Edited by H. Araki, Kyoto, J. Ehlers, Minchen, K. Hepp, Zirich R. Kippenhahn, Minchen, H. A. Weidenmiller, Heidelberg and J. Zittartz, Koln 186 Critical Phenomena Proceedings, Stellenbosch, South Africa 1982 Edited by F J. W. Hahne Springer-Verlag Berlin Heidelberg GmbHLecture Notes in Physics For information about Vols. 1-99, please contact your bookseller or Springer-Verlag. Vol. 100: Einstein Symposion Berlin. Proceedings 1979. Edited by H, Nelkowski etal, Vil, 550 pages. 1979. Vol. 101: A. Martn-Lot, Statistical Mechanics and the Foun dations of Thermodynamics. V, 120 pages. 1979. Vol. 102: H. Hora, Nonlinear Plasma Dynamics at Laser Irradiation. Vil, 242 pages. 1979. Vol. 103: P. A. Martin, Modéles en Mécanique Statistique des Processus Irréversibles. IV, 184 pages. 1979. Vol. 104: Dynamical Critical Phenomenaand Related Topics. Proceedings, 1979. Edited by Ch. P. Enz. XIl, 390 pages. 1979. Vol. 108: Dynamics and Instability of Fluid Interfaces. Pro- ‘ceedings, 1978. Edited by T. S. Sorensen. V, 315 pages. 1979. Vol, 106: Feynman Path Integrals, Proceedings, 1978. Edited by S. Albeverio etal. XI, 451 pages. 1979, Vol. 107: J. Kijowski, W. M. Tulezyjew, A Symplectic Frame- work for Field Theories. V, 257 pages. 1979, Vol. 108: Nuclear Physics with Electromagnetic Interactions, Proceedings, 1979. Edited by H. Arenhovel and D. Drechsel IX, 509 pages. 1979. Vol. 109: Physics of the Expanding Universe. Proceedings, 1978. Edited by M. Demiaiski. V, 210 pages. 1979. Vol. 10: D. A. Park, Classical Dynamics and Its Quantum Analogues. Vil, 338 pages. 1978. Vol. 111: H-J. Schmidt, Axiomatic Characterization of Physi- cal Geometry. V, 163 pages. 1979. Vol. 112: Imaging Processes and Coherence in Physics. Proceedings, 1979. Edited by M. Schlenker etal. XIX, 577, pages. 1980. Vol. 13: Recent Advances in the Quantum Theory of Poly: ‘mers. Proceedings 1979. Edited by J-M. André et al. V, 308, pages. 1980. Vol, 14: Stellar Turbulence. Proceedings, 1979. Edited by D.F. Gray and J. L Linsky. IX, 308 pages. 1980. Vol. 118: Modern Trends in the Theory of Condensed Matter. Proceedings, 1979. Edited by A. Pokalskiand J. Praystawa, 1X, 597 pages. 1980, Vol. 116: Mathematical Problems in Theoretical Physics. Proceedings, 1979. Edited by K. Osterwalder. Vil, 412, pages. 1980. Vol. 1t7: Deep-Inelastic and Fusion Reactions with Heavy lons. Proceedings, 1979. Edited by W. von Oertzen, Xill, 394 pages. 1980. Vol. 118: Quantum Chromodynamics. Proceedings, 1979. Edited by J. L Alonso and R. Tarrach. IX, 424 pages. 1980, Vol. 119: Nuclear Spectroscopy. Proceedings, 1979. Edited by G. F. Bertsch and D. Kurath. Vl, 250 pages. 1980. Vol. 120: Nonlinear Evolution Equations and Dynamical Systems. Proceedings, 1979. Edited by M. Boi, F. Pemp- nelli and G., Soliani. VI, 368 pages. 1980, Vol. 121: F. W. Wiegel, Fluid Flow Through Porous Macro: ‘molecular Systems. V, 102 pages. 1980, Vol. 122: New Developments in Semiconductor Physics. Proceedings, 1979. Edited by F. Beleznay et al.V,276 pages. 1980. Vol, 123: D. H. Mayer, The Ruelle-Araki Transfer Operator in Classical Statistical Mechanics. Vill, 154 pages. 1980. Vol. 124: Gravitational Radiation, Coilapsed Objects and Exact Solutions. Proceedings, 1979. Edited by C. Edwards. VI, 487 pages. 1980. Vol. 126: Nonradial and Nonlinear Stellar Pulsation. Pro- ceedings, 1980. Edited by H. A. Hill and W. A. Dziembowski Vill, 497 pages. 1980. Vol. 128: Complex Analysis, Microlocal Calculus and Rel ativistic Quantum Theory. Proceedings, 1979, Edited by D. lagolnitzer. Vil, 602 pages. 1980. Vol. 127: E. Sanchez-Palencia, Non Homogeneous Media and Vibration Theory. IX, 398 pages. 1980. Vol. 128: Neutron Spin Echo. Proceedings, 1979. Edited by F. Mezei. VI, 253 pages. 1980. Vol. 129: Geometrical and Topological Methods in Gauge ‘Theories, Proceedings, 1979. Edited by J. Hamad and S. Shnider. Vil, 185 pages. 1980. Vol. 190: Mathematical Methods and Applications of Scat- tering Theory. Proceedings, 1979. Edited by J. A. DeSanto, A.W. Sdenz and W. W. Zachary. Xl, 331 pages. 1980. Vol. 131: H. C. Fogedby, Theoretical Aspects of Mainly Low Dimensional Magnetic Systems. Xl, 163 pages. 1980. Vol. 192: Systems Far from Equilibrium, Proceedings, 1980. Edited by L Garrido. XV, 403 pages. 1980. Vol. 183: Narrow Gap Semiconductors Physics and Applica tions. Proceedings, 1979. Edited by W. Zawadzki. X, 872 pages. 1980. Vol. 134: yy Collisions. Proceedings, 1980. Edited by G. Cochard and P. Kessler. XIll 400 pages. 1980. Vol. 135: Group Theoretical Methods in Physics. Proceed ings, 1980. Edited by K. B. Wolf. XXVI, 629 pages. 1980. Vol. 136: The Role of Coherent Structures in Modelling Turbulence and Mixing. Proceedings 1980. Edited by J. Jimenez. Xill, 89 pages. 1981. Vol. 197: From Collective States to Quarks in Nuclei. Edited by H. Arenhével and A.M. Saruis. Vi, 414 pages. 1981. Vol. 188: The Many-Body Problem. Proceedings 1980. Ezited by R. Guardiola and J. Ros. V, 374 pages. 198! Vol. 139: H. D. Doebner, Differential Geometric Methods in Mathematical Physics. Proceedings 1981. Vil, 329 pages. 1981. Vol. 140: P. Kramer, M. Saraceno, Geometry of the Time Dependent Variational Principle in Quantum Mechanics. IV, 98 pages. 198. Vol. 141: Seventh International Conference on Numerical Methods in Fluid Dynamics. Proceedings. Edited by W. C. Reynolds and R. W. MacCormack. Vil, 485 pages. 1981. Vol. 142: Recent Progress in Many-Body Theories. Pro: ‘ceedings. Edited by J. G. Zabolitzky, M. de Llano, M. Fortes and J. W. Clark, Vil, 479 pages. 1981, Vol. 143: Present Status and Aims of Quantum Electro: dynamics. Proceedings, 1980. Edited by G. Griff, E. Klempt and G. Werth. VI, 302 pages. 1981.Lecture Notes in Physics Edited by H. Araki, Kyoto, J. Ehlers, Miinchen, K. Hepp, Ziirich R. Kippenhahn, Munchen, H. A. Weidenmiller, Heidelberg and J. Zittartz, KéIn 186 Critical Phenomena Proceedings of the Summer School Held at the University of Stellenbosch, South Africa January 18-29, 1982 Edited by F J.W. Hahne Springer-Verlag Berlin Heidelberg GmbH 1983Editor FJW. Hahne University of Stellenbosch ‘The Merensky Institute for Physics ‘Stellenbosch, 7600 South Africa ISBN 978-3-540-12675-1 ISBN 978-3-540-38667-4 (eBook) DO! 10.1007/978-3-540-38667-4 This work is subject to copyright. lights are reserved, whether the whole or part ofthe material is concerned, specifically those of translation, reprinting, re-use of illustrations, broadcasting, reproduction by photocopying machine or similar means, and storage in data banks. Under §§ 54 of the German Copyright Law where copies are made for other than private use, a fee is payable to “Verwertungegesellschatt Wort”, Munich. ‘© Springer-Verlag Berlin Heidelberg 1983 Originally published by Springer-Verlag Berlin Heidelberg New York in 1983 2189/3140-543210Table of Contents M.E. Fisher Scaling, Universality and Renormalization Group Theory ..sseeeeessseeeseeseseeees Introduction ...+. Critical Phenomena in Magnets and Fluids: Universality and Exponents . Scaling Microscopic Models ....ccesseeeeseeeeeesseeeeeeceseeesssseeeeeessseeeeeeeees Renormalization Group Theory . Dimensionality Expansions .........++ Acknowledgments . Appendix A The Kac-Hubbard-Stratonovich Transformation . Appendix B Details of the c-Expansion Calculation . Appendix C Dimensionality as a Continuous Variable ....sseeeeeeeeeeeeeeees Appendix D Hyperscaling and Dangerous Irrelevant Variables Bibliography References . 4H. Thomas Phase Transitions and Instabilities . Historical Introduction ... Classical Theory . Symmetry Aspects . Mean-Field Approximation .. Introduction to Driven Systems . Description of Driven systems Bifurcation from the Steady State . Onset of Turbulence ......+++eseeeeeeeees A. Aharony Multicritical Points . General Review: Tricritical Points General Review: Lifshitz and Bicritical Points Landau Theory: Tricritical Scaling Landau Theory: Other Cases ... Renormalization Group and Scaling . Continuous Spins, Wilson's Renormalization Group and the Gaussian Model . Tandau and Lifshitz Point Theories .. The e-Expansion . Results for Multicritical Points .. The Cubic Problem ......++ a 46 63 100 116 117 tar 128 132 136 137 141 142 . 143 - 159 165 2173 477 = 185, = 200 + 209 = 210 - 216 = 222 = 228 = 232 = 236 = 242 . 247 = 252 = 255;M.J. Stephen Lectures on Disordered systens . 259 Percolation .. = 260 Random Magnets ....66006 : an Random CondUCtOrs .essseseeeeseeeeeeses se eeeeeeecseeeeeessseeeeeeses 275 spin Glasses - : : 280 Pocalization seeeesesesceseesseeeessseeeeseseeeeessseees 289 A.L. Fetter Lectures on Correlation Functions ... see : ses 301 X-Ray Scattering .... seeeeeeeeeeees 302 wo-Body Correlations .....+ : cet eceseeeeesseeeesseeeeeseses 307 Neutron Scattering .. : : 313 Linear-Response Theory ....- 318 Dynamic Compressibility .... fee eeeeeeeteeeeetteeeeeees sesees 322 Model Calculations for Density Correlation Functions ... 327 Fluctuation-Dissipation Theorem .....ssssesesssssssseeeeeees = 332 Magnetic Phenonena; Perturbation Calculations 336 Magnetic Phenomena; Hydrodynamics Description seeeeeteeeeeeees 341 Light Scattering in Fluids . 347A. Aharony, Tel Aviv University A.L, Fetter, Stanford University ME, Fisher, Cornell University M.J. Stephen, Rutgers University H. ‘Thomas, University of Basel ORGANIZING COMMITTEE .A, Engelbrecht, University of Stellenbosch. F.J.W. Hahne (Chairman), University of Stellenbosch W.D. Heiss, NRIMS, CSIR, Pretoria (now at University of the Witwatersrand) RH. Lemer, University of the Witwatersrand W.S. Verwoerd, University of South Africa, Pretoria P. du T. Van der Merwe, AEB, Pelindaba ©.A. van der Westhuysen, CSP, CSIR, Pretoria Mrs. E. Blum (Secretary), CSIR, Pretoria PARTICIPANTS D. Bedford, University of Natal, Durban M.W.H. Braun, University of Pretoria J.H, Brink, AEB, Pelindaba J.D. Comins, University of the Witwatersrand, Johannesburg E.D. Davis, University of Cape Town P.R. de Kock, University of Stellenbosch O.L. de Lange, University of Natal, Pietermaritzburg S.J. Donovan, University of the Witwatersrand, Johannesburg C.A. Engelbrecht, University of Stellenbosch E,A. Evangelidis, AEB, Pelindaba A.G. Every, University of the Witwatersrand, Johannesburg D. Hyre, NRINS, CSIR, Pretoria G.M, Field, University of Cape Town G.C.K. Félscher, University of the Witwatersrand, Johannesburg P.J. Ford, University of the Witwatersrand, Johannesburg W.E, Frahn, University of Cape Town W.L, Gadinabokao, University of Bophuthatswana, Maf ikeng M, Gering, University of the Witwatersrand, Johannesburg F.J.W. Hahne, University of Stellenbosch S. Hart, NPRL, CSIR, Pretoria W.D. Heiss, NRIMS, CSIR, Pretoriavi J.J. Henning, AEB, Pelindaba J.D. Hey, University of Cape Town M.J.R. Hoch, University of the Witwatersrand, Johannesburg D.P. Joubert, University of Stellenbosch S. Klevansky, University of the Witwatersrand, Johannesburg P.J. Kok, University of Pretoria RH, Lemer, University of the Witwatersrand, Johannesburg P.E, Lourens, AEB, Pelindaba L. Matthews, University of Pretoria RE. Nettleton, University of the Witwatersrand, Johannesburg P.E. Ngoepe, University of the Witwatersrand, Johannesburg G.N. v/a H Robertson, University of Cape ‘Town F.G. Scholtz, University of Stellenbosch G.J. Shepherd, Rhodes University, Grahamstown L.C.A. Stoop, University of South Africa, Pretoria J.H. van der Merwe, University of Pretoria P. du T, van der Merwe, AEB, Pelindaba E. van der Spuy, AEB, Pelindaba . van Niekerk, AEB, Pelindaba W.S. Verwoerd, University of South Africa, Pretoria J. du P. Viljoen, AEB, Pelindaba D.H. Wiid, Rand Afrikaans University, JohannesburgPREFACE ‘The study of critical phenomena and phase transitions has received considerable attention during the past ten years, and many new achievenents have been made. Due to the smallness of our physics community, we in South Africa have not been able to participate in this endeavour in any significant way. It thus became apparent that in order to acquaint physicists in general with this field an advanced course on these subjects within our programme of sumer schools was very opportune. The second school, which had as its topic “critical phenomena," was held at the University of Stellenbosch from January 18 to 29, 1982 and was both well attended and enthusiastically received by students and practising physicists alike. We consider ourselves very fortunate in having had five outstanding experts in the field present the material in very clear terms. The enthusiasm and clarity of the lectures was surely a unique experience for many of the students, and such exposure to excellent physics is most Likely to be the best counter to the waning interest in pure science resulting from competition from financially more rewarding disciplines. on behalf of all the participants, I wish to thank all the lecturers for performing their task admirably. ‘This venture was only possible as a result of financial and organisational support from the Council for Scientific and Industrial Research (CSIR). The CSIR has sponsored these schools since the initiative was taken by the Organization for Theoretical Physicists (OTP) and the South african Institute of Physics (SAIP). ‘The venue in the university town of Stellenbosch was well suited for holding the course, and the use of the facilities, as well as other support, is gratefully acknowledged. These lecture notes consist of manuscripts either supplied by the lecturers themselves or, in two cases, compiled from notes by participants who had further lengthy contact with the lecturer concerned. In all cases these notes provide a very readable account of the courses presented. We are grateful to the editors that these notes can appear in the Springer series “Lecture Notes in Physics. Stellenbosch, South Africa F.J.W. Hahne May 1983SCALING, UNIVERSALITY AND RENORMALIZATION GROUP THEORY By Michael E. Fisher Baker Laboratory, Cornell University Ithaca, New York 14853, U.S.A. Lectures presented at the \dvanced Course on Critical Phenomena” held during January 1982 at The Merensky Institute of Physics University of Stellenbosch, South Africa Lecture Notes prepared with the assistance of Arthur G. Every Department of Physics, University of the Witwatersrand, Johannesburg, South Africa On leave at Department of Physics, University of Illinois ‘at Urbana-Champaign Urbana, Illinois 61801, U.S.A.1 2. 3. CONTENTS: Introduction Gritical Phenomena in magnets and fluids: Universality and Exponents 2.1 The gas-Liquid critical point 2.2 Universal behavior 2.3 Binary flutds 2.4 Critical exponents defined precisely 2.5 Specific heats 2.6 The order parameter 2.7 Fluld-magnet analogy 2.8 Magnetic susceptibility 2.9 Critical 1sothera Sealing 3.1 Introduction: thermodynamic functions 3.2 The classical phenomenological approach or Landau theory 3.3. The scaling concept 3.4 Scaling of the free energy 3.5 Fluctuations, correlations and scattering 3.6 The correlation length 3.7. Decay of correlations at and below criticality 3.8 Scaling of the correlation functions 3.9 Anomalous or critical dimensions: General definitions and relations Microscopic Models 4a 4.2 43 44 45 4.6 47 ‘The need for models Ising model Solution of the one-dimensional Ising model ‘The two-dimensional Ising model Ising model in three dimension The n-vector spin models Continuous spin models Series expansions Renormalization Group Theory Sel 5.2 5.3 Preanble ‘A renormalization group for the one-dimensional Ising model 5.2.1 Spatial rescaling and spin correlations 5.2.2 Unitarity Flow equations, recursion relations, and fixed points 5.3.1 Linearization about a fixed point5.4 5.5 5.6 5.3.2 A second variable and scaling General renormalization groups 5.4.1 The space of Hamfltontans 5.4.2 Renormalization group desiderata 5.4.3. Quasilinear renormalization groups Flows, universality and scaling 5.5.1 Universality, relevance and irrelevance 5.5.2 Continuous flows 5.5.3 The fixed point spectrum 5.5.4 Scaling of the free energy and hyperscaling The construction of renormalization groups 5.6.1 Kadanoff's block spin renormalization group 5.6.2 Nleneijer and van Leeuven's majority rule 5.6.3 Wileon'e momentum shell integration . Dimensionality Expansions 6.1 Transformation of the Hamiltonian 6.2 Computing with continuous spins 6.3 Implementation of momentum shell renormalization 6.4 The Gauseian fixed point 6.5 The renormalization group to order © 6.6 The n-vector fixed point 6.7 Some numerics 6.8 Further developments in brief ACKNOWLEDGMENTS Appendix A The Kac-Hubbard-Stratonovich Transformation Appendix B Details of the €-Expansion Calculation Appendix C Dimensionality as a Continuous Variable Appendix D —Hyperscaling and Dangerous Irrelevant Variables BIBLIOGRAPHY REFERENCES1, Introduction My aim in these lectures will be to describe some of the more interesting and important aspects of critical phenomena. I will, in particular, be discussing the ideas of scaling and critical exponents and emphasizing the idea of universality. Following this I will be dealing with the microscopic formulation of statistical mechanics and certain series expansion methods that have been extensively used in the past. These are not only applicable to critical phenomena, but are useful in other areas of physics and engineering as well. The main emphasis and focus of the lectures will, however, be on the collection of rather subtle ideas which underlie renormal tion group theory and its applications to critical phenomena. I will be approaching them in roughly two stages: from the microscopics will cone some Antroductory concepts; then I will be describing the general renormalization group ideas which are essentially topological in nature. I will be aiming, in describing these general concepts, at applications beyond critical phenomena. They have for example been used to handle the Kondo problem and to study various aspects of field theories. Finally, I will discuss some of the first practical successes of the renormalization group, based on the so-called “epsilon expansions". These expansions are generated in terms of the parameter c = 4-d, where d is the spatial dimensionality of the physical system. They were some of the first sweet fruite of the renormalization group ideas! What is the task of theory? It is worthwhile, when enbarking on theory to have some viewpoint as to what theory is. There are different opinions on this subject. Some people feel the task of theory is to be able to calculate the results of any experiment one can do: they judge a theory successful if it agrees with experiment. That is not the way I look at a theory at all. Rather, I believe the task of theory is to try and understand the universal aspects of the natural world; first of all to identify the universals; then to clarify what they are about, and to unify and inter-relate them; finally, to provide some insights into their origin and nature. Often a major step consists in finding a way of looking at things, a language for thinking about things ~~ which need not necessarily be a calculational scheme. This aspect of renormalization group theory, which I view as very important, is underplayed in a number of articles and books on the subject. "Shape here, the geometrical properties of the circle have been known for a long time, and are aspects I often regard as important. To make an illustrative point we tend to take them for granted. The ratio of the circumference to the diameter is called , which is only the first of many Greek letters that will be introduced in these lectures! Its value, which today we know as 3.14159265358... was, from very early times, felt to be the same for all circles; i.e., that it was a universal property. This is true if space is Euclidean; and, to a very high degree of accuracy, the space we inhabit is, indeed, Buclidean. The value of this ratio is of 1 course of great interest. The Bible has an unanbiguous statement! that the value of1 {8 3, although the people to whom that is attributed probably knew that it was not exactly equal to 3. This “Biblical” theory is the analogue to the so-called “classical” theory of critical exponents that will be referred to frequently in these lecture: We know that the Ancient Greeks already had very good inequalities for ™. Of course, the numerical value of 1 is now known to very many decimal places indeed, and there are nunerous series expansions which converge to the exact value, euch : ap Also, there are explicit formulae that relate to the other transcendental numbers, the most famous being (a2) In a similar way, in the theory of critical phenomena there is a set of important numbers, the critical exponents, and they are also believed to be universal in character. In these lectures evidence will be presented to show that this {s so. In addition sone formulae, in the form of series expansions, have been derived for these critical exponents, but, so far, only the first few teras in the expansions are known. Also, while the expansion (1.1) for ™ 1s convergent, the &~ expansions for the critical exponents are almost certainly not convergent in general, unless they are treated in @ special way. We will also see that there are a number of formulae like (1.2) which relate the various critical exponents to one another, although perhaps with not quite the mathematical rigor and generality of (1.2). These remarks more or less sum up the attitude I will be taking towards the subject matter of these lectures. 2. Gritical Phenomena in magnets and fluids: Universality and Exponents 2.1 The gas-liquid critical potnt The first critical point to be discovered was in carbon dioxide. Suppose one 2 examines? a seated tube containing CO, at an overall density of about 0.5 ga/ec, and hence @ pressure of about 72 atm. At a temperature of about 29° C one sees a sharp meniscus separating liquid (below) from vapor (above). One can follow the behavior of Liquid and vapor densities if one has a few spheres of slightly different densities close to 0.48 gn/ec floating in the system. When the tube is heated up to about 30° C one Finds a large change in the two densities since the lighter sphere floats up to the very top of the tube, i.e., up into the vapor, while the heaviestfluid Fig. 2.1 (a) (p,T) dtagram for a typical physical eysten; (b) corresponding plot of particle number density p versus T. The ver: “tie-lines” Link coexisting liquid and vapor densities, and span the region of liquid vapor coexistence.one sinks down to the bottom of the liquid. However, a sphere of about “neutral” density (in fact “critical density") remains floating “on” the meniscus. There is, indeed, still a sharp interface between the two fluids, but they have approached one another closely in density. Further slight heating to about 31°C brings on the striking phenomenon of critical opalescence. If the carbon dioxide, which is quite transparent in the visible region of the spectrum, is illuminated from the side, one observes a strong intensity of scattered light. This has a bluish tinge when viewed normal to the direction of illumination, but has a brownish-orange streaky appearance, Iike a sunset on a smoggy day, when viewed from the forward direction (.e., with the opalescent fluid illuminated from behind). Finally, when the temperature is raised a further few tenths of a degree, the opalescence disappears and the fluid becomes completely clear again. Moreover, the meniscus separating “Liquid” from "vapor" has completely vanished: no trace of it remains! All differences between the two phases have gon indeed only one, quite homogeneous, “fluid” phase renains above the critical temperature (1, = 31-04° ¢). These phenomena are best interpreted in the pressure-tenperature (p,T) phase diagram shown in Fig. 2.1. The first three stages are represented by the points a, band c on the vapor pressure curve. Note that T, and p, are the critical temperatures and pressures respectively at which critical opalescence is observed. As the temperature is raised further, the system follows a contour of constant overall density (the “critical isochore"). | The whole process is completely reversible. Significantly, it is possible to go from liquid (point 1) to vapor (point 2) either smoothly via a route along which the properties of the fluid always change smoothly and continuously, or through the vapor pressure curve, at which first order transition takes place with a discontinuity in density, internal energy, etc. Any point inside the shaded region of Fig. 2-1(b) corresponds to liquid and vapor coexisting with one another. As the critical point {s approached the tvo densities, o44q(T) and pygp(T) becone closer and closer to each other til they match at T= Te 2.2 Universal behavior One finds that the actual variation of py4q(T) and Pyap(T) is close to universal for gases such as argon, krypton, nitrogen, oxygen, etc., in the sense that if the temperature is normalized by the critical temperature, T,, and the fe density by the critical density, p,, then the data for the different gases all fit very nearly on the same coexistence curve. The shape of this coexistence curve will be one of the first objects of our investigation. The simplest curve which has the T vs p is, of course, the parabola y = Ax’, The assertion that the coexistence curve is parabolic (in the same basic shape as the coexistence curve graphed critical region) in fact represents the “Biblical” or classical theory of thecoexistence curve. At firet sight, it seems to be @ most natural and unprejudiced starting point. But what really is the shape of this curve near T,? ‘That is the question one must ask! To that end we introduce here a vartable that will be greatly used, namely, the reduced temperature (2.1) which measures the deviation of the temperature from critical in dimensionless Now as T approaches T, from below, the difference between the liquid and g units densities, py4q and pyap respectively, is going to vanish as, we might reasonably 114 'vap expect, some power 6 of |t|. Thus we write 8 ate. PrtgPyap” (HIP a8 T+ Toe (2.2) The exponent 8 is the first of the eritical exponents that will be introduced in these lectures. It is the analogue of because it directly describes the shape of the coesxistence curve. From the way the parabola is oriented in Fig. 2.1(b), ve see that the classical or “Biblical” theory prediction is simply f=1/2. How does this compare with the value of 8 measured in the real world? The experiments that have been done in this connection are some of the most precise experiments ever performed in Physics. A notable example is provided by the work of Balzarini and Ohrn? who measured the coexistence curves for xenon and sulphur hexafluoride using very sensitive optical methods. These two fluids are obviously very different chemically but, nevertheless, their critical behavior 1s found to be essentially the sane. The data on the density jump 4p = o4q ~ Pyap Span the range fron t = 3 x 10% down to 3 x 1076 and on a log-log plot lie very accurately on two straight and parallel lines. This first confirms the power law behavior and then yields a value of 8 which is quite close to 1/3. The precise value lies somewhere in the interval 0.32 ~~ 0.34, perhaps closer to 0.32. Despite the experimental accuracy and the great range of the data one cannot, unfortunately, actually determine such critical exponents to much better than 40.02. We are certain now that it 1s not a simple fraction, or at least not a very simple fraction like 1/2 or 1/3! Clearly, therefore, the “Biblical” value is quite outrageously wrong. Finally, in Line with 1 being independent of the size of the circle, it is found that 6 is also quite independent of the type of fluid; the same values are found for water, a highly associated liquid, for liquid metals, and for the ‘quantal liquids’ helium three and four at their liquid-vapor critical points.2.3 Binary fluids Another type of system which has been much investigated is that of a mixture of two chemical compounds, say A and B, that at high enough temperatures are mutually soluble, but at lower temperatures separate out into two phases as ofl separates from water, which we will call a and @ (see Fig. 2.2). There are a great many combinations that can be used: organic liquids such as aniline and cyclohexane or carbontetrachloride and perfluoroheptane are favourites because the interesting behavior occurs (under atmospheric pressure) at temperatures close to room temperature. The vapor phase is usually present, a showm in Fig. 2.2, but plays no essential role. The denser phase at the bottom could be, for example, A-rich, while the less dense one floating above it would then be Brrich. As the temperature is increased @ liquid-liquid critical point or consolute potnt is reached and critical opalescence is exhibited just as for a one-component fluid. Beyond this point only a single, homogeneous liquid phase existe. Vapor B-rich phase A-rich phase Fig. 2.2 Illustrating phase separation in a binary liquid mixture of two chemical species A and B. Now what shovld one focus on instead of the density difference? We will use syabols such as x to denote the mole fraction of A molecules in the Acrich phase « and so on. As the critical temperature 1s approached from below one observes that the composition difference between a and 8 phases varies as « 8 xe ~ lel + (t+ 0-). (2.3) The question is “Does ® have the same value as before?” The answer is an10 unequivocal “yes” as can be seen from experiments on very many binary fluid systens (including molten metal mixtures). An interesting comparison has been published by Stenko*, He finds, for example, that on a normalized log-log plot the coexistence curve for CCl, and C7F,, shows @ form which is almost indistinguishable from that of the Liquid-vapor coexistence curve for G02, so that 8 again lies close to 1/3. Stenko and coworkers also studied the metal-annonia systems in which alkali metals Na, Li and Ca are dissolved in NHy. At first sight these mixtures appear to provide an exception to the @ = 1/3 rule. For temperatures deviating from (below) T, by from 1% to 10% (1.e., t = 0.01 -~ 0.1) the coexistence curve on a log-log plot has a steeper slope than for other systems and, indeed, seems to conform to a 8 = 1/2 relation as predicted by the “Biblical” theory. But accurate data that are taken closer into the critical (or consolute) point fall clearly into line with all the other systems: the slope changes quite rapidly around t = 0.007 to 0.009 and decreases to yield again 8 = 1/3. S0 we are forced to accept this universality of behavior, but we learn that the universality does not extend indefinitely out of the critical region. Indeed it is really a matter of extrapolating in towards the critical point if one wants to determine the true, universal, asymptotic behavior. So when I discuss critical behavior it 1s always a matter of approaching clo enough to the critical point. It 1s worthwhile to embody this point in @ formal definition of a critical exponent which can then be used for more exact and rigorous theoretical arguments and analyses. 2.4 Critical exponents defined precisely Generally, when we say a function f(x) behaves like x, or write f(x) ~ 2 as x + OF, (2.4) it will be taken to mean that Lim In{£(x)) mOF ine =X (2.5) d In this way we can avoid introducing a constant for the coefficient of x as would be essential if we wrote f(x) = Ax or f(x) = x. At a more subtle level suppose we have a function such as wd £(x) = Alinx|"x*. (2.6) This does not vary as a simple or “pure” power law but rather has a “confluent” Logarithnic singularity. Fron a theoretical viewpoint one can still use eqn. (2.4), and in this way one obtains a critical exponent equal to A. Thus even functions of" this type with more complex singularities are covered. One of the important contributions of renormalization group theory is that it reveals the circumstances under which such logarithmic factors should be anticipated. One must always expect, of course, that over any finite range there will be gone correction terms: thus even for an asymptotically pure power law one will generally have fx) =a + (2.7) where the confluent “correction” exponent, 9, may well be le than unity (although it must be positive for the form written to make sense). On a log-log plot corrections such these can and do actually alter the slope and lead to erroneous values for measured critical exponents. The most serious correction terms are those where 8 <1, the smaller the value of 9 the wor: the problem. In fact, values of around 1/2 are expected on theoretical grounds in many real situations. This assertion reflects another valuable contribution of the renormalization group since it has enabled us to give a sensible estimate of the exponent 8 and to explain why this sort of behavior {s what one should expect in most circumstances. ‘There have been people in the past who have questioned whether nature really is required to conform to power law behavior near a critical point. The evidence, both experimental and theoretical, 1s now compelling that, apart from logarithmic factors in spectal cases and certain correction terms, power law behavior is the ule, One would have to be a brave scientist indeed to hold out against this conviction and this point. Nevertheless there are still those — some would call them “cranks” -- who argue that perhaps the “Biblical” theory is still correct if one goes really close to T,, so that 8 = 1/2 after all! However, I am afraid that in sctence, new and more correct ideas often win out only after their opponents die or retire. Evidently many people are not as open to rational conviction by new thoughts, as might be desirable! Another problem that arises in the handling of experimental data is that the critical temperature T, is, of course, not known in advance. Usually one treats T, in the expression t = (T-T,)/T, as a fitting paraneter. When the data extends over several decades, the data close in to the transition point will sometimes be used to determine T,, while that further out then serves to determine the critical exponents. Sometimes T, will be determined separately from both sides in sintlar or distinct experiments. All in all, great care has to be exercised when interpreting even the very best data if one is not to assign misleadingly small “error” estimates to parameters such as T,, @, and the amplitudes A, etc.12 2.5 Specific heats In 1963 Voronel’ (then in the Soviet Union) and his coworkers made some historically important measurenents of the specific heat at constant volume, Cy(T) of argon in the vicinity of its critical point. More precisely, they observed the specific heat at constant overall density along the critical isochore p = p,. Below T, the system will, as seen, then consist of a mixture of vapor and liquid, and the proportions of the two will actually change as the temperature is varied. So this “specific heat” actually contains a latent heat contribution. Nevertheless that 1 both experimentally and theoretically, the most appropriate function to measure for the study of critical behavior. Now the “Biblical” or classical theory predicts that (I will not say “this” anymore) the specific heat merely has a jump discontinuity at the critical point, i.e, G,(T.-) # Cy(Ty#). Actually Cyto) > G(TgH) 18 predicted as indicated by the dashed curve in Fig. 2.3. Voronel' was the first one to do sufficiently careful and accurate measurenents to show unambiguously that this was not so! On the contrary, Cy(T) rose up smoothly but very steeply on both sides of T, as sketched in Fig. 2.3. Asymptotically the variation has the form et) ~ |e], (t+ OF), (2.8) where the specific heat exponent a has a value in the region of 1/8 to 1/9 for most fluids. Because of the small value of a, correction terms now assume much greater importance and make a hard to determine precisely. Also one might question whether Cy does, indeed, diverge to infinity, or whether it just has a sharp spike or cusp at Ty. On this latter question microscopic mdels are able to provide us with sone definite guidance. These models come in various shapes and sizes: but the most famous is undoubtedly the Ising model, which I will be discussing in more detail later in these lectures. Onsager's celebrated solution of the 2-dinensional Ising model in 1944 gave the specific heat as Cy(T) = A Int] + finite “background” terms. (2.9) The singular behavior is carried mainly by the leading logarithmic term (although terms like tin|t| appear in the “background”). As 16 readily confirmed by application of the formal definition (2.5), a logarithaic divergence corresponds to the limiting case of a + 0+. [Consider the function t,(t) = (t|“-1)/a -] To draw attention to the fact that the logarithm is present, this case is usually reported as a = 0 (log).cy} Fig. 2.3 13 Specific Heat b Nn. | divergence to + @ 1 1 1 i 1 1 4 — temperature Te T Sketch showing the variation of the specific heat, Cy(T), of argon and other fluids through the gas-liquid critical point. The dashed curve represents the prediction of the classical (or "Biblical”) theories.14 One small detail that Fig. 2.3 suggests one should take into account is that the specific heat does not mirror itself around the critical point. Thus one a’ for T <7, and a for T> 7. The should, properly, define two exponent: convention is that primed exponents refer to T <7, and unprined to T > 7, (except where, like 8, the definition makes sense only for T < T,). Nowdays it is rather well established on both experimental and theoretical grounds that a = a’, so the distinction is often dropped unless one has reason for being circumspect. Modern experiments on critical specific heat obtain temperature resolutions of 1076 or 1077 in t. Some of the best experiments are those of Ablers® on liquid helium at its lambda point, T, = T, = 2.18 K, where the normal fluid becomes superfluid. The transition {s seen to remain sharp down to a tenth of a microdegree. More recently Lipa’ has pushed the resolution still further down to only tens of nanodegrees. The specific heat seens to continue rising down to these very small deviations from T,. Tt 18 worthwhile asking the question at this point if, with continuing experimental refinements, one can expect to observe the specific heat continuing to diverge indefinitely close to T,. Naturally, precautions must be taken to allow for gravity and other small disturbing factors. However, ultimately the basic theoretical answer 1s "No, the specific heat cannot increase without bound”. The Feason is that in the laboratory one would always be dealing with a finite syste with a finite number of atoms confined in a bounded region of spac A perfectly sharp phase transition can take place only in a truly infinite system, 1.e., in the thermodynamic limit where the system is infinitely large in extent but its density, pressure, and all other intensive quantities are fixed and finite. However large a system is in practice, it will still be finite and, ultimately then one will reach the point where the specific heat singularity is seen to be rounded off. Experiments deliberately done on emall samples certainly show these rounding effects. So in talking about @ phase transition one really should always have in mind the thermodynamic limit. The specific heat anomaly at the lambda transition in Het is now believed to be very close to logarithmic. Thus Ahlers quotes a = a’ = ~0.02 +2 (the uncertainty being in the last decimal place) signifying that a is probably very slightly negative. This suggests that the specific heat does not quite diverge to » but rather cones up to form a sharp cusp at which point C(T,) is finite but the slope (ac/aT), 18 infinite. Similar behavior is also observed at magnetic phase transition: being the specific heat of the ferromagnet nickel near its Curie or critical a notable c point, T,. Magnetic systens are in many ways much simpler to think about theoretically because magnetic field H=0 {sa point of symmetry. One finds that the zero-field specific heat of nickel displays a sharp cusp, but it is much less strong than in the case of superfluid helium or some of the other fluid systems. In this15 case, and that of other magnetically isotropic magnetic systems, one finds that a is 10 to ~0.15, high definitely negative although still quite emall say a = precision being again difficult to attain. 2.6 ‘The order paraneter In the case of simple fluids the parameter of apparently central interest is the density, p. Following Landau's general conception of phase transitions, we nane this special quantity the order parameter and denote it generally as ¥. So for single-component fluids we write ¥ = p. For fluid mixtures we saw that what mattered was the difference between the mole fractions, 4x, which measures differences in composition: so here we have ¥ = Ax. For superfluid He’, the crucial theoretical concept, which enbodies our understanding of superfluidity, 1s an effective macroscopic wave function, ¥ = ¥' + 1 y!'. As a wavefunction this has both real and imaginary parts. While p and Ox are both simple scalar quantities, a complex number is best thought of as a two-component vector. Thus the superfluid order parameter, ¥ = 9, is a two-component vector which has the symmetry of a circle, i.e., can point in any direction in the complex plane. Tt is the phase of which 1s in fact responsible for the existence and nature of superfluidity. In the magnetization, f, which should be the order parameter. In the case of a magnet Like nickel, the magnetization can point freely in any direction; i.e., nickel is of ferromagnetism, there are various possibilities, but certainly it is the spatially, highly teotropies then the magnetization can be thought of as a three- component vector H= (My, Mys My)s In summary, we see that the order parameter, ¥, has a tensorial character which may depend on the class of systems considered. Theoretically it is natural to distinguish between these various cases, and the renormalization group has enabled us to make this distinction meaningful and effective. In particular we often refer ton, the number of components of the order parameter. Then we hav n= 1 for simple fluide, binary fluide, uniaxial ferromagnets, binary alloys, ete. n= 2 for superfluid He’ and He? + He" mixtures, XY-magnets (easy plane of magnetization). n= 3 for isotropic magnets, etc. As regards values of the critical exponents, none of which conform to “Biblical Specifically, one has a(n=1) = 0.11, a(m=2) = 0.0 and a(n=3) = 0.14 + 4, Similar or classical theory, there is found to be a subtle dependence on n.16 slight differences are found for the critical exponent 8, viz. (n=l) = 0.32, B(m=2) = 0.34 and B(n=3) = 0.35 - 0.37. The n = 2 value applies to xY-magnets but ible to the corresponding superfluid order parameter is essentially inacce experiment. Clearly then, the symmetry or tensorial character of the order Parameter is important. The three cases described above are often referred to as Ising-like (n=l), X¥-1ike (n=2), and Heisenberg-like (n=3). Larger values of n are not just of theoretical interest; they are also required for describing real physical systems, in particular various magnetic crystals of more complex structure and symmetry. 2.7 Fluid~magnet analogy The close analogies that exist between flutds and ferronagnets are worth enphasizing, even though ferromagnets have an intrinsic symetry that makes then easier to think about. Conjugate to the order paraneter, ¥, in any thermodynamic system, is a “thermodynamic field” variable, h. In the case of fluid the pressure, P, has traditionally been treated as this conjugate variable, but often it is better to regard h as the chemical potential, u. The pressure p or chemical potential u 18 the variable that directly allows one to alter the density (at constant temperature). The analogous variable for a magnet should therefore be the magnetic field H, which fs the variable primarily coupled to the magnetization. Fig. 2.4 illustrates clearly how far the analogy can be taken. In the case of the magnet, in the (h,T) plane, there is a line of first order transitions separating the “up” and “down” ferromagnetized states; this line ends at the critical (or Curie) point. The firet order transition line is analogous to the vapor pressure curve, but differs from it in one minor respect in that it is entirely confined to the H=0 or T-axis. This, of course, is a consequence of symmetry under H+ -H. In the (¥,T) plane there 1s a coexistence curve in both cases. Inside this curve the magnet breaks up into domains; this 1s analogous to gas-liquid coexistence in fluids. For the magnet the coexistence or “spontaneous magnetization” curve is symetric about the T-axis while for the fluid this symetry is apparently absent. Below T, the order Parameter variation for the fluid is given by 444 - Pyap ~ |t/® while for the magnet it is Mo(T) ~ |t|*®, where the spontaneous magnetization should be defined as Mj(T) = Lim M(H,T). (2.10) Hoot This careful definition of Mp(T) is neces: values depending on whether H=0 is approached from positive or negative values. The ry because M takes different limiting specific heat exponent is also defined in an analogous way for the two systems, and 80 on. Thus while most emphasis will be placed on magnetic systens, analogous effects and similar results hold for other types of systems in nearly all casetht Hi te oT Fig. 2.4 Phase and coexistence diagrams illustrating the magnet-fluid analogy. Note magnetization corresponds to density and magnetic field to pressure or, better, chemical potential.‘The question of how the perfect symmetry of the spontaneous magnetization curve is reflected in the less than fully symmetric nature of the fluid coexistence curve is a fairly subtle one. For the magnet the natural field varfable to take, because of the symmetry, is H. One suspects that for a fluid the most suitable variable by analogy should be he ppt), (@.y where p,' is the Limiting slope of the vapor pressure line at T,. In this way h would measure the deviation from the Limiting tangent (shown dashed in Fig. 24), which one expects might be the analogue of the H=O symmetry axis of the magnet. This 1s sometines called a scaling axis. A remarkable feature of the coexistence curve is that the line of mid points between the liquid and vapor phases is surprisingly straight. Furthermore, one can clearly define two different exponents, B_ and 84, with respect to deviations below and deviations above critical density, Per A obvious (or known) symmetry between liquid and gas that should tell us a priort that these two exponents should be the same; yet to an exceedingly high degree of » for the vapor and liquid sides of the coexistence curve. There is no accuracy they are identical in value! Somehow the system builds itself an asymptotic symmetry from a Ham{ltonian which does not, in the first place, possess this symmetry at all. Again, the renormalization group is able to explain how a system is able to build up a symmetry on approach to a critical point, and to decide when a symmetry can be built (or, on the contrary, when a weakly broken near symmetry of the Hamiltonian is amplified). 2.8 Magnetic susceptibility. Above T, the spontaneous magnetization of a ferromagnetic material is identically zero, but magnetization can be induced by applying a magnetic field, H. Fig. 2.5 illustrates the type of isotherms observed. The isothermal susceptibility 1s defined quite generally as a function of H and T by x(t) = (3). (2.12) T One usually measures, and is most interested in, the so-called init: susceptibility lim xg) = aoe XqCTD, (2.13) which measures the slope of the magnetization isotherm at zero field (as shown by19 Fig. 2.5 Typical ferromagnetic magnetization curves (isotherms) above T,, at T, ‘and below T, (for a scalar,Ising-like or n=l system). the tangents in Fig. 2.5). In practice one often drops both the superscript o and the adjective “initial” and just’ refers to “the _—_ susceptibility”. Clearly x measures the ease of magnetizing a ferromagnet and hence is expected to grow large and, indeed, diverge at the Curie point where, after all, a ferromagnet essentially magnetizes itself! This divergence can be seen in Fig. 2.5: the slope of the critical, M7, isotherm is actually infinite at zero field. For theoretical Purposes it © is’ usually —conventent_ © to.~= define the_-— reduced susceptibility x = xq/xq 4° eet ideal paramagnet (with no spin-spin interactions). Evidently, x, which 1s > where x, is the isothermal susceptibility of an dimensionless, measures the enhancement in magnetic responsiveness caused by the interactions, which are, of course, responsible for the ferromagnetic critical behavior. The analogous reduced susceptibility for a fluid is x = K,/K, @°*?, where (2.14)is the isothermal compressibility of the fluid and Kiel = i/p ts the corresponding quantity for an ideal gas. Since, as explained, x measures the ease with which the order parameter 1s changed in response to the conjugate field, it is often known as the response function. (See also the lectures by A. L. Fetter). The divergence of x(T) at criticality is very strong and is characterised by an exponent Y defined as expected via x~ rey, (to0#s h = 0). (2.15) Me ured values of y are typically y(n=l) = 1.23 —~ 1.24, y(ne2) = 1631 1.32 and y(n=3) = 1.35 — 1.38. In the case of superfluid He’, one does not know how to measure x: thus only the exponent a can be measured (of the thermodynanie Properties we have defined). As can be seen, y has a small n-dependence, but in all cases deviates markedly from the “Biblical” value which is simply y=1. Below 7, the situation 1s more complex. Even at W=0 there is a nonzero spontaneous magnetization, Mo(T). Nevertheless, (as mentioned), one can still define the initial susceptibility as the limiting slope of the magnetization curve when H+ O+, The temperature dependence of x, so defined, provides one with the further exponent y', These last remarks apply, however, only to the Ising-like case of nel, If ne2 or 3, so that a continuous (rotational) symmetry is present it can be shown theoretically, although experimentally it {s not so easy to observe, that this Limiting slope 1s infinite, so that xq(T,H) diverges as H+O+ for T < T, and the exponent y' cannot be defined in the usual way. 2.9 Critical isotherm ‘The order parameter variation on the critical isotherm is generated by fixing the temperature precisely at T,, varying the order field, h, and observing the change in ¥, t.e., Mor as the case may be. For a magnet one finds that for small H this varfation is given by (see Fig. 2.5) mete, Go, mr), (2.16) which defines the critical exponent §. Values of 6 are typically: (n=l) = 4.8, S(n=2) = 4.7, and 6(n=3) = 4,6, These should, perhaps, be regarded as more theoretical than experimental, since 6 is extremely difficult to measure accurately owing to the steepness of the critical isotherm. The classical value is 6 = 3 which corresponds to a cubic curve for the critical isotherm. Of course, this is just the simplest analytic function which has the correct shape. Naturally the critical {sotherm near a fluid critical point displays completely2 analogous behavior. ‘The relation is often written in reverse form as 8 Ip-p,| ~ lael”s (ret), (2.17) where 4p = p-p,, but this clearly corresponds precisely to the expected magnet-fluid analogy. Likewise, “Biblical” theory (in this case the original prophet is van der Waals) predicts 6=3, a cubic relation, but experiment yields 6 = 4.2 to 4.8. 3. Scaling 3.1 Introduction: thermodynamic functions The “Biblical” or classical theories break down completely in the region of a critical point. What then, can replace them? It turns out that the simplest phenomenological theories that come anywhere close to explaining critical behavior embody the concept of scaling. In order to make the discussion reasonably comprehensive one needs to couch it in terms of the full thermodynamics. Let us consider a ferromagnet since its symmetry allows us to make certain convenient (but inessential) simplifications. The Helmholtz free energy, F(T,H), is associated with the basic differential thermodynamic relation aF = -SdT ~ Mai, GD where S is the total entropy. From this one can, by means of a Legendre transformation, generate the alternative free energy function, A(T,M) = FH, and it ie then a simple matter to show that the basic differential relation becomes dA = -SaT + HEM. (3.2) The magnetic field and susceptibility are obtained from A by differentiation according to and (3.3) Note that the susceptibility will diverge when T+ T,, but it 4s intrinsically non- negative: Indeed a negative static compressibility or magnetic susceptibility is thermodynamically inconceivable. This is equivalent to the statement that the free energy A as a function of M must be a convex function: although the graph of A versus M can have a flat portion, ite curvature must, otherwise, be strictly positive (See Fig. 3.1 below).2 3.2 ‘The classical phenomenological approach or Landau theory. The simplest type of phenomenological theory in this context derives from mean f1eld theory; {t was developed to a fine art by Landau and now frequently goes under his name. It consists, firet of all, in identifying the order paraneter, ¥, (physically if this 1s possible but otherwise just as an abstract quantity), and then expanding the appropriate free energy as a Taylor series in powers of the order parameter. For a magnet the issue is straightforward: we have ¥ = M and the power series expansion reads ACT = Ag + ACD + ACH #34) By symmetry under M ++ -M no odd powers of M can be present. At high temperatures this expansion can be justified for all reasonable models on fully rigorous grounds, but near T, it turns out to be dangerous! By differentiating twice one obtains the Anverse susceptibility, which in zero field above T, 1s thus given by = 2A,(1) for T>T,, (HM = 0). (3.5) The next assumption ie that the coefficients Aj(T) can also be expanded in powers of t = (T-T,) so that, in particular, we may write = Ay g + IA, t+ O(t7). (3.6) When T+ T,+ the susceptibility, by definition of T,, diverges to infinity, so that x71 + 0 as t+ O+, and hence Ay 9 = 0. The predicted behavior of x near T, is thus x= C/t ast + OF, (HM = 0). (3.7) This, of course, corresponds to y = 1. The fact that this theory gives an incorrect value for y can be traced directly to the unjustifiable assumption that A(T,M) can be expanded in a power series near and, indeed, at a critical point. Neverthele: this seems to be a very natural assumption of the sort which is frequently made in physics and engineering. Furthermore, it can also be shown to be the essentially imevitable outcome of any of the wide variety of more microscopically-based mean field theories that have been proposed in this and many other related contexts. In spite of the evident shortcomings of the classical phenomenological theory, let us continue to explore its consequences by considering the effect of the term of fourth order in M. Its coefficient is AD = Ay g + (4) = Gu + CE), G.8)23 _ Mo +My M Fig. 3.1 Variation of the free energy A(T,M) according to classical phenomenolog- ical theory. The non-convex section of the isotherms for T < T, must be orrected” by drawing in the flat, tangential segment, so forming the “convex cover” of the underlying, approximate function.24 where the replacement of Ayo by Zu 1 purely @ matter of convenience, We will assune u > 0 to ensure thermodynamic stability (although, in fact, the case u <0 is required for dealing with tricritical points). Let us now examine the equation of e, which 1s the relationship connecting T, H and M, near T,. It {8 obtained by differentiating A with respect to M and is easily seen to be of the form He Mot + ut), (3.9) where we have put 2A); = ¢ 80 that, from (3.5) and (3.6), Ag(T) = Let. For a fluid the corresponding equation would, for example, follow from van der Waal's equation with M replaced by p - 9, and H by p- Pos on setting t = 0, we obtain the critical isotherm as H~ M? and, thence, the For T < T, and H+ 0- one obtains an equation with erroneous prediction § = three roots, namely, M=o ‘ and Mm £ MCT) = Ble] ~ e with Be (c/uy’ . (3.10) The first root turns out to have a higher free energy than the other two (see Fig. 3.1) and therefore is of no real physical interest. The other tuo roots provide two equivalent states of equilibrium spontaneous magnetization. We see clearly that the 1 predicted value of the exponent 8 is , the incorrect classical result. One of the difficulties of the classical theory is associated with the necessary convexity of the free energy. If one follows through in graphical terms the arguments just presented, one obtains for the variation of A as a function of M for various values of T the resulte shown in Fig. 3.1. Above T, the variation predicted by (3.4) de quadratic in M for small M and obviously convex. At T, the coefficient of the quadratic term vanishes and A hi on M. The graph ds extremely flat but still convex as it should be. Below T., however, the coefficient of M2 is negative, so the curve starts off at M= 0 like an a pure fourth power dependence inverted parabola, although it is ultimately turned around by the positive quartic term. The resulting concave portion of the curve for small M is clearly unphysical, and this should be taken as an indication that the theory has gone wrong! This defect in the theory can, however, be repaired in a more-or-less ad hoc way by means of the "Maxwell construction”, which essentially consists of drawing a straight line between the two minima at -Mo(T) and 4%)(T). This process generates the so-called “convex cover” of the original A(M) plot. But what can be done about the totally incorrect values of the critical exponents that cone fron this theory? Can anything8 be salvaged? The temptation is to somehow or other graft on the correct values! This desire brings one naturally to the idea of scaling. 3.3 The scaling concept There are several ways in which the desired modifications of Landau theory can be introduced. One of the earliest and most direct approaches was that of Widon. The gist of the argument goes cl follows: Consider the exponent 8. The incorrect ical value of Yj, arises from the presence of the M? term in (3.9). Let us therefore try to patch up the theory by replacing we by 1/8 where 8 is now a free parameter that can be fitted to experiment. If this were all, the equation of state would thus become He mee + unl!) Gab and so the spontaneous magnetization below T, would come out correctly! Likewise, however, one might try to get the susceptibility exponent, y, right by replacing t in (3.9) by tY, For T> 7, it follows that H~ cMtY and so x ~ t”Y as desired. If this modification 1s to apply also for negative values of t then t should obviously be replaced by |t|. This, however, is easily seen to lead one into trouble since it introduces non-analytic behavior into the equation of state everywhere on the critical isotherm t = 0 (even for H or M nonzero). This has quite unphysical consequenc since, in fact, the equation of state is, both theoretically and experimentally, completely free of singularities on crossing the critical isotherm avay from H=M~= 0, Similar problems arise in (3.11) for small values of M above T,, where the expansion (3.4) should be valid but 4s not unle; integer! 1/8 1s an even To avoid these problens let us rewrite (3.9) by dividing through by clt|2/? to obtain the equivalent fore (3.12) where we have replaced 3/2 by A while B and D are simply related to the original constants ¢ and u, However, fron this point on we may release A fron its constrained value and treat it as a second free exponent, which, hopefully, can be adjusted to get, say, y correct. Now the spontaneous magnetization varies as |t|® so the quantity M/B|t|® can be viewed as the magnetization scaled by the spontaneous magnetization, Mj(T), Similarly, on the left hand side of (3.12) we have the magnetic field, H, scaled by a characteristic power of the temperature, nanely, |t|*. Next we notice that the full equation of state is a relation connecting M,T and H which we could express as M=/{(T,H). Widon's original suggestion was that,26 perhaps, when T+ T,, and M and H becone small the equation of state in general simplifies if M is replaced by the suitably scaled magnetization, namely M/B|t|° and H is replaced by a suitably scaled field, namely Dx/|t|*. This evidently applies to the special case (3.9) which embodies classical theory but perhaps it also holds asymptotically for the true equation of state in the critical region! More explicitly, the nature of the proposed simplification is that, in the critical region, the equation of state reduces from a function W{(T,H) of two variables to function of only one variable, but which relates the two scaled variables together. In other words, we make the scaling postulate H+ BD), (3.13) [el lel where Wis sone sufficiently general function of a single argunent. This assertion, the scaling ani we shall see, a renarkably successful guess! tz, must, at this stage, be regarded purely as a guess, albeit, as In the classical theory we have @ = 1/2 and A = 3/2 and these values are universal for all systems: they simply arise from the integral exponents in the assumed Taylor series expansion. Additionally in classical theory, as one sees from (3.12), the full scaling function, Wy), is generally that § and \ are universal exponents and W(y) is a universal function, universal. Thus we may expect more even though the values will differ from their classical counterparts. On the other hand, the parameters B and D, like T, iteelf, must reflect the details of the amplitudes. The particular ferromagnet: thus they are referred to as non-univer exponent 4 is often termed the gap exponent. Let us now examine some of the implications of this simple but, in fact, far— Teaching assumption. The susceptibility for t > 0 and H+ 0 te given by 8-4 x= GR = [e]P sow'(o), (3.14) on ‘T,R=0 ~ where W'(0) must just be some number. Since, by definition, we have x ~t’, we see that AaB+y. (3.15) This shows how A should be chosen to give the right value of y. Otherwise it tells us nothing new. To find a new result let us look at the critical isotherm, T= T,, for which Purpose the limit t + 0 must be studied. In this limit the scaled magnetic field evidently diverges since we t+o (3.16)a In the spirit of the enterprise let us then assume that Wy) also varies as some power when y becomes large, i-e., suppose Ww yr as yee, (a7) where q, and A are constants, It follows that am fel? teu, ye, (3.18) When t + 0, the temperature variable should drop out of this expression since M then becomes a function of H only; consequently we must demand 14 = 6 which fixes the exponent as A= B/A. (3.19) Thus there 1s, in reality, no free choice of 1! Moreover, from (3.17) we now see that MoH; but, by definition we have M~ HS for T= Ty. Thue we conclude § = 1/, and hence a14+%, 1+h (3.20) This novel equation relating the three exponents 8, y, and 6 is known as Widon's relation. It is our first nontrivial scaling law or, simply, exponent relation. In a similar way, by integrating M = -(3F/H)q to obtain the free energy F(T,H) and then differentiating with respect to T one derives expressions for the entropy and specific heat and hence establishes the so-called Essam-Fisher relation at + 2B y= 2 3.21) A little further investigation using the fact that there must be no singularities as one crosses the critical isotherm, t = 0, at nonzero H or M reveals that one must also have ana, (3.22) for the specific heats above and below T, and yor (3.23) for the susceptibilities. Evidently there are four relations connecting the six exponents a, a', 8, Y, Y' and 6, and go only two of them can be independent: this is a striking prediction, by now verified many times experimentally!Fig. 3.2 28 (a) A schematic plot illustrating equation of state data. i.e., M versus H isotherns, for a ferromagnet through the critical region; (b) a scaled plot of the same data illustrating the “collapse” of the data onto a single scaling function W(y), with two branches W(y) and We(y) corresponding to t 2 0.29 The success of scaling can be illustrated graphically by replotting equation of state data for magnets, fluids, etc. in scaled form. Thus consider Pig. 3.2(a) where M versus H isotherms are sketched for a ferromagnet. This data may be replotted in scaled form as M/|t|® versus y = H/|t|®° (recalling that 4 = 86 by the scaling laws) for appropriate choice of the exponents 6 and 5, which might be determined separately from the spontaneous magnetization curve and critical isotherm. Scaling is confirmed if one observes, as in fact is found ,® a “collapse” of the data for the different isotherms onto a common locus, which represents the scaling function W(y). Actually in this representation one finds two branches, asymptotically matching as y + =, corresponding to Wy(y) and Wy(y), the scaling function for T % Ty When the procedure is repeated for different magnets one finds similar results with, indeed, the same scaling function? up to different scaling amplitudes B and D. The scaling function that emerges for fluids is, likewise, the same for all fluids, and furthermore it agrees, with that found for magnete!! do the exponent: To sum up then, the scaling postulate proves to be a remarkably successful guess. Our theoretical task from here on is to set scaling theory in a broader context, to explain why it works, and to ask if we can actually calculate the exponents and, also, the scaling functions. The renormalization group approach provides many of the explicit sewers and, further, explains the circumstances under which scaling can break down and how it fails. 3.4 Scaling of the free energy It is useful at this stage to recapitulate by taking a somevhat different approach to scaling, and to be a little more precise. Specifically we will again use the symbol "~" to mean “behaves ike” and take the symbol "=" to mean “asymptotically equal to” f.e., tf f(x) ~ g(x) as x > 0, the ratio, £(x)/g(x) approaches unity when x* 0. As before, the discussion will be couched in magnetic language but, as previously emphasized, the same types of behavior are to be found in many other systems tf one merely identifies the analogous quantities properly. In the critical region the free energy, F(T,H), will have a singular part which embodies the leading critical behavior. Let AF be the deviation of the free energy from its value at the critical point with other non-singular contributions (the “background” terms) aleo subtracted off. We define a normalized or reduced free energy by "We restrict attention here to unfaxial, n = 1 or Ising-like magnets. For isotropic, n = 3, Heisenberg-like or XY, n = 2 magnets of different symmetry the exponents differ slightly, (see later below) and, necessarily, the scaling functions must also differ slightly for these distinct “universality cli .30 FP ‘cingular "1" (3.24) Division by kT produces a dimenstonli quantity which, however, is still proportional to the size of the system; thus, we have also divided by the volume, V, in order to obtain an intensive quantity which contains the bulk thermodynamics. ‘The dimensions of f are thus inverse volume or number density. As we will be working in the thermodynamic limit, £ is independent of V (which we suppose becomes infinite through a sequence of domains of reasonable shape). In the previous section scaling was introduced via the equation of state in (3.13). TE we integrate M(T,H) with respect to H, this leads to the free energy which (after background subtraction) will be simtlarly scaled. Alternatively, we could introduce a scaling postulate directly for the free energy. In this way, the scaling ansatz becomes the assertion # 2a (TR) » Agi]? ¥ (D 0 ie tH > 0, (3.25) Feing. where Ap and D are non-universal scaling amplitudes which depend on the details of the system. The first, Ap, sets the scale of the free energy while D sets the scale of the magnetic field. As before, there appear two universal exponents a and 4. A technical point that arises here concerning the scaling function Y(y) was already alluded to before: specifically, the universal function Y(y) should really be considered in two parts: Y,(y) for t > 0 and Y(y) for t <0. ‘These two parts must match analytically as y+ ©, but to pursue that point here would be unecessarily distracting. The reason for writing the power of the temperature prefactor in (3.25) as 2 - a 46 to get the specific heat exponent correct, as is easily seen. Let us set H = 0 and normalize the scaling function by setting ¥(0) = 1, which we may do because of the presence of the factor Ag. Recalling that entropy is given by S =~ @F/2T)yeq it follows that the singular part of the entropy varies a as(t) = Fhe ae, (Heo). (3.26) ‘The internal energy behaves similarly. The specific heat then follows as ocr) = 1(35) « “. att (3.27) In these expressions A; and Ay are anplitudes proportional to Ay. (Note that the variation of the prefactor T in the definition of C(1) is sooth and so does not affect the critical behavior of the specific heat. For t > 0 the symbol C referred to here could be eubseripted either Mor H since in zero field (i=) one has Gy = Cy this ts 0 spectal feature resulting from the symetry of a simple31 ferromagnet. ore generally, one should consider Cy(T) or, for a fluid systen, Cy(T) and 60 on. The reader should sketch the variation of the zero field entropy S(T), noting that $ is monotonic, ahd also continuous through the critical point, T = T,, but exhibits a vertical tangent there, varying in the vicinity as + |t|!~* where, as is typically true, we have supposed a has a positive, albeit small value. Moreover, the internal energy and many other quantities ‘driven’ by the critical behavior, such as the resistance, exhibit precisely the sane forn near T,. Thus it ie not these quantities themselves but rather their temperature derivatives which diverge at T, (or, if -1 2, (3.37) was a thermodynamic necessity. Note that this is a rigorous result that does not sumption as scaling theory does. Similarly, Griffiths later proved depend on any the inequality at +a(1 +8) > 2, (3.38) corresponding to (3.34). Evidently, then, scaling theory certainly does not conflict with thermodynamics even though it asserts that the rigorous inequalities hold as equalities. Nor, however, can the scaling laws be obtained by pure thermodynamic arguments although quite a few theorists have been tempted to think eo and to try to demonstrate it! Occasional reports in the past of measured values of critical exponents violating the above inequalities have all proved to be poorly founded (which is just as well, since otherwise a violation of the Second Law of ‘Thermodynamics would have been observed!) It seems that, at least as far as systems belonging to the same symmetry class are concerned, the critical exponents are universal quantities satisfying the 1 function only of the scaled field y for such systems. However, as mentioned before, one does expect some scaling laws. Similarly, the scaling function ¥ is a univer: change in Y as n, the number of components of the order parameter and d, the epatial33, dimensionality, are varied. So we can write Y= Y(y;n,d). To emphasize this point, note, as will be shown, that there are good grounds for believing that classical theory is correct when d > 4. To see how the scaling function depends on d, consider the behavior of the (zero-field) susceptibility above and below criticality. We can write xy as T+ 74, ecylty as t+ (3.39) where the amplitude ratio, C*/C™, should be universal but, clearly, depends on the particular form of the scaling function. Within Landau theory it is an easy exercise to prove Ct/c” = 2 (which is, indeed, universal). We can accept this for d > 4, but for the Ising model (n = 1) and d < 4, we find Ct/c™ = 5.03 for d = 3 while for d = 2 one knows the exact univeral value Ct/C” = 37.693562 3.5 Fluctuations, correlations and scattering What is the ‘cause’ of the failure of mean field theory and Landau theory? Why do they yield wrong exponents and wrong scaling functions? The short answer is “Because they neglect fluctuations", To understand the significance of this piece of now conventional wisdom and to explore further striking critical phenomena that provide a key to the renormalization group approach, let us study fluctuations in the critical region and introduce the correlation and scattering functions which serve to quantify them and to describe relevant observations. Much can be learned about criticality by scattering radiation --- light, x- off the system of interest. In a standard scattering raya, neutrons, ete. — experiment, a well-collimated beam of light, or other radiation, with known wavelength, 2, is directed at the sample, fluid, magnetic crystal, etc., and one ures the intensity, 1(8), of the light scattered at an angle @ away from the “forward” direction of the main beam. The radiation undergoes a shift in wave vector, k, which is simply related to 8 and \ by (3.40) The scattered intensity (6) 1s determined by the fluctuations in the medium. If the medium were perfectly uniform (.e-, spatially homogenous) there would be no scattering at all! If one has in mind light scattering from a fluid, then the relevant fluctuations correspond to regions of different refractive index and, hence, of particle density (R). For neutron scattering from a magnet, fluctuations34 im the spin or magnetization density are the relevant quantities, and 60 on. We need to study the normalized scattering intensity 1(0,7,H...)/I%#1(9), where 1(8;T,H..) de the actual scattering intensity observed at an angle 6, which will normally depend on such factors as temperature, magnetic field, etc. while 1/deal(gy is the scattering that would take place if the individual particles (spins, etc.) doing the scattering could somehow be taken far apart go that they no longer interacted and thus were quite uncorrelated with one another. Now this normalized scattering intensity is proportional to the fundamental quantity * tek Gk) -fa e” “G(R), (3.41) which represents the Fourler transform of the appropriate real space correlation function G(R) (of density-density, spin-spin, etc.) As the critical point of a fluid or fluid mixture is approached one observes enormously enhanced values of the scattering, especially at low angles, corresponding via (3.40) and (3.41), to long wavelength density fluctuations in the fluid. In the inediate critical region the scattering 18 60 large as to be visible to the eye, particularly through the phenomenon of critical opalescence. This behavior is not, however, Limited to fluids. Thus if, for example, one scatters neutrons from tron in the vicinity of the Curie point one likewise sees a dramatic growth in the low-angle neutron scattering intensity as sketched in Fig. 3.3. (with neutrons care mst be taken to ensure that the total elastic scattering 1s observed since the proportionality of 1(8) to G(k) holds only if inelastic scattering Processes can be neglected.) As can be seen, for small angle scattering there is a pronounced peak in 1(8,7) as a function of temperature, and this peak approaches closer and closer to T, as the angle is decreased. Of course, one could never actually observe zero-angle scattering directly, since this would mean picking up the oncoming main beam, but one can extrapolate to zero angle. When this is done one finds, in fact, that the zero-angle scattering 1(0,7), actually diverges at Ty This ts the most dramatic manifestation of the phenomenon of critical opalescence and is quite general, being observed whenever the appropriate scattering experiments can actually be performed. In order to understand these effects we need to examine the correlation function for the relevant quantity, which, in general, is the locally fluctuating order parameter, ¥(R), for the transition in question. Thus ¥(R) could, for instance, describe how the spin varies from lattice site to lattice site in a magnetic crystal. The overall spatial average of this quantity is what was previously referred to as the (total) order parameter, ¥. We will define the correlation function Gyy(R), or, for brevity, just G(R) by Gyy(R) = HDR? = <(Q)>H(R)>~ (3.42)Fig. 3.3 8-0 “zero- angle scattering” small @ Schematic plot of the elastic scattering intensity of neutrons scattered ‘at fixed angle, @, from a ferromagnet, such as iron, in the vicinity of the Curie or critical point. The small arrows mark the smoothly rounded maxima (at fixed @) which actually occur above T, (in contrast to ical and most mean field theories which yield a honanalytic maximum36 We will always presuppose a macroscopically large system, so that there is translational symmetry. Likewise, we suppose that inhomogeneous effects due to gravity, etc. can be ignored. This means that the two average quantities, <¥(Q)> and <¥(R)> will be equal to one another and to the overall, bulk, thermodynamic order parameter ¥. We may thus let ov) = 4) ~ >, (3.43) represent the deviation or fluctuation of ¥ about its uniform mean value; then it is a matter of simple algebra to show that the correlation function directly measures the fluctuations since G(R) = <5¥(Q)5¥(R>, (3.44) For simplicity we will often assume an isotropic system so that G is a function of R rather than R. 3.6 The correlation length If one thinks of @ lattice of spins above T, in zero field, H = 0, one has ¢¥> = 0 by symmetry, A ferromagnetic exchange coupling between neighboring spins then tends to align the spins parallel to one another whereas thermal energy works to randomize them. Thus at high temperatures one expects the spin-spin correlation function, G(R), to falloff rather_— rapidly with —_ the distance, R, separating the spins, whereas at lower temperatures the spins should become correlated with each other over longer and longer distances, the correlation function then decaying more slowly with R. What should the law of correlation decay be? On fairly general grounds one can show" that away from T, the correlation function should fall off exponentially with R for large distances, i.e., that the leading behavior te given by RIE G(R) ~ e (3.45) where £ is a quantity that has the dimensions of length, and is thus called the correlation length. It evidently tells us the scale on which the correlations decay. At high temperatures £ will be just a few angstroms, but near a critical point it becomes very large. This ties in well with our earlier comments on critical opalescence, since if £ becomes comparable with the wavelength of the ‘One must assume that the interactions themselves are of finite range or decay rapidly.37 radiation, the medium will then contain fluctuations or inhomogeneities on that scale, and this will give rise to strong low angle scattering 1.e., to critical opalescence. There is another, very general theoretical route that tells us that £ must become large near a critical point. This utilizes the fluctuation-susceptibility relation which reads COT) = fe GORST,H) = x(T,HD. (3.46) For simplicity we consider here only the magnetic case. Note that G(QsT,H) is the limiting value as k + 0 of the Fourier transform of G(R;T,H). It thus depends only on T and H and {s therefore a thernodynanic function: via statistical mechanics one finds it is just the reduced susceptibility, x(T,H) = k,T xq/m”, where m is the magnetic moment per spin. Now when T > T, for H = 0, we know that x diverges; somehow this divergence must also cone out of the integral in (3.46). Since G(R) 1s a bounded function it cannot, itself, diverge [In the case of S = ly spine one has G(R) < Zs thus a divergence of the integral can only mean that G(R) decays very slowly when T= T,, certainly more slowly than an exponential. Consequently we are forced to conclude that £(T) diverges to infinity when T > T,. The variation of © near T, can, naturally, be deseribed by Er) ~1/t, — (= 0), (3.47) where for three-dimensional systems the new exponent, v, has values around 2/3. This contrasts with the classical prediction v = 1/2 (which follows from an extension of phenomenological, Landau theory to inhomogeneous situations). More concretely one has v= 0.63 for Ising-like (n = 1) systems, particularly fluids, increasing to v = 0.70 for Heisenberg-like systems. For the two-dimensional Ising model the divergence of £(T) was established by Onsager along with his original calculation of the zero-field free energy which revealed the logarithmic divergence of the specific heat; his results yield v = 1. In experiments on fluids such as carbon dioxide, the correlation length has been measured down to t = 10 or 1075 by when £ is thousands of angstroms in magnitude. The divergence of the correlation length is one of the crucial clues to our general understanding of critical phenonen: the renormalization group approach, in particular, focuses on the behavior of the correlation length. 3.7 Decay of correlations at and below criticality At T, the correlation length is infinite. If it were not, then the integral in (3.46) would nece: rily converge and be finite: then x would be bounded at T,38 which 1s certainly not the case! Thus precisely G(R) cannot be an exponential function of R. Moreover, it must, in general, still T= 7, the correlation function, decay to zero and one should, in fact, anticipate an “algebraic” or inverse power law form such ai G(R) as R> ©, (3.48) pan (where the nonuniversal amplitude D should not be confused with our previous use of this symbol). The reason for writing the decay exponent in this rather special way is that G(R) frequently appears, as in (3.46), in volume integrals of the form Je CCR)X(R) «f Re Te ¢R)x(R) aR, 0 0 that the d drops out. Evidently n 1s a new critical exponent which describes how G(R) behaves at Ty. Its numerical value is alvays rather small and in classical theory, which, as already mentioned, will be found to apply for spatial dimenstonalities d > 4, one has n= 0. (Of course this result provides another good reason for writing (3.47) in the form given.) For real three-dimensional systems one finds n = 0.03 to 0.06, but it proves to be a very difficult parameter to measure reliably and accurately in experiments. For 4 = 2 Ising-like systems the theoretical value is n = 1/4; this can even be confirmed by experiments on (effectively) two-dimensional systems. Since n 1s in all cases small, the integral in (3-46) necessarily diverges and the susceptibility is indeed infinite at T,. Beneath T, there is a subtlety that has to be taken into account. The correlation function in general now exhibits long range order, i.e., G(R) does not decay to zero as R + © but rather approaches a nonzero value, say G(#). This appearance of long range order is, in fact, one of the notable characteristics of most phase transitions. In a magnet the zero field spin-spin correlation function when R + © becomes proportional to [§Mj(T)]?, the square of the spontaneous magnetization. Via the ecattering theory this leads to a magnetic Bragg peak in the scattering of strength proportional to [Mo(T)]?. In systems such as antiferromagnets, where ¥(R)is a “staggered magnetization”, this provides a means of measuring the spontaneous order which would, otherwise, be inaccessible to experimental observation. If one subtracts the limiting value, , from the correlation function one obtains a net correlation function which again decays to zero. In Ising-like systems there is then also an exponent v' for the correlation length beneath T.. Experimentally, one finds v' = v and theoretically, according to scaling, the two exponents should be exactly the same. Experimentally the scattering intensity 1(8) = I(k) provides us with the information needed to determine £. It is not hard to show that G(k), which we recall is essentially Proportional to I(k), is an even function of k which, rather generally, can be expressed in the formshebtie te soa, >t, G(k) (3.49) In @ so-called Ornstein-Zernike analysis one thus plots 1/6(k) [or 1/T(k)] in the critical region versus 2, The data for small k (such that ka £ 0.1 where a is a typical molecular dimension) usually fall close to a straight line whose intercept with the k? = 0 axis determines the susceptibility x(T). As T+ T, this intercept falls to zero but the successive isotherms remain more or let parallel on the Ornstein-Zernike (or 02) plot. The reduced slopes evidently serve to determine £(T). Close to T, the plots in the case of very pood experiments show a slight downward curvature: this is an indication of a nonzero and positive value of the exponent n. Thus at T= T, we have, by (3-48), the power law decay 1/R?-?#", and on Fourier transformation this yields . (3.50) asymptotically for small k. On an Ornstein-Zernike plot the curvature of the critical isotherm thus measures n. It should be stressed, however, that since n has such a small value, it is difficult to measure this curvature unambiguous]; extensive data are needed and careful corrections for multiple scattering and other extraneous effects are called for. Nevertheless, a small positive value is definitely established. 3.8 Scaling of the correlation functions Our treatment of correlation functions hi evidently introduced two new exponents, v and n. Are these independent of each other? Are they independent of the thermodynamic exponents a, 8, y, and 5? Or are all the exponents somehow linked together? Let us see what the idea of scaling has to say in this context. Accordingly, with no loss of generality we write the correlation function and its Fourier tranform as Dost») 2=n Der (Rit) = yoo and G(ksT,B) = @.51) ® which serves to pull out the critical point behavior. Now we expect (or hope!) that G and G will exhibit some simplified behavior as T+ T,. Scaling means that there should be some reduced description, some compression or collapse of the multivariable data, Thus the dependence of the correlation functions on three variables might, perhaps, reduce to a dependence on only two, properly scaled, variables. The behavior at T, has been extracted in terms of the functions D and®, In line with our previous application of scaling, it is thus natural to