Scribd
Upload
2K views206 pages

H E Stanley Introduction To Phase Transitions and Critical Phenomena

Introduction to Phase Transitions and Critical Phenomena

Uploaded by

Carlos Torres Echeverría
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
2K views206 pages

H E Stanley Introduction To Phase Transitions and Critical Phenomena

Introduction to Phase Transitions and Critical Phenomena

Uploaded by

Carlos Torres Echeverría
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
You are on page 1/ 206
PART I INTRODUCTION 1 WHAT ARE THE CRITICAL PHENOMENA? A SURVEY OF SOME BASIC RESULTS Our purpose in this first chapter is to provide the reader with a brief qualitative description of what actually happens near the critical point, and with an appreciation of why the study of these critical phenomena has burgeoned in recent years. 1.1. Classical era of critical phenomena Many of the basic facts of critical phenomena were observed fifty or even a hundred years ago, while other aspects have been discovered. only in very recent years. Hence it has become customary to divide the field of critical phenomena into an early ‘classical’ era and a more recent ‘modern’ era. Although a wide variety of physical systems exhibit critical pheno- mena, we shall concentrate on the liquid-gas and magnetic critical points for the sake of simplicity. 1.1.1, Fluid systems Our discussion of fluid systems will focus first on the equation of state, a functional relationship of the form f(P, p, 7) = 0, which relates the thermodynamic parameters—pressure, density, and temperature. The equation of state thus defines a surface in a three-dimensional space whose coordinates are P, p, 7’; each of the points on this surface corresponds to an equilibrium state of the system. In order to aid in visualization of this PpT' surface, it is convenient to consider its pro- jections on to the P7’, Pp, and p7' planes; these are shown schematically in Figs. 1.1(a), 1.2(a), and 1.3(a) respectively. We see from Fig. 1.1(a) 2 CRITICAL PHENOMENA: BASIC RESULTS 1,§1 that the projection on to the P7' plane produces three separate regions, corresponding to the three familiar phases of matter—the solid, liquid, and gaseous phases. The solid and gaseous phases are in equilibrium along the sublimation curve, the solid and liquid phases are in equilib- rium along the fusion curve, and the liquid and gaseous phases are in equilibrium along the vapour pressure curve. Each point on these three curves represents an equilibrium state in which two or more phases can coexist—and the triple point represents an equilibrium state in which all three phases coexist. We notice, however, that the vapour pressure curve does not extend forever, as the fusion curve appears to do, but rather that it terminates in a point. This point is called the critical point, and its coordinates are bs | Fusion _ | R curve ony . Solid ttt - Z Vapour H=0) prosure curve rae | Sublimation: Gas Hy st |Surve ' ‘Triple point | (a) Fluid (b) Magnet Fie. L.1. (a) Projection of the PVT’ surface in the PT’ plano. (b) Projection of the HMT’ surface in the HT’ plane. denoted by (P,, p., 7'.), where P,, p,, and 7’, are the critical pressure, critical density, and critical temperature respectively. The fact that the vapour pressure curve terminates in a critical point means that one can convert a liquid to a gas continuously, without crossing the phase transi- tion line, as is indicated by the dotted path shown in Fig. 1.1(a). In this sense there is no fundamental difference between the liquid and gaseous phases. It is widely believed that the fusion curve does not also terminate in a (second) critical point. However, thus far it has not been possible to prove the non-existence of a liquid-solid critical point. That the vapour pressure curve terminates in a critical point was not appreciated until about a hundred years ago. Prior to that time scientists regarded certain gases as being ‘permanent’ in the sense that these gases could not be made to condense no matter how much pres- sure was applied (compressing a gas was a standard laboratory pro- cedure for liquefaction). Presumably the work on these gases was carried. 1§1 CLASSICAL ERA OF CRITICAL PHENOMENA 8 out at a temperature 7’ > 7',, while a prerequisite for obtaining droplets of the condensed phase is that the material be brought to a temperature lower than the critical temperature 7',. Thus, for example, helium cannot be liquefied no matter how great the applied pressure unless the temperature is reduced below the critical temperature of 5-2K. In addition to the PT projection, it is also useful to consider the pro- jections in the Pp and pT’ planes; these are shown schematically in Figs. 1.2(a) and 1.3(a) respectively. These projections tell us a great deal (a) Fluid (b) Magnet Fra. 1.2. (a) Isothermal cross-sections of the PVT surface (or, literally, the PpT’ surface). The exponent § is a measure of the degree of the critical isotherm. The fact that 3 > 1 reflects the fact that tho compressibility Ky is infinite at ',. The fashion in which Ky diverges at 1’ -> 7, is described by the exponents y and »/. (b) Isothormal cross- sections of the HM’ surfece. At high temperatures all isotherms approach straight lines, corresponding to the non-interacting limits, P = pkT'/m and H = TM/C, where C is Curic's constant (of. Chapter 6). concerning the qualitative features of the critical point. From Fig. 1.2(a) and also from Fig. 1.3(a) we see that at low temperatures there is a rather large difference between the liquid and gas densities, p,, and pg, but that as the critical temperature is approached this density difference tends to zero. The existence of a quantity which is non-zero below the critical temperature and zero above it will be seen to be a common feature associated with the critical points of a wide variety of physical systems. We say that p,, — pg is the order parameter for the liquid-gas critical point. A second striking feature of Fig. 1.2(a) concerns the shape of the iso- therms as the critical point is approached. At very high temperatures the ideal gas law is obeyed and the isotherms are therefore the straight lines given by the ideal gas equation of state P = pkT/m, (1.1) 4 CRITICAL PHENOMENA: BASIC RESULTS 1,§1 where & is the Boltzmann constant and m is the mass of a molecule. Hence one might suspect that the curvature of the isotherms which becomes apparent as 7’ decreases towards the critical temperature is a manifestation of interactions among the constituent molecules of the fluid. This is in fact the case, and we can discuss what is going on as the critical point is approached rather more easily if we utilize an analogy between a fluid and a ferromagnetic system, the lattice-gas model (see Appendix A for a more detailed discussion). According to this model, we imagine that the macroscopic volume V containing the fluid is par- titioned into fixed microscopic cells whose volume » is roughly the size (a) Fluid (b) Magnet ot u| Liquid Coexistence : j curve No stable Two-phase - states in - 2 Tegion T this region T Te / ~_---L_ Gas Fig, 1.3, (a) Projection of the Pp’ surfaco in the pT’ plane. Noto that as 7’ > 1’, from below, tho vapour density pg increases and the liquid density p,, decreases. Sometimes one writes p, — pa ~ |«|*o and p, — po ~ |€|*. However, in practice fo ~ f,, and we may simply write p, — po ~ |«|*. (b) Projection of the HM surface in the MT plane, Note that the spontaneous magnetization M plays the role of p — py. Unlike tho fluid case, this projection is the same as the H = 0 cross-section of the HM surface. of the constituent molecules of the fluid. Next we construct an analogous magnetic system by considering each cell to be a lattice site on which a magnetic moment is situated. This magnetic moment is chosen to point in the upward direction if the corresponding cell of the fluid is occupied by the centre of a molecule, and the magnet is assumed to be pointing downward otherwise. Thus for temperatures well above the critical temperature, the free motion of the gas molecules of our fluid system will correspond to a rapid and random ‘flipping’ of the magnetic moments from one orientation (up or down) to the other. However, as the temperature is lowered toward the critical point, small ‘droplets’ of correlated spins appear. As the critical point is approached still 1§1 CLASSICAL ERA OF CRITICAL PHENOMENA 5 closer, the droplets grow in their dimensions as shown in Fig. 1.4. In fact, it is experimentally feasible to bring a fluid close enough to its critical point for these droplets to acquire lateral dimensions on the order of the wavelength of light, whereupon the light is scattered strongly. This phenomenon, called critical opalescence, was discovered by Andrews just over a century ago in the course of his measurements T>>T, Fic. 1.4. Sketch of the lattico-gas model of a fluid system as the temperature approaches tho critical temperature. Each cell is coloured black if it is occupied by the centre of a molecule and we associate an ‘up’ spin with this site. The correlation longth £ may be thought of as being roughly the diameter of an ‘island’ of aligned spins. By courtesy of D. L, Njus. of the critical behaviour of carbon dioxide. To this day it provides one of the more striking manifestations of the critical point. Fig. 1.4 may be misleading in that we may get the impression that the system becomes completely ordered as 7’ approaches 7, from 6 CRITICAL PHENOMENA: BASIC RESULTS 1,§1 above (in the sense that all the moments are oriented parallel to one another). Of course this is not true; the system is completely ordered only at a temperature of absolute zero, and the net order (measured by the order parameter, which is the spontaneous magnetization in this case—ef. Fig. 1.3(b)) is zero at the critical temperature 7 = 7',. Hence Fig. 1.4 is to be interpreted as a small portion of a large system. To clarify this remark further, Fig. 1.5 shows a system consisting of a two-dimensional lattice of cells with 64 cells on each side (so that there are altogether 212 = 4096 cells). This sequence of diagrams may be interpreted as representing the approach to the critical point from the low-temperature side. Thus (a) represents the completely ordered state at 7 = 0 while in (b) the temperature of this two-dimensional lattice gas might be roughly }7',. We notice that relatively few of the spins have flipped from their up-configuration to their down-configuration, and this fact corresponds to the extreme flatness of the Ising model magnetization curve (cf. Fig. 9.9). In (c) the temperature is, say, $7’, and a larger number of spins have flipped orientations. In (d) the tem- perature is getting close to 7',, and we observe that the overturned spins appear to cluster together in small ‘islands’. In (e) T ~ 7, while in (f) 7 2 7, and the islands of aligned moments are quite large. Notice that even though above the critical temperature the net magnet- ization is zero (i.e. there are roughly as many up spins as there are down spins), nevertheless there exists a considerable degree of ‘order’ in the system, as reflected in the large dimension of the islands of aligned ‘Spins. This order is frequently called short-range order in order to dis- tinguish it from spontaneous magnetization or long-range order. The phenomenon of critical opalescence is captured in the sequence of six photographs shown in Fig. 1.6. In part (a) the temperature is well above 7, and we see the single, clear phase. The fluctuations are weak and there is no appreciable light scattering. In parts (b), (c), and (d) the temperature is successively lowered and we see that the sample tube takes on a dramatic glow. In (e) the temperature is just below 7’, and although a meniscus between the liquid and gaseous phases is apparent there are still liquid globules of higher density raining out of the gas phase above the meniscus, and vice versa. Finally, (f) shows complete separation of the liquid and gas phases. The concept of density changes near the critical point is illustrated in a sequence of four photographs in Fig. 1.7. These show a sealed glass bulb which has been filled with a quantity of carbon dioxide chosen such that the average density inside the bulb is very close to the critical Fic. 1.5. Schematic indication of the lattice-gas model of a fluid system, Fig. 1.4 (which contains 8 x 20 cells) should be interpreted as representing a relatively small portion of this lattice (which contains 64 x 64 cells). (a) is the completely ordered state (which exists only at T= 0); (b), T= 37.3 (¢), TX 3T es (@), T= BT (), TX Tos (1), L & T.. This illustration and the associated temperatures are to be regarded as purely schematic. In fact, the figure was constructed from a computer simulation of the time- dependent aspects of the two-dimensional Ising model and actually represents rather different phenomena (ef. Appendix E). After Ogita et al. (1969). d e F lowered past the critical temperature: (e), P< To3and (f), 7 « T.. The fter Ferrell (1968). Fic. 1.6. Behaviour of a fluid as the temperature is (a), T > 7,5 (b), TZ Tes (0, T~ Tos (0, TS fluid shown is the binary mixture L§1 CLASSICAL ERA OF CRITICAL PHENOMENA 7 density p,. Also inside the bulb are three balls with pre-determined densities that are slightly less than, approximately equal to, and slightly larger than p,. In (a), the temperature is greater than 7’, and the fluid consists of a single, relatively homogeneous ‘phase’ with slight density variations due to gravity. Hence the heaviest ball is at the bottom and the lightest ball is at the top of the tube. In (b) the temperature is lowered to just above 7’, and we observe the critical opalescence as the entire tube takes on a milky white colour. Since the compressibility is large, the density distribution is very sensitive to pressure gradients, and the middle ball—in this photograph—is not precisely at the centre of the tube. In (c) the temperature has fallen just below 7, and a meniscus is seen to form, while in the final stage (d) the temperature is well below 7’, and all three balls float at the, by now, well-defined meniscus. The fact that this meniscus forms near the centre of the tube supports the classic law of rectilinear diameter according to which Hex + pa) & Por After the preceding qualitative discussion, it should come as no surprise that the isotherms in Fig. 1.2(a) indicate a significant depar- ture from the linear behaviour predicted by the ideal gas law; we are now in a position to understand something of the nature of this devia- tion. Specifically, we see from Fig. 1.2(a) that the isotherms actually acquire a flat portion in the immediate vicinity of the critical point, ie. the slope (@P/@p) becomes zero as 7'—> T}. Now the isothermal compressibility Ky of a fluid is defined to be p~*(@p/@P)y, and hence the flat portion of the isotherm in Fig. 1.2(a) corresponds to an isother- mal compressibility which diverges to infinity as the critical point is approached. Of course an infinite value of (@p/@P)y means that the response of the density to a very small pressure fluctuation is infinite. Thus we might expect that this divergence in the isothermal com- pressibility is connected to the huge density fluctuations which we associ- ated with critical opalescence. In fact, we shall see in Chapter 7 that the density fluctuations are mathematically related in a direct fashion to the isothermal compressibility and hence to the derivative (@p/@P)z. Thus far our discussion of the classical era of fluid critical phenomena has centred on experimental results. This is not to say that there were no theories, however. In fact, barely three years had passed after Andrews’ classic 1869 publication on the critical point of carbon dioxide when van der Waals, in his Ph.D. dissertation, published a theoretical description of the critical region which to this day provides an intriguingly accurate description of the critical region for 8 CRITICAL PHENOMENA: BASIC RESULTS 1,§1 temperatures that are not too close to 7’, (cf. Chapter 5). The pheno- menon of critical opalescence was also discussed in the early years of the twentieth century by many theoreticians—among them von Smoluchowski, Einstein, Ornstein, and Zernike. 1.1.2, Magnetic systems It is customary to stress the analogies between magnetic and fluid transitions for pedagogical reasons. For example, if we apply pressure to a fluid system the density increases, while if we apply a magnetic field H to a ferromagnetic system the magnetization M increases. Hence in a sense H is analogous to P and M to p and the equilibrium PpT surface of a fluid system corresponds, after a fashion, to the HMT surface of a magnetic system. Figs. 1.1(b), 1.2(b), and 1.3(b) show the projections of the HMT sur- face on to the HT, HM, and MT planes respectively. Much of the qualitative discussion presented above also applies to the magnetic system. In particular, the response function Ky is analogous to the isothermal susceptibility yp = (@M/@H)z, which approaches infinity near 7’, corresponding to the ‘flattening’ of the critical isotherm (7 = T.) in Fig. 1.2(b). Associated with this divergent response func- tion is an enormous increase in the fluctuations of the magnetization, and a relation, analogous to that for the fluid system, relates xp to these magnetization fluctuations (cf. Appendix A). In the early years of the twentieth century, theoretical progress was made in understanding magnetic transitions, and the approach taken was not altogether unlike the classic work of van der Waals on fluid systems. In 1907, just a few years after the pioneering experimental work of Curie, Hopkinson, and others, Pierre Weiss proposed a theory of ferromagnetism in which he assumed that the constituent magnetic moments interact with one another through an artificial ‘molecular field’ which is proportional to the average magnetization (see Chapter 6). More specific models of the interacting magnetic moments were put forward some years later. These models have in common the feature that they assume the magnetic moments to be localized on fixed lattice sites and that they influence one another through pairwise interactions with an energy that achieves its maximum value, J, when the moments are parallel. Two particular forms of the interaction are particularly interesting to this day. In the first of these, due to Wilhelm Lenz but called the Ising model, the magnetic moments are assumed to be classical, one- (8961) s1oBu0g [Oar] pur sieBuag wos L2G) 8G «Kb “(pd “uvyy toqvasB X] 34s pur ‘oy penbo ynoqe suey? soy AYsYS sos syreq coat YL, ya 04 esojo 40a Aysuap oBer0A ue YIN poly st oqny sse[8 oy, “qutod-jwonto ot} avou oprxorp uoqavo jo anorauyog “ET “Dg 1,82 MODERN ERA OF CRITICAL PHENOMENA 9 dimensional ‘sticks’ capable of only two orientations. Thus the Ising model is the magnetic model analogous to the lattice-gas model of a fluid system discussed in the previous subsection. A second model, called the Heisenberg model, regards the magnetic moments as being related to quantum-mechanical three-component spin operators, and assumes that the energy is proportional to the scalar product of these operators. Although the original Weiss model of @ magnetic system is quite easy to solve exactly, neither the Ising nor the Heisenberg model has yielded as yet to an exact solution for a three-dimensional lattice. Nevertheless, these models appear to repre- sent reasonable theoretical descriptions of certain physical systems, and therefore their continued study has provided us with considerable insight regarding magnetic phase transitions. 1.2. Modern era of critical phenomena 1.2.1. Critical point exponents There is no clear-cut development which delineates the beginning of the modern era of critical phenomena. Some might say that it began in the 1940s with Guggenheim’s realization that the coexistence curve of a fluid system is not parabolic and with Onsager’s exact solution of the two-dimensional Ising model, while others might set the date somewhat later, for example in the early 1960s when Heller, Benedek and Jacrot (together with theoreticians such as Domb, Rushbrooke, Fisher, and Marshall) came to recognize that certain entities, the critical point exponents, were significant entities worthy of special attention in their own right. In any case, our brief introduction to the modern era will begin by considering the subject of critical-point exponents. Consider, for example, the now classic Guggenheim plot, reproduced in Fig. 1.8, which shows the temperature dependence of the liquid-gas density difference p;, — pq for eight different simple fluids (Ne, A, Kr, Xe, Ng, Og, CO, and CH,). We call p, — pq the order parameter because it is non-zero only in the ordered phase. The fact that the data, properly normalized, fall on one and the same curve is in accord with the law of corresponding states. What is interesting is that the shape of the solid curve—supposedly a ‘best fit’ to the data—is a cubic function rather than the quadratic function that the van der Waals theory would predict. For the analogous magnetic system the appropriate order parameter 10 CRITICAL PHENOMENA: BASIC RESULTS 1,§2 1-00 i, oP Es, 0-95, + 0-90) | 0-85 nn, | 0-80; | +Ne 4N, 0-75 oA 90, "Kr CO 70 xXe 0 CHy 0-65; 0-60) 0-551 aL 1 1 4 1 4 m ae 00 02 04 06 O08 I0 12 14 16 18 20 22 24 26 p/P. Fra, 1.8, Measurements on eight fluids of the coexistence curve (a reflection of the Ppl’ surface in the pT’ plane analogous to Fig. 1.3). The solid curve corresponds to a fit to a cubic equation, i.e. to the choice B = }, where p — po ~ (—«)*. From Guggenheim (1945). is the zero-field magnetization M because M is a measure of the degree to which the magnetic moments are aligned throughout the crystal. Here again the classic Weiss theory predicts a quadratic dependence M? x (T, — 1), whereas we see that M? oc (T, — T) would seem to be an appropriate fit to the measurements of Heller and Benedek shown in Fig. 1.9. It is customary to say that the order parameter varies as (—«)* where T-T, ae e (1.2) e= and where the critical-point exponent f typically has a value in the range 0-3-0-5. It is important to stress that it is not necessary to have a strict proportionality between the order parameter and (—e)? in order to be able to define a critical-point exponent. In fact, if we knew that a simple relation of the form M = @(—e)* were valid, then three meas- 1,§2 MODERN ERA OF CRITICAL PHENOMENA 1 urements in the critical region would suffice to determine the exponent ! In practice there are frequently correction terms, so that M might have the form #,(—«)*{1 + B(—«)* +...} with x > 0. Hence a more natural definition of the critical-point exponent f is Init Ba Mins (3) where the correction terms will drop out in taking the limit. In fact, critical-point exponents are frequently determined by measuring the slopes of log-log plots of experimental data, since !’Hospital’s rule, together with eqn (1.3), implies that 8 = d(n M)/d{In (—«)}. Although this is a particularly quiek—method-of-determining the exponent, it. requires a prior knowledge of the.critical temperature,so_that in prac- tice one. must frequently resort to plotting M1? for several trial values of A until a value is found which produces a straight line. At one time many workers believed that all materials have the same exponents. For example, we remarked above that all eight fluids shown in the Guggenheim plot, Fig. 1.8, appear to have roughly the same exponent, 8 ~ 4. Hence it was rather satisfying when the first accurate measurements of f for a magnetic system, those of Heller and Benedek in 1962, produced the value B = 0-385 + 0-005 (cf. Fig. 1.9), and 200 000 a b b b 150 000+ Cube of N.M.R. frequency (MHz) Tx =67-336 K Poa or rn ro 500 66-000 66500 67-000 67-500 Temperature (K) Fra. 1.9. Dependence upon temperature of the cube of the zero-field magnetization for MnF,. Since MnF, is an antiferromagnet instead of a ferromagnet, the critical tom- perature is denoted by Ty rather than by T,. After Heller and Benedek (1962). 12 CRITICAL PHENOMENA: BASIC RESULTS 1,92 subsequent measurements on other magnetic systems also appeared to yield similar values of 8. However, this once-hoped-for universality has yet to be more rigorously demonstrated, and there now exists a growing list of materials for which f = } is definitely outside the experimental error. For example, particularly accurate measurements supporting 8 = 0-364 for helium are shown in Fig. 1.10. 10° | r Jexeel _. {ee | 7087 < I le | a | B | § 104 or 4 | 3 we © Vapour 4 zg 4 Liquid | | | Te = 51888 K pe= 0-0690 g em-3 10-46 - ! J lot los lot Reduced temperature Fic. 1.10, Measurements of the coexistence curve of helium in the neighbourhood of its critical-point. The critical-point exponent 8 has a value of 0-354. After Roach (1968). For some of these materials, however, the source of the discrepancy may be due to complicating factors such as the lattice compressibility. In Fig. 1.11, for example, are shown some recent data on the magnetiza- tion of CoO, which contracts suddenly on cooling through the critical temperature so that the exchange energy between neighbouring atomic moments increases. Hence when the critical temperature is approached from below, the system finds the exchange energy and hence the effective critical temperature decreasing (kT, is generally thought to be a linear function of J, as one might imagine from dimensional analy- sis), and the measured critical-point exponent f is decreased below what one would expect for an incompressible lattice. Thus the value B = 0-244 + 0-015 is obtained from the slope of the log-log plot of the CoO data in Fig. 1.11 whereas, when corrected for this lattice contrac- tion effect, the data indicate B = 0-290 + 0-025. 1,82 MODERN ERA OF CRITICAL PHENOMENA 13 In Chapter 3 we shall define a good many of the commonly used critical-point exponents—suffice it to say here that there are essentially ag many exponents as there are singular functions, and the Greek alphabet is fast being exhausted. Three of the most common critical- point exponents—a’, 8, and y’—are defined for fluid and magnetic systems in Table 1.1. Note that minus signs are associated with the el | 2 20-1081 ep | & i 100 - 1 eg 50 F weighted LSF Jp=C (72-7) 7 Ss | | i 20 ¢ 88.9825 K | m 0-488+ 0-030 | 2 10 + 4 i | LSF corrected for variation Ty=Ty (7) j fos + 2B come 0°575 0-050 ; / { O01 9% 05 10 20 50 100°C (Ty-7) ‘Temperature difference from 7'y Fic. 1.11. Dependence of the logarithm of the magnetic Bragg peak intensity from neutron scattering from CoO upon the logarithm of (Tx — 1). This intensity is propor- tional to the square of the spontaneous magnetization. The upper curve ia a least squares fit (LSF) to a power law, assuming that the lattice is incompressible. The lower curve is a similar least squares fit to data that have been corrected for the lattice compressibility. The critical exponent obtained from the lower curve is clearly in better agreement than the upper curve with the anticipated value of B ~ 0-3. After Rechtin, Moss, and Averbach (1970). exponents for response functions such as the specific heat, compressi- bility, and susceptibiliff’ which are expected, theoretically speaking, to diverge to infinity at the critical point; hence the exponents «’ and y’ are defined such that they are positive quantities. Of course, no one has ever measured an infinite value for any of these response functions. This is not only because we never can make measurements arbitrarily close to 7’, (measurements for « < 10~ or closer than one part in a S—1r.7. i“ CRITICAL PHENOMENA: BASIC RESULTS 1,§2 Tasye 1.1 Representative critical-point exponents for fluid and magnetic systems. For simplicity we have only considered here the approach to T’, from the low-temperature side. More complete tables are shown in Chapter 3 Definition a B y w+ 2B+y’ Fluid Crave~(~2)"" py pa~(—e? Kew (—0)-Y = Magnet Cuao~(—6)-* Mueo~(—¢)? xr~(—«)-” aad Typical experimental values Fluid or magnet 0-0-2 03-05 <2 Theories van der Waals or Weise 0 (discontinuity) } 1 2 ‘Two dimensional Ising model 0 (log) + t 2 million from the critical temperature are extremely rare) but also be- cause some sort of rounding-off of the data is frequently found, as we see, for example, in the specific-heat data shown in Fig. 1.12. 45 T T T aa { CoC, 6H,0 7 = sandy CoC, BH, Gash. rT oe 5 E 3.0 be Tey Ne 4 Fest ON, 4 20 L th ly 4 g if i ~ S15 4 (+ TTy Mi 4 05 1 eee al 00 L \ ! —L 1 J 10+ 10? 10 107 10° 10! IP=TyldK) Fic. 1.12, Dependence of the specific heat upon the logarithm of | 7’ — 7'y| for cobalt chloride. The data appear to be fitted fairly well by @ logarithmic divergence except within a few millidegrees of T'y. After Kadanoff et. al. (1967). 1.2.2. Results from model systems The number of model systems which have been studied as a means of gaining insight into the nature of phase transitions and critical pheno- mena is extremely large and therefore we shall limit our remarks here 1,§2 MODERN ERA OF CRITICAL PHENOMENA 15 to the two models we introduced above, the Ising model and the Heisen- berg model. Although both these models were proposed in the early years of this century, it is only within the last two decades that much of their richness has come to be appreciated. The highlight in any discussion of the Ising model is perforce 40 3-0 Culk 20 00 05 10 15 20 T/T. Fie, 1.13, The aclid curve shows the specific heat of the two-dimensional Ising model as obtained from the exact solution of Onsager (solid curve), from the Bethe approximation (dotted curve), and the Kramers-Wannier and Kikuchi approximation (broken curve). After Domb (1960). Onsager’s solution, in 1944, for the H = 0 partition function of a two- dimensional lattice. From the partition function he was able to demon- strate that the specific heat possesses a logarithmic divergence at 7’, when approached from either side of the transition. This result stood in dramatic contradistinction to the predictions of the mean field theory and other theories of cooperative phenomena of that day which pre- dicted a simple discontinuity in the specific heat (of. Fig. 1.13). In 16 CRITICAL PHENOMENA: BASIC RESULTS 1,§2 particular, it showed that Ehrenfest’s classification according to which derivative of the free energy undergoes a discontinuity was, in a sense, inadequate for the description of the two-dimensional Ising model phase transition. (In this monograph we shall follow Ehrenfest in so far as we shall denote by ‘first-order transitions’ phase transitions in which the first derivative of an appropriate thermodynamic potential is discontinuous, while we shall denote by ‘higher-order’ or ‘continuous’ transitions the characteristic behaviour near the critical points of fluids and magnets for which the first derivatives are continuous while the second derivatives are either discontinuous or infinite. This classifi- cation, due to Fisher, is somewhat more general than Ehrenfest’s.) Because Onsager’s solution is for the zero-field partition function, it is not possible to obtain directly the exponents f, y, and y’. The spontaneous magnetization and hence the exponent f was, however, obtained a few years later by Onsager, who announced his result in 1948 as a discussion remark following a paper of Tisza on the general theory of phase transitions. Onsager never published his derivation, however, and four years later Yang published a complete derivation that abounds in complicated elliptic integrals and represents a true mathematical tour de force. “Phe susceptibility of the two-dimensional Ising model has not yet been caloulated, so Table 1.1 shows only an approximate (though an extremely reliable) value for the exponent y’. This approximate value is not obtained from the usual sort of closed-form approxi- mations, but rather from extrapolations based upon a method of successive approximations in-which successively longer-range corre- ken inte-aeeountAs we shall see in Chapter 9, closed mations have so far not been capable of describing the subtle and fascinating physical feature of the critical point—namely that although the interaction between the constituent magnetio mo- ments is generally of extremely short range (for example, extending only to neighbouringmoments), this interaction neverth gates’ from one moment she next, tending to create a preferred direction for all magnetic moments ts and in fact the resulting order becomes infinite in range as t po sritical point is approached. In particu- lar, these approximation methods in which longer and longer correlation ranges are successively taken into account have demonstrated the crucial role of lattice dimensionality in determining critical behaviour. Hence the fact that the critical-point exponents shown in Table 1.1 for the two-dimensional Ising model do not agree with most experi- 4g2 - MODERN ERA OF CRITICAL PHENOMENA 7 mental measurements on simple fluids does not necessarily indicate a breakdown of the intuitively plausible lattice-gas model. Since no one has yet succeeded in solving the three-dimensional Ising model exactly (and since some workers have claimed that it may in fact be an insoluble problem!), we must perforce consider alternate models. Now the Heisenberg model is not even soluble for two-dimen- sional lattices, so this avenue of approach would appear fruitless. How- ever, some years ago Berlin and Kac succeeded in solving, for both two- and three-dimensional lattices, a rather unphysical model called the spherical model. This model is in some respects similar to the Ising model but, in the critical region, the exponents of the spherical model agree with experiments as poorly as do those of the two-dimensional Ising model. The principal virtue to this day of the spherical model result is that it represents one of the very few non-trivial examples of a many-particle system that can be solved exactly for three dimensions. This statement is not intended to deprecate model systems which as yet have no physical system as a counterpart—there is generally much to be learned from a theoretical model when considered in its own right, and frequently the predictions of the theoretical model motivate a successful search for the appropriate physical system. 1.2.3. Exponent inequalities and the scaling hypothesis At this point, the reader may be wondering why so much effort has been directed towards the investigation of the critical-point exponents and he may be led to inquire why more attention is not focused on finding a theory that fits experimental data over the entire range of temperatures. Our answer to his question has two parts. The first is that the physical phenomena that come into play—namely, the building up of extremely long-range correlations extending to perhaps a billion constituent particles—are absent when one is far from the critical point. The second part of our answer is that there exist among the exponents certain relations which arise from fundamental thermo- dynamic and statistical mechanical considerations and thus transcend any particular system under consideration. As an example of this, we show in Chapter 4 that a large number of inequalities among the various critical-point exponents can be obtained from elementary thermodynamic reasoning, together with certain additional assump- tions. One of these inequalities states that a’ + 28 + y’ > 2, and Table 1.1 shows the values of the quantity «’ + 28 + y' for several theoretical models and for typical experimental systems. We see that 18 CRITICAL PHENOMENA: BASIC RESULTS 1,§3 «' + 28 + y’ apparently equals 2 for the theoretical models and either equals or falls just short of 2 for the experimental systems. However, the equality can frequently be obtained by reconsideration of the error bounds placed on the data. Hence one might be led to conjecture that the inequality «’ + 28+ y' = 2 is in fact obeyed as an equality. Attempts to prove this rigorously have thus far been unsuccessful. An alternate approach which does predict that the inequalities be satisfied as equalities involves making an assumption that certain thermodynamic functions are_homegeneous functions of theirargu- ments. This assumption, which has come to be called the statie-scaling_ hypothesis, leads not only to the prediction that most of the exponent inequalities are satisfied as equalities, but also to additional predictions concerning the equation of state of a substance in the very immediate vicinity of the critical point. Many of these predictions have been partially corroborated by experimental measurement and theoretical calculation on model systems, and the scaling law approach has thereby enjoyed quite a remarkable success in providing a unified approach to many phenomena associated with the critical point. It should be pointed out, however, that certain examples still stand in contradiction to the predictions of the scaling hypothesis, and hence the ultimate validity of the hypothesis must be regarded as an open quéstion. More important, perhaps, is the fact that the scaling hypothesis is at best unproved, and indeed to some workers represents an ad hoc assumption entirely devoid of physical rationale. 1.3. Phase transitions in other systems Phenomena analogous to those described above are found to be associated with critical points in a wide variety of other physical sys- tems. For example, many features of the superfluid, superconducting, and ferroelectric transitions are quite similar to those found for liquid— gas and magnetic critical points. In particular, the analogy between the phase transitions discussed above and the phase transition in a binary mixture (or a binary alloy) is relatively easy to describe. A binary mixture of two different fluids exhibits a critical temperature below which the two components do not mix homogeneously in all proportions, and one can define a two-component region, bounded by a coexistence curve, which is in many respects analogous to the two-phase region of a simple one-component fluid. Analogies among phase transitions in superfluids, superconductors, 1§3 PHASE TRANSITIONS IN OTHER SYSTEMS 19 oO Ty, =2:17K Fro. 1.14, Sketch (not to scale) of the equilibrium PVT surface of ‘He in the low- temperature region, and the projection of this surface on to the P7’ plane. Contrast this phase diagram with that of a normal fluid shown in Fig. 1.1(a). From Huang (1963). 10 j 4 2 we d SE 3 BoE 2} 2 OE 3 [ { | : ! 1 J 1 Wo 10-+ 105 10> 1077 10° 1 -T (K) Fra. 1.15, Dependence of the logarithm of the superfluid density upon the logarithm of the temperature difference from the A temperature for ‘He. The superfluid density is the square of the order parameter for the A-transition, and one sees from the slope of this plot that the appropriate exponent has a value of about § or roughly that of M3, where M is the order parameter for the ferromagnotio transition. After Kadanoff et. al. (1967). 20 CRITICAL PHENOMENA: BASIC RESULTS 1,83 and ferroelectrics must, however, be made with care. For example, there is actually an entire line of critical points in a superfluid along which various critical phenomena take place, and as we see from the PT-projection of the PVT surface of helium shown in Fig. 1.14, this _(eateeel Lou. 4 4 too oe —20-10 0 10 20 30 T-Tudeg Fro, 1.16. Specific heat of ‘Ho as a function of 7’ — 1’, in K. Notice that the shape of the specific heat curve is rather like the Greek letter A, whence the origin of the term ‘- transition’. The fact that the specific heat is only about ten times its ‘normal’ value even at temperatures only a fow microdegrees from 7’, is correlated with the fact that the critical-point exponent is extremely small (in fact, « is probably zero, corresponding to a logarithmic divergence). ‘The width of the small vertical line just above the origin indi- cates the portion of the diagram that is expanded in width in the curve directly to the right. After Buckingham and Fairbank (1965). ‘A-line’ terminates at the vapour pressure curve in a ‘)-point’. These phase transitions appear to be characterized by an order parameter which is non-zero below 7’, and zero above (cf. the data for the order parameter in the superfluid transition in Fig. 1.15) and an anomaly in the specific heat (of. Fig. 1.16, which shows the temperature dependence of Cs, the specific heat under saturated vapour pressure, for *He; note that the temperature scales are progressively expanded by factors of 1000). Table 1.2 gives a brief list of phase transitions and the order para- meters of each. For pedagogical reasons, we shall restrict most of our 1,§3 PHASE TRANSITIONS IN OTHER SYSTEMS 21 Tasue 1.2 Partial list of phase transitions and the order parameters

associated with each. Adapted from Kadanoff et al. (1967) ‘Transition Meaning of

Free choice in liquid-ges P— bo p > 0 = liquid # p< 0 = vapour (2 choices) ferromagnetic magnetization M if n equivalent applied magnetic ‘easy axes’ field, H, along easy 2n choices axes antiferromagnetic sublattice if n ‘easy axes’ not physical magnetization 2n choices Heisenberg model magnetization M direction of M H ferromagnet (can choose any value on surface of sphere) Ising model ferro- 8, 2 choices magnet - superconductors A (complex gap phase of A not physical parameter) suporfiuid <#> (condensate wave phase of <> not physical function) ferroelectric lattice polarization _ finite number of electric field choices phase separation concentration 2 choices @ difference of chemical potentials discussion in this monograph to the simpler cases of fluid and magnetic phase transitions. Suggested further reading Andrews (1869). Hopkinson (1890). Weiss (1907). Temperley (1956). Brout (1965). Mendelssohn (1966). Landau, Akhiezer, and Lifshitz (1967). Widom (1967). Stanley (1971a). 2 USEFUL THERMODYNAMIC RELATIONS FOR FLUID AND MAGNETIC SYSTEMS Tuer material in this chapter is intended to serve as @ brief review of those aspects of thermodynamics that will be used to describe phase transitions and critical phenomena. We shall first review the thermo- dynamic relations appropriate for a fluid system, following which we shall discuss the analogous relations for a magnetic system. 2.1. The thermodynamic state functions U, E, G, and A We begin by recalling the first law of thermodynamics, which states that the differential dU, defined by the relation aU = dQ - aw, (2.1) is exact. Here dQ is the differential quantity of heat absorbed by the system and dW = PY is the differential quantity of work done by the system, where P is the pressure and V is the volume. That dU is exact means that there exists a function U representing the internal energy of the system, such that the change in U, AU = i dU, is the same for all processes leading from a given initial state i to a given final state f. We say that U is a state function because, providing we choose U = 0 for some reference state i, U depends only on the state f. We shall usually write U = U(S, V), where S is the entropy. Three other state functions are useful—the enthalpy H, the Gibbs potential G, and the Helmholtz potential A. They are defined by the relations E = E(S, P) = U + PY, (2.2) @ = OT, P) =U - TS + PY, (2.3) and A=A(T,V) =U —- TS. (2.4) 2,§2 THE THERMODYNAMIC SQUARE 23 2.2. Differential relations for the state functions: the thermo- dynamic square We will frequently use the following two mathematical lemmas, which the reader can easily prove for himself. Luma 1. Let df be an exact differential of the function f(x, y). Then df = u(x, y) de + v(x, y) dy (2.5) with wen=(Z) 5 on = (2) s (2-04) and ou) OF oF fv (i).- yae~ way ~ es) Luma 2. Let x, y, and z satisfy some functional relationship f(x, y, z) = 0(eg. P, V, and 7 satisfy an equation of state of the form f(P, V, 7’) = 0). Let g be a function of any two of x, y, and z. “ (2), -{@)}° an 2,2) (2), --» es (=), - @),); i We begin with the differential expressions for the four state functions. The first relation is obtained from eqn (2.1) with the substitutions dQ = TdS and dW = PdV; the remaining relations are obtained from (2.2)-(2.4) by writing (PV) = PaV + VdP and d(7'S) = Td + SAT. Thus we obtain dU = Tas - Pav, (2.10a) dz = TdS + VaP, (2.10b) dG = —SdT + VapP, (2.100) d4 = -SdT — Pdv. (2.10d) 24 FLUID AND MAGNETIC THERMODYNAMICS 2,§2 On using (2.6a), we find A convenient mnemonic device for recalling these equations is the thermodynamic square shown in Fig. 2.1(a). On each side of the square voAoTr oxo S E P (a) PoE & P) E 8 ofan U 7 A m a4 (b) Fro. 2.1. (a) The thermodynamic square. Note that the arrows enable one to get the sign right in eqns (2.10) and (2.11). (b) Two thermodynamic squares rotated 180° as required to obtain the Maxwell relation, eqn (2.124). appears one of the state functions, surrounded by its two natural in- dependent variables. The reader can easily discover how to use the square to construct the relations (2.10) and (2.11). On applying eqn (2.6b) of Lemma 1 to each of eqns (2.10) and (2.11) we obtain the four ‘Maxwell relations’: a aP) of) _ _(#)\, 1 Gr : ak (2.128) (=), = (3), (2.12b) 2,53 TWO BASIC RESPONSE FUNCTIONS 25 G2), -, em (- en The Maxwell relations are also easily obtained from the thermodynamic square, providing we rotate it to the appropriate position. Thus, by rotating the square 180° (see Fig. 2.1(b)) we can read off the last relation, eqn (2.124). Clearly a knowledge of any one of the four state functions U, H, G, and A for all values of its natural variables ((S, V), (S, P), (7, P), and (T, V) respectively) is sufficient to determine the remaining three state functions. For example, U can be obtained from A by using U=44+T8=A- (Fn), e -™cal7)),- (2.18) 2.3. Two basic response functions: the specific heat and the compressibility Two types of response function will be introduced in this section: (i) the specific heats Cp and C,, which measure the heat absorption from a temperature stimulus, and (ii) the compressibilities K, and Ks, which measure the response of the volume to a pressure stimulus. (i) The specific heat at constant x ( = Por V) is defined by _ (42) _ (28 a Ifthe volume is held constant, then os U) eA or= (Fn), = (én), ~"(am)y 9 using (2.10a) and (2.11d). Similarly, if the pressure is held constant, thendQ = dU + PdV =d(U + PV) = dH, sothat c= (8), (B).- 8) (ii) The isothermal and adiabatic compressibilities are defined by the relations 1(aV\ _1(a\ __ 1 /a@ Kr=-% (=), ~ =(3), “> (er), (2.17) 26 FLUID AND MAGNETIC THERMODYNAMICS 2,§3 _ 1 (av) _1/a\ _ _ 1 (eB Ke=—7(ap),~;(@8),-~ 7 (ge), 9 where p = mn is the mass density, and n = N/V is the average particle density. A third response function which receives considerably less attention is the coefficient of thermal expansion, ap, defined by es y(n), (2.19) The response functions are not all independent of one another. Two particularly useful relations among them are and K,(Cp — Oy) = TVa2 > (2.208) and O,(Kp — Ks) = TV08. > (2.20b) To derive eqns (2.20a) and (2.20b), we begin with the relations as as aS\ (a (ar), ~ (ex), + (er) (a), eat) and ev av av\ (a& (ze), ~ (ar), + (os) (@2)_ ey) which are special cases of the general result ew ew" aw (3) ) .() + (2) (2). 2,22 (i). (&).* @),@). aa not T (a as\ (ev Ky(Op — Oy) = -5.(5p), (a0), (sr), (2.288) and T (aS\ (eV\ (ee ls — K) = ~7(5m),(53),(zp), 29) Next we observe that eqn (2.8) of Lemma 2 implies that OV\ (aP\ (at (=e),(n),(=r), ~~ ee) Using the Maxwell relation (2.12d) and eqn (2.7) of Lemma 2, eqn (2.24) becomes (=), (7), - - (x); (25 2,53 TWO BASIC RESPONSE FUNCTIONS 27 On substituting eqn (2.25) into eqn (2.23a) and using the definition (2.19) of the thermal expansion coefficient op, we obtain the first desired result, eqn (2.20a). Equation (2.20b) is derived in an analogous fashion. Thus by eqn (2.8) ()G8),-—- om which may be rewritten using eqns (2.7) and (2.12b) as (én),(3), a (a), (2.27) Finally, eqn (2.23b) together with eqns (2.27) and (2.12c) result in eqn (2.20b). T'>T P'>P (a) (b) Fro, 2.2, Using the principle—often attributed to Le Chatelier—that any spontancous change in the paramoters of a system that is in stable equilibrium will give rise to pro- cesses that tend to restore the system to equilibrium, the reader can complete the following heuristic arguments that (a) 0 = dQ/dT > 0, and (b) K ~ —dV/dP > 0: {a) suppose there has occurred a spontaneous temperature fiuctuation in which tho temperature of the aystem isincreased from T' to 7” ;(b) suppose there occurs a spontaneous pressure fluctuation P—> P’. The non-negativity of these response functions also follows directly from the fluctuation-dissipation relations derived in Appendix A. Now in order that our fluid system be thermally and mechanically stable, the specific heat and compressibility should be positive for all 7’ (see Fig. 2.2). Hence eqns (2.20) imply that Op = Oy (2.28a) and Ky = Ks (2.28b) for all temperatures. In particular, as 7’—> 7’, we shall see that Cp >» Cy and Ky >» Ks. (Note that the relations (2.28) reduce to equalities only (i) when 7’ = 0 or (ii) when ap = 0, as, for example, in the case of water at about 4 °C.) Equations (2.20) and their magnetic analogues will prove extremely useful in our discussion below. In 28 FLUID AND MAGNETIC THERMODYNAMICS 2,§4 particular, it is elementary to obtain from eqns (2.20) the following corollaries: Cp|Cy = Ky/Ks, (2.29) TV Op = = 2.30) "Ke Ky ae and C, TVeKs (2.31) ¥* Kr(Kr — Es) 2.4. Stability and convexity relations In this section we shall see that the positivity of the specific heat and compressibility, as required for thermal and mechanical stability, implies certain convexity properties of the thermodynamic potentials A and @. These properties will be used in Chapter 4. Dermirioy. A function f(x) is a convex function of x providing the chord joining the points f(x,) and f(x) lies above or on the curve (2) for all xin the interval a, < @ < a. For example, the function f(x) = «? is convex (see Fig. 2.3) as is the Fra. 2.3. The parabola f(z) = 2? is a convex function since, for example, (i) any chord lies above or on the parabola, (ii) any tangent lies below or on the parabola, and (iii) the second derivative f"(z) is positive. Therefore the parabola f(z) = —2? is a concave function, function f(x) = |x|. Note that this definition does not require f(z) to be differentiable. However, should the derivative f’(x) exist, then it follows that a tangent to a convex function always lies below the function except at the point of tangency. If, moreover, the second derivative of a convex function exists, then f” (x) > 0 for all x. 2,84 STABILITY AND CONVEXITY 29 Dermtrioy. A function f(x) is a concave function of x if the function —f(«) is convex. Thus, for example, f(z) = —27is concave. In the previous section we noted that for a mechanically stable system, the specific heat and the compressibility must be positive for all temperatures. Thus we are led to the following theorem. TuEorEM. (i) The Gibbs potential G(T, P) is a concave function of both temperature and pressure, and (ii) the Helmholtz potential A(T, V) ig a concave function of the temperature and a convex function of the volume. Proof. We begin by noting from eqns (2.11¢) and (2.11d) that Hence ue as 1 (iz), - -(ér), = ~20<° ae and A os 1 (sr), = - (3), = - pr $0. (2.34) Since the second derivatives exist (except for 7 = 7',) and are non- positive, it follows that both G(7’, P) and A(T, V) are concave functions of temperature. Next we observe, again from eqns (2.11c) and (2.114), that @@ @ (=), = (=), = -VKr<0 (2.35) and eA aP\ _ on Fr), 5 - Gr), = (VEp)-1 > 0, (2.36) from which it follows that G(T’, P) is a concave function of pressure and A(T, V)is a convex function of volume. Corottary. The curvature of G(T', P) with respect to P (for fixed T) is the negative reciprocal of the curvature of A(T’, V) with respect to V (for fixed 7’). Proof. Directly from eqns (2.35) and (2.36) it follows that (=), - - (er), * ean 4—ievr. 30 FLUID AND MAGNETIC THERMODYNAMICS 2,85 2.5. Geometrical interpretation of the Gibbs and Helmholtz potentials First let us note from eqns (2.3) and (2.4) that G(T, P) and A(7, V) are simply related by G = A + PV. Therefore, given one of these potentials, we may easily find the other by means of the simple geo- metric construction shown in Fig. 2.4 (for a fixed temperature above T..). Note that the convexity relations for A and G imply that the second derivatives with respect to any argument have the same sign for all values of the argument, thereby guaranteeing that no ambiguities will arise in this construction. More familiar quantities are the volume and vo 'P Fie, 2.4, Geometrical interpretation of the relation betweon the Gibba and Helmholtz potentials at a fixed temperature T > T. This construction, adapted from Morse (1969) and Griffiths (unpublished lecture notes), is based on the identities A = @ — PV and V = (8@/@P);. Tho vertical distance betweon the dashed lines in (a) and (b) is the product PY, and on subtracting this from @ we obtain A. Note that the construction indicated is unambiguous because G(T, P) ia a concave function of P (for all P) and A(T, V) is a convex function of V (for all V), Also shown are the volume as @ function of pressure (obtained from the pressure derivative of G(T, P)) and the pressure as a function of volume (obtained from the volume derivative of A(T, V)). the pressure, given by eqns (2.1lc) and (2.11d). These quantities are shown in Figs. 2.4(c) and 2.4(d). By definition, a first-order phase transition is characterized by a finite discontinuity in a first derivative of the Gibbs potential G(T’, P). In Fig. 2.5(a) @ is plotted as a function of P for a fixed temperature less than T. We see that at a certain pressure the tangent to the curve 2,55 THE GIBBS AND HELMHOLTZ POTENTIALS 31 changes discontinuously. This discontinuous change in the first deri- vative of @ corresponds to an infinite curvature in @ and hence, by eqn (2.37), to a zero value for the curvature of A. Therefore we expect that a plot of A as a function of V will have a corresponding straight-line portion, as in Fig. 2.5(b). P-V (pressure-volume) and V—P isotherms are shown in Fig. 2.5(c) and (d) respectively. T-4, (2.38) P+ 4. (2.89) 2,88 MAGNETIC STATE FUNCTIONS 33 The minus sign in (2.38), though bothersome, is absolutely necessary. For example, the response function xy = 21/0H is positive (the magnet- ization increases with magnetic field), whereas the analogous derivative for the fluid system, 2V/@P, is negative (the volume decreases with pressure). There is a second magnet-fluid analogy, in which M —> p — p, and H-> — p, where here » denotes the chemical potential. This is pedagogically the sounder analogy, since » is the thermodynamic variable conjugate to the density and the density is the order parameter. However for our purposes it is more convenient to develop the analogy stated in eqns (2.38) and (2.39). 2.7. The thermodynamic state functions for a magnetic system The first law of thermodynamics for a magnetic system may be written either as aU, =dQ—- MdaH (2.402) or as dU, = dQ + Ham. (2.40b) The distinction is carefully discussed by Kittel (1958), Guggenheim (1967), Callen (1960), Reif (1965), Wannier (1966), and Fay (1965). Here we shall use eqn (2.40b). The remaining three state functions are then defined, in analogy with the fluid case, as E = E(8,H) =U — MH, (2.41) G = QT, H) =U —- TS —- MH, (2.42) and A=A(T,M) =U - TS. (2.43) 2.8. Differential relations and the thermodynamic square for magnetic state functions The substitutions V > —M and P > H are sufficient to obtain the results of this section from the corresponding expressions in § 2.2 for fluid systems. Thus the thermodynamic square is as shown in Fig. 2.7, and the differential expressions for the magnetic state functions are dU = 7TdS8 + Ham, (2.440) az = TdS — MH, (2.44b) d@ = -SdT - MaH, (2.44¢) /d4= -SaT + Hay. (2.44d) 84 FLUID AND MAGNETIC THERMODYNAMICS 2,98 (e) (b) UsU (S,V) (AU=dQ-P dV) U=U (8,M) (dU=dQ+H dM) A=A (T,V)=U-TS A=A (7\M)=U-TS au=T as-P av = (er), =-(&), dU=T7 d8+H aM >(F5), = G53), dE=T a8+V dP = (e)- (ss), ab=T4S—M dH >()= ~G), dG=-S aT+V dP > (6) --(e7), ag=—sar—aan = (FF) = en), d4=-s ar-P av, (Fr), (sy, d4=-8 arsnan = (55) =-(en) VOA T -M. A T >< >< sz P s’ eH Fra. 2.7. Comparison of thermodynamic relations for (a) fluid aystems and (b) magnetic systems. Note that the two are related by the aubstitutions V > —M and P > H. Similarly, we have T= (),3 _M= (a), (2.45b) _g= (5), -M= (iz), (2.45) oe (=), 3 H = (Si), (2.454) The four corresponding Maxwell relations are om =, a 2,§9 MAGNETIC RESPONSE FUNCTIONS 35 (zz), ~ (=n), (2.46¢) (a), =- (@),: (2.46d) 2.9. Magnetic response functions: specific heat and susceptibility (i) Corresponding to Cy and Cp we introduce Cy and Cy, defined by as’ aU" aA ou = (5m), = (5m), =~? (zr), ae and os ok ee o="(m),-(a),--T(ar)e (ii) Instead of the isothermal and adiabatic compressibilities K; and Ks, we now have the isothermal susceptibility w= (),--G, a and the adiabatic susceptibility om) 3) =(2) =--(4). 2.50 xs Ge 7 0H), (2.50) Note from eqns (2.17) and (2.18) that the ‘normalizing’ factor V, which occurs in the definition of the compressibilities, does not appear in the corresponding definitions of the magnetic susceptibilities. It is now straightforward to verify the relations xr(Cu — Cy) = Toy? > (2.51a) and Culxr — Xs) = Tox?, » (2.51b) which are analogous to eqns (2.20). Here «,, is defined by eM ox (2. com) In analogy to eqn (2.29), we have Cu|Cu = xelxs- (2.58) We mention, for later reference, that eqn (2.51a) may be written in the form Oy — Oy = Tots xr (2.54) 36 FLUID AND MAGNETIC THERMODYNAMICS 2,§10 where here _ (2H\ _ _ oy ae (=), -- (2.55) 2.10. Convexity relations for magnetic systems In the case of a magnetic system we cannot argue, as we did in § 2.3 for fluid systems, that x and C are both non-negative. In fact we know T>T. Fie. 2.8, Geometrical relation betwoen G(T, H) and A(T, M) for a fixed temperature T > T,. Also shown are the magnetization as a function of field (obtained from M = —(8G/aH);) and the field as a function of magnetization (obtained from H = (2A/aM)z). that y < 0 for a large number of diamagnetic materials. However, it can be shown (Griffiths 1964) that for a system whose Hamiltonian is of the form H=H,-HM (2.56) (where H is the magnetic field and .# is the magnetization operator) that (i) the Gibbs potential @(7', H) is a concave function of both temperature and magnetic field and (ii) the Helmholtz potential A(T, M4) is a concave function of the temperature and a convex function 2, §11 THE POTENTIALS FOR A MAGNETIC SYSTEM 37 of the magnetization. It is reasonable that many real ferromagnets are closely “approximated by a Hamiltonian of the form of (2.56); for example, familiar magnetic models, such as the Heisenberg and Ising models, can be written in the form (2.56). 2.11. Geometrical interpretation of the thermodynamic poten- tials for a magnetic system For magnetic systems G(7',H) = A(T, M)— MH. Hence the potentials G and A are related to each other by geometric constructions T 7’, and in Fig. 2.9 for T < 7',. Also shown are M-H isotherms (obtained from the relation M = —(2@/@H);) and H-M isotherms (obtained from H = (24/2M)z). Two slight differences between the magnetic and the fluid cases 38 FLUID AND MAGNETIC THERMODYNAMICS 2,91 are worth noting here: (i) A(7’, M/) is an even function of M, correspond- ing to the physical idea that M(H) should be an odd function of H. We will use this fact when we consider the Landau theory in Chapter 10. (ii) The straight-line segment of the A—M isotherm is horizontal. This makes proving the Griffiths inequality easier for a magnetic system than for a fluid system (cf. Chapter 4). Suggested further reading Kittel (1958). Callen (1960). Fay (1965). Reif (1965). Wannier (1966). Guggenheim (1967). Kubo (1968). Kittel (1969). PART II CRITICAL-POINT EXPONENTS AND RIGOROUS RELATIONS AMONG THEM 3 CRITICAL-POINT EXPONENTS 3.1. Definition of a critical-point exponent Iy recent years the study of critical phenomena has come to focus more and more on the values of a set of indices, called critical-point expo- nents, which describe the behaviour near the critical point of the various quantities of interest. We begin with a precise and general definition of a critical-point exponent to describe the behaviour near the critical point of a general function f(e), where T-T7 _T oe? (3.1) serves as a dimensionless variable to measure the difference in tem- perature from the critical temperature. We assume that this function f(€) is positive and continuous for sufficiently small, positive values of ¢, and that the limit d= lim BL) > (3.2) oo Ine exists. This limit, A, is called the critical-point exponent to be associated with the function f(e). As a shorthand notation, we fre- quently denote the fact that A is the critical-point exponent for the function f(e) by writing f(e) ~ &'. It is important to stress that the relation f(<) ~ <* does not imply the Telation f(e)= Ar [ez =A), (3.3) 40 CRITICAL-POINT EXPONENTS 3,§1 although of course the converse is true. In fact, it is relatively rare that the behaviour of a typical thermodynamic function is as simple as (3.3); in general we find that there are correction terms, and eqn (3.3) is replaced by a functional expression such as fle) = Ac(L + BY +--+) [y > 0}. (3.4) ‘We see that the definition (3.2) of a critical-point exponent does not distinguish the functional forms (3.3) and (3.4)—for both functions, A=a, We may rightfully question why we should focus on a quantity such as the critical-point exponent, which contains considerably less infor- mation than the complete functional form. The answer seems to lie in the experimental fact that sufficiently near the critical point the behaviour of the leading terms dominates. Therefore log-log plots of experimental data, such as that of Fig. 1.10, display straight-line behaviour suffi- ciently near the critical point, and hence the critical-point exponent is easily determined as the slope of this straight-line region. Hence critical- point exponents are measurable while the complete function may not be. A second reason for focusing on the critical-point exponents is that, as we shall see, there exists a large number of relations among the expo- nents that arise from fundamental thermodynamic and statistical mechanical considerations and thus transcend any particular system. We conclude this section with a discussion of four simple examples (cf. Fig. 3.1). These examples are designed to illustrate the extent to which there is a simple one-to-one relationship between the magnitude of a critical-point exponent and the qualitative behaviour of the func- tion under consideration near the critical point, « = 0. In the first two of these examples, illustrated in Fig. 3.1(a) and (b), we show that nega- tive values of the critical-point exponent A defined in eqn (3.2) corres- pond to a function f(e) which diverges to infinity at the critical point, while positive values of \ correspond to a function f(e) which approaches zero. More important, perhaps, is the fact that the smaller the magnitude of A, the ‘sharper’ is the behaviour of f(c) in the sense that for negative the divergence is more sudden while for positive the approach to zero becomes steeper. The intermediate case when A is zero does not correspond to a unique type of behaviour; in fact, 4 = 0 can correspond to a logarithmic divergence as in Fig. 3.1(c), to a cusp-like singularity as in Fig. 3.1(d), or to a perfectly analytic function with no anomalous behaviour worse than, say, a jump discontinuity. Since these three possibilities differ 3,§1 DEFINITION OF CRITICAL-POINT EXPONENT 41 sharply from one another, we are motivated to introduce yet another sort of critical-point exponent that serves to distinguish the case of the logarithmic singularity from that of the cusp singularity. To find the exponent ) that describes the singular part of a function f(e) with a flo~e flare 1234 5 6 Or 02 03 x 10-4 (a) (b) fl)=Alln [+B BS Dest se! Ae oor x10 (c) (a) Fic. 3.1. Examples of the various types of behaviour near ¢ = T/T, — 1 (a) fle) ~ & with A < 0. (b) fl) ~ & with A> 0. (0) fle) ~ [In f(e)|, 80 that A (d) f(e) ~ const. — 4 ao that A = 0, but A, = 4. cusp-like singularity, we first find the smallest integer j such that the derivative d/f/d = f(e) diverges as «> 0. We then define Ay = j + tim BLO! (3.6) <0 For example, the function plotted in Fig. 3.1(d) is f(e) = const. —e?, 42 CRITICAL-POINT EXPONENTS 3,§2 and the first derivative diverges as «~* when «> 0. Hence from eqn (3.5) it follows that 4, = 1 — 4 = } for this example. The remaining sections of this chapter are designed to introduce the most commonly used critical-point exponents for fluid and magnetic systems. In general, the same notation for a critical exponent is used for the analogous fluid and magnetic functions, so that it is convenient to treat them together. It is important to stress that not all the exponents are defined along paths on which T' is varying (cf. Table 3.1), and the discussion of this section applies to any function f(x) where « might denote M or H as well as e. TasieE 3.1 Notation for measurements at constant P(or H), V(or M), T, and S Fluid ‘Magnet H = const. (=0) (critical) isochamp p Pe) M = const. (=0) (critical) isochore (critical) isomeg (critical) isotherm 8 = const. (=S,) (critical) isentrope Note that measurements along the vapour pressure curve P = Pyy,(7) are not at constant pressure, nor at constant volume. 3.2. The critical-point exponents a, 8, y, 5, v, and 7 We begin our survey of a few selected critical-point exponents with a further analogy between fluid and magnetic phase transitions. Fig. 1.3 shows the temperature dependence of the liquid-gas density difference and the zero-field magnetization M,(7’) of an idealized (single-domain) ferromagnet at constant magnetic field H = 0. From the shape of these curves it would seem natural to postulate temperature dependences in the neighbourhood of 7’, which have the form {eu(T) — palT)}2p. = BL — T/T) +--+ (3.6) and. M,(P)|Mo(0) = BL — TIP.) + +++. (3.7) As a shorthand notation, we commonly write Ap = py — pa ~ (—<€)? (or M ~ (-«)*). 3,§2 THE EXPONENTS «, 8, y,8,v, AND 43 The normalization constants p, and M(0) are included in order that the coefficient @ vary only slightly from system to system (generally @ is of the order unity). Fig. 1.2 shows PV isotherms and HM isotherms for fluid and mag- netic systems respectively. The slopes of these isotherms are propor- tional, respectively, to the inverse isothermal compressibility Ky * and the inverse isothermal susceptibility yz1. Although both Ky and yp diverge to infinity as 7’ —> 7’, we must distinguish whether the critical point is being approached from above or from below. Thus we define the (not necessarily equal) exponents y’ and y by C(-) ML te) (P< 7, p= pl) or polT)] Mae ray) Pet hae "88) for the fluid case, and 0 _ [O(-9-"(Lte) [P< 7, H=0) Xelxe = eae ))) fot, Hag &9) for the magnetic case. Here Kp. = 1/P> = m/kTp, is the compressibility of an ideal gas of density p, at 7’ = T,. Similarly, y> is the suscepti- bility of a system of non-interacting magnetic moments (paramagnet) eyaluated_at-the-critical point, Measurements often come sufficiently close to 7’, that Ky/K> is as large as 10°. Note that we have included a minus sign in the definitions of y’ and y in order that both exponents be positive. The exponent 5 describes the variation of P — P, with p — p, (and of H and M) along the critical isotherm, 7 = T,: (P — P.)P3 = Dlplpo — 1\*sgn(p — po) [= Teh (8:10) HH = O|M,AT = TMT = 0) [P= TJ. (8.11) Here P? = &T',p,/m is the pressure that the system would have at p =p,and T = 7, if its particles did not interact; also, H = kT',/m, where here mmo is the magnetic moment per spin. Thus the index 5 serves to describe the degree of the critical isotherm, and from casual examination of Fig. 1.2 we might expect 8 to be an odd integer such as 3 or 5. Experimentally, 8 is found to have values (for different systems) which are between ~4 and ~6 and are generally non-integral—whence the complicated form of the defining equations (3.10) and (3.11). As a 44 CRITICAL-POINT EXPONENTS 3,§2 shorthand notation, we frequently write AP ~ (Ap)’ and H ~ M’. The larger the value of 8, the ‘flatter’ the critical isotherm (and the harder it is to obtain an accurate experimental determination of 5); for example, 8 ~ 15 for the two-dimensional Ising model, corresponding to an extremely flat critical isotherm. The specific heat exponents «’ and « are defined by the relations oy = (tM ED < Top = pu (2) oF pol 7)(3-12) VA Vetee(Ltee) [2 > Top = pel (3.13) for the fluid systems, and _fot(-yr ete) (P<) Cen geedany pony Ha) Cas) for magnetic systems. Note that we have departed from our strict fluid-magnet analogy Veo —M and P< FH in our definition (3.14). Fisher (1967) has argued TasLe 3.2 Summary of definitions of critical-point exponents for fluid systems. Here = T/T, -1 Expo- Definition Conditions Quentity nent « P=P, p=pe we Cyn (-0-" <0 0 0 specific heat at constant volume V = Vs a Cyne >0 0 0 Bo pu— pan (—e* <0 0 #0 liquid-gas density dif- ference (or shape of coexistence curve) Y Kp~(-o7" <0 0 #0 isothermel compressibility y Krve? >0 0 0 8 P-P, 0 £0 #0 critical isotherm ~ |p. — pal? san (p. — pa) vo gx (-9-" <0 0 #0 correlation length » éwer >0 0 0 2 Or) ~ fej ae 0 0 0 pair correlation function (@ = dimensionality) BE sam nw (~9-*1Ge-? <0 0-0 successive pressure deriva- ope tives of the Gibbs poten- tial G(T, P) bn OS sae co unae-m 50 0 0 ope — 3,§2 THE EXPONENTS « f, y,8,»,AND_ — 45 (a) BS cy] (b) P 7, while for J’ < T,, the two-phase M = 0 specific heat also corresponds to H = 0 since only then can oppositely magnetized domains coexist to yield zero total magnetization. Suppose we measure Cy as a function of temperature for a system whose volume V9 is not equal to V,, for example along a path parallel to, but slightly below, the line p = p, in Fig. 1.3, corresponding to Vo > V. Or po < pe. Our expectations are shown in Fig. 3.2: Cy, wy, TABLE 3.3 Summary of definitions of critical-point exponents for magnetic systems. Heree = T/T, -—1 Expo- Definition Conditions Quantity nent € H M a” On ~ (-9-" <0 0 0 specific heat at constant magnetio field «Ores >0 0 0 B M~(-#) <0 0 #0 zero-field magnetization Y we (-o" <0 0 0 zero-field isothermal susceptibility yooxwwey >0 0 (0 3 -H~ |M|*egn(M) 0 #0 #0 critical isotherm Yo gw(-—" <0 0 <0 correlation length vy gwe® >0 000 an Dir) ~ |r |-e-aem 0 0 0 pair correlation function (d = dimensionality) A, SE awn (—a-*tae-y <0 0 0 successive field dovivatives of the Gibbs potential G2, H) 2 Ay pe SOM w ME GR-D 0 0 0 bier. 46 CRITICAL-POINT EXPONENTS 3,§3 will undergo a jump discontinuity at the temperature 7')(<7',) at which the isochore p = po passes through the coexistence curve. Experimentally, either the values of the exponents a’ and « are zero or they are small positive numbers (2 ~ } is the three-dimensional Ising model prediction). The case a = 0 corresponds in some measure- ments to a logarithmic singularity, and in others to a cusp-like sin- gularity (in which case a, # 0). The exponents v’, v, and 7 refer to the behaviour of the pair correlation function @ (r) in the critical region. Since a rather extensive discussion of @ (r) will be presented in Chapter 7, at this time we give only a brief definition of the exponents themselves. The correlation length ¢ is a measure of the range of the correlation function. It is assumed that pa {Ol T/TIY (P< PH =O} f(T/T,- 1" [PT > TH = 0} In many cases the pair correlation function at 7’ = 7, falls off to zero with distance r with the simple power law form (3.15) [7 =7,; P=P, (rH =0)}, (3.16) Ww paaten ar) ~ 5 thereby defining the critical-point exponent ». Here d is the dimen- sionality of the system. The exponent definitions are summarized for fluid and magnetic systems in Tables 3.2 and 3.3 respectively. 3.3. Numerical values of critical-point exponents Values of the exponents a’, «, 8, y’, y, 8, v’, v, and 7 are given in Table 3.4 for selected fluid and magnetic systems, and for a few theoretical models. Despite the fact that the critical temperatures change sub- stantially from material to material (see, e.g. Table 3.5), we notice a distinct similarity among the values found experimentally for each exponent. At one time many workers believed that all materials have the same exponents, though this once-hoped-for universality has not been borne out by more recent experiments. For example, many of the early measurements on magnetic systems indicated that 6 had a value extremely close to }, but very recently Ho and Litster (1969) have shown unequivocally that 8 = 0-368 + 0-005 for the insulating ferro- magnet CrBr,. What can theoretical models say about the critical-point exponents? First of all, a given model predicts a given set of critical-point 47 NUMERICAL VALUES 3,§3 OL-0~ wI~ 0- ~ (e80-0~) an ae — — iave-0~) = (qjeusem) Jepour Brequesion, § = P 8€9-0~ iw t~ TH0-0~ a~ al He ed ftw Gpmp) 30 F~ 40 f~ yepow Susy ¢ = P suoypunrxoddy T I~ (Bot) 0 t st~ I i~ t (201) 0 pow Busy Z = P I zg I-=* 0 9 ae — t _ Tepour jwouoyds § = P I z I- =" 0 ¢ = = oxfuze—ar04su3Q) t 1 (ost) 0 0 8 t T t (o8tp) 0 Teorsse]o spp mas _ STe1 _ _ eh _ _ 898-0 _ "gt a a 90-0 = = aa = 86-0 s10- = sng a oe 0 al wer a al a¥-0 80- = fe IN sreuboyy ad eI = a aa 00 BIN 98-0 w0> ex a ser Tow - or al ot~ ¥e0 To~ 00 spent a 4 » u 8 A é g 2 “hb D woyskg suapshis payoajas sof spuauodxa yurod-yoornso fo sonyo 4 ye wavy, 48 CRITICAL-POINT EXPONENTS 8,93 TABLE 3.5 Critical-point parameters (critical temperature, critical pressure, and critical density) for selected fluids Fluid 1, (K) P,(atm) pp (gem Water 647-5 218-5 0-325 Aleohol 516-6 63-1 0-28 Ether 467-0 36-5 0-26 Carbon dioxide 304-20 72-8 0-46 Xenon 289-75 57-64 1-105 Oxygen 154-6 49-7 0-41 Argon 160-8 48:34 0-53 ‘Hydrogen 32-98 12-76 ‘Helium-4 519 225 0-069 Helium-3 3-324 1.15 0-04178 exponents, independent of the location of the critical point. Moreover, the van der Waals theory of a fluid, the molecular field theory of a mag- net, and the Landau theory (at least in their original forms) all give the same values for each of the exponents. These theories we call classical to suggest the fact that until fairly recently they were essentially the only theories of critical-point exponents. However, it is clear from Table 3.4 (and Fig 3.3) that these classical theories fail to predict the observed values of the exponents. Also failing in this regard are the Ornstein-Zernike theory and the various exactly soluble models such as the two-dimensional Ising model, the spherical model, and the ideal Bose gas. On the other hand, the predictions of the three-dimensional Ising model and the three-dimensional classical Heisenberg model do appear to mirror to a certain extent the data on, respectively, fluid and mag- netic systems. Certainly a first goal of any theory of critical phenomena is to find such theoretical models and, moreover, to understand which features of the models are relevant in determining the values of the critical-point exponents and which are not. Even if we should eventually succeed in this first goal, we will still be left with the question, ‘Isn’t there some underlying theory of all these exponents which tells us how they hang together?’ A second goal, then, is to study relations among the various exponents. One step in this direction is taken in the follow- ing chapter, where we shall demonstrate certain rigorous inequalities which relate the various exponents. In Chapter 11 we shall study in some detail the recent homogeneous function or scaling law theory of 3,§4 THE EXPONENTS 4 AND @ 49 exponents which, among other results, predicts that the inequalities be satisfied as equalities. 3.4. The exponents A and @ We introduce in this section additional critical-point exponents that will be needed elsewhere. (a) dl ——van der Waals I theory i Cy Chel] ACyWe® —— Experiment (ACyay)~e 1 11 I ial /\ / \ Te Te (Experiment) (van der Waals) van der Waals Fra, 3.3. Schematic comparison of the predictions of the classic van der Waals theory with typical experimental measurements—on fluid systems—of the exponents a (specific heat), B (coexistence curve), y (isothermal compressibility), and 8 (critical isotherm). Experiments of the sort shown in (b) have been carried out on xenon by ‘Habgood and Schneider (1954). 3.4.1. The gap exponents A; and Age Recall that for a magnetic system, the spontaneous magnetization and the zero-field susceptibility are proportional, respectively, to the first 50 CRITICAL-POINT EXPONENTS 3,94 and second derivatives of the Gibbs potential G(7', H) with respect to the magnetic field H (and evaluated at H = 0). In 1963 Essam and Fisher suggested that one might consider higher field derivatives of G(T, H), thereby defining a sequence of exponents A; called gap exponents. Thus (0G/0H)p = QM ~ (1 = T/T,)~*1G (3.17) and (@G@/0H?), = G® ~ (1 — T/T,)~“2Q™, (3.18) In general, then, (¢4|H\p = GY ~ (1 — T/T,)-%AE-¥, > (3.19) In eqns (3.17)-(3.19), H = Oand 27> Tf. Since GY ~ M ~ (1 — T/T), and the specific heat Cyio ~ @G/8T? diverges as (1 — 7/T,)~“, it follows that @ ~ (1 — T/T.)?-” and hence AL =2-a' -B. > (3.20) Similarly, the fact that G® ~ yp ~ (1 — T/T,)-” means that Ab=B+y. > (3.21) For the molecular field theory, A; = $ for all 7(e.g.Aj = 2-0’ ~ B= 3.Ah=B+y' =4%,---). For the two-dimensional Ising model, a’ = 0 and f = } are rigorous results. Hence A; = Az = 44 exactly providing y’ = 3. Numerical studies of Essam and Fisher (1963) find Aj = 1-87 + 0-05 ~ 48, so that it is certainly plausible that Ay = 4 for all for the two-dimensional Ising model. Note that the conjectured equality Aj = Aj (or 2 — a’ — 8 = B + y’) serves to relate the expo- nents «’, B, and y’: a + B+ y' = 2 (3.22) The conjectured equality (3.22) is valid for the molecular field theory and for the two-dimensional Ising model, and it is not inconsistent with numerical studies on the three-dimensional Ising model. Above T,,, all odd field derivatives of @(J', H) are zero when evaluated at H = 0. Hence we generalize (3.19) to (O%G]@H™)p = G2 ~ (T/T, — 1)-%4aG%-®, (3.23) Numerical calculations for the Ising model give Ag & Ag & Ag =e (3.24) 3,95 RELATIONS FOR CRITICAL-POINT EXPONENTS 51 for two-dimensional lattices and Ay ~ Ag % Ay © 1-56 + 0-03 ~ 28 (3.25) for three-dimensional lattices (Essam and Hunter, 1968). For the three- dimensional Heisenberg model, Ay ~ Ag ~ Ay ~ 181 (3.26) for spin quantum number S = } and Ay ~ Ag +++ 1-78 (3.27) for § =o, the classical Heisenberg model (Baker, Gilbert, Eve, and Rushbrooke 1967, Stephenson and Wood 1968). Thus we see that these model systems lend support to the general conjecture that the gap exponents are equal for all orders. It should be remarked that the gap exponents are largely of theoretical interest at the present time, since the higher derivatives of the Gibbs potential have not yet been measured experimentally. 3.4.2, The vapour pressure curvature exponent 0 The specific hest C, may be written, directly from eqn (2.10c), as Cy = TV(2P/0T)y — TN(8%./8T)y, (3.28) where , = G/N is the chemical potential. Yang and Yang (1964) pointed out that if C, is singular at 7’ = 7',, then from eqn (3.28) (22P/87)y or (24 /2T?)y—or both—will be singular. We define the exponent @ by 0°P’ (=) ~ (1 — 2/T,)~*. (3.29) Thus 0 is a measure of the degree of divergence (if any) of the curvature of the vapour pressure curve as J’ —> 7’, from below (see Fig. 1.1). Yang and Yang argue that for the lattice-gas model of a fluid (2%1/2T?)y— yo approaches zero at the critical point, so that @ = a’. However, for real gases the curvatures of » and P might both be singular, so that 6 might differ from a’. Recently Moldover (1969) has found experimentally that for “He (21/277), is less singular than (2?P/27), and may even be constant near 7’,. In any case, the specific heat singularity in *He appears to be dominated by (02P/07),. 3.5. Useful relations among critical-point exponents To conclude this chapter, we present general relations among critical- point exponents that will be of use, for example, in proving inequalities among the various critical-point exponents in Chapter 4, 52 CRITICAL-POINT EXPONENTS 3,§5 Lemma 3. Let f(x) ~ 2%, g(x) ~ 2, and let f(x) < g(x) for sufficiently small positive x. Then A = p. The proof of Lemma 3 is quite straightforward, but for the sake of completeness we shall reproduce it here. The assumption f(x) < g(x) implies that In f(z) < In g(x). Now for x < 1, nz < 0, and Inf) , Inge), (3.30) Ing Ing where the inequality sign is reversed. The relation (3.30) is valid as a-—»0, whence from the fundamental definition of a critical point exponent, eqn (3.2), it follows that A = 9. Lemma 3 is basic to the proofs of exponent inequalities so it is perhaps useful to keep some simple examples in mind. In particular, we note that Lemma 3 is valid regardless of the signs of \ and gy. For example, suppose f(x) = 1/x and g(x) = 1/x *. Since f(x) < g(x) for 0 o (of. Fig. 3.1(a)). Additional relations that are useful in proving exponent inequalities have been given by Griffiths (1965). Suggested further reading Guggenheim (1945). Fisher (1967). Heller (1967). 4 EXPONENT INEQUALITIES THE only rigorous relations thus far proposed among the critical-point, exponents are a set of inequalities. In this chapter we shall present in detail the proofs for magnetic systems of a few representative examples of such inequalities; most of the remaining inequalities that have been proved thus far are discussed in §4.3 and are listed in Table 4.1. 4.1, The Rushbrooke and Coopersmith inequalities 4.1.1. The Rushbrooke inequality: H = 0, T+ Tz We begin by recalling from §2.9 that the specific heat Cy must be positive so that eqn (2.51a) leads to the inequality cre a(S fe Using the definitions of the critical-point exponents from Chapter 3, Cy ~ (-8)7", xr ~(-6)-”, and (@M/0T)y ~ (—«)f~? and using Lemma 3 (§3.5), we see that (4.1) implies that —a’ < 28 — 1) + y’ or equivalently, a + B+y' > 2 » (4.2) The relation (4.2) is generally called the Rushbrooke inequality (Rush- brooke 1963), although actually it was first conjectured, on the basis of heuristic arguments, in the form of an equality (Essam and Fisher 1963). Using the definitions of the gap exponents A; presented in §3.4, eqn (4.2) may be written equivalently as Ay > Ay. (4.3) After the style of Onsager, Rushbrooke first proposed the inequality (4.2) as a discussion remark at the end of a seminar by someone else, namely a seminar which presented the latest (1963) estimates of the exponents a’, 8, and y' obtained from the detailed Padé approximant analyses: a = 0, B = x%, y' = 3. (Numerical approximation methods, B4 EXPONENT INEQUALITIES 4,§1 useful for models that have not been solved exactly, are discussed in Chapter 9.) Rushbrooke calmly questioned the speaker as to whether he had any explanation as to why the quantity « + 28 + y’ (=48) fell short of 2. The speaker replied that he knew of no reason why they should add up to 2 in the first place, whereupon Rushbrooke produced a two-line derivation of the now famous inequality (4.2). The first efforts were directed toward finding some loop-hole in Rushbrooke’s thermodynamic argument. Not surprisingly, thermodynamics held up under scrutiny and the errors were then presumed to lie in the numerical approxima- tions. Subsequent arguments based upon series extrapolations now favour the value a’ = 4, so that (4.2) would appear to be obeyed as an equality (Gaunt 1967, Gaunt and Domb 1968, Garelick and Essam 1968, Guttmann and Thompson 1969). For the two-dimensional Ising model it has been proved rigorously (Onsager 1944) that a’ = 0 (corresponding to a logarithmic singularity in the zero-field specific heat) and B = 4 (Yang 1952). Hence the Rushbrooke inequality implies that y’ > 4. Using arguments that fall barely short of being fully rigorous, it has been shown that in fact y' = y = & (Fisher 1964). Hence for the two-dimensional Ising model the Rushbrooke inequality would appear to be satisfied as an equality, as it is for the classical theories for which «’ = 0 (corresponding to a jump discontinuity) 8 = }, and y’ = 1. For the relatively few materials on which a’, 8, and y’ have all been measured, the experimental values often ‘fail to add up to two’ unless the error bars are taken into account. Thus experimental results, as well as theoretical work on the Ising model for d = 2 and 3, suggest that the Rushbrooke inequality may well be obeyed as an equality. This is, in fact, one of the predictions of the (non-rigorous) scaling law theory of exponents, which we shall develop in Chapter 11. We now show that (4.2) fails to be an equality if and only if the quantity R= tim CulCn (4.4) has the value unity (Rushbrooke 1965). We begin by writing eqn (2.51a) in the form 1 = Cyn = Taf|Cuxr. (4.5) (a) Consider first the case R = 1, and define the positive exponent + by the (assumed) dependence near 7’ = T,, OylCg ~ RB - (-)(L + +) = 1 (-*(L Fe). (4.8) 4,§1 OF RUSHBROOKE AND COOPERSMITH 55 Then, on substituting into (4.5), we obtain PH) (-e(1 = (Hg terra 4 (4.7) or 2Z+e=a' t+ Bt+y’ (4.8) (b) Next consider the case R = 1 — y < 1 (i.e. y > 0). Then near T, eqn (4.5) leads to VfL = yh + p= yl toe) = (Hartera tee) (49) and Qaa + Bty’. (4.10) 4.1.2. The Coopersmith inequality: T = T,, H > 0* The Rushbrooke inequality was obtained by setting H = 0 in (4.1) and letting 7’ —> 7. Suppose instead that we set 7 = 7’, and then let H -> 0. In this case, M ~ H}! so that xp ~ H™*-1, Next introduce the exponents p and y% by means of the definitions Cy~ H-?~M-% [P= 7) (4.0) and. S(H)~ -H¥. MY [P= 7). (4.12) Finally, we use the Maxwell relation (2.46c) to replace (2M/27),, on the right-hand side of (4.1) by (28/2H), ~ —H¥-1. Hence (4.1) becomes, for T=T,, H-O(1 +++) > HW-D-ale-9y 4...) (4.13) or p+ %—-1ds21, m» (4.14) Equation (4.14) was first derived by Coopersmith (1968a). Unfortunately the exponents @ and 4 are not readily measured experimentally. The exponent £ is defined by means of the relation S~-mtt (T=T,). (4.15) From eqn (4.12) we have that f= 78-1, (4.16) so that the Coopersmith inequality may be written in terms of £ as pt (264+ 1/821 or pd + 24+128. (4.17) Many authors prefer to use £ instead of The mean field approximation (cf. Chapter 6) predicts that » = 0, 56 EXPONENT INEQUALITIES 4,§2 8 = 3, and ¢ = % (go that £ = 45 — 1 = 1); hence the Coopersmith inequality reduces to an equality for this model. Unfortunately, the new exponents » and are not known for most model systems, thereby precluding extensive tests of the Coopersmith inequality. 4.2. The Griffiths inequality The proof of the Griffiths inequality (Griffiths 1965a, 1965b), a’ + B(1 + 8) > 2, (4.18) illustrates the utility of the convexity relations described in Chapter 2. Te. T Fic. 4.1, Zoro-field magnetization as a function of temperature. M, is the value of the magnetization when T = 7. This diagram is useful in proving the Griffiths inequality (4.18). ‘We consider here a magnetic system as the proof is simpler than for a fluid. Choose an arbitrary temperature 7’ = 7’, which is less than T, and let M, denote the value of the zero-field magnetization at tem- perature 7’, (cf. Fig. 4.1), that is, M,(7,) is just the spontaneous magnetization which is assumed to vanish as (7, — 7',)® when T, > T,. Now since for all M < My, (2A/0@M),; = H = 0 (see eqn (2.46d)), it follows that A(T, M) = A(T,,0), M < U,(P,). (4.19) Similarly, the fourth Maxwell relation for magnetic systems, eqn (2.46d), implies that S(T, M)=S(7,,0), Ms M7). (4.20) 4,§2 THE GRIFFITHS INEQUALITY 87 Now define the functions A*(T, M) = {A(T, M) - A} + (T - T,)S, (4.21) and " S*(P, M) = S(T, M) - 8, (4.22) where A, = A(T, M) and S, = S(T’, M). With these definitions, oA* 8*(7T, M) = -(=) - 4,23) rm) = - (Fp) (4.23) The function A* is shown as a function of 7 and M in Fig. 4.2. Note from (4.21) that 4* has the same convexity properties as does A, since the addition of a term linear in 7’ cannot affect the second derivatives. We next utilize the convexity properties of A* to obtain our first inequality. Choose some temperature 7, < 7, and, as before, let M ¥ra, 4.2, General appearance of the function 4* (’, M); note that the point (7 = T., ‘M = 0) is a saddle point. After Griffiths (19658), 4M,(7) denote the saturation magnetization corresponding to T,. Then consider the function A*(7', M) for constant M = M, (cf. Fig. 4.3). Draw a tangent to this curve at the point 7 = 7’; the equation of the tangent is @A* ND) = AMT, My) + (P-~T(Fr) (28 THT; Then since A* is a concave function of 7 for constant M, all points on the A* curve must lie below (or on) the tangent curve (4.24); in particu- lar, A*(7', M,) for T = T, must lie below (4.24), that is, A*(Z',, M,) < f(T.) or, on using (4.23) and (4.24), ANT, My) < A*(T,, My) — (PT. — Ty)S8*(T1, My). (4.25) Because of the relations (4.19) and (4.20), eqn (4.25) is equivalent to A*(T., My) < A*(T,, 0) — (T. — Ts)8*(L4, 0). (4.26) 58 EXPONENT INEQUALITIES 4,92 Next observe that the concavity property of A* implies A*(7y, 0) < A*(T,, 0) since the tangent to A*(7’, M,) is horizontal at T = T,. Finally, we note from the definition (4.21) that 4*(7',, 0) = 0; hence (4.26) implies that A(T. My) < —(T. — Ti)S*(T1, 0). (4.27) Assuming that H(M,, 7.) ~ M% (recall eqn (3.11)), the relation H = (0A*/0M), implies A*(T,, My) ~ M3** ~ (1, — Ty)PO*?, (4.28) A | (a) MoM, (b) \ ' | ' 1 a aaeeeeneneen Te T Fre. 4.3. (a) A(T, M,) as a function of 7 for fixed M = My, Note that the slope is always negative, as is required for the entropy to be positive. (b) A*(7', My) as a function of 7 for M = My, The slope of this plot is given by —(8A*/8T yy, = S*(T, My) = S(T, M,) — So. Consider next the right-hand side of (4.27). Since 1’ is slowly varying near T,, we may write oC os ae c= (5m), ~ (T,— 7)", (4.29) or S*(T, 0) = S(Ty, 0) — S(T.,0) ~ —(T. — Tis*. (4.30) Hence the right-hand side of (4.27) varies as (7, — 7',)?~*’; comparing with (4.28) and using Lemma 3 (§ 3.5) yields BS +1) 22-2, > (4.31) which may be recognized as the Griffiths inequality (4.18). Additional complications arise for a fluid system because a plot of A as a function of V has a straight-line portion whose slope, given by — P, changes with temperature as 7’ —> 7’,, and the derivation of the Griffiths inequality for the fluid case is rather more difficult (Griffiths 19658). Note from Table 3.4 that (4.31) holds as an equality for the classical 4,92 THE GRIFFITHS INEQUALITY 59 theories. Observe, further, that for the two-dimensional Ising model, it provides the rigorous result § => 15, which is consistent with the value 8 = 15 + 0-08 provided by numerical approximation methods (Gaunt, Fisher, Sykes, and Essam 1964), The Griffiths inequality may be tested by comparison with experi- ments on a system such as helium-4 for which all three exponents «’, 8, and 8 have been measured. Specifically, Roach and Douglass (1967) and = 400 0-08 0-09 0-10 0-11 0-12 p (g-em-*) Fra. 4.4, Pressure-density-temperature surface of “He near the critical point, showing paths along which Roach and Douglass obtained their data. After Roach (1968). (Note that 1 atm = 760 torr.) Roach (1968) have attempted to map out the Pp surface of *He by making measurements along the paths shown in Fig. 4.4. They obtained B = 0-354 + 0-010 (4.32) and 38 <8 < 41. (4.38) In order to test the Griffiths inequality, Roach and Douglass calculated a = 0-017 + 0-008 by analysing specific heat data taken by Moldover 60 EXPONENT INEQUALITIES 4,§3 (1969). Using these values of the exponents, we obtain 1-660 < a’ + A(S + 1) < 1-881, that is {o’ + B(S + Wax = 1881 < 2. (4.34) Thus the experimental results as reported would appear to be inconsis- tent with the requirements of thermodynamics. However, in any situ- ation such as this there is always one easy way out, namely to conclude that the experimental measurements did not extend in temperature sufficiently close to 7, that the limiting behaviour could be ascer- tained. If this is the case, thermodynamics has told us something very useful indeed: the experiment should be repeated. In the particular case of the Roach—Douglass helium data, several other possible interpre- tations of the discrepancy (4.34) have been suggested (Moldover 1969, Coopersmith 19686). In particular, Moldover (1969) has recently pub- lished a detailed account of his own specific heat measurements and he concludes that «’ ~ 0-15, whereupon {a’ + B(S + Imax = 2.01 > 2. (4.35) 4.3. More inequalities In the preceding sections we discussed in detail three of the more easily proved inequalities that relate various of the critical-point expo- nents. These inequalities are of particular utility for reasons indicated above and also because each has as a counterpart an equality predicted by the non-rigorous scaling-law theory of exponents. There exist several more useful inequalities which we shall collect in this section. Since most of these remaining inequalities cannot be demonstrated easily, we shall refer the interested reader to the original work for details of the proofs. 4.3.1, The remaining Griffiths inequalities What is generally termed the Griffiths inequality, (4.18), is actually only one of many inequalities proved by Griffiths. For example, two other inequalities Griffiths (19655, 1968) has proved are y’ = BS — 1) (4.36) and. (8 + 1) = (2 - «(3 — 1). (4.37) The remaining Griffiths inequalities are listed in Table 4.1 (Nos. 3-9). Most of these require for their ‘proofs’ certain plausible assumptions 4,93 MORE INEQUALITIES 61 TABLE 4.1 Rigorous inequalities among critical-point expo- nents for magnetic systems. Assumptions are listed in Table 4.2; exponents not defined in Table 4.2 are defined in Table 3.2. Adapted from Griffiths (19658). Inequality Assumptions 1. w+ Bt y a2 - 2. a + AS +1)>2 = 3. (2— ae) +12 (1 — «8 B 4, vB + 1) > (2 — @\(8 — 1) B,C, 5 of = BB — 1) AG 8+ =5 = Tt 8, = min (8, f + 0] B &t (2-ao> 3,41 B oF 7s + 1) = (2 — @)(8, — 1) B,Cy 10. ety —1/s=1 — I. a8 Vo 4 1) =2— DF 12. ay'| = @) > dy'/2p + 7) 22-4 DF 13, Q-mey DF 14, da’ > 2-2 DF 15. 2dpd/(S + 1 + #8) = aol +e)e2— ne DF 16. other 17. * > 22 « other TIf «> a, inoqualitios 3, 6, 7, and 8 lead to tho result 8, = 8 (Griffiths 19656), that have not been established yet for general systems; these auxiliary assumptions are given in Table 4.2. For example, (4.37) is obtained by making essentially the following three assumptions: (QM/eT)y <0 for H> 0, (4.38) (@M/0H?)p < 0 for H > 0, (4.39) and a> a’, (4.40) Assumption (4.38) has recently been proved for Ising ferromagnets, and is plausible for other systems (cf. Fig. 4.5(a)). Assumption (4.39) states that the M-H isotherms are concave and, although it has not been proved either generally or for any particular systems, it is certainly plausible (of. Fig. 4.5(b)). Assumption (4.40) states that the specific heat singularity is not stronger when 7’, is approached from below than from above. It is valid for the Ising model if d = 2 (and probably also valid if d = 8). Thus the inequality (4.37) may well hold for some systems. One must note the difference in proving a general thermodynamic 6-17, 62 EXPONENT INEQUALITIES 4,§3 TaBLe 4.2 Assumptions referred to in Table 4.1 A. (@H/@T)x_ = —(88/2M), = 0 for M > 0. Cf. Fig. 4.5(a). . (8S/@M)y < 0 for M > 0; T = T.. (20) for Fs Cf. Fig. 4.5(b) (20) for T > T,. Cf. Fig. 4.5(b) . (@H/@M)r is non-decreasing function of . (@H|@M)z is a non-decroasing function o} . Positivity of the correlation functions—for all temperatures 7’ and any non-negative field (i.e. H > 0), Py > 0, Pa(r) —TP> 0, and P(rsrayrs) — Ta(rs)Pa(ts — 72) = 0, where here I’, denotes the j-spin correlation function (and hence I’; is proportional to the spontaneous magnetization). E. Monotonicity with magnetic field—for any fixed valuo of 7, T,(7, H), V'a(7, H), and T,(Z, H) are monotonic non-decreasing functions of H for H->. 0 (in particular, for T > T,, 1, = 0 provided H = 0). F, Monotonicity with temperature—for any fixed (non-negative) value of H, I'y(7, H), T,(Z, H), and T',(7, H) are monotonic non-incroasing functions of 7’. Exponent definitions: (a) So — S ~ Mit (bo) T — 1, ~ Me (c) H~ Mt (a) Cy ~ H (c) SH) ~ —H* @) Tap ~ ro 3+ poon 8 = 98-1) 8 8 T T T af) (a) H=0* Fre. 4.5, (a) General form of the magnetization curves required by assumption (4.38), that M be a non-increasing function of 7 for fixed non-negative H. (b) Assumption (4.39) requires that the M-H isotherms be concave. relation, such as (4.1) and (4.18), and one valid for particular systems for which auxiliary assumptions such as eqns (4.38)-(4.40) must be satisfied. Notice that if we further assume that 2 — « — y > 0 (as is generally true), then (4.37) implies 2-a+y ——— Al J—a = 8, (4.41) which is a useful upper bound on the critical isotherm exponent 8. 4,53 MORE INEQUALITIES 63 For example, for the three-dimensional classical Heisenberg model, one set of numerical estimates is (« = —0-10, y = 1-40) and another set is (a = —75, y = 44). Equation (4.41) predicts § < 5 for both sets of data (Stanley 19716). We can combine (4.41) and (4.31) to obtain the relation 2-at+y 2-a'-p doa-y 262 > (4.42) or equivalently, A2-—aty) 2 (2a —pl@—a—y). (4.48) Note that (4.42) and (4.43) contain both high-temperature and low- temperature exponents. Note also from Table 3.4 that both (4.42) and (4.48) are satisfied as equalities for the classical theories. For the d = 2 Ising model «’ = « = 0; also assumption (4.38) has been proved rigorously. Hence the only assumptions that are still needed to obtain the value 5 = 15 from (4.42) are (4.39) and y = 4, which is almost rigorous. Note that the values « = (4.41), while the values of «’ = 74, = 4,8 = #5, and y = § are incon- sistent with (4.43). These values were at one time the best numerical estimates for the d = 3 Ising model (Baker and Gaunt 1967). Note also that we can restore the inequality (4.41) by setting 8 = 5 and the inequality (4.43) by setting «’ = 3. Of course there are many other sets of values that are capable of satisfying (4.41) and (4.43) but current estimates support decreasing 6 to 5 and increasing a’ to 4 (Gaunt 1967, Gaunt and Domb 1968, Garelick and Essam 1968). With these values the inequalities (4.31), (4.41), and (4.43) become equalities; also in this case both the Rushbrooke inequality (4.2) and the inequality (4.36) predict that y’ > §. It is therefore still possible that y’ = 24, as is favoured by some recent work (Guttmann and Thompson 1969). = $,and 8 = 5-2 are inconsistent with 4.3.2. The Buckingham-Gunton inequalities We begin by recalling that the critical exponent 7, as defined in eqn (3.16), characterizes the decay with distance r of the pair correlation function I',(r). Buckingham and Gunton (1969) and Fisher (1969) have proved the following two inequalities relating » and the dimension- ality d of the system (see also Stell 1968a): 8-1 @ 8+1 22-7 (4.44) 64 EXPONENT INEQUALITIES 4,§3 and aoe 2B + y 2 A corollary to (4.45) is easily obtained from the Rushbrooke inequality, 2a < B+y, —4 (4.45) dy’ po em (4.46) The proofs of (4.44) and (4.45) depend on three further assumptions concerning the behaviour of the correlation functions of the system. D. Positivity of the correlation functions—for all temperatures 7’ and any non-negative field (i.e. H > 0), T, 20, (4.47) T(r) — T? = 0, (4.48) and Ta(ra, ta, 13) — Ta(ra)Valrs — r2) = 0, (4.49) where I’, denotes the j-spin correlation function (and hence I’, is pro- portional to the spontaneous magnetization.) E. Monotonicity with magnetic field—for any fixed value of T, T,(7, H), 1(7,H), and [,(7,H) are monotonic non-decreasing functions of H for H = 0 (in particular, for 7 > T,, [', = 0 provided H = 0). F. Monotonicity with temperature—for any fixed (non-negative) value of H, T,(1, H), .(T, H) and I',(7', H) are monotonic non-increasing functions of 7. Assumptions D-F have been proved rigorously for the Ising model with arbitrary ferromagnetic interactions of arbitrary range and with arbitrary spin quantum number § (Kelly and Sherman 1968, Griffiths 1967a, 1968). For other magnetic models and for real magnets, assump- tions D-F are at least intuitively plausible. For fluid systems, on the other hand, assumption D (positivity) is certainly not valid; rather, T(r) oscillates in sign. Moreover, it is not yet known whether or not the spin correlation function develops a monotonically-decaying tail as TT, (Fisher 1969). Hence the Buckingham-Gunton inequalities have not been proved for fluid systems—but neither have they been disproved! The inequality (4.44) is especially useful since it provides for 7 @ lower bound which depends only on 5. For example, if § = 5 for a three-dimensional system (as is true for the spherical model and as is 4,93 MORE INEQUALITIES 65 extremely likely for both the Ising and Heisenberg models), then (4.44) requires that 7 = 0. Moreover, many experimental measurements of 8 yield a value smaller than 5, whence (4.44) implies that » must be larger than zero. Note, however, that if 5 = 4-3 for a system (as for CrBr,—cf. Table 3.4), then we cannot conclude that 7 must be larger than 0-132 unless the assumptions D—F could be verified for the system. We observe that the inequalities (4.44)-(4.46) differ from the in- equalities we have considered earlier in this chapter in that they are not satisfied by the van der Waals and mean field theories. They are, however, satisfied (as equalities!) by the two-dimensional Ising model, providing we choose § = 15 and y’ = } (Fisher 1964). 43.3. The Fisher inequalities Fisher (1969) has recently extended and refined the arguments for the Buckingham-Gunton inequalities. He has also proved, subject to the same assumptions (D-F above), the following inequalities: @-nyey (4.50) do’ = 2— nz, (4.51) and a ® 2 2-o (4.52) Tel ps pte where the exponent 7, characterizes the decay of the energy—energy correlation function at 7 = T,; v, 7, p, #, and ¢ are defined, respec- tively, in eqns (3.15), (3.16), (4.11), (4.12), and 4.16. We can easily derive the following corollary to (4.52), ip >= 2- (4.53) from the Coopersmith equality (4.14), (p + ¥) => 4(1 + 1/8 + 9). As we can see from the exponents presented in Table 3.4 the inequality (4.50) is satisfied as an equality for (a) the classical theories, (b) the d = 2 Ising model, and (c) the d = 3 spherical model. Unlike the other inequalities discussed in this chapter, (4.51) and (4.52) are not satisfied as equalities when the scaling hypothesis (cf. Chapter 11) is made. 4.3.4. The Josephson inequalities Josephson (1967) has proved the inequalities dy >2—a! (4.54) and dv >2-a, (4.55) 66 EXPONENT INEQUALITIES 4,§3 subject to certain additional assumptions which we shall not discuss here. Observe that if we could prove the analogue of (4.50) for 7 < T, (2 — )’ = y’, then combination with (4.46) would produce the first Josephson inequality (4.54). The Josephson inequalities, (4.54) and (4.55), fail for the classical theories for d = 3 (which predict the same values of the critical exponents regardless of the lattice dimensionality d). Observe from Table 3.4 that the Ising model satisfies the Josephson inequalities as equalities for d = 2, and as inequalities for d = 3. Similarly, the spherical model satisfies (4.55) as an equality. Suggested further reading Fisher (1964). Griffiths (19655). Fisher (1967). Fisher (1969). Griffiths, Hurst, and Sherman (1970). PART III CLASSICAL THEORIES OF COOPERATIVE PHENOMENA 5 THE VAN DER WAALS THEORY OF LIQUID-GAS PHASE TRANSITIONS VAN DER WAALs presented in his doctoral thesis ‘On the continuity of the liquid and gaseous states’ (1873) an approximate equation of state for a fluid system. His work provided one of the earliest qualitative explanations of critical phenomena, although we shall see that many quantitative predictions are not borne out by detailed experimental work. 5.1. Heuristic derivation of the van der Waals equation of state The simplest theory of a fluid system corresponds to assuming that the particles of the system interact so weakly with one another that these interactions may be neglected altogether. Such a treatment results in the ideal gas law, PV = NkT = »?, (5.1) where WN is the number of molecules and « = N/N, is the number of moles in the system, N, is Avogadro’s number, k is the Boltzmann constant, and @ = kN, is the ideal gas constant. For a gas at very low density and/or high temperatures eqn (5.1) is fairly well obeyed. Therefore it might seem reasonable to attempt to describe an interacting gas at normal densities by means of simple corrections to the non-interacting result (5.1). To this end, van der Waals attempted to take into account (i) the non-zero size of the molecules and (ii) the attractive force between the molecules. 68 THE VAN DER WAALS THEORY 5, §1 (i) The ideal gas law assumes that the volume of the molecules may be neglected; for example, the equation P = »#T/V means that as we increase the pressure at fixed temperature, the volume decreases without limit. Actually there is a limit if we think of the molecules as being like rigid spheres; i.e. there is a minimum volume Vmin = N x volume of a rigid sphere molecule. Of course real molecules are not rigid spheres, yet it may be a reasonable approximation to replace V in eqn (5.1) by V — Vann, where Vinin is regarded as a phenomenological parameter. Now in order that V — Vinin be an extensive quantity, V,,.,, must be linear in NV. We call the pro- portionality constant ¢/N, and therefore make the substitution VV — (¢/N,)N into eqn (5.1), with the result P = nRT|(V — nb). (5.2) Equation (5.2) is sometimes called the Clausius equation of state. (ii) The attractive force between the molecules of a gas clearly leads to a decrease in the pressure that the gas would otherwise exert on the walls of the container. This decrease AP has contributions from the decrease in momentum per molecule, and from the decrease in the number hitting the walls. If we assume that both contributions are proportional to the density of molecules, then AP oc (p/m)?. We call the proportionality constant 2/N% and therefore subtract the term (2/N3) (N/V)? from the value of the pressure given by eqn (5.2) obtaining P = nAT KV — nt) — an®[V?. (5.3) Equation (5.3) is the van der Waals equation of state and it is usually written in the form (P + an3/V?) (V — xd) = xT, > (5.4) or else in the form (P + af¥2)(V - 4) = AT, (5.5) where V = V/ is the volume per mole or specific volume. Clearly the van der Waals equation of state reduces to the non-interacting result (5.1) in the limit of low density, i.e. when both a and ¢ are negli- gible compared to 7. It is important to stress that « and ¢ are pheno- monological parameters characteristic of the fluid. Values of 2 and ¢ that provide a fairly satisfactory fit of (5.4) to the thermodynamic data (at least far from 7’) are listed in Table 5.1 for a few selected substances, 5,§2 ISOTHERMS AND THE MAXWELL CONSTRUCTION 69 Tasiz 5.1 The van der Waals constants a and é for selected fluid systems, together with the ratio Z, = P.V.[#T,. The van der Waals constants appear in the equation of state (5.4) ; because of the dimensions of a and é used here, we should measure P in atm, V in litres, and take B ~ 0-08206 litre-atm] mole-degree. Note that a is a measure of the attractive force between particles, and ¢ measures the non-zero volume of the molecules. Since both aand é vary with temperature, the values shown here are only approzi- mate; they are obtained from the CRC Handbook, 45th edn. The systems are listed in order of increasing Z,. The van der Waals theory predicts Z,, = } = 0-375, while in the non-interacting limit Z = 1. Fluid a(Patmmol-?) é(Imol-!) Zp = PpVo@T > Water, H,O 5-464 0-230 Sulphur dioxide, SO. e714 0-269 Ethylene, C,H, 4471 0-270 Acetylene, CzHz 4390 0.274 Carbon dioxide, CO, 3-592 0-275 Ethane, C,H, 5489 0-285 Xenon, Xe 4-194 0-288 Methane, CH, 2-253 0-290 Nitrogen, Nz 1-390 0-201 Argon, Ar 1-345 0-291 Oxygen, O2 1-370 0-202 Carbon monoxide, CO 1-485 0-204 Hydrogen, H, 0.2444 0-304 Helium, *He 0-03412 0-308 van der Waals theory - _ 0375 5.2. van der Waals isotherms and the Maxwell construction Note that the van der Waals equation of state (5.4) may be written, for 1 mole (# = 1), a8 a ab V8 - (6+ AT/P)V? + 5V- 5 = 9, (5.6) so that for a given value of P and T there will correspond in general three values of V instead of only one. For small T all three roots of (5.6) will be real (see Figs. 5.1, 5.2(a)). As we increase the temperature these roots move together and for 7 > 7’, one pair of roots becomes complex; at very large 7, eqn (5.6) becomes V* — (#7'/P)V? = 0, which is just the ideal gas law, eqn (5.1). For all subcritical (7 < 7.) isotherms there will be a region in 70 THE VAN DER WAALS THEORY 5, §2 which the slope @V/@P is positive corresponding to the unphysical occurrence of a negative isothermal compressibility Ky = —V~1 (a0 /@P)q. This dramatic breakdown of the van der Waals theory for T < T, was recognized from the outset, and less than two years had passed yl A —\ = = P (a) t — Ss SS ve LX TA AY 4 A ai =|] — tea NX OB rX8 : | i ——"* f = = Fre.5.1, The PVT surface for the van der Waals theory. Any point on this surface satisfies the van dor Waals oquation of stato (5.4) and hence is predicted to correspond to an. equilibrium state of the system. The reader should determine that portion of the surface that corresponds to unstable states. Aftor Kubo (1968). Fra, 5.2, (a) A single subcritical (or 7 < 7.) isotherm, as given by (6.4). (b) The Max- well construction restores the convexity of the Helmholtz potential A(T, V) as a function of V. After Huang (1963). == 1 to. Se from the date of van der Waals’ thesis when Maxwell proposed an ad hoc remedy which is commonly called the Maxwell equal-area con- struction. Following Huang (1963), we construct in Fig. 5.2(b) the 5, §3 THE VAN DER WAALS CRITICAL POINT m Helmholtz potential A(7'5, V) for a fixed temperature 7) < 7, and we find that the region in which @P/@V > 0 corresponds to a region where 474/dV? < 0 (cf. eqn (2.36)). Now A(7'o, V) must be a convex function of V for all V in order that the system be stable. The Maxwell construction consists in replacing the Helmholtz potential in the entire region V; < V < V2 (which is somewhat larger than the unstable region referred to above) by the dashed line shown in Fig. 5.2(b). This modification of the Helmholtz potential corresponds to a hori- zontal straight line connecting points 1 and 2 of the P-V isotherm of Fig. 5.2(a). It is easy to show that the horizontal line is at just such a value of the pressure that the areas labelled A and B of Fig. 5.2 (a) are equal, and for this reason the Maxwell construction is sometimes called the equal-area construction. An equivalent formulation of the Maxwell construction focusses instead on the Gibbs potential—whose dependence upon pressure the reader can easily construct from Fig. 5.2 using the methods developed in Fig. 2.4, One finds that the Gibbs potential becomes a three-valued function in the neighbourhood of the phase transition, and that the Maxwell construction is equiva- lent to deleting all but the lowest branch of this function. 5.3. The van der Waals critical point: P,, V., and T, The P-V isotherm for T = 7, is characterized by a horizontal tangent and an inflection point when P = P, and V = V,. Hence we may obtain the three parameters P,, V,, and 7’, (which define the critical point) by solving simultaneously the three equations: (i) (@P/V) rar, = 0, (ii) (@P/2V%)par, = 0, and (iii) the van der Waals equation (5.4), where in all three equations we set P = P,, V = V,, and T = 7,. An alternative procedure is to observe that we may equate coefficients of identical powers of V in the two equations: (V -— V.)? = V8 — 3V?V, + 3V V2 - V3 =0 (5.7) and Ve — (6+ &T./P.) V? + (a/P,.)V — at/P, 0 (5.8) Equation (5.8) merely states that for P = P, and T = T,, all three 72 THE VAN DER WAALS THEORY 5§4 roots of eqn (5.7) are the same and in fact are equal to the critical volume V,. We thereby obtain the three equations: —3V, = —(¢+ &T./P.), (5.9) +302 = a[Py (6.10) and -V2 = —at|P,. (6.11) Equations (5.10) and (5.11) may be combined to yield P, = a/21@ (5.12) and V, = 3¢. (5.18) Finally, substitution of (5.12) and (5.13) into (5.9) results in RT, = 80/276. (5.14) Table 3.5 lists values of P,, p,, and 7’, for a few representative fluids. 5.4. The law of corresponding states We begin by observing that eqns (5.12)-(5.14) provide three relations among the two parameters @ and ¢ which characterize a given fluid. Hence there must be one relation involving P,, V,, and T, which is independent of the parameters a and ¢; clearly Z, = P.V/AT, = (a/276?) (34) (276/82) = § = 0375 (5.15) Typical values of Z, are listed in the last column of Table 5.1. Note that for an ideal gas (which has no critical point), Z= PV/AT =1 (6.16) for all values of P, V, and 7 so that the van der Waals theory has achieved some improvement over the non-interacting result, although from Table 5.1 we see that Z, < § for real fluids. Suppose we multiply both sides of the van der Waals equation by (276/a) and then use eqns (5.12)-(5.14) to eliminate the parameters aand é ;we thereby obtain (P + 3/74) (a — 1) = 8f, > (6.17) where P=P/P,, V=ViV,, and P= T/T. (5.18) This new form of the van der Waals equation, (5.17), means that if we 5, §4 THE LAW OF CORRESPONDING STATES 73 measure pressure, volume, and temperature, respectively, in units of P,, V., and T,, then the equation of state is the same for all substances. Thus any two fluids with the same values of P, V, and 7 may be said to be in corresponding states; eqn (5.17) is generally called the law of corresponding states. One might expect that the law of corresponding states is obeyed by real gases only to the extent that the van der Waals equation (5.4) is obeyed. However, experimentally we find remarkable confirmation of the (more general) notion that materials should behave similarly pro- x Methane Oar ® Iso-pentane i oFthylene 0 n-Heptane a Ethane Nitrogen | O3F @ Propane ® Carbon dioxide o n-Butane ¢ Water | 2p i | o1 L 1 1 — 1 0005 10 15 20 25 30 35 40 45 50 55 60 65 70 Reduced pressure, > Fra. 5.3, Dependence of the compressibility ratio Z = PV/RT upon reduced pressure P for different reduced temperatures 7’. The fact that the data for a wide variety of fluids fall on identical curves supports the law of corresponding states. After Su (1946). vided that P, V, and 7 are measured in units of P,, V,, and 7. For example, in Fig. 5.3 we display measurements of the compressibility ratio Z = PV/&T for a variety of fluids. Even though these materials do not obey the van der Waals equation near 7’, (e.g. the value of Z, is not 3), the data for all materials nevertheless lie on the same curve for a given reduced temperature 1’. See also Fig 1.8. We shall discuss behaviour analogous to the law of corresponding states in Chapter 11. 4 THE VAN DER WAALS THEORY 5,85 5.5. Critical-point exponents for the van der Waals theory One of the simpler methods of obtaining critical-point exponents for the classical van der Waals theory is to begin with eqn (5.17), the van der Waals equation in terms of reduced variables P, 7, and 7. We then expand each variable about the critical point; thus, on de- fining p=P-1=(P- P,P, (5.19) v=V-1=(V-V)Vo (5.20) e=f-15(7-T,/T. (5.21) we find that eqn (5.17) becomes {(1 + p) + 3(1 + v)-73} (1 + v) — I= 8(L + €). (5.22) Multiplying both sides by (1 + v)®, we get {4 + 2v + 0? + pl + 2v + v7} {2 + Bo} = B(1 + €)(1 + Qo + 0), (5.23) and on expanding both sides and combining terms, we get 2p(l + 7/2 + 4v? + 3v9/2) = — 303 + 8e(1 + 20 + 02). (5.24) To obtain the critical isotherm exponent 5 and the coefficient 2 defined in eqn (3.10), we set « = Oin (5.24), p = —$3 (1 + 70/2 +--+)-t = —$0%(1 — 70/2 +--+). (5.25) The leading term on the right-hand side is cubic; hence § = 3. To obtain the coefficient 2, we observe that =o (5.26) where we have used the fact that P? = #7',/V, =8 P, from eqn (5.15). ‘Thus from (3.10) and (5.26) we see that 9 =—— 27, 9=%G (6.27) Consider next the behaviour of K, as 7’—> Tt along the critical isochore (v = 0). From (2.17), (5.19), and (5.20), 5, §5 CRITICAL-POINT EXPONENTS 15 where the last equality in (5.28) follows from (5.24) On using (Kz)~* = P? = $P., we find Ky _8/1\_ 41 R73 (52) “oC (6.29) Hence y = 1 and the coefficient @ defined in eqn (3.8) has the value §. We can show 1n a similar fashion that when we approach the critical temperature from above, we obtain the values y’ = 1 and@’ = §. Thus the inverse compressibility is linear in e, but it rises twice as fast below T,, as above T, (cf. Fig. 5.4). This result can be remembered easily by means of the following mnemonic device: when the system is below (a) (b) I | i L qT. T Frc. 5.4. (a) Inverse isothermal compressibility predicted by the van der Waals theory. Note that »’ = 7 = 1, yot @(=8) < €(=4). (b) A PV projection for throo isotherms, 1 = T,,and T = T, + To. The circlos indicate the regions where the slope is propor- tional to tho inverse compressibility. That 6’ < @ is often said to correspond to the fact that the lowor circle is further from the critical point than tho upper circle. the critical point, it is removed both in temperature and in density, whereas when it is above 7',, we are on the critical isochore and hence are removed only in temperature. Similarly, we can calculate the shape of the coexistence curve and we find that the exponent f has the value 8 = 4, while the coefficient defined in eqn (3.6) has the value # = 2. Finally, the specific heat predicted by the van der Waals theory is GNK (1 — $e +--+) (7 < 7.) tb IP > 7 where Cy = $N/ is the constant volume specific heat predicted for the non-interacting (high-temperature) limit or ideal gas. Thus a’ = « = 0. Note that the slope of Cy vs. T is finite as T’ —> 7’, from below, so that a! = Oas well. ACy = Cy — CF = (6.30) 76 THE VAN DER WAALS THEORY 5, §6 It is clear from Table 3.4 and Fig. 3.3 that the predictions of this classical theory disagree with most experimental results on fluid and magnetic systems near the critical point. On the other hand, we see that the d = 3 Ising and Heisenberg models agree rather better with experiment. It is by now generally believed that the classical exponent values are to be associated with the assumption, inherent in all of the classical theories, that the interparticle forces are of extremely long range. In fact, the essential results of the van der Waals theory can be derived by making the approximation that each particle moves in a mean field due to all the other particles. Correspondingly, in the next chapter we show that the molecular field (or mean field) theory of magnetism can be derived by making the assumption that each spin interacts equally with all the other spins in the system. Presumably these rather drastic assumptions are unrealistic for real fluid and mag- netic systems. 5.6. The van der Waals equation of state as a mean field theory In this section we show that the van der Waals equation of state also results if we approximate the interactions among the particles comprising the fluid by an effective potential of the form shown in Fig 5.5(a). The technique of approximating the effects of molecular interactions by means of the assumption that a particle moves in such U(r) vm (a) (b) Fie. 5.5. (a) The effective potential U(r) assumed in the Reif-derivation of the van dor Waals equation of state consists of a hard core of radius ry and a long-range tail for which the potential is approximately constant. (b) The effective potential U(r) = 00 for r < 79 and U(r) = —xe-™" for r > ro. A ayatom of particles interacting with this potential behaves, in the limit x > 0, aa a van der Waals gas. 5, §6 AND MEAN FIELD THEORY 7 a mean field due to all the other particles is a trick that we shall use again in the following chapter to develop the so-called mean field theory of magnetism. Our present mean field theory of fluids follows Reif (1965). Assume, then, that the particles are independent, yet that they move about in an effective potential of the form co r 0, the potential becomes both infinite in range and in- finitely weak. Kac et al. have shown that in this limit, the model becomes essentially the same as the van der Waals theory, with one noteworthy improvement—there are no unstable regions (i.e., the ‘Maxwell construction’ of § 5.2 is no longer necessary). The interested reader is referred to this important work for the details of the argument. Suggested further reading Huang (1963). Kac, Uhlenbeck, and Hemmer (1963). Baker (1963b). Reif (1965). Widom (1967). Kittel (1969). 6 THE MEAN FIELD THEORY OF MAGNETIC PHASE TRANSITIONS In 1907—two years after the appearance of Langevin’s theory of para- magnetism—Pierre Weiss proposed a phenomenological theory of ferromagnetism in which he assumed that the spins interact with one another through a molecular field proportional to the average magnetiz- ation. It was only some years later, after the Heisenberg exchange inter- action had been proposed, that the molecular field was given an interpretation in terms of pair-wise exchange interactions J,, between spins S, and S, situated on sites i andj in the lattice. Fairly recently this mean field theory has received yet another interpretation. It has been shown that if we consider a model system in which every magnetic moment interacts with every other moment—with an equal strength— then the properties of this model are identical to those of the mean field theory. In this chapter we shall consider each of these three inter- pretations of the mean field theory in turn: (i) the Weiss molecular field, (ii) the molecular field theory as an approximation to the Heisenberg model, and (iii) the solution of an infinite interaction range model. We shall see that the critical-point exponents predicted by the molecular field theory are identical to those of the van der Waals theory. This should come as little surprise, as we saw in § 5.6 that the van der Waals theory could be interpreted as an infinite interaction range model also. Most systems in nature have relatively strong, short- range interactions, and we shall see in Chapter 9 that the nature of the cooperative phenomena observed depends crucially on the fashion in which these interactions ‘propagate order’ from one particle to another. Hence we will come to appreciate why it is that the mean field theory provides an inadequate description of phenomena in the critical region. We begin our discussion with the derivation of the partition function and the thermodynamic properties of a system for the limit in which the 80 MEAN FIELD THEORY 6, §1 constituent magnetic moments interact so weakly with one-another that these interactions may be neglected altogether. We shall find that the techniques illustrated in this calculation for the ‘perfect para- magnet’ prove useful in deriving the predictions of the mean field theory. 6.1, The non-interacting magnetic system The Hamiltonian for a system of N non-interacting magnetic mo- ments in an external magnetic field H is N = -9e > S,-H, (6.1) a where the product S, - H may assume the values »S— 1,8) (6.2) S,-H = mH, (m, = -S, -S +1,..., and where we have used the relation —y. = gu,S between the magnetic moment yp. and the spin S. In eqn (6.1) g is the Landé factor and py, = efi/2me is the Bohr magneton; since the combination gyg occurs in many of the formulas to be presented, we shall introduce the notation B= 9s. (6.3) The partition function or Zustandsumme is obtained by weighting each state by the appropriate Boltzmann factor from (6.1) and then summing over all (28 + 1)" states of the system. Thus we have Z= . Pa 2 exp (© 5m) (6.4) where ° - x = pH|kT. (6.5) The summations in eqn (6.4) are particularly elementary for the case of a two-level system (S = 4), and we have Z= Ulf $ exp (em) tat nde N = [ [2 cosh (42) t-1 = 2" cosh” ($2). (6.8) 6,§1 THE NON-INTERACTING MAGNETIC SYSTEM 81 For general S, the solution becomes Z = [exp (—a8) [1 — exp {(28 + 1)z}]/{1 — exp (x)}]" sinh {(S + })2}]* - (Fae) (6-7) Thus it is possible to obtain a closed form expression for the partition function in the non-interacting limit. We say that we can ‘solve’ the paramagnetic system in the sense that since we can obtain the partition function, we can also obtain a great many thermodynamic functions that are directly related to the par- tition function. For example, the Gibbs potential G(7’, H) is, directly from eqn (6.7) (Wannier 1966), QT, H) = -kTnZ _ sinh {(S + $)2}], = -NkT In oe” oh fe (6.8) The magnetization is related to the Gibbs potential by eqn (2.45c), whence a@ a U(T, H) = - (3), = Mer Sn = My Bs (Sz), (6.9) where M, = M(T = 0, H = 0) = NSp = NSgup (6.10) is the maximum value of the magnetization, and 28 +1 1 7 28 +1 1 — sy eoth (ss) (6.11) Bay) = St coth ( t 28 is called the Brillouin function. This function, which relates the magnet- ization of the system to the applied field, is plotted in Fig. 6.1. Note that for S = } (that is, the single electronic spin value), B, (42) = 2 coth (x) — coth (2/2) _ coth? (x/2) + 1 ~ ~~ eoth (x/2) = tanh (x/2). (6.12) — coth (2/2) 82 MEAN FIELD THEORY 6, §2 Fic. 6.1. Dependence of the reduced magnetization « = M/My = M(T, H)/M(0, 0) upon jHS/k7 for a magnetic system in the non-interacting limit, as given by eqn (6.9). The curves are those of the Brillouin function Bs(Sz), which reduces to the hyperbolic tangent and to the Langevin function in the limits S = } and S = co respectively. 6.2. The assumption of an effective molecular field We note from Fig. 6.1 that M = 0 when H = 0; we say that the spontaneous magnetization M(7, H = 0) of the paramagnet is zero. This prediction is not supported by observation—in fact, for all ferro- magnetic materials there is some temperature below which there exists a spontaneous magnetization. The basic assumption of the mean field theory is that the inter- actions among the spins give rise to a magnetic field H,, in addition to the external field H. If one assumes that H,, is proportional to the magnetization, H,, = \M(T', H), then the effective field seen by each spin is Aggy = H + AM(T, H). (6.13) The parameter A in (6.13) is called the molecular field parameter. The derivation of the partition function proceeds as before, and all the above expressions remain valid providing we replace the magnetic field 1 (which enters into the parameter x of eqn (6.5)) by the effective field of eqn (6.13). In particular, eqn (6.9) becomes M = MoBz {RiS (H + AM)}, (6.14) 6, §2 EFFECTIVE MOLECULAR FIELD 83 where M denotes M(T, H), My = M(T = 0, H = 0) as in (6.10), and B = 1k. For H = 0, eqn (6.14) reduces to M = My By(ppSAM). > (6.15) Equation (6.15) is an implicit equation in the sense that M appears in the argument of the Brillouin function as well as on the left-hand side of the equation. We consider therefore a graphical solution, and refer the reader to Fig. 6.2 which shows both sides of (6.15) plotted as functions of M. Equation (6.15) possesses the trivial solution M = 0 for all values of 7’, but there exists a second solution with M ¥ 0, providing Fra. 6.2. Plotted as functions of M aro the left-hand and the right-hand sides of eqn (6.15). The solution M = 0 exists for all 7, but the solution M # 0 exists only for T sufficiently small that the initial slope of the Brillouin function is larger than 1. the initial slope of the right-hand side of (6.15) is larger than the initial slope of the left-hand side. To study the initial slope, we consider the small-argument expansion of the Brillouin function, ae S +1282 +2941 BSy) = Sg-¥ - g- oge P t (6.18) Therefore the initial slope of a graph of the right-hand of (6.15) as a function of M is /S + 1) aSA a OH a ) er cA (6.17) where we have used eqn (8.10) to eliminate M, and where Np28(8 +1) le 3k (6.18) 84 MEAN FIELD THEORY 6,§3 is called the Curie constant. We see from (6.17) that non-trivial solutions exist for 7’ < AC, whence we have the result that the critical temperature for the molecular field theory is given by T, = XC. (6.19) Thus 7’, is proportional to the molecular field parameter A and ap- proaches zero when A approaches zero. This should not be surprising, since for A = 0 we recover the paramagnetic or non-interacting result, T, = 0. 6.3. Critical-point exponents for the mean field theory We begin by considering the case S = }, so that the equation of state in the mean field theory is, from (6.14) and (6.12), M = M, tanh (387 (H + AM)}. (6.20) If we introduce the reduced variables M _ M(T,H) °= i, * MO, 0) em) and 7 = T/T, then (6.20) becomes re « = tenh (se + oi?) > (6.22) In order to facilitate the calculation of the critical-point exponents, it is convenient to write the equation of state (6.22) in the form pH ) o — tanh (0/7) = tanh (22) _ _¢ — tanh (0/7 : fala (& 1 — otenh (o/T) (6.28) where we have made use of the trigonometric identity tanh « + tanhy tanh (@ + 9) = “Ganh 2) (anh g) (6.24) Near the critical point (H = 0, M = 0, T’ = T,) the arguments of all the hyperbolic tangent functions in eqn (6.23) are small, and we can therefore use the expansion tanha =a - fae + a+... (6.25) to obtain h = o(1 — 1/7) + o{1/(37%) + (1 — 1/7)/T} + O(o8). (6.26) 6,§3 CRITICAL-POINT EXPONENTS 85 Most of the critical-point exponents for S = } are obtainable from (6.26). 6.3.1. Magnetization exponent B From eqn (6.23) we see immediately that in zero field, h = 0, and for T < T, eqn (6.26) becomes gu TIT —\ ~ 3 13) _ THOT) + (TIT — TIP) * (oan T\?7,-T 29(7) “7 Thus the square of the zero-field magnetization vanishes linearly with T, — T, so we have B = 3, just as in the van der Waals theory of a fluid system. Notice, however, that the coefficient # defined in eqn (3.7) has the value */3 here, in contrast to the van der Waals prediction B=2 6.3.2. Critical isotherm exponent 5 To obtain the curvature of the M-H isotherm at the critical temper- ature, we set 7 = 1 in (6.26) and consider both H and M to be small. Thus we have MopH + OP) = 09/3 + O(08), (6.28) where 8, = 1/kT,. Hence & = 3, just as in the van der Waals theory. However the coefficient again differs, and we find that D = % (cf. eqn (5.27). 6.3.3. Specific heat exponents « and a! For S=}, a straightforward though somewhat lengthy calculation leads to the expression A(T, M) = NkT[-In2 + $In(1 — 0?) + Joln {(1 + )/(1 — o)} —07/27'), (6.29) From (6.29) and (2.47), we see that Cy = —7'(22A/8T),,=0 for all 7. We can find CO; either from (2.48), C,, = —1(2G/8T),, from (2.51a), = T xr” * {(@M/6T),3}?, or from (2.54), Cy = T xp {((0H/2T)y}?. A de calculations lead to the same expression: for H = 0, Oy, = 0 for T > T,, while for T $ T, ig = ENE( — O(T, — TT, +--+}. (6.30) 86 MEAN FIELD THEORY 6,§3 Thus the limiting slope of C,, vs. J’ remains finite as 7’ — 7',, and we have the results «, = a, = 0, as well as «’ = « = 0. Further, we see that Cy > 3NK/2 as T>T;, (6.31) so that there is a jump discontinuity in the specific heat of magnitude AC, = 3NK/2 for S = } (cf. Fig. 6.3). 6.3.4. Susceptibility exponents y and y' We begin by observing that the zero-field isothermal susceptibility Xr = (2M/@H); satisfies the relation C (80 ),~ 7 (ae 099 ue (), (=). (=), = (Na) (282) where we have used (6.18) for the Curie constant for S = }. Hence we a Tf, 10 Fic. 6.3, Dependence of Cy for = 0 upon reduced temperature f' = 1/7. that although there is a simple jump discontinuity for all values of § (i.0. a = a’ = 0), the magnitude of the discontinuity depends weakly upon S. Notice also that the behaviour for the classical limit (S = <0) is only slightly different from the finite-§ results in the critical region, but is qualitatively different at low temperatures. After Mattis (1965). differentiate both sides of (6.26) with respect to h, obtaining, for 7 ~ 7, 1 = (80/@h) {(1 — 1/7) + 30? (1/37) + O(o4}} (6.33) or, on using (6.32), 2 -1 ae #5 +mt over} 3 (6.34) Now for 7 > T,,0 = OwhenH = Oand (6.34) reduces to _¢/?, 7-7,\-__€ (7 Te ) “Fo, Ce! 6, §3 CRITICAL-POINT EXPONENTS 87 Hence y = 1 and the coefficient @ defined in eqn (3.9) has the value unity also (cf. Fig. 6.4). However for 7’ < 7, we must substitute o? ~ —3e from (6,27), whereupon xe 23S (-97 (6.38) Thus we find y’ = 1 but the coefficient @ is only half as large below T, as above 7, (@’ = 4). “—Non-interacting limit “—— Mean field theory k—— ac ——_+ T 1 Fra. 6.4. Dependence upon temperature of the inverse isothermal susceptibility xr~ for the non-interacting limit (perfect paramagnet) and for the moan field theory. Only the behaviour for 7 > 7, is shown (of. Figs. 5.4 and 10.2), 6.3.5. The gap exponents A, and A; The gap exponents were defined in § 3.4 in terms of successive field derivatives of the Gibbs potential. Thus, for example, we have AL =2-0 —B=% (6.37) A=R+y¥=h Ag=Hy + 2-e) = (6.38) To obtain Aj, we need to differentiate the Gibbs potential three times, or, equivalently, we need to find the field derivative of the susceptibility evaluated at H = 0. From (6.34) we have (#). = a (e+ ay? {20 (3), + ove} (6.39) For T > T,, H = 0 implies c = 0, Hence the right-hand side of (6.39) reduces to zero, as we would already have suspected because the odd- order field derivatives of the free energy vanish for T > T. 88 MEAN FIELD THEORY 6, §3 For 1 < Tq, (d0/dh) = (T/C) xp ~ $(—e)-} from (6.36) and o ~ 3#(—e)# from (6.27). Hence c -G® = (#8), oF le — Be)-2(—Be)'B(—e)“t w (a) (6.40) and, since —@? = yp ~ (—«)~}, using (3.19) we have the result ag = 8. (6.41) 10 1-0 Perr, Fic. 6.5. Dependence of reduced magnetization ¢ = M/M, = M(7, H)/M(0, 0) upon reduced temperature 7 = 1'/T',. Tho fact that there is a slightly different curve for each value of the spin quantum number § means that this law of corresponding states is valid only for a given valuo of 8. The solid circles reprosent typical experimental data for Gd (S ~ }), Fo (8 ~ 1), and Ni (S ~ 4). After Martin (1967), 6.3.6. Law of corresponding states For general angular momentum quantum number S we can use (6.18) and (6.19) to eliminate the molecular field parameter ) from the equation of state (6.14) with the result, 3S_ oo o= 2B, (+555 5) where H = £j.SH is a dimensionless magnetic field. The critical-point exponents are the same for all values of angular momentum quantum (6.42) 6,§4 APPROXIMATION FOR THE HEISENBERG MODEL 89 number S, although the coefficients do depend upon S. For example, we find 10 (‘S + 1)? o- Teper (6.48) and (28 + 1)? - 1 AC =§Nk Qe Fat (6.44) Notice that eqn (6.42) is analogous to the law of corresponding states, eqn (5.17), in the van der Waals theory. However, (6.42) predicts that even if we measure H, M, and 7 in the proper units, two materials will behave differently if they have different spin quantum number S (see Fig. 6.5). 6.4. The mean field theory as an approximation for the Heisen- berg model In the preceding section we obtained a set of critical indices for the molecular field theory without assuming anything concerning the nature of the coupling constant A. Further, the theory does not give a microscopic argument for the introduction of the effective molecular field. This can be done, for example, through the Heisenberg model for a magnetic system. The basic idea of the Heisenberg model is that there exists an exchange interaction between the magnetic moments S, and S, localized on sites i andj, which can be represented by the form Egg = —2Ty S,+S;. (6.45) The minus sign is included in eqn (6.45) in order to conform to the convention that if the exchange parameter J, in (6.45) is positive the configuration in which the spins S, and S, are parallel is energetically favoured. (The case in which J,, is negative favours anti-parallel align- ment of S, and S,, giving rise to an antiferromagnetic ordering.) In order to render manageable the complexity of calculations based upon (6.45), it is often assumed that J,, = 0 except when sites i and j are neighbouring sites of the lattice. However, within the framework of the molecular field approximation it is possible to consider interaction potentials J,, of arbitrary range. The Hamiltonian for the entire system is then uN y H= -2 270 S,-S, — aD S,-H. > (6.46) 90 MEAN FIELD THEORY 6 §4 The mean field approximation to the Heisenberg Hamiltonian of eqn (6.46) is one of a large class of cluster approximations (Smart 1966) for which we treat only the interactions among the spins within a cluster exactly, whereas for the remaining spins in the system we set S,, = (8. and S,, = Sy = 0. For the molecular field approximation, we choose the cluster to consist of a single spin, so that the cluster Hamiltonian is simply w= (-2 Su 68> aH) 8. (6.47) Clearly the dynamics described by (6.47) are those of a single spin situated in a magnetic field of magnitude Hop = H + 2(Jo/fi) = H + 25,/Np2)M, (6.48) where Jo = Sj Ji, with Ji, = 0 for ¢ = j. Comparing this equation with (6.13), we see that the molecular field coupling constant dis given by d= Wo/Np?. (6.49) Hence it follows from (6.19) that the critical temperature is 1, = Rt S(S + 1). > (6.50) We see that for the case of nearest-neighbour interactions (Jj, = J only when site j is one of the q sites that are nearest neighbours of site ¢), J = q@J and the critical temperature is given by the expression ETT = 3gS(S + 1). > (6.51) When we compare eqns (6.50) and (6.51) with experimental results on materials that are thought to be described fairly accurately by the Heisenberg model and also when we compare with numerical calcu- lations (to be described in Chapter 9), we find that eqns (6.50) and (6.51) significantly overestimate the value of 7’,—by as much as 50 per cent (see Fig. 6.6). What is perhaps more disconcerting about (6.51) is that it predicts that two lattices will have the same value of the critical temperature if they have the same coordination number (i.e. the same value of ¢)- For example, the simple cubic lattice is predicted to have the same value of 7’, as the plane triangular lattice, since for-both lattices g = 6 (cf. Fig. 6.7). Thus the mean field theory is evidently too crude to take 6,85 INFINITE INTERACTION RANGE a1 into account the dimensionality of the lattice, which is believed to be one of the crucial features determining critical behaviour. T Te rT (experiment) (MFT) Fra. 6.6. Schematic comparison of typical experimental measurements on a Heisenberg ferromagnet (such as EuS) with the predictions of the molecular field theory (ef. Fig. 3.3). Note that the curve for 1/yr is shown only for 7 > T. (a) Fra, 6.7. (a) The plane triangular lattice, and (b) the simple cubic lattice. The fact that both lattices have the same coordination number (q = 6) means that within the mean field approximation they are predicted to have the same critical temperature (cf. (6.51)). ‘Thia prediction is at odds with the more accurate approximation procedures to be dis- cussed in Chapter 9, and it is currently believed that the critical temperature depends rather strongly upon lattice dimensionality. After Ziman (1964). 6.5. Equivalence of the mean field theory and an infinite inter- action range In this section we show that the mean field theory results follow froma model which assumes that each spin interacts equally with all the other spins in the lattice. The argument (Kac 1968) is most easily illustrated for an ‘Ising’ interaction, #=(-2/N) > 43, > (6.52) isi 4 =I-¥/( “)° (6.54 WAR tw Ae mT Wee Hence the partition function is 1 1 N 2 Zeon eee > exp {#(5,22"-4) } (6.55) atti Sta i where J = J/kT = J is a dimensionless coupling constant. Using the identity exp (a*) = (2n)-#[”_ oxp (—$2? + 2az)de, (6.56) we can write the partition function in the form © 1 N Zy = (2n)7* ef e772 > exp {e200} > “} de. “oe a=-1 sy=-1 i=l (6.57) The summations now can be performed independently, and we have Zy = (2n)-to-F J" 0 #79 (2 cosh (2F/N)tx" dz. (6.58) If we substitute z = yN, we obtain . Zy = (N]2n) oF 2® [°_ {0-2 cosh (2F)ty}" dy, (6.59) which suggests application of the saddle-point method. Thus Zyo Nte-%2" max {e-¥? cosh (2f)ty}™ (6.60) -o 2J/k), the maximum occurs aty = 0and@(7,H) = —kT'In2fromoqn (6.61).For J > $(or 7 < 2J/k), the maximum occurs at y = yo and there is an additional contribution to the Gibbs potential G(7', H). This figure is adapted from Kae (1968). To find the maximum in eqn (6.62), we calculate the maximum of the logarithm of the bracketed factor, Sly) = —4y? + Incosh (2f)ty. (6.63) Setting f’(y) = 0 results in the equation y = (29)! tanh (29)ty. (6.64) Equation (6.64) is recognized as being essentially the molecular field equation for H = 0; in particular, for J < 3 (or T > 2J/k), the only solution of (6.64) is y, = 0, whereas for J > } (or T < 2J/k) there exists a solution y, # 0 (see Fig. 6.8). Suggested further reading Weiss (1907). Van Vleck (1945). Ziman (1964). Kittel and Shore (1965). Brout (1968). Smart (1966). Kao (1968). Ser, 7 THE PAIR CORRELATION FUNCTION AND THE ORNSTEIN-ZERNIKE THEORY In this chapter we shall consider a third classical theory—the Ornstein- Zernike theory. In the course of our development, we shall introduce the pair correlation function for a fluid system. This function plays a particularly important role in almost all discussions of critical phenom- ena. 7.1, The density-density correlation function for a fluid system Consider a fluid confined in a macroscopic volume V. If the fluid is in equilibrium and can exchange particles with an external reservoir, the probability density for the system to have N particles in a particular configuration of position and momenta (r,,..., Ty, Py--+, Py) is given by the grand canonical ensemble to be —BUA(T,,.-- Ty, Pi. -- + Pyltyy c+ yy Daye By) = SPLCBU ale es Pal Bu, (7.1) where 8 = 1/kT’, h is Planck’s constant, » is the chemical potential, Uy(t1, Ta,+++ Ty, Pis-++ Dy) is the energy of the system, and 2 Bun 2 = ¥ ange [an-.. dry dp, ... dPy x x exp{—BUy (f1,.+- Ty, Pis+++ Pw} (7.2) is the grand partition function. What will concern us here are the average values of variables when they are weighted with the probability factor (7.1). Let us first consider the density nr) = > ar - 1) (7.3) i=l at a point r in the fluid; here r, is the spatial coordinate of the ith 7§2 DENSITY-DENSITY CORRELATION FUNCTION 95 particle. On using (7.1) we obtain for the average value of the density the expression © dn(r)> = 2-4 (NAR) f dr dp n(r) exp(—BUy + Bul), (7.4) v=o where we have adopted the shorthand notations Uy = Uy(ry,...Ty, Pi,--- Py), dr = dr, dra... dry and d¥p = dp, dp,... dpy. In a uniform system, is independent of position and is therefore simply i Xn(t)y = G> =n (7.5) A quantity that reflects in a more detailed fashion the microscopic properties of the system is dn(r) n(r')) = 2-2 5 (NTAPN)-* f d*r dp n(r) n(r’) x reo x exp(—BUy+fyN). (7.6) The quantity is proportional to the probability of finding a particle at the position r if we know that there is a particle at the position r’. In this sense is a measure of a conditional probability. A function of more immediate interest is G(x, r’) = C{n(r) — G(r — r’). Further- more, = , and eqn (7.7) may be written in the equivalent form - G(r — 2’) = — 1. = (7.8) As |r — r’| 00, the probability of finding a particle at r’ becomes independent of what is happening at r; that is to say, the densities become uncorrelated. Hence > = 2? [as |r — r’| 00], (7.9) and from eqn (7.8), Qe — 41’) +0 [as |r — r'| +o]. (7.10) 96 THE PAIR CORRELATION FUNCTION 1,§2 G(r) (b) Fic. 7.1, Lennard-Jones model of a fluid. (a) Two-body potential U(r). (b) Density— density correlation function G(r). The dependence upon |r — r’| of G(r — r’) is shown in Figs. 7.1 and 7.2 for two different interparticle potentials U(r), a Lennard-Jones potential and a hard-sphere potential. Ue) ee vin={o 32 om | | i a r t Tr ° (a) j (b) Fis. 7.2, Hard-sphere model of @ fluid. (a) Two-body potential U(r). (b) Density— density correlation function G(r). 7.2. Relation between density fluctuations, the isothermal com- pressibility, and the density-density correlation function In this section we shall show that the isothermal compressibility Ky can be directly related to the fluctuations in the total number of particles in the system—which in turn is directly related to G(r — r’). We shall see that the divergence of Ky as T > T, is equivalent to the correlation function G(r — r’) becoming very long range (i.e. G(r — r’) approaches zero very slowly as |r — r’| increases without limit). We first note that the fluctuation in the total number of particles is «(NW — = <> — -(=) > NY \ 6/2, 6 whence eqn (7.16) becomes N>*kT daw — ayy = x, (7.18) = - (7.21) If we introduce the definition (7.7) of G(r — r’) into (7.21), we obtain (WW — (Ny) = far f dr’ G(r — r’) = V far" ar’), (7.20) (7.22) where the last equality in (7.22) follows from the uniformity (trans- lational invariance) of the system. Combining eqns (7.20) and (7.22), we have a =n? far G(r). = (7.23) From eqn (7.23) we see that a divergent compressibility near_7', corresponds_mathematiesHy—te-an—inereasa_in the range of the. -pair correlation function G(r — r’). Sufficiently close to 7’, the correlation length becomes as large as the wavelength of light and the density inhomogeneities scatter light strongly; this is the phenomenon of critical opalescence, which is displayed in Figs. 1.6 and 1.7. In summary, then, we see from eqns (7.20), (7.22), and (7.23) that the three phenomena that are observed near the critical point, (i) increase in the density fluctuations, (ii) increase in the compressibility, and (iii) increase in the range of the density-density correlation function are all interrelated phenomena. In particular, eqn (7.23) is the analogue for fluid systems of the fluctuation—dissipation relation (A.20) which relates the isothermal susceptibility y7\to the summation over all spins of the two-spin correlation function of a magnetic system. 7.3. The structure factor: relation between the pair correlation function and the scattering of electromagnetic radiation In this section we relate the above discussion of the correlation function G(r) tothe intensity distribution of elastically-scattered electro- 7,§3 THE STRUCTURE FACTOR 99 magnetic radiation. Imagine that we irradiate our critical fluid with a monoenergetic beam of X-rays, neutrons, or light. Assume that the scattering takes place quasielastically, that is, assume that the energy of the radiation is much larger than the typical excitation energies of the-syatem. Hence we can write for the momentum transfer vector q = k, — k, the expression (see Fig. 7.3), a q=|a| = %sing; k= |k|~|k,|. (7.24) We wish to calculate J(q), the intensity of the scattered radiation. Now 1a) = <2 > where a,(q) is the scattering amplitude for scattering from particle j. (7.25) Fic. 7.3. Soattering geometry used to determine eqn (7.24). Now we know that the scattering from two different particles in the system is simply related by a phase factor; thus a,(q) = a,(q) e's" FD. (7.26) 1a) = » (7.21) = |a,(a)? : (7.28) Now if there were no correlation between the particles, the scattering intensity I°(q) would be given by I°(q) = Na,(q)|?. (7.29) Hence from eqns (7.28) and (7.29) Z(a) mse, rey {Semen ”» Hence ¥ a,(a) > esr gel star, (7.30) 100 THE PAIR CORRELATION FUNCTION 1g4 which may be written as dq) _1 , , -ta@-r P@ 7 wf far CE oe - 1) 8(r’ — 1) eI" » 5 wf* fares eee <{ ar - w}{> Ber" — nah» (7.31) On introducing the number density n(r) through (7.3), (7.31) becomes my = Hy fae fae’ ome day n(e’)> (7.32) = xfer far'ermenn {Or — 1’) + 0%, where the second equality in (7.32) follows from (7.8). Equation (7.32) may be written as AD _F age g-sa-e Gerry 4 nt : Pa 75 dr” o-!e-r" G(r") + 5 n* 8(a). (7.33) The last term in eqn (7.33) contributes only when q = 0, and hence, by (7.24), only for forward scattering (0 = 0). Therefore it is customary to omit this term in many expressions, and to simply write na = = far ee" G(r) = 25a), » (7.34) where the second equality in (7.34) serves to define the structure factor 8(q) as the spatial Fourier transform of the density-density correlation function. Equation (7.34) tells us that the intensity of radiation scattered through a wave vector q is changed from the value I°(q) (predicted if the particles of the fluid were not interacting) by an amount proportional to the Fourier transform of the density—density correlation function. In particular, as the critical point is approached, the integral in (7.34) becomes extremely large for small values of q. This dependence upon q is displayed in Fig. 7.4 for scattering from a binary alloy at several temperatures in the critical region. 7.4. Ornstein-Zernike theory of the scattering amplitude The first qualitative explanation of the huge increase of forward scattering called critical opalescence was due to Ornstein and Zernike (1914). Two particularly readable reviews of their original work and of 4m sin 8 q= Soin 5 Fra. 7.4. The q dependence of the structure factor S(q) for several temperatures just above the critical temperature 7’, = 361-5°C. After the work of Miinster and Sagel (1958) on Al-Zn alloy. The ‘lattice-gas’ model of a simple fluid (cf. §1.1) also provides useful model for a binary alloy. 102 THE PAIR CORRELATION FUNCTION 1§4 numerous subsequent developments are Miinster (1966) and Fisher (1964). In order to develop the Ornstein—Zernike theory, we must make use of the density-density correlation function G(r — r’) introduced in eqn (7.7). We begin by substituting eqn (7.3) for the number density n(r) into eqn (7.8), with the result NON Gr -r)= Ks a 8(r — r,) 8(r’ — rj) )—2, (7.35) Notice that the terms with i = j are not excluded in eqn (7.35). This is reflected in the fact that G(r — r’) includes not only correlation between different particles, but also the correlation of a particle with itself. Therefore it is convenient to partition G(r — r’) into two terms, Or — 1!) = nd(r — rv’) +n? T(r - vr’), (7.36) where the first term is the contribution to G(r — r’) due to correlation of a particle with itself and the second term is the contribution to G(r — r') due to correlation between different particles. Notice that the presence of the factor n? in the second term implies that the function T(r — r’) is dimensionless. We next introduce the ‘direct correlation function’, C(r), by means of its Fourier transform, O(q) = C(x) exp (—iq - r) dr, where C(q) is defined by =—f@ O@) = ara (7.37) here T(q) = {T'(r) exp (-iq‘r) dr is the Fourier transform of the function I(r). Note that O(q) thus defined does not depend very strongly on temperature: at high temperatures where f(q) is almost zero, C(a) ~ fq), while as T+ 7, where f(q = 0)>0,C(q = 0) = fO(r) dr ~ n-1. Thus this new function O(r) must remain compara- tively short-range even at 7’,! It is perhaps convenient to think of C(r) as being some function with a range similar to that of the interaction potential U(r), and then to imagine that I(r) becomes long-range near 7, a8 a result of the propagation of this direct correlation—see for example, Fig. 7.5. This intuitive picture is supported by the Fourier transform of eqn (7.37), Dee — r') = Or — 2) + afr - re" —r’)dr”. (7.38) LS4 THE SCATTERING AMPLITUDE 103 Equation (7.38) has come to be called the Ornstein—Zernike integral equation, and of course it may equally well be chosen as the defining equation for C(r), since (7.37) and (7.38) are equivalent (as may be seen from the convolution theorem). 7.4.1, The Ornstein-Zernike assumption Our goal is to calculate the structure factor S(q) = » + n?f(q), since this gives us the scattered intensity I(q). We can obtain the qualitative form of S(q) by following Ornstein and Zernike and assuming that Clr) A Clrag) SOT TD Prin) Fic. 7.5. Schematic diagram to illustrate the difference between the total correlation function I'(r) (which can be long-range) and the direct correlation function C(r) (which is short-range). O(q) may be expanded in a Tayor series about g = 0 for all temperatures right up to 7',. Thus, for an isotropic system, O(a) = C(0) + 2 den, Tye. (7.39) The coefficients é,(n, 7’) in eqn (7.39) are given by Maclaurin’s ex- pansion, én, T) = ate CC) af, pf dy [ete dr, (7.40) where the first integral in eqn (7.40) arises from the angular integration and is zero for odd values of the integer ¢. Although we know from the definition of O(q) that O(r) is integrable at T = T., we have no reason to expect that all of the moments of C(r) are also integrable at T',. Yet let us follow Ornstein and Zernike and assume for now that the coefficients ¢; (n, 7',) are finite for all ¢. Then it follows at once from the defining equation for C(q) that for small q, the scattered intensity is very nearly a Lorentzian. To see this, notice from (7.36) and (7.37) that £8(q) =1 + af = Tost’ » (7.41) 104 THE PAIR CORRELATION FUNCTION T§d Hence 5a = 1 nO(q) = 1 - n{0(0) + éa(n, 7) g? + O(0*)} = te, 7) MOO) _ nt + or} = Bch + P+ OG), (7.42) where BR? = —nég(n, T) o [r°0(r) dr (7.43) is directly related to the second moment of C(r), and Ks 1= 2600 (7.44) is related to the zeroth moment. From eqn (7.42) we would predict that a plot of inverse scattering intensity, 1°(q)/I(q) = n/S(q), against q? for various temperatures would result in a family of curves which for sufficiently small g? are linear, with slopes given by R? = —né,(n, T) and intercepts by 2x? = 1 — nO(0) = (1 + nf(0)}-? = n8(0)-? = KZ/Ky. (7.45) Note that x? ~ ¢’, assuming R? is not singular at 7 = 7’. The slope parameter R is sometimes called the Debye persistence length but here we regard it as phenomenological parameter since we have seen no way in which C(r) and hence R? may be calculated. A plot of inverse scattering intensity against q? (called an Ornstein-Zernike-Debye or OZD plot) is given in Fig. 7.6 for argon. That the limiting slopes of the curves in Fig. 7.6 change only slightly as 7’ > 7’, is consistent with our assumption that C(r) remains reasonably short-range in the critical region. 7.4.2, The Ornstein—Zernike approximation Next we make what we shall call the OZ approximation—we truncate the expression (7.42) by neglecting all terms of order g* and higher. In this approximation (which presumably is good for small q) 1(q) is indeed a Lorentzian, @ _ Sq) _ _R PQ) on +e » (7.46) The Fourier inversion of eqn (7.46) shows that the OZ approximation 7,84 THE SCATTERING AMPLITUDE 1085 2. [ Argon | sy 4 | 207 | ret z Fer Fal eld g 14 g z 312 3 3 g lot J ost 06 4 04 4 T.= 1508 K 02} 4 i | 005-39, 40D ~~«BO~~«d100~=«2120~«140~—~*160~—+180 (Scattering angle (rad))® x 10-6 Fic. 7.6. Dependence on g? of the inverse scattering intensity for argon. Apparently the predictions of the Ornstein—Zernike theory are borne out by these measurements, Taken from Thomas and Schmidt (1963). Note that the slopes of the lines increase as T’ —> T'g, corresponding to an increase in the parameter R? defined in eqn (7.43). to the asymptotic form for large r of the total correlation function G(r) is on ar) ~ z . > (7.47) From the definition (7.44), we see that x2 0 as T’ > 7',. Clearly, x7? has the dimensions of length and we set a (7.48) where « is called the inverse correlation length. K=k 106 THE PAIR CORRELATION FUNCTION 1,85 From eqns (3.8), (7.45), and (7.48) we find that as 7 T,,x ~ ed, Hence from the exponent definitions in eqn (3.15), we have v = by’ (7.49) vel. (7.50) 7.5. Further developments of the Ornstein-Zernike theory 7.5.1. Failure of OZ theory for two-dimensional systems Suppose we consider applying the OZ approximation to a fluid system of arbitrary dimensionality d. The development of the proceeding section proceeds through eqn (7.46) essentially unchanged. To get G(r) we must invert eqn (7.46) for general dimensionality d. The d-dimensional Fourier transform, Gr) = [S(a)e“*"* dq with G=(Gdn--.94) (7.81) has a volume element dq = g*~* dg (sin 0,)¢~? d@, dQg_4, where q = |q|, and dQ,_, is an element of solid angle in d — 1 dimensions. Hence eqn (7.51) becomes, for an isotropic system, G(r) = fsa) { f etr0o8 8, (gin 6,)¢~? 46} dQ4_1 2-1 dg. (7.52) Now the integral inside the braces in (7.52) can be evaluated in terms of Bessel functions (Arfken 1970), whence J, Gtr) cc [° say “AB ae? a+ dep (7.83) where J;,(z) is a Bessel function of the first kind. Note that the kernel multiplying §(q) in the integrand reduces for d = 3 to the familiar expression (sin gr)/ gr. Fisher (1962) has carried out the Fourier inversion in detail for general d, with the following results: Case (i). Fixed T > 7, (and hence fixed x > 0) but let r — a0. 7 d-3 Qn) a a mi! + o(S = ) (7.54) Case (ii). Fixed r, but let T’ > T+ (x —> 0). (nr) emf + °(s5)} d=2 Greco d=3 (7.58) = sa {l + own} d>3. and $5 FURTHER DEVELOPMENTS 107 For d = 3, we see that the higher order terms in eqn (7.54) vanish identically; thus case (i) and case (ii) both give the same answer, G(r)oc rt em, The correlation function, for large values of r, is found from (7.55) to behave as Inr d=2 (7.56) Gr)|p-7, © { ea r d>3. (7.57) From eqn (7.56) we see that the Ornstein-Zernike theory predicts that at the critical temperature of a two-dimensional fluid, the correlation increases with distance—this is clearly a nonsensical result! Indeed, for the two-dimensional Ising model, we can obtain the two-spin correlation function exactly (Kaufman and Onsager, 1949, Stephenson 1964) and find that Ar) |rero cr 1* [d = 2 Ising]. (7.58) 7.5.2. Fisher’s Modification of Ornstein—Zernike Theory For many systems one finds experimentally that plots of inverse scattering intensity against q? are not straight lines but rather that as Fig. 7.7. Sketch of the normalized inverse scattering intensity as found for many systems. The curvature is smaller for many systems (es, e.g. for Ar in Fig. 7.6.) and the result 7 = 0 is not outside the experimental error. By eqn (7.34), 1°(q)/1(g) = n/S(q)- T-»T, they become concave as sketched in Fig. 7.7. Fisher (1964) explains this downward turn of the isotherms for small g by means of the assumption that S@|rer,~9-7*" — [q ~ O], (7.59) where the index » is zero for the Ornstein-Zernike theory. Equivalently, on Fourier inverting (7.59), we have GAr)lrar, ~ 72-24 — [r large), (7.60) which is the definition of 7 used in eqn (3.16). 108 THE PAIR CORRELATION FUNCTION Ls For the two-dimensional Ising model, eqn (7.60) is rigorously true and, using (7.58), we have 7 = }. Moreover, v = 1, and y = j so that the relation y = (2 — 4) vis satisfied. For the three-dimensional Ising model, the best numerical approximations (Moore, Jasnow, and Wortis, 1969) suggest that y = 1.26, (7.61) v = 0-638 +8:002, (7.62) and 7 = 0-041 +9:998, (7.63) The estimate of 7 was obtained from the estimates of y and v and using the relation y = (2 — 9)». For the spherical model (the one exactly-soluble three-dimensional magnetic model), y = 2,v = 1, and 7 = 0. Most experimental measure- ments suggest a value of 7 between zero and 0-1, but the experimental error in 7 is frequently so large as to make excluding y = 0 impossible. Suggested further reading Ornstein and Zernike (1914). Fisher (1964). Miinster (1966). Egelstaff (1967). Rowlinson (1969). Stell (1969).

You might also like