.54
x3 = 33 + 13 + 23 ;
7.55
184
=T = z , 1 lnx3 =x2 , z lny=Z

= z , 1 ln 1 + wz 1 , = , z lny=Z and, from 7.18, with r = 33 , 7.46 7.47 and 7.54, P=T = z , 1 ln x3 , 1 z ln 33 2 = 1 z , 2 ln x3 + z lnx3 =33 2 7.57 = 1 fz , 2 ln1 , + z ln y + Z wz,1 + Z=y g : 2 Since n = 3 in the sense of Sect. 1.8 and H = 0, we have, after T is speci ed, only one remaining independent intensive variable, which can be , or P . It follows that for the ordered region, where w is a variable parameter, a relation between y and w is needed. This is given, from 7.51 and 7.54, by z,1 + X 7.58 w = 1w X wz,1+ ZZyy ; + + or z,1 z y = w ,Zww,, w , 1X : 7.59 1 In the disordered state, where w = 1, 7.59 is indeterminate but we can obtain y as a function of T and by solving 7.55 as a quadratic equation in y and retaining the positive root to give y = ,1 f1 , 2 Z + 1 , 2 2 Z2 + 2 1 , 1 + X 1=2 g : 7.60 From 7.55, y = 0 implies = 1 and the theory becomes identical with that of Sect. 7.3 for the zeroeld Ising model. This is con rmed by putting y = 0 in 7.58 which then becomes equivalent to 7.33 when is substituted for wz,1 . Now consider the ; T plane. It can be veri ed from 7.55 and 7.58 that, if there is a solution w 6= 1, then w,1 is also a solution for the same value of . On the = 1 axis there is such a reciprocal pair of solutions for T Tc, where Tc is the Ising model critical temperature given by 7.34. Taking w 1 for de niteness and xing T Tc, it is found that w decreases with until a value = is reached where w = 1. This is the critical value of at the given T ; that is the density at which the order-disorder secondorder transition occurs. To obtain as a function of T , rst take the limit of the right-hand side of 7.59 as w ! 1 giving y = z , 2 , zX=Z : 7.61 into 7.55 yields Substitution of w = 1 and y = y , 1 + = 2zzy yZ y2Z 2 : 7.62 1, + 7.56
7.4 Phase Transitions in Amphipathic Monolayers 185 Since, by 7.52 and 7.61, Z and y are functions of T , 7.62 is the equation
of the critical curve in the ; T plane. From 7.57 the corresponding surface pressure is 7.63 P = 1 T fz , 2 ln1 , + z ln 1 + 2Z=y g : 2 Equations 7.62 and 7.63 give the order-disorder transition point on any T Tc isotherm in the ; P plane. There is a discontinuous decrease in @P=@ T as increases through . We now consider vapour liquid rst-order transitions, which involve equilibrium between two disordered phases. In the disordered state, w = 1 so that 11 = 22 ; 12 = 22 ; 13 = 23 = Z y22 : 7.64 By substituting in 7.11 it can be shown that, apart from an extra term ln2 in the entropy per molecule, the model is equivalent to a simple lattice gas with an e ective value of "=T given by "=T exp ,"=T e = exp,J=T : 7.65 cosh It follows that, if the density of one conjugate phase is 1 , then that of the other is 1 , 1 and that the chemical potential for conjugate phases at temperature T is given by 7.66 =T = , ln2 , 1 z"=T e : 2 The critical point for disordered phase separation is thus de ned by exp,"=T 1=2 z , 2 1 = 2; = z : 7.67 coshJ=T The tricritical point occurs on the critical line at the temperature T = Tt where @P=@ T = 0 at = + 0. For T Tt , @P=@ T 0 in a range of above implying, by 6.57, that @=@ T 0, giving an instability loop like that shown in Fig. 6.9, with x replaced by , and leading to separation between ordered and disordered phases. With some manipulation it can be deduced from 7.55, 7.57, 7.59, 7.61 and 7.62 that the relation between temperature and the parameter y at the tricritical point is given by Zy 2 , 2 2z , 2 , 2z , 1Z2 y + 8Z = 0 : 7.68 From 7.61, this has a solution T = Tt 0 for any "=J ,1. For the highest range of "=J the tricritical point falls into the instability region initiated by the vapour liquid critical point and does not appear on the equilibrium phase diagram which displays a critical end-point and is similar to that shown in Fig. 6.8a. When "=J is reduced, the phase diagram rst becomes similar to that of Fig. 6.8b, with a triple point and a tricritical point, and then to that of Fig. 6.8c with a tricritical point only. The latter disappears at "=J = ,1.
186
P J Tt =J
3.95 3.9 3.85 0.4 1.0 1.3 1.6 a=a0 Fig. 7.2. Surface pressure-area isotherms for z = 6, "=J = 3:2. Each isotherm is labelled with the value of T=J . 4.05
0.7
For "=J ,1 there is no rst-order transition line and the critical curve meets the -axis at = z , 1=2z , 3, Example 7.5. The parameter ranges giving a- or b-type behaviour, where there are two successive transitions on certain isotherms, are relevant for air water interface monolayers just as the c range is for 3 He 4 He mixtures, see Sect. 6.8. The attraction between the carbon chains of the monolayer molecules is likely to be much diminished when the chains are in a hydrocarbon oil medium rather than air. In fact, the parameter ratios relevant to monolayers at the oil water interface lie in the range "=J ,1, where all transitions are second order.2 For z = 6 it is found that the changeover from the a-type phase pattern to b occurs at "=J = 3:6129, and from b to c at 2:8192. These values can be compared to the meaneld values of 3:8019 and 2:7747 respectively obtained in Sect. 6.8. Figure 7.2 shows pressure area isotherms calculated for z = 6 and "=J = 3:2, for which the tricritical temperature is Tt = 4:0134J . Equations 7.59 and 7.60 are used for the ordered and disordered states respectively. The area per monolayer molecule and the minimum area per monolayer molecule are denoted respectively by a and a0 so that a=a0 = ,1 e e and the surface pressure eld P = Pa0 , where P is the surface pressure. The isotherms for T Tt and T = Tt display a rst-order vapour liquid-extended transition and then, at a lower value of a, a second-order liquid-extended liquid-condensed transition. The next two isotherms display two rst-order 2 Sublattice segregation of molecules and holes is possible for "=J ,1 but is
unlikely for monolayers at a uid interface.
7.5 A Lattice Gas Model for Fluid Water
187
transitions whereas for the lowest one, the vapour liquid-extended and liquidextended liquid-condensed transitions have merged into a single rst-order transition. It can be seen that comparatively small alterations in T=J or "=J can change a rst-order liquid-extended liquid-condensed transition to a second-order one or vice versa. By using 7.62 and 7.63, values of X and Z, and hence of J and ", can e be deduced from the observed values of P and a at a second-order transition point on any isotherm, provided that a value can be assumed for a0 . The transitions on the 10, 15 and 20 C isotherms for a phospho-lipid di C1 6 lecithin monolayer at an oil n-heptane water interface were used in this way by Bell et al. 1978 who found negative " and positive J varying appreciably with temperature, "=J values being about ,3 or ,4. The temperature variation is not surprising since monolayer molecules, especially the phospho-liquids, have a large number of internal degrees of freedom so that ," and ,J are properly free energy di erences containing entropy terms rather than simple energy parameters. Since both ," and ,J increase with temperature it follows that at the oil water interface there is entropy loss involved in bringing two molecules to their minimum separation and then in adjusting their mutual orientation to that of minimum free energy. Baret and Firpo 1983 used a meaneld method on a monolayer model similar to that of this section, which they termed a `spin-1 Ising model'. They considered the e ect of incorporating a term equivalent to 6.97, which removes the symmetry between states 1 and 2. Another `spin-1 Ising model' for monolayers was treated by Banville et al. 1986. Albrecht et al. 1978 considered the possibility of a tricritical point in a monolayer model.3 The model of Nagle 1986 gives rise to an unusual kind of tricritical point. Tricritical points were emphasized by Firpo et al. 1981 whose model, like a number of other monolayer lattice gas models, involved molecular states in which more than one lattice site is occupied. A simple monomer-dimer model of this type is treated in Volume 2, Chap. 8.

The feature of the water molecule whose cooperative e ect leads to the anomalous properties of liquid and supercritical water is the hydrogen bond see Appendix A.3. A number of statistical mechanical models, attempting to incorporate this feature have been developed. The simplest of these is the twodimensional triangular lattice model of Bell and Lavis 1970 see Example 7.6. In this section we consider a somewhat more realistic three-dimensional model, based on the body-centred cubic lattice and due to Bell 1972. This is a lattice gas model of M molecules on N M sites with each molecule
3
For comments, see Bell et al. 1981, who reviewed theories of amphipathic monolayers up to that date.
188 8
7. Cluster Variation Methods 5 + 6 7 10 4 + 1 20 2
Fig. 7.3. Sites on the body-centred cubic with a water molecule placed on site 10 . Reprinted from Bell 1972, by permission of the publisher IOP Publishing Ltd.
,
3
having four bonding arms pointing to the vertices of a tetrahedron. These arms are directed towards four of the eight nearest-neighbour sites and two arms have positive polarity and two negative, implying twelve distinguishable orientations Fig. 7.3. Two nearest-neighbour molecules are connected by a hydrogen bond when a positive arm of one and a negative arm of the other lie along the segment connecting their sites. This occurs in 18 of the 144 possible orientational states of the two molecules. An open ice Ic structure see Appendix A.3, with one diamond sublattice of the body-centred cubic vacant and the other occupied by a bonded network, is thus possible. In the present uid state theory this arrangement occurs only as short-range order, but it is of crucial importance since it is the breakdown of the local open structure by increasing temperature which produces anomalous e ects. The di erence between the locally similar ice Ic diamond and ice I wurzite structures is not likely to be signi cant in the uid state. Locally ordered regions with the close-packed ice VII structure will exist at high pressures together with a range of other types of short-range order including the `cage' structure in which unbonded molecules lie in the interstices of an open bonded network. However, the types of local order in real water are considerably more complex. The reduction of the ten or so observed ice structures Eisenberg and Kauzmann 1969, Fletcher 1970 to two indicates the approximations made in neglecting hydrogen bond distortion from the tetrahedral angles and the possibility of ordered hydrogen positions. An interaction energy ," " 0 is assumed for each unbonded nearestneighbour pair of molecules which changes to ,",w w 0 when a hydrogen bond is formed. To ensure that the ice VII structure is stable only at high pressures, an energy penalty against close packing is needed. This is imposed 1 by associating an energy 3 u u 0 with the occupation by three molecules of a triad of sites where two second neighbours share a nearest neighbour for example, the triads 1210 or 10 20 1 in Fig. 7.3. There are no occupied triads in the open structure, but a total of 12N when all sites of the body-centred cubic lattice are occupied. As pointed out in connection with the one-dimensional water-like model of Sect. 5.5, density maxima occur on isobars for pressures at
189
which the open structure is stable when T = 0. By considering ground-state enthalpies it can be shown that, if " 2u " + w ; 7.69 then, at T = 0, the open ice Ic structure is stable for 0 P P0 and the close-packed ice VII structure for P0 P where P is the pressure eld and P0 = 4u , 2" : 7.70 A number density , equal to 1 for close packing, is de ned by = M=N: 7.71 A meaneld approximation would smear out the short-range order phenomena in which we are interested so we use a rst-order approximation based on a non-planar group of four sites, like 121020 in Fig. 7.3, two belonging to each of the diamond sublattices of the body-centred cubic lattice. There are four nearest-neighbour site pairs in the group, 110, 120, 210 and 220, so 1 that p = q = 4. Since z = 8, 2 z=q = 1 and 1 zp=q = 4. For uid states it 2 can be assumed that the twelve orientational molecular species are equally 1 likely, a mole fraction 12 of each being present together with a fraction 1 , of the hole species. Hence the orientational species do not have to be considered separately provided that appropriate statistical weights are assigned. Table 7.1 shows, for each group con guration `, the con guration number or weight !` , the number of molecules p` and the energy "` . The line segments in the con guration diagrams represent hydrogen bonds. As an example of the reasoning involved in deriving the weights, we shall evaluate !4 and !5 . Now, !4 + !5 = 4 144 = 576 since there are four ways of placing a nearest-neighbour pair of molecules on the group of sites and any two molecules have 144 distinguishable orientational states. Since hydrogen bonding occurs in one eighth of the orientational states of an nearest-neighbour pair, !4 = 576=8 = 72 and hence !5 = 504. There are four site triads in the group i.e. 1210, 1220, 10 201, 1020 2 in the group 121020 in Fig. 7.3 so that the ratio of nearest-neighbour site pairs to triads is 1:1. Since in the lattice as a 1 whole this ratio is 1:3 we adjust by using 3 3 u = u as the triad energy in our approximation. This ensures that the correct high-pressure ground-state energy is attained at T = 0. Using the grand distribution we have, in the notation of Sect. 7.2, i = for molecules and i = 0 for holes while the equilibrium f = ,P . A parameter is de ned by 3=2 2 = 8 = exp 2"4 , 2"8 + ; 7.72 2T 121 , 4 where the last expression follows from 7.17 with ` = 8 and r = 4. Reference to Table 7.1 shows that 2 expresses the ratio of ice VII-type to ice Ic-type con gurations on the basic group of four sites. Applying 7.17 again and substituting from 7.72 we have, for any con guration `,
190
7. Cluster Variation Methods Table 7.1. Reproduced from Bell 1972, by permission of the publisher IOP Publishing Ltd. Con guration Con guration !` "` p` index ` 1 2 3 4 5 6 7 8 9 10 1 48 288 0 0 0 0 1 2 2 2 3 3
, , ,, ,
72 504 1728 5184 648 9072
,w , " ," ,w , 2" + u ,2" + u
,2w , 4" + 4u 4 ,w , 4" + 4u 4

4
11016 ,4" + 4u
2T For this system, equations 7.3 become, using 7.73, X = 1 p`!` ` = 4 X,4 Z,1 ,2 X ; 4 f`g X 1 , = 1 4 , p` !` ` = 4 X,4 Z,1 ,2 Y ; 4 where
f`g
` = 4 p` ,2 exp 4 , p` "4 + p` , 2"8 , 2"` :
7.73
7.74
191

u
X = exp , 2" ; Y = exp , 2w ; Z = exp , T : 7.77 T T Dividing the rst relation of 7.74 by the second yields X 7.78 1 , = Y : It can be shown that, at given T , the ratio X =Y is a monotonically increasing function of . From 7.72 and 7.74 the con gurational chemical potential is given by 3
7.79 exp T = 4 X12 Y4 Z,8 12Y : X Putting f = ,P and r = 4 in equation 7.18 for the equilibrium free energy density yields h
i ,P = T ln 4 + "4 , 3 T ln 12 + ln1 , , 1 : 7.80 2 2 After a little manipulation, including adding the two relations of 7.74 to obtain an expression for 4 , we obtain 3 exp P = 5 X : 7.81 T Z X + Y 2 Equations 7.78 and 7.81 constitute an implicit equation of state giving in terms of P and T . Best results are found when u lies in the middle of the range given by 7.69. That is, when 2u = 1 w + ", which gives, from 2 7.70, P0 = w . There is liquid vapour phase separation, and density maxima occur on ; T isobars in the liquid state and in the supercritical state for P P0 , as shown in Fig. 7.4. The temperature of maximum density varies very little with pressure. A compressibility minimum also occurs on isobars in the liquid state, and the mean number of nearest neighbours of a given molecule is between 4 and 5 at low pressures Bell 1972. It can be seen that reasonable qualitative agreement with experiment is obtained in spite of the simplications inherent in the model. The same model was used by Lavis and Christou 1977 to study dielectric properties and Wilson and Bell 1978 for an investigation of aqueous solutions. Bell and Salt 1976 used the meaneld approximation on a similar model to discuss transitions to ordered ice-like states. Weres and Rice 1972 used a model based on the body-centred cubic lattice, making rather di erent assumptions and were mainly concerned with local properties. These authors had reservations about Bell's triad term u
and
X = 12X3 YZ3 + 3XZ1 + 7Y2 + 4X2 Y2 + 108X2Y1 + 3Y2 2 + 54XZ1 + 14Y2 + 17Y43 g ; Y = Z5 + 36X3YZ3 + 36X4Z1 + 7Y2 + 4X2Y2 2 + 432X5Y1 + 3Y2 3
7.75 7.76
192 1.0
7. Cluster Variation Methods 1.25 0.75 0.1
0.5
0.00024
0.01
0.0516
0.5 1.0 T=w 1.4 Fig. 7.4. Density temperature isobars for "=w = 2 and u=w = 5=4. Each isobar is labelled with the value of P=w. Reprinted from Bell 1972, by permission of the publisher IOP Publishing Ltd.
which they considered made excessively large contributions to the energies of less dense structures. Other authors who used the body-centred cubic lattice with a di erent approach were O'Reilly 1973 and Fleming and Gibbs 1974 see also the review by Stillinger 1975. A square lattice model with polarized bonds was discussed by Lavis and Christou 1979. Whitehouse et al. 1984 examined the Bell 1972 model treated above using Monte Carlo numerical methods and concluding that `the cooperativity of the model is seriously underestimated by the rst-order approximation, although the general behaviour of the model is correctly represented'. These authors consider that it would be more realistic to have u = 0, " 0, pointing out that the inequality 7.69 could still be satis ed and a positive P0 obtained. Shinmi and Huckaby 1986 have used a similar model with non-polar bonding arms and repulsive bond energy to represent carbon tetrachloride. Both the two-dimensional bonded lattice model of Bell and Lavis 1970 and the three-dimensional non-polar version of the Bell 1972 model, introduced by Shinmi and Huckaby 1986, can be expressed in Hamiltonian form as modi ed versions of the dilute Ising model introduced in Sect. 6.6.1. The modi cations consist in the introduction of certain terms involving more that two sites. Based on this approach Southern and Lavis 1980 and Lavis and Southern 1984 used nite-size cluster real-space renormalization group methods to investigate the two models.4
4
For an account of real-space renormalization group methods see Volume 2, Chap. 6.
7.6 1:1 Ordering on the Face-Centred Cubic Lattice
193

We consider an A B mixture on the face-centred cubic lattice with nearestneighbour pair energies obeying the inequality 6.21. The canonical distribution is used and it is assumed that Ma = Mb . The eld H in the equivalent antiferromagnet is thus zero. The face-centred cubic lattice is close-packed but can be divided into equal, but not interpenetrating, sublattices a and b, which are alternate planes of sites perpendicular to one of the cubic axes. The sublattices are each composed of two of the four equivalent simple cubic sublattices of the face-centred cubic lattice and hence, in the notation of Sect. 4.5, zaa = zbb = 4, zab = zba = 8. The con gurational energy of an ordered mixture with the A atoms concentrated on a sites and the B atoms on b sites will thus be lower than that of a random mixture. This layer type of sublattice ordering for an equimolar mixture will be studied in this section. In Sect. 6.5 a meaneld theory for 1:3 sublattice ordering on the facecentred cubic was developed and it was found that transitions are rst-order, which is in accordance with experiment on the Au Cu system. However, as we pointed out in Sect. 6.5, meaneld results are unsatisfactory in other ways and in particular they predict a second-order transition for an equimolar mixture. The rst-order pair method is even more unsatisfactory, and the smallest group which adequately represents the symmetry of the sublattice structure is a tetrahedron of two a sites and two b sites. This is shown in Fig. 7.5 where the full lines are the six face-centred cubic lattice edges connecting nearest-neighbour pairs in the group. It can be seen that there are one aa, one bb and four ab pairs, giving the same pair ratios as in the whole lattice. In the notation of Sect. 7.2, z = 12, p = 4 and q = 6. The various con gurations on the tetrahedral site group are shown in Table 7.2. Using 6.9 for the con gurational energy the canonical Hamiltonian is b H = E = constant , 1 Nab j"j ; 2 7.82 j"j = "aa + "bb , 2"ab 0 :
Fig. 7.5. The basic tetrahedron group
of sites on the face-centred cubic lattice. Sites on the a sublattice are denoted by and those on the b sublattice by .
194
7. Cluster Variation Methods Table 7.2. Tetrahedron site group con gurations on the face-centred cubic lattice. Con guration Con guration !` 2"` =j"j index ` aajbb 1 AA|BB 1 ,4 2 AA|AB 2 ,3 3 AB|BB 2 ,3 4 AA|AA 1 0 5 BB|BB 1 0 6 AB|AB 4 ,4 7 AB|AA 2 ,3 8 BB|AB 2 ,3 9 BB|AA 1 ,4
where s is an order parameter. Again, by symmetry in the equimolar mixture, 2 = 3 ; 4 = 5 ; 7 = 8 : 7.84 Equations 7.83 and 7.84 are connected since, by using 7.19, the relations xaa = xbb and xab = xba can be deduced from 7.84. It also follows that 1 + s = + 3 + + 2 + ; 1 2 4 6 7 2 7.85 1 , s = + + 2 + 3 + : 2 4 6 7 9 2
The constant can be disregarded and the energy ratios in Table 7.2 are simply the numbers of unlike pairs multiplied by ,1. Since xa = xb the sublattice mole fractions can be written as xaa = xbb = 1 + s ; 2 7.83 1,s ; xab = xba = 2
From 7.21, 7.83 and 7.84 the variation in the Helmholtz free energy per site can be written as a = T ln 1 + "1 1 + 4T ln 2 + "2 2 + 2T ln 4 + "4 4 +4T ln 6 + "6 6 + 4T ln 7 + "7 7 + T ln 9 + "9 9 , 3 T ln 1 + s s : 7.86 2 1,s It is useful to de ne parameters and X by
4 = 1 ;
9
j X = exp 2"j : T
7.87
195
We regard 1 and 9 as dependent variables given in terms of the independent variables 2 , 4 , 6 , 7 and s by 7.85. Then, equating @a=@ ` to zero for ` = 2; 4; 6; 7 2 = 9 X 3 ; 4 = 9 X4 2 ; 7.88 6 = 9 2 ; 7 = 9 X : Substitution in 7.85 yields the relation 1 + s = 3 + 3X 2 + X4 + 2 + X : 7.89 1 , s X 3 + X4 2 + 22 + 3X + 1 It follows that s increases monotonically with . Provided that 7.88 is satis ed, 7.86 and 7.85 yield 1 @a 1 1 , 3 ln 1 + s 7.90 @s T = 2 T ln 9 1 , s = 2 T ln ; where, from 7.89, 3 + 4 2 + 22 3 = X + 3X 2 + X4 + 3X + 1 : 7.91 3 X + 2 + X It can be seen that as given by 7.91 obeys the equilibrium condition = 1 ; 7.92 which is satis ed by = 1 for all T , with any other solutions in reciprocal pairs ; ,1 . The solution = 1 gives s = 0 by 7.89, whereas and ,1 with 1 give s and ,s s 0 respectively. We now have to determine whether the latter type of solution exists and, if so, whether it is stable and how the transition occurs. Although, numerical solutions of 7.92 are ultimately necessary, useful information can be obtained analytically in the neighbourhood of s = 0. Equation 7.89 can be rewritten in the form sinh2 2X s = cosh2 + +coshsinh4 + 2 ; 4X + X 7.93 1 1 ln = ln = 4 : Hence is an odd function of s and can be expanded in a series of odd powers of s, as can @a=@s. Obtaining the rst two terms in the expansion of , substituting in 7.90 and integrating with respect to s yields as , a0 = B2 T s2 + B4 T s4 + Os6 ; 7.94 where
9
196
7.95 1 T XX + 12 X2 , 2X + 32 5X , 2X2 , X3 , 24 : B4 T = 96 A Landau expansion has been derived for the rst-order approximation rather than, as is more usual, for the meaneld approximation. Since B2 T 0 and as thus has a minimum at s = 0 for all T 0 it follows that there can be no second-order transition at which the equilibrium value of s changes continuously from zero to non-zero values. However, B4 T changes from positive to negative as T decreases, suggesting the possibility that @a=@s may become negative for a range of s 0. If this happens, a maximum and a minimum will appear in the curve of a against s between s = 0 and s = 1, since @a=@s ! 1 as s ! 1. At the temperature when the value of a at this minimum becomes equal to a0, a rst-order transition will take place. Since a,s = as, corresponding phenomena will occur for s 0 and the transition can be to either one of a pair of values ,s; s. Numerical solution of 7.92 and calculation of the curve of a against s con rms the existence of a rst-order transition which occurs at T = 0:365j"j and takes s from zero to 0:983 Li 1949. The rst-order transition is thus of the symmetrical type as opposed to the unsymmetrical rst-order transition to a unique value of s which was found in Sect. 6.5 for 3:1 ordering on the face-centred cubic. This treatment is con ned to the xa = xb = 1 mixture, but Li 1949 used 2 a similar rst-order method, based on a tetrahedron of sites, to investigate the whole composition range. He found that all transitions are rst-order and are associated with changes in composition as in Sect. 6.5 except at 1 3 the compositions xa = 4 , 1 and 4 . There are three distinct sets of phase 2 separation curves with gaps between them and azeotropic-type maxima like 3 point P in Fig. 1.7 at xa = 1 , 1 and 4 respectively. Kikuchi 1974 treated 4 2 the whole composition range using a version of the Kikuchi method with the site tetrahedron as basic group. The chief di erence between his phase diagram and Li's is the disappearance of the gaps. Danielian 1961 discovered using the language of the antiferromagnet that the completely ordered layer sublattice structure is not a unique ground state for the equimolar mixture see also Phani et al. 1980. There are 0 states of minimum energy where ln 0 =2 N 1=3 so that N ,1 ln 0 ! 0 as N ! 1. It is di cult to predict the e ect on an exact solution since the situation di ers from that in the exactly solved triangular lattice antiferromagnet see Sect. 8.12. In the latter, the ground state is also in nitely degenerate but 0 is so large that N ,1 ln 0 tends to a non-zero limit as N ! 1 and there is a zero-point entropy. It is interesting to note Binder 1987 that Monte Carlo methods give a phase diagram like Li's but that with the introduction of second-neighbour interactions it becomes more like Kikuchi's. In the Cu Au alloy there are phase separation curve maxima at the Cu3 Au, CuAu and CuAu3 compositions, but the CuAu3 transition temperature is much lower than that for Cu3 Au, indicating that the model's A B
B2 T = 1 T XX2 , X + 1 ; 2
7.7 Homogeneous Cacti
197
symmetry is unrealistic. For the equimolar alloy, the layer sublattice structure, known as CuAuI, involves an atomic distribution in the layer planes quite di erent from that in planes perpendicular to the layers. This causes a change in lattice symmetry type from cubic to tetragonal, a-a or b-b nearestneighbour distances becoming larger than a-b distances. This change may stabilize the layer structure relative to other low-energy con gurations. For an interval above the temperature of transition to CuAuI, an orthorhombic structure, known as CuAuII, is stable. This corresponds to one of the alternative minimum energy states and resembles CuAuI, except that at every fth unit cell in one lattice direction parallel to the layers, the planes occupied by Cu and Au are interchanged, creating a `stepped' layer structure, Irani 1972, Parsonage and Staveley 1978. The lattice retains its cubic symmetry when the 3:1 or 1:3 sublattice structures are stable, near the Cu3 Au or CuAu3 compositions.

Rushbrooke and Scoins 1955 adapted the Mayer cluster expansion approach to lattice systems and found that with all closed circuits of nearest-neighbour pairs excluded from consideration their results were equivalent to those of the rst-order pair approximation. This implies Domb 1960 that the pair method is accurate for Bethe lattices. It is convenient at this point to introduce some graph theory terminology,5 identifying sites with vertices and lines connecting the members of a nearest-neighbour pair with edges, the valency of a vertex being the number of edges incident there. A Cayley tree is a graph with no closed circuits of edges, and the Bethe lattice is a Cayley tree homogeneous in the sense that all except the outer vertices have the same valency z. The case z = 3 is illustrated in Fig. 7.6 where the outer vertices are indicated by circles. In a regular lattice of N sites with a `reasonable' shape the number Nb of outer or boundary sites is proportional to N 1=2 for d = 2 and N 2=3 for d = 3 so that Nb =N ! 0 in the thermodynamic limit N ! 1. This is not so for the Bethe lattice. Suppose that there are m edges between a site of the central pair and an outer site in a graph of the type shown in Fig. 7.6 where m = 2. Then Nb = 2m+1 ; 7.96 N = 2 + 22 + + 2m+1 = 22m+1 , 1 and hence Nb=N ! 1 as N ! 1. For general z, this limiting ratio is z , 2 2=z , 1 so that the limiting value of Nb=N is zero only for the z = 2 Bethe lattice which is just the open linear lattice of Sect. 2.4.1. For any Bethe
5
For detailed accounts of graph theory terminology see Temperley 1981 and Volume 2, Appendix A.7.
198
Fig. 7.6. Cayley tree with z = 3.
lattice the total number of edges is N , 1 so that in the thermodynamic limit the mean coordination number or valency is 2. This results from the high proportion of sites which are outer vertices with valency unity. For z 2, a Bethe lattice system is thus likely to have very di erent properties from those of a regular lattice system. This is con rmed by Eggarter's 1974 derivation of the partition function for the zeroeld Ising model. Suppose we rst sum over the spin variables i for the outer sites i.e. those indicated by circles in Fig. 7.6, each of which is connected by a single edge to a site in the next layer. Since 2.77 applies, a factor 2 coshJ=T is obtained for each outer site irrespective of the spin variable values for the next layer. Summing over the latter again produces a factor 2 coshJ=T for each site, and so on till the relation 2.79 is reproduced for the Bethe lattice with any value of z. This implies a continuous heat capacity and no spontaneous magnetization for T 0 which is at variance with the pair method results obtained in Sect. 7.3 and appears to contradict the inference mentioned at the beginning of the present section. However, in a treatment of the nearestneighbour exclusion model on a Bethe lattice, Runnels 1967 pointed out that it is results for interior sites deep within the lattice which can be obtained from the rst-order pair method. These are quite di erent from results obtained by averaging over all sites of the lattice. Using Eggarter's partition function for the zeroeld Ising model is equivalent to averaging over the entire Bethe lattice. Baxter 1982a showed that for this model, pair method relations are applicable to sites deep within the lattice in the limit N ! 1. We now generalize by showing that the rst-order theory of Sect. 7.2 is exact for interior sites of a lattice equivalent to a graph which may be termed a homogeneous Husimi tree or homogeneous cactus. This consists entirely of gures identical to the site group of Sect. 7.2, a central group being surrounded by m layers of other groups connected so that all vertices except the outer ones of the last layer have the same valency z. When the group is simply a pair of vertices joined by an edge, the Bethe lattice of Fig. 7.6 is regained. Figure 7.7 shows a cactus composed of squares of four sites with z = 4 and m = 2. The squares are shaded, and the decrease in their size is due to the exigencies of drawing on a plane. Temperley 1965 considered a nearest-neighbour exclusion model on a cactus of triangles using the
199
Fig. 7.7. Cactus of squares of four sites with

z = 4.
Rushbrooke-Scoins technique. For the general group, using the notation of Sect. 7.2, zp=2q groups are attached at each inner vertex. For a polygon group this number reduces to 1 z, since p = q, and two edges from each group 2 are incident at the vertex. In Fig. 7.7, with z = 4, each inner vertex belongs to just two groups. We construct a partition function for the -species model of Sect. 7.2, using the = n , 1 distribution. The partition function is based on the central group. Each site of this group belongs to zp=2q , 1 other groups and may be regarded as a `root' for zp=2q , 1 sub-cacti of m layers, each joined to the central group at the root. Let gj m be the partition function for a subcactus of m layers with the root occupied by a member of species j , the summation being taken over all possible occupations of the subcactus sites apart from the root. Then, for the entire cactus, the partition function is
Z=
X
f`g
!` exp,"` =T
Yn
j =1
gj m 2q ,1 expi =T
zp
opj`
7.97
where the summation is over all con gurations on the central group of sites. The ratio of the probabilities of con gurations of types ` and r respectively is thus given by
` r =
exp,"` =T exp,"r =T
Yn
j =1 Yn j =1
gj m 2q ,1 expi =T
zp
opj`
opjr zp gj m 2q ,1 expi =T

j = 1; : : : ; , 1:
7.98
Now de ne j m = gj m=g m; Then, from 7.98 and 7.99,
7.99
200
r
7. Cluster Variation Methods Y ` = exp "r , "` ,1 exp j ,
The partition function for the entire cactus can be constructed in another way, based on one particular site of the central group for example, that indicated by a black circle in Fig. 7.7. As before, this site is a root for zp=2q , 1 subcacti of m layers. However, it can also be regarded as a root for a subcactus of m + 1 layers, the rst layer being the central group itself. Hence
j =1
,1 j m 2 q
zp
pj` ,pjr
: 7.100
Z=
X j =1
zp exp j =T gj m 2q ,1 gj m + 1 :
7.101
We now suppose that j m ! , where is nite, when m ! 1. Equaj j tions 7.100 and 7.102 become respectively
It follows that zp ,1 xj = exp j , j m 2q j m + 1; j = 1; : : : ; , 1 : 7.102 x T
Substitution of 7.104 into 7.103 then yields 7.17 again so that the latter is an exact result for the central group of a homogeneous cactus. By adapting this method it can be shown that 7.17 applies to any group in layer n, provided that n is kept xed as m ! 1. This condition, in fact, de nes an interior group and we have thus shown that rst-order results are exact for the interior of a homogeneous cactus. Equations 7.103 and 7.104 are similar to the basic relations of the Bethe approach, though the latter was developed as an approximation for regular lattices. In the Bethe method a factor Zz is postulated in the probaj bility of occupation of a site by a member of species j to allow for the e ect of the lattice environment. For a site of a group it is assumed that the corresponding factor is Zz,2q=p , z , 2q=p being the number of edges incident j at the site from outside the group. Equations 7.103 and 7.104 result if we substitute for Zj =Z 2q=p in the Bethe relations. j
Y ` = exp "r , "` ,1 exp j , zp ,1 pj` ,pjr ; 2q j r T T j , j=1 zp xj 2q ; j = 1; : : : ; , 1 : j x = exp T
7.103 7.104
Examples
7.1 Apply the pair approximation to the Ising model ferromagnet on the d = 1 lattice by putting z = 2 in Sect. 7.3. Show that, for all T and H, 7.31 gives only one physically meaningful value for which is
= 21 Y2 , 1 + Y2 , 12 + 4X2 Y2 1=2 ; X where Y = expH=T . Hence show that the pair approximation yields the accurate relation 2.99 between m and H. 7.2 With de ned by 7.32, show that d = 1 + X z,2 X , 12 + X + 1zX , z + 2 : d X + z Hence show that when X z , 2=z the only solution of the equation = 1 for positive is = 1 but that a pair of other solutions appears when z , 2=z.
7.3 For the Ising model ferromagnet in the pair approximation, show that the internal energy per site J 1 , X u = , z21 + X ; when H = 0 and X z , 2=z. Hence show that the corresponding zeroeld con gurational heat capacity per site is given by J=T 2 du = 2kb zX J 2 = 1 k z c0 = kb dT 1 + X2 T 2 b coshJ=T : Show that c0 is a monotonically decreasing function of T when X z , 2=z and z 3 and that z2 , 2 z 2 c0 ! kb 8zz, 12 ln z , 2 as X ! z , 2 + 0: z
Examples
201
7.4 Apply the pair approximation to the Ising model ferromagnet on the square lattice by putting z = 4 in Sect. 7.3. Show that 7.33 has solutions = 1 for all temperatures, but that for X Xc = 1 2 there is a pair of positive real solutions 0 ; ,1 for which 0 0 + ,1 = 1 , 3X2=X3 : 0 1 Hence show, using 7.29, that for X 2 , u 8 4 J = , 1 , X2 + 1 , 2X2 + 2 ; and 1 J 2 1 2 c0 = 32kb X 1 , 2X22 , 1 , X2 2 T :
202
By using these results with those of Example 3, show that u is con1 tinuous at X = Xc = 2 and nd the magnitude of the discontinuity in c0 . 7.5 When " ,J , show from 7.61 and 7.62 that ! 0 as T ! 0 on the critical curve. When " ,J , show that ! c 0 as T ! 0 on the critical curve, where c = 2z , 1=z2 , 2 for " = ,J and c = z , 1=2z , 3 for " ,J . It is to be assumed in all cases that J 0. 7.6 In the two-dimensional water-analogue lattice gas model of Bell and Lavis 1970, mentioned in Sect. 7.5, each molecule has three identical bonding arms in the triangular lattice plane at angles of 2=3 to each other. These bonding arms point to three of the six nearest-neighbour sites so that the molecule has two distinguishable orientations. When the bonding arms of a pair of nearest-neighbour molecules lie on the same line, a bond is formed and the pair interaction energy is ,"+w where " 0 and w 0. For an unbonded nearest-neighbour pair the interaction energy is ,", so that w is the energy associated with a bond. The reader can verify, with the aid of a sketch, that it is possible to form a completely bonded network on a honeycomb sublattice of the triangular lattice. The open ground state consists of such a network with the remaining one third of the lattice sites vacant. There are a large number of close-packed ground states with all lattice sites occupied, but in all of them, exactly one of the three nearestneighbour pairs on every triangle of sites is bonded. Show that at T = 0 the open ground state is stable for P w , 3", where the e two-dimensional pressure eld P = Pa0 , a0 being the area per site. Apply a rst-order method, with a triangle of sites as basic group, to this model, adopting a similar approach to that of Sect. 7.5, where the d = 3 model is considered. Let be the ratio of the probability of a triangle con guration with all three sites occupied and one pair of molecules bonded to that of a con guration with one site vacant and the other two occupied by a bonded pair. Also, let X = exp,"=T ; Y = exp,w=T : Show that the number density is given by = X ; 1, Y where X = 2 3 + Y2 + 1 + 3Y + YX,1 ;
Y = 1 + 3Y2 + 4YX,1 + YX,3 ;
Examples
203
and that the chemical potential and two-dimensional pressure respectively are given by
= ,6" , T 2 ln 21 , , 3 ln ; P = T ln Y 1 , + w , 3" :
8. Exact Results for Two-Dimensional Ising Models
8.1 Introduction
We already have exact results for the one-dimensional Ising model Chap. 2 and Sect. 4.1 and exact transcriptions between models, for instance from the zeroeld Ising ferromagnet to the zeroeld antiferromagnet Sect. 4.2 or from the Ising model to the simple lattice uid or two-component mixture Chaps. 5 and 6. In Sects. 8.2 8.6, dual and star-triangle transformations are derived, connecting the partition function for the zeroeld Ising model on a given lattice to that of the zeroeld Ising model on a related lattice at a di erent temperature.1 Some associated relations connecting nearest-neighbour correlations are obtained, a particularly important one being derived in Sect. 8.6. We then derive exact results for critical conditions and thermodynamic functions for the two-dimensional zeroeld Ising model. There are a large number of approaches to this, all mathematically intricate.2 The original method of Onsager 1944, Kaufman 1949 and Kaufman and Onsager 1949 depended on transfer matrices, of which the 2 2 matrix appearing in the theory of the one-dimensional Ising model in Sect. 2.4 is a simple example. Other workers have employed the Pfa an see Volume 2, Chap. 8, Green and Hurst 1964, McCoy and Wu 1973 or combinatorial methods Kac and Ward 1952, Vdovichenko 1965a, 1965b. We shall use a method due to Baxter and Enting 1978, which is closely related to the preceding transformation theory and has the advantage of giving results simultaneously for the square, triangular and honeycomb lattices. The exact expressions for con gurational energy and heat capacity depend on elliptic integrals and these protean quantities appear in the literature in a number of di erent forms. We have tried to use expressions valid for the whole temperature range and closely related to the formulae given for the partition function or free energy. For most of this chapter we assume, for de niteness, a ferromagnetic interaction. However in the last section, the antiferromagnet is discussed, especially the antiferromagnet on the triangular lattice. This lattice is close1 2
Such transformations are discussed by Syozi 1972 and Baxter 1982a. There is an excellent short survey of this work by Temperley 1972. For other applications of transformation methods see Maillard 1985.
206
packed so the behaviour of the zeroeld antiferromagnet cannot be deduced from that of the ferromagnet and it displays some curious properties. Exact solutions exist for various two-dimensional models apart from the Ising model and are considered in Chaps. 9 and 10 and Volume 2, Chaps. 5 and 8. However, exact methods have their limitations. There are no exact solutions for three-dimensional lattices and, not even for the two-dimensional Ising model in non-zero eld.
8.2 The Low-Temperature Form and the Dual Lattice

Consider the zeroeld Ising model ferromagnet whose Hamiltonian is given by 2.66 with H = 0. This can also be written as
b H = E = ,Ne J + J
n:n: X
where Ne is the number of nearest-neighbour site pairs, which may be re1 placed by 2 zN in the thermodynamic limit. Each unlike nearest-neighbour pair of spins makes a contribution 2J to the sum in 8.1 whereas a like nearest-neighbour pair gives a zero contribution. Substitution into 2.67 gives the con gurational partition function in the form X Z = expK Ne exp,2KNu ; 8.2 gurations and Nu denotes the number of unlike nearest-neighbour pairs. Each term involves a power of the variable exp,2K , which is zero at T = 0 and unity at T = 1. Since this variable is small at low temperatures, 8.2 is a low-temperature form. If terms in Z or ln Z are arranged in increasing powers of exp,2K then we have a low-temperature series see Volume 2, Chap. 7. The terms of the sum in 8.2 can be given a geometrical interpretation by using the idea of a dual lattice. First note that the graph theory terminology introduced for the Bethe lattice in Sect. 7.7 can be used for the regular two-dimensional lattices shown in Appendix A.1. Each site is a vertex, the members of each nearest-neighbour pair are connected by an edge and the interior of each elementary polygon of edges for example, a square of four edges for the square lattice is a face. The resulting graph is termed the full lattice graph. The dual lattice corresponds to the graph theoretical dual Temperley 1981 of the full lattice graph. It is obtained by placing a point
3
fi;jg
1 , i j ;
8.1
where, as in 2.1463 K = J=T , the sum is over all spin con
Parameters formed from ratios of energies or elds to temperature are often called couplings. These will be used extensively in this chapter and Chap. 9. An alternative formulation of thermodynamics to the eld-density representation of Sect. 1.8 is a coupling-density representation see, for example, Volume 2, Sect. 1.4..
8.2 The Low-Temperature Form and the Dual Lattice
207
a b Fig. 8.1. Dual lattices: a square self-dual; b honeycomb-triangular.
representing a dual lattice site in each face of the original lattice. Points in adjacent faces of the original lattice form a nearest-neighbour pair of sites in the dual lattice and are connected by a dual lattice edge, shown by a broken segment in Fig. 8.1. It can be seen from Fig. 8.1a that the square lattice is self-dual, since its dual is also a square lattice. The dual of the honeycomb lattice is the triangular lattice, as can be seen from Fig. 8.1b, and since the duality relation is a reciprocal one, the dual of the triangular lattice is the honeycomb lattice. Denote the numbers of sites and edges respectively in the original lattice by N and Ne , and the number of sites in the dual lattice by N . From the method of construction, Ne is also the number of edges in the dual lattice. Since N is the number of faces of the original lattice, Euler's relation for planar graphs Temperley 1981 yields4 N + N = Ne : 8.3 Take a distribution of positive and negative spins on the original lattice, each site of which lies in a face of the dual lattice with nearest-neighbour pairs occupying adjacent faces of the dual lattice. When the spins of a nearestneighbour pair are unlike, replace the broken line segment representing the dual lattice edge separating them see Fig. 8.2 by a full line segment. Thus each spin distribution on the sites of the original lattice is represented by a set of lines on the edges of the dual lattice. Figure 8.2 gives an illustration for the self-dual square lattice. The lines and the vertices they connect form a con guration graph which is a subgraph of the full dual lattice graph. The form of the con guration graphs can be deduced by noting that when the sites on the boundary of any face of the original lattice are traversed in order, there is either no change of spin sign or an even number of changes. Hence the number of lines incident at the dual lattice vertex inside the face is either zero or even. Apart from sequences terminating at the lattice boundary, the lines of any con guration graph thus form closed polygons see Fig. 8.2. Reversal 4 For nite plane graphs, the right-hand side of 8.3 is N + 1 or Ne + 2, but in e the thermodynamic limit of very large N and N we can use Ne .
208 + +
8. Exact Results for Two-Dimensional Ising Models Fig. 8.2. A spin con guration on the + , , , + square lattice and the corresponding con guration graph on the dual lattice. , + , , + + +
, , ,
+
,
+
+ +
, ,
+
+ +
,
+
,
+
, ,
, ,
of all spins on the original lattice sites does not change the distribution of like and unlike nearest-neighbour pairs, and hence leaves the polygon graph on the dual lattice unaltered. Thus each polygon graph corresponds to a pair of spin distributions. Putting Nu = , where is the number of lines in a polygon graph, 8.2 can be expressed in the form exp,2K ; where the summation is over all polygon graphs.
Z = 2 expK Ne
p:g: X
8.4
8.3 The High-Temperature Form and the Dual Transformation

Since the product i j can take only the two values 1, expK i j = coshK + i j sinhK = coshK 1 + i j ; 8.5 where = tanhK : 8.6 Thus, when H = 0, the con gurational partition function 2.67 for the Ising model can be expressed as
Z = coshK N e
X n:n: Y
fi;jg
1 + i j :
8.7
Each of the products in 8.7 contains Ne factors and can be expanded binomially into 2N e terms. However, many of these terms vanish when the sum over spin values is performed. For instance, a linear term i j disappears when either i or j is summed over the values 1. The condition for a term
8.3 The High-Temperature Form and the Dual Transformation
209
to have a non-zero value after the summation over spin values is easily understood geometrically. Suppose we draw the lattice with broken segments for edges. Now, after binomial expansion, a typical term in 8.7 can be written as i1 j1 i2 j2 in jn n 8.8 and can be represented by replacing the broken segments corresponding to the nearest-neighbour pairs i1 j1 ; i2 j2 ; : : : ; in jn by full line segments. If the nearest-neighbour pairs have no sites in common then the term vanishes after summation. The simplest type of non-vanishing term on the square lattice is 2 2 2 i j j ` ` m m i 4 = i2 j ` m 4 ; 8.9 where the lines form an elementary square. Summation over i , j , ` and m yields a factor 16, and summation over the spin variables not present in this term yields a further factor 2N ,4 so that nally 2N 4 is obtained as the contribution from a single square. It is now not di cult to derive the general rule for non-vanishing terms. Each spin variable i must be present as an even power which means that either no lines or an even number of lines meet at each lattice site. Thus only polygon graphs as de ned in Sect. 8.2 contribute to 8.7, which can be expressed as
; Z = coshK N e 2N 8.10 where denotes the total number of lines composing a given graph. In this
p:g: X
context, polygon graphs are referred to as zeroeld graphs Volume 2, Appendix A.7. The variable = tanhK = tanhJ=T is unity at T = 0 and zero at T = 1. Since is thus small at high temperatures, 8.10 is a hightemperature form. If the terms in Z or ln Z are arranged in ascending powers of we have a high-temperature series see Volume 2, Chap. 7. In 8.4 the low-temperature form was expressed in terms of a sum over certain sets of lines or graphs on the dual lattice. Except at the boundaries, the rules for permissible sets of lines are exactly the same as for the hightemperature form 8.10 for an Ising ferromagnet on the dual lattice. Thus for large N and N , that is in the thermodynamic limit, the sum in 8.4 is equal to the sum in 8.10 for the dual lattice Ising model, provided that exp,2K = tanhK ; 8.11 where K denotes the value of K for the dual lattice. This relation, which is meaningful only for positive K and K , implies that exp,K Ne Z N; K = 2,N coshK ,N e Z N ; K ; 8.12 where we have introduced the dependence of Z on N and K explicitly, and Z is the con gurational partition function for the dual lattice. A factor 1 , which 2 has a negligible e ect in the thermodynamic limit, is omitted from the lefthand side of 8.12. Both 8.11 and 8.12 can be expressed in symmetrical
210
form, as would be expected from the symmetrical nature of the lattice-dual relation. From 8.3 and 8.11, sinh2K sinh2K = 1 ; 8.13
2N coshK N e exp,K Ne = 2 sinh2K N =2 2 sinh2K ,N =2
8.16 i j n:n: N @ K Nu J zJ e where u is the con gurational energy per site. Combining this result with 8.15 see Example 8.1 gives 8.17 coth2K coth2K which is a relation between nearest-neighbour correlations. For the self-dual square lattice, 8.15 becomes sinh2K ,N =2 Z N; K = sinh2K ,N =2 Z N; K : 8.18 = K c then, from 8.13, K c is given by If K = K sinh2 2K c = 1 : 8.19 The transformation 8.13 connects any value of K in the high-temperature interval 0; Kc with one in the low-temperature interval K c ; 1 , and vice versa. A critical point arises from a singularity in Z , and thus from 8.18 a critical point in 0; K c implies a critical point in K c; 1 . Thus, if there is a unique critical point for the square lattice, it must correspond to p K = Kc ; Tc = J=K c = 2J= ln1 + 2 : 8.20 Intuitively a single critical temperature seems likely since for the Ising ferromagnet there is only one simple type of long-range order and this corresponds to the T = 0 ground state. For a proof, see Sect. 8.9. Substituting K = K = K c in 8.17, the critical point values of the nearest-neighbour correlation and the con gurational energy per site for the square lattice are given by uc = , 1 zJ h i j in:n: c = ,J coth2K c ; 8.21 2 where z = 4. Numerical values for Tc and uc are given in Table 8.1.
and 8.12 becomes 2 sinh2K ,N =2 Z N; K = sinh2K ,N =2 Z N ; K : From 2.20 and 2.66 , U = , 2u = h i = 1 @ ln Z ;
8.14 8.15
i j in:n: +
i j i :n: = 1 ; n
8.4 The Star-Triangle Transformation Table 8.1. Critical parameters for the Ising model ferromagnet of dimension d and coordination number z =. Results for d = 3 are those obtained by series methods and are taken from Domb 1974. The rst-order pair values can be derived from formulae given in Sect. 7.2. 1 Lattice d z Tc =zJ ,uc = 2 zJ ln2 , sc = ln2 Honeycomb 2 3 0.5062 0.7698 0.6181 Plane square 2 4 0.5673 0.7071 0.5579 Triangular 2 6 0.6068 0.6667 0.5235 Diamond 3 4 0.6761 0.4370 0.2640 Simple cubic 3 6 0.7518 0.3308 0.1951 Body-centred cubic 3 8 0.7942 0.2732 0.1603 Face-centred cubic 3 12 0.8163 0.2475 0.1485 Mean- eld 1.0 0.0 0.0 First-order pair 3 0.6068 0.5000 0.2831 First-order pair 4 0.7213 0.3333 0.1634 First-order pair 6 0.8221 0.2000 0.0871 First-order pair 8 0.8690 0.1429 0.0591 First-order pair 12 0.9141 0.0909 0.0358
211
8.4 The Star-Triangle Transformation

The dual transformation maps the triangular lattice into the honeycomb and vice versa, and hence does not yield a high-temperature low-temperature relation like 8.13. However such relations can be obtained by using the star-triangle transformation together with the dual transformation. The startriangle transformation is obtained as follows. First note from Appendix A.1 that the honeycomb lattice is loose-packed and divides into two triangular sublattices, whose sites are represented by white and black circles respectively in Fig. 8.3. Consider a particular white site and denote the spin variable there by . The spin on the white site interacts only with the spins on the three neighbouring black sites, where the spin variables are denoted by ; and . These interactions give a factor exp G + + ; 8.22 where G = J=T honeycomb 8.23 in each term of Z . In zero eld this is the only factor involving , and summing over = 1 gives X exp G + + = 2 cosh G + + : 8.24 The star-triangle transformation depends on de ning parameters R and K such that, for = 1; = 1; = 1, 2 cosh G + + = R exp K + + : 8.25
=1
212
Fig. 8.3. The honeycomb lattice, sublattices and the startriangle transformation.
K G
This is possible because 8.25 corresponds to only two independent relations. Since both sides are invariant under a simultaneous change of sign of , and and any permutation of ; and , these two relations are obtained by putting = = = 1 and = = , = 1 respectively giving 2 cosh3G = R exp3K ; 8.26 2 coshG = R exp,K : Eliminating R and K successively, exp4K = 2 cosh2G , 1 ; 8.27 R4 = 16 cosh3G cosh3 G = exp4K exp4K + 3 2 : The right-hand side of 8.25 corresponds to an Ising model with interaction parameter K between nearest-neighbour pairs on the triangular sublattice of black sites, such neighbouring pairs being represented by the broken segments in Fig. 8.3. It follows that if, in the con gurational partition function for a honeycomb lattice of 2N sites, we sum over all spin variables corresponding to white sublattice sites and take the thermodynamic limit so that boundary e ects can be disregarded then we obtain the star-triangle transformation relation Z h 2N; G = RN Z t N; K : 8.28 Here `H' and `T' refer to honeycomb and triangular lattices respectively and the parameters G, K and R are connected by 8.27. Now suppose that the dual transformation is performed on the triangular lattice Ising model, mapping it into a honeycomb lattice Ising model. From 8.15, using K 0 and G in place of K and K respectively, 2 sinh2K 0 ,N =2 Z t N; K 0 = 2 sinh2G ,N Z h 2N; G ; 8.29
If the star-triangle transformation is now performed on the honeycomb lattice, the triangular lattice is recovered and, combining 8.28, 8.29 and 8.30, sinh2K 0 ,N =2 Z t N; K 0 = 2,N =2 sinh2K 0 N RN Z t N; K : 8.31 Eliminating G between 8.30 and the rst relation of 8.27 gives exp4K , 1 exp4K 0 , 1 = 4 : 8.32 Equation 8.31 can now be expressed in symmetrical form. Using 8.32 and the second relation of 8.27, sinh2K 0 4 R4 = 4= sinh2K 2 8.33 and, substituting in 8.31, sinh2K ,N =2 Z t N; K = sinh2K 0 ,N =2 Z t N; K 0 : 8.34 This is identical in form to the square lattice dual transformation relation 8.18, but the two K values are now connected by 8.32 instead of 8.13. Like 8.13, equation 8.32 relates values of K in a high-temperature interval 0; K c with those in a low-temperature interval K c ; 1 where K c is now given by exp4Kc , 1 2 = 4 : 8.35 By similar reasoning to that used in the square case, if there is a unique critical point it must be at K = K c where the two intervals meet and the critical temperature is then Tc = J=K c = 4J= ln3 : 8.36 By a star-triangle transformation followed by a dual transformation it can be shown Example 8.2 that, for the honeycomb lattice, sinh2G0 ,N =2 Z h N; G0 = sinh2G ,N =2 Z h N; G ; 8.37 where cosh2G , 1 cosh2G0 , 1 = 1 : 8.38 The meeting point of the high- and low-temperature intervals connected by 8.38 is at G = Gc, where cosh2Gc , 1 2 = 1 8.39 and, if there is a unique critical point, it occurs at temperature p Tc = J=Gc = 2J= ln2 + 3 : 8.40 It can be veri ed Example 8.3 that 8.21 for the square lattice critical point energy applies also for both the triangular and honeycomb lattices. Numerical values for Tc and uc are given in Table 8.1.
8.4 The Star-Triangle Transformation 213 since in this case N = 2N . By 8.13, K 0 and G satisfy the relation sinh2K 0 sinh2G = 1 : 8.30
214
8.5 The Star-Triangle Transformation with Unequal Interactions

In the next three sections, following Baxter and Enting 1978, we develop a functional equation for the nearest-neighbour correlation in two-dimensional lattices. A necessary preliminary is to consider the star-triangle transformation when the interactions are di erent in the three honeycomb lattice directions so that the parameter G of Sect. 8.4 is replaced by G1 , G2 and G3 as shown in Fig. 8.4. Equation 8.25 is now generalized to 2 cosh G1 + G2 + G3 = R exp K 1 + K 2 + K 3 : 8.41 As parameter K has to be replaced by K 1 , K 2 and K 3 , which correspond to the three triangular lattice directions, there are four unknowns on the righthand side of 8.41. However, unlike 8.25, 8.41 is not invariant under permutations of , and and is equivalent to four independent relations which may be obtained by putting = = = 1, , = = = 1, = , = = 1 and = = , = 1 respectively. It is useful to de ne x = cosh G1 + G2 + G3 ; xi = cosh G1 + G2 + G3 , 2Gi ; i = 1; 2; 3; 8.42 and then 8.41 gives R expK 1 + K 2 + K 3 = 2x ;
X = pxx1 x2 x3 ;
R expK 1 , K 2 , K 3 = 2x1 ; R exp,K 1 + K 2 , K 3 = 2x2 ; R exp,K 1 , K 2 + K 3 = 2x3 :

K3 G1 K2 G2 G3 K 1
8.43
Fig. 8.4. The star-triangle transformation with

unequal interactions.
8.5 The Star-Triangle Transformation with Unequal Interactions
215
Taking i; j; ` to be a permutation of 1; 2; 3, the parameters on the righthand side of 8.41 can now be expressed in terms of G1 , G2 and G3 by xx exp4K i = x xi ; i = 1; 2; 3; j ` 8.44 p R = 2 X: The aim is now to express 8.44 in an alternative form and to obtain inverse relations giving the honeycomb parameters Gi in terms of the triangular parameters K i . Let ci = cosh2Gi ; i = 1; 2; 3: 8.45 si = sinh2Gi ; Standard hyperbolic function relations then give 1 xxi = 1 cosh2Gi + cosh2Gj + 2G` = 2 ci + cj c` + sj s` ; 2 8.46 1 cosh2G + cosh2G , 2G = 1 c + c c , s s : xj x` = 2 i j ` 2 i j ` j ` By putting i = 1, j = 2, ` = 3 in 8.46 and substituting in the expression for X given by 8.42, q 8.47 X = 1 2c1 c2 c3 + c2 + c2 + c2 , 1 : 1 2 3 2 From 8.43 R2 exp2K i = 4xxi ; 8.48 R2 exp,2K i = 4xj x` and then, using 8.46, it is easy to prove that js sinh2K i = s2X` ; i = 1; 2; 3: 8.49 ci + cj c` ; cosh2K i = 2X It can now be shown Example 8.5 that cosh2Gi = cosh2K j cosh2K ` + sinh2K j sinh2K ` coth2K i ; i = 1; 2; 3 : 8.50 This is the inverse relation giving Gi in terms of K 1 , K 2 and K 3 . From the rst relation of 8.49, sj s2 sinh2K i sinh2Gi = si2Xs` = s12Xs3 ; 8.51 since i; j; ` is a permutation of 1; 2; 3. The last expression is clearly independent of i and hence 8.51 can be written
216
8.52 which is a symmetrical form of the relation between the honeycomb and triangle parameters. The last expression of 8.51 gives k,1 in terms of the honeycomb parameters Gi . Before obtaining k in terms of the triangle parameters we de ne i = cosh2K i ; c i = 1; 2; 3: 8.53 i = sinh2K i ; s Then, putting i = 1 in 8.52 and 8.50, 1 k2 = + c , + + : 8.54 c2 c3 s1 s2 s3 1 s1 c2 c3 s1 s2 s3 c1 s1 Using standard hyperbolic function relations, k can then be expressed in symmetrical form as 2 1 2 , 2 3 k = 4 + , 1 1 2 1 ,1+ ; 8.55 1 2 3 2+ 3 1 3+ 1 2 1 2 3 1=2 where 8.56 i = tanhK i :
8. Exact Results for Two-Dimensional Ising Models sinh2K i sinh2Gi = k,1 ; i = 1; 2; 3;
8.6 A Linear Relation for Correlations

Let p ; ; be the probability that the spin variables on the triangle of sites shown in Fig. 8.3 or Fig. 8.4 have given values , and respectively. Then Z ; ; ; p ; ; = 8.57 Z where Z ; ; denotes the part of the partition function Z corresponding to the given , and . It is important, in this section and the next one, to note that p ; ; and indeed the distribution probability for any group of black sites is not a ected by a star-triangle transformation since the latter involves summation over white site spin variables only. If p ; ; ; similarly denotes a probability distribution for the set of four sites p ; ; ; = p j ; ; p ; ; 8.58 where p j ; ; is the conditional probability of taking a given value for prescribed , and . Since spin interacts with spins , and only and since p1j ; ; + p,1j ; ; = 1, exp G1 + G2 + G3 p j ; ; = 2 coshG1 + G2 + G3 = 1 1 + tanhG1 + G2 + G3 : 8.59 2
8.6 A Linear Relation for Correlations
217
Let
tanhG1 + G2 + G3 = w1 + w2 + w3 , w ; 8.60 where w1 , w2 , w3 and w are parameters determined below. Equation 8.60 is invariant under a simultaneous change of sign of , and and hence, like 8.41 above, is equivalent to four independent relations. In terms of the variables y = sinhG1 + G2 + G3 ; 8.61 yi = sinhG1 + G2 + G3 , 2Gi ; i = 1; 2; 3; these are y = xw1 + w2 + w3 , w ;
y1 = x1 ,w1 + w2 + w3 + w ; y2 = x2 w1 , w2 + w3 + w ;
evaluated. From 8.62, 4w = , y + y1 + y2 + y3 x x x x
1 2 3
8.62
y3 = x3 w1 + w2 , w3 + w ; where x and xi are de ned by 8.42. The coe cients w and wi must now be
= ,yx1 , y1 xx2 x3 + y2 x3 + y3 x2 xx1 : 2
Using 8.46 with standard hyperbolic function relations,

2s
3s
1s s w = s1 xx1 ,2 x2 x3 = s1 s22 3 = s 2Xs s2X3 s2X1 s2X2 : 8.64 4X 4X 1 s2 3 Similarly, w1 = 1 y , y1 + y2 + y3 = s1 c1 +2c2 c3 = wc1s+ c2 c3 : 8.65 4 x x1 x2 x3 4X 2 s3 ,1 , using 8.49 and, generalizing Then the last expression of 8.51 gives k from w1 to wi , we have w = k sinh2K 1 sinh2K 2 sinh2K 3 ; 8.66 wi = w coth2K i ; i = 1; 2; 4 : Combining 8.58, 8.59 and 8.60, 1 p ; ; ; = 2 1 + w1 + w2 + w3 , w p ; ; : 8.67 Now multiply by and sum over ; ; ; = 1. Since 2 = 2 = 1,
8.63
218

p
+ w3 , w p ; ; 8.68 which gives the linear relation between correlations h i = w1 h i + w2 h i + w3 , wh i : 8.69 The expectation h i is an nearest-neighbour correlation for the honeycomb Ising model with parameters G1 , G2 and G3 . Each expectation on the righthand side can be regarded either as a second-neighbour correlation for the honeycomb lattice or as a nearest-neighbour correlation for a triangular lattice Ising model with parameters K 1 , K 2 , K 3 related to G1 ; G2 ; G3 by the equations in Sect. 8.5.5
f ;;;g
; ; ; = 1 2
f ;;;g
+ w1
+ w2
8.7 Baxter and Enting's Transformation and the Functional Equation

From Fig. 8.3 it can be seen that instead of performing the star-triangle transformation by summing over the spin variables associated with the white sites the summation could equally well have been taken over those associated with the black sites. In other words, up-pointing stars could have been used instead of down-pointing stars. In the reverse triangle-star transformation there is similarly a choice between placing new spins in down-pointing or up-pointing triangles. These alternatives are exploited very ingeniously in a transformation illustrated in Fig. 8.5. The starting point is the honeycomb lattice, illustrated by the full lines in Fig. 8.5a, which is regarded as wound round a cylinder with the right-hand side joined to the left. The central row of vertical edges containing sites i, j , m and n is denoted by R. A star-triangle transformation is performed using up-pointing stars above R and down-pointing stars below R and producing the edges represented by dotted segments in Fig. 8.5a. The new lattice consists of the unchanged vertical edges of R, the dotted edges and the original honeycomb edges at the upper and lower boundaries. It embodies a row of rectangular faces containing R, triangular faces above and below R and, at the boundaries, kite-shaped quadrilaterals formed by the original honeycomb boundary edges and the top and bottom rows of dotted triangle edges. The next step is a triangle-star transformation in which a new spin is put into all down-pointing triangles above R and all up-pointing triangles below R. The result is the lattice of Fig. 8.5b. Both the above transformations leave untouched the original
5
The derivation of 8.69 given here is that of Baxter and Enting 1978, but it can also be obtained by putting sg = s3 in equation 81 of Fisher 1959a. The expressions for w and wi , though not the derivation above, are due to Baxter and Enting.
8.7 Baxter and Enting's Transformation and the Functional Equation
219
n j
m i
n j
m i
n j
m i
Fig. 8.5. The e ect of repeated star-triangle and triangle-star transformations on

the honeycomb lattice. Reprinted from Baxter and Enting 1978, by permission of the publisher IOP Publishing Ltd.
a
b
c
sites of R including i, j , m and n and preserve symmetry about R. This will also be true of all subsequent transformations. Star-triangle and triangle-star transformations of the kind described above are now performed alternately, each pair of transformations producing a new row of rectangular faces above and below R and a new row of upward or downward pointing kites at each boundary. This process is continued until the entire lattice is occupied by a rectangular region and two regions of kite-shaped faces, apart from two rows of pentagons where the regions meet. This situation is shown in Fig. 8.5c. Suppose that the interaction parameters for the original hexagonal lattice are G1 , G2 and G3 , as shown in Fig. 8.4, and that they are all positive. The interaction parameter for vertical edges in the rectangular parts of the nal lattice is then G3 and, for horizontal edges, K 3 , where K 3 is given by 8.44 or 8.49 with i = 3. The rectangular region is thus equivalent to a square lattice with unequal interactions in the two mutually perpendicular lattice directions. In the thermodynamic limit, the numbers of faces in each row and column of the original honeycomb lattice tend to in nity and thus so do the numbers of faces in each row and column of the rectangular region. The sites i, j , m and n lie deep in the interior of the latter and hence the spin correlations between them must be equal to those of the square lattice. Let gK ; G denotes the horizontal nearest-neighbour correlation in a square lattice with interaction parameter K in the horizontal direction and G in the vertical direction then h i j i = gK 3 ; G3 . By considering a 90 rotation of
220
the square lattice, the vertical nearest-neighbour correlation is gG; K and thus h i m i = gG3 ; K 3 . The sites of R, including i, j , m and n, have remained unchanged by the transformations and hence the spin distribution probabilities on them are those of the original honeycomb lattice, where i and j are a horizontal second-neighbour pair corresponding to the sites with spin variables and in Fig. 8.4. Thus h i = gK 3 ; G3 ; h i = gK 1 ; G1 i ; h i = gK 2 ; G2 ; 8.70 where the last two relations follow by symmetry considerations. Similarly, since i and m are a vertical pairs of neighbouring sites in the original honeycomb lattice corresponding to and in Fig. 8.4, h i = gG3 ; K 3 ; h i = gG1 ; K 1 ; h i = gG2 ; K 2 : 8.71 However, as was seen in Sec. 8.5, the second-neighbour correlations in the honeycomb lattice with interaction parameters G1 , G2 and G3 are equal to nearest-neighbour correlations in a triangular lattice with interactions K 1 , K 2 and K 3 . So 8.70 and 8.71 give nearest-neighbour correlations in the triangular and honeycomb lattices respectively in terms of those of a square lattice with unequal interactions in the two lattice directions. The parameters Gi and K i are connected by the relations given in Sect. 8.5 and, from 8.49, our assumption that the Gi are all positive implies that the K i are also all positive. Substitution of 8.70 and 8.71 into the linear correlation relation 8.69 yields the functional equation gG3 ; K 3 = w1 gK 2 ; G2 + w2 gK 1 ; G1 + w3 , wgK 3 ; G3 : 8.72 Baxter and Enting 1978 showed that 8.72 e ectively determines the form of gK ; G and thus yields the nearest-neighbour correlations and hence the con gurational energies for the zeroeld Ising model on the square, triangular and honeycomb lattices. In the next section an outline is given of the derivation of gK ; G from equation 8.72 following the general lines of the proof given by Baxter and Enting.
8.8 The Solution of the Functional Equation

In terms of the square lattice nearest-neighbour correlation gK ; G we de ne gK ; f K jk = coth2G ; 8.73 K where K is termed the amplitude of f K jk and k = sinh2K 1sinh2G 8.74
221
is its modulus. For xed k, K = 1 implies G = 0 and vice versa. Now since gK ; G is h i j i for a pair of neighbouring sites with interaction parameter K , g 1; 0 = 1 and g 0; 1 = 0. It follows that, where k is kept xed, f 1jk = 1; f 0jk = 0 : 8.75
8.8.1 A Preliminary Result for f K jk

Now substitute 8.73 in the functional equation 8.72, noting that, by 8.52, the modulus k is the same for all terms. After using the second relation of 8.66 coth2G3 f G3 jk = ,wff K 3 jk coth2K 1 coth2K 2 + coth2K 3 , coth2K 1 coth2K 2 f K 1 jk + f K 2 jk + f K 3 jk , coth2K 3 g : 8.76 From 8.50 and 8.52 with i = 3 and the rst relation of 8.66, coth2G3 = w coth2K 1 coth2K 2 + coth2K 3 : 8.77 We now use 8.77 to substitute for w coth2K 3 on the right-hand side of 8.76. After using 8.66 for w, 8.76 can then be rearranged to give bK 3 jksech2K 1 sech2K 2 sech2K 3 = k f K 1 jk + f K 2 jk + f K3 jk , 1 ; 8.78 where bK 3 jk = coth2K 3 coth2G3 f K 3 jk + f G3 jk , 1 : 8.79 That G3 can be omitted from the arguments of bK 3 jk follows from 8.52 which gives G3 as a function of K 3 and k. However, by permuting the interactions in Fig. 8.4 we could derive 8.78 with K 2 in place of K 3 . Since all other factors in 8.78 are symmetric in K 1 , K 2 and K 3 it can be deduced that bK 2 jk = bK 3 jk : 8.80 From 8.55 and 8.56 it is possible for K 2 to vary with k and K 3 xed. In fact with K 3 = 0 and 3 = 0, K 2 can be made to vary over its whole range from 0 to 1 by varying 1 from 1 to 0. Hence bK jk must be independent of the value of its argument K and can be written as bk. The f K jk functions in 8.78 have four distinct arguments K 1 , K 2 , K 3 and G3 . We aim to derive a relation for f with one argument only. We now wish to obtain an expression for f 0 K jk, the partial derivative of f K jk with respect to K at constant k. Taking K 3 and k as constant, di erentiation of 8.78 with respect to K 1 gives @ K 2 = , kf 0K 1 jk + 2bk tanh2K 1 sech2K 1 sech2K 2 sech2K 3 : @K1 kf 0K2 jk + 2bk tanh2K 2 sech2K 1 sech2K 2 sech2K 3 8.81
222
With K 3 and k constant, G3 is also constant and di erentiation of the expression for cosh2G3 given by 8.50, together with substitution from the corresponding expressions for cosh2G1 and cosh 2G2 , yields @ K 2 = , sinh2K 2 cosh2G2 : 8.82 @K1 sinh2K 1 cosh2G1 Eliminating @ K2 =@ K1 between 8.81 and 8.82 and using 8.50 gives aK 1 jk = aK 2 jk ; 8.83 where 8.84 aK jk = 1 f 0 K jk coth2G + bk tanh2 2K : 2 Equation 8.83 is true for all values of K 1 and K 2 , and reasoning similar to that used for bk shows that aK jk is likewise independent of the modulus K and may be written as ak . Then 8.84 gives bk tanh2 tanh2 f 0 K jk = 2 ak ,coth2G 2K = 2 ak , bk 2 2=K : 8.85 1 + k2 sinh 2K 1 2 Integration of 8.85 yields f K jk = akAK jk , bkB K jk ; 8.86 where, since f 0jk = 0, Z 2K q dx 2 ; AK jk = 0 1 + k2 sinh x 8.87 Z 2K tanh2 xdx q : B K jk = 0 1 + k2 sinh2 x Since f 1jk = 1, 8.86 gives 1 = akA1jk , bkB 1jk : 8.88 From 8.73, 8.86 and 8.88, AK jk , bk B K jk , B 1jk AK jk : gK; L = coth2K A1jk A1jk 8.89 We thus have the required formula for the nearest-neighbour correlation once we have formulae for A1jk, B 1jk and bk.
223
8.8.2 Expressions for A1jk and B1jk We now derive formulae for A1jk in terms of AK jk and AGjk and B 1jk in terms of B K jk and B Gjk, which depend solely on the relation
between k, K and G given in 8.74. The rst step is to put y = sinh x in 8.87 which yields Z sinh2K p1 + y2dy + k2y2 ; AK jk = 1 0
B K jk =
Z sinh2K
0
For AK jk, a second transformation y = 1=kz then gives Z1 p1 + z2dz + k2z2 AK jk = Z1=k sinh2K dy 1 1 p = ; 8.91 sinh2G 1 + y 2 1 + k 2 y 2 where we have replaced the dummy variable z by y in the last expression. This is equivalent to A1jk = AK jk + AGjk : 8.92 For B K jk, the transformation y = 1=kz yields Z1 dy p1 + y21 + k2y23 : 8.93 B K jk = sinh2G Now, replacing K by G in the second relation of 8.90 and integrating by parts,
p1 + y231 + k2y2 :
y2 dy
8.90
y B Gjk = , p 2 y 2 1 + y 2 1 + k 0 Z sinh2G dy p1 + y21 + k2y23 : 8.94 + 0 Since B 1jk is the limit of B Gjk as G ! 1 for xed k, 8.94 yields Z1 dy p1 + y21 + k2y23 : B 1jk = 8.95 0 Adding 8.93 to 8.94 and substituting for k2 in the integrated part from
8.73
"
sinh2G
B 1jk = B K jk + B Gjk + tanh2K tanh2G: 8.96 Expressions for AK jk and B K jk in terms of elliptic integrals are given in
Appendix A.2.
224
8.8.3 An Expression for bk

The square lattice Ising model with horizontal and vertical interaction parameters K and G respectively has a zeroeld Hamiltonian given by
b H=T = ,K
nX : :n:h
where the rst and second summations are over horizontal and vertical nearest-neighbour pairs respectively. Since gK ; G and gG; K denote horizontal and vertical correlations respectively, gK ; G = 1 @ ln Z K ; G ; gG; K = 1 @ ln Z K ; G ; 8.98 where Z K ; G denotes the con guration partition function for a square lattice with unequal interactions. Hence @gK ; G = 1 @ 2 ln Z K ; G = @gG; K : 8.99 From 8.73, @gK ; G = coth2K @f K jk @k
fi;jg
i j ,G
nX : :n:v
fi;jg
i j;
8.97
@K
@G
@G @G
@K@G
@K
Derivation of a similar expression for @gG; K j@ K and substitution in 8.99 gives @f K jk = @f Gjk : 8.101 From 8.79, without the index 3, b f Gjk = coth2Kk G + 1 , f K jk : 8.102 coth2 Di erentiating both sides of this relation with respect to k at constant G and using 8.101 and 8.84 gives after some rearrangement, @f K jk = 1 ak , bk + K dbk C K jk ; k @k 8.103 2 dk where C K jk is de ned by A.26. Also it follows from 8.86 that @f K jk = dak AK jk , dbk B K jk @k dk dk @AK jk , bk @B K jk : 8.104 + ak @k @k With some manipulation it can be shown from 8.87 that
= ,2k coth2K coth2G @f K jk : @k
@k
@G
8.100
@k
@k
225
0 2B k @ k @kK jk = k0 2 B K jk , AK jk + C K jk ; p where k0 = 1 , k2 see A.6. Combining 8.103, 8.104 and 8.105 we
K k @A@k jk = B K jk , AK jk + C K jk ;
8.105
i B K jk + 1 C K jk ,k dbkk + ak , 1 + k2 k0 ,2 bk 2 d dak 0 ,2 bk = 0 : + AK jk k dk , ak + k 8.106 AK jk and B K jk + 1 C K jk are linearly independent and since 8.106 is 2 1 true for all K it follows that the coe cients of AK jk and B K jk+ 2 C K jk are zero. This yields a pair of rst-order di erential equations for ak and bk, or equivalently for ak and ck = k0 ,2 bk. k dakk = ak , ck ; d 8.107 0 2 dck = ak , k0 2 ck : kk dk For k 1 these equations can be solved in terms of complete elliptic integrals. Eliminating ak, d kk0 2 dck , kck = 0 : 8.108 dk dk
From A.10, the general solution of 8.108 gives bk = k0 2 ck = k0 2
Kk + Kk0 ; 8.109 where
and are constants. By de nition, jgK ; Gj 1 for all K and G, and hence all K and k. However Kk0 ! 1, as k ! 0, k0 ! 1. It follows from 8.73, 8.86 and 8.87 that, if 6= 0, jgK ; Gj will become unbounded as k ! 0 for xed K . Thus it must be the case that = 0. Then, from the rst relation of 8.107, 8.109 and A.8, ak =
Ek; bk =
k0 2 Kk: 8.110 The constant
can be evaluated by putting K = 1 in A.24 and substituting it and 8.110 into 8.88 to give
EkKk0 + Ek0 Kk , KkKk0 = 1 : 8.111 From the identity A.11,
= 2= giving bk = 2k0 2 Kk= : 8.112 Since the modulus of an elliptic integral cannot be greater than 1, this procedure must be modi ed for k 1. It can be shown that then
nally obtain
226
An expression valid for all k can be obtained by using the modulus p p 2 k = 2 m; k1 = 1 + k 1 + m 8.114 where k1 1 for all k 6= 1 and k1 = 1 for k = 1. Using Landen's transformation in the form A.13, 8.112 and 8.113 both take the form bk = 21 , kKk1 = : 8.115
8. Exact Results for Two-Dimensional Ising Models bm = ,2m02 Km=m; where m = k,1 :
8.113
8.9 Critical Behaviour

For equal interactions in the various lattice directions the con gurational energy per site is 1 8.116 u = , 2 zJ h i j in:n: = , 1 zJgK ; G ; 2 where gK ; G is given by 8.89; K = G for the square lattice, and K and G are connected by 8.27 for the triangular lattice. For the honeycomb lattice, gK ; G is replaced by gG; K . Since AK jk and B K jk are analytic except at K = 1, a critical point can arise only from a singularity in bk. However, bk itself is analytic unless k1 = 1. From A.5, near k1 = 1, 1 16 1 8 1 Kk1 ' 2 ln k162 = 2 ln 1 , k2 ' 2 ln 1 , k : 8.117 0 1 1 1 Hence, near k1 = 1, 8.115 gives 8 bk ' 1 , k ln 1 , k : 8.118 1 Now k increases steadily from 0 at T = 0 to 1 at T = 1, passing through the value 1 at a temperature denoted by Tc, while k1 attains its maximum value of 1 at Tc. Hence 1 , k is proportional to Tc , T whereas 1 , k1 is proportional to Tc , T 2 near Tc. Hence, from 8.89, 8.116 and 8.118, u = uc + DT , Tc ln jT , Tcj 8.119 near Tc, where uc is the value of u at T = Tc and D is a constant appropriate to the particular two-dimensional lattice. So u is continuous at T = Tc but the zeroeld con gurational heat capacity c0 , which is proportional to du=dT , diverges to in nity as , ln jT , Tcj, both above and below the critical point. Since xz ln x ! 0 as x ! 0 for any z 0 it follows that the critical indices = 0 = 0. Since = 0 = 0 also for the meaneld theory, where c0 has a nite discontinuity, it is obvious that these critical index values can correspond to quite di erent types of behaviour. The analysis presented above con rms the existence of a unique critical point for a given lattice. It is easy to verify that the singularity condition k = 1
8.9 Critical Behaviour
227
gives the same values of Tc as those deduced from transformation theory, on the assumption of a unique critical point for each lattice, in Sect. 8.3 and 8.4. It is convenient to adopt a notation used by Domb 1960 in which the highand low-temperature parameter values connected by a transformation are written as K ; K inv . Thus for the square lattice, K inv = K , given by 8.13, for the triangular lattice, K inv = K 0 , given by 8.32 and for the honeycomb lattice, Ginv = G0 , given by 8.38. For the square lattice with K = G, from 8.74 k = 1= sinh2 2K and hence, by 8.13, inv k = sinh2KK : 8.120 sinh2 For the triangular and honeycomb lattices k is given by 8.74 where K and G are connected by the star-triangle transformation 8.27. However, for the triangular lattice sinh2G = 1= sinh2K 0 and hence 8.120 applies. Again for the honeycomb lattice, sinh2K = 1= sinh2G0 and 8.120 thus applies if J=T for this lattice is also denoted by K . The critical point condition used in Sects. 8.3 and 8.4 was K = K inv , and it follows immediately from 8.120 that this is equivalent to k = 1. Values of Tc =zJ are given in Table 8.1 together with those of other critical parameters, discussed below. Corresponding quantities for d = 3 lattices, obtained by series methods and given in Domb 1974, are included for comparison, as are results from standard approximations. The expression for the energy uc at the critical point given by 8.21, applies for the triangular and honeycomb lattices as well as for the square lattice Example 8.3 and it will be shown in Sects. 8.10 and 8.11 that 8.21 follows from 8.89 and 8.116. At T = 1, where the spin distribution is random, u = 0 whereas, in the ground state at T = 0, u = , 1 zJ . The ratio 2 ,uc= 1 zJ , given in Table 8.1, is thus the proportion of the total energy 2 change from T = 1 to T = 0 which occurs when T Tc and is therefore due entirely to short-range ordering e ects. In the meaneld approximation, uc has the random value zero, as can be seen from 3.14. It is useful to de ne a partition function per spin or per site by Z = N : 8.121 Since the constant magnetic eld distribution is being used with H = 0 it follows from 2.59 that, with H = 0, f = u , Ts = ,T ln ; 8.122 where f and s are respectively the con gurational free energy and entropy per site. Thus s = u=T + ln: 8.123 At T = 1, where the spin distribution is random, = 2 and u = 0 so that s = ln2. At T = 0, where the spins are completely ordered, s = 0. Hence if sc denotes the value of s at the critical point T = Tc the ratio ln2 , sc= ln2,
228
values of which are presented in Table 8.1, gives the proportion of the total entropy change from T = 1 to T = 0 which occurs when T Tc. Exact expressions for sc are obtained in Sects. 8.10 and 8.11. The direct method of deriving the spontaneous magnetization is to use 3.11, as was done in meaneld theory. This requires an expression for the equilibrium magnetization as a function of T and H. However such an expression is not available since the exact partition function is known only for H = 0. One way out is to calculate the dependence of the zeroeld correlations h i j i on the separation between sites i and j and to use 3.2. We shall not attempt to present the long and complicated algebraic derivation McCoy and Wu 1973 but simply give the compact and elegant result ms = 1 , k2 1=8 = 1 , sinh2 2K inv sinh2 2K 1=8 ; 8.124 which applies for the square, triangular and honeycomb lattices. Since 1 , k 1 is proportional to Tc , T for Tc , T Tc this immediately yields = 8 , where the critical exponent is de ned in 3.55. To derive the critical exponent , de ned by 3.58, we need the limit of @m=@ H as H ! 0, just above the critical point. Since the partition function is available only for H = 0, a direct derivation of @m=@ H is impossible. However, it is possible to use 2.115 which was shown in the one-dimensional case to yield the same expression for @m=@ HH=0 as the direct method. For T Tc the second member of 2.115 is true for any dimension and with 2.116 gives
@m T @H = 1 +
where j denotes a typical site and the summation is over all other sites. It has been assumed that in the thermodynamic limit all sites can be regarded as equivalent. Suppose that the distance between sites j and `, in lattice constant units, is r r = 1 for a nearest-neighbour pair and that the angle made by the vector from j to ` with a xed direction is . Then it is deduced from transfer matrix considerations that, for large r and T , Tc Tc, h j ` i ' wr,1=4 exp ,DT , Tcr : 8.126 This implies a large correlation range just above Tc and the rst term on the right-hand side of 8.125 is negligible in comparison to the sum. Replacing the latter by an integral Replacing the variable r by x = DT , Tcr and integrating rst with respect to x and then with respect to ,
fk6=jg
j ki ;
8.125
@m T @H =
Z 1 Z 2
0
dr
d wr3=4 exp ,DT , Tcr :
8.127
@m T , T ,7=4 : c @H
8.128
8.10 Thermodynamic Functions for the Square Lattice
229
and thus = 7 .6 There remains the di culty that in a direct derivation of 4 @m=@ H the thermodynamic limit would in principle be approached before letting H ! 0, whereas in the method discussed, H is set equal to zero before the thermodynamic limit is taken. However Abraham 1973 has shown that this inversion of limit operations makes no di erence. The values = 0, 1 = 8 and = 7 satisfy the Essam Fisher scaling law 3.87. If 3.77 is 4 assumed then = 15, which is the generally accepted value.

Thermodynamic functions for the two-dimensional Ising model are discussed at length by Domb 1960, Green and Hurst 1964, McCoy and Wu 1973 and Baxter 1982a. The fundamental result used here is the expression 8.89 for the nearest-neighbour correlation. For the square lattice with equal interactions in the two lattice directions K = G and 8.92 and 8.96 become A1jk = 2AK jk ; 8.129 B 1jk = 2B K jk + tanh2 2K and 8.89 then becomes 1 8.130 gK ; K = 2 coth2K 1 + bk tanh2 2K : Substituting k = 1= sinh2 2K in the expressions 8.115 for bk and 8.114 for k1 , and z = 4 in 8.116 u = ,J coth2K 1 + 2 Kk 2 tanh2 2K , 1 ; 8.131
8.132 1 + sinh 2K applicable for the entire temperature range. From 8.118, bk = 0 at T = Tc and equation 8.21, for the critical value of u, is thus veri ed for the square lattice. It is of interest to look at the two extremes of the temperature range T = 0 and T = 1, for both of which k1 = 0 and hence from A.4, 2Kk1 = . As T ! 0, K ! 1 and hence tanh2K ! 1 so that, by 8.131, u ! ,2J , con rming that the zero-temperature state is completely ordered. The T ! 1 case is more troublesome since K ! 0 and coth2K ' 2K ,1 ! 1. However, using A.4 with k1 ' 4K , the curly bracketed part of 8.131 is, to order K 2 , 1 + 8K 2 , 11 + 4K2 ' 4K2 : 8.133
6
k1 = 2 sinh2K = 2 sinh2K sech2 2K ; 2
This derivation is essentially that of Fisher 1959b.
230
Thus u=J ' ,2K ! 0, as T ! 1, as would be expected in a random distribution. The con gurational heat capacity can easily be deduced from 8.131 see Example 8.7. Using 2.26, and noting that in zero eld the enthalpy per site h = u, 8.134 u = , J @ lnZ = ,J @ ln ; where is de ned by 8.121. The form of u and the condition = 2 at T = 1, K = 0 completely determine since, from 8.134,
N @K
@K
J ln=2 = ,
ZK
0
udK :
8.135
It can be veri ed Example 8.6 that when substituted in 8.134 the expression 1 Z =2 ln n 1 1 + 1 , k2 sin2 1=2 o d ln=2 = lnfcosh2K g + 1 2 0 8.136 yields the form of u given by 8.131. Also, putting K = 0 in 8.136 gives = 2. Hence 8.136 is the required integral of 8.131. An alternative formula for the partition function is 1 Z Z ln=2 = 22 d d lnfcosh2 2K , sinh2K cos + cos g 0 0 o n 1 Z Z = ln cosh2K + 22 d d ln 1 , 1 k1 cos + cos : 2 0 0 8.137 For k1 1 the integrand is nite throughout the domain of integration, but for k1 = 1 it tends to ,1 when ! 0, ! 0. Since k1 1 for T 6= Tc but k1 = 1 at T = Tc the divergence in the integrand is associated with the critical point. It can be shown that 8.137 is equivalent to 8.136 by rst changing the variables of integration to 1 !1 = 2 + ; !2 = 1 , : 8.138 2 The domain of integration transforms to a square in the !1 ; !2 plane bounded by the lines !1 + !2 = , !1 + !2 = 0, !1 , !2 = and !1 , !2 = 0. By using the symmetry of the expression cos + cos = 2 cos !1 cos !2 ; 8.139 it can be shown that an alternative domain of integration is the rectangle 1 0 !1 , 0 !2 2 . Since dd = 2d!1d!2 Z =2 Z ln=2 = ln cosh2K + 1 d! d! ln 1 , k cos ! cos ! :
2
8.140
231
3 u ,J 2
3 c0 kb 2
0
ous line
Fig. 8.6. Con gurational internal energy broken line and heat capacity continunet.
3 T=J
for zero eld plotted against temperature for the square lattice ferromag-
Integration over !1 can be performed using the standard result y + y2 , x2 1=2 Z lny , x cos !d! = ln ; x y; 8.141 2 0 1 and a trivial transformation from !2 to = 2 , !2 then recovers 8.136. A di erent single integral expression is obtained by applying 8.141 directly to 8.137, without the preliminary transformation from ; to !1 ; ! . Integration with respect to yields 1
ln=2 = lncosh 2K + 2 ln 1 k1 4 Z ,1 ,1 2 1=2 + 21 0 d lnf2k1 , cos + 2k1 , cos , 1 g Z h1 i ,1 = 1 ln 2 sinh2K + 21 2 0 d arccosh2k1 , cos : 8.142 This is the form obtained directly by several authors from the largest eigenvalue of the transfer matrix Domb 1960, Green and Hurst 1964 although in the version of the transfer matrix method given by Baxter 1982a the result is an expression like 8.136 but with replaced by 1 , . 2 The con gurational entropy can be obtained by substituting 8.131 and 8.136 in 8.123. As T ! 0 both terms become in nite. However, it may be shown from 8.136 that ln , 2K ! 0 as T ! 0 K ! 1 and from 8.131,
232
the critical point, using 8.123, 8.21 and 8.136, 1 Z =2 d ln1 + cos sc = ,K c coth2K c + 1 ln 2 cosh2 2K c + 2 0 p Z =2 + 1 d ln1 + cos ; 8.143 = ln2 , ln1p 2 + 2 0 the critical parameter values being given by 8.19 and 8.21. The critical entropy ratio is given in Table 8.1. The con gurational energy and heat capacity for the square lattice ferromagnet are plotted against temperature in Fig. 8.6. Using 8.13 the general relation 8.124 for the relative magnetization of the square lattice ferromagnet takes the form 4 1=8 ms = sinh 2K , 1 : 8.144 sinh4 2K
8. Exact Results for Two-Dimensional Ising Models using A.4, that T ,1u + 2K ! 0 as T ! 0. Hence, as expected, s ! 0. At
8.11 Thermodynamic Functions for the Triangular and Honeycomb Lattices

By 8.70, 8.71 and 8.116 we require gK ; G for the triangular lattice and gG; K for the honeycomb lattice, G and K being connected by the star-triangle relation 8.27 in both cases. It is thus necessary to use for the coe cients AK jk and B K jk not only the equations 8.92and 8.96, which depend on the relation 8.74, but also equations A.39 and A.41 which arise from the star-triangle relation between K and G. Combining these 83AKjk ; A1jk = : 3 8.145 AGjk ; 2 83BK jk + tanh2K tanh2G 1 + tanh2K ; h i 8.146 B 1jk = : 3 B Gjk + tanh2K tanh2G 1 , 1 tanh2K : 2 2 Substitution into 8.89 then yields 1 gK ; G = 3 coth2K f1 + bk tanh2K tanh2G 1 + tanh2K g ; 8.147 n h 1 io 2 gG; K = 3 coth2G 1 + bk tanh2K tanh2G 1 , 2 tanh2K : 8.148
8.11 Thermodynamic Functions for Triangular and Honeycomb Lattices
233
From 8.114, 8.115, 8.116 and 8.27 it follows that, for the triangular lattice with equal interactions in the three lattice directions, the con gurational energy per site is given by 4Kk1 XX , 3 ; 8.149 u = ,J coth2K 1 + 1=2 4X + X , 13=2 X + 31=2 where X = exp4K ; 8.150 , 3=4 2 + 3 1=4 k1 = 4X4=X + 1 , X3=2 XX 31=2 1 2 X 1 + and for the honeycomb lattice, with equal interactions in the three lattice directions, K , 2 u = ,J coth2G 1 + sck1 c + 2cc,+ 1=2 ; 8.151 , 1 11 where c = cosh2G, s = sinh2G and 1=2 c 1=2 c 11=4 k1 = 2ssc, , 1 22, , 1=2 : 8.152 1 + c 1 These expressions are valid over the entire temperature range. In both cases, k1 has its maximum value of 1 at T = Tc, giving rise to the critical point singularity, while k1 = 0 at T = 0 and T = 1. As for the square lattice, it 1 can be shown that 8.149 and 8.151 give u = 0 at T = 1 and u = , 2 zJ at T = 0. It is easy to verify that 8.21 applies at T = Tc in both cases. The expressions for the free energy are connected by 8.135 with the forms derived above for u. Since we are no longer concerned with the startriangle transformation though see Example 8.8 the symbol K is used to denote J=T for both the triangular and honeycomb lattices. For the triangular lattice 1 Z 2 d Z 2 d lnf3 + 3 , cos + cos + cos + g ; c s s ln=2 = 82 0 0 8.153 where c = cosh2K , = sinh2K . For the honeycomb lattice, s Z 2 Z 2 c ln=2 = 1612 d d ln 3 + 1 , 2 cos + cos + cos + s 0 0 1 ln2 8.154 ,4 It follows immediately that = 2 for T = 1 K = 0. We shall now sketch a proof that 8.134 is satis ed by the expressions given for and u. A rst step is to put 8.153 and 8.154 into a common form by writing
234
1 s 1 ln=2 = 2 ln + 4 ln + 82 where =

0
Z 2 Z 2
d
0
d ln , cos , cos , cos + ; 8.155 8.156
1 for the triangular lattice,
However, cos + 1 can be replaced by cos in 8.155 because the inte2 gration with respect to is taken over a whole period 2. Then use of 8.141 enables an integration over to be performed, giving 1 s 1 ln=2 = 2 ln + 4 ln Z h i + 2 d ln 1 f , cos + , cos 2 , 4 cos2 =2 1=2 g : 2 0 8.159 Then, with the substitution x = cosh , 8.134 gives u = , coth2K , d Z 1 p dx J 2 dK ,1 1 , x2 x , 2 , 2x , 2 ; 8.160 which is easily transformed to 8.149 or 8.151 using A.28. From 8.123, 8.21 and 8.159 the critical point entropy is given by 1 sc = ,K c coth2K c + 4 ln 16 sinh2 2K c Z h i + 2 d ln 1 f3 , cos + 3 , cos 2 , 2 cos , 2 1=2 g : 2 0 8.161
= 2X , 1 for the triangular lattice, s 8.157 = 3 + 1 2 , + 1 c c c for the honeycomb lattice. : 2 = , 1 s c In both cases, = 1 at T = 0 and T = 1, and has a minimum value of 3 at T = Tc. For non-zero T 6= Tc the integrand in 8.155 is therefore nite over the whole domain of integration. The singularity at T = Tc arises because of the in nity in the integrand when = = 0 or = = 2. We shall now express ln=2 in terms of an integral with respect to a single variable. First write
1
1 8.158 cos + cos + = 2 cos + 2 cos 2 :
: 1 for the honeycomb lattice, 2 8 3 + 3 X2 + 3 c s
8.12 The Antiferromagnet
235
0.8 0.6
FM
FM
0.8 0.6
s
0.4 0.2 0
AFM FM AFM
c0 kb 0.4
0.2
1 2 3 T=J 4 5 Fig. 8.7. Con gurational entropy broken lines and heat capacity continuous lines for zero eld plotted against temperature for the triangular lattice ferromagnet FM and antiferromagnet AFM.
The entropy ratios presented in Table 8.1 are derived from this. In both 1 cases, T ,1 u + 2 zK and ln , 1 zK tend to zero with T so that s ! 0. Plots of 2 entropy and heat capacity against temperature for the triangular ferromagnet and antiferromagnet are shown in Fig. 8.7. For the triangular and honeycomb lattices respectively, 8.124, 8.32 and 8.38 yield X + 13X , 3 1=8 ms = X , 13 X + 3 ; 8.162 , 2 c c ms = + 1 , 13 c c
3
1=8
8.163

For the antiferromagnet the nearest-neighbour pair interaction ,J i j of the ferromagnet is changed to J i j J 0, which corresponds to the transformation K ! ,K K 0. In Sect. 4.2 it was shown from the form of the Hamiltonian that, for loose-packed lattices, the zeroeld con gurational
236
At K = 0 T = 1, 0 = 1, and 0 decreases steadily as K increases, attaining its least value of 3 at K = 1 T = 0. Now the minimum value of 23 cos + cos + cos + is , 2 for = = 2 . Thus a singularity appears in 3 the integrand in 8.164 only at T = 0, indicating that there is no non-zero critical temperature. A transformation similar to that producing equation 8.159 gives Z 1 1 s ln=2 = 2 ln + 21 d ln 2 0 + cos 0 p + + cos 2 , 4 cos2 =2 : 8.166 Then 8.134 and the substitution x = cos yield u = , coth2K , 1 d0 Z 1 p dx J 2 dK ,1 1 , x2 x + 0 2 , 2x , 2 : 8.167 The quadratic x + 0 2 , 2x , 2 is positive forp x when 0 3 . It may be all 2 written as x + x + where = 0 , 1 + i 20 , 3 and is its complex conjugate. Using A.30, 4X3 , X1=2 8.168 u = ,J coth2K 1 , 1 + X3=2 Kk ;
partition function is the same for the antiferromagnet as for the ferromagnet. It follows that thermodynamic quantities such as the zeroeld internal energy and heat capacity and the critical temperature are also the same. The square and honeycomb lattices are loose-packed, and we now verify that the expressions for the corresponding partition functions are invariant to the transformation K ! ,K . This follows for the honeycomb lattice since, in 8.154, the quantities 3 + 1 and 2 are unaltered by the transformation. In c s the rst expression for the square lattice partition function, given in 8.137, the change from K to ,K alters the integrand to lnfcosh2 2K + sinh2K cos + cos g : However a change of variables to 0 = , , 0 = , restores the integral to its original form, and the required invariance property is thus demonstrated. For the triangular lattice, which is close-packed, the situation is quite di erent. Equation 8.153 for the partition function becomes 1 Z 2 d Z 2 d lnf3 , 3 + cos + cos + cos + g c s s ln=2 = 82 0 Z0 2 Z 2 1 = 1 ln + 82 d d ln 0 + cos + cos + cos + ; 2 s 0 0 8.164 where 3 s3 X,2 + 3 0 = c , = 21 , X,1 : 8.165 s
237
where now X = exp,4K and 3=2 1=2 8.169 k = 1 , X3=2 3 + X1=2 : 1 + X 3 , X The modulus k = 0 at K = 0 and increases monotonically with K to approach the value 1 as K ! 1. There is, therefore, no singularity and no critical point at a non-zero temperature. The heat capacity is a continuous function of T with a maximum and approaches zero as T ! 0 or T ! 1. For X 1, p k0 4= 3X and using A.5 it can be seen that u ! ,J as K ! 1. It is not di cult to prove the slightly stronger result that lim u + J = 0 : 8.170 So far the behaviour of the triangular antiferromagnet, with a maximum rather than a singularity in the heat capacity, looks rather like that of a linear Ising model. However, there is an important di erence which is crucial to physical understanding of the behaviour of the model. From 8.166, since 0 ! 3 as T ! 0, 2 J = 1 Z d ln n 3 + cos lim ln , T 2 T !0 2 0
3 1=2 o 2 2 =2 + 2 + cos , 4 cos : 8.171 From 8.170 and8.171 it follows that the entropy at T = 0 has a non-zero value7 given by
u s = Tlim0 ln + T !
T !0
1 Z d ln n 3 + cos +
3 + cos 2 , 2 cos , 2 1=2 o = 2 2 2 0 = 0:32306 : 8.172 Plots of the zeroeld antiferromagnetic entropy and heat capacity against temperature are shown in Fig. 8.7. For the explanation of this zero-point entropy we must look at the con gurations giving the zero-temperature energy ,J per site. A minimum energy state for the triangular antiferromagnet is one in which each triangle of neighbouring sites is occupied by two + spins and one , spin or vice versa. Out of the total of 3N neighbouring site pairs, 2N will then be occupied by like spins and N by unlike spins, giving U = ,2NJ + NJ = ,NJ . Hence u = ,J is the least possible energy per site. It might be thought that this minimum energy is associated with a highly ordered con guration like that shown in Fig. 8.8a. However there are many other minimum energy con gurations, as can be seen from Fig. 8.8b where
7
We agree with Domb 1960 rather than Wannier 1950 about the magnitude of this quantity.
238 +
8. Exact Results for Two-Dimensional Ising Models +
+ , + , , + , + + + + + + + , + , , + , + , , , , b a Fig. 8.8. Minimum energy con gurations for the triangular lattice antiferromagnet.
a triangle lattice is divided into a honeycomb sublattice and the sites at the centres of the hexagons, which are shown as open circles. If the honeycomb sublattice is occupied alternately by + and , spins as shown in Fig. 8.8b, then any distribution of spins on the circle sites gives u = ,J . There are 2N =3 minimum energy con gurations of this type which thus give a contribution 1 of 3 ln2 to s. Other minimum energy con gurations make up the entropy value to that given in 8.172. The ground states are frustrated in that, although the interaction between the spins of each nearest-neighbour pair is antiferromagnetic in character, one third of the pairs are forced to adopt a parallel or ferromagnetic con guration. It is natural to assume that the resulting in nite degeneracy of the minimum energy state causes the absence of a transition point for T 0. However, from other examples it appears that the connection between frustration, or in nite degeneracy of the ground state, and the absence of a transition to a long-range ordered state is not a simple one Mouritsen 1984. Expressions for ln=2 for the Ising model on the honeycomb, triangular and square lattices with unequal interactions are derived by the Pfa an method in Volume 2, Chap. 8.
Examples
8.1 From 8.15, show that 1 ln 1 ln N , coth2K @@ KZ = ,N + coth2K @@ KZ ; where K and K are connected by 8.13. Hence prove 8.17. 8.2 By a star-triangle transformation from a honeycomb lattice of N sites 1 with interaction parameter G to a triangular lattice of 2 N sites and then a dual transformation to a honeycomb lattice of N sites with interaction parameter G0 , prove 8.37 and 8.38. 8.3 Given that Z N; K is the partition function for the triangular lattice, show, from 8.34 and 8.32, that @ ln Z = N coth2K 0 , @ ln Z 0 1 , exp,4K 0 ; ,N coth2K + @ K @ K 0 1 , exp,4K
Examples
239
where Z 0 denotes Z N; K 0 . Hence, using 8.16 and noting that 1 Ne = 2 zN = 3N , show that 1 , exp,4K h i j in:n: + 1 , exp,4K0 h i j i0n:n: = 1 : 2 + exp,4K + exp,4K 0 3 By a similar procedure, show that for the honeycomb lattice, where 1 Ne = 2 zN = 3 N , 2 sinh2Gh i j in:n: + sinh2G0 h i j i0n:n: = cosh2G + 1 cosh2G + 1 2 cosh2G + cosh2G : 3 cosh2G + 1 cosh2G0 + 1 Hence prove that, for both the triangular and honeycomb lattices, the critical value of the con gurational energy is given by uc = ,J coth2K c ; assuming a unique critical point and replacing K c by Gc in the honeycomb case. 8.4 Show that the relations 8.44 between triangle and honeycomb parameters for the unequal interactions star-triangle transformation reduce to 8.27 if K i = K , Gi = G i = 1; 2; 3. 8.5 Show using 8.47 and 8.49 that, if K i and Gi i = 1; 2; 3 are respectively the triangle and honeycomb parameters connected by the star-triangle transformation, then cosh2Gi = cosh2K j cosh2K ` + sinh2K j sinh2K ` coth2K i ; where i; j; ` is a permutation of 1; 2; 3. 8.6 With k1 for the square lattice de ned by 8.132, show that 1 dk1 = 2f1 , 2 tanh2 2K g : k1 dK tanh2K Obtain the formula 8.131 for u by substituting the expression for given by 8.136 into the relation @ ln = , u :
0 8.7 Show that the complementary modulus k1 to the k1 de ned by 8.132 satis es the relation 0 k1 2 = 2 tanh2 2K , 1 2 : Show that the zeroeld con gurational heat capacity c0 per site is given by
@K
240
Hence, using 8.131 and A.8, deduce that, for the square lattice, c = 4kb K coth2K 2 fKk , Ek
0
u=J c0 = ,kb K 2 @ @ K :
+ tanh2 2K , 1 2 + Kk1 f2 tanh2 2K , 1g : The rst term in the square brackets gives the singular part of c0 . 8.8 Show that the star-triangle relation 8.28 between the partition functions of the triangular and honeycomb lattices may be written 1 ln h G = 2 ln R + 1 ln t K : 2 Use this relation with the star-triangle relation 8.27 between lattice parameters to deduce equation 8.154 for h from equation 8.153 for t .
h1
io
9. Applications of Transform Methods
9.1 The Decoration Transformation

In this chapter we describe how a number of models, mathematically equivalent to the Ising model, can be derived by transformation methods and, in particular, by the decoration transformation. Accurate results will be derived for cooperative phenomena, for example transitions on critical curves and the associated exponents, ferrimagnetic rst-order transitions in the H 6= 0 region and water-like behaviour when d 1 which have until now been treated by approximation methods. New phenomena, such as lower critical points and critical double-points, will also arise. Decoration is based on a regular lattice of N sites with coordination number z, which is called the primary lattice. An additional decorating or secondary site, is then placed between each nearest-neighbour pair of primary sites see Fig. 9.1. Hence there is one secondary site for each of the 1 zN 2 edges of the primary lattice. An assembly on a decorated lattice is equivalent to a standard Ising model on the primary lattice when: i There are two occupation states for any primary site, represented by i = 1 for the primary site labelled by i. ii A microsystem on a secondary site can interact only with the microsystems on the two adjacent primary sites.1 iii A microsystem on a primary site can interact only with the microsystems on the adjacent secondary sites and the nearest-neighbour primary sites. The two types of interaction are termed primary secondary and primary primary respectively. Using the discrete variable ij to label the occupation states of the secondary site between the nearest-neighbour primary sites i and j , the Hamiltonian for the decorated lattice model can then be written
1
The single secondary site on each edge of the primary lattice can be replaced by a group of secondary sites, Fisher 1959a, Syozi 1972. This condition must then be augmented with the phrase òr with the microsystems on the other secondary sites of the same group'. However in the decorated models treated in this chapter there is only one secondary site per edge.
242
9. Applications of Transform Methods Fig. 9.1. Decorated square lattice: white sites are primary and black sites are secondary.
b H=
nX : :n:p
where the rst summation is over primary nearest-neighbour pairs, the second is over primary sites and the i are eld related quantities. The primary primary interaction is not present in all decorated models. b Since each ij occurs in only one term of H a partial summation over the ij can be performed to yield, for the partition function Z of the decorated model, X b Z= exp,H=T
f i ; ij g
fi;jg
^ i ; j ; ij + X i ; h
p:
fig
9.1
where i; j =
f ig
exp ,
p: h X
fig
i =T
:n: i nYp:
fi;jg
i; j ;
9.2 9.3
X
f ij g
^ exp ,h i ; j ; ij =T :
With condition i, there are three forms of i ; j , +1; +1, ,1; ,1 and +1; ,1. From 9.2, X Z= +1; +1 N ++ ,1; ,1 N ,, +1; ,1 N +,
f ig
where
expf, N+ +1 + N, ,1 =T g ;

p: X
9.4 9.5
N = 1 2
are the numbers of primary sites for which i = 1 and
fig
X 1 i = 1 N i 2
p:
fig
N++ = 1 4
1 N,, = 4
nX : :n:p
fi;jg
h X 1 + i 1 + j = 1 N z + 2z i +2 8
p:
9.1 The Decoration Transformation

nX : :n:p
fig
fi;jg
nX : :n:p
i j
i i
243
; ;
nX : :n:p
fi;jg
h X 1 , i 1 , j = 1 N z , 2z i +2 8
p:
fig
fi;jg
i j
N+, = 1 2
nX : :n:p
are the respective numbers of ++, ,, and +, nearest-neighbour pairs on the primary lattice. From 2.67, Z = RN =4 Zpi:m K e; Le ; 9.7 where R = +1; +1 ,1; ,1 z=2 +1; ,1 z expf,2 +1 + ,1 =T g 9.8 i:m: is the partition function for an Ising model on the primary lattice and Zp with the quantities J=T and H=T , which appear in 2.67, replaced by the e ective couplings K e and Le respectively, where +1 ; exp4K e = +1; +1; ,,12 ,1 ; 1 9.9 +1; +1 z exp4Le = ,1; ,1 expf2 ,1 , +1 =T g : The general form of these relations is due to Fisher 1959a. The equivalent primary lattice Ising model with couplings K e and Le is a ferromagnet or antiferromagnet according as K e is positive or negative. For K e 0, any critical point or curve in the space of decorated model thermodynamic variables maps into the point K e = K c, Le = 0 in the K e ; Le space of the Ising model while any rst-order transition line or surface maps into the segment Le = 0 K e K c . Two useful relations follow immediately from 2.67. The rst one is i:m: 9.10 g K ; L = h i = 2 @ ln Zp = , 2 ui:m: K ; L ;
p e e
fi;jg
1h 1 , i j = 4 N z , 2
nX : :n:p
fig
i j
9.6
i j p
where i and j are a nearest-neighbour pair on the primary lattice and ui:m:K e ; Le is the con gurational energy per site in the equivalent Ising model. Similarly we have i:m: h i = N ,1 @ ln Zp = mi:m: K ; L ; 9.11
ip
zN
@Ke
zN
e e
@ Le
e e
244
where mi:m: K e; Le is the relative magnetization for the equivalent Ising model.
9.2 Dilute Decorated Models

9.2.1 A Superexchange Model
We consider a mixture model with all primary sites occupied by Ising spins and with a fraction y of the secondary sites occupied by solute atoms, the remaining secondary sites being vacant. The solute atoms have no magnetic moment but they have the superexchange property that a solute atom between the nearest-neighbour pair of primary sites i and j induces a primary primary interaction ,Jp i j Jp 0. When the secondary site between i and j is unoccupied, the spins on i and j do not interact. Denoting the solute chemical potential by , we use the distribution corresponding to the choice of , H and T as independent intensive variables. Let ij = 1 when the secondary site is occupied by a solute atom, and ij = 0 when it is unoccupied. Then ^ h i ; j ; ij = , ij + Jp i j 9.12 and 9.3 gives +1; +1 = ,1; ,1 = expK p + 1 ; 9.13 +1; ,1 = exp,K p + 1 ; where K p = Jp =T ; = exp=T : 9.14 Since the primary sites are occupied by Ising spins, 1 = H; 9.15 e where H = m0 H, m0 being the primary site dipole moment see Sect. 1.7. From 9.9, K e and Le are given by +1 +1 expK p exp2K e = +1;; ,1 = exp,K + 11 ; Le = H=T : 9.16 p + Since 0, it follows from 9.16 that K e K p . From 9.7, 9.8, 9.13 and 9.15 the partition function for the decorated model is Z = +1; +1 +1; ,1 zN =4 Zpi:m:K e ; Le : 9.17 Since, in each term of Z there is a factor for each occupied secondary site, the equilibrium value of y is given by
1 ,1 y = 2 zN @ lnZ 9.18 @ T ;H
9.2 Dilute Decorated Models
245
and from 9.10, 9.16 and 9.17, noting that @ Le =@ = 0, i:m:
,1 K 1 y = 2 @ ln +1; +1 +1; ,1 + 1 zN @@e @ ln@Zp 2 @ Ke 1 @ ln +1; +1 +1; ,1 =2 @ 1 + 2 gp K e; Le @ ln +1; +1= +1; ,1 : 9.19 @ This can be re-expressed as y = 1 h1 + i j ip @ ln +1; +1 + 1 h1 , i j ip @ ln +1; ,1 ; 2 2 @ @ 9.20 where i and j are a nearest-neighbour pair on the primary lattice. The terms 1 in the nal expression have a physical interpretation: 2 h1 , i j ip is the fraction of primary lattice edges for which i 6= j while @ ln +1; ,1=@ is the fraction of secondary sites on such edges which are occupied by solute atoms. The rst term corresponds similarly to the case i = j , noting that +1; +1 = ,1; ,1.
9.2.2 The Equilibrium Bond Dilute Ising Model

Since the only property of the solute atom in Sect. 9.2.1 is to induce an interaction between the spins on the adjacent primary sites, the model is equivalent to the equilibrium bond dilution model of Sect. 6.11.1 Rapaport 1972. Occupied and unoccupied secondary sites correspond respectively to unbroken and broken bonds between nearest-neighbour pairs on the primary lattice, and y is the fraction of unbroken bonds. The solute activity can be expressed in terms of K e and K p from 9.16, and substitution in 9.19 yields y = CK e ; K p + gp K e ; Le SK e; K p ; 9.21 where cosh CK e; K p = sinhK esinhK Kp , K e ; p 9.22 sinhK e sinhK p , K e : SK e; K p = sinhK p Equation 9.21 is an implicit relation giving K e in terms of y, T and H. In the equivalent Ising model there is a second-order transition when K e passes through the critical value K c at Le = 0. With H = 0 and K e = K c in 9.21 y = sinhK c fcothK p coshK c , gpK c ; 0 sinhK c + gpK c; 0 coshK c , sinhK cg 9.23
246
T T
Fig. 9.2. The critical curve for equilibrium

bond dilute model.
yc
y
is the curve of second-order transitions critical curve in the y; T plane, which is sketched in Fig. 9.2. The bond dilute model is equivalent to the standard Ising model when y = 1, and it can be veri ed from 9.23 that y = 1 corresponds to K p = K c T = Jp =K c. As T decreases from Jp =K c , y decreases steadily and dy=dT = 1 when the curve meets the T = 0 axis at y = yc, given by ; yc = 1 + gpK cK0 : 9.24 1 + coth c For y yc there is no transition at any value of T . The internal energy per primary site is 1 zJ
1 zN ,1 @ lnZ 1 zJ h u = , 2 p ij i j i = , 2 p 2 @ K p ;H 1 zJ fSK ; K + g K ; L CK ; K g : =2 p 9.25 e p p e e e p
9.3 Heat Capacity and Exponent Renormalization

In the bond dilute Ising model, the response functions behave in a rather surprising way at the critical curve shown in Fig. 9.2. To analysis this behaviour we begin by de ning T as the zeroeld critical temperature given by 9.23 for y = y where yc y 1. Suppose that the critical curve in the y; T plane is approached along the line y = y Fig. 9.2 from the disordered region where T T . The heat capacity per primary site is Now de ne
k K2 @u cT; y = kb @T = , b p @@u : Jp Kp y y
9.26
9.3 Heat Capacity and Exponent Renormalization

Ke Ce K e ; K p = @ C@ K; K p ; Ke Cp K e; K p = @ C@ K; K p
247
K 0 gpK e = @gp@ K e ; 0 ;
e
9.27
with similar de nitions of SeK e ; K p and Sp K e; K p . From 9.25
@Ke 1 0 @ K p y = , 2 zJp Sp + gp Cp + @ K p y Se + gpCe + gp C : 9.28 With y = y = constant and H = 0, di erentiation of 9.21 with respect to K p gives @K 0 = Cp + gpSp + @ K e
@u
"
Eliminating @ Ke =@ Kp between 9.28 and 9.29 and using 9.26, 0 = 1 zkb K 2 Sp + gp Cp , Cp + gp Sp Se + gpCe + geC cT; y 2 p 0 Ce + gpSe + ge S 1 = 2 zkb K 2 Sp + gp Cp , CCp + gpSp p S Cp + gp Sp SSe , CCe + geSCe , CSe : , 9.30 0 SCe + gp Se + gpS From 9.10 the heat capacity per site in the equivalent Ising model is 1 0 ci:m:K e = 2 zkb K 2 gpK e: 9.31 e Since it is known that, in two and three dimensions, ci:m: K e diverges to 0 in nity as K e ! K c, gp K e ! 1 as 4Ke = K c , K e ! 0. As all other quantities on the right-hand side of 9.30 are nite at K e = K c it follows that cT ; y , the bond dilute model heat capacity at the transition point for y = y 1, is itself nite.
p y
0 Ce + gp Se + gpS:
9.29
9.3.1 Three-Dimensional Lattices

0 In three-dimensional lattices gpK e 4Ke , and hence gpK e ; 0 : 4K e1, as 4Ke ! +0 with the Ising model critical exponent = 1 8 Sect. 3.6. De ning 4cT; y as cT; y , cT ; y , so that 4c includes 1 , the leading any singular part of c, it follows from 9.30 that since 2 0 K e and hence to 4Ke giving term in 4c is proportional to 1=gp 4cT; y 4Ke ; as 4Ke ! 0 . 9.32
248
9. Applications of Transform Methods Let 4Kp = K , K p , where K = Jp =T . Then 9.29 yields p p
Hence 4Kp 4Ke 1, and 4Ke 4K p1=1, ; as 4Kp ! 0 . With 9.32 this gives 4cT; y 4Kp =1, T , T =1, ;
@ 4Ke = @ K e 4K ; e @Kp @Kp
as 4Kp ! 0 .
9.33 9.34
as T ! T + 0 . 9.35 Since the only magnetic moments in the system belong to the primary site spins, the relative magnetization is given by mK p ; H = h i ip = mi:m: K e; Le : 9.36 It follows, using the second relation of 9.16, that @m 1 @mi:m: = 4K , ; as 4K ! +0 ,
@H
H=0
@ Le
Le =0
9.37 as T ! T + 0 . 9.38
where is the Ising model exponent. With 9.34 this gives
@m
@H
H=0
= 4Kp , =1, T , T , =1, ;
Again, since mi:m:K c ; Le L1= ; as Le ! 0 , 9.39 e where is the Ising Model exponent, we have, from 9.16 and 9.36, mT ; H L1= H1= : 9.40 e 1 by Labelling exponents in the d = 3 bond dilute model at xed y = y the index r, we have, from 9.35, 9.38 and 9.40, 9.41 r= : 1, ; The exponent transformations of 9.41 and 9.45 below are termed exponent renormalization. For T T K p K , similar methods yield relations like 9.34, 9.35 p and 9.38 with 0 , 0 and T , T replacing , and T , T respectively. From 9.35,
r= r = ,1 ,
@c
,1,2 =1, ; @T y T , T
as T ! T + 0 ,
9.42
critical point is replaced by a cusp with vertical tangents. For K p K the spontaneous magnetization of the bond dilute Ising p model is, from 9.36 and 9.16, ms K p = H!0 K p ; H = mi:m: K e : lim 9.43 s
9.3 Heat Capacity and Exponent Renormalization 249 and there is a similar relation with replaced by 0 when T ! T , 0. Since : 0 = = 1 , the Ising model in nite singularity in the heat capacity at the
8
Now mi:m: K e j4Ke j as 4Ke ! ,0, being the Ising model exponent. s Thus, from 9.43, ms T , T =1, 0 ; as T ! T , 0 . 9.44 Hence the critical exponents, when T T , are given by 9.45 r = 1, : 1, 1, 0 ; From 9.41 and 9.45 it is easily veri ed that, if the Ising exponents satisfy the Widom scaling law 3.77 and the Essam Fisher scaling law 3.87 then 0r = r r , 1 ; 9.46 0r + 2 r + 0r = 2 : These exponents also satisfy the scaling laws. Also 0r = r and 0r = r if 0 = and 0 = . The critical curve can also be approached along the line T = T , see Fig. 9.2. With K p = K , 9.21 yields p @K e = C + g 1 + g0 S 4Ke ; as 4Ke ! 0 , 9.47 @y T =T e p Se p where 4Ke = 0 now corresponds to 4y = y , y = 0. Hence we have, 4Ke 4y1=1, ; 9.48 which is similar to 9.34. From 9.32, 4c 4y =1, as 4y ! 0 and we again nd r = , =1 , . For y y, the spontaneous magnetization ms 4K e j4yj =1, 0 so that, again, r = =1 , 0 . It can similarly be shown that the other constant-temperature exponents are equal to those at constant y. For y = 1 the exponents must revert to the standard Ising values. In the present formalism, putting y = 1 in 9.21 or 9.23 yields K e = K p so that 9.34 is replaced by 4Ke = 4Kp. There are thus nite discontinuities in the exponents, since the limits as y ! 1 , 0 are not equal to the values at y = 1. It is remarkable that for any non-zero value of 1 , y, however small, there is a qualitative change in the properties of the model, the in nite discontinuity in the heat capacity being replaced by a cusp. However, the smaller the value of 1 , y, the narrower becomes the range around the critical point
0 =, r 0 0; 0= r 0
250
where the exponent changes have a signi cant numerical e ectFisher and Scesney 1970.2 The exponent renormalization results from taking y as an independent variable which can change at constant T or remain constant when T changes. It is not simply a consequence of the existence of broken bonds. If the latter are regarded as defects with an activation energy 4", then = exp4"=T and K e , as given by 9.16, depends only on T . Thus 4Ke 4Kp and no renormalization occurs. The concentration y in zero eld becomes a function of T . In fact, it is more realistic physically to take y as an independent variable in the mixture interpretation of this model given in Sect. 9.2.1 than in the broken bond interpretation in Sect. 9.2.2. In the mixture case, the mole fraction of solute is zy=2 + z and is a legitimate thermodynamic density.
9.3.2 Two-Dimensional Lattices

In two dimensions = 0 and there is no renormalization. However, as we deduced from 9.30, the transition-point heat capacity is nite for y 1. 0 From 8.119 and 9.10, gp K e , ln j4Ke j as 4Ke ! 0 and thus 1 4cT; y , ln j4T j ; as 4T = T , T ! 0 . 9.49 1 dj4cj dT j4T jln j4T j2 ; There is thus a cusp with vertical tangents at the transition point, as in the d = 3 case. For the d = 2 case this represents a further variety of = 0 behaviour discussed in Sect. 3.4. Renormalization relations like 9.41 and 9.45 were derived by Essam and Garelick 1967 for the Syozi dilution model Syozi 1972, Essam 1972 and Example 9.1, which is slightly more elaborate than the one treated here. Widom 1967 derived the expression =1 , for the compressibility exponent in a decorated mixture model with the density of one component xed. Fisher 1968 used thermodynamic arguments, in conjunction with examples from statistical mechanics, to establish general conditions for exponent renormalization. These were the existence of a `hidden variable' subject to a constraint, for instance the presence of an equilibrium distribution of a xed number of impurity atoms. Ìdeal' exponents are obtained when the largest possible number of elds are taken as independent variables, that is = n , 1, and are renormalized when a eld is replaced as an independent variable by a density. The replacement of by y in the model of Sect. 9.2 is an example. Version II of the dilute Ising model in Sect. 6.6.2 can be regarded as a compressible Ising model since the volume per spin v = v0 = , where v0 is the volume per site, and v can thus be adjusted by altering the pressure
2
These authors used a di erent model, but the same general principles must apply here.
9.4 Fisher's Decorated Antiferromagnetic Model
251
or chemical potential. Renormalization is to be expected if v is taken as an independent variable. The same applies to the compressible Ising model of Baker and Essam 1971. Renormalization will not occur in a closed-form approximation since there is then a nite discontinuity in the heat capacity, giving = 0 for d = 3.

Suppose that on a decorated square lattice Fig. 9.1 the microsystems on the N horizontal edge secondary sites have di erent properties from those on the N vertical edge secondary sites. Labelling quantities connected with the horizontal and vertical edges by the subscripts h and v respectively, we can de ne h 1; 1 and v 1; 1 by sums like 9.3 but restricted to the secondary microstates on the horizontal and vertical edges respectively. Then, by a derivation similar to that of 9.7, Z = RN =4 Zpi:m K eh; K ev ; Le ; 9.50 hv where Rhv = h +1; +1 v +1; +1 h ,1; ,1 v ,1; ,1 h +1; ,1 v +1; ,1 2 expf,2 +1 + ,1 =T g ; 9.51 where Zpi:m: refers to an anisotropic Ising model on the primary sites with interaction parameters K eh and K ev respectively in the horizontal and vertical directions see 8.97 for the zeroeld case. Here x, exp4K ex = x +1; +1 ; ,1 1; ,1 ; x = h; v ; 2 +1 x 9.52 +1; +1 +1; +1 h v exp2Le = expf ,1 , +1 =T g : h ,1; ,1 v ,1; ,1 Suppose that both primary and secondary sites are occupied by Ising spins with primary secondary interactions. The interaction constants are Jh and Jv , and the magnetic moments of the secondary spins rh m0 and rv m0 for horizontal and vertical edges respectively, m0 being a standard magnetic moment. Then ^ hx i ; j ; ij = ,Jx ij i + j , rx ij H ; x = h; v ; 9.53 ij = 1 being the spin component on the secondary site. Let Gx = Jx =T; F x = rx H=T ; x = h; v ; 9.54 and then, from 9.3, x 1; 1 = 2 cosh2Gx F x ; x = h; v : 9.55 +1; ,1 = 2 coshF x ; x
252
where h i ; j ip;x for x=h,v are the nearest-neighbour correlations in the equivalent Ising model for horizontal and vertical edges respectively. Using 9.50, the relative magnetizations for the two types of secondary sites are given by m = 1 @ ln Z
x
Generalizing 9.10, @ ln Zpi:m: = h ; i = g K ; K ; L ; i j p;x p;x eh ev e @ K ex
x = h; v ;
9.56
and substituting from 9.11, 9.52, 9.55, 9.56 and into 9.57 1 mx = 4 1 + gp;xK eh; K ev ; Le tanh2Gx + F x , tanh2Gx , F x , 2 tanhF x + tanhGx + 1 mi;m K eh; K ev ; Le tanh2Gx + F x + tanh2Gx , F x ; 2 x = h; v : 9.58 The basic assumptions of Fisher's 1960 model are: a The magnetic moments of all primary site spins are zero and those of all secondary site spins are equal, so that rh = rv = 1. Hence +1 = ,1 = 0 ; F h = F v = F = H=T : 9.59 b The horizontal Ising interactions are ferromagnetic in character whereas the vertical Ising interactions are antiferromagnetic, the interaction constants being of equal magnitude. Thus Jh = ,Jv 0 and Gh = ,Gv = G = Jh =T : 9.60 Substitution from 9.59 and 9.60 into 9.55 yields h 1; 1 = v 1; 1 = 2 cosh2G F ; 9.61 h +1; ,1 = v +1; ,1 = 2 coshF : From 9.52 and 9.59, 1 h +1; +1 h ,1; ,1 ; K eh = K ev = K e = 4 ln h +1; ,1 2 9.62 Le = 0 :
N @F x = 1 @@ ln x +1; +1 + ln x ,1; ,1 + 2 ln x +1; ,1 4 Fx i:m: @ L i:m: x = h; v + @ ln Zp @ K ex + @ ln@Zp @ F e ; @ K ex @ F x Le x
9.57
9.4 Fisher's Decorated Antiferromagnetic Model ferromagnetic model.
253
Fig. 9.3. The ground state of Fisher's anti-
This shows the remarkable feature of this model, which is that the entire G; F plane is mapped into the K e ; 0 line of the Ising model ferromagnet.3 The exact results for the decorated model at all values of T and H are obtainable from the known exact results see Sect. 8.10 for the square lattice Ising ferromagnet in zero eld. In the H = 0 ground state for the decorated model, the primary site spins, which have zero magnetic moment, are all parallel whereas the secondary site magnetic dipoles on the horizontal and vertical edges respectively are oppositely oriented see Fig. 9.3. This is an antiferromagnetically ordered state with magnetization on two sublattices cancelling to give zero overall magnetization. The ground-state energy per primary site is u = ,4Jh. By similar considerations to those of Sect. 4.2 it can be shown that there is a critical eld Hc = 2Jh and that the antiferromagnetic ground state is stable at T = 0 for ,Hc H Hc. For H Hc or H ,Hc the stable state at T = 0 is paramagnetic with all magnetic moments aligned with the eld. The Fisher model is sometimes referred to as a superexchange antiferromagnet since the only function of the primary site spins is to couple the secondary site magnetic moments. The behaviour of the Fisher model is similar to that indicated for an Ising antiferromagnet on a loose-packed lattice by series or meaneld methods. By substitution from 9.61 into the rst relation of 9.62 the latter can be expressed in the form sinh2 2G = cosh2 F exp4K e , 1 : 9.63 This gives K e 0 for all G and F , so that, although the decorated model is an antiferromagnet, its equivalent Ising model is a ferromagnet. Since Le = 0 for all T and H, mi:m: in 9.58 must be replaced by mi:m: , the spontaneous s magnetization. Again, from 9.62, gp;h = gp;v = gpK e ; 0 as de ned by 9.10. Substituting 9.59 and 9.60 into 9.58 then yields
3
It is shown below that K e 0 for all G and F .
254 2

Ke Kc
Fig. 9.4. The critical curve for

Fisher's model.
J1
H
Ke Kc
Paramagnetic Antiferromagnetic
T=Tc
1 m = 2 mh + mv
= 1 1 + gpK e ; 0 tanh2G + F , tanh2G , F , 2 tanhF 4 + tanhF ;
= 1 mi:m: K e; 0ftanh2G + F + tanh2G , F g ; 2 s where m is the overall relative magnetization and a sublattice order parame1 ter de ned as in 4.32. Since the primary lattice is square, gp = , 2 ui:m: =Jh, where ui:m: is given by 8.131 and 8.132 and mi:m: is given by 8.144 s with K e replacing K in both cases. For K c K e , mi:m: = 0. s From 9.50, 9.61 and 9.62 the con gurational energy per primary site is given by u = , Jh @ ln Z = ,J 1 + g K ; 0
1 = 2 mh , mv
9.64
N @G
The condition K e = K c yields, from 8.20, p 9.66 sinh2 2G = cosh2 F exp4K c , 1 = 2 cosh2 F 1 + 2 between G = Jh =T and F = H=T . It can be shown that H is a monotonically decreasing function of T on the curve in the T; H plane de ned by 9.66; sketched in Fig. 9.4. Inside the curve K e K c and hence by 9.64, 6= 0 while outside the curve, K e K c and = 0. It follows that the curve de ned by 9.66 is the critical curve between the antiferromagnetic sublattice ordered and paramagnetic regions in the T; H plane. From 9.66, F =2G ! 1 as T ! 0 and the critical curve meets the positive H-axis at H = 2Jh = Hc , as would be expected from ground-state considerations. The critical curve meets the T -axis orthogonally at T = Tc , where Tc is the N el temperature. e Putting F = 0 in 9.66, gives
tanh2G + F + tanh2G , F :
9.65
sinhGc = sinh2Jh =Tc = 21 + 21=2 ;
255
Tc = 1:3084Jh : On the T -axis it can be seen, from 9.64 with F = 0, that m = 0; mi:m K e; 0 tanh2G ; T s As T ! 0, mh = ,mv ! 1. @m 16T @ H mh = ,mv = :
9.67
80 ;
Tc , Tc .
9.68
We now consider the magnetic susceptibility. From 9.64,
+ 8 1 , gp K e ; 0 sech2 F 0 + gpK e tanh2G + F , tanh2G , F , 2 tanhF 2 ; 9.69 0 gp K e being given by 9.27. With F = 0, the zeroeld susceptibility
0 =C
= 4 1 + gpK e; 0 sech2 2G + F + 4sech2 2G , F
@m
9.70 where C is de ned by 4.46. From 9.70, 0 is continuous for all T and 0 tends to zero as T ! 0 or T ! 1. From 8.119, gp K e ln j4Ke j as 4Ke = K c , K e ! 0 so that the 0 against T curve has a vertical tangent at T = Tc. Its maximum is at a temperature T Tc , as series methods indicate is the case for a loose-packed lattice Ising antiferromagnets. Take a point T ; H , with H 6= 0, on the critical curve see Fig. 9.4. 0 From 9.69, @m=@ H gpK e ln j4Ke j as 4Ke ! 0, provided that H 6= 0. 0 on the constanteld line H = H 6= 0, the magnetic Hence as T ! T 0 susceptibility T T gpK e ln jT , T j. Thus T has a logarithmic in nity at the critical temperature for all H such that 0 jHj Hc and its continuity on the H = 0 axis is an exceptional case. It follows similarly that as H ! H 0 on the constant-temperature line T = T then @m=@ H ln jH , H j. Thus if m is plotted against H for T = T Tc the curve has a vertical tangent at the critical eld value H , instead of the nite slope discontinuity predicted for the loose-packed lattice Ising antiferromagnet by meaneld theory Sect. 4.3. It can be deduced from 9.65 that the constanteld heat capacity cT; H ln jT , T j as T ! T 0 for all H such that jH j Hc , including H = 0 when T = Tc. Again from 9.64 the order parameter T , T 1=8 or H , H1=8 as the critical curve is approached along the lines H = H or T = T respectively.
@H T
= 2C fsech2 2G 1 + gp K e ; 0 + 1 , gpK e ; 0g ; T
256
Versions of the Fisher model are possible for d = 3. For instance, a bodycentred cubic lattice of N sites can be divided into planes of sites passing through one cubic axis and making angles of =4 with the other cubic axes. There are 2N edges within these planes and 2N edges connecting sites in adjacent planes. For the susceptibility and heat capacity the ln jT , T j or ln jH , H j dependence near the critical line is replaced by dependence on : jT , Tj, or jH , H j, where = 1 . The order parameter T , T 8 : 5 . If m rather than H is kept constant as T varies, or H , H where = 16 renormalization of the exponents occurs Fisher 1968. Accurate transition curves away from the H = 0 axis are also obtainable for the ferrimagnetic model of the next section, although the entire T; H plane does not map into the Le = 0 axis, as it does in the Fisher model.
9.5 The Decorated Lattice Ferrimagnet

In this type of model Syozi and Nakono 1955, Syozi 1972, Bell 1974b the secondary and primary sites are occupied by Ising spins with magnetic moments m0 and rm0 respectively. There is an antiferromagnetic primary secondary interaction. The model considered here can be related to the ferrimagnetic model treated in Sect. 4.5 by putting Na = N , Nb = 1 zN and, thus, s = z=2, 2 where z is the coordination number of the primary lattice. For conformity with Sect. 4.5 the labels à' and `b' will be used in this section for quantities connected with the primary and secondary sites respectively.4 From condition 2 of Sect. 9.1 there is no interaction between the b spins but we assume a direct ferromagnetic interaction between the spins of nearest-neighbour pairs on the a lattice. With interaction constants Jab and Jaa ^ h i ; j ; ij = Jab ij i + j , Jaa i j , H ij ; Jab ; Jaa 0 : 9.71 Unless stated otherwise, it will be assumed that z 2r. For small jHj the ground state will be ferrimagnetically ordered with the b spins aligned with the eld and the a spins against it. For larger jHj all spins will be aligned with the eld in the ground state. At T = 0 transitions between the ground states will occur at H = Hc. The critical eld Hc can be obtained by putting 1 s = 2 z, zba = 2 in 4.63, giving Hc = zJab=r : 9.72 De ning
4
The model shown in Fig. 4.8 is an example of a decorated lattice ferrimagnet based on a triangular lattice of a sites with the b sites decorating the edges of the a lattice.

G = Jab =T ; G0 F
257
= Jaa =T =
G;
= Jaa =Jab ;
9.73
9.74 +1; ,1 = 2 coshF exp,G0 and, since the a site magnetic moment is equal to rm0 , 1 = rH : 9.75 The relative magnetization ma is equal to that of the equivalent Ising model whereas mb is given by 9.58 with Gx = ,G, F x = F and gp;x = gp, de ned by 9.10. Hence ma = mi:m: K e ; Le; 1 mb = tanhF + 4 1 + gpK e; Le
= H=T ; the partition function is given by 9.7 with 1; 1 = 2 cosh2G F expG0 ;
tanh2G + F , tanh2G , F , 2 tanhF
9.76
The overall relative magnetization is given by +z 9.77 m = 2rma + zmb : 2r As T ! 0, m ! z , 2r=z + 2r when jHj Hc whereas m ! 1 for jHj Hc . Now consider the H = 0 axis. From 9.74, 9.75 and 9.9, H = F = 0 implies Le = 0 and then putting Ke = Kc, cosh2Gc exp2 Gc = exp2K c ; 9.78 for the critical temperature Tc = Jab =Gc . At H = 0, from 9.76, ma = mi:m: K e ; 0 = mi:m:K e ; s 9.79 mb = , tanh2Gmi:m: K e ; 0 = tanh2Gmi:m:K e ; s
1 , 2 mi:m: K e; Le tanh2G + F + tanh2G , F :
where mi:m: K e 0 is the spontaneous magnetization of the Ising ferros magnet. Hence 9.77 yields, for the spontaneous magnetization of the ferrimagnet, z ms = mi:m: K e 2r , 2rtanh2G : 9.80 s +z
258
To avoid ambiguity we specify the limit H ! +0 so that the sign must be chosen to make ms 0. When T Tc, K e K c so that mi:m: = 0 while, s as T decreases from Tc to 0, tanh2G increases from tanh2Gc to 1. Since r 1 z, ma = ,mi:m: = ,1 at T = 0, in accordance with our discussion of s 2 the ground state. We can distinguish three cases: 1 a r 2 z tanh2Gc , b r = 1 z tanh2Gc , 2 1 c 2 z r 1 z tanh2Gc . 2 The plots of m against T are qualitatively similar to the curves for r = 2 ,r0 and 2:6 respectively of Fig. 4.9. For case c there is a compensation temperature Tcomp Tc, given by 9.81 r = 1 z tanh2Jab =Tcomp ; 2 where ms = 0 because the rst factor on the right-hand side of 9.80 is zero. For Tcomp T Tc ma = +mi:m: K e 0; mb 0; 9.82 s but, for 0 T Tcomp, ma = ,mi:m: K e 0; mb 0: 9.83 s Both ma and mb change sign discontinuously at T = Tcomp although jma j and jmb j are continuous. Since mi:m: K e j4Kej as 4Ke = K e , K c ! 0, being the Ising s exponent, 9.80 gives ms Tc , T as T ! Tc , 0, except in case b. For case b, r , 1 z tanh2G Tc , T as T ! Tc , 0 and hence ms Tc , T +1 . 2 Since 0 1 for both d = 2 and d = 3 lattices it follows that the plot of ms against T at H = 0 has a vertical tangent at T = Tc except for case b where the tangent is horizontal. The zeroeld susceptibility can be deduced from 9.76, Example 9.2. We now consider the T; H plane of the decorated ferrimagnet. In addition to the segment 0; Tc of the H = 0 axis, where m undergoes a discontinuous change of sign, there are rst-order transition lines on which jHj 6= 0 and for which exact relations can be derived Bell 1974b. This arises from the fact that the mapping from the T; H plane of the decorated ferrimagnet into the K ,1 ; Le plane of the Ising model is not one-to-one. Putting Le = 0 e in the second relation of 9.9 and substituting from 9.74 and 9.75 gives cosh2G , F 9.84 cosh2G + F exp4rF =z = 1 : As would be expected, F = 0 maps into Le = 0. However, when r 1 z there 2 is another branch in the T; H plane which maps into Le = 0 and which is given, after rearrangement of 9.84, by
259
Jab 4
r=1
r = 2:7
2 4 T=Jab 6 Fig. 9.5. The Le = 0 locus and rst-order transition heavy lines for the decorated ferrimagnet with Jaa = Jbb . Reprinted from Bell 1974b, by permission of the publisher IOP Publishing Ltd.
This curve is clearly symmetrical about H = 0 and meets the T = 0 axis where H = zJab =r = Hc . The branch 9.85 of the Le = 0 locus cuts the T -axis orthogonally. Taking the limit of the right-hand side of 9.85 as F ! 0 gives, after a little rearrangement, the formula r = 1 z tanh2G 9.86 2 for the temperature at the point of intersection. The intersection of the branch 9.85 of the Le = 0 locus with the T -axis occurs at a temperature T Tc for case a and at T = Tcomp Tc for case c. The two cases are illustrated by the curves for r = 1 and r = 2:7 respectively in Fig. 9.5. These are derived from the model of Fig. 4.8, where the b sites decorate the edges of a triangular lattice of a sites, with Jaa = Jab , which gives Tc = 4:4354Jab and 1 z tanh2Gc = 1:2679. In the H 0 half-plane, Le is negative between 2
exp4G = sinh 1 + 2r=zF : sinh 1 , 2r=zF
9.85
260
the T -axis and the curve 9.85 and positive outside this curve, these signs being reversed in the H 0 half-plane. By substituting 9.74 into the rst relation of 9.9 and a little manipulation, the K e = K c curve can be obtained in the form 2 exp 4K c , G0 , 1 = sinh 22G : 9.87 cosh F Equation 9.78 is reproduced by putting F = 0. As jF j ! 1, G0 ! K c which implies that T ! Jaa =K c and de nes the asymptotes of the K e = K c curves as jHj ! 1. The broken line in Fig. 9.5 represents the K e = K c curve, with K e K c to the left and K e K c to the right. The rst-order transition lines in the T; H plane of the decorated ferrimagnet are the images of the segment 0 K ,1 K ,1 , Le = 0 from the e c K ,1 ; Le plane of the Ising model. So these lines of discontinuity are the e parts of the Le = 0 locus which lie to the left of the K e = K c curve. They are represented by heavy lines in Fig. 9.5 and include the segment 0; Tc of the T -axis in all cases. For case a there are two other rst-order transition lines which are mirror images of each other in the T -axis. They start from the points H = Hc on the H-axis and terminate at critical points at the intersections of the Le = 0 locus with the K e = K c curve. For case c the entire branch of the Le = 0 locus de ned by 9.85 lies in the K e K c region and is thus a rst-order transition line crossing the 0; Tc segment on the T -axis at T = Tcomp. The compensation point is thus the intersection of rst-order transition lines as indicated by the meaneld theory of Sect. 4.5. The con gurational energy is continuous over the segment 0; Tc of the T axis but, together with the magnetization m, it is discontinuous at rst-order transition points for which H 6= 0. The phase pattern is di erent when Jaa = 0. Although the Le = 0 locus is una ected there is a signi cant change in the K e = K c curve. Putting G0 = 0 in 9.87, it can be seen that this curve no longer goes to F = 1 for nonzero T but turns to meet the H-axis at H = 2Jab . This is illustrated by the broken curve in Fig. 9.6 which is drawn for a model similar to that of Fig. 9.5 1 but with Jaa = 0 so that Tc = 1:7449Jab and 2 z tanh2Gc = 2:4495. Since 1 z, the transition points on the H-axis at H = Hc 2Jab Hc when r 2 are now isolated since K e K c in the adjacent parts of the T; H plane. For case a there are no transitions except on the T -axis since the branch of the Le = 0 locus given by 9.85 lies entirely in the K e K c region of the T; H plane. Case c with r = 2:7 is illustrated in Fig. 9.6. The rst-order transition segment 0; Tc on the T -axis is crossed at the compensation point T = Tcomp by another rst-order transition line on the branch of the Le = 0 locus given by 9.85. This transition line terminates at critical points like B in Fig. 9.6 and its mirror image at the intersection of the Le = 0 locus with the K e = K c curves. These critical points are reached by reducing the temperature from Tcomp and so are the rst examples we encounter of lower critical points.
261
Jab ma 0
2
B
ma 0
1 2 T=Jab Fig. 9.6. As Fig. 9.5 for Jaa = 0, r = 2:7. Reprinted from Bell 1974b, by permission of the publisher IOP Publishing Ltd.
1 We now consider brie y the situation when r 2 z. The rst factor on the right-hand side of 9.80 is now positive and hence mi:m: = +mi:m: for the s entire range 0 T Tc with H = +0. For r = 1 z, ms ! 0 as T ! 0 and the 2 plot of ms against T is qualitatively similar to the r = 3 curve of Fig. 4.9ii 1 whereas for r 2 z the plot is similar to the r = 4 curve of Fig. 4.9ii. 1 z , 0 the branch of the Le = 0 locus given by 9.85 shrinks to a As r ! 2 segment on the H-axis and it is absent for r 1 z. The only transition line 2 is the 0; Tc segment of the T -axis. There is general qualitative agreement as regards the phase diagrams between the exact theory of the present section and the meaneld theory of Sect. 4.5, with Jbb = 0. However, critical parameter values di er considerably. For instance when Jaa = 0 the exact theory above gives r 1 z tanh2Gc = 2 2:4495 as a condition for the occurrence of a compensation point in the model ps = p3. of Fig. 4.8. The meaneld equation 4.73 gives r r0 =
262
9.6 The Kagom Lattice Ising Model e

As well as the decoration transformation discussed in Sect. 9.1 the startriangle transformation of Sect. 8.4 is useful for relating other models to the Ising model on lattices for which accurate results are available. In the present section the partition function of the zeroeld Ising model on the two-dimensional Kagom lattice see Appendix A.1 is related to the partie tion function of the zeroeld Ising model on the honeycomb lattice by applying decoration and star-triangle transformations successively. In Fig. 4.8 the b sites, which decorate the edges of the triangular a site lattice, form a Kagom lattice. Since the triangular and honeycomb lattices are dual to each e other there is a one-to-one correspondence between their edges. As shown in Fig. 9.7, the secondary sites decorating the edges of a honeycomb lattice also form a Kagom lattice. e Take a honeycomb lattice of N sites the white sites of Fig. 9.7 and decorate each edge with a secondary site one of the black sites of Fig. 9.7. Suppose all sites are occupied by Ising spins and that there is a primary secondary interaction with interaction constant J 0 0. By 9.3, in zero magnetic eld, +1; +1 = ,1; ,1 = 2 cosh2G0 ; 9.88 +1; ,1 = 2 ; where G = J 0 =T 9.89 and +1 = ,1 = 0 : 9.90 Hence, by 9.9,
Fig. 9.7. The decorated honeycomb lattice.
9.6 The Kagom Lattice Ising Model e
263
exp2K e = cosh2G ; Le = 0 9.91 and, from 9.7, the partition function Z of the decorated model is given in terms of the partition function Z h of the zeroeld honeycomb lattice Ising ferromagnet by Z = 64 cosh3 2G N =4 Z h N; K e 9.92 noting that, by 9.91, K e 0. The next step is a star-triangle transformation which removes the white sites those of the original honeycomb lattice and replaces the interactions along the full-line segments of Fig. 9.7 by interactions along the broken segments. By 8.28 and 8.27, Z = RN Z kg 3N=2; K ; 9.93 where exp4K = 2 cosh2G , 1 ; 9.94 R4 = exp4K exp4K + 3 2 : Since this implies K 0, Z kg is the partition function for a zeroeld Ising ferromagnet on the Kagom lattice and K = J=T kg . Combining 9.92 and e 9.93 and replacing N by 2 N yields the relation 3 N =6 1 3 exp Z h 2N=3; Ke ; 9.95 Z kg N; K = 8 exp4K +K + 3 2 ,4K exp4 where, from 9.91 and 9.94, exp2K e = 1 exp4K + 1 : 9.96 2 From 8.39 the honeycomb Isingp model critical temperature is given by cosh2K e = 2 or exp2K e = 2 + 3. Thus, from 9.96 the Kagom Ising e ferromagnet critical temperature is given by p exp4K c = 3 + 2 3 ; Tc = 2:1433J : 9.97 This is slightly lower than the value Tc = 2:2692J for the square lattice, which also has coordination number z = 4. From arguments similar to those used in Sects. 9.4 and 9.5, the zeroeld relative magnetization on the decorating black sites is tanh2Gmh K e s where mh is the spontaneous magnetization for the honeycomb lattice Ising s model. Since the black sites form the Kagom lattice the spontaneous mage netization mkg K = tanh2Gmh K e s s 1 + 3X1 , X 1=2 31 3 1 3 2 1=8 ; = 1 , 128 X1 ,+ X1 ++ XX2 1 + X2 X6 3 9.98
264
where X = exp,4K. The tanh2G factor was transformed using 9.91 and 9.96 while the expression 8.163 for mh , with K e substituted for G, was s transformed using 9.96. The result 9.98 was obtained by Naya 1954.5
9.7 A Modi ed Star-Triangle Transformation and Three-Spin Correlations

In Sect. 8.4 the zeroeld Ising model partition function for the honeycomb lattice was related to that for the triangular lattice by using a star-triangle transformation to remove half the honeycomb sites. In this section we generalize this procedure to the situation where the honeycomb Ising model is acted on by a eld H, it being assumed that all spins have the same magnetic moment m0 . In the limit H ! 0 this yields a simple method of determining the correlation for the three spins on an elementary triangle of the triangular lattice which is used in the alternative ferrimagnetic model of the next section. Putting L = H=T , the expression cosh G + + in 8.24 and 8.25 is modi ed to cosh G + + + L which has four distinguishable values, corresponding to = = = 1 and = = , = 1. Accordingly four adjustable parameters are needed on the right-hand side of the relation corresponding to 8.25 which is modi ed to 2 cosh G + + + L = R exp K + + + L1 + + + K 3 : 9.99 The new parameters L1 and K 3 correspond respectively to a eld acting on the triangular lattice and a three-spin interaction energy. De ning x; y = coshxG + yL ; 9.100 9.99 yields the four relations 23; +1 = R exp3K + 3L1 + K 3 ; 23; ,1 = R exp3K , 3L1 , K 3 ; 21; +1 = R exp,K + L1 , K 3 ; 9.101
21; ,1 = R exp,K , L1 + K 3 : Hence the four parameters R, K , L1 and K 3 are given in terms of G and L by
See Syozi 1972 for some related transformations.
9.7 A Modi ed Star-Triangle Transformation and Three-Spin Correlations
265
When L = 0, 9.102 reduces to L1 = 0, K 3 = 0 and the relations for R and K given by 8.27. Taking the number of original honeycomb sites as 2N , half of these the white sites of Fig. 8.3 are removed by performing the star-triangle transformation over all the downward pointing triangles. From 9.99 this yields
3 exp8K 3 = 3;; +1 1;; ,1 3 : 3 ,1 1 +1
R8 = 2563; +13; ,1 1; +11; ,1 3 ; 3 +13 ,1 exp8K = 1;; +11;; ,1 ; +1 1 +1 exp8L1 = 3;; ,11;; ,1 ; 3
9.102
Z h 2N; G; L = RN
t X
b H i =T = ,K
n:n: X
f ig
b exp ,H i =T ; X
fig
j , K3
The sum on the right-hand side of the rst of these equations is the partition function for Ising spins on the triangular lattice of N sites the black sites of Fig. 8.3 with a eld and a three-spin interaction. The latter is represented by the term with coe cient K 3 in which the summation is over the N downward pointing triangles in the triangular lattice. In the eld coe cient 3L1 + L, the term L is present because each black site spin has a magnetic moment m0 and the 3L1 term because each black site belongs to three downward pointing triangles. For L = 0 the triangular lattice partition function reduces to that of a standard pair interaction Ising model in zero eld. From 9.103, h mh = 21 @ ln Z N @L 1 @ ln R + 3 h i @ K + mt 1 + 3 @ L1 + h = 2 @L 2 i j @L i @@K 3 : i j ` L @L 9.104 From 9.102, @ R=@ L = @ K =@ L = 0 at L = 0 and, with L = 0 in 9.104, 3 1n mh G = 2 mt K + 4 mt K tanh3G + tanhG s s s o 1h + 4 i j ` i tanh3G , 3 tanhG ; 9.105
fi;jg
i j , 3L1 + L
X
fi;j;`g
i j `:
9.103
where mh and mst are spontaneous magnetizations for the Ising ferromags net on honeycomb and triangular lattices respectively. However, since the triangular lattice is formed from one sublattice of the honeycomb lattice by
266
the star-triangle transformation it follows that mh G = mt K where G s s and K are connected by 8.27.6 The three-spin correlation in zero eld is therefore given by G h i j k i = mt K 4 , 3 tanh3, 3+ tanhG ; s tanh3G tanhG 1=2 t K 3 + 1 + X , 2 1 + 3X = ms ; 9.106 1,X 1 , X3 where X = exp,4K . Here h i j ` i is evaluated for a downward pointing triangle of sites, but for an isotropic triangular lattice the value for an upward pointing triangle of sites will be the same by symmetry. Using a di erent method, Baxter 1975 obtained the result 9.106 and the generalization to unequal interactions in the three triangular lattice directions. Enting 1977 developed an alternative approach and gave a brief outline of the variant of his method which is used here.
9.8 The Star-Triangle Ferrimagnet

An alternative ferrimagnet based on the triangular lattice is shown in Fig. 9.8.7 The a sites form a triangular sublattice of the original triangular lattice and the b sites a honeycomb sublattice of the original triangular lattice so that Nb = 2Na and s = 2. Magnetic moments rm0 and m0 are associated respectively with the spins on the a and b sites. There are antiferromagnetic interactions between the spins of ab nearest-neighbour pairs, that is along the edges indicated by full-line segments in Fig. 9.8, but no bb nearest-neighbour interactions. Hence each b spin interacts only with the a spins at the vertices of the triangle with the b site at its centre. A startriangle transformation can thus be performed to transfer the interactions to the broken line segments in Fig. 9.8, which form the edges of the triangular lattice of a sites. The formalism of the previous section can be used, with some modi cations. Since there is a b site inside every triangle of a sites, R, K and 3L1 in 9.103 must be replaced by R2 , 2K and 6L1 respectively. As the ab interaction is antiferromagnetic, G must be replaced by ,G which, from 9.100 and 9.102 leaves R and K invariant but changes L1 and K 3 to ,L1 and ,K 3 respectively. Hence the partition function for the ferrimagnet is given by
6 7
This result can be veri ed from 8.162 and 8.163. This is model i of Bell 1974a, 1974b. See also Lavis and Quinn 1983, 1987.
b a b a b b b
a b a b a
b a b a b
a X
9.8 The Star-Triangle Ferrimagnet
267
Fig. 9.8. The star-triangle ferrimagnet.
Z G; L = R2N a
b H i =T = K 0
where
K0
n:n: X
X X i j + L0 , 6L1 i , K3 i j `; fi;jg fig fi;j;`g
f ig
b expf,H i =T g ;
9.107
= 2K ; L0 = rL ; 9.108 R, K , L1 and K 3 being de ned by 9.102. The sum over fi; j; `g is taken over all the 2Na triangles of a sites. Treating L and L0 , for the moment, as independent variables the zeroeld relative magnetizations ma and mb , are given by m = 1 @ ln Z = mt K 0 ;
s @ L0 L=L0 =0 1 @ ln Z t K 0 @ L1 = ,3ms mb = 2N @ a @ L L=L0 =0 @ KL L=0 , h i j ` i @ L3 : L=0
Na
9.109
From 9.108 and either 9.102 or 8.27 exp2K 0 = 2 cosh2G , 1 : 9.110 The critical temperature for the ferrimagnet can be obtained by putting K 0 = K c where K c is the critical parameter value for the triangular lattice. Thus p 1 cosh2Gc = 2 exp2K c + 1 = 1 3 + 1 ; 2 9.111 Tc = J=Gc = 2:4055J : From 9.109 the overall spontaneous magnetization ms is given by
268
where c = cosh2G. The last expression of 9.106 can be used for h i j ` i, with K 0 substituted for K so that X = exp,4K 0 = 1=2c , 12 . Finally t 0 ms = ms Kr +r2, f c ; 9.113 where " c2 , 11=2 3 + 2c , 1 c3 + 1 1=2 : 9.114 f c = cc + 1 c3 , c It has been assumed for simplicity that the triangular lattice spontaneous magnetization takes the positive value mt . However, putting H = +0 ims plies m 0, as in Sect. 9.5, and hence +mt must be used in 9.113 when s t when r r f c and ,ms f c. It can be veri ed that f c = 0 at T = 1 c = 1 and that f c increases steadily as T decreases, assuming the value 2 at T = 0 c p 1.1 Atp critical temperature T = Tc, where = the 1 p c = 2 3 + 1, f c = 4 3=2 2 2 3 , 3 = 1:72758. Equation 9.113 was obtained by Lavis and Quinn 1983 using the results of Baxter 1975. For 2 r 1:72758, r = f c at some temperature Tcomp in the range 0; Tc and, from 9.113, this de nes a compensation point. The shapes of the curves of ms against T are similar to those obtained in Sects. 4.5 and 9.5.
r + 2ms = rma + 2mb n o = mt K 0 r , 3 tanh3G + tanhG s 2 1h , 2 i j ` i tanh3G , 3 tanhG ; c , 1=2 = rmt K 0 , 2c 2 1 c + 1 s 1 h t 0 , i 3cms K , h i j ` ic , 1 ;
9.112
9.9 The Unsymmetrical Ising Model

When the Ising model represents a magnetic system it is natural to assume, as has been done until now, that the energies of ++ and ,, pairs are equal and that the two spin states have equal statistical weight. However, in some other cases, it is useful to modify these conditions. It will be shown that the resulting unsymmetrical model is equivalent to a standard Ising model in a temperature dependent eld. Suppose that the i = 1 and i = ,1 states represent di erent species of the same component and that ++ and ,, nearest-neighbour pairs have unequal interaction energies. The zeroeld Hamiltonian is then
9.9 The Unsymmetrical Ising Model
269
b 4 H=1
which can be rearranged to give
fJ++ 1 + i 1 + j + J,, 1 , i 1 , j fi;jg + J+, 1 + i 1 , j + 1 , i 1 + j g ;

n:n: X
n:n: X
9.115 9.116
b H = zNR , Je
where
fi;jg
i j + 4J
X
fig
i;
1 R = 8 2J+, + J++ + J,, ; 1 4J = 4 zJ++ , J,, :
Je = 1 2J+, , J++ , J,, ; 4
9.117
Suppose also that the states i = 1 and i = ,1 are composed of substates, the degeneracies being !+ and !, respectively. Then the partition function is X b 9.118 Z = !N + !N , exp,H=T ; + , where
f ig
1 N = 2
X
fig
1 i :
9.119
By comparison with 2.67 Z = !+ !, N =2 exp zNR Z i:m: T; He ; 9.120 where 1 T ln !+ , 4J : 9.121 Hc = 2 !, Here Z i:m: is the partition function of a standard Ising model on the same lattice and at the same temperature as the unsymmetrical model but acted on by the eld He, given as a linear function of T by 9.121. The interaction constant Je of the Ising model and the quantity 4J in 9.121 are given by 9.117. Now assume Je 0 so that the equivalent Ising model is a ferromagnet. When 4J and ln!+ =!, have the same sign, the line H = He in the T; H plane of the Ising model crosses the T -axis at T = T0 where 2 T0 = ln!4=J! : 9.122 + ,
270
Let Tc be the zeroeld Ising ferromagnet critical temperature for the given value of Je . Then if T0 Tc the point T = T0 lies on the segment of discontinuity 0; Tc on the T -axis of the Ising model and the unsymmetrical model undergoes a rst-order transition at T = T0 . Since the con gurational energy contains the term N 4J h ii there is a latent heat 4U = 2N j4J jmsT0 = N j ln!+ =!, jT0 ms T0 ; 9.123 where ms T0 is the spontaneous magnetization of the equivalent Ising model at T = T0 . When T0 Tc or 4J and ln!+ =!, have opposite signs or one of these quantities is non-zero but not the other then no transition occurs. Thus there is either a rst-order transition or no transition at all, except in the physically improbable case T0 = Tc. This theory is due to Bell and Fairbairn 1961b who were concerned with the low temperature second-order transition in solid ortho-hydrogen. Denoting the axial-component angular momentum quantum number by M, under certain assumptions the M = 1 states can be identi ed with i = +1 and the M = 0 state with i = ,1, giving an unsymmetrical Ising model with !+ =!, = 2. However, as we have seen, such a model does not display a second-order transition, and Bell and Fairbairn 1961b also showed that, with the parameters given appropriate values, T0 Tc. They concluded that the transition cannot be a rotational one, and in later work Bell and Fairbairn 1967 discussed sublattice ordering in the system. When the i = 1 and i = ,1 states represent occupation by atoms of components A and B with degeneracy factors !a and !b the analysis showing equivalence to the Ising model is similar to that of Sect. 6.2 but the eld He is now given by the relations 1 He = 2 4 ; 9.124 1 z" , " , T ln! =! : 4 = a , b , 2 aa bb a b For any T Tc, where Tc is the critical temperature for the equivalent Ising ferromagnet, there is a rst-order transition at the value of a , b where He = 0, the mole fractions in the conjugate phases being xa = 1 1 msT . 2 The coexistence curve in the xa ; T plane is thus una ected by the degeneracy factors, whose only e ect is to add on a temperature dependent term to a , b . For J = 0 the dilute Ising model of Sect. 6.6.1 becomes a system of this type since, with the two spin orientations possible on sites occupied by A, !a =!b = 2. In this case the meaneld result that the coexistence curve is symmetrical about the line xa = 1 is exact. Gri ths 1967b showed that 2 when J = 0 the dilute Ising model is equivalent to a standard Ising model in a temperature dependent eld.
9.10 A Competing Interaction Magnetic Model
9.10 A Competing Interaction Magnetic Model
271
Consider a decorated loose-packed lattice with the primary sites occupied 1 by standard Ising spins with s = 2 and the secondary sites by Ising spins of quantum number s1 , where s1 can take any integer or half-integer value. As well as the primary secondary interaction there is an antiferromagnetic primary primary interaction. We assume that, in zero eld, ^ h i ; j ; ij = ,J=q ij i + j + J i j ; 9.125 where J 0, 1 0 and ij = q; q , 2; : : : ; ,q with q = 2s1 and ^ i ; j = 1. The rst term in h is scaled so that its largest magnitude is 2J . Although the primary secondary interaction term in 9.125 is taken to be ferromagnetic, an interaction of either sign tends to align the spins on the primary sites i and j . The primary secondary interaction is thus in con ict with the antiferromagnetic primary primary interaction which tends to make the spins on the primary sites i and j antiparallel. From 9.125 the energy of a perfect ferromagnetic state with all primary site i = 1 and all secondary site ij = q or all i = ,1 and all 1 ij = ,q is , 2 zN 2 , J . It is also possible to have a state of perfect antiferromagnetic order on a loose-packed primary lattice in which all primary 1 nearest-neighbour pairs have i + j = 0 and the energy is , 2 zN J . Since we assume 1, the ground state attained at T = 0 is ferromagnetic. We thus expect ferromagnetic long-range order to be stable in a range 0; Tc, where Tc is a critical temperature. However all the q + 1 values of ij are equally probable when the spins on primary sites i and j are antiparallel, contributing a term kb lnq + 1 to the entropy. States with primary spins antiferromagnetically ordered have a high entropy. Ordered states with this unusual property can be destabilized by either an increase or a decrease in temperature but may be stable in a range Tc2 ; Tc1, where Tc1 and Tc2 are respectively termed the upper and lower critical temperatures. For appropriate parameter values, this situation exists for the present model with the antiferromagnetic lower critical temperature Tc2 Tc1 where Tc is the critical temperature for ferromagnetic order. Equations 9.3 and 9.125 give 1; 1 = exp, G
q X `=0
exp q , 2`2G=q ; 9.126
+1; ,1 = q + 1 exp G ; where G = J=T . Since +1 = ,1 = 0 in zero eld, 9.9 yields
= exp, G sinh 2q + 1G=q ; sinh2G=q
272
2 exp2K e = exp,2 G sinh 1 q + 1G=q ; q + sinh2G=q 9.127 Le = 0 for the equivalent Ising model on the primary lattice. From 9.127, exp2K e ' exp 21 , G ; as G ! 1 . 9.128 q+1 Hence K e ! 1 as T ! 0, in accordance with the previous deduction of a ferromagnetic ground state. However, K e does not increase monotonically with G since it follows from 9.127 that for small G high T , q + 2 G2 + OG4 : Ke = , G + 9.129 3q For small G, K e 0 but, for large G, K e 0. To proceed further we need dK e = , + q + 1 coth 2q + 1G=q , 1 coth2G=q ; 9.130 dG q q
2 2 q+1 d2 K e = , 2 q sinh 2q + 1G=q : 9.131 2 dG q sinh2G=q 2 From the power-series expansion of the sinh function, sinh 2q + 1G=q q + 1 sinh 2G=q ; 9.132 2 K e =dG2 0. It follows that K e decreases from 0 as for all G 0 so that d G increases from 0 to reach a unique minimum K e = K min 0 at G = Gmin T = Tmin. Then K e increases monotonically, passing through 0 at G = G0 T = T0 and through K c the critical value for the primary lattice Ising ferromagnet at G = Gc T = Tc T0 to approach in nity with G as G ! 1. From the above it follows that K e K c for the temperature interval 0; Tc giving long-range ferromagnetic order when T Tc . If K min ,K c then K e = ,K c at temperatures Tc1 and Tc2 , with T0 Tc2 Tmin Tc1 . Thus K e ,K c in the interval Tc2 ; Tc1 . Since we have speci ed a loose-packed primary lattice, there is long-range antiferromagnetic order when Tc2 T Tc1 . It follows that Tc1 and Tc2 are respectively upper and lower critical points. The value of K min, which depends on q and and which is given by the relation dK e=dG = 0, is crucial. It can be shown from 9.130 that, with a xed value of q, K min has a greatest lower bound equal to , 1 lnq + 1 2 for the range 0 1. This bound is approached as ! 1 , 0. Hence antiferromagnetic critical points occur over some range of if 1 lnq + 1 2 K c , where K c , of course, depends only on the primary lattice. With q = 1 this condition is satis ed for the d = 3 simple cubic and body-centred cubic lattices. With q = 2 it is also satis ed for the square and diamond lattices, and with q = 3 for the honeycomb lattice as well see Table 8.1. As an example,
9.11 Decorated Lattice Mixtures of Orientable Molecules
273
with a square primary lattice, q = 3 and = 0:8 we have Tc1 = 1:0150J , Tc2 = 0:4387J , T0 = 0:1443J and Tc = 0:0882J . For the general case, since dK e=dG 6= 0 when K e = K c, 4G 4Ke and hence the heat capacity singularity at each critical point is of similar type to that at the critical point of the primary lattice Ising model. The q = 1 form of the present model was investigated by Nakono 1968 see also Example 9.3. Syozi 1972 discusses related models. The temperature T = T0 , where K e changes sign, is of some interest. Since K e = 0 at T = T0 the primary lattice spins are completely disordered, as they are at T = 1. Suppose that h 0 ri is the correlation between primary lattice spins whose relative position is represented by r. As the temperature increases through T0 , K e changes from positive to negative and h 0 r i changes from a positive quantity which decreases with jrj to one which alternates in sign, though its magnitude still decreases with jrj. At T = T0 itself h 0 r i = 0 for all r. For the d = 1 case of the present model, from 2.88, 8tanh K ` ; T T0 , e h i ; i+` i = : 9.133 ,1`tanh jK e j` ; T T0 . The point T = T0 is an example of a disorder point, other instances of which occur in the triangular Ising model with unequal interactions. Early work on disorder points is discussed by Stephenson 1970, who cites the d = 1, q = 1 case of the present model as one example see also Georges et al. 1986. As in the present model, disorder points in other cases are due to competing interactions which promote di erent kinds of short-range order. Disorder points in the eight-vertex model are discussed brie y in Volume 2, Chap. 5.

In Sects. 9.2 9.10 we were mainly concerned with magnetic systems. In the remainder of the chapter we shall consider non-magnetic mixtures and lattice gases.8 The present section is concerned with mixtures which have closedloop coexistence curves in the composition temperature plane and thus both upper and lower critical solution temperatures. Phenomena of this type occur in real mixtures, for example glycerol-ethylbenzylamine and water-nicotine. A decorated lattice is assumed with each primary and each secondary site occupied either by a molecule of component A or one of component B. A restricted grand distribution is used with the Hamiltonian given by 6.2. There is an interaction energy "aa for each primary secondary pair of A molecules with "ab and "bb being similarly de ned. The existence of a lower
8
For a review of decorated models of such systems, see Wheeler 1977.
274
critical point is likely to be an entropic phenomenon, as we saw in Sect. 9.10. To achieve the necessary high entropy it is postulated that each A or B molecule on a secondary site has ! distinguishable orientation states. In two of these states an active arm is directed towards the molecule on one of the adjacent primary sites and terms aa , ab and bb are then added to "aa , "ab and "bb respectively. In the other ! , 2 states the secondary site molecule is said to be disoriented. The orientations of the molecules on the primary sites are taken as xed so that these sites have only two distinguishable states, occupation by A or B respectively, which can be represented by = +1 and = ,1. The system is thus equivalent to an Ising model on the primary lattice. Symmetry between the A and B components can be obtained by putting9 "aa = "bb = 0 ; "ab = "; 9.134 = bb = ,1 "; = ,2 ": aa ab Summing over the secondary site states +1; +1 = ! , 2 + 2 exp1 Q + exp,2Q ! , 2 + 2 exp2 Q ; ,1; ,1 = exp,2Q ! , 2 + 2 exp2 Q + ! , 2 + 2 exp1 Q ; +1; ,1 = + 1 exp,Q ! , 2 + exp1 Q + exp2 Q ; where Q = "=T; = exp a , b =T : 9.136 To evaluate the partition function we also need +1 = ,a , b ; ,1 = 0 : 9.137 For phase equilibrium it is necessary to transform to the zeroeld state of the equivalent Ising model by putting a = b which gives = 1 and, from 9.135, +1; +1 = ,1; ,1. Then substitution in 9.9 gives Q + exp2K e = ! , 2 cosh, 2+ exp 1 Q1Q + exp 2 , 1Q ; ! + exp1 + exp2 Q 9.138 Le = 0 : An expansion in powers of Q yields the relation ! , 22 , 1 + OQ3 ; 9.139 K e = Q2 4! 9 If "aa = "bb = 0 it is found that results depend only on the di erence "ab , "aa , 6 so that no generality is lost by taking "aa = "bb = 0. 9.135
275
9.141 the spontaneous magnetization of the primary lattice Ising T Tc2, mi:m: 0 and using the + and , signs in 9.141 s gives the compositions of A-rich and B-rich conjugate phases. Outside this range of T , mi:m: = 0 and 9.141 gives xa = 1 . Thus in the xa ; T plane s 2 there is a closed-loop coexistence curve symmetrical about the line xa = 1 2 with upper and lower critical temperatures for phase separation at T = Tc1 and T = Tc2 respectively. There is a disorder point, with K e = 0, at some temperature T0 Tc2 see Fig. 9.9. For a loose-packed primary lattice there is a transition to a sublattice ordered state when K e passes through the value ,K c . The 1 = 0 case of the above theory was developed by Wheeler 1975, 1977 and Anderson and Wheeler 1978a tted theoretical results to the where mi:m: is s model. For Tc1
which is useful at high temperatures. We now assume " 0 ; 1 0 ; 2 0 ; 2 1 + 1 : 9.140 The lowest energy for a primary secondary primary triad of sites is ,2 , 1", attained for occupational states AAB or ABB with the active arm of the secondary site molecule directed towards the unlike adjacent primary site molecule. Thus, on the primary lattice, AB nearest-neighbour pairs are favoured at low temperatures, corresponding to antiferromagnetic local order in the equivalent Ising model. For a loose-packed primary lattice the ground state is one of long-range sublattice order with As occupying one sublattice 1 and Bs the other. The ground-state energy is , 2 zN 2 , 1", where N is the number of primary lattice sites and z the primary lattice coordination number. However for a primary secondary primary triad of sites there are also occupation states AAA or BBB, with the molecule on the secondary site disoriented, which have zero energy but an entropy kb ln!,2. Thus for high enough ! and T , AA or BB nearest-neighbour pairs on the primary lattices will be favoured and separation into A-rich and B-rich phases will occur, corresponding to ferromagnetic order in the equivalent Ising model. This is borne out by the behaviour of K e. With the conditions 9.140, K e ! ,1 as T ! 0 Q ! 1. However, from 9.139, K e 0 at high temperatures if ! 22 , 1 . The T; Ke plot then has the form shown in Fig. 9.9. For su ciently large !, the maximum value of K e K c, the critical Ising parameter value for the primary lattice, as shown in Fig. 9.9. The values of T at which K e = K c are denoted by Tc1 and Tc2 Tc1 Tc2 . By di erentiating ln Z with respect to and then putting = 1, which implies Le = 0, the mole fractions xa in the conjugate phases are given by i:m: xa = 1 ms+ 2 2 z z ! , 2 sinh 1 + 2 ! , 2 cosh Q + exp Q1 + 1 , exp Q2 , 1 ; Q + exp Q + 1 + exp Q , 1
1 2
276
Ke
Kc
T0
Tc2
Tc1
Fig. 9.9. Orientable molecule mixture with the inverse reduced temperature for
the equivalent Ising model plotted against temperature. Tc1 and Tc2 are the upper and lower critical temperatures and T0 is the disorder point.
experimental closed-loop curves for some organic mixtures, for example glycerol-ethylbenzylamine. They required the very large value ! = 5000 for the number of secondary site orientation states. Although the model is highly schematic, this value of ! may re ect the large number of con gurations available to fairly complex organic molecules. Bell and Fairbairn 1967 had earlier obtained a closed-loop coexistence curve from a regular lattice model with orientation dependent interaction energies. This is not equivalent to an Ising model, and a rst-order approximation was used. Closed-loop coexistence curves are observed experimentally for some aqueous mixtures in, for example, water-nicotine but are highly unsymmetrical with respect to the two components. Anderson and Wheeler 1978b tted theoretical results obtained by taking "aa 6= "bb and aa 6= bb . Some interesting new phenomena appear when the symmetrical model is used without the last condition of 9.140 Wheeler and Anderson 1980. For 1 + 1 2 but with the rst three conditions of 9.140 still satis ed, the lowest energy for a primary secondary primary triad of sites is attained for occupations AAA or BBB with the active arm of the secondary site molecule directed to one of the adjacent primary site molecules. The ground state thus consists of pure A and pure B phases, the energy being , 1 zN1 ". First, 2 let 1 + 1 = 2 . Then, from 9.138, K e ! 0 as T ! 0 and the T; K e

Ke Kc
277
c a b
Tc3
Td
Tc2
Tc1
Fig. 9.10. As Fig. 9.9 but with curve b giving an additional upper critical temperature Tc3 and curve c a critical double-point temperature Td .
plot assumes the form sketched as curve a in Fig. 9.10. Then suppose that 1 + 1 is slightly greater than 2 . The T; Ke plot is sketched as curve b in Fig. 9.10 and remains close to curve a except for low values of T=" where now K e ! 1. There is thus a low temperature interval 0; Tc3 where Ke Kc, so that phase separation occurs. The coexistence diagram broken curves is sketched in Fig. 9.11. There is a closed-loop coexistence curve with upper critical temperature Tc1 and lower critical temperature Tc2. Separated from the closed-loop curve by the temperature interval Tc3; Tc2 is a low temperature coexistence curve with upper critical temperature Tc3.
Tc1 T
Tc2 Td Tc3
0:5
Fig. 9.11. Coexistence curves in the

xa
1:0
mole fraction temperature plane; the full and broken curves correspond respectively to cases c and b of Fig. 9.10.
278
For both intervals Tc2 ; Tc1 and 0; Tc3 there is separation into A-rich and B-rich phases, but for the upper temperature interval the phase separated state is entropically stabilized whereas for the lower temperature interval it is energetically stabilized. Now suppose that 1 + 1 is increased again relative to 2 , raising the minimum of the T; K e plot till it occurs at K e = K c , as shown in curve c of Fig. 9.10. Denote the temperature of the minimum at K e = K c by Td . The lower critical point of the upper branch of the coexistence curve and the upper critical point of the lower branch have coalesced into a single point at T = Td . This is a critical double-point. In the full-line sketch in Fig. 9.11 the two branches of the coexistence curve are shown as meeting with a common tangent at the critical double-point. This is justi ed by the analysis below. Let 4xa denote the di erence between the values of xa in the conjugate phases. By 9.141, 4xa mi:m: as 4Ke = K e , K c ! +0 and mi:m: s s 4K e where is the Ising model critical exponent. Now K e has a minimum at K e = K c , T = Td so that 4Ke T , Td 2 . This applies for both positive and negative T , Td , and hence on both sides of the critical double-point we have 4xa jT , Tdj d where d = 2 . This result was derived for a somewhat di erent decorated solution model see Sect. 9.13 by Bartis and 1 for both d = 2 and d = 3, Hall 1974. Since 1 and the two d 2 coexistence branches meet with a common tangent parallel to the xa -axis.10 If 1 +1 is again increased relative to 2 the minimum K e moves up to a value greater than K c. The two branches of the coexistence curve are joined to form a rather bizarre `neck' shape and there is now only one critical temperature.
9.12 The Decorated Lattice Gas

9.12.1 General Properties
We now treat liquid vapour phase phenomena by using a decorated lattice gas model, which is more exible in its properties than the simple lattice gas model of Sects. 5.3 and 5.4. In the one-component model of this section each primary site or secondary site is either vacant or occupied by a molecule. The grand distribution of Sect. 2.3.3 will be used. Let the occupation states , and for a primary secondary 1 primary triad of sites have energies , 2 ", ,w and ," , w + w1 respectively with all other occupation states having zero energy. For additivity of pair energies, w1 = 0 but it is useful to include the non-additive case. Representing the occupied and vacant states for primary sites by i = +1 and i = ,1 respectively and using the grand distribution
10
If it were true that 1 , the coexistence curves would form a double cusp at 2 their meeting point.
279
+1; +1 = Z1 + X2 Z1 ; ,1; ,1 = 1 + ; +1; ,1 = 1 + X ; 9.142
+1 = , ; ,1 = 0 ; where is the chemical potential and = exp=T ; X = exp "=2T ;
Z = expw=T ; Z1 = exp,w1 =T : Then, from 9.9, 2 1 exp4K e = Z1 +1X ZX1 + ; + 2
exp2Le =
9.143
: 1+ The density equal to 1 when all primary and secondary sites are occupied is given by @ ln Z ; 9.145 = 1 + 1 zN @ T 2 where Z is the partition function for the decorated model, given by 9.7 and N the number of primary lattice sites. Using 9.142 and 9.144 in 9.145
1 1 1 X2Z1 + 2X 1 + 2 z = 2 + 8 z 1 + X2 Z + 1 + 1 + X 1 1 zg K ; L @ K e + mi:m: K ; L @ Le ; +2 p e e e e @ T @ T 9.146 gp K e ; Le and mi:m: K e; Le being given by 9.10 and 9.11 respectively, and where K 1 2Z @@e = 4 1 +X X21Z + 1 + , 1 2XX ; 1 + 9.147 @ Le T 1 1 X2 Z 1 , @ = 2 + 4 z 1 + X2 Z 1 + : 1 T The rst-order transition lines in the T; plane are the parts of the Le = 0 locus lying in the region where K e K c, that is lying to one side of the
Z1 + X2Z1 z=2
9.144
280
Ke
Let 4T = Tc ,T and 4Ke = K e ,K c . Since mi:m: 4Ke as K e ! K c +0 s and since K e and Le are analytic functions of and T it follows that as T ! Tc , 0 . 9.149 ` , g 4T ; This is similar to the behaviour of the simple lattice gas described in Chap. 5, but there are important di erences both globally and in the neighbourhood of the critical points. The hole particle symmetry of the simple lattice gas is lost except for a special case to be discussed in Sect. 9.12.3 and the coexistence curves in the ; T plane are no longer symmetrical about = 1 . 2 The coexistence diameter 1 9.150 d = 2 ` + g 1 for the simple lattice gas, but for the decorated is, from 5.37, equal to 2 model, from 9.146,
1 1 1 X2Z1 1 + 2 z d = 2 + 8 z 1 + X2 Z + 1 + + 1 2XX + @1K + 1 zgpK e; 0 @e : 9.151 2 T The empirical law of the rectilinear diameter see Sect. 3.4 implies that d d =dT remains nite as T ! Tc , 0. This law is satis ed by the simple lattice gas, since d d =dT = 0 for all T Tc. However, for the decorated model, the right-hand side of 9.151 is not a linear function of T near the critical point. Unless @ K e=@ = 0, the derivative d d =dT contains a term 0 proportional to gpK e de ned by 9.27. Now 8 , 0 0 : 1 0 K e 4K c ; with = 8 for d = 3 , 9.152 gp :, ln j4Kej ; for d = 2 .
= K c locus which is always a single connected curve. When, as will be assumed in the present section, the Le = 0 locus is also a single connected curve, the part of it in the K e K c region forms a vapour liquid transition line. The conjugate phase densities are given by taking mi:m: = +mi:m: and s mi:m: = ,mi:m: , mi:m: K e being the spontaneous magnetization of the s s equivalent Ising model. The value of is that which satis es the relation Le = 0 at the given temperature T while mi:m:K e and gp K e ; 0 are calculated s for the zeroeld Ising model at the corresponding value of K e . The critical temperature Tc is de ned by the intersection of the Le = 0 and K e = K c curves. Let the densities of the conjugate liquid and vapour phases are denoted by ` and g respectively. Then, by 9.146, the coexistence width 2 @ Le : 9.148 ` , g = mi:m: K e s 1 1 + 2 z @ T
281
, 0 0 : 1 d d 4T ; with = 8 for d = 3 , 9.153 dT :, ln j4T j ; for d = 2 . In both cases, d d=dT ! 1 as T ! Tc , 0. Mermin 1971a, 1971b see also Wheeler 1977 derived similar results for two types of lattice model, the rst being the `bar' model and the second the w = w1 = 0 case of the decorated model discussed here. This indicates that a non-singular diameter is an artifact of the simple lattice uid. As pointed out in Sect. 3.4, more general arguments for the existence of the singularity have been developed. However, experimentally the situation is still not completely resolved Rowlinson and Swinton 1982.
Hence
9.12.2 A Water-Like Model

Bell and Sallouta 1975 developed a decorated model for water by taking the open, completely hydrogen-bonded structure as one in which all primary sites are occupied and all secondary sites empty. The interaction energy ,w w 0 of a primary primary pair of molecules with the intervening secondary site empty is attributed to the formation of an hydrogen bond whereas 1 the primary secondary pair energy , 2 " " 0 is due to the non-bonding interaction. It is assumed that no hydrogen bond can form between an nearestneighbour primary pair of molecules with the intervening secondary site occupied so that w1 = w. The close-packed state is one in which all primary and all secondary sites are occupied. This model is less realistic structurally then the water model treated in Sect. 7.5 and less satisfactory for uid phases in that the open structure is imposed ab initio rather than arising as short-range order due to bonding between appropriately placed and oriented molecules. However, it does have the advantage that, because of the equivalence to the Ising model, some accurate results can be obtained for comparison with those derived from standard approximations. The decorated model can be looked on as one of the class of interstitial models see, for instance, Perram 1971 in which the open structure is disordered by displacement of molecules from `framework' onto interstitial sites. The results of Sects. 5.5 and 7.5 indicate that water-like properties are associated with the existence of a pressure range in which the open structure is stable at T = 0. In the present model the energies of perfect open and 1 1 close-packed structures are respectively , 2 zNw and , 2 zN" or, in terms 1 zMw and , 1 zM"=1 + 1 z. Denoting the of the number of molecules M , 2 2 e 2 volume area for d = 2 per primary site by v0 , V = V =v0 and the reduced volume per molecule is v = V=M . For the open and close-packed states v = 1 1 e and v = 1=1 + 2 z respectively. The pressure eld P = Pv0 is taken as an independent variable and from 1.84,
282
Hence, in the open and close-packed states o = P , 1 zw ; 2
f = = u , Ts + Pv = h , Ts ; 9.154 where u and h are respectively the energy and enthalpy per molecule. For given P at T = 0 the equilibrium state is thus that of least = h = u + Pv. P , 1 z" 2 : 1 + 1z 2
9.155
cp =
An open close-packed transition occurs when o = cp . From 9.155, the pressure and chemical potential at the transition are given by
1 P0 = 1 + 2 z w , " ; 9.156 0 = w , " : There is an interval 0 P P0 of stability for the open structure if P0 0, which is equivalent to
1 1 + 2z w ": 9.157 If 9.157 applies there is a vapour open transition at P = 0 when T = 0 and = o = , 1 zw. The inequality 9.157 is in fact equivalent to the 2 condition , 1 zw 0 . It can be shown from the second relation of 9.144 2 with Z1 = 1=Z that the Le = 0 locus in the T; plane then meets the 1 T = 0 axis at = , 2 zw. The open close-packed transition point = 0 on the T = 0 axis is an isolated singularity. It cannot be the terminus of any branch of the Le = 0 locus because at low temperatures the primary lattice is nearly full in both the open and close-packed states and so both correspond to Le 0. Coexistence curves in the ; T plane can be obtained from 9.146. Curve 1 A in Fig. 9.12 is for the diamond lattice d = 3, z = 4, 1 + 2 z = 3 with i:m: K and g K ; 0. The " = 2:8w, accurate series results being used for ms e p e terminal points = 0 and = 1 correspond to the vapour and open structure 3 phases respectively. The most striking feature, apart from the marked lack 1 of symmetry about = 2 , is the pronounced maximum on the liquid density branch. Curves B and C respectively, given for comparison, show the results of using rst-order and meaneld formulae for mi:m: K e and gpK e; 0. s Unfortunately the calculation of ; T isobaric curves requires Ising model results away from the H = 0 axis. However, near the maximum on the liquid branch of the coexistence curve, the rst-order plot is quite close to the accurate one. First-order calculations show maxima on the adjacent isobaric ; T curves. By considering perturbations from the perfect open structure, Bell and Sallouta 1975 showed that near T = 0 the density increases with T on isobars for 0 P P0 , provided that " w. This implies a density
9.12 The Decorated Lattice Gas 1.5 3 1.0 A B C
283
Fig. 9.12. Coexistence
curves in the temperature density plane for the water model. Reprinted from Bell and Sallouta 1975, by permission of the publisher Taylor and Francis Ltd.
0.5
0.0 0.0 0.2 0.4 0.6 0.8 1.0 T=w 1.2
maximum on any continuous isobar in this pressure range. Since P0 is considerably greater than the critical pressure Pc , density anomalies occur well into the supercritical region, in accordance with experimental results. Compressibility minima are also found. There is thus good qualitative agreement between the results for this model and those for the model of Sect. 7.5, where rst-order approximation methods had to be used exclusively.
9.12.3 Maxithermal, Critical Double and Cuspoidal Points

When the K e = K c and Le = 0 loci in the T; plane of the decorated model are both single curves and have a unique intersection, the phase diagram in this plane is topologically similar to that of a simple lattice gas, though as we have shown in Sect. 9.12.1 there are important di erences in properties. However, for large negative values of " the mapping from the T; plane of the decorated model to the K ,1 ; Le plane of the Ising model is no longer e one-to-one and the Le = 0 locus has several distinct branches. The resulting phenomena are complicated and best approached by considering the transitions on the -axis. For the general case, 9.155 and 9.156 are still valid provided " is replaced by " + w , w1 . However we also have to consider an
9. Applications of Transform Methods Table 9.1. The values of P and at possible transitions between the vapour V, open O, close-packed CP and intermediate states I. Transition P VO 0 , 1 zw 2 +w w V CP 0 , 1 z " 1 + 1, w 1 2 2z 1 zw , " + w1 O CP w1 , " 22 zw zw OI z,2 z,2 1 z2 w1 , " , w 1 zw1 , " , w I CP 4 2
284
ìntermediate' state in which all secondary sites are occupied and all primary sites are empty. The intermediate state energy is zero and v = 2=z so that hi = i = 2P=z. With this relation and the modi ed 9.155, Table 9.1, showing the values of P and at the possible transitions, is constructed. It will be assumed that w 0 and w w1 . These conditions can be shown to imply that K e 0 for all and T . The condition 9.157 for a range of open state stability and a vapour open state transition at T = 0 now becomes 1 zw " , w ; 9.158 1 2 If 9.158 is not satis ed then the close-packed state is stable at T = 0 for all P 0. The vapour close-packed transition point is the only one on the -axis and is the terminus of the Le = 0 locus. We turn from this rather uninteresting case to ask whether a range of stability is possible for the intermediate state at T = 0. Such a range exists if the value of or P at the intermediate close-packed transition point is greater than the value of or P at the open intermediate transition point. From Table 9.1 this is true if w w1 , " zz, 2 : 9.159
Since w 0, the condition 9.159 implies 9.158. So when 9.159 is satis ed 1 there are three transition points on the -axis, vapour open at = , 2 zw, open intermediate at = zw=z , 2 and intermediate close-packed at = 1 zw1 , " , w. These points are denoted by Q, R and S respectively in 2 Fig. 9.13. Now, at low temperatures, the primary lattice is nearly empty in the vapour and intermediate states so that Le 0. The points Q, R and S all correspond to changes in the sign of Le and it can be con rmed from 9.144 that, when 9.159 applies, Q, R and S are termini of the Le = 0 locus. Thus the Le = 0 locus has more than one branch.
285
+ 1 " Close-packed 2
S Intermediate
Ke = Kc
Le
Le = 0
Fe 0
R Fe Open Q Vapour
0
A
Le
Fig. 9.13. The symmetric case of the decorated lattice gas showing the loci of
= 0 and K e = K c . The broken curves are the two rst-order transition lines which do not lie on the T -axis.
Le
11
We now consider the case w1 = 0 when the pair energies are additive.11 De ning Y = X = exp "=2T = exp 2 + "=2T ; 9.160 the rst relation of 9.144 with Z1 = 1 becomes ,1 ,1 + 9.161 exp4K e = ZX YY + XY + 1 X : + 1Y,1 For given T , K e is unaltered when Y is replaced by Y,1 . The K e = K c locus in the T; plane is given by the relation X + ,1 exp4K c , w=T , 1 = Y + X ,1 , 2 : 9.162 Y +2 This curve i is symmetrical about the line = , 1 " Y = 1, 2 ii has the line T = w=4K c as asymptote when ! 1, iii lies entirely to the right of this asymptote, iv has its point of maximum T when = , 1 ". 2
This has been treated extensively by Mulholland and Rehr 1974.
286
The general form is shown by the K e = K c curve in Fig. 9.13. The Le = 0 curve has no similar symmetry for arbitrary "=w, but for the particular case z 9.163 ," = z2,w2 the second relation of 9.144 can be written as z=2 exp2Le = Y 1X+ YX ; 9.164 +Y giving Le T; Y = ,Le T; Y,1 : 9.165 The locus Le = 0, and thus the whole phase diagram in the ; T plane, are 1 symmetrical about the line = , 2 ". This is the case shown in Fig. 9.13. From i:m: K e ; Le and gp K e ; Le are respectively odd the above, the Ising model m and even functions of + 1 " at given T . It can then be shown from 9.146 2 that T; Y + T; Y,1 = 1 : 9.166 Thus there is particle hole symmetry and the phase pattern in the ; T plane is symmetrical about = 1 . 2 1 When " is given by 9.163 the entire line = , 2 " Y = 1 is part of the Le = 0 locus in the T; plane, ending at R. As 9.159 is satis ed, Q and S are also termini of the Le = 0 locus. It can be shown that two other solutions of Le = 0 separate from the solution Y = 1 at X = z , 2=z + 2, which is the point P shown in Fig. 9.13. When K e is calculated at P it proves to be less than K c for all common lattices. Hence, as shown by the broken lines in Fig. 9.13, there are three distinct rst-order transition curves, RB on the line 1 = , 2 " with QA and SC placed symmetrically on either side. Since P = 0 at Q, QA can be regarded as the vapour liquid transition line. By symmetry, 1 + 2 = 1 on RB, where 1 and 2 are the densities of conjugate phases, 1 and hence d = 2 and there is no singularity in the diameter. This can be attributed to the fact that, for w1 = 0, @ K e =@ = 0 when Y = 1. However, @ K e=@ 6= 0 on the lines QA and SC so that in both cases there is a singular diameter. Though the symmetrical case is exceptional, it serves as a useful guide to the situation when " 6= ,2zw=z , 2. It can be shown that the curve of exp2Le against Y is monotonically increasing for high temperatures in all cases but that as T is lowered, an interval of negative slope develops for Y 1 when ," 2zw=z , 2 and for Y 1 when ," 2zw=z , 2. Thus for ," 2zw=z , 2 the Le = 0 locus consists of a loop from Q to R and a single branch going from S to T = 1. At ," = 2zw=z , 2 we have the intermediate stage shown in Fig. 9.13 whereas for 2zw=z , 2 ," zw=z , 2 there is a closed loop from R to S and a single branch going from Q to T = 1. As ," ! zw=z , 2 + 0 the interval RS of intermediate state stability on the

R e
287
d c b a
Fig. 9.14. QMR is the closed-loop branch of the Le = 0 locus. The curves a
decreases.
e show the relative positions of the lower part of the K e = K c locus as j"j=w
-axis shrinks to zero and with it the RS loop of the Le = 0 locus. For the
simple cubic, body-centred cubic and face-centred cubic lattices, the loop QR of the Le = 0 locus lies entirely in the K e K c region when " = ,1 see Example 9.6. The position of part of the lower half of the K e = K c locus relative to the loop QR at successive stages as ," decreases from in nity towards 2zw=z , 2 is indicated by curves a e in Fig. 9.14. For simplicity, the loop QR is shown as xed. The point M on the loop where dT=d = 0 is a maxithermal point. Curve a represents the case " = ,1. At the next stage b, the two curves touch at D which is a critical double-point. The coexistence curves in the ; T plane in the neighbourhood of D and M at stage b are sketched in Fig. 9.15. At stage c the critical double-point is replaced by two critical points separated by a homogeneous region, and at d the K e = K c curve intersects the loop QR at M, which becomes a cuspoidal point. The ; T plane con guration here is sketched in Fig. 9.15. By stage e the K e = K c locus cuts the loop QR in two ordinary critical points. Except at stage a there is another rst-order transition line, not shown in Fig. 9.14, which starts at S on the -axis and ends at an ordinary critical point on the K e = K c locus. A number of accurate coexistence curves and other diagrams
288
Fig. 9.15. Coexistence curves in the density temperature plane. The left-hand
diagram displays a maxithermal point M and a critical double-point D and the right-hand diagram shows a cuspoidal point M.
are given by Mulholland and Rehr 1974. These authors also discuss the phase diagram when w 0 for the additive case w1 = 0. The water-like model treated in Sect. 9.12.2 was an example of the nonadditive case, with w1 = w. The assumption that the Le = 0 locus for this model is a single connected curve is con rmed from the condition 9.159, which is not satis ed when w1 = w and " 0. The general non-additive model can be discussed only brie y. One important di erence from the additive model is that the K e = K c locus in the T; plane is no longer symmetrical about the line = , 1 ". However, for w1 w the entire locus lies to the right 1 2 of the line T = 4 w , w1 =K c , which forms the ! 1 asymptote. When w1 = w the locus bends round to terminate on the -axis. For w1 w there are regions where K e 0 and the properties of the decorated model depend on those of the Ising antiferromagnet. The present model can be transcribed to a decorated mixture by using a restricted grand, rather than a grand, distribution and substituting a , b for . The molecules of the present model are labelled A and the holes are replaced by molecules of another component B so that we can set = x where x denotes the mole fraction of A. In this transcription the maxithermal point in Fig. 9.15 becomes an azeotropic point whereas the cuspoidal point becomes a critical azeotropic point. Denote the values of x and T at such a point by xc and Tc, and let x1 and x2 be the mole fractions in the coexisting phases when T is just less than Tc. Then x1 , xc and x2 , xc have the same sign, contrasting with the opposite signs found near an ordinary critical point.12
12
A similar remark can be made about density di erences near a cuspoidal point.
9.13 Decorated Lattice Gas Mixtures
289
If the two-dimensional x; T or a , b ; T space is extended into three dimensions by adding an "=w axis, the critical azeotropic point occurs where a line of azeotropic points ends on a line of critical points. In spaces of three intensive thermodynamic variables this phenomenon occurs experimentally at high pressures Rowlinson and Swinton 1982, Sect. 6.2 and theoretically in the decorated lattice gas mixtures described in the next section.
9.13 Decorated Lattice Gas Mixtures

We now consider brie y two component decorated lattice gas models, taking the term `lattice gas' to imply that both primary and secondary sites can be vacant. Thus for condition i of Sect. 9.1 to apply, molecules of only one of the components can occupy primary sites. This component is termed the solvent and the other component the solute. There are then two possibilities: 1 Only solute molecules can occupy secondary sites, so that the pure solvent is a simple lattice gas as de ned in Sect. 5.3. 2 Both solute and solvent molecules can occupy secondary sites. A solute component could be added to any of the lattice gas models of Sect. 9.12 to produce a system of type 2. From condition ii of Sect. 9.1 there is no direct solute-solute interaction so that there can be no phase separation in pure solute in either type 1 or type 2 models. Such models are, therefore, applicable only when the critical temperature of the solute is much lower than that of the solvent or when the solute mole fraction is fairly small. Work on models of type 1 was initiated by Widom 1967; for further developments, see Wheeler 1977. They can display a number of cooperative phenomena, including maximum critical solution temperatures and critical azeotropy. Bartis and Hall 1974 obtained critical double-points in the pressure mole fraction plane from a model of type 2. They needed to use a solvent-solvent interaction which is attractive for primary secondary pairs and repulsive for primary primary pairs. This implies a potential which is negative at small separations and positive at larger ones. Sallouta and Bell 197613 modelled aqueous solutions with a system of type 2, a solute being added to the decorated lattice gas water model discussed in Sect. 9.12.2. They found that with repulsive solute-water interaction the water structure breaks down at small solute concentrations. With attractive solute-water interaction the water structure is stable for quite high solute concentrations, and large attractive interactions give rise to azeotropy and critical azeotropy. The decoration transformation has given exactly solvable models displaying a variety of critical phenomena. The two models solved by the star-triangle transformation Sect. 9.8 and Example 9.4 yield results similar to those
13
There is an error in the table on p. 341 of this paper. For conformity with the rest of the paper, ,u should be replaced by u and the statement u 0 deleted.
290
of the corresponding decorated models Sect. 9.5 and Example 9.1 respectively. However there are limits to the versatility of the decoration method. It has so far failed to produce any exact theories of phenomena dependent on three-phase equilibrium, such as triple points, tricritical points or critical end-points. This situation is discussed by Wheeler 1977 who attributes it to the fact that a decorated model must transcribe to a zeroeld standard Ising model for exact results to be obtained. However this underlying Ising model is limited to two phase equilibrium. Although the ferrimagnetic models of Sect. 9.5 and 9.8 give, for appropriate parameter values, compensation points where four phases are in equilibrium, these phases become identical in pairs as T ! Tcomp 0. This is because the H = +0 phase for T = Tcomp + 0 becomes the H = ,0 phase for T = Tcomp , 0 and vice versa. Each identical pair corresponds to a single phase in the underlying Ising model, so supporting Wheeler's argument.
Examples
9.1 The Syozi dilution model is a mixture of Ising spins A and magnetically inert atoms B on a decorated lattice. All primary sites are occupied by As but a secondary site can be occupied by either an A or a B. There are primary secondary ferromagnetic interactions, with the constant denoted by J , between A atoms. Using a restricted grand partition function, show that, in zero eld, cosh2 exp2K e = 1 + 21 + 2 G ; Le = 0 ; where = exp a , b =T , G = J=T . Hence show that the fraction x of secondary sites occupied by A is given by x = 1 + cosh2GG 1 + gp K e ; 0 + 1 + 2 1 , gpK e ; 0 ; cosh2 where gp K e ; Le is de ned by 9.10. Show that the critical curve in the T; x plane is c x = 2exp2KG , 1 cosh2 , 1 exp,2K c cosh2G1 + gpK c ; 0 + 1 , gpK c ; 0 and deduce that the critical value of x is 1 xc = 2 1 , exp,2Kc 1 + gp K c ; 0 :
Examples
291
Show, using 8.13 and 8.17, that, if x denotes the critical value c of xc for the dual primary lattice, then xc + x = 1 and deduce that c 1 xc = 2 for the square lattice. 9.2 For the decorated ferrimagnet of Sect. 9.5, show that n o
1 @m r + 2 z T @H = 1 z 1 , 1 1 + gpK c; 0 tanh2 2G 2 H=0 2 @m i:m: h 1 i2 + @L r , 2 z tanh2G ; e Lc =0 where m and gpK c ; Le are respectively de ned by 9.77 and 9.10. Consider a d = 1 lattice with alternate sites labelled a and b. All sites are occupied by Ising spins and there is an interaction term Jab i j , Jab 0, for each nearest-neighbour ab pair. The spins on a and b sites have magnetic moments rm0 and m0 respectively. This model can be regarded as a decorated ferrimagnet with the b sites decorating the segments connecting the a sites and Jaa = 0. By using the formula deduced above and results from Chap. 2, show that @m 2 G r 2 = r , 1 exp22r+ 1+ 1 exp,2G ; @ H H=0 + T where G = Jab =T . Verify that this reduces to equation 4.5 when r = 1. Show that @ H=@mH=0 plotted against T has a horizontal tangent at T = 0. Show that the paramagnetic Curie temperature is ,4rJab =r2 + 1. 9.3 For the q = 1 case of the competing interaction model of Sect. 9.10, show that 1 K min = , 4 1 + ln1 + + 1 , ln1 , : Show that in the range 0 1, K min decreases as increases and 1 that K min ! , 2 ln2 as ! 1 , 0. 9.4 Using the notation of question 1, suppose that the black sublattice sites of the honeycomb lattice in Fig. 8.3 are all occupied by As but that each white site can be occupied by either an A or a B. There are ferromagnetic interactions with interaction constant J between a spin on a white site and the spins on the neighbouring black sites. Show that in zero eld, the white sites can be removed by a star-triangle transformation with equation 8.25 replaced by 1 + 2 cosh + + G = R exp K + + : Hence show that
292
exp4K = 1 + 2c3 =1 + 2c1 ;
9.5 From equation 9.144 for the decorated lattice gas deduce that, when w1 = 0 and " = ,2zw=z , 2, equation 9.165 applies. 9.6 For the decorated lattice gas, show from 9.144 that, when w1 = 0 and " = ,1, exp4K e = expw=T 1 + ;
R4 = 1 + 2c3 1 + 2c1 3 ; where c` = cosh`G. Show that the critical curve in the x; T plane, x being the fraction of white sites occupied by As, is c , x = exp4K,1 1 f3 1 , gp K c ; 0 8c2 + 2c2 , 1 1 + 3gpK c ; 0 exp,4K c g ; K c and gp K c ; 0 having the values appropriate to the triangular lattice. Hence show that the critical mole fraction xc is given by 1 xc = 4 1 , exp,4Kc 1 + 3gpK c; 0 = 1 : 2
exp2Le = exp zw=2T 1 + ,z=2 : By considering the two possibilities, ! 0 as T ! 0 and ! 1 as 1 T ! 0, show that for w 0 the points = , 2 zw and = zw=z , 2 are the termini of the Le = 0 locus on the -axis of the T; plane. Show that on the Le = 0 locus there is a maximum value Tmax of T given by exp zw=2Tmax =
1 2
1 z 12 z,1 z 2
1z , 1 2
and that the Le = 0 locus is a loop between the termini on the -axis. Show that on the Le = 0 locus the minimum value of exp4K e is fz= z , 1z,1=z g2 and then, using Table 8.1, show that the entire Le = 0 locus lies in the K e K c region of the T; plane for simple cubic, body-centred cubic and face-centred cubic primary lattices.
10. The Six-Vertex Model
10.1 Two-Dimensional Ice-Rule Models

In this chapter we consider exact solutions for a class of two-dimensional models which have completely di erent behaviour from the Ising model. These models are based on a square lattice version of a completely hydrogen-bonded network, like that of ice I see Appendix A.3. With all bond con gurations having equal energy, the free energy is related to the ground-state zeropoint entropy of a d = 2 ice analogue square ice, which was evaluated by Lieb 1967a. As well as water, certain other substances, such as KH2 PO4 potassium dihydrogen phosphate, crystallize as four-coordinated hydrogenbonded networks. KH2 PO4 is a ferroelectric with a spontaneous electrical polarization in zero eld below a critical temperature. Square lattice analogues of ferroelectrics and antiferroelectrics can be constructed by appropriate assignment of energies to the various bond con gurations. Lieb 1967b, 1967c obtained expressions for the free energy of these models by an extension of the method used for square ice. In contrast to the Ising model, some results can be obtained for non-zero elds electric elds in this case Sutherland 1967, Lieb 1969. The development in this chapter is based on the extensive review article of Lieb and Wu 1972. We rst introduce an arrow convention for the polarity of the bonds. A water molecule see Appendix A.3 can be regarded as having two negative and two positive àrms' Fig. 10.1a, and two molecules are hydrogen bonded by the junction of a positive arm from one and a negative arm from the other. Now distinguish the positive arms, that is those corresponding to the positions of the two protons, by arrows directed away from the molecule's centre Fig. 10.1b. There is then a single arrow on the segment between
+ +
,
a
b c Fig. 10.1. Water molecule polarity and bonding: arrow notation.
294
the centres of any bonded pair of molecules Fig. 10.1c. When a completely hydrogen-bonded network is represented by a graph with the centres of the molecules at the vertices and one arrow on each edge, there are inward arrows on two of the four edges meeting at each vertex and outward arrows on the other two. This `two in, two out' law is known as the ice rule. Since the coordination number z = 4 for the square lattice, the latter can be used as a two-dimensional analogue of a completely hydrogen-bonded network. With the application of the ice rule there are six permitted types of vertices and the model is referred to as the six-vertex model. The most general vertex model on the square lattice is the sixteen-vertex model, in which all 24 possible arrow con gurations about a vertex are allowed. This model, which can be shown to be equivalent to an Ising model with two, three and four site interactions and an external eld Suzuki and Fisher 1971, is unsolved. The model in which the ice rule is modi ed to allow in addition the two vertices when all the arrows are pointing inwards or outwards is the eightvertex model. The zeroeld case of this model was solved by Baxter 1972. It is discussed in Volume 2, Chap. 5 where we show that both the zeroeld Ising model and six-vertex models are special cases see also Baxter 1982a, Lavis 1996. Each site of the six-vertex model is identi ed with a vertex of one of the six types permitted by the ice rule. These are shown, with assigned index numbers, on the rst line of Fig. 10.2. They fall into pairs 12, 34 and 56 such that reversal of all arrows interchanges the two members of each pair. In the second line of Fig. 10.2 the double-headed arrows indicate the resultant electric dipole moments. Vertices 1 and 2 have oppositely directed dipole moments as do 3 and 4 whereas 5 and 6 have no resultant dipole moments in the plane of the diagram. These unpolarized vertices reverse the ` ow' of arrows along any row of edges and so occur alternately see Fig. 10.3. Thus, in the thermodynamic limit, the numbers of 5 and 6 in each row, and hence in the whole lattice, are equal. The con gurational energy is assumed to be
1 2 3 4 5 6
Fig. 10.2. Vertex ty-
pes in the completely hydrogen-bonded square network.
10.2 Parameter Space
295
Fig. 10.3. Row of vertices showing reversal
of horizontal arrow direction by types 5 and 6.
the sum of vertex energies ep, p = 1; 2; : : : ; 6, one of which can always be taken as zero since this involves changing the total con gurational energy by only a constant. We impose the conditions e1 = e2 ; e3 = e4 ; e5 = e6 : 10.1 Since the numbers of 5 and 6 are equal, the last relation involves no loss of generality.1 From the rst two relations, the energy of a polarized vertex is unaltered by a 180 rotation of the dipole moment. These relations therefore imply the absence of an electric eld. For the ice problem, ep = 0 for p = 1; : : : ; 6. The entropy of d = 2 ice is thus kb ln N where N is the number of arrow con gurations satisfying the ice rule at all the N vertices of the lattice. As will be seen below, the derivation of N is not easy. However a good approximation was obtained by Pauling 1935. For large N the number of edges is 1 zN = 2N and so the 2 number of unrestricted arrow con gurations is 22N . Since the ice rule allows only six out of the sixteen possible con gurations at each vertex, Pauling put
6 N
N N = 22N 16 = 3 ; 10.2 2 giving the value kb ln3=2 for the entropy per molecule. Other systems originally studied were the KDP Lieb 1967c and F Lieb 1967b models which respectively represent a ferroelectric and an antiferroelectric. As can be seen from Table 10.1, the vertex energies for the KDP model favour vertices 1 and 2 whereas those for the F model favour the unpolarized vertices 5 and 6.
10.2 Parameter Space

We consider the most general zeroeld case with the ep satisfying 10.1. The vertex energies expressed in terms of "1 , "2 "3 are given in the `zeroeld' line of Table 10.1. The model is referred to as a ferroelectric or antiferroelectric according to whether the lowest energy con guration ground state consists of polarized or unpolarized vertices.2 We introduce the classi cation
In fact it is often convenient, as in the models in Table 10.1, to impose also the condition e5 = e6 = 0. 2 The terms intrinsically ferroelectric or intrinsically antiferroelectric are sometimes used to emphasize that it is the zeroeld ground states which are the basis of the classi cation. However the word ìntrinsically seems no more necessary here than it is for the Ising model.
without loss of generality
296
10. The Six-Vertex Model Table 10.1. Vertex energies for di erent realizations of the six-vertex model. Model
e1
Vertex Energies e2 e3 e4 e5 e6 "1 = "2 = "3 = 0 "1 0; "2 = "3 = 0 "1 0; "2 = "3 = 0 "2 0; "1 = "3 = 0 "2 0; "1 = "3 = 0
Zero Field "2 , "1 "2 , "1 "2 "2 "3 "3 Square Ice 0 0 0 0 0 0 KDP ,"1 ,"1 0 0 0 0 IKDP ,"1 ,"1 0 0 0 0 F "2 "2 "2 "2 0 0 IF "2 "2 "2 "2 0 0
i "1 0, "1 "2 , "3 , e1 e3 , e1 e5 : ferroelectric, ii "1 0, "2 "3 , e3 e1 , e3 e5 : ferroelectric, iii "2 "3 , "2 "1 + "3 , e5 e3 , e5 e1 : antiferroelectric. For case i, the ground-state vertices are of type 1 or 2, for case ii of type 3 or 4, and for case iii of type 5 or 6. From Table 10.1, the KDP and F models are examples of i and iii respectively. The IF inverse F model is transitional between i and ii whereas the IKDP inverse KDP model is transitional between ii and iii. We now de ne the vertex weights Zp = exp,ep =T , p = 1; : : : ; 6 and using 10.1, the more compact notation, a = Z1 = Z2 = exp "1 , "2 =T ;
b = Z3 = Z4 = exp,"2=T ;
10.3
c = Z5 = Z6 = exp,"3 =T : As indicated above we could, without loss of generality, set "3 = 0, c = 1. However, in view of the discussion of the staggered six-vertex model in Sect. 10.14 it is not convenient to do so. But, since a, b and c are all positive, the parameter space is the positive quadrant of the plane of the scaled variables a=c; b=c see Fig. 10.4 where the point 1; 1 corresponds to T = 1 for all "1 and "2 . For any model de ned by prescribing values of "1 "2 and "3 a point a=c; b=c is speci ed by 10.3 at each value of T . As T varies from in nity to zero, the point traces out a trajectory in the appropriate part of parameter space. For case iii, a c, b c, so that the unit square bounded by the lines a=c = 1 and b=c = 1 is the antiferroelectric domain. The origin a=c = b=c = 0 corresponds to T = 0 for any antiferroelectric model. The
b
1
FE
b a
10.3 Graphical Representation and Ground States 297 Fig. 10.4. Parameter space for the sixvertex model showing ferroelectric FE and antiferroelectric AFE regions. "1 = 0
AFE
T =1
FE
a b
0 1
region outside the unit square is the ferroelectric domain. It is divided into two by the line a = b, which corresponds to "1 = 0. Below this line, a c and a b so that the inequalities i apply, whereas above it, b p c and b a so that the inequalities ii apply. For all ferroelectric models, a2 + b2 =c ! 1 as T ! 0. Consider a square lattice with an equal number of rows and columns and arrows on its edges in a particular ice-rule con guration. Suppose that the lattice is rotated in its own plane through an angle of 90 in an anticlockwise sense. This produces a di erent ice-rule con guration in which the original vertices of types 1, 2, 3, 4, 5 and 6 become vertices of types 4, 3, 1, 2, 6 and 5 respectively. Thus a and b are exchanged in the corresponding term in the partition function. Now when this rotation is applied to all ice-rule con gurations successively, the result is to produce the same set of con gurations in a di erent order. It follows that the partition function Z expressed in terms of a, b and c satis es the symmetry condition Z a; b; c = Z b; a; c: 10.4 Thus no generality is lost by putting a b "1 0. Henceforth we shall take a b "1 0, unless stated otherwise, leaving the case a = b "1 = 0 for separate consideration. So, for ferroelectric models, we consider only case i, where the ground state consists of vertices of type 1 or 2 and the ground-state energy is N "2 , "1 .
10.3 Graphical Representation and Ground States

In Chap. 8 we represented Ising model spin con gurations by graphs on the lattice or its dual. For arrow models we obtain a con guration graph by putting a line on any horizontal edge with a left pointing arrow and on any vertical edge with a downward arrow. The corresponding line patterns at the six types of ice-rule vertex are shown on the bottom row of Fig. 10.2.
298
Before considering the form of the con guration graphs it is helpful to think of the square lattice as wrapped round a vertical cylinder in such a way that the last site of each row becomes a nearest neighbour of the rst. The only boundary sites are now those on the top and bottom rows. Also we adopt the convention that the `cross' pattern of vertex 2 is to be regarded as a pair of `corners' like those at vertices 5 and 6. A typical graph is shown on Fig. 10.5a. It consists of three paths which can be regarded as followed in the sense of the arrows, which is the same as that of the original arrows on the lines making up the path. The apparently isolated piece of path in the bottom right-hand corner is a continuation on the cylinder of the left-hand path. From the permitted vertex patterns, no sequence of lines can terminate except at a boundary site. Only steps downward or to the left are allowed so that a path leaving the top row of sites must nish at the bottom. So any path passes through a given row of vertical edges exactly once. This has an important consequence. If n denotes the number of downward arrows on a row of vertical edges in the original arrow con guration, then in the con guration graph the number of paths passing through this row is n. Thus n is the same for all rows of vertical edges. In Fig. 10.5a, for instance, n = 3 but the constant n rule also applies for alternative path patterns. Fig. 10.5b shows a con guration graph consisting of a single path spiralling round the cylinder. It can be seen that n = 1. Any closed path must encircle the cylinder. Since no upward steps are allowed, no downward steps can occur either, and the path is con ned to a single row of horizontal edges. Con guration graphs composed of these ring paths correspond to n = 0. The vertices on the ring paths are all of type 4 whereas the remaining rows consist of vertices of type 1. The null graph is a particular case where all vertices on the lattice are of type 1.
a b Fig. 10.5. Con guration graphs on the cylinder wrapped square lattice.
10.4 Free Energy and Transfer Matrices
299
In the ferroelectric ground state, all vertices are identical and are either of type 1 or of type 2. The null graph corresponding to the type 1 ground state can be changed only by the addition of a ring path or a path like one of those shown in Fig. 10.5. Thus no small perturbation of the ground state is possible. Since the energy is invariant under the reversal of all arrows, the same is true for a type 2 ground state. This contrasts with the Ising ferromagnetic ground state where the complete spin alignment can be perturbed by overturning one spin. From Sect. 8.2 this corresponds to a dual lattice con guration graph consisting of a single 1 1 square. No such closed polygons can occur in the con guration graphs of ice-rule models. In the antiferroelectric ground state the con gurational energy is zero and all vertices are of types 5 and 6. Since the latter must occur alternately along every row and column the only possible arrangement is one of sublattice order see Appendix A.1 with one sublattice consisting of vertices of type 5 and the other of type 6. The con guration graph is shown by the full lines in Fig. 10.6. Small perturbations can occur, and one is shown in Fig. 10.6. The lines marked with a cross are replaced by the broken lines, and the numbers are the indices of the new vertices introduced by the perturbation. Although the new vertices are polarized ones, the resultant dipole moment is zero.
2 3
4 1
Fig. 10.6. Con guration graph for antiferro-
electric sublattice ordered ground state with a perturbation.
10.4 Free Energy and Transfer Matrices

It has been shown that n, the number of downward pointing arrows, is the same for all rows of vertical edges in a given con guration. Denote the number of sites in each row by N1 and the number in each column by N2 so that the total number of sites N = N1 N2 . Let n be the sum of the terms in the zero electric eld partition function which correspond to con gurations with the given value of n. Using the procedure of Sect. 2.6 the whole partition function can be equated to the maximum n . Since, for zero eld, the con gurational energy is unchanged by reversal of all arrows we have the symmetry relation
300

n = N 1 ,n :
10.5 Thus the maximum n is unique only if it corresponds to n = 1 N1 . Num2 bering the vertical edges of row j from 1 to N1 , suppose that the n downward arrows are on edges 1 1 2 n N1 . The arrow distribution is then speci ed by the vector j = 1 ; 2 ; : : : ; n and similarly that on the j +1-th row of vertical edges by a vector j+1 . Taking the j-th row of vertices as lying between the j-th and j + 1-th rows of vertical edges, we de ne X V j ; j+1 = exp,Ej =T ; 10.6 where Ej is the sum of the energies of the j-th row of vertices, and the summation is over all vertex con gurations on row j compatible with the given j and j+1 . The V j ; j+1 are elements of a transfer matrix V of dimension N1 != n!N1 , n! , and a method similar to that of Sect. 2.4.1 will be used though the single sites are now replaced by rows of vertices. In Sect. 2.4.2 the line of sites was converted into a ring by making the rst and last sites into nearest neighbours. We now make the sites of row 1 into nearest neighbours of the corresponding sites of row N2 , thus bending the cylinder on which the lattice is wrapped into a torus. Then X V 1 ; 2 V 2 ; 3 V N 2 ,1 ; N 2 V N 2 ; 1 n= = TracefV N 2 g : 10.7 The right eigenvector of V with elements g and corresponding eigenvalue satis es the relation X V ; 0 g0 = g ; 10.8 where and 0 specify the distributions of downward arrows on successive rows of vertical edges. An important theorem due to Perron Gantmacher 1979 has the following form: A matrix of positive non-zero elements has an eigenvalue which is real and positive and of magnitude larger than that of any other eigenvalue. This eigenvalue has left and right eigenvectors with all elements real and positive non-zero. Since V is a matrix of this type then it has a unique real maximum eigenvalue denoted by n. As we shall see below, only one right eigenvector can have all its elements real and positive, and we shall use this fact to identify n . Equation 10.7 yields an expression like 2.92 but with N1 != n!N1 , n! terms instead of two in the sum. The leading term n N 2 becomes dominant for large N2 and 10.9 lim ln n = ln :
N 2 !1 N2
n
f1 ;2 ;:::;N 2 g
f 0 g
10.5 Transfer Matrix Eigenvalues
301
Although the torus condition is convenient, it is not essential and, with a little more trouble, 10.9 can be proved for an open-ended cylinder. Since the temperature and electric eld have been taken as the independent intensive thermodynamic variables, with the eld xed at zero, ,T lnn is a non-equilibrium free energy per row of sites, depending on the òrder variable' n. Minimizing ,T lnn with respect to n is equivalent to nding the maximum term n . In the thermodynamic limit, when N1 ! 1, n must be replaced by the relative vertical electrical polarization component p, de ned by 10.10 p = N1 , 2n ; which is also regarded as an order parameter. A zeroeld non-equilibrium free energy f T; p per site is de ned by lnn f T; p = ,T lim ; 10.11
N 1 !1
N1
the ratio n=N1 being kept constant during the limiting process. From 10.9 and 10.11 : 10.12 f T; p = ,T lim 1 lim ln n
N 1 !1 N1 N 2 !1 N2
N1 n=N 1 1,p=2
It can be seen that the thermodynamic limit is approached in two stages.3 From the symmetry relation 10.5 and from 10.10, f T; p = f T; ,p : 10.13 To obtain the equilibrium zeroeld free energy per site, f T; p must be minimized with respect to p.4
n=N 1 1,p=2

Suppose that the lattice is wrapped on a vertical cylinder but that the step of bending the cylinder into a torus has not yet been taken. Let the cylinder be rotated by an angle 2s=N1 about its axis, s being an integer. Then each i ! i + s, mod N1 and rearrangement of the resulting numbers in ascending order yields a new vector s . Since the energies Ej are not changed by the rotation, V s ; 0s = V ; 0 : 10.14
3 4
See Lieb and Wu 1972, p. 360, for proof of validity of this procedure. It should be noted that if p, which is a density in the sense of Sect. 1.8, were taken as an independent variable instead of the eld then f T; p would become the equilibrium Helmholtz free energy density.
302
Fig. 10.7. Part of a con guration graph for one row of vertices: the n = 1 case.
It follows that, for all , gs =
s g ; 10.15 for some
s . Since N 1 = ,
N 1 = 1. Also
s
r =
r+s so that there is a number
such that
s =
s ;
N1 = 1 : 10.16 The determination of the eigenvalues of V for general n is complicated. In this section we give the derivation for the cases n = 0 and n = 1 but present only the results for general n. The detailed analysis is contained in Appendix A.4.
10.5.1 The Case n = 0 The matrix V is 1 1 and hence equal to 0 . On the intervening row of
horizontal edges, either all arrows point to the right or all arrows point to the left, giving vertices of type 1 or 4 respectively. From 10.6 and 10.3 0 = aN1 + bN 1 : 10.17 The same result applies for n = N1 , in accordance with 10.5, but the vertices are of type 2 or 3. Since a b, the rst term on the right-hand side of 10.17 is dominant for large N1 . Hence, from 10.9, 10.11 and 10.13, f T; 1 = f T; ,1 = ,T lna = "2 , "1 ; 10.18 corresponding to the ferroelectric ground state.
10.5.2 The Case n = 1

N1 X
The vector reduces to a scalar , which denotes the position of the single downward arrow, and the matrix V is N1 N1 . Equation 10.8 becomes For 6= 0 there is only one term in the sum on the right of 10.6 and its form depends on whether 0 or 0 , the relevant portions of con guration graph for the two situations being shown in Fig. 10.7. See also Fig. 10.5b, which shows a particular n = 1 con guration. It can be seen that the two situations are still distinct in the limiting cases where 0 =
0 =1
V ; 0 g0 = g :
10.19
303
and there are accordingly two terms in V ; , corresponding respectively to rightward and leftward ows of arrows on the row of horizontal edges. For the eigenvector elements g, we shall try the simplest form which satis es 10.15 and 10.16: g =
;
= expik ; k = 2=N1 ; = 0; 1; : : : ; N1 , 1 : 10.20 Each eigenvector is characterized by one of the N1 possible values of the wave number k and we have to nd the corresponding k. We now write
N1 X
where the index i corresponds to 1 0 and ii to 0 N1 . Using Figs. 10.2 and 10.7
i = abN 1 ,1
+ c2
0 =1
V ; 0 g0 =
N1 X
0 =1
V ; 0
0 =
,1 X
i + ii ;
10.21
0 =1
a,0 ,1 bN 1 ,+0 ,1
0 ; aN 1 ,0 +,1 b0 ,,1
0 :
ii = aN 1 ,1 b
+ c2
N1 X
10.22
After summing the geometrical progressions and some rearrangement,

i = M
bN 1
+ c2 bN 1 a
aa , b
,1 ; b 10.23
a N 1+1 ii = L
aN 1
, c2 bN 1 ,1 ; aa , b
b
where 2 2 ,1 2 2 L
= b a,bc,,
ab1
; M
= a b,ac,,
ab
: 10.24 , a b Substituting 10.23 into 10.21 and using the second relation of 10.16,
N1 X
0 =+1
so that equation 10.19 is satis ed by the assumed form of g. The eigenvalue is the coe cient of
on the right-hand side of 10.25 and it is desirable to express it in terms of the wave number k. Accordingly we de ne ,1 10.26 Lk = L expik = 2 , b ,1a , expi,ik ; 1 , b a exp, k where
,1 Mk = M expik = 2 , a ,1b , expik ; 1 , a b expik
0 =1
V ; 0
0 =
L
aN 1 + M
bN 1 ;
10.25
10.27
304

2 2 2 = a +2b , c = cosh"1 =T , 1 exp 2"2 , 2"3 , "1 =T : 2 ab
10.28
The eigenvalues can now be written as k = LkaN 1 + MkbN 1 ; 10.29 where k can take any of the N1 values speci ed in 10.20. As indicated in Sect. 10.4, the largest eigenvalue can be deduced from Perron's theorem since the corresponding eigenvector can be taken to have real positive elements and from 10.20 this occurs only when k = 0 = 0. Hence, by 10.29, 1 = 0 = L0aN 1 + M0bN 1 : 10.30 Since a b, the rst term in the last expression becomes dominant for large N1 , just as in 0 .
10.5.3 The General n Case
We assume the Bethe ansatz that the elements of a given eigenvector depend on n real wave numbers k1 ; k2 ; : : : ; kn according to the relationship X g = Az1 z2 zn ; 10.31 1 2 n where zm = expikm and is a permutation f1 ; 2 ; : : : ; n g of the numbers 1; 2; : : : ; n and the sum is over all n! permutations. In Appendix A.4 it is shown that I z1 z2 zn =
; 10.32 where
is the quantity appearing in 10.16. Results for n = 1 are regained by putting 1 = , z1 =
and k1 = k. In Appendix A.4 two important results are proved for the coe cients A. De ning operators R and Sj;j+1 acting on the permutation by R = f2 ; 3 ; : : : ; n ; 1 g; 10.33 Sj;j+1 = f1 ; 2 ; : : : ; j+1 ; j ; : : : ; n g; these are , II AR = z1N1 A = exp,ik1 N1 A ; 10.34 III ASj;j+1 = ,A exp ikj ; kj+1 ; where Lq Mq , 1 exp iq1 ; q2 = Lq2 Mq1 , 1
1 2
fg
10.35
= 1 + expfiq2 + q1 g , 2 expiq2 : 1 + expfiq + q g , 2 expiq

1 2 1
10.36
305
Since q1 and q2 are real, j expfiq1 ; q2 gj = 1 so that p; q is real. Also, from 10.36, q1 ; q2 = 0 ;q ; q ; q1 = q2 ,. 10.37 , 2 1 q1 6= q2 The results I and II are direct consequences of symmetry with respect to rotation about the axis of the cylinder. Together with III they imply see Appendix A.4 that the eigenvalue corresponding to the eigenvector 10.31 is IV k1 ; k2 ; : : : ; kn = aN1 Lk1 Lk2 Lkn + bN1 Mk1 Mk2 Mkn : 10.38 Any permutation can be obtained from any other by a sequence of operations Sj;j+1 . For instance , R = Sn,1;n Sn,2;n,1 S1;2 and comparing 10.34 and 10.35 exp ,ik1 N1 = ,1n,1 exp i
h X
However, k1 can be any of the wave numbers k1 ; : : : ; kn and, from 10.37, kj ; kj = 0 so that 10.39 is equivalent to exp ikj N1 = ,1n,1 exp , i
n h X r=1
fm 6=1 g
k1 ; km :
10.39
kj ; kr :
10.40
Equation 10.38 is a generalization of 10.29 and it might be expected that could be maximized by putting all kj = 0. Unfortunately this is not so, except when = 1, and it will now be shown that, for 6= 1, all kj must be distinct. For de niteness, consider k1 and k2 . Let E12 be an operator acting on a permutation which transposes the indices 1 and 2. Suppose that, in , 1 occurs before 2 and that they are separated by s other symbols. Then E12 is the product of 2s + 1 adjacent transposition operators Sj;j+1 , and, from 10.35, AE12 = ,12s+1 A expi where is the sum of 2s +1 terms. One of these is k1 ; k2 and the other terms can be arranged in pairs k1 ; ki + ki ; k2 . Then, if k1 = k2 , k1 ; k2 = 0 and k1 ; ki + ki ; k2 = 0 from 10.37. Hence = 0, AE12 = ,A and, since z1 = z2 , the terms corresponding to and E12 cancel in 10.31. As the whole set of n! permutations can be divided into pairs di ering by a 1 $ 2 transposition it follows that g = 0 when k1 = k2 . The same reasoning is valid for any pair of wave numbers ki ; kj . The exceptional case = 1 will be discussed in Sect. 10.6. No method is known for nding and maximizing n for a nite lattice at an arbitrary value of T . However the problem can be solved in the thermodynamic limit and the equilibrium f T obtained. In the thermodynamic limit it is necessary to consider a continuous spectrum of k values, and this theory is discussed in Sects. 10.7 10.11.
306
10.6 The Low-Temperature Frozen Ferroelectric State

In the present and subsequent sections it will be shown that the values = 1 for the parameter de ned in 10.28 give the transition temperatures for ice-rule models. The line b = c , a on which = ,1 and the lines b = a , c and b = a + c on which = 1 are shown in Fig. 10.8. Comparison with Fig. 10.4 shows that each line is relevant to only one type of model, the = ,1 line to the antiferroelectric case iii and the = 1 lines b = a , c and b = a + c to ferroelectric cases i and ii respectively. We concentrate on the ferroelectric and, since we assume a b, need consider only the line b = a , c. For given values of "1 and "2 the trajectory crosses this line at T = Tc where, by 10.3, exp,"2 =Tc = exp "1 , "2 =Tc , exp,"3 =Tc : 10.41 The trajectory for the KDP model, where b = c for all T , is shown in Fig. 10.8. Now we look at the equations for general n given in Sect. 10.5. The value = 1 has a special signi cance for 10.36 since the right-hand side becomes indeterminate for q1 = q2 = 0. To nd the limiting value we expand in a Maclaurin series. Then as q1 ! 0, q2 ! 0 . 10.42 exp iq1 ; q2 ' iq1 , q2 + ! ,1 iq2 , q1 + It follows that 10.40 is satis ed both sides being equal to 1 when kj = 0, j = 1; : : : ; n. Again, 10.35 reduces to ASj;j+1 = A and the eigenvector g does not vanish, as it does for 6= 1 unless all the kj are distinct. In fact, from 10.31, all eigenvector elements have an equal value which can be set equal to 1. Thus for T = Tc = 1 not only does the set of wave numbers kj = 0 j = 1; : : : ; n give an eigenvalue but, by Perron's theorem see Sect. 10.4, this is the maximum eigenvalue. Hence, by 10.38, n = aN 1 L0 n + bN 1 M0 n : 10.43 However, this is not all. From 10.26 and 10.27, L0 = M0 = 1 when = 1 and, by comparison with 10.17, we have
=1 T =1 = ,1 =1
1
KDP
b
1 0
Fig. 10.8. The six-vertex parama

eter space with lines of critical points and KDP trajectory.
10.7 Wave-Number Density
307
for all n . 10.44 From 10.11 it follows that, at T = Tc, f T; p is independent of p. Now suppose that 1, implying a ferroelectric with T Tc see Example 10.1. Take N2 and N1 so large that n = N 2 and 0 = aN 1 n without signi cant error. Then 10.45 0 = exp ,N "2 , "1 =T ; with N "2 ,"1 being the ferroelectric ground state energy. Now for n 6= 0 and n 6= N1 the terms in n correspond to con gurational energies N "2 , "1 + 4E , where 4E 0. It follows that the ratio n = 0 is the sum of terms each depending on T through a factor exp,4E=T . Hence d n = 0 =dT 0. However, at T = Tc, from 10.44, N 10.46 n = 0 = n =0 2 = 1 : Thus, for T Tc, when n 6= 0 and n 6= N1 . 10.47 n 0 = N1 Thus, from 10.12 and 10.13 we have, for the non-equilibrium free energy per site, f T; p f T; 1 = f T; ,1 when p 6= 1 . 10.48 The equilibrium free energy per site for the ferroelectric with T Tc is thus f T = f T; 1 = ,N ,1 T ln 0 = ,T lna = "2 , "1 : 10.49 The equilibrium state is thus one of perfect long-range order with either p = 1 all vertices of type 1 or p = ,1 all vertices of type 2. This is a remarkable result. The kind of completely ordered con guration which exists only at T = 0 for the Ising ferromagnet is here `frozen in' over the nite temperature range 0; Tc. Although Tc is clearly a transition or critical temperature for the ferroelectric we know nothing about the type of singularity there until we have studied the equilibrium state for T Tc.
n = 0 ;
10.7 Wave-Number Density

We can take 1, which applies for a ferroelectric when T Tc and for an antiferroelectric at all temperatures see Example 10.1. Now 10.40 implies
N1 kj = 2`j , n , 1 ,
n X r=1
kj ; kr ;
j = 1; : : : ; n ;
10.50
where the `j are integers. Since, from 10.10, n ! 1 in the thermodynamic limit for p 1, we can take n as even. It has been shown that the kj are distinct for 1 and, as the required largest of the eigenvalues given by 10.38 is real, they must occur in equal and opposite pairs. The dominant
308
term on the right-hand side of 10.38 then depends on the real products LkL,k. By 10.26, 2 jLkj2 = LkL,k = 2 , X++ 1 , 22 ,kX cosk ; 10.51 1 X2 , 2X cos where X = a=b = exp"1 =T : 10.52 For 1, jL0j 1 and jLkj decreases as k increases see Example 10.2. So, for the largest eigenvalue, the n wave numbers kj must be packed as closely as possible in a distribution symmetrical about k = 0. To this end we put `j = j , 1 and the term 2`j , n , 1 then varies from ,n , 1 to n , 1. Since ,p; ,q = ,p; q it follows that kn,j+1 = ,kj .5 Now consider the thermodynamic limit in which N1 and n tend to in nity while p, de ned by 10.10, remains constant. From 10.50, with `j = j , 1, the number j of values of j corresponding to the interval k; k + k is given, for large N1 , by
N 1 k = 2 j ,
n X r=1
k + k; kr , k; kr :
10.53
In the thermodynamic limit there is a continuous wave-number distribution symmetrical about k = 0 and we introduce a density function k, such that j = N1 k k and ,k = k. Since there are n = 1 N1 1 , p wave 2 numbers in total, the normalization relation Z q0 1 qdq = 2 1 , p 10.54 ,q0 is satis ed, where ,q0 ; q0 is the range of the wave numbers. In view of the condition n 1 N1 we must take p 0 but, from the symmetry relation 2 10.13, no generality is lost. Dividing 10.53 by N1 k and letting k ! 0 we obtain Z q0 @k; q 2 k = 1 + qdq ; 10.55 where summation with respect to r is replaced by integration with respect to q. Since a b the right-hand side of 10.38 can be replaced by its rst term for large N1 . As the distribution k corresponds to the largest eigenvalue, we replace by n . Then, from 10.11, the zeroeld non-equilibrium free energy per site is given by
,q0 5 The problem of the kj distribution is the same as for the one-dimensional Heisenberg Model; see the contribution by Barouch in Sect. IV D of Lieb and Wu 1972. ,q0
@k
f T; p = ,T lna , T
Z q0
ln jLqj qdq :
10.56
10.8 The Solution of the Integral Equation for ,1 1
309
@x If Lk transforms to ex, then equation 10.56 becomes L T Z x0 ln jexjRxdx : L f T; p = ,T lna , 2 ,x0
where x0 corresponds to q0 . Suppose that k; q transforms to x; y where q has the same relation to y as k has to x. The integral equation 10.55 then becomes Z x0 @x; y dk Rx = dx + 21 10.59 ,x0 @x Rydy : Fourier techniques can be used to solve 10.59 for suitable x0 if x; y depends on x and y through the di erence x , y. The last term in 10.59 is then a convolution integral with a di erence kernel, and the equation can be written as Z x0 dk Rx = dx + 21 ,x0 K x , yRydy ; 10.60 @x , y : K x , y = 10.61
To minimize f T; p with respect to the order variable p, q0 is adjusted to maximize the integral in 10.56, subject to 10.55. Then the equilibrium value of p is deduced from 10.54. To solve 10.55 a transformation from k to another variable x is necessary. De ne a transformed density Rx by dk Rx = 2 k dx 10.57 and the normalization equation 10.54 becomes 1 Z x0 Rxdx = 1 1 , p ; 10.58 2 2 ,x0
It turns out that the form of solution depends on whether ,1 1 or ,1. The value of lies in the rst range for both the ferroelectric and antiferroelectric models at high temperatures, but the second range is relevant only for the antiferroelectric see Figs. 10.4 and 10.8.
10.8 The Solution of the Integral Equation for ,1 1

For this range, a real parameter can be de ned by cos = ,; with 0 , 10.62 where ! , 0 as ! 1 , 0 and ! +0 as ! ,1 + 0. From 10.51, jLkj = 1 when cosk = . So ln jLkj 0 for jkj , but changes sign
310
when jkj = , from which it follows that q0 = , is the largest value which q0 can assume without the integrand in 10.56 becoming negative. Thus putting q0 = , minimizes f T; p, provided that this value of q0 is compatible with sensible results for k and p from 10.54 and 10.55. We assume that q0 = , corresponds to the equilibrium state. If x0 = 1 the convolution integral in 10.60 has, from A.72, a Fourier transform which is the product of the transforms of K x and Rx. We try the transformation from k to x de ned by expi expx = expik + k + ; 10.63 1 + exp i which yields a real x for real and k. Also x = 1 for k = , so that q0 = , corresponds to x0 = 1. From 10.63, sinh , cos sink = coshxx sin 10.64 cosk = 1coshx,coshx ; cos , cos and hence, dk sin 10.65 dx = coshx , cos : Using 10.63 and a similar relation with y and q replacing x and k respectively, 10.36 becomes , exp x exp2i exp ,ik; q = exp ,ix; y = 1expy ,yx,,exp2i : 10.66 Since the last expression depends on x and y through the di erence x , y, the aim of a di erence kernel has been achieved. After logarithmic di erentiation of this expression with respect to x and substitution from 10.65, equation 10.60 becomes Z1 10.67 Rx = S ; x , 21 ,1 S 2; x , yRydy ; where S ; x = coshsincos : 10.68 x , The Fourier transform S ; ! of this function is, from A.70, given by 10.69 S ; ! = 2 sinh !!, : sinh and using A.72 with10.67 S ! R ! = 22+ S 2;;! : 10.70 This gives
10.9 The Free Energy of the Disordered State
311
Rx = 2 sech x : 10.72 2 With x0 = 1, the left-hand side of 10.58 is equal to R 0=2. However, from 10.71, R 0 = . So it follows from by 10.58 that the equilibrium value of p = 0. By 10.10, this corresponds to n = 1 N1 or an equal num2
and inverting the transformation
R ! = sech!
10.71
ber of up and down arrows on each row of vertical edges. The domain of the a=c; b=c plane for which ,1 1 see Fig. 10.8 thus represents the disordered state, which is reasonable since it corresponds to the hightemperature range for both ferroelectric and antiferroelectric, with in nite 1 temperature corresponding to a = b = c where = 2 .
10.9 The Free Energy of the Disordered State

We now derive the equilibrium zeroeld free energy for the disordered state with p = 0, ,1 1. We rst de ne a real temperature dependent parameter ' by + X expi 10.73 expi' = 1X + expi ; where X is given by 10.52 and by 10.62. This is equivalent to 2 1 2 cos' = 1X , X+ X 2 : 10.74 + ,2 + X From 10.74, cos' cos allowing the condition 0 ' . At T = 1, where X = 1 and = 1 , ' = 0 and = 2 . For the ferroelectric at T = Tc, 2 3 where = 1, ' = = . To obtain jexj, substitute from 10.64 into L 10.51. The resulting expression can be simpli ed by deriving the real part of expi' , 2i from 10.73 and also using 10.74. This gives , , 10.75 jexj2 = coshx x cos2 ' ' : L cosh , cos Substituting 10.75 and 10.72 into 10.61 and putting x0 = 1, which corresponds to p = 0, the equilibrium free energy per site takes the form f T = f T; 0 = ,T lna + T ; '; 10.76 where Z 1 1 1 sech x ln coshx , cos2 , ' dx : 10.77 ; ' = , 8 2 coshx , cos' ,1 This applies to both the ferroelectric and antiferroelectric in temperature ranges for which ,1 1. Replacing x by 2x gives the alternative form
312
cosh2x , cos' dx : 10.78 Another form for ; ' can be derived directly from 10.61 with x0 = 1. The Fourier transform of Rx is given by 10.71 and , sin F! fln jexjg = 2 sinh !! sinh! ! , ' L 10.79 Baxter 1982a, p. 148. It follows from Parseval's formula A.73 that Z 1 1 sinh y , sinh y , ' dy: ; ' = , 2 10.80 y coshy sinhy ,1 We now consider two particular cases:
,1
; ' = , 4
10. The Six-Vertex Model Z 1 1
sechx ln
cosh2x , cos2 , '
10.9.1 The KDP Model

Here "2 = 0 and, by 10.28 and 10.52, X = 2 so that cos' = 3 , 43 = cos3 : 10.81 Hence we put ' = 3 , 2 10.82 2 and ' = for = . Glasser 1969 showed which yields ' = 0 for = 3 that for the KDP model the integral in 10.80 can be expressed in terms of elementary functions to give, from 10.76, 2 2 f T = ,"1 , T ln sin cot 2 : 10.83
10.9.2 Square Ice

Here "1 = "2 = 0 and a = b = c, so that square ice represents the in nite temperature limit of any ice-rule model. Accordingly, putting = 2 in 3 10.83, the entropy of the square ice model on N sites is given by Lieb's 1967a formula
4 =4 kb ln N = Nkb ln 4 cot3=3 = 3 Nkb ln 3 ; 10.84 2 3 sin2 which yields
4 3N =2 = 1:5396N ; 10.85 N = 3 a result which is remarkably close to the Pauling estimate 10.2.
10.10 The Ferroelectric Transition in Zero Field
313
10.9.3 Non-Zero Polarization

An explicit form for f T; p is available only for p = 0. However, an intricate perturbation analysis Lieb and Wu 1972 yields, for ,1 1, 1 T , cos ' p2 + Op4 : f T; p , f T; 0 = 4 10.86 2 Since and ' the coe cient of p2 is positive, so that f T; p has a minimum at p = 0, as would be expected. The zeroeld susceptibility can be obtained from 10.86. With a non-zero vertical electric eld6 E, the constanteld non-equilibrium free energy per site becomes f T; E; p = f T; p , Ep : 10.87 Minimizing f T; E; p with respect to the order variable p, we obtain the equilibrium condition E = @f T; p : 10.88 Substitution from 10.86 yields
@p
@p
@E
The equilibrium free energy f T; E can be obtained by substituting the value of p given by 10.88 in 10.87. The alternative symmetry condition f T; E = f T; ,E 10.90 follows from 10.87, 10.88 and 10.13.
E=0
2 = E!0 E = T , cos'=2 : lim p
10.89

In Sect. 10.6 it was shown that for T Tc, 1 a ferroelectric is frozen into a completely ordered con guration with p = 1. For T Tc, the equilibrium polarization is p = 0 so that p switches from 0 to 1 at T = Tc, implying a rst-order transition. We now look at other aspects of this transition. Consider a temperature just above Tc putting T = Tc + T ; Tc T 0 : 10.91 Since ! , 0 and ' ! , 0 as T ! 0, 0 =,' 1 = , 1; 10.92 in this limit and from 10.62,
6
Here E is a ` eld' in the sense of Sect. 1.7, and is the vertical electric eld multiplied by the vertical electric dipole moment component of a vertex of type 1 or 4.
314
,T = cos = , cos = ,1 + 1 2 + O 4 ; 2

1 cos' = , cos 0 = ,1 + 2 0 2 + O 0 4 : It is necessary to relate 0 to . From 10.74 and 10.93 1 ,X , 12 + 2 1 + X2 2 + O 4 cos' = X , 12 + X 2 + O 4 1 = ,1 + 2
2
10.93 10.94
4 10.95 X , 1 + O and comparison with 10.94 then yields 0 = X + 1 + O 3 10.96 X,1 and 10.97 cos2 , ' = , cos2 , 0 = ,1 + 1 2 , 0 2 + O 4 : 2 We now need a relation between and T . Using a Taylor expansion, 1 , T = Tc , T = 1 A T + O T 2 ; 10.98 2 where, by 10.28 and 10.41, ,2 ,1 A = ,2 d 10.99 dT T =Tc = 2Tc Xc , 1 "1 , "2 1 , Xc 0 ; with Xc = exp"1 =Tc. Comparison with 10.93 then yields = A1=2 T 1=2 1 + O T : 10.100 We now investigate the behaviour of the free energy f T as T ! Tc + 0. Using 10.94 10.97, the expansion of the logarithmic term in the integrand of 10.77 yields coshx , cos2 , ' 4 2 ln coshx , cos' = X , 1 coshx + 1 + O 4 : 10.101 To derive the essential results about the transition we need the 3 term in the expansion of ; '. Accordingly, using a binomial expansion of 1 , =,1 and a Taylor expansion of the sech function, we expand the other factor in the integrand to obtain 1 sech x = 1
1 , ,1 sech 1 x
1 , ,1 2 2 n o 1 = sechx=2 1 + 2 2 , x sechx=2 sinhx=2 + O 2 : 10.102
X + 1 2
315
Substituting from 10.101 and 10.102 and changing the variable of integration from x to y = 1 x, 2 Z1 1 2 y sinhy dy + O 4 1 + , coshy ; ' = , 2X , 1 3 ,1 cosh y 2 2 = 4X , 1 1 + 3 + O 4 : 10.103 From 10.100, an expression accurate to terms in T 3=2 can be obtained by replacing in 10.103 by A1=2 T 1=2 and X by Xc. From 10.77 and 10.103, f T = , lna + ; '
By comparison with 10.49 the free energy is, as would be expected, continuous at T = Tc. At zero eld the con gurational energy and enthalpy are equal. Dividing both sides of 1.66 by N and substituting from 10.104, the con gurational energy per site u is given by u = ,T 2 @ f T
= "2 , "1 , 4X A, 1 T + 2A T 3=2 + O T 2: 10.104 T 3

1 =2 c
T 1=2 ATc2 = "2 , "1 + 4X , 1 1 + A T 1=2 + O T : 10.105 c Since u = "2 , "1 for T Tc, there is a nite discontinuity in u, giving a
latent heat per site
@T
= 1 "1 , "2 1 , X,1 : 10.106 c 2 1 For the KDP model, where "2 = 0, 4u = 2 "1 . For Tc T the leading term in the zeroeld con gurational heat capacity c0 per site is given by 3=2 T 2 du b c c0 = kb dT = 8kA , 1 T ,1=2: 10.107 Xc Thus, c0 diverges as T ,1=2 when T ! Tc + 0 and, since c0 = 0 for T Tc, it has an in nite discontinuity at T = Tc. From 10.89 and 10.92, , 0 ,1 @p 2 = 1 0 , : 10.108 @ E E=0 = 2 T cos 2 , T sin 2 Using 10.96, the leading term is given by @p ' 2Xc , 1 ' 2Xc , 1 T ,1 : 10.109
ATc 4u = uTc + 0 , uTc , 0 = 4X , 1

c
@E
E=0
Tc
TcA
316
The zeroeld electric susceptibility thus diverges as T ,1 when T ! Tc +0. One very unusual feature of the foregoing analysis is that exact results
have been derived for the behaviour of thermodynamic functions near a rstorder transition.
10.11 The Antiferroelectric State

The range ,1 is relevant only for an antiferroelectric see Sect. 10.6 and Example 10.1 and then corresponds to T Tc0 where Tc0 is given, from 10.3, by aTc0 + bTc0 = cTc0 ; 10.110 exp "1 , "2 =Tc0 + exp,"2 =Tc0 = exp,"3=Tc0 : For ,1, jLkj 1 for all real k see Example 10.2 and hence f T; p, as given by 10.56, is minimized by putting q0 = . Because of this change in the equilibrium value of q0 , the analysis is rather di erent from that applicable when ,1 1. For ,1 the values of and ', given by 10.62 and 10.73 respectively, become imaginary and we de ne new real parameters = ,i; = ,i'; 10.111 where 0. From 10.62 and 10.73, cosh = , ; 10.112 + X exp exp = 1X + exp : To keep the variables real, we replace the variable x by ix and the functions R and K by i,1 R and i,1 K respectively. The equations 10.57 10.61 are unchanged in form and 10.63 becomes exp k + 1 expix = exp + iexpik : 10.113 + It can be seen that x = for k = . Hence q0 = transforms to x0 = . The integral equation 10.67 is thus replaced by 1 Z 2; x , yRydy ; Rx = ; x , 2 10.114 , where 10.115 ; x = coshsinhcosx = i,1 S i; ix : , With x, y, and ' replaced by ix, iy, i and i respectively, 10.75 becomes
10.11 The Antiferroelectric State
317
Although the integral in 10.114 still has a di erence kernel, the change in the range of integration to ,; means that it is necessary to use Fourier series rather than Fourier integrals. Applying the series transformation A.74 A.76 to 10.114 gives Rn = 22n 2 ; 10.117 + n where the transform of ; x is n = 2 expjnj : 10.118 Thus Rn = sechn; for all n. 10.119 When x0 = is substituted into 10.58 the left-hand side becomes equal to R0 =2. Since R0 = by 10.119 this implies p = 0 which is consistent with the antiferroelectric ground state, where the two sublattices are composed of unpolarized vertices of types 5 and 6 respectively see end of Sect. 10.3. Substitution of x0 = and the expression for jexj given by 10.116 L into 10.61 yields the equilibrium free energy in the form f T = f T; 0 = ,T lna + T ; ; 10.120 where 1 Z Rx ln cosh2 , , cosx dx : 10.121 ; = , 4 cosh , cosx , It can be shown Example 10.4 that f 0 = 0, as would be expected for the antiferroelectric. The Fourier coe cients of the logarithmic part of the integrand in 10.121 are given by 8 4 , ; for n = 0, cosh2 , , cosx Fn ln cosh , cosx = 4 sinh n , : n expjnj ; for n 6= 0, 10.122 and thus, from Parseval's formula A.77, 1 X sinh 1 ; = , 2 , , exp,ncoshn , 10.123 n n n=1 for the antiferroelectric when T Tc0, ,1. From 10.120 and 10.123 the free energy depends on T both directly and through the parameters and so that the con gurational entropy is non-zero. So the antiferroelectric zeroeld ground state with perfect 5 6 sublattice order is only attained at
, , cos jexj2 = cosh2 cosx x : L cosh ,
10.116
318
energy and hence zero con gurational entropy. This di erence is related to the fact that, as we saw in Sect. 10.3, the antiferroelectric ground state admits small perturbations whereas the ferroelectric ground state does not. The zeroeld correspondence in the Ising model between orientationally ordered and sublattice ordered states below the critical temperature clearly does not exist for ice-rule models. Now consider a perturbation of the polarization from zero to a value p, where 0 p 1. Suppose that x0 changes from to , and that Rx is modi ed to Rx + Rx. Then 10.58 becomes 1 Z , Rx + Rx dx = 1 1 , p : 10.124 2 2 ,+ Subtracting this from the corresponding relation for p = 0, we have to the rst order, denoting the n = 0 Fourier coe cient of Rx by R0 , , R0 + R = = 1 p : 10.125 2 Performing a similar variation on 10.60 and on 10.61 and using A.74 A.77 we obtain, for the non-equilibrium free energy f T; p, the relation f T; p , f T; 0 = jpj T ; jpj 1 ; 10.126 where " , 1 T , + T ln cosh ,, 1 2 T = , 2 1 cosh 2 1 X n ,2n sinh , T ,1 expn coshn n , : 10.127 n=1 The replacement of positive p by jpj is justi ed by the symmetry relation 10.13. Further analysis using elliptic functions shows Lieb and Wu 1972 that T 0 for 0 T Tc0 and that, as ! ,1 , 0 ! +0, 10.128 X+1 2 so that T ! 0 as T ! Tc0 , 0 and ! +0. Since T 0 for ,1, 10.126 shows that f T; p has a downward pointing cusp at p = 0. This is quite unlike the smooth minimum which f T; p has at p = 0 when ,1 1 see 10.86. It can be shown Lieb and Wu 1972 that the next term in the Maclaurin expansion of f T; p is 1 3 T jpj3 , where 3 T 0. Hence, for ,1, f T; p has the form 3 sketched in Fig. 10.9. Its least value occurs at the tip of the cusp, which corresponds to the equilibrium state with p = 0. For comparison, the limiting form at T = 0 see Example 10.4 and a sketch of f T; p for the disordered region where ,1 1 are also given in Fig. 10.9.
T = 0. This contrasts with an ferroelectric in zero eld when 0 T Tc where the ground state is frozen in with constant con gurational free
2 T ' 4T sin exp , ;
10.12 Field-Induced Transitions to the Completely Polarized State 319 Fig. 10.9. Dependence of f free energy density on reduced polarization: 1 = -1 0 1 p ,1, 2 ,1, 3 ,1 1. 1 2 3
The investigation of the antiferroelectric transition at T = Tc0, requires analysis based on regarding ; as a function of the complex variable . Lieb and Wu 1972, Glasser et al. 1972. It transpires that there is only a very weak singularity at T = Tc0 and that all derivatives of f T with respect to T are continuous there. The transition is said to be of in nite order. If the zeroeld heat capacity is plotted against T , the curve resembles that found for the d = 1 Ising model, with a maximum at a temperature well below Tc0 . The reader may feel that the antiferroelectric transition is a non-event, in contrast to the ferroelectric transition, which could hardly be more dramatic. However, signi cant changes in the properties of the antiferroelectric do occur at T = Tc0. In zero eld for T Tc0 , there is sublattice ordering of the vertices of types 5 and 6. Also the cusped form of f T; p has important implications for the response to an applied electric eld, as will be seen in Sect. 10.13.
10.12 Field-Induced Transitions to the Completely Polarized State

From Sect. 10.6 the completely polarized state p = 1 is attained in zero eld at a non-zero temperature for some ice-rule models. Such a state can also be attained at a nite value of jEj if the right-hand side of 10.88 remains nite at p = 1. Taking E 0 for de niteness, there is then a curve in the T; E plane, E = E T = @f T; p ; 10.129
1
where p reaches the value 1. Since p varies with E for E E1 T but is xed at 1 for E E1 T , 10.129 de nes a transition curve. From the symmetry condition 10.90 there is a mirror image curve E = ,E1 T in the lower half of the T; E plane.
@p
p=1
320
We now evaluate the right-hand side of 10.129 and check that it is nite. , Suppose that the number N1 of sites in each row is so large that ,TN1 1 ln n can be equated to f T; p, p being given by 10.10. Then f T; 1 , f T; 1 , 2N1,1 = N1,1 T ln1 =0 : 10.130 ,1 and let N1 ! 1. Since a b it follows from Now divide both sides by 2N1 10.17 and 10.30 that @f T; p 1 10.131 @p p=1 = 2 T ln L0 :
Substituting 10.131 into 10.129 and using 10.26 and 10.52, 1 T ln 1 + X , 2 : E1 T = 2 10.132 X,1 We assume "1 0 so that X 1 and hence E1 T 0 for ,1 1. It may be shown Example 10.5 that dE1 =dT 0 for ,1 1. Also, since 1 ! 2 as T ! 1, it follows from 10.132 that E1 T ' 1 T lnT="1 ! 1. 2 To investigate the transition on the line E = E1 T , information is needed about further derivatives of f with respect to p. From 10.58, p = 1 corresponds to x0 = 0. By di erentiating equations 10.58, 10.60 and 10.61 with respect to p and then letting x0 ! 0 it is possible to reproduce 10.131 and to show that, where ,1 1, @ 2f T; p @ 3f T; p = 0; 2B T = 0; 10.133 where the last relation de nes B T , Lieb and Wu 1972. Putting p = 1 , p, where p 1, and using 10.129 a Taylor expansion then gives f T; 1 , p = "2 , "1 , E1 T p , 1 B T p3 + O p4 ; 10.134 3 since f T; 1 = "2 , "1 . From 10.88, the equilibrium relation for p is E = E1T + BT p2 + O p3 : 10.135 Thus as E ! E1 T , 1 , p = p ' E1 T , E 1=2 jB T j,1=2 10.136 and it follows that @p 1 ,1=2 ,1=2 : 10.137 @ E T ' 2 jB T j E1 T , E Let E = E1 T so that T ; E is a point on the transition line E = E1 T in the T; E plane. Suppose that this point is approached from below on the line T = T see Fig. 10.10 for a sketch of the ferroelectric case. Putting T = T and E = E , E, E 0, in 10.137 it follows that @p=@ E E,1=2 as E ! E , 0. For T = T , E E the system is completely polarized with
@p2
p=1
@p3
p=1
10.12 Field-Induced Transitions to the Completely Polarized State
321
E E
0
p=1
P
0 p 1
Tc p = ,1
T
,1 p 0
Fig. 10.10. Transition curves in the temperature electric eld plane for a ferroelectric.
approach along the line E = E . Take a point P in Fig. 10.10 with coordinates T + T; E where 0 T T . By a Taylor expansion E1 T + T , E = A T + O T 2 ; 10.138 0: A = dE1 dT T =T With T = T + T , E = E in 10.136 and 10.137
p = 1 so that @p=@ E = 0. There is therefore an in nite discontinuity in @p=@ E at E = E although p itself is continuous. Thus a second-order transition on the curve E = E1 T is indicated. This will be con rmed by considering an
1 ,1=2 ,1=2 as T ! T + 0 , 10.140 @ E T ' 2 A jB j T where B = B T . From 10.87, the non-equilibrium free energy is f T + T; E ; 1 , p = f T + T; 1 , p , 1 , pE : 10.141 Substituting the equilibrium value of p from 10.139 and using 10.134 and 10.138, the equilibrium free energy at T = T + T and E = E is f T + T; E = "2 , "1 , E , 2 A 3=2 jB j,1=2 T 3=2 + O T 2 : 10.142 3
@p
A 1=2 p ' jB j T 1=2
as T ! T + 0 ,
10.139
322
= "2 , "1 , E + T A 3=2 jB j,1=2 T 1=2 + O T : 10.143 It follows that the constanteld con gurational heat capacity is of the order of T ,1=2 in the limit T ! T + 0 on the line E = E . From 10.140 this behaviour is similar to that of @p=@ E. For E = E , T T , the system is completely polarized with p = 1 and h = "2 , "1 , E = constant so that the heat capacity is zero. Thus h is continuous at the transition but the constanteld heat capacity has an in nite discontinuity there. The transition is second-order and the lines E = E1 T are critical curves. This treatment of behaviour near the critical curves applies to both ferroelectric and antiferroelectric, but the form of the critical curves is di erent for the two types of model. We now specialize to the ferroelectric, where ! 1 , 0 as T ! Tc + 0. Hence E1 = 0 for T = Tc so that the mirror image critical curves meet at the ferroelectric critical point on the T -axis see Fig. 10.10. There is, therefore, a second-order transition on any line along which E has a non-zero constant value, and the transition temperature increases with jEj. The rst-order transition, with latent heat, on the E = 0 axis see Sect. 10.10 is thus an exceptional case. From the form of the curve of f against p when ,1 1 see Fig. 10.9, the relative polarization p increases steadily with E from p = 0 at E = 0 to p = 1 at E = E1 T for any temperature T Tc. For T Tc, p = 1 when E 0 and p = ,1 when E 0 so that the segment 0 T Tc of the E = 0 axis in the T; E plane is a rst-order transition line. This line bifurcates into the critical second-order transition curves E = E1 T at the point Tc ; 0. The latter thus resembles a bicritical point Aharony 1983, a type of multicritical point we have not previously encountered. However, the situation is abnormal in that inside the ordered regions of the phase plane there is a frozen state of perfect order with p = 1.
The con gurational enthalpy per site is then given by hT + T; E = ,T 2 @ f T =T ;
@T
10.13 An Antiferroelectric in an Electric Field

We consider the case E 0, p 0. From 10.90 the part of the phase diagram in the lower half of the T; E plane is the mirror image of that in the upper half. We rst look at the situation when T = 0. In the perfectly ordered antiferroelectric state see the end of Sect. 10.3, the con gurational energy and polarization are zero and hence the con gurational enthalpy per site, h = u , pE = 0 for any E. For a completely polarized state with all vertices of type 1, p = 1 and h = h1 = u , pE = "2 , "1 , E : 10.144
10.13 An Antiferroelectric in an Electric Field
323
E Ec
p=1 p=0
0 p 1
Fig. 10.11. Transition
curves in the temperature electric eld plane for an antiferroelectric.
p=0
Tc0
,Ec
p = ,1
,1 p 0
Recalling that "2 , "1 0 for an antiferroelectric case iii of Sect. 10.2 we have h1 0 for E Ec and h1 0 for E Ec where Ec = "2 , "1 : 10.145 Since the equilibrium state at T = 0 is that of minimum h, the state of perfect sublattice order is stable for E Ec , and the completely polarized state for Ec E. This behaviour is similar to that of the Ising antiferromagnet at T = 0, but there are important di erences for T 0, as we shall see. For an antiferroelectric, the parameter varies from 1 at T = 1 to 2 ,1 at T = 0. Hence, by 10.132, the critical curve for transitions to the p = 1 state never reaches the E = 0 axis. It can easily be deduced from the inequality "2 "1 0 that E1 T , "2 , "1 ' 1 T exp,"1=T as T ! 0 . 10.146 2 So E1 = "2 , "1 and dE1 =dT = 0 at T = 0. Thus, by 10.145, the critical curve terminates at the critical eld point E = Ec on the axis T = 0 and the tangent there is horizontal see Fig. 10.11. When T Tc0 , where Tc0 is the antiferroelectric critical temperature given by 10.110, ,1 and the form of the curve of f T; p against p is the same as for a ferroelectric when T Tc see Fig. 10.9. So p increases continuously from 0 at E = 0 to 1 at E = E1T . However, when T Tc0, ,1 and the f T; p curve has the cusped form shown in Fig. 10.9. At p = +0, @f=@pT = T where T , given by 10.127, is positive and is the least value of the slope of the f T; p curve in the range 0 p 1. Thus for E T there is no solution to the equilibrium relation 10.88. It is easy to see why by considering the form of the non-equilibrium constanteld free energy f T; E; p, de ned by 10.87.
324
When 0 E T , the term ,Ep makes f T; p , Ep unsymmetrical about p = 0 but its least value is still at the tip of the cusp, that is at p = 0. It follows that the equilibrium value of p is zero, as it is for E = 0. As E increases through T , the least value of f T; p , Ep moves from p = 0 to a minimum in the range 0 p 1. There is thus a region of the T; E plane, bounded by the transition curve E = T and its mirror image E = , T , in which p = 0 see Fig. 10.11. Now, for an antiferroelectric, 2"2 , "1 "1 . Hence, from 10.28 and 10.112, ' 2"2 , "1 =T and ' "1 =T as T ! 0. From 10.127 it follows that 0 = "2 , "1 = Ec . The part of the T = 0 axis bounding the p = 0 region of the T; E plane coincides with the range of E for which a state of sublattice order is stable at T = 0. It is thus reasonable to regard the p = 0 region as that of sublattice order, with vertices of type 5 situated preferentially on one sublattice and vertices of type 6 on the other. This ordering is perfect only at T = 0; there is no frozen state like that of an ferroelectric below its critical temperature. The existence of a region in which p = 0 for E 6= 0 is linked to the fact that, as we saw in Sect. 10.3, the perfect sublattice ordering admits small perturbations which carry zero polarization. It can be shown from 10.112 and 10.128 that T exp ,C Tc0 , T ,1=2 as T ! Tc0 , 0 , 10.147 C being a constant. It follows that all derivatives of T , as well as T itself, approach zero as T ! Tc0 , 0. The boundary E = T of the p = 0 region in the T; E plane thus has an in nitely sharp cusp at T = Tc0 on the T -axis. The weakness of the singularity at the zeroeld critical point is presumably connected with the very rapid decrease in the width of the ordered region as T ! Tc0 , 0. This contrasts with the Ising antiferromagnet where the boundary of the sublattice ordered region in the T; H plane has a vertical tangent at T = Tc , H = 0 see Sects. 4.3 and 9.4. Also, for the Ising antiferromagnet, m 6= 0 for all T 0, H 6= 0 so that there is no analogue of the zero-polarization region. On the p = 0 side of the curve E = T in the T; E plane the equilibrium free energy is f T; 0 while all thermodynamic functions depend on T only and have no singularities for T Tc0. On the other side where 0 p 1, a Maclaurin expansion yields, using results from Sect. 10.11, 1 f T; p = f T; 0 + T p + 3 3T p3 + Op4 ; 3 T 0 : 10.148 Thus, from 10.88, the equilibrium value of p satis es the relation E = T + 3T p2 + Op3: 10.149 These relations are of the same form as 10.134 and 10.135, and by arguments like those in Sect. 10.12 it can be shown that the thermodynamic functions behave similarly on the disordered sides of the respective curves E = E1 T and E = T . Hence E = T is a line of second-order transitions or critical curve. In particular the con gurational enthalpy is continuous
10.14 The Potts Model
325
at any point T ; E on the curve E = T , but the heat capacity diverges as T , T ,1=2 when the point is approached on a line E = E 0. The continuity of the heat capacity along the line E = 0 is thus exceptional. Hitherto we have assumed "1 0, but now we consider the F model, which is an antiferroelectric with "1 = 0. When "1 = 0, X , 1 = 0 for all T 0 and 10.132 gives E1 T = 1 so that the transition lines E = E1 T are absent in the F model. From Table 10.1 the KDP and F models look like the simplest forms of ferroelectric and antiferroelectric respectively. However, whereas the KDP model is a typical ferroelectric, the F model is an untypical antiferroelectric. The reason is that the condition "1 = 0 puts it on the symmetry line for two-dimensional ice-rule models see Fig. 10.4. For any value of E, vertices 1 and 4 are equally probable as are vertices 2 and 3. So, as we have just seen, there is no transition to a completely polarized state at any nite E because the system is unable to `choose' between the alternative con gurations available. A rather similar situation exists for the zeroeld IF model, see Example 10.3, where no transition of any kind occurs. By symmetry the reaction of an ice-rule model to a horizontal eld is the same as to a vertical eld. The phenomena associated with a eld with both a horizontal and a vertical component are interesting but too complicated to discuss here Lieb and Wu 1972.

The Potts model was de ned in Example 2.2. In this section we consider the anisotropic Potts model on a square lattice N of N sites. With the parameter R of Example 2.2 taking di erent values Rh and Rv for nearest-neighbour horizontal and vertical pairs respectively, the Hamiltonian can be expressed in the form
b H Rh ; Rv ;
i = ,Rh
hX : :n:n
10.150 where the rst and second summations are over all nearest-neighbour horizontal and vertical pairs respectively and Kr is the Kronecker delta function. The variable i at any lattice site takes one of distinct integer values. In this section we demonstrate, using the method of Baxter 1973b, 1982b, the equivalence between this anisotropic square lattice -state Potts model and a staggered six-vertex model, with di erent vertex weights on the two equivalent interpenetrating sublattices of the square lattice. This equivalence was rst demonstrated by Temperley and Lieb 1971.
fi;jg
Kr
i , j , Rv
vX : :n:n
fi;jg
Kr
i , j ;
326

Polygon lines
Fig. 10.12. Polygon con-
a
b
struction on N white sites0 from the medial graph N black sites. a corresponds to the cases where a heavy black line of g passes through the site of N 0 and b where it does not.
10.14.1 The Staggered Six-Vertex Model

It is not di cult to show that the partition function with Hamiltonian 10.150 is
where the summation is over all N points in the space of values of the variables i and p p 10.152 v = expRv =T , 1 = : h = expRh =T , 1 = ; This form for the partition function is similar to the high-temperature series form for the Ising model in Sect. 8.3 and, as in that case, a binomial expansion of the products in 10.151 gives a sum of terms. Each term corresponds to a subgraph g of the full lattice graph N . The subgraph is obtained by connecting, by a line, each nearest-neighbour pair with microsystems in the same state. Suppose that g has qh horizontal lines, qv vertical lines and c components. Then the degeneracy associated with g will be c and the partition function can be expressed in the form X q q Z h ; v = c+qh +qv =2 h h v v : 10.153 The medial lattice N 0 of N is de ned to be the set of sites given by placing a site at the centre of each nearest-neighbour pair of N . This new square lattice, which has 2N sites, divides in a natural way into two sublattices, of N sites each, denoted by h and v, with the sites classi ed according to whether they respectively lie on a horizontal or vertical line of N . From the nearest-neighbour lattice lines of N 0 we now construct a set of polygons in the following way. If a line of g passes through a site of N 0 the lattice lines are moved as shown in Fig. 10.12a and otherwise they are moved as
fgg
8 :n: 9 = X hYn: p Kr Z h; v = : 1 + h i , j ; fi;jg 9 8v:n:n: Y p Kr , = ; : 1+ v i j ; fi;jg
10.151
327
shown in Fig. 10.12b. The nal e ect of this procedure is to give a set of closed polygons. For a nite lattice without toroidal boundary conditions the number p of polygons constructed in this way is given by p = 2c + qh + qv , N : 10.154 To see the truth of equation 10.154 begin with the extreme case where no members of a nearest-neighbour pair on N are in the same state. Then c = N , q = qh + qv = 0 and p = N . Any subgraph g can be constructed from this extreme situation by a systematic alteration in the states of the microsystems on N . Suppose a particular site is in a state di erent from all its neighbours. If these neighbours are themselves all in di erent states then changing the centre site to one of these will decrease c and p by one and increase q by one which leaves 10.154 satis ed. If two neighbours are in the same state and the centre site is changed to this state q is increased by two. If the two neighbours belonged to the same component one polygon would be destroyed and one created leaving p unchanged with c decreased by one. If the two neighbours belonged to di erent components c and p would both be decreased by two. In each case 10.154 continues to hold. The reasoning for three and four neighbours in the same state is left as an exercise for the reader. It is clear that any subgraph g corresponds to a polygon decomposition of the type described above and, using 10.154, the partition function 10.153 can be re-expressed in the form 10.155 where the summation is over all polynomial decompositions. The factor qh qv h v in 10.155 is `local' in the sense that it is made up of a contribution of either 1 or h from every site of h and 1 or v from every site of v. We now 2 need to reduce the factor 1 p to a product of local contributions. Arrows are placed on the edges of all the polygons so that around any polygon all the arrows are pointing in the same direction. The arrow con gurations at each site of N 0 obey the `two in, two out' ice rule of the six-vertex model Sect. 10.1. Now consider the partition function 10.156 where the inner summation is over all allowed arrow con gurations and the product is over all polygon corners, with !0 = ! if, following in the direction of the arrows, the corner represents a 90 turn to the left and !,1 if it represents a 90 turn to the right. It is clear that the e ect of the new summation and product in 10.156 is to give a factor !4 + !,4 for every polygon and the partition functions 10.155 and 10.156 are equal if 2 10.157 1 = !4 + !,4 :
2 Z h ; v ; ! = 1 N 1 Z h; v = 2 N
p:d: X
2 q q 1 p h h v v;
p:d: a:c:
p:c: XX Y
q q !0 h h v v ;
328
Consider a site of sublattice h. There are eight possibilities for the con guration of arrows and the presence or not of a line of g. If the polygons are reconnected at the site, there are six vertex con gurations with the weights Zh 1 = Zh2 = h ; Zh 3 = Zh 4 = 1 ; 10.158 Zh 5 = !2 h + !,2 ; Zh 6 = !,2 h + !2 : The ordering of the vertex con gurations is that of Fig. 10.2. A similar analysis of a site of v yields Zv 1 = Zv 2 = 1 ; Zv 3 = Zv 4 = v ; 10.159 Zv 5 = !,2 v + !2 ; Zv 6 = !2 v + !,2 : The two summations in 10.156 are together equivalent to one summation over arrow con gurations at the sites of N 0 which satisfy the ice rule and Z h ; v ; ! = !4 + !,4 N Zs6v h ; v ; ! ; 10.160 where Zs6v h ; v ; ! is the partition function on the medial lattice of 2N sites of a model which di ers in two important respects from the standard six-vertex model: i The model is a staggered six-vertex model with di erent vertex weights on the two sublattices h and v. ii The vertex weights for vertices 5 and 6 are di erent, unless ! = 1 = 4. For the lattice as a whole, with toroidal boundary conditions, we know that the numbers n5 and n6 of vertices 5 and 6 are equal. It is, however, not in general the case that n5h = n6h , n5v = n6v and the inequality of the vertex weights 5 and 6 for each individual sublattice is a genuine generalization of , , the model. Replacing h and v by h 1 and v 1 in 10.158 and 10.159 is equivalent to dividing the weights in 10.158 by h and in 10.159 v and interchanging the sublattices. It follows that the partition function satis es the duality relationship , , Z h ; v ; ! = h v N Z h 1 ; v 1 ; ! ; 10.161 which maps the model between high and low temperatures. Except in two special cases the staggered six-vertex model with weights 10.158 and 10.159 has not been solved. Let 4 4 h = h !+ +41 ; v = v !+ +41 : 10.162 h ! v ! In terms of these variables, it was shown by Baxter 1982b that the Bethe ansatz can be extended to solve the cases with constraints h v = 1 ; 10.163 which we now treat separately
329
10.14.2 The Solvable Case hv = 1

The constraint here reduces to 10.164 h v = 1; which was the case considered by Baxter 1973b. The vertex weights 10.158 and 10.159 are proportional to each other and with h and v positive the model is equivalent to the regular six-vertex model. The corresponding Potts model is ferromagnetic with its temperature xed by the condition = expRh =T , 1 expRv =T , 1 : 10.165 When Rh = Rv 10.165 gives the critical temperature of the ferromagnetic Potts model on a square lattice. That this is the case can be seen by imposing the isotropy condition = h = v on 10.161 giving Z ; ! = 2N Z ,1 ; ! : 10.166 The mapping ! ,1 is the duality transformation rst obtained by Potts 1952 for his model. Given that the ferromagnetic Potts model has a phase transition between a low-temperature ordered state and the hightemperature disordered state and assuming that there is a single phase transition on the T -axis, it will correspond to a singularity in the partition function, which from 10.166 must occur at = 1 which gives the critical temperature p Tc = R= ln + 1 ; R 0: 10.167 For = 2 and R = 2J this reduces to the formula 8.20 for the critical temperature of the square lattice Ising model. At the critical temperature the staggered nature of the equivalent six-vertex model disappears. From 10.158 and 10.159, p a = b = 1; c = !2 + !,2 = 2 + 1=2 10.168 and from 10.28 p = , =2 : 10.169 From Table 10.1, 10.3 and 10.168 "1 = "2 = 0, "3 0. By subtracting "3 from each vertex energy we see that this is the case of an F model. At its critical temperature the -state ferromagnetic Potts model can now be interpreted with variations of the parameter playing the role of temperature variation. In doing so we must remember that the six-vertex model is on the medial lattice of 2N sites. The Range 0 4. From 10.169 this gives ,1 1, which is the disordered region considered in Sects. 10.8 and 10.9. From 10.52 and 10.73 ' = 0 and, from 10.62, is given by p = 2 cos ; 0 1 : 10.170 2
330
In deriving the free energy per site of the Potts model at the critical temperature from 10.76 and 10.80 we must remember that the corresponding six-vertex model has 2N lattice sites and we must also include the factor 2 , 1 ln which arises from the prefactor of !4 + !,4 N = 1 N in 10.160. 2 This nally gives Z 1 sinh y , tanhy f Tc = , 1 Tc ln , 2Tc dy : 10.171 2 y sinhy 0 The Range 4. From 10.169 this gives ,1, which is the disordered region considered in Sect. 10.11. From 10.52 and 10.112 0 = 0 and is given by p = 2 cosh ; 0: 10.172 From 10.120 and 10.123, 1 X f Tc = , 1 Tc ln , Tc , 2Tc exp,n tanhn : 10.173 2 n n=1
The Boundary = 4. In this case the free energy can be obtained by taking the limit ! 0 in equation 10.171 or ! 0 in 10.173. Using 10.167, the result is
f Tc = ,Tc ln2 , 2Tc

8R = , ln3 ln
Z1
;
5 2 28 1 Gradshteyn and Ryzhik 1980, p.327.
" ,1 0 ,
4
exp,y tanhydy ; 10.174
10.14.3 The Solvable Case hv = ,1

Using 10.157, the constraint here reduces to p p h + 2 v + 2 = 4 , ; 10.175 which, from 10.152, xes the temperature of the anisotropy Potts model with the condition 4 , = expRh =T + 1 expRv =T + 1 : 10.176 Although it is of interest to consider non-physical situations Baxter 1982b, 10.176 has a physical solution only when 3 and then only when at least one of Rh and Rv is negative. When Rh = Rv = R 0 10.176 gives the critical temperature p Tc = R= ln 4 , , 1 ; R 0: 10.177 of the antiferromagnetic square lattice Potts model. For = 2 and R = 2J this formula gives the result 8.20 for the antiferromagnetic Ising model.
331
For = 3 the critical temperature of the antiferromagnetic 3-state Potts model is Tc = 0. Unlike the ferromagnetic case the staggered nature of the underlying six-vertex model does not disappear at the critical temperature and evaluation of the free energy in this case is rather more complicated Baxter 1982b.
10.14.4 The Polarization and Internal Energy

It is clear from the above discussion that the -state Potts model di ers in its behaviour according to whether is greater of less that four. One way to understand more clearly the e ect on the nature of the phase transition is to calculate the internal energy as the critical point is approached from above and below. For the ferromagnetic case this has been achieved by Baxter 1973b. In our present notation 1 u = ,R1 + , 2 1 p0 tanh=2 ; 10.178 where p0 is the spontaneous polarization and the + and , signs in 10.178 apply according to whether the critical point is approached from below or above. The spontaneous polarization was calculated by Baxter 1973a, 1973c and is zero when 0 4 with
p0 =
1 Y
Twice the second term in 10.178, therefore, represents the latent heat at the transition for 4, the transition being rst-order. In all cases the rst term in 10.178 gives the internal energy per site averaged over the values below and above the transition. This was rst calculated by Potts 1952.
n=1
tanh2 n ;
4:
10.179
and independently by Nienhuis et al. 1980 and Pearson 1980 that +2 = 12 ; 10.181 where is given by 10.170. Using the scaling laws 3.77 and 3.87 this gives 2,8 +4 2 2, + 2 = 7 6 2 ; = 15 2 16 2 4 : 10.182 , , 4 The formula 10.180 for has now been establish by Black and Emery 1981 by further exploiting the relationship to the staggered F model. From
10.14.5 Critical Exponents For 0 4 it has been conjectured by den Nijs 1979 that = 2 , 8 3 , 6
10.180
332
10.170, with = 2, = =4 and the exponent values for the Ising model, derived in Sect. 8.9, are recovered. When = 3, = =6 = 1 ; = 13 ; = 14: 10.183 = 1; 3 9 9 These values are strongly supported by the nite-size scaling analysis of Blote and Nightingale 1982 and by conformal invariance arguments Friedan et al. 1984, Cardy 1996, see also Volume 2, Sect. 2.17. They are the same as those of the hard hexagon model Baxter 1982a and support the argument of Alexander 1975 that the 3-state Potts model and hard hexagon model lie in the same universality class see Volume 2, Sect. 5.12.
Examples
10.1 With de ned by 10.28, show that: a For case iii of Sect. 10.2, increases steadily with T from ,1 1 at T = 0 to 2 at T = 1. 1 b For case i of Sect. 10.2, with "2 2 "1 , decreases steadily, 1 at T = 1. from +1 at T = 0 to 2 1 c For case i of Sect. 10.2, with "2 2 "1 , decreases steadily from +1 at T = 0 to a value min , where 0 min 1 , and 2 1 then increases to 2 at T = 1. 10.2 From 10.28 and 10.52, show that, given "1 0, X , 0 ; 1 + X , 2 0 ; L0 0 ; for all T . From 10.51 show that jLkj2 = 1 + 41X+,X cosk ,k : 2 , 2X cos Hence show that jLkj decreases as k increases. For ,1 1, show that jLkj 1 for 0 k , and jLkj 1 for , k , where is de ned by 10.62. For ,1, show that jLkj 1 for all k. 10.3 Using a diagram similar to Fig. 10.8, draw trajectories for the F, IKDP and IF models, showing that no transition occurs in the two latter models with T 0. Sketch trajectories for models where A "2 = "1 , "1 0; B "2 = 2"1, "1 0; C "2 = 2"1, "1 0. Verify that A, B and C belong respectively to the categories i, ii and iii of Sect. 10.2. Show that, for A, B and C, the critical p temperature is j"1 j= ln 5 + 1=2 .
Examples
333
10.4 For an antiferroelectric with "1 0, show that , lim0 , 2"2 T "1 = 0 ; lim0 , "1 = 0 ; T! T! T
where and are de ned by 10.112. Hence show, using 10.58 with x0 = and p = 0 and 10.121, that f T ! 0 as T ! 0. Also show, using 10.58 and 10.61 with jexj given by 10.116, that L f T; p = "2 , "1 jpj at T = 0 . Hence show that f O; E; p = "2 , "1 jpj , Ep = "2 , "1 , jEjjpj is minimized by p = 0, for ,Ec E Ec, by p = 1, for Ec E, and by p = ,1, for E ,Ec , where Ec = "2 , "1 . 10.5 Show that equation 10.132 can be expressed as 1 T ln 1 , exp,"1=T + expf2"2 , "1 =T g : E1T = 2 exp"1 =T , 1 Derive the relation dE1 = E1 + "1 exp"1 =T dT T 2T exp"1 =T , 1 "2 1 exp 2 , =T + "1 exp , 22T ,1"+exp 2""2, ""1=T , exp," ,"1=T f 2 1 1 =T g and hence show that for an ferroelectric "1 0, "1 "2 , dE1 =dT 0 for T Tc. Show that, for a ferroelectric at T = Tc, dE1 = T ,1f" , " 1 , exp," =T g : 1 2 1 c c dT Prove that an alternative expression for E1 is E1T = "2 , "1 "2 =T 1 + 2 T ln 1 + exp "11, 2exp,", exp,2"2=T , 1 =T and hence that E1 T "2 ,"1 for all T 0. Derive equation 10.146 for the antiferroelectric "2 "1 0 and show that dE1 =dT 0 for all T 0 by di erentiating the alternative expression with respect to T . 10.6 Consider the parameter value = 0, a2 + b2 = c2 which corresponds to a disordered state of the antiferroelectric. Show that q1 ; q2 , given by 10.36, is zero for all q1 and q2 , k = 1=2 and the limits of integration q0 in equation 10.54 are given by 1 q0 = 2 1 , p. Hence show that
334
Prove that the relative vertical polarization p and the vertical eld E are connected by 2 2 1 3 X + 1 + 2X sin 2 p 5 1 E = 4 T ln 4 2 : X + 1 , 2X sin 1 p 2 Verify 10.132 and 10.89 for the case = 0. Putting p = 1 , p, show that, as p ! 0, 2 1 E1 , E ' 3 T2 XXX ,+ 1 p2 : 2 12
T Z 21,p=2 ln X2 + 1 + 2X cosk dk : f T; p = "2 , "1 , 2 X2 + 1 , 2X cosk

0
A. Appendices
A.1 Regular Lattices

This appendix contains diagrams of the one-, two- and three-dimensional regular lattices. Sites are shown by black or white circles and nearest-neighbour pairs are connected by line segments. In graph theory language the sites are the vertices and the segments are the edges of what is called the full lattice graph.1 Throughout the book d denotes the dimension of the lattice and z the coordination number, which is the number of nearest neighbours of a lattice site. The body-centred cubic lattice of Fig. A.7 can be obtained from the simple cubic of Fig. A.6 by placing a new site at the centre of each cube of eight sites. The face-centred cubic lattice of Fig. A.8 can be obtained from the simple cubic by placing a new site at the centre of each square of four sites. The diamond lattice of Fig. A.9 can be obtained from the body-centred cubic by eliminating alternate sites. A loose-packed lattice is one which can be divided
Fig. A.1. The linear lattice; d = 1, z = 2.
Fig. A.2. The square lattice; d = 2, z = 4.

into two equal sublattices a and b in such a way that all the nearest neighbours of a site of one sublattice belong to the other sublattice; a close-packed
1
For detailed accounts of graph theory terminology see Temperley 1981 and Volume 2, Appendix A.7.
336
A. Appendices
Fig. A.3. The triangular lattice; d = 2,

z = 6.
Fig. A.4. The honeycomb; d = 2, z = 3.
Fig. A.5. The kagom lattice; d = 2, e

z = 4.
Fig. A.6. The simple cubic lattice;

d = 3, z = 6.
Fig. A.7. The body-centre cubic lattice; d = 3, z = 8.
A.2 Elliptic Integrals and Functions
337
Fig. A.8. The face-centred cubic lattice; d = 3, z = 12.
Fig. A.9. The diamond lattice; d = 3,

z = 4.
lattice is one for which this is not possible. The linear, square, honeycomb, simple cubic, body-centred cubic and diamond lattices are loose-packed and the triangular, Kagom and face-centred cubic lattices are close-packed. In e the diagrams of the loose-packed lattices the sites of sublattice a are denoted by and those of sublattice b by . Nearest-neighbour pairs of sites are linked by full line segments and second-neighbour aa pairs by broken line segments. The corresponding bb links are not shown.

Certain results in the theory of elliptic integrals and functions are needed for the treatment the Ising model in Chap. 8 and these together with additional material on nome series are needed for the discussion of the eightvertex model in Volume 2, Chap. 5. For convenience a brief discussion of this material is presented in this appendix and further results are given in Volume 2, Appendix A.6. For a extensive treatments the reader is referred to Bowman 1953, Lawden 1989 or Baxter 1982a Chap. 15. Comprehensive lists of formulae are given by Gradshteyn and Ryzhik 1980 and Abramowitz and Segun 1965.
338
A. Appendices
A.2.1 Elliptic Integrals

Elliptic integrals of the rst and second kind, respectively, are de ned by Z ; F jk = q d! 0 1 , k2 sin2 ! A.1 Z q 2 ! d! ; 1 , k2 sin E jk =
where 0 k 1 is the modulus. Complete elliptic integrals of the rst and second kind, respectively, are de ned by
1
1 Kk = F 2 jk ; Ek = E 2 jk : A.2 Alternative forms for F jk and E jk are given by the substitution x = sin!, x0 = sin. Then Z x0 p1 , x2dx , k2x2 : F jk = A.3 1 0 From A.1 and A.2 K0 = E0 = 1 . By binomial expansion of the 2 integrands
2
1 2 Kk = 1 + 1 k2 + 2::3 k4 + ; 2 2 4 A.4
1 2 1
1:3 2 1, 2, 4 + : Ek = 2 2 k 3 2:4 k As k ! 1, Kk ! 1. It can be shown Bowman 1953, pp. 21 and 22 that Kk = ln 4 + k ; A.5 where k ! 0 as k ! 1 and k0 is the complementary modulus de ned by p k0 = 1 , k2 : A.6 It also follows from A.1 and A.2 that E1 = 1 : A.7 From A.1 and A.2, by di erentiating under the integral sign, dKk = Ek , k02 Kk ; dEk = Ek , Kk : A.8 02 dk kk dk k Interchanging k and k0 and using A.6 gives dKk0 = k2 Kk0 , Ek0 ; dEk0 = k KM 0 , Ek0 : A.9 02 dk kk dk k02 From A.8 and A.9 it can be veri ed that Kk and Kk0 are solutions of the di erential equation
k0
d 02 A.10 dk kk dk = kf: The general solution is therefore

Kk+ Kk0 for arbitrary constants
and . An important formula relating complete elliptic integrals is the Legendre relation EkKk0 + Ek0Kk , KkKk0 = 1 : A.11 2 That the left-hand side is indeed a constant can be established from A.8 and A.9. That the value of the constant is =2 can be established by taking the limit k ! 0 and using A.4, A.5 and A.7. There are a number of formulae connecting elliptic integrals of di erent moduli. In Chaps. 8 and 10 we need the Landen transformation given by p 2 k1 = 1 + k ; A.12 k for which Kk1 = 1 + kKk : A.13 From A.6 and A.12, , 0 k1 = 1 + k A.14 1 k and 1 0 A.15 Kk1 = 2 1 + kKk0 :
df
339
A.2.2 Elliptic Functions

In A.3, F is de ned as a function of x0 . Writing u = F arcsinx0 jk A.16 the inverse relation de nes the Jacobian elliptic function sn. That is to say x0 = snujk : A.17 The associated Jacobian elliptic cn and dn functions are de ned by p p cnujk = 1 , sn2 ujk ; dnujk = 1 , k2 sn2 ujk : A.18 From A.3, A.16 and A.17 snuj0 = sinu ; snuj1 = tanhu A.19 and thus, from A.18, cnuj0 = cosu ; cnuj1 = sechu ; A.20 dnuj0 = 1; dnuj1 = sechu: A.21
340
A. Appendices
In the limit k ! 0 the elliptic functions reduce to trigonometric functions and in the limit k ! 1 to hyperbolic functions. A generalization of the duplication formulas for sin and tanh is k j dn sn2ujk = 2snuj,cnu2kujk ujk : A.22 1 k2 sn The elliptic integral of the second kind, de ned in A.1, can be regarded as a function E ujk of u through the relation = arcsinx0 = arcsin snujk . It then satis es the duplication formula E 2ujk = 2E ujk , k2 sn2 ujksn2ujk : A.23
A.2.3 Results Required for Chapter 8

Many integrals can be reduced to standard forms involving elliptic integrals. p These include those with integrand 1= Y y where Y y is a quartic in y, positive over the range of integration with respect to y. The substitution y = tan in equations 8.90 gives, when k 1, AK jk = F jk0 ; A.24 B K jk = k0,2 F jk0 , E jk0 ; where tan = sinh2K . When k 1, an alternative treatment is needed. Putting ` = 1=k and y = ` tan AK jk = `F j`0 ; A.25 B K jk = ,C K jk + E j`0 , `2 F j`0 =`02 ; when tan = `,1 sinh2K = k sinh2K ;
`02 = 1 , `2 ; : C K jk = q tanh2K 1 + k2 sinh2 2K If k is given by 8.74 then C K jk = tanh2K tanh2G : Two other integrals needed in Chap. 8 are Z1 2 p1 , x2 dx, x , x = p ,Kk + 1 ; 1 ,1 where
A.26
A.27 A.28
341
and are real and the intervals ; and ,1; 1 are disjoint, and Z1 p1 , x2 dx+ x + x = q 12Kk ; A.30 ,1 + 2 where 1 2 2 = + 1 + 2 + 2 , ; k + A.31 p 1 + + + 8 = 2 and and are conjugate complex numbers with positive real part. For the thermodynamic functions of the triangular and honeycomb lattices we need relations between AK jk and AGjk, and between B K jk and B Gjk, where K and G are connected by the star-triangle formula 8.27 and k is given by 8.73. Writing AK jk simply as A, it follows from A.24, A.16 and A.17 that snAjk0 = sin = tanh2K A.32 and thus, from A.18 and 8.73, cnAjk0 = sech2K ; A.33 q A.34 dnAjk0 = 1 , k02 tanh2 2K = sech2K coth2G : Substituting from A.32 A.34 into A.22 and using 8.73 A.35 sn2Ajk0 = 2 cosh2G sinh2G tanh2K : sinh2 2G + tanh2 2K cosh2 2G Now the star-triangle relation 8.27 can be expressed in the form tanh2K = cosh2GG, 1 A.36 cosh2 and substituting in A.35 we have sn2AK jkjk0 = tanh2G ; A.37 which is equivalent to 2AK jk = F 1 jk0 ; A.38 where tan1 = sinh2G and sin1 = tanh2G. From A.24 2AK jk = AGjk : A.39 From A.24, A.32, A.37 and A.38
, k2 = 2 1 + 1 ; ,
A.29
342
A. Appendices B Gjk = k0,2 2A , E 2Ajk0 ;
Then, from A.23, B Gjk = 2B K jk + tanh2 2K tanh2G : A.41 The derivations of A.39 and A.41 are valid only when k 1. It may, however, be shown using a similar method and A.25 that the results are also valid for k 1.
B K jk = k0,2 A , E Ajk0 :
A.40
A.3 The Water Molecule and Hydrogen Bonding

The hydrogen H nucleus is a single proton and the ground state of the single electron is designated 1s, that is principal quantum number unity and angular quantum number zero. The oxygen O nucleus has an eight proton charge, and of the eight electrons two are in the 1s state with, by the Pauli exclusion principle, oppositely directed spins and form an ìnner layer'. There is an òuter layer' of six electrons which have principal quantum number two, two electrons being in the 2s state and four in the 2p state, with angular momentum quantum number unity. When chemical combination takes place between oxygen and hydrogen atoms, the wave function solutions of Schrodinger's equation for the original atoms are, of course, no longer valid. The resulting situation can be represented approximately by regarding the 1s orbital of each hydrogen atom as mixing or hybridizing with one of the oxygen 2p orbitals to form a molecular orbital occupied by two electrons with opposite spins. A covalent bond is then said to exist. In the water molecule there are two covalent O H bonds with the two vectors from the oxygen nucleus to the hydrogen nuclei each of length 0:957 10,10 m and at an angle of 104:5 to each other. Owing to the shifting of the negative charge probability density, which surrounded the hydrogen nuclei in the atomic state, there is now a resultant positive charge in the neighbourhood of each hydrogen nucleus. The two covalent bonds take up four electrons, but the wave functions of the four remaining electrons from the oxygen outer layer must also be modi ed. They now occupy two socalled lone-pair orbitals, the electrons of each pair having opposite spins. The charge probability distribution due to each lone pair has its densest part in the form of a `lobe' symmetrically distributed with respect to the plane bisecting the H O H angle and on the opposite side of the oxygen nucleus from the hydrogen nuclei. The two negatively charged lobes are mirror images of each other in the plane of the oxygen and hydrogen nuclei, and the angle between them is about 120 . The two negative and two positive outer charges in the water molecule can thus be picturesquely regarded as situated on four àrms' radiating out from
A.3 The Water Molecule and Hydrogen Bonding
343
the oxygen nucleus in roughly the directions of the vertices of a tetrahedron with the oxygen nucleus at its centre.2 From this description of the water molecule it is obvious that it is highly polar, with signi cant electric dipole and quadrupole moments. What is not so obvious is that speci c and directed links, called hydrogen bonds, can form between water molecules. Such a bond is created when the oxygen nuclei of two molecules are about 2:75 10,10 m apart and positive arm of one molecule and a negative arm of the other are approximately directed along the line segment connecting the oxygen nuclei. Each bond therefore involves one hydrogen nucleus which is situated roughly in the line of the two oxygen nuclei, but is much nearer the oxygen nucleus with which it has a covalent link. Since the positive hydrogen nucleus of one molecule is close to the negative lone-pair charge region of the other molecule, the hydrogen bond energy contains a large negative electrostatic term. There is also a covalent or electron delocalization term, due to distortion of the molecular electronic wave functions, and a dispersion interaction term, both of which are negative. They are partially o set by a positive term due to overlap or repulsive forces Rao 1972. Hydrogen bonds owe their stability to the very small size of a hydrogen atom when partially stripped of its single electron. They can form between other molecules containing hydrogen atoms and electronegative regions, for instance KH2 PO4. Any water molecule can form four hydrogen bonds with other water molecules, two of which involve its own hydrogen atoms. It is therefore possible for a completely hydrogen-bonded water crystal to have a tetrahedrally coordinated array of oxygen nuclei. This means that any oxygen nucleus has four nearest-neighbour oxygen nuclei situated at the vertices of a tetrahedron with itself at the centre. From Appendix A.1, the oxygen nuclei could then be on the sites of a diamond-type lattice, and this is the case in ice Ic, which is a metastable form occurring between ,120 C and ,140 C. The structure is an open one with a good deal of empty space. However, the form called ice I, which is stable under normal conditions, has a wurtzite-type lattice, which is also tetrahedrally coordinated. Diamond has cubic symmetry, and the sites can be regarded as arranged in layers perpendicular to an axis which makes equal angles with the three cubic directions indicated by the broken lines in the diamond lattice diagram of Appendix A.1. These layers are composed of puckered rings which are like the hexagons of sites in a honeycomb lattice except that, owing to the tetrahedral angles of the edges, the six sites of a ring are not coplanar. A shift in the relative position of successive layers changes the lattice to the hexagonally symmetric wurzite type. The density of ice I is nearly equal to that of ice Ic so it is an equally open structure, but the empty spaces now become shafts perpendicular and parallel to the layers. Again, the nearest-neighbour environments of a water molecule are the same
2
The vectors from the centre of the tetrahedron to the vertices make angles of 109:47 with each other. Note that in Fig. 6.2, either the four b1 sites or the b2 sites are the vertices of a tetrahedron with the a1 site as centre.
344
A. Appendices
in the two forms, the di erences starting with second neighbours Eisenberg and Kauzmann 1969, Fletcher 1970. A water molecule bonded tetrahedrally to four other water molecules can take up six orientations, depending on the pair of bonds where its own hydrogen atoms are situated. However, in a completely hydrogen-bonded network, any reorientation must be cooperative since there must always be exactly one hydrogen on each bond. Even with this restriction there are about 1:507M ways of placing the protons on the bonds of a network of M molecules, so that ice I has a residual con gurational reduced entropy per molecule, s = 0:410 Nagle 1966. There are a number of high pressure polymorphs of ice, in all of which each water molecule is still hydrogen bonded to four others. Ices II-V achieve a greater density than that of ice I by distortion of the hydrogen bonds. In ice II the positions of the protons, and hence the orientations of the molecules, are ordered and the residual entropy is absent. The highest density forms are ices VI, VII and VIII, which are composed of two intertwining fully-bonded networks. In ice VII the oxygen nuclei occupy the sites of a body-centred cubic lattice and are linked by hydrogen bonds into two diamond networks.3 To force the molecules of one bonded network into the interstices of another involves a good deal of work against repulsive forces and hence ice VII is stable only at pressures above about 22; 000 atm. Ice VIII is similar to ice VII except that the proton positions are ordered. In a condensed water phase the rupture of some of the hydrogen bonds decreases the amount of interstitial space and increases the density. So ice I at its melting point is less dense than the liquid with which it is in equilibrium. Ice I, therefore, oats on its melt, and its melting point temperature decreases as the pressure increases. However, just above the melting point temperature, large regions of ice-like open structure persist in the liquid. As the temperature increases, more hydrogen bonds are broken, and the open structure regions diminish in size. The consequent closer packing o sets the usual decrease in density and results in the density increasing with temperature until a maximum is reached. At 1 atm pressure this occurs at 4 C. Again, instead of the steady increase in compressibility with temperature observed in other liquids, the compressibility of water decreases until a minimum is attained, this being at 46 C for 1 atm pressure. These anomalous properties of uid water persist up to pressures well above the critical value of 218 atm but eventually disappear.
A.4 Results for the Six-Vertex Model

The proofs are presented for the results I IV given in equations 10.32, 10.34, 10.35 and 10.38 respectively.
3
In Fig. 6.2 it is possible to construct one diamond sublattice from the sites of the a1 and b1 face-centred cubic sublattices, while the a2 and b2 sites form the other diamond sublattice.
345
A.4.1 The Proof of I
Equation 10.31 is valid for all vectors of the type de ned in Sect. 10.4. In particular, if n N1 , X g1 = Az1 +1 z2 +1 zn+1 = z1 z2 zn g ; A.42 1 2 n and so, from 10.15 and 10.16, z1 zn =
:
fg
A.43
A.4.2 The Proof of II

Now suppose we choose 1 s 2 , then X , N g,s = Az2 ,s zn,1s zn1 +1 ,s 1 n =
fg X fg
N ARz11 +1 ,s z2 ,s zn,s ; 2 n
A.44
which, from A.42 and A.43, gives g,s =

,s g X N 2 = z11
,s ARz1 z2 zn : 1 n Comparing this with 10.31 we have , AR = z1N 1 A :
fg
A.45 A.46
There are two possibilities for the components of the vector 0 in 10.8: 0 0 i 1 1 1 ; j,1 j j ; j = 2; : : : ; n: 0 0 ii j j j+1 ; j = 1; : : : ; n , 1; n n N1; and, as for the case n = 1 see 10.21 we divide the left-hand side of 10.8 into two parts corresponding to these two cases. X X V ; 0 g = A i ; + ii ; ; A.47
0
A.4.3 The Proof of III
where
f0 g
fg
m 0 0 m ; = X V ; 0 z 1 z n ; 1 n f0 g
m = i; ii
A.48
346
A. Appendices
are, the respective sums restricted as indicated above. They are generalizations of i and ii de ned in 10.22 for the case n = 1. In order to avoid separate evaluations of i and ii we de ne the transformation 8 a ! b;
L=
and we have ii ; =

L
: z ! 1=z j n+1,j j ! N1 + 1 , n+1,j

i ; ;
b ! a;
A.49
A.50
LMzj = Lzn+1,j ; A.51 the functions Lz and Mz being de ned in 10.24. For simplicity we consider i I; , where I is the identity permutation. The summation in A.48 is a geometrical progression as in 10.22, and after summation we have an expression of the form i I; = bN ,n X H z H z ; A.52 1 1 n n where the sum is over the di erent forms for the functions Hj z given by 8 a=bb1 z1 ; 1 H1 z = : A.53 b1 z 1 Mz , a=b + a1 Lz , b=a 1 ; 8 a=bbj ,j,1 zj j ; bj ,j,1 z j Mz , a=b Hj z = + aj ,j,1 z j,1 Lz , b=a j ;
N1 j 1:
A.54
0 0 In i and ii the condition that neighbouring members of the set 1 ; : : : ; n cannot be equal must hold and this is ensured in A.52 by imposing the rule that an function cannot be followed on the right by a function. Every term in A.52 contains a particular set of factors aj ,j,1 , bj ,j,1 and we gather together those terms with the same set of factors to give i I; = X f I; Q pq1 aqr ,qr,1 bN1 ,qr ; A.55 r r
: b=aaj ,j,1 zj,1 j ;
where p = a or b according as r is even or odd. The summation in A.55 is over all subsets Qr = fq1 ; q2 ; : : : ; qr g of the integers 1; : : : ; n including the empty subset denoted by Q0. Let
fQr g
WI; j = zj+1 zj j Lzj+1 Mzj , 1 ; 81 j2 j1 , 81

j =j1
347
A.56 A.57
j2 XI; j1 ; j2 = Y j : zj Mzj 1 j1 j2 n,
j 2 j1 , j1 = j2 = 1,
j =2
Lz1 , b=a;
j2 Y j,1
j2 YI1; j1 ; j2 = Lz , b=a Y z j,1 Lz ; 1 = j j n, A.58 1 j 1 2 j
with corresponding expressions for the permutation . Then, from A.52 A.58, f0 I; Q0 = XI; 1; n; A.59
: j=j1
zj Lzj ;
1 j1 j2 n,
f1 I; Q1 = YI; 1; q1XI; q1 + 1; n; fr I; Qr = XI; 1; q1 , 1WI; q1 YI; q1 + 2; q2

j =2
A.60
2 r=2 3 Y 4 XI; q2j,2 + 1; q2j,1 , 1WI; q2j,1 YI; q2j,1 + 2; q2j 5

A.61
X1; qr + 1; n; r 2 and even, fr I; Qr = YI; 1; q1

j =1
2r,1=2 3 Y 4 XI; q2j,1 + 1; q2j , 1WI; q2j YI; q2j + 2; q2j+1 5

XI; qr + 1; n r 3 and odd.
A.62 The primary aim of this appendix is to establish the conditions for the righthand side of A.48 to be of the form of the right-hand side of 10.8. From A.55 A.62 it can be seen that a su cient condition for this to be the case would be
348
X
fg
A. Appendices
Afr P; Qr = 0 ;
A.63
for all r 1. In fact, as we shall see below, this is stronger than is needed. It is su cient that A.63 is true for all r 1. We see from A.61 and A.62 that all the functions fr I; Qr for r 1 and similarly when the permutation replaces I contain W functions. For A.63 to hold for r 1, it is therefore su cient that the coe cients A satisfy the condition ACj;j+1 Lzj Mzj+1 , 1 = A Mzj Lzj+1 , 1 : A.64 This is equivalent to 10.35
A.4.4 The Proof of IV

From A.55, A.59 A.63
X
fg
A
i ; = bN 1 X A Y j =1 fg
h j
zj Mzj
+ and from A.49 A.50
X
fg
A
n X r=1
bN 1 ,r ar Y; 1; rX; r; n
A.65
X
fg
A
ii ; = aN 1 X A Y j =1 fg
h j
zj Lzj
X
fg
A
n X N1,r r L b a Y; 1; rX; r; n : r=1
Now, from 10.24, Lz , b=a = z b=a a=b , Mz A.67 and using this result together with 10.33, A.46 and A.49 it can be shown, after some manipulation, that
A.66
X
fg
A
n X N1,r r L b a Y; 1; rX; r; n = r=1 n X X N1,r r
From A.47, A.65, A.66 and A.68 we then have
fg
A
r=1
a Y; 1; rX; r; n:
A.68
X
f0 g
V ; 0 g0 = aN 1
n Y j =1
A.5 Fourier Transforms and Series
Lzj + bN 1
n Y j =1
n Y
X
fg
j =1
Mzj
349
A
zjj ;
A.69
which is equivalent to 10.8 with given by 10.38.
A.5 Fourier Transforms and Series

We summarize the Fourier transform and series formulae needed for Chap. 10. A more detailed discussion of these transforms in d dimensions is given in Volume 2, Appendix A.1.
A.5.1 Fourier Transforms The Fourier transform F! fGxg of a function Gx is de ned by Z1 ! = F! fGxg = G Gx exp,ix!dx ;
where 1 Z 1 G ! expi!xd! : Gx = 2
,1 ,1
A.70 A.71
The transform of a convolution integral of two such functions F x and Gx is given by
F!
Z1
which gives Parseval's formula Z1 1 Z 1 F !G ! expi!xd! : F x , yGydy = 2 ,1 ,1
,1
F x , yGydy = F !G ! ;
A.72 A.73
A.5.2 Fourier Series The coe cients Fn fgxg for the complex Fourier series of a piecewise continuous function gx de ned in a range ,; are de ned by Z gn = Fn fgxg = gx exp,inxdx ; A.74
where
,
gx = 21
1 X
n=,1
gn expinx :
A.75
350
A. Appendices
The coe cients of a convolution integral of two such functions f x and gx are given by
Fn
Z
which gives Parseval's formula Z 1 1 X f g expinx : f x , ygydy = 2 n n , n=,1
f x , ygydy = fn gn ; ,
A.76 A.77
References and Author Index
The page numbers where the references are cited in the text are given in square brackets. Abraham D.B. 1973: Susceptibility and uctuations in the Ising ferromagnet. Phys. Lett. A 43, 163 166, 229 . Abramowitz M. and Segun I.A. 1965: Handbook of Mathematical Functions. Dover, New York, U.S.A. 337 . Adkins C.J. 1983: Equilibrium Thermodynamics. C.U.P., Cambridge, U.K. 1 . Aharony A. 1983: Multicritical points. In Critical Phenomena Ed. F.J.W. Hahne, 209 258, Springer, Berlin, Heidelberg, Germany, 162,322 . Albrecht O., Gruber H. and Sackmann E. 1978: Polymorphism of phospholipid monolayers. J. de Phys. 39, 301 313, 187 . Alexander S. 1975: Lattice gas transitions of He on grafoil. A continuous transition with cubic terms. Phys. Lett. A 54, 353 354, 332 . Anderson G.R. and Wheeler J.C. 1978a: Directionality dependence of lattice models for solutions with closed-loop coexistence curves. J. Chem. Phys. 69, 2082 2088, 275 . Anderson G.R. and Wheeler J.C. 1978b: Theory of lower critical solution points in aqueous mixtures. J. Chem. Phys. 69, 3403 3413, 276 Baker G.A. and Essam J.W. 1971: Statistical mechanics of a compressible Ising model with application to brass. J. Chem. Phys. 55, 861 879, 251 . Barker J.A. and Fock W. 1953: Theory of upper and lower critical solution temperatures. Disc. Faraday Soc. 15, 188 195, 270,276 . Bartis J.T. and Hall C.K. 1974: A lattice model of gas-gas equilibria in binary mixtures. Physica 78, 1 21, 278,289 . Banville M., Albinet G. and Firpo J.L. 1986: Fatty alcohol monolayers at the air water interface a spin-one model. J. Chim. Phys. et Phys. Chim. Biol. 83, 511 517, 187 . Baret J.F. and Firpo J.L. 1983: A spin-1 Ising model to describe amphile monolayer phase transitions. J. Coll. Inter. Sci. 94, 487 496, 187 . Baxter R.J. 1972: Partition function of the eight-vertex lattice model. Ann. Phys. 70, 193 228, 294 .
352
Baxter R.J. 1973a: Spontaneous staggered polarization of the F model. J. Phys. C: Solid State Phys. 6, L94 L96, 331 . Baxter R.J. 1973b: Potts model at the critical temperature. J. Phys. C: Solid State Phys. 6, L445 L448, 165,325,329,331 . Baxter R.J. 1973c: Spontaneous staggered polarization of the F-model. J. Stat. Phys. 9, 145 182, 331 . Baxter R.J. 1975: Triplet order parameter of the triangular Ising model. J. Phys. A: Math. Gen. 8, 1797 1805, 266,268 . Baxter R.J. 1982a: Exactly Solved Models in Statistical Mechanics. Academic Press, London, U.K. 174,198,205,229,231,294,312,332,337 . Baxter R.J. 1982b: Critical antiferromagnetic square lattice Potts model. Proc. Roy. Soc. A 383, 43 54, 325,328,330,331 . Baxter R.J. and Enting I.G. 1978: 339th solution of the Ising model. J. Phys. A: Math. Gen. 11, 2463 2473, 205,214,218 220 . Baxter R.J., Temperley H.N.V. and Ashley S.E. 1978: Triangular Potts model at its transition temperature, and related problems. Proc. Roy. Soc. A 358, 535 559, 165 . Bell G.M. 1953: Statistical thermodynamics of regular ternary assemblies. Trans. Faraday Soc. 49, 122 132, 151,157 . Bell G.M. 1958: Dilution e ects in regular assemblies. Proc. Phys. Soc. 72, 649 660, 166,167 . Bell G.M. 1969: One-dimensional bonded uids. J. Math. Phys. 10, 1753 1760, 128,130 . Bell G.M. 1972: Statistical mechanics of water: lattice model with directed bonding. J. Phys. C: Solid State Phys. 5, 889 905, 187,188,190 192 . Bell G.M. 1974a: Ising ferrimagnetic models I. J. Phys. C: Solid State Phys. 7, 1174 1188, 110,266 . Bell G.M. 1974b: Ising ferrimagnetic models II. J. Phys. C: Solid State Phys. 7, 1189 1205, 256,258,259,261,266 . Bell G.M. 1975: The diluted Ising model on linear and Bethe lattices. J. Phys. C: Solid State Phys. 8, 669 682, 168 . Bell G.M. 1980: A continuous one-dimensional uid as the limit of a lattice uid. J. Phys. A: Math. Gen. 13, 659 673, 124,128,131 . Bell G.M., Combs L.L. and Dunne L.J. 1981: Theory of cooperative phenomena in lipid systems. Chem. Rev. 81, 15 48, 187 . Bell G.M. and Fairbairn W.M 1961a: The critical temperatures of binary alloys with one magnetic component. Phil. Mag. 6, 907 928, 168 . Bell G.M. and Fairbairn W.M 1961b: Regular models for solid hydrogen: I. Mol. Phys. 4, 481 489, 270 . Bell G.M. and Fairbairn W.M 1967: Orientable ordering in solid hydrogen. Phys. Rev. 158, 530 536, 270,276 . Bell G.M. and Lavis D.A. 1965: Sublattice order in binary alloys with one magnetic component. Phil. Mag. 11, 937 953, 168 .
353
Bell G.M. and Lavis D.A. 1968: Statistical e ects of super-exchange in binary mixtures with one magnetic component. Phys. Rev. 168, 543 549, 168 . Bell G.M. and Lavis D.A. 1970: Two-dimensional bonded lattice uids: II. J. Phys. A: Math. Gen. 3, 568 581, 187,192,202 . Bell G.M. and Lavis D.A. 1989: Statistical Mechanics of Lattice Models. Vol.1: Closed-Form and Exact Theories of Cooperative Phenomena. Ellis Horwood, Chichester, U.K. VI . Bell G.M. and Sallouta H. 1975: An interstitial model for uid water: accurate and approximate results. Mol. Phys. 29, 1621 1637, 281 283 . Bell G.M. and Salt D.W. 1976: Three-dimensional lattice model for the water ice system. J. Chem. Soc. Faraday Trans. II 72, 76 86, 191 . Bell G.M., Mingins J. and Taylor R.A.G. 1978: Second-order phase changes in phospholipid monolayer at the oil water interface and related phenomena. J. Chem. Soc. Faraday Trans. II 74, 223 234, 182,187 . Bethe H.A. 1935: Statistical theory of superlattices. Proc. Roy. Soc. A 150, 552 575, 173 . Bidaux R., Carrara P. and Vivet B. 1967: Antiferromagnetisme dans un champ magnetique I. Traitement de champ moleculaire. J. Phys. Chem. Solids 28, 2453 2469, 108 . Binder K. 1987: Theory of rst-order transitions. Rep. Prog. Phys. 50, 783 859, 28,196 . Black J.L. and Emery V.J. 1981: Critical properties of two-dimensional models. Phys. Rev. B 23, 429 432, 331 . Blote H.W.J. and Nightingale M.P. 1982: Critical behaviour of the twodimensional Potts model with a continuous number of states; a nite size scaling analysis. Physica A 112, 405 465, 332 . Blume M. 1966: Theory of the rst-order magnetic phase change in UO2 . Phys. Rev. 141, 517 524, 151 . Blume M., Emery V.J. and Gri ths R.B. 1971: Ising model for the transition and phase separation on 3 He-4 He mixtures. Phys. Rev. A 4, 1071 1077, 151,157,160 . Bowman F. 1953: Introduction to Elliptic Functions with Applications. E.U.P., London, U.K. 337,338 . Bragg W.L. and Williams E. 1934: The e ect of thermal agitation on atomic arrangement in alloys. Proc. Roy. Soc. A 145, 699 730, 139 . Brankov J.G. and Zagrebnov V.A. 1983: On the description of the phase transition in the Husimi Temperley model. J. Phys. A: Math. Gen. 16, 2217 2224, 77 . Buchdahl H.A. 1966: The Concepts of Thermodynamics. C.U.P., Cambridge, U.K. 1 . Burley D.M. 1972: Closed-form approximations for lattice systems. In Phase Transitions and Critical Phenomena. Eds. C. Domb and M. S. Green, Vol. 2, 329 374. Academic Press, London, U.K., 173 . Capel H.W. 1966: On the possibility of rst-order phase transitions in Ising systems of triplet ions with zeroeld splitting. Physica 32, 966 988, 151 .
354
Cardy J.L. 1996: Scaling and Renormalization in Statistical Physics. C.U.P., Cambridge, U.K. 332 . Chandler D. 1987: Introduction to Modern Statistical Mechanics. O.U.P., Oxford, U.K. 31 . Danielian A. 1961: Ground state of an Ising face-centred cubic lattice. Phys. Rev. Lett. 6, 670 671, 196 . De Boer J. 1974: Van der Waals in his time and the present revival. Physica 73, 1 27, 9 . De Neef T. and Enting I.G. 1977: Series expansions from the nite-lattice method. J. Phys. A: Math. Gen. 10, 801 805, 173 . Den Nijs M.P.M. 1979: A relationship between the temperature exponents of the eight-vertex and q-state Potts model. J. Phys. A: Math. Gen. 12, 1857 1868, 331 . De Jongh L.J. and Miedema 1974: Experiments on simple magnetic model systems. Adv. in Phys. 23, 1 260, 44 . Domb C. 1960: On the theory of cooperative phenomena in crystals. Phil. Mag. Supp. 9, 149 361, 173,197,227,229,231,237 . Domb C. 1974: Ising model. In Phase Transitions and Critical Phenomena. Eds. C. Domb and M. S. Green, Vol. 3, 357 484. Academic Press, London, U.K., 75,103,104,211,227 . Domb C., 1985: Critical phenomena: a brief historical survey. Contemporary Physics 26, 49 72, V . Domb C. and Sykes M.F. 1961: Use of series expansions for the Ising model susceptibility and excluded volume problem. J. Math. Phys. 2, 63 67, 88 . Eggarter T.P. 1974: Cayley trees, the Ising problem and the thermodynamic limit. Phys. Rev. B 9, 2989 2992, 198 . Ehrenfest P. 1933: Phase changes in the ordinary and extended sense classi ed according to the corresponding singularity of the thermodynamic potential. Proc. Acad. Sci. Amsterdam 36, 153 157, 72 . Eisenberg D. and Kauzmann W. 1969: The Structure and Properties of Water, Clarendon Press, Oxford, U.K. 188,344 . Elcock E.W. 1958: Order-Disorder Phenomena. Methuen, London, 141 . Enting I.G. 1977: Triplet order parameters in triangular and honeycomb Ising models. J. Phys. A: Math. Gen. 10, 1737 1743, 266 . Essam J.W. 1972: Percolation and cluster size. In Phase Transitions and Critical Phenomena. Eds. C. Domb and M. S. Green, Vol. 2, 197 270. Academic Press, London, U.K., 167,250 . Essam J.W. 1980: Percolation theory. Rep. Prog. Phys. 43, 833 912, 167 . Essam J.W. and Fisher M.E. 1963: Pad approximant studies of the lattice e gas and Ising ferromagnet below the critical point. J. Chem. Phys. 38, 802 812, 86,88 . Essam J.W. and Garelick H. 1967: Critical behaviour of a soluble model of dilute ferromagnetism. Proc. Phys. Soc. 92, 136 149, 250 .
355
Firpo J.L., Dupin J.J., Albinet G., Baret J.F. and Caille A. 1981: Model for the tricritical point of the chain-melting transition in a monomolecular layer of amphiphilic molecules. J. Chem. Phys. 74, 2569 2575, 187 . Fisher M.E. 1963: Perpendicular susceptibility of the Ising model. J. Math. Phys. 4, 124 135, 105 . Fisher M.E. 1959a: Transformations of Ising models. Phys. Rev. 113, 969 981, 218,241,243 . Fisher M.E. 1959b: The susceptibility of the plane Ising model. Physica 25, 521 524, 229 . Fisher M.E. 1960: Lattice statistics in a magnetic eld I. A two-dimensional super-exchange antiferromagnet. Proc. Roy. Soc. A 254, 66 85, 252 . Fisher M.E. 1967: Theory of equilibrium critical phenomena. Rep. Prog. Phys. 30, 615 730, 73 . Fisher M.E. 1968: Renormalization of critical exponents by hidden variables. Phys. Rev. 176, 257 272, 250,256 . Fisher M.E. and Essam J.W. 1961: Some cluster size and percolation problems. J. Math. Phys. 2, 609 619, 167 . Fisher M.E. and Scesney P.E. 1970: Visibility of critical-exponent renormalization. Phys. Rev. A 2, 825 835, 250 . Fisher M.E. and Sykes M.F. 1962: Antiferromagnetic susceptibilities on the simple cubic and body-centred cubic Ising lattices. Physica 28, 939 956, 104 . Fleming M.E. and Gibbs J.N. 1974: An adaption of the lattice gas to the water problem. J. Stat. Phys. 101, 157 173, 351 378, 192 . Fletcher N.H. 1970: The Chemical Physics of Ice. C.U.P., Cambridge, U.K. 188,344 . Fowler R. and Guggenheim E.A. 1949: Statistical Thermodynamics. C.U.P., Cambridge, U.K. 76,145 . Friedan D., Qiu Z. and Shenker S. 1984: Conformal invariance, unitarity and critical exponents in two dimensions. Phys. Rev. Lett. 52, 1575 1578, 332 . Gantmacher F.R. 1979: Theory of Matrices. Chelsea, New York, U.S.A. 300 . Gotoh O. 1983: Prediction of melting pro les and local helix stability for sequenced DNA. Adv. in Biophys. 16, 1 52, 58,62 . Georges A., Hansel D., Le Doussal P. and Maillard J.M. 1986: The vicinity of disorder varieties: a systematic expansion. J. Phys. A: Math. Gen. 19, 1001 1005, 273 . Glasser M.L. 1969: Evaluation of the partition functions for some twodimensional ferroelectric models. Phys. Rev. 184, 539 542, 312 . Glasser M.L. Abraham D.B. and Lieb E.H. 1972: Analytic properties of the free energy for the ice" model. J. Math. Phys. 13, 887 900, 319 . Gradshteyn I.S. and Ryzhik I.M. 1980: Table of Integrals, Series, and Products. 4th Ed. Academic Press, London, U.K. 330,337 . Green H.S. and Hurst C.A. 1964: Order-Disorder Phenomena. Interscience, London, 205,229,231 .
356
Gri ths R.B. 1967a: Thermodynamic functions for uids and ferromagnets near the critical point. Phys. Rev. 158, 176 187, 81 . Gri ths R.B. 1967b: First-order phase transitions in spin-one Ising systems. Physica 33, 689 690, 270 . Gri3 ths R.B. 1970: Thermodynamics near the two- uid critical mixing point in He -He4 . Phys. Rev. Lett. 24, 715 717, 151,161 . Gri ths R.B. 1972: Rigorous results and theorems. In Phase Transitions and Critical Phenomena. Eds. C. Domb and M. S. Green, Vol. 1, 7 109. Academic Press, London, U.K., 33 . Gri ths R.B. and Wheeler J.C. 1970: Critical points in multicomponent systems. Phys. Rev. A 2, 1047 1064, 16 . Gunton J.D., San Miguel M. and Sahai P.S. 1983: Dynamics of rst-order transitions. In Phase Transitions and Critical Phenomena. Eds. C. Domb and J. L. Lebowitz, Vol. 8, 269 466. Academic Press, London, U.K., 28 . Guggenheim E.A. 1935: The statistical mechanics of regular solutions. Proc. Roy. Soc. A 148, 304 312, 173 . Guggenheim E.A. 1952: Mixtures. O.U.P., Oxford, U.K. 176 . Guggenheim E.A. and McGlashan M.L. 1951: Statistical mechanics of regular mixtures. Proc. Roy. Soc. A 206, 335 353, 173 . Hemmer P.C. and Stell G. 1970: Fluids with several phase transitions. Phys. Rev. Lett. 24, 1284 1287, 82 . Hemmer P.C. and Lebowitz J.L. 1976: Systems with weak long-range potentials. In Phase Transitions and Critical Phenomena. Eds. C. Domb and M. S. Green, Vol. 5b, 107 203. Academic Press, London, U.K., 77 . Hijmans J. and de Boer J. 1955: An approximation method for order-disorder problems I III. Physica 21, 471 484,485 498,499 498, 151,161 . Hijmans J. and de Boer J. 1956: An approximation method for order-disorder problems IV V. Physica 22, 408 428,429 442, 173 . Houska C.R. 1963: A theoretical treatment of atomic con gurations in some iron-aluminum solid solutions. J. Phys. Chem. Solids 24, 95 107, 168 . Huang K. 1963: Statistical Mechanics. Academic Press, London, U.K. 31 . Husimi K. 1953: Statistical mechanics of condensation. In Proc. Int. Conf. on Theoretical Physics, 531 533, Kyoto and Tokyo, 77 . Irani R.S. 1972: The thermodynamics and mechanisms of ordering systems. Contemporary Physics 13, 559 583, 147,197 . Izyumov Yu. A. and Skryabin Yu. N. 1988: Statistical Mechanics of Magnetically Ordered Systems, Plenum Press, New York, U.S.A. 42 . Jancel R. 1969: Foundations of Classical and Quantum Statistical Mechanics. Pergamon, Oxford, U.K. 32 . Kac M. 1968: Mathematical mechanisms of phase transitions. In Statistical Physics, Phase Transitions and Super uidity. Vol. 1 Eds. M. Cretien et al.. 241 305, Gordon and Breach, New York 77 . Kac M. and Ward J.C. 1952: A combinatorial solution of the two-dimensional Ising model. Phys. Rev. 88, 1332 1337, 205 .
357
Kasteleyn P.W. and van Kranendonk J 1956a: Constant-coupling approximation for Heisenberg ferromagnetism. Physica 22, 317 337, 173 . Kasteleyn P.W. and van Kranendonk J 1956b: Constant-coupling approximation for antiferromagnetism. Physica 22, 367 385, 173 . Kasteleyn P.W. and van Kranendonk J 1956c: Constant-coupling approximation for Ising spin systems. Physica 22, 387 396, 173 . Kaufman B. 1949: Crystal statistics. II Partition function evaluated by spinor analysis. Phys. Rev. 76, 1232 1243, 205 . Kaufman B. and Onsager L. 1949: Crystal statistics. III Short-range order in a binary Ising lattice. Phys. Rev. 76, 1244 1252, 205 . Kikuchi R. 1951: A theory of cooperative phenomena. Phys. Rev. 81, 988 1003, 173 . Kikuchi R. 1974: Superposition approximation and natural iteration calculation in cluster-variation method. J. Chem. Phys. 60, 1071 1080, 196 . Kincaid J.M. and Cohen E.G.D. 1975: Phase diagrams of liquid helium mixtures and metamagnets: experiment and mean eld theory. Phys. Rep. 22, 57 143, 162 . Kirkwood J.G. 1943: Phase transitions in monolayers due to hindered molecular rotation. Pub. Am. Assoc. Advant. Sci. 21, 157 160, 182 . Klein M.J. 1974: The historical origins of the van der Waals equation. Physica 73, 28 47, 9 . Knobler C.M. and Scott R.L. 1984: Multicritical points in uid systems. In Phase Transitions and Critical Phenomena. Eds. C. Domb and J. L. Lebowitz, Vol. 9, 163 231. Academic Press, London, U.K., 164 . Landsberg P.T. 1978: Thermodynamics and Statistical Mechanics. O.U.P., Oxford, U.K. 1 . Lavis D.A. 1996: The derivation of the free energy of the Ising model from that of the eight-vertex model. Rep. Math. Phys. 37, 147 155, 294 . Lavis D.A. and Bell G.M. 1967: The e ect of sublattice order in binary alloys with one magnetic component: III. Phil. Mag. 15, 587 601, 168 . Lavis D.A. and Bell G.M. 1999: Statistical Mechanics of Lattice Systems. Vol.2: Exact, Series and Renormalization Group Methods. Springer, Berlin, Heidelberg, Germany, V . Lavis D.A. and Christou N.I. 1977: The dielectric constant of a bonded lattice model for water. J. Phys. A: Math. Gen. 10, 2153 2169, 191 . Lavis D.A. and Christou N.I. 1979: A two-dimensional bonded lattice model for water. J. Phys. A: Math. Gen. 12, 1869 1890, 192 . Lavis D.A. and Fairbairn W.M. 1966: The e ect of sublattice order in binary alloys with one magnetic component. Phil. Mag. 13, 477 492, 168 . Lavis D.A. and Quinn A.G. 1983: An Ising ferrimagnetic model on a triangular lattice. J. Phys. C: Solid State Phys. 16, 3547 3562, 266,268 . Lavis D.A. and Quinn A.G. 1987: Transfer matrix and phenomenological renormalization methods applied to a triangular Ising ferrimagnetic model. J. Phys. C: Solid State Phys. 20, 2129 2138, 266 .
358
Lavis D.A. and Southern B.W. 1984: Renormalization group study of a threedimensional lattice model with directional bonding. J. Stat. Phys. 35, 489 506, 192 . Lawden D.F. 1989: Elliptic Functions and Applications. Springer, Berlin, Heidelberg, Germany, 337 . Lawrie I.D. and Sarbach S. 1984: Theory of tricritical points. In Phase Transitions and Critical Phenomena. Eds. C. Domb and J. L. Lebowitz, Vol. 9, 2 161. Academic Press, London, U.K., 162 . Lebowitz J.L. and Penrose O. 1966: Rigorous treatment of the van der WaalsMaxwell theory of the liquid-vapour transition. J. Math. Phys. 7, 98 113, 77 . Le Guillou J.C. and Zinn-Justin J. 1980: Critical exponents from eld theory. Phys. Rev. B 21, 3976 3998, 88 . Li Y.Y. 1949: Quasi-chemical theory of order for the copper gold alloy system. J. Chem. Phys. 17, 447 452, 173,196 . Lieb E.H. 1966: Quantum-mechanical extension of the Lebowitz-Penrose theorem on the van der Waals theory. J. Math. Phys. 7, 1016 1024 77 Lieb E.H. 1967a: Residual entropy of square ice. Phys. Rev. 162, 162 171, 293,312 . Lieb E.H. 1967b: Exact solution of the F model of an antiferroelectric. Phys. Rev. Lett. 18, 1046 1048, 293,295 . Lieb E.H. 1967c: Exact solution of the two-dimensional Slater KDP model of a ferroelectric. Phys. Rev. Lett. 19, 108 110, 293,295 . Lieb E.H. 1969: Two dimensional ice and ferroelectric models. In Lectures in Theoretical Physics, Vol. 11 Eds. K.T. Mantappa and W.E. Brittin, 329 354, Gordon and Breach, New York, U.S.A. 293 . Lieb E.H. and Mattis D.C. Eds. 1966: Mathematical Physics in One Dimension. Academic Press, New York, U.S.A. 124 . Lieb E.H. and Wu F.W. 1972: Two-dimensional ferroelectric models. In Phase Transitions and Critical Phenomena. Eds. C. Domb and M. S. Green, Vol. 1, 332 490. Academic Press, London, U.K., 293,301,308,313,318 320,325 . Ma S.K. 1985: Statistical Mechanics, World Scienti c, Philadelphia, U.S.A. 31 . Maillard J.M. 1985: Star-triangle and inversion relations in statistical mechanics. In Critical Phenomena Eds. V. Ceausescu et al., Progress in Physics, Vol. 11, 375 401, Birkhauser, Boston, U.S.A. 205 . Martin D.H. 1967: Magnetism in Solids. Ili e, London, U.K. 42,44,96,105,110 . Mattis D.C. 1985: Theory of Magnetism II. Springer, Berlin, Heidelberg, Germany 42,44,105 . Mattis D.C. Ed. 1993: The Many-Body Problem: An Encyclopedia of Exactly Solved Models in One Dimension. World Scienti c, Singapore, 124 McCoy B.M. 1972: Exact calculation on a random Ising model. In Phase Transitions and Critical Phenomena. Eds. C. Domb and M. S. Green, Vol. 2, 161 195. Academic Press, London, U.K., 167 . McCoy B.M. and Wu T.T. 1973: The Two-Dimensional Ising Model. Harvard U.P., Cambridge, Mass. U.S.A. 205,228,229 .
359
Meijering J.L. 1950: Segregation in regular ternary mixtures, I. Philips Res. Rep. 5, 333 356, 151 . Meijering J.L. 1951: Segregation in regular ternary mixtures, II. Philips Res. Rep. 6, 183 210, 151 . Mermin N.D. 1971a: Solvable model of a vapour-liquid transition with a singular coexistence-curve diameter. Phys. Rev. Lett. 26, 169 172, 82,281 . Mermin N.D. 1971b: Lattice gas with short-range pair interactions and a singular coexistence-curve diameter. Phys. Rev. Lett. 26, 957 959, 82,281 . Mermin N.D. and Rehr J.J. 1971: Generality of the singular diameter of the liquid-vapour coexistence curve. Phys. Rev. Lett. 26, 1155 1156, 82 . Morita T. 1984: Convergence of the cluster variation method for random Ising models in the thermodynamic limit. Prog. Theor. Phys. Supp. 80, 103 107, 173 . Mouritsen O.G. 1984: Computer Studies of Phase Transitions and Critical Phenomena. Springer, Berlin, Heidelberg, Germany, 238 . Mukamel D. and Blume M. 1974: Ising model for tricritical points in ternary mixtures. Phys. Rev. A 10, 610 617, 162 . Mulholland G.W. and Rehr J.J. 1974: Coexistence curve properties of Mermin's decorated lattice gas. J. Chem. Phys. 60, 1297 1306, 285,288 . Nagle J.F. 1966: Lattice statistics of hydrogen bonded crystals. I The residual entropy of ice. J. Math. Phys. 7, 1484 1491, 344 . Nagle J.F. 1975: Chain model theory of lipid monolayer transitions. J. Chem. Phys. 63, 1255 1261, 182 . Nagle J.F. 1986: Theory of lipid monolayer and bilayer chain-melting phase transitions. Faraday Disc. Chem. Soc. 81, 151 162, 182,187 . Nakono H. 1968: Ordering in certain statistical systems of Ising spins. Prog. Theor. Phys. 39, 1121 1132, 273 . Naya S. 1954: On the spontaneous magnetization of honeycomb, Kagom and e Ising lattices. Prog. Theor. Phys. 11, 53 62, 264 . Nienhuis B., Riedel E.K. and Schick M. 1980: Magnetic exponents of the twodimensional q-state Potts model. J. Phys. A: Math. Gen. 13, L189 L192, 331 . Onsager L. 1944: Crystal statistics. I Two-dimensional model with order-disorder transition. Phys. Rev. 65, 117 149, 205 . O'Reilly D.E. 1973: Scaled particle-quasilattice model of liquid water. Phys. Rev. A 7, 1659 1661, 192 . Pallas N.R. and Pethica B.A. 1985: Liquid-expanded to liquid-condensed transitions in lipid monolayers at the air water interface. Langmuir 1, 509 513, 182 . Palmer R.G. 1982: Broken ergodicity. Adv. in Phys. 31, 669 735, 57,75 . Parsonage N.G. and Staveley L.A.K. 1978: Disorder in Crystals. O.U.P., Oxford, U.K. 197 . Pauling L. 1935: The structure and entropy of ice and other crystals with some randomness of atomic arrangement. J. Am. Chem. Soc. 57, 2680 2684, 295 . Pearson R. 1980: Conjecture for the extended Potts model magnetic eigenvalue. Phys. Rev. B 22, 2579 2580, 331 .
360
Perram J.W. 1971: Interstitial models of water in the random mixing approximation. Mol. Phys. 20, 1077 1085, 281 . Phani M.K., Lebowitz J.L. and Kalos M.H. 1980: Monte Carlo studies of a face-centred cubic Ising antiferromagnet with nearest- and next-nearest-neighbour interactions. Phys. Rev. B 21, 4027 4037, 196 . Pippard A.B. 1957: Elements of Classical Thermodynamics. C.U.P., Cambridge, U.K. 1,26,72 . Potts R.B. 1952: Some generalized order-disorder transformations. Proc. Camb. Phil. Soc. 48, 106 109, 164,329,331 . Pynn R. and Skjeltorp A. Eds. 1984: Multicritical Phenomena. Plenum Press, New York, U.S.A. 162 . Rao C.N.R. 1972: Theory of hydrogen bonding in water. In Water; A Comprehensive Treatise., Vol. 1 Ed. F. Franks. 93 114, Plenum Press, New York, U.S.A. 343 . Rapaport D.C. 1972: The Ising ferromagnet with impurities: a series expansion approach, I. J. Phys. C: Solid State Phys. 5, 1830 1858, 245 . Rigby M. 1970: The van der Waals uid: a renaissance. Quart. Rev. Chem. Soc. 24, 416 432, 9 . Rowlinson J.S. 1973: Legacy of van der Waals. Nature 244, 414 417, 9 . Rowlinson J.S. 1988: English translation with introductory essay of J.D. van der Waals: On the continuity of the gaseous and liquid states of matter, Studies in Statistical Mechanics, Vol. 14. North-Holland, Amsterdam, New York, 362 . Rowlinson J. and Swinton F.L. 1982: Liquids and Liquid Mixtures. Butterworths, London, U.K. 22,119,164,281,289 . Ruelle D. 1969: Statistical Mechanics. Benjamin, New York, U.S.A. 33 . Runnels L.K. 1967: Phase transition of a Bethe lattice gas of hard molecules. J. Math. Phys. 8, 2081 2087, 198 . Runnels L.K. 1972: Lattice gas theories of melting. In Phase Transitions and Critical Phenomena. Eds. C. Domb and M. S. Green, Vol. 2, 305 328. Academic Press, London, U.K., 120 . Rushbrooke G.S. and Scoins H.I. 1955: On the Ising problem and Mayer's cluster sums. Proc. Roy. Soc. A 230, 74 90, 197 . Sallouta H. and Bell G.M. 1976: An interstitial model for aqueous solutions of non-electrolytes. Mol. Phys. 32, 839 855, 289 . Sato H., Arrott A. and Kikuchi R. 1959: Remarks on magnetically dilute systems. Phys. Chem. Sol. 10, 19 34, 167 . Schlijper A.G. 1983: Convergence of the cluster variation method in the thermodynamic limit. Phys. Rev. B 27, 6841 6848, 173 . Shinmi M. and Huckaby D. 1986: A lattice gas model for carbon tetrachloride. J. Chem. Phys. 84, 951 955, 192 . Shockley W. 1938: Theory of order for the copper gold alloy system. J. Chem. Phys. 6, 130 144, 147 . Sinha K. and Kumar N. 1980: Interactions in Magnetically Ordered Solids, O.U.P., Oxford, U.K. 42 .
361
Southern B.W. and Lavis D.A. 1980: Renormalization group study of a twodimensional lattice model with directional bonding. J. Phys. A: Math. Gen. 13, 251 262, 192 . Stanley H.E. 1987: Introduction to Phase Transitions and Critical Phenomena. O.U.P., Oxford, U.K. 47,83 . Stephenson J. 1970: Ising model with antiferromagnetic next-nearest-neighbour coupling: spin correlations and disorder points. Phys. Rev. B 1, 4405 4409, 273 . Stillinger E.H. 1975: Theory and molecular models for water. Adv. Chem. Phys. 31, 1 101, 192 . Stinchcombe R.B. 1983: Dilute magnetism. In Phase Transitions and Critical Phenomena. Eds. C. Domb and J. L. Lebowitz, Vol. 7, 152 280. Academic Press, London, U.K., 167 . Straley J.P. and Fisher M.E. 1973: Three-state Potts model and anomalous tricritical points. J. Phys. A: Math. Gen. 6, 1310 1326, 166 . Sutherland B. 1967: Exact solution of a two-dimensional model for hydrogenbonded crystals. Phys. Rev. Lett. 19, 103 104, 293 . Suzuki M. and Fisher M.E. 1971: Zeros of the partition function for the Heisenberg, ferroelectric and general Ising models. J. Math. Phys. 12, 235 246, 294 Syozi I. 1972: Transformation of Ising models. In Phase Transitions and Critical Phenomena. Eds. C. Domb and M. S. Green, Vol. 1, 270 329. Academic Press, London, U.K., 205,241,250,256,264,273 . Syozi I. and Nakono H. 1955: Statistical models of ferrimagnetism. Prog. Theor. Phys. 13, 69 78, 256 . Temperley H.N.V. 1954: The Mayer theory of condensation tested against a simple model of the imperfect gas. Proc. Phys. Soc. 67, 233 238, 77 . Temperley H.N.V. 1965: An exactly soluble lattice model of the uid-solid transition. Proc. Phys. Soc. 86, 185 192, 198 . Temperley H.N.V. 1972: Two-dimensional Ising models. In Phase Transitions and Critical Phenomena. Eds. C. Domb and M. S. Green, Vol. 1, 227 267. Academic Press, London, U.K., 205 . Temperley H.N.V. 1981: Graph Theory and Applications. Ellis Horwood, Chichester, U.K. 197,206,207,335 . Temperley H.N.V. and Lieb E.H. 1971: Relations between the percolation and colouring problems on the plane square lattice. Some exact results for the percolation problem. Proc. Roy. Soc. A 322, 251 280, 325 . Thompson C.S. 1972a: One-dimensional models short range forces. In Phase Transitions and Critical Phenomena. Eds. C. Domb and M. S. Green, Vol. 1, 177 226. Academic Press, London, U.K., 45 . Thompson C.S. 1972b: Mathematical Statistical Mechanics. Macmillan, New York, U.S.A. 77 . Thompson C.S. 1985: Random spin systems. In Critical Phenomena Eds. V. Ceausescu et al. Progress in Physics Vol. 11. 136 186, Birkhauser, Boston, U.S.A. 167 . Thompson C.S. 1988: Classical Equilibrium Statistical Mechanics. O.U.P., Oxford, U.K. 31 .
362
Van der Waals J.D. 1873 Over de Continuiteit van der Gas- en Vloeistoftoestand, Doctoral Thesis, Leiden; for an English translation see Rowlinson 1988, 9 . Vdovichenko N.V. 1965a: A calculation of the partition function for a plane dipole lattice. Soviet Phys. JETP. 20, 477 479, 205 . Vdovichenko N.V. 1965b: Spontaneous magnetization of a plane diploe lattice. Soviet Phys. JETP. 21, 350 352, 205 . Wada A., Yubaki S. and Husimi Y. 1980: Fine structure in the thermal denaturation of DNA: high temperature-resolution spectrophotometric studies. CRC Crit. Rev. Biochem. 9, 87 144, 60 . Wannier G.H. 1950: Antiferromagnetism: the triangular net. Phys. Rev. 79, 357 364, 237 . Wartell R.M. and Benight A.S. 1985: Thermal denaturation of DNA molecules: a comparison of theory and experiment. Phys. Rep. 126, 67 107, 58,60 . Wartell R.M. and Montroll E.W. 1972: Equilibrium denaturation of natural and of periodic synthetic DNA molecules. Adv. Chem. Phys. 22, 129 203, 58,60 . Weres O. and Rice S.A. 1972: A new model for liquid water. J. Am. Chem. Soc. 94, 8983 9002, 191 . Wheeler J.C. 1975: Exactly solvable" two component lattice solution with upper and lower critical solution temperatures. J. Chem. Phys. 62, 433 439, 275 . Wheeler J.C. 1977: Decorated lattice-gas models of critical phenomena in uids and uid mixtures. Ann. Rev. Phys. Chem. 28, 411 443, 82,273,275,281,289,290 . Wheeler J.C. and Anderson G.R. 1980: Some interesting new phase diagrams in hydrogen-bonded liquid mixtures. J. Chem. Phys. 73, 5778 5785, 276 . Wheeler J.C. and Widom B. 1970: Phase equilibrium and critical behaviour in a two-component Bethe lattice gas or a three-component Bethe lattice solution. J. Chem. Phys. 52, 5334 5343, 151 . Whitehouse J.S., Christou N.I., Nicholson D. and Parsonage N.G. 1984: A grand ensemble Monte Carlo investigation of the Bell lattice model for water. J. Phys. A: Math. Gen. 17, 1671 1682, 192 . Widom B. 1964: Degree of the critical isotherm. J. Chem. Phys. 41, 1633 1634, 85 . Widom B. 1967: Plait points in two- and three-component liquid mixtures. J. Chem. Phys. 49, 3324 3333, 250,289 . Widom B. and Rowlinson J. S. 1970: New model for the study of liquid-vapour phase transitions. J. Chem. Phys. 52, 1670 1684, 82 . Wiegel F.W. 1983: Models of conformal phase transitions. In Phase Transitions and Critical Phenomena. Eds. C. Domb and J. L. Lebowitz, Vol. 7, 101 149. Academic Press, London, U.K., 62 . Wilson G.L. and Bell G.M. 1978: Lattice model for aqueous solutions of nonelectrolytes. J. Chem. Soc. Faraday Trans. II 74, 1702 1722, 191 . Wood D.W. 1975: Thermodynamic behaviour in the critical region. In Statistical Mechanics, Vol. 2 Ed. K. Singer, 55 187, Chem. Soc. Specialist Periodical Report, London, U.K. 44,76 . Wu F.Y. 1982: The Potts model. Rev. Mod. Phys. 54, 235 268, 164 166 .
363
Yang C.N. 1945: A generalization of the quasi-chemical method in the statistical theory of superlattices. J. Chem. Phys. 13, 66 76, 173 . Young A.P. 1981: The Kikuchi approximation: a reformulation and application to the percolation problem. J. Phys. A: Math. Gen. 14, 873 881, 173 . Ziman J. 1979: Models of Disorder. C.U.P., Cambridge, U.K. 63,67,173 .
Subject Index
alloy annealed 57, 146 quenched 57 amphipathic monolayers 181 antiferroelectric 293, 295 ground state 299 antiferromagnet critical curve 140 critical eld 97 aqueous solution 289 azeotropic point 23, 135, 147, 288 critical 288 line of 289 azeotropic-type maximum 196 azeotropy 22, 289 critical 289 Bethe ansatz 304, 328 Bethe lattice 167, 197 Bethe method 173 Bethe pair method 178 bicritical end-point 162 point 322 binary mixture, equivalent Ising model 137 Blume Emery Gri ths model 151 Boltzmann distribution 31 bond percolation 167 broken ergodicity 57 broken symmetry 75 canonical distribution 32 partition function, con gurational 34 carbon tetrachloride 192 Cayley tree 197 cell model 121 closed-form approximations 173
closed-loop coexistence curve 275 cluster variation methods 173 coe cient of thermal expansion 4 coexistence curve 80 region diameter 81, 280 width 81, 280 compensation temperature 112 compressibility 4 minimum in water model 191, 283 con guration graph 297 conformal invariance 332 constant-coupling method 173 constanteld grand distribution 40 correlation in antiferromagnet 93 length 74 two-site function 46 covalent bond 342 critical curve 100 double-point 241, 278, 287 end-point 107, 156, 157, 161, 163, 185 exponents 75, 83 classical 86, 173, 181 eld in antiferromagnet 94, 98 opalescence 8 parameters pair approximation 180 point, in ferrimagnet 115 points, line of 159 temperature lower 241, 271, 272 upper 271, 272 Curie temperature 44 in ferrimagnet 112 paramagnetic 105, 291 Curie Weiss law 105
366
Subject Index line of 106, 287 surface of 159 uctuation response function relation 50, 52 free energy Gibbs 3 Helmholtz 3 Gibbs phase rule 20, 159, 163 Gibbs potential 40 Gibbs Duhem equation 4, 18 Gibbs Helmholtz relations 29, 32 generalized 15 glycerol-ethylbenzylamine 273, 276 grand partition function restricted 135 grand potential 4, 40 graph edge 335 vertex 197, 335 ground state 52 frustrated 238 Hamiltonian 35 con gurational 37 hard hexagon model 332 hard shoulder model 133 heat capacity 5 anomaly 139 constanteld 46 cusp in 249 Heisenberg model 42, 44, 167, 168 one-dimensional 308 helium, 3 He 4 He mixtures 151, 157, 186 hidden variable 250 high-temperature series 209, 326 Husimi tree 198 Husimi Temperley model 77 hydrogen bond 57, 187, 281, 343 ice rule 294, 327 structures 188 IF model 296, 325, 332 IKDP model 296, 332 instability, intrinsic 26, 106 interstitial model 281 irreversible process 7 Ising model 20, 42, 293 antiferromagnet 93, 137, 324, 330 binary mixture equivalence 137 bond-dilute 166 compressible 251
cuspoidal point 287 decoration transformation 241, 262 density 16 density maximum in water model 191, 283 Dieterici equation 30 disorder point 273, 275 disordered chain conformation 182 displaced parabolic well model 128 distant-independent model 77 DNA 57 double tangent rule 10 dual lattice 206 dual transformation 205, 211 duality relationship 328 edge of graph 197 eight-vertex model 294 elliptic integrals 338 complete 338 ensemble average 32 enthalpy 3 entropy 2 increase in 7 principle of increasing 7 equal-areas rule, Maxwell's 9 equation of state 1 equilibrium, thermodynamic system in 1 ergodic hypothesis 32 Euler's relation 207 exclusion model n , 1-th neighbour 121 nearest-neighbour 120 exponent renormalization 248, 256 F model 295, 325, 329, 332 staggered 331 fatty acid 182 ferrimagnet, simple 112 ferrimagnetism 110 ferroelectric 293, 295 ground state 299 ferromagnet antiferromagnet equivalence 139 ferromagnetism 42 eld 13 con gurational 35 nite-size scaling 332 rst-order approximation 147, 276 rst-order transitions 147, 196 in monolayers 182 in solid mixture 138
Subject Index correlation 67, 74 critical temperature honeycomb lattice 263 Kagom lattice 263 e square lattice 213, 263 Curie temperature 71 dilute 151, 157, 192 ferromagnet 78, 137 rst-order transition in 70 for real ferromagnets 44 honeycomb lattice spontaneous magnetization 235 lattice gas 126, 182 lattice gas equivalence 125 local order 74 long-range order in 72 meaneld critical temperature 68 one-dimensional 54 short-range order in 67 site-dilute 166 spin-s 43 spin-1 64, 90, 149, 187 spontaneous magnetization 69, 235 susceptibility 73 two dimensions critical entropy 227 critical exponents 226, 228, 332 critical point energy 227 critical point entropy 234 exact results 72 spontaneous magnetization 228 zeroeld 51, 52, 184 isomerism 182 Jacobian elliptic function 339 KDP model 295, 306, 325 Landau expansion 143, 196 Landen transformation 339 lattice close-packed 95, 337 loose-packed 95, 335 lattice gas 119, 121 critical isochore 128 rst-order transition line 128 Ising model equivalence 125 one-dimensional 121 simple 120 Lebowitz Penrose interaction 125 Lebowitz Penrose theorem 77 Legendre relation 339 Legendre transformation 3, 14
367
liquid-condensed phase, in monolayers 182 liquid-expanded phase, in monolayers 182 lone-pair orbital 342 low-temperature series 206 lower critical point 260 macrosystem 31 maxithermal point 287, 288 Maxwell's thermodynamic relations 29 Mayer cluster expansion 197 Mayer factors 123 meaneld approximation 165 Weiss eld approach 76 medial lattice 326 metamagnet 20, 157 metastable 27, 57 state 145 microsystem 31 monolayer air water 151, 182, 186 oil water 151, 182 N el temperature 96, 101, 254 e oil water interface monolayer 186 order long-range 47 short-range 47 orientable molecule 274 ortho-hydrogen 270 pair approximation 178, 180 paramagnetic state 96 Parseval's formula 349, 350 particle hole symmetry 286 perfect gas 9 Perron's theorem 300, 304, 306 phase energetically stabilized 278 entropically stabilized 278 phase transition continuous 73 Ehrenfest's classi cation 72 rst-order 80 second-order 73 phospho-lipid monolayer 182, 187 polygon decomposition 327 polygons 326 potassium dihydrogen phosphate 293 Potts model 63, 162, 164, 165, 172 anisotropic 325
368
Subject Index superexchange antiferromagnet 253 super uid 4 He 151 superlattice formation 168 supersaturated 28 susceptibility, isothermal magnetic 6 Syozi dilution model 250, 290 Takahashi model 121, 124 thermodynamic inequalities 26 limit 16, 33, 49 potential 3 Tonks gas 124 transfer matrix 44 eigenvalues 48 transition rst-order 20 higher-order 102 second-order 103 transition line, rst-order 158 in ferrimagnet 112 tricritical point 106, 156 158, 161, 163, 165, 185, 187 in monolayer 187 triple points 20, 156, 158, 185 line of 159, 161, 163, 165 universality class 332 valency of vertex of graph 197 van der Waals equation 9, 30, 125 coexistence curve 12 conjugate phases 12 critical isotherm 11 critical point 11 scaled form 88 van der Waals theory 77 variables conjugate 2 extensive 1 intensive 1 water compressibility minimum 344 density maximum 344 molecule 293 water-nicotine 273, 276 Weiss eld equation 90 zero-point entropy 53 Pauling's approximation for square ice 295 zeroth-order approximation 68
conjectured exponents 331 critical temperature antiferromagnetic 330 ferromagnetic 329 duality transformation 329 rst-order transition in 331 internal energy 331 latent heat 331 one-dimensional 63 spontaneous polarization 331 quadruple point 161, 165 quasi-chemical method 173 randomized approximation 68 rectilinear diameter law 81, 280 relaxation time 28 response function 82 scaling law Essam Fisher 86, 88, 249 Widom 85, 87, 249 second law of thermodynamics 2 short-range order e ects 173 site percolation 167 six-vertex model 294 antiferroelectric free energy 317 antiferroelectric state 316 ferroelectric ground state 307 rst-order transition 316 in nite-order transition 319 perfect long-range order 307 staggered 325, 328 transition temperature 306 sixteen-vertex model 294 solute 289 solvent 289 spinodal 27 spinodal decomposition 27 square ice 293 zero-point entropy, Lieb's formula 312 square lattice, self-duality 207 staggered eld 162 star-triangle transformation 205, 211, 227, 262, 341 stoichiometric ratio 139 striped randomness 167 sublattice interpenetrating 95 order 96 rst-order transition 136 supercooled 28 superexchange 244

Volume 1

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Volume 1

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Volume 1

Uploaded by

Copyright:

Available Formats

D. A. Lavis and G. M.

Bell Department of Mathematics King's College, London

Statistical Mechanics of Lattice Systems

1. Introduction to Thermodynamics and Phase Transitions : : 1

2. Statistical Mechanics and the One-Dimensional Ising Model : : : : : : : : : : : : : : : : : : : : 31

3. The Mean-Field Approximation, Scaling and Critical Exponents : : : : : : : : : : : : : : : : : : : : : : : : : : : 67

4. Antiferromagnets and Other Magnetic Systems : : : : : : : : : : : 93

5. Lattice Gases : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 119

6. Solid Mixtures and the Dilute Ising Model : : : : : : : : : : : : : : : 135

7. Cluster Variation Methods : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 173

8. Exact Results for Two-Dimensional Ising Models : : : : : : : : : 205

9. Applications of Transform Methods : : : : : : : : : : : : : : : : : : : : : : : 241

10. The Six-Vertex Model : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 293

References and Author Index : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 351 Subject Index : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 365

1. Introduction to Thermodynamics and Phase Transitions

1.1 Thermodynamic Variables: Simple Fluids

1. Introduction to Thermodynamics and Phase Transitions

1.2 Change of Variable and Thermodynamic Potentials

1.2 Change of Variable and Thermodynamic Potentials

1. Introduction to Thermodynamics and Phase Transitions

1.3 Response Functions and Thermodynamic Relations

1.3 Response Functions and Thermodynamic Relations

From 1.25, 1.26, 1.20 and 1.23

e where V is the molar volume Ma v. ~

1.4 Magnetic Systems

1. Introduction to Thermodynamics and Phase Transitions

1.5 Stationary Properties of Thermodynamic Functions

1.5 Stationary Properties of Thermodynamic Functions

1. Introduction to Thermodynamics and Phase Transitions

1.6 Phase Equilibrium in the Van der Waals Gas

1.6 Phase Equilibrium in the Van der Waals Gas

1. Introduction to Thermodynamics and Phase Transitions

v2 v ~ ~ e Tc ; equal-areas rule for conjugate

1.6 Phase Equilibrium in the Van der Waals Gas

1. Introduction to Thermodynamics and Phase Transitions

Fig. 1.3. Equilibrium pressure volume isotherms; C is the critical point.

1.7 The Field-Extensive Variable Representation of Thermodynamics

1.7 The Field-Extensive Variable Representation of Thermodynamics

1. Introduction to Thermodynamics and Phase Transitions

By taking the di erential of 1.57 and comparing it with 1.56

i Qi ; i=1 n X dA = ,S dT + i dQi : i=1

1.7 The Field-Extensive Variable Representation of Thermodynamics

1. Introduction to Thermodynamics and Phase Transitions

,P = 0. On the other hand, a system can be isolated magnetically by putting

1.8 The Field-Density Representation of Thermodynamics

i=1 n,1 X i=1

1.8 The Field-Density Representation of Thermodynamics

which can be combined with 1.73 to give du = T ds +

From 1.77 and 1.75 it is easy to show that

1. Introduction to Thermodynamics and Phase Transitions

i=1 n, X i=1

The last relation can be re-expressed in the form

which is a generalized form of the Gibbs Duhem equation 1.16.

1.9 General Theory of Phase Equilibrium

1.9 General Theory of Phase Equilibrium

i1 , i2 dQi :

1.9.1 A One-Component Fluid

1. Introduction to Thermodynamics and Phase Transitions

1.9 General Theory of Phase Equilibrium

From 1.25, 1.26, 1.20 and 1.23

By taking the di erential of 1.57 and comparing it with 1.56

i Qi ; i=1 n X dA = ,S dT + i dQi : i=1

which can be combined with 1.73 to give du = T ds +

From 1.77 and 1.75 it is easy to show that

i=1 n, X i=1

which is a generalized form of the Gibbs Duhem equation 1.16.

i1 , i2 dQi :

e! e ! b @ T = @ V ; e e @P @S e !S eP! d @ V = , @ S : e e @T P @P T

e e e 1.4 De ning CH as in 1.35 and CM = T @ S =@ T M show that @ M 2 " @ M ! ,1 e : CH , CM = T e e @T H @H T

f1 = u , Ts + Pv = a + Pv = ; df1 = ,sdT + vdP ; da = ,sdT , P dv :

Then, from 2.2, Z = Z1 Z2 ; where X Zi = exp ,Ei i =T ;

Thus the probability of a given set of values of by X p 2 = p = exp ,E2 2 =T

= exp ,E2 2 =T

1 exp ,E1 1 =T

with the generalized partition function X b Z T; 1 ; : : : ; ; Q+1 ; : : : ; Qn = exp ,H =T :

00 X b b Xc i Qi = Hrc + Ekin + T Mj ln j T ; i=1

c ic Qi = , @ ln Z1 = Fc + TSc ; @ T, X i=1 n X

urational enthalpy. From 2.38 and 2.39 dFc = ,Sc dT ,

The con gurational free energy density is de ned by f = ,T lim ln Zc :

0 n,1 X fc dQn = @ ic

2.3.2 The Constant-Pressure Isobaric Distribution

2.52 2.53 2.54

and the elds which are not chemical potentials are = ,T @ ln Z = @D ; j = + 1; : : : ; n :

j , dMj + , P dN ;

From 2.45 df = ,sdT ,

Ix six sjx + Iy siy sjy + Iz siz sjz ; ^ ^ ^ ^ ^ ^

F = ,NT fln2 + ln coshJ=T g