Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
3 Critical Phenomena and Phase Transitions INTRODUCTION One of the most active areas of physics in the last decade has been the subject of critical phenomena. Enormous progress was made during the decade, both theoretically and experimentally, and research in the field was honored with the award of the 1982 Nobel prize in physics. It seems safe to predict that the study of critical phenomena and closely related subjects will remain a major activity of condensed- matter physics throughout the 1980s and that much further progress will occur. WHAT ARE CRITICAL PHENOMENA, AND WHY ARE THEY INTERESTING TO PHYSICISTS? The term critical phenomena refers to the peculiar behavior of a substance when it is at or near the point of a continuous-phase transition, or the critical point. A continuous-phase transition, in turn, may be defined as a point at which a substance changes from one state to another without a discontinuity or jump in its density, its internal energy, its magnetization, or similar properties. The critical point or continuous-phase transition may be contrasted with the more familiar case of a first-order phase transition, where the above-mentioned properties do jump discontinuously as the temperature or pressure 75
76 A DECA DE OF CONDENSED-MA TTER PH YSI CS passes through the transition point. Continuous-phase transitions in many cases, but not all, are associated with a change of symmetry of the system. Although the critical point was first discovered more than 100 years ago, a good understanding of behavior near a critical point has only emerged recently. The peculiarities of the critical point arise because there are, in each case, certain degrees of freedom of the system that show anomalously large fluctuations on a long-wavelength scale, compared with those of a normal substance far from a critical point. These large fluctuations cause a breakdown of the normal macroscopic laws of condensed-matter systems, in some dramatic ways and in some subtle ways, and it has been a major challenge to learn what are the new special laws that describe the systems at their critical points. The challenge has been difficult for theorists because the large fluctuations could not be handled by the old calculational schemes, which depended implicitly on long-wavelength thermal fluctuations being small. The challenge has been difficult for experimentalists, because in order to make measurements sufficiently close to a critical point, to test existing theoretical calculations, or to discover directly the laws of critical behavior where no theory exists, it is necessary to have extremely precise control over the sample temperature, and frequently over the pressure and purity as well. The study of critical phenomena has been rewarding in spite of its difficulties, and the understanding gained has proved useful to the understanding of other types of systems-including quantum field theories in elementary-particle physics, analyses of phenomena in long polymer chains, and the description of percolation in macroscopically inhomogeneous systems in which fluctuations play an important and subtle role but where precise direct experiments may be even more difficult than in the case of critical phenomena. The techniques of renormalization-group analysis, developed in the theory of critical phenomena, have had a profound impact on an entire branch of mathematics, for example in the study of iterative maps, which has applications to economics, biology, and other sciences, as well as to the study of nonlinear fluid dynamics and other problems in condensed- matter physics. Experimental research on critical phenomena has also had an impact both inside and outside condensed-matter physics. The requirements of experiments on critical phenomena have often stimulated the synthesis of samples with a new degree of perfection and of materials with special properties such as magnetic systems with anisotropic
CRITICAL PHENOMENA AND PHASE TRANSITIONS 77 spin interactions. The precision measurement techniques developed for the study of critical phenomena have also found application, for example, in the study of the onset of fluid convection. In the following sections, we further define the features of a critical point, and we give some examples of properties that show critical-point anomalies. We outline the progress that has been made in the field, and we give a few selected examples of important problems that are still unsolved. EXAMPLES OF PHASE TRANSITIONS AND CRITICAL POINTS Several examples may illustrate the difference between a first-order transition and a continuous transition or critical point. One example of a critical point is the Curie point of a ferromagnetic substance such as iron (TC = 770°C for iron). At temperatures below Tc, a single-domain sample of iron has a net magnetization M that points arbitrarily along one of several directions that are energetically equivalent, in the absence of an external orienting magnetic field. The strength of the magnetization M(~ decreases with increasing temper- ature until the Curie temperature Tc is reached. Above Tc the magne- tization is zero in the absence of an applied magnetic field, and we say the material is in a paramagnetic state. In most magnetic systems (including iron) the magnetization M(D decreases continuously to zero as the temperature approaches TC from below; then we say that there is a continuous phase transition, or critical point, at Tc In some cases, however, the magnetization of a substance approaches a finite, non- zero value, as T approaches TC from below, and the magnetization jumps discontinuously to zero, as the temperature passes through Tc In these cases there is a first-order transition at Tc. In the magnetic example there is a symmetry difference between the phases involved, since the ferromagnetic phase has a lower symmetry than the paramagnetic phase. The familiar boiling transition, from liquid to vapor, is a first-order transition. Thus, when water boils at 1 atmosphere pressure and a temperature of 100°C, there is a decrease in density by a factor of 1700. However, if the pressure is increased, the boiling temperature in- creases, and the difference in density between the liquid and vapor becomes smaller. There exists a critical pressure PC where the density difference between liquid and vapor becomes zero; at this pressure the phase transition is no longer first order, and the transition temperature TC at pressure Pc is described as the gas-liquid critical point of the
78 A DECADE OF CONDENSED-MATTER PHYSICS substance. For pressures greater than Pc there is no distinction at all between liquid and vapor. We may note that there is no symmetry difference between the liquid and vapor phases. In order to facilitate the comparison between critical points in various systems, it has proved convenient to introduce the concept of an order parameter associated with each phase transition. For systems like the ferromagnet, where there is a broken symmetry below Tc' the order parameter is a quantity like the magnetization, which measures the amount of broken symmetry in the system. For systems without broken symmetry, one chooses some quantity that is sensitive to the difference between the two phases below the critical temperature and measures the difference of this quantity from its value at the critical point. For the liquid-vapor critical point, we may choose the order parameter as the difference between the actual density of the fluid and the density precisely at the critical point. HISTORY The earliest theories of critical phenomena, developed near the end of the last century and at the beginning of this century, gave a good qualitative description of the behavior of a system near its critical point. However, it gradually became clear in the mid-twentieth century that these classical theories were incorrect in important details. A most important step in this realization occurred in the 1940s, when Onsager found a remarkable exact solution of a model of a magnetic system in two dimensions (known as the two-dimensional, or 2-D, Ising model) and showed that its phase transition did not follow all the predictions of the classical theories. In the 1960s, experiments on actual three-dimensional (3-D) systems began to show more and more clearly that their critical behavior was also different from that predicted by the classical theories and different from that of the 2-D Ising model as well. At the same time, there appeared a certain regularity to the behavior of different 3-D systems, which was encouraging to the search for some general theory of these transitions. Other evidence for this viewpoint, and hints at the shape that the new theory must take, were provided by various types of numerical calculations (one might call them computer experiments), which included both computer simula- tions of thermal fluctuations in simple magnetic models and also numerical extrapolations of the properties of these magnetic systems from temperatures far above the critical temperature, where accurate calculations could be done. An important step forward in our understanding of critical phenom
CRITICAL PHENOMENA AND PHASE TRANSITIONS 79 ena occurred in the mid-1960s, with the development of a set of empirical scaling laws, which were successful in describing certain relations between different critical properties of a system, although they could not predict all these properties from the beginning. The concept of universality classes developed, as it appeared that systems could be divided into certain broad classes, such that all members of a given class had identical critical properties but that these same properties varied from one class to another. One important factor that affects the critical behavior is the spatial dimensionality of the sys- tem- e.g., 3-D systems have different critical behavior than 2-D systems but there are other factors that are relevant, including the symmetry differences between the states at the phase transition, the presence or absence of certain long-range interactions, and other factors that will be discussed below. A proper understanding of the factors that determine the universality class of a system had to await the developments of the 1970s, however, and in fact, a classification in the more difficult cases remains one of the tasks for the 1980s. The most important theoretical advance of the 1970s was the development of a set of mathematical methods known as renormaliza- tion group techniques. These methods are not limited to critical phenomena they are useful whenever one has to deal with fluctua- tions that occur simultaneously over a large range of length scales (or energy scales or time scales, for example). The methods proceed by stages, in which one successively discards the remaining shortest- wavelength fluctuations until only a few macroscopic degrees of freedom remain. The effects of the short-wavelength fluctuations are taken into account (approximately) at each stage by a renormalization of the interactions among the remaining long-wavelength modes. The renormalization group techniques have made possible a number of achievements. 1. They have given us a justification for the scaling laws of the 1960s. 2. The renormalization group methods enable one to predict with high reliability which features of a system are relevant to determining its universality class for critical behavior and which features of the microscopic description become irrelevant in the vicinity of the critical point. 3. The renormalization group methods enable us to calculate prop- erties of any given universality class. In the simpler cases, these critical properties have been calculated with a high degree of accuracy; and these predictions of the renormalization group have been confirmed in turn by some beautiful experiments of high precision. In more compli
80 A DECADE OF CONDENSED-MATTER PHYSICS cased cases, the numerical accuracy of existing renormalization group calculations is not high, and further improvements in them are badly needed. WHAT DOES ONE MEASURE? In order to make more precise our discussion of critical phenomena, it is useful to give some examples of quantities measured and to give some examples of the laws that describe them. Perhaps the most fundamental measurement to make in the vicinity of a critical point is to determine the way in which the magnitude of the order parameter approaches zero, as the critical point is approached from the low-temperature side. According to the classical theories of phase transitions, such as the van der Waals or mean-field theories, the order parameter should approach zero as the square root of the temperature difference from Tc. We may write this as M= Mo(Tc- T)h, (1) where M is the order parameter on the coexistence curve (i.e., for a ferromagnet, M is the magnetization in zero magnetic field; near the gas-liquid critical point M is proportional to the density discontinuity between liquid and vapor). Me is a constant that will vary from one system to another, and the exponent ~ is equal to 1/2 for all critical points, in the classical theories. Now the result of the modern theory of critical phenomena is that the classical theory is not correct close to Tce We can still write the temperature dependence of the order parameter in the form of Eq. (1), but the value of the exponent ~ is not equal to 1/2. For the 2-D Ising model of magnetism, and for other 2-D systems in the same universality class, the result is ,8 = 1/8, as given by the Onsager solution. For the gas-liquid critical point in three dimensions, as well as for the 3-D version of the Ising model, the result of the most accurate experiments and renormalization group calculations is ,B = 0.325, with an estimated uncertainty of +0.001. Other 3-D systems may belong to different universality classes, but their values of ~ are typically in the range 0.3-0.4. The forms of the power law Eq. (1), for various values of the exponent is, are illustrated by the curves in Figure 3.1. The curves for ~ = 1/8, 0.325, and 1/2 are all qualitatively similar, and indeed the quantitative differences appear small on this linear scale. The differ- ences may actually be quite large, however, if precision measurements are made sufficiently close to Tc. For example, if the constants Me are chosen so that the various curves have unit magnetization at a
CRITICAL PHENOMENA AND PHASE TRANSITIONS 81 \\' M \~,3~5 \ ~\ in, 1_$ p=2i Tc FIGURE 3.1 Power-law M °° ~TC - To, for various values of the exponent p. Data points are experimental measurements of the order parameter of the antiferromagnet MnF2, in the range 1.8 degrees below the critical temperature TC = 67.336 K. temperature 10 K below Tc, then the curves for ~ = 1/8, 0.325, and 1/2 take on the respective values M = 0.316, 0.050, and 0.010, at a temperature 0.001 K below Tc. Thus there is a difference of a factor of 5 between the values in the last two cases. The temperature variation of the order parameter on the coexistence curve is certainly not the only quantity that can be studied with experiments in critical phenomena. Another important quantity is the order-parameter susceptibility, defined as the derivative (i.e., the rate of change) of the order parameter with respect to a small change in the field to which it is coupled, while the temperature is held constant. For a magnetic system this quantity is the magnetic susceptibility (deriva- tive of the magnetization with respect to magnetic field); for the gas-liquid critical point the order-parameter susceptibility is the iso- thermal compressibility (derivative of the density with respect to pressure, at constant temperature). These quantities become extremely large near the critical point, and we may write, for example, the zero-field magnetic susceptibility as X = Xo/~T- TO Y. (2) where the exponent ~ is the same for all members of a universality class. The coefficient Xo varies from one system to another, and it is
82 A DECADE OF CONDENSED-MATTER PHYSICS different above and below Tc; however, the ratio of its value above Tc to its value below Tc is a universal number i.e., it is the same for all members of a universality class. Another important quantity is the specific heat, defined as the derivative of the internal energy of the system with respect to a small change in temperature. The specific heat is found to become infinite at the critical point in some systems; for some other universality classes one finds that the specific heat is finite but has a sharp cusplike maximum at the critical point. In either case, one may define an exponent ax that characterizes the anomalous behavior of the specific heat at the critical point. An example of the specific heat behavior is shown in Figure 3.2. Although the critical exponents a, if, and fly defined above may be independent in principle, they were found empirically, in the 1960s, to obey a scaling law: = 2-~-2~. (3) This scaling law is one of the consequences of the more recent renormalization group theories. 80 60 o E 40 :~. rig. it. Bow t. O. T < Tc 0 a. ·Cb . 20 _ o ~ ma °. ° o ° °°oOOk 1 1 1 1 1 1 IO-6 10-5 10-4 10-3 IO-2 10 1 | T-TC |/TC FIGURE 3.2 Specific heat Cs of liquid 4He at saturated vapor pressure, near the temperature T`. = 2.172 K of the onset of superfluidity, as a function of IT - TO on a logarithmic scale. A straight line on this plot would correspond to a logarithmic temperature dependence for Cs or a critical exponent cx = 0. Careful analysis of the data gives the result cx = -0.026 + 0.004.
CRITICAL PHENOMENA AND PHASE TRANSITIONS 83 A property of great interest near the critical point is the statistical correlation for fluctuations in the order-parameter density, at two nearby points in space, as a function of the distance between the points. This correlation function can be measured by neutron- scattering experiments in magnetic systems, and it can be measured by light-scattering experiments or small-angle x-ray-scattering experi- ments near the liquid-vapor transition. Near to the critical point, the correlation length, which characterizes the range of correlations for the order-parameter fluctuations, becomes extremely large relative to the typical spacing between atoms in the substance. This large correlation length is directly related to the large amount of long-wavelength fluctuations that were mentioned earlier and that give to critical phenomena their special subtleties and complexities. Naturally, there is great interest in studies of the variation of the correlation length with temperature, pressure, and other parameters, near the critical point. It should also be mentioned that the integrated order-parameter correlation function, which can be directly measured by a scattering experiment in the limit of small angles, is related by a theorem of statistical mechanics to the order-parameter susceptibility X defined above; thus a scattering experiment may be a convenient method of measuring X. Also, for systems like an antiferromagnet, in which the order parameter describes a quantity that oscillates as a function of position in space, a scattering experiment may be the only direct way of measuring the value of the order parameter M(1) in the broken- symmetry phase below TC. Many other experimental techniques have also been used to study properties of various systems near critical points. For example, the temperature coefficient of expansion of a solid, which can be measured with great precision, has similar behavior to the specific heat near a critical point; the index of refraction of a fluid has been used as a measure of its density; the rotation of polarized light by a transparent ferromagnet (Faraday effect) has been used to study the temperature dependence of the magnetization. The quantities discussed above are all equilibrium or static quanti- ties; they can be measured in a time-independent experiment, under conditions of thermal equilibrium, and any correlation functions in- volved refer to the correlations of fluctuations at a single instant of time. The majority of theoretical studies and of experiments on critical phenomena are concerned with these static measurements, and the usual division of systems into different universality classes is based on these static phenomena. There are other properties of systems, known as dynamic properties, which require a more detailed theoretical
84 A DECADE OF CONDENSED-MATTER PHYSICS analysis and which require a further subdivision of the universality classes- i.e., two systems that belong to the same universality class for their static properties may show quite different behaviors in their dynamic properties. Examples of dynamic properties are various relaxation rates when the system is slightly disturbed from equilibrium, correlations involving fluctuations at two different instants of time, and transport coefficients, such as the thermal and electrical conductivities. Among the experiments used to study dynamic properties are mea- surements of sound-wave attenuation and dispersion, widths of nuclear or electron magnetic resonance lines, and inelastic-scattering experi- ments, in which the energy change of the scattered particle is deter- mined along with the scattering angle. Typically, one finds that the relaxation rate of the order parameter becomes anomalously slow at a critical point. Some other relaxation rates are found to speed up, however, and transport coefficients become large in a number of cases. In some cases, the results of a dynamic experiment may be interpreted as an indirect measurement of a static property of the system. In fact, some of the most precise measurements of static critical properties have been obtained by dynamic means. Examples here are measurements of the superfluid properties of liquid helium, the low-frequency sound velocity of a fluid, and the frequency of nuclear magnetic resonance in a magnetic system. WHAT DETERMINES THE UNIVERSALITY CLASS? We now know what determines the static and dynamic universality classes in most cases, although there remain a number of difficult cases that are not resolved. Since there is no simple rule that is completely general, we shall describe here only a few simple cases and give a few examples of factors that do or do not change the universality class of these systems. Consider an idealized magnetic system in which magnetic atoms sit on the sites of an elementary periodic lattice such as a simple cubic lattice. Each magnetic atom has an elementary magnetic moment or spin, of fixed magnitude, which can point c' priori in one of several directions of space. In the using model, which we have referred to several times above, we suppose that the atomic moments can point in only two directions- either parallel or opposite to some fixed axis, which we shall take to be the z axis. Then the microscopic state of this model is determined by specifying for each atom whether the magnetic moment along the z axis is positive or negative. In the Ising model we assume an interaction
CRITICAL PHENOMENA AND PHASE TRANSITIONS 85 between nearest neighbor atoms, tending to align the moments in the same direction. At temperatures above the critical temperature, the aligning force cannot overcome the disordering tendency of random thermal motion, and there are equally many positive and negative moments. At temperatures below Tc' however, one of the two possible directions acquires a majority of the moments, and there is a net magnetization +M(I) that measures the size and direction of this majority. Thus, the Ising model has a plus-minus symmetry that is broken below TC. Another model of magnetism is the Heisenberg model. Here it is assumed that the atomic moments can point in any direction of space. The aligning force between neighboring spins is assumed to be derived from an energy that depends only on the angle between the two spins and not on their orientation with respect to any fixed axis in space. The energy of this model is unchanged if we rotate all the spins in the system by the same angle, about any direction, and we say that the model is symmetric under arbitrary rotations of the spins. This symmetry is broken below Tc' when there is a magnetization that spontaneously picks out some direction in space. A system intermediate between the Heisenberg model and the Ising model is the XY model, where the directions of the magnetic moments are restricted to lie in a single plane (say the XY plane). Again, the energy is taken to be unchanged if all spins are rotated by the same angle in this plane. The Ising model, the XY model, and the Heisenberg model may be said to have order parameters that are, respectively, a 1-D vector, a 2-D vector, and a 3-D vector. We have already seen that spatial dimensionality is crucial in determining the universality class of a system the Ising model on a 2-D lattice has different critical exponents from the 3-D Ising model, for example. The dimensionality (or symmetry) of the order parameter also turns out to be important. The critical exponents of the Ising model, XY model, and Heisenberg model in three dimensions differ from each other by a small but significant amount. (For example, the exponent ~ defined above takes on the values 0.325, 0.346, and 0.365 in the three cases.) In two dimensions the differences are more dramatic. We believe that the 2-D Heisenberg model has no phase transition at all it remains paramagnetic (disordered) at all tempera- tures other than zero. The 2-D XY model is believed to have a phase transition of a peculiar type (see below), for which the critical exponents ax, is, and my cannot be properly defined. We may next ask what happens if a model has a 3-D vector order parameter similar to the Heisenberg model, but the interactions give a
86 A DECADE OF CONDENSED-MATTER PHYSICS lower energy to spins in the z direction than to spins in the other direction of space. In this case the order parameter has the symmetry of the Ising model, and the critical exponents are those of the Ising model rather than those of the Heisenberg model. If the energy favoring the z direction is small compared to the Heisenberg-type interactions favoring parallel alignment of spins without regard to the particular direction in space, then we expect to observe a crossover behavior: close to the critical point (say IT - TO ' 10-5 Tc) we will see the critical exponents of the Ising model, but farther from TC there may be a range of temperatures (say 10-2 TC > IT - TO > 10-4 Tc) where the system appears to have the critical exponents of the Heisenberg model. An experimentalist seeking to measure accurately the critical exponents of some universality class will naturally try to avoid using systems that have a crossover in the middle of the accessible temperature range. In more complicated systems, with multicomponent order parame- ters, there is a variety of possible higher-order symmetry breaking terms, which may favor some discrete subset of the possible orienta- tions of the order parameter. In some cases these terms lead to a change in critical behavior; in some others they lead to a small fluctuation-induced first-order transition, even though the classical theory predicts a continuous transition. There are also many factors that are known to be irrelevant to deciding the universality class. The precise nature of the spatial lattice is unimportant for example, an Ising model on a hexagonal lattice in two dimensions will have the same critical exponents as on a square or rectangular lattice. The exponents are also unaffected if the interac- tions are stronger along one spatial direction than another. The universality class of a magnetic model is unchanged if the interaction between spins extends beyond nearest neighbors on the lattice, provided that the interaction falls off sufficiently rapidly with separation. In real magnetic systems, however, there is an important long-range interaction that does not fall off rapidly with distance the magnetic dipole interaction, which decreases only as the inverse cube of the distance between atoms. This is sufficiently long range to change the universality class of a ferromagnet. Particularly in an Ising system, the dipole interaction has a drastic effect on the critical behavior. For a system like iron, where the magnetic dipole interaction is weak compared with the quantum-mechanical exchange interactions respon- sible for the ferromagnetism of the material, the dipole interaction only becomes important close to the critical point. However, there also exist cases where the dipole interaction is large and one readily sees an effect on the critical behavior. Neutron scattering and specific-heat measurements on one of these systems (LiTbF4) have provided dra
CRITICAL PHENOMENA AND PHASE TRANSITIONS 87 matic confirmation of the peculiar critical behavior predicted theoret- ically for the Ising model with dipolar forces. We may remark here that the long-range magnetic dipole interactions are irrelevant to the critical behavior of antiferromagnets, because of the cancellations arising from the alternating directions of the spins in this case. We have already noted that the liquid-vapor critical point has the same critical exponents as the 3-D Ising model. The liquid-vapor order parameter, which we take as the difference from the density at the critical point, is a real quantity, which can be positive or negative like the magnetization of the Ising model, but the fluid does not possess a precise symmetry between positive and negative values of the order parameter. It is a prediction of renormalization group calculations that this remaining asymmetry is irrelevant for the critical behavior, and indeed experiments confirm with high precision the identity of the critical exponents for the fluid and Ising critical points. In two-component fluid mixtures, there is often a critical point for phase separation, which is closely analogous to the liquid-vapor critical point. This critical point also falls in the Ising universality class, and it has been studied in many experiments. The critical behavior of the XY model is particularly interesting because this model is predicted to fall in the same universality class as the superfluid transition of liquid helium (4He). In the latter case the order parameter is a complex number representing the quantum- mechanical condensate wave function of the superfluid, and the relation to the XY model results from the mathematical representation of a complex number as a vector in the XY plane. Because liquid 4He can be obtained with great purity, and because temperatures near the superfluid transition can be controlled with high precision, critical exponents have been measured with high accuracy in this system. The excellent agreement with calculations for the XY model provide both a confirmation of the modern theory of critical phenomena and an important confirmation of the theory of superfluidity as well. EXPERIMENTAL REALIZATIONS OF LOW-DIMENSIONAL SYSTEMS Although the world we live in is three dimensional, theoretical studies of 2-D systems have direct applications to systems in nature. For example, a transition between commensurate phases of a layer of atoms adsorbed on a crystalline substrate, or the melting of a commen- surate adsorbate phase, will generally fall into the same universality class as some simple 2-D model with a discrete order parameter, such
88 A DECADE OF CONDENSED-MATTER PHYSICS as the 2-D Using model or the three-state Potts model (an Ising-like model, realized by some gases adsorbed on graphoil, in which each spin can take on three, rather than two, values; the energy of an interacting pair of spins is lower if they have the same value, and higher if they are different). Recent experimental developments, including improved substrates, and the availability of synchrotron x-ray sources have made possible new precise measurements of phase transitions in adsorbed gas systems. In general, a film that is extended in two dimensions, but thin in the third dimension, will show the same critical behavior as some 2-D models. Slightly thicker films may show a crossover behavior from 3-D behavior to 2-D behavior as one gets closer to the critical point. Two-dimensional behavior can also be studied in 3-D layered systems, when the interactions between layers are sufficiently weak. In this case, one typically sees a crossover from 2-D to 3-D behavior as one gets closer to the critical point. Interesting phenomena also occur in quasi-l-D materials such as crystals with chains of magnetic atoms and only weak interactions between chains, even though a true 1-D system does not show a phase transition at finite temperature. There has been a variety of experi- ments in quasi- 1 -D and quasi-2-D systems that demonstrates the expected crossover behaviors. MULTICRITICAL POINTS Although the gas-liquid critical point of a pure fluid occurs at a single point in the pressure-temperature plane, this same critical point becomes a critical line in the three-parameter space of pressure, temperature, and composition in the case of a two-component mixture. The transition temperature of an antiferromagnet may also become a line of critical points when a uniform applied magnetic field is included as a parameter. There exists in nature a variety of special multicritical points, where several lines of critical points come together. Multicritical points have been studied experimentally in multicomponent fluid mixtures, in magnetic systems, and at the tricritical point of superfluidity and phase separation in a liquid ~He-4He mixture. SYSTEMS WITH ALMOST-BROKEN SYMMETRY Some of the most interesting phase transitions involve systems in which the low-temperature phase has a special type of order, where there is almost, but not quite, a broken symmetry. It had been noted,
CRITICA L PHk'NOMk'NA AND PHASE 7'RANS'/T/ONS' 89 as early as 1930, that thermally excited long-wavelength fluctuations should have the eject of destroying the long-range order and the broken symmetry of certain types of 2-D systems. (These include the 2-D Heisenberg and XY magnets and the 2-D superfluid.) It was noted similarly that thermally excited long-wavelength vibrational modes must destroy the periodic translational order of a 2-D crystal. Al- though, for many years, it was believed that the absence of broken symmetry implied, in turn, the absence of a phase transition in all those cases, this conclusion began to be questioned in the 1960s. In partic- ular, it was proposed that the XY magnet, the superfluid, and the crystal might have a distinct low-temperature state, in two dimensions, where the order parameter has a kind of quasi-order, in which there are correlations over arbitrarily large distances that fall off only as a small fractional power of the separation between two points. This behavior has recently been proven with complete mathematical rigor, in the case of the XY model at low temperatures. Since this power-law behavior is different from the exponential fallow of the correlation function (short- range correlations) that one finds at high temperatures in the same systems, there must be a definite temperature separating these two behaviors, which is by definition a phase transition temperature. (in the 2-D Heisenberg model, however, it is believed that there are only short-range correlations at all temperatures above zero, and, hence, there is no phase transition.) Two-Dimensional Superfluid and XY Model In the 1970s, there was developed a theory in which the transition to short-range order in the 2-D X Y model and superfluid occurs as a contin- uous transition (i.e. not first order) that can be described by the prolif- eration of point-like topological defects in the order parameter of the system. This theory makes a number of specific predictions about both static and dynamic properties near the phase transition, which differ significantly from the behaviors at other critical points. These predic- tions have been confirmed to some extent by experiments on thin films of superconductors and of superfluid helium and by numerical simulat- ions of the 2-D XY model, but the accuracy of these comparisons is not yet sufficient to be considered incontrovertible support for the theory. Melting of a Two-Dimensional Crystal An application of the point-defect mechanism to the melting of a 2-D crystal again makes a number of striking predictions, the most inter- esting of which is that there should be a new hexatic liquid-crystal
90 A DECADE OF CONDENSED-MATTER PHYSICS phase, existing in a narrow temperature region between the crystal and the isotropic liquid. The hexatic phase would possess quasi-order in the orientation of bonds between neighboring atoms but would have only short-range correlations in the positions of atoms, as one finds in the true liquid state. Many workers in the field believe, however, that some other melting mechanism (perhaps grain boundaries) will necessarily intervene and produce a first-order melting before the melting temper- ature for the point defect (dislocation) mechanism is reached. A first-order transition could make it impossible to reach the hexatic liquid crystal phase. (In 3-D bulk systems, melting is always a first-order transition.) This question continues to generate controversy, as computer sim- ulations tend to favor a first-order transition, while scattering experi- ments on incommensurate crystalline layers of argon, xenon, or methane, adsorbed on a graphite substrate, provide strong evidence for a continuous melting transition, for some range of coverages. The xenon experiments also provide evidence for the existence of a hexatic phase. Further work is necessary, however, particularly to clarify the possible erects of the crystal substrate on the melting transition. In layered phases of certain organic molecules (smectic liquid crystals), there are phase transitions arising from a change in the order within a layer, which may be considered as generalizations of the 2-D melting transition. A phase describable as a stack of hexatic layers has been observed by x-ray experiments in several smectics. Smectic A-to-Nematic Transition Although the most elementary examples of systems with almost- broken symmetry (quasi-order) are the 2-D systems discussed above, the phenomenon also occurs in certain 3-D liquid-crystal phases. The simplest of these is the smectic A phase, in which long organic molecules are arranged with their axes parallel to a particular direction of space, and, in addition, the centers of gravities of the molecules tend to be arranged in a series of equally spaced layers perpendicular to the molecular axes, i.e., there is a periodic modulation of the density in one direction. In the directions parallel to the planes there is only short- range, liquidlike order. Because it costs little energy to excite long-wavelength bends in the molecular layers, thermal fluctuations in these modes are large, and the resulting displacements reduce the periodic translational order to quasi-long-range order (power-law correlations), as in a 2-D solid. When heated, a smectic A may lose its remaining translational order,
CRITICAL PHENOMENA AND PHASE TRANSITIONS 91 while retaining the orientational order of the molecular axes. We then say the material has undergone a transition to the nematic phase. The theory of this phase transition is complicated by the coupling between the translational- and orientational-order parameters and also by the large differences in the microscopic properties of the nematic phase, when measured along different directions. Experimentally, well-defined critical exponents have been seen, holding over several decades in the distance from the transition temperature, for such properties as the translational correlation length, measured by x-ray scattering, in the nematic phase. However, the critical exponents are different in the directions parallel and perpen- dicular to the molecular axis and also vary from one material to another. QUENCHED DISORDER The discussion, until this point, has focused on systems in thermal equilibrium, where the important fluctuations arise from the intrinsic thermal population of excitations, required by the laws of statistical mechanics. In many solid-state systems of interest, however, there may be additional, frozen-in disorder quenched into the system on formation of the sample. Depending on the nature of the phase transition, the nature of the quenched disorder, and the way in which the quenched disorder couples to the order parameter of the phase transition, the critical behavior may or may not be changed by the disorder. In the most extreme cases, the phase transition may be smeared out or eliminated entirely. The effects of quenched disorder appear to be well understood in many cases, but there remain others that are poorly understood and are the subject of active investigation. One particularly interesting case occurs when the quenched disorder couples linearly to an Ising-like order parameter, as would be the case if there were a local magnetic field of random sign on each site of the Ising ferromagnet. (This situation has been realized experimentally in an Ising-like antiferromagnet, with a uniform magnetic field and randomly missing magnetic atoms.) Different theoretical approaches have led to opposite expectations for whether or not there should be a sharp phase transition in this case, and experimental measurements have not yet resolved the issue in a satisfactory manner. The question may also have implications for elementary-particle physics, because one of the theoretical approaches takes advantages of a close mathe- matical analogy between the random magnetic-field problem and so-called supersymmetric models in quantum field theory.
92 A DECADE OF CONDENSED-MATTER PHYSICS Extreme examples of systems with quenched disorder are the spin-glass phases, discussed in Chapter 4. There are many open ques- tions related to phase transitions in these systems. PERCOLATION AND THE METAL-INSULATOR TRANSITION IN DISORDERED SYSTEMS There are a number of problems in condensed-matter physics that bear a qualitative resemblance to systems at a continuous phase transition and that may indeed be understood by methods of analysis similar to those used in the theory of critical phenomena but where the source of disorder is entirely quenched randomness and not thermal fluctuations. Among these are various problems concerned with geom- etry and transport in disordered systems, including metal-insulator transitions in disordered systems where quantum mechanics plays a critical role, as well as the classical problem of percolation in a mixture of macroscopic conducting and insulating particles. NONEQUILIBRIUM SYSTEMS A number of problems resembling critical phenomena have been observed in systems out of equilibrium. As one example, there have been experimental and theoretical studies of the phase-separation critical point of a binary fluid mixture, under strong shear flow. The renormalization group concept has already had a profound impact on the theoretical ideas and mathematical techniques (e.g., iterative maps) used to describe changes of the state of motion in nonequilibrium fluids, at moderate Reynolds numbers (Chapter 111. Turbulence in fluids at high Reynolds numbers has certain features of universality and scaling that suggest that theoretical techniques used to understand critical phenomena may also have a bearing here. .. . . . FIRST-ORDER TRANSITIONS In general, one does not find at a first-order phase transition the rich variety of phenomena that one finds at a critical point, and first-order transitions have consequently received relatively less attention in recent years. Several classes of universal phenomena associated with first-order transitions do deserve mention, however, because they have been the subject of continuing research, and because they still contain outstanding puzzles. These include the areas of nucleation phenomena, limits of superheating and supercooling, spinodal decomposition, and
CRITICAL PHENOMENA AND PHASE TRANSITIONS 93 mathematical questions concerning the nature of the singularities in thermodynamic functions at a first-order transition. The past few years have seen significant advances in research on the dynamics of phase transitions in fluid mixtures, but many basic questions remain unanswered. It is now clear that the transition from spinodal decomposition (from an unstable region) to nucleation and growth (from a metastable regions is a gradual one; there is no abrupt changeover at a spinodal curve. Some progress has been made in providing a general theoretical description of the phase separation process, and machine and laboratory experiments have provided some clues about the underlying physics. A global scaling procedure, first recognized in computer simulations, has been described by simple models whose validity has been demonstrated experimentally. A long-standing doubt about the validity of nucleation theory in the neighborhood of critical points has apparently been laid to rest by a theory that demonstrates that the anomalous behavior is the result of critical slowing down of growth and by experiments that are consistent with the theory. Of course, there are many important open questions of a nonuniver- sal nature concerning first-order transitions, as there are for continuous transitions. These include such matters as understanding the micro- scopic mechanisms for various transitions and calculations of the location of the transition and of the sizes of the discontinuities of various physical quantities. Such questions are discussed elsewhere in this report, in the chapter appropriate to the particular transition. Indeed, the reader will find that phase transitions are featured in virtually every chapter of the report. OUTLOOK Work on critical phenomena and related problems, in the 1980s, should lead to progress in a number of directions, among which we may expect the following: 1. There will be more precise experimental tests of the predictions of the renormalization group theories and of some of its consequences (e.g., scaling laws among exponents and the universality of certain relations among absolute values of properties near Tc). This is neces- sary, because it is important to test thoroughly the underpinnings of a theoretical approach that is seeing such widespread application in modern physics. 2. There will be an extension of our understanding to some of the
94 A DECADE OF CONDENSED-MA TTER PHYSICS more complicated types of critical points. We may also expect a better understanding of crossover phenomena and, more generally, of the corrections to simple power-law behavior that are necessary to under- stand measurements that are some distance away from the critical point. Combinations of renormalization-group approximations and accurate microscopic models will be used increasingly to calculate entire phase diagrams, including the locations of first-order transitions, in a variety of systems. 3. Some of the outstanding problems mentioned above may be solved, perhaps by means of some important new calculational meth- ods, or perhaps by the development of some new physical ideas, or by refinements in experimental techniques. For example, it seems likely that in the 1980s considerable progress will be made in our understand- ing of 2-D melting and related phenomena, through experiments on a variety of systems, including liquid crystals, both in bulk and in suspended films of several layers thickness; adsorbed layers; the electron crystal on the surface of liquid helium; and perhaps synthetic systems, such as a film containing colloidal polystyrene spheres. The role of the substrate in the transition, in the case of adsorbed layers, will be investigated. The construction of a theory of the smectic A-to-nematic transition is one of the most challenging unsolved prob- lems in critical phenomena. Interest in this problem is heightened by its possible connection to phase transitions in idealized superconductors and in certain quantum-mechanical models of interest to elementary- particle theories. A variety of other phase transitions among other liquid-crystal phases is also poorly understood at present and will undoubtedly be the subject of major investigations in the next few years. Problems of disordered systems in which the disorder is quenched into the system during its formation, and is not due to thermal fluctuations, remain an important area where new ideas are necessary, and the theoretical methods of critical phenomena need further development. 4. The theoretical and experimental methods used to study the problems of critical phenomena will be applied to the study of other problems in condensed-matter physics.